kernel - Fix serious issue w/ smp_invltlb(), plus other issues (3)

[dragonfly.git] / sys / platform / pc32 / i386 / mp_machdep.c

/*
 * Copyright (c) 1996, by Steve Passe
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. The name of the developer may NOT be used to endorse or promote products
 *    derived from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 *
 * $FreeBSD: src/sys/i386/i386/mp_machdep.c,v 1.115.2.15 2003/03/14 21:22:35 jhb Exp $
 * $DragonFly: src/sys/platform/pc32/i386/mp_machdep.c,v 1.60 2008/06/07 12:03:52 mneumann Exp $
 */

#include "opt_cpu.h"

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/sysctl.h>
#include <sys/malloc.h>
#include <sys/memrange.h>
#include <sys/cons.h>	/* cngetc() */
#include <sys/machintr.h>

#include <vm/vm.h>
#include <vm/vm_param.h>
#include <vm/pmap.h>
#include <vm/vm_kern.h>
#include <vm/vm_extern.h>
#include <sys/lock.h>
#include <vm/vm_map.h>
#include <sys/user.h>
#ifdef GPROF 
#include <sys/gmon.h>
#endif

#include <sys/mplock2.h>

#include <machine/smp.h>
#include <machine_base/apic/apicreg.h>
#include <machine/atomic.h>
#include <machine/cpufunc.h>
#include <machine/cputypes.h>
#include <machine_base/apic/mpapic.h>
#include <machine/psl.h>
#include <machine/segments.h>
#include <machine/tss.h>
#include <machine/specialreg.h>
#include <machine/globaldata.h>
#include <machine/pmap_inval.h>

#include <machine/md_var.h>		/* setidt() */
#include <machine_base/icu/icu.h>		/* IPIs */
#include <machine_base/isa/intr_machdep.h>	/* IPIs */

#define FIXUP_EXTRA_APIC_INTS	8	/* additional entries we may create */

#define WARMBOOT_TARGET		0
#define WARMBOOT_OFF		(KERNBASE + 0x0467)
#define WARMBOOT_SEG		(KERNBASE + 0x0469)

#define BIOS_BASE		(0xf0000)
#define BIOS_BASE2		(0xe0000)
#define BIOS_SIZE		(0x10000)
#define BIOS_COUNT		(BIOS_SIZE/4)

#define CMOS_REG		(0x70)
#define CMOS_DATA		(0x71)
#define BIOS_RESET		(0x0f)
#define BIOS_WARM		(0x0a)

#define PROCENTRY_FLAG_EN	0x01
#define PROCENTRY_FLAG_BP	0x02
#define IOAPICENTRY_FLAG_EN	0x01


/* MP Floating Pointer Structure */
typedef struct MPFPS {
	char    signature[4];
	u_int32_t pap;
	u_char  length;
	u_char  spec_rev;
	u_char  checksum;
	u_char  mpfb1;
	u_char  mpfb2;
	u_char  mpfb3;
	u_char  mpfb4;
	u_char  mpfb5;
}      *mpfps_t;

/* MP Configuration Table Header */
typedef struct MPCTH {
	char    signature[4];
	u_short base_table_length;
	u_char  spec_rev;
	u_char  checksum;
	u_char  oem_id[8];
	u_char  product_id[12];
	void   *oem_table_pointer;
	u_short oem_table_size;
	u_short entry_count;
	void   *apic_address;
	u_short extended_table_length;
	u_char  extended_table_checksum;
	u_char  reserved;
}      *mpcth_t;


typedef struct PROCENTRY {
	u_char  type;
	u_char  apic_id;
	u_char  apic_version;
	u_char  cpu_flags;
	u_long  cpu_signature;
	u_long  feature_flags;
	u_long  reserved1;
	u_long  reserved2;
}      *proc_entry_ptr;

typedef struct BUSENTRY {
	u_char  type;
	u_char  bus_id;
	char    bus_type[6];
}      *bus_entry_ptr;

typedef struct IOAPICENTRY {
	u_char  type;
	u_char  apic_id;
	u_char  apic_version;
	u_char  apic_flags;
	void   *apic_address;
}      *io_apic_entry_ptr;

typedef struct INTENTRY {
	u_char  type;
	u_char  int_type;
	u_short int_flags;
	u_char  src_bus_id;
	u_char  src_bus_irq;
	u_char  dst_apic_id;
	u_char  dst_apic_int;
}      *int_entry_ptr;

/* descriptions of MP basetable entries */
typedef struct BASETABLE_ENTRY {
	u_char  type;
	u_char  length;
	char    name[16];
}       basetable_entry;

struct mptable_pos {
	mpfps_t		mp_fps;
	mpcth_t		mp_cth;
	vm_size_t	mp_cth_mapsz;
};

typedef	int	(*mptable_iter_func)(void *, const void *, int);

/*
 * this code MUST be enabled here and in mpboot.s.
 * it follows the very early stages of AP boot by placing values in CMOS ram.
 * it NORMALLY will never be needed and thus the primitive method for enabling.
 *
 */
#if defined(CHECK_POINTS)
#define CHECK_READ(A)	 (outb(CMOS_REG, (A)), inb(CMOS_DATA))
#define CHECK_WRITE(A,D) (outb(CMOS_REG, (A)), outb(CMOS_DATA, (D)))

#define CHECK_INIT(D);				\
	CHECK_WRITE(0x34, (D));			\
	CHECK_WRITE(0x35, (D));			\
	CHECK_WRITE(0x36, (D));			\
	CHECK_WRITE(0x37, (D));			\
	CHECK_WRITE(0x38, (D));			\
	CHECK_WRITE(0x39, (D));

#define CHECK_PRINT(S);				\
	kprintf("%s: %d, %d, %d, %d, %d, %d\n",	\
	   (S),					\
	   CHECK_READ(0x34),			\
	   CHECK_READ(0x35),			\
	   CHECK_READ(0x36),			\
	   CHECK_READ(0x37),			\
	   CHECK_READ(0x38),			\
	   CHECK_READ(0x39));

#else				/* CHECK_POINTS */

#define CHECK_INIT(D)
#define CHECK_PRINT(S)

#endif				/* CHECK_POINTS */

/*
 * Values to send to the POST hardware.
 */
#define MP_BOOTADDRESS_POST	0x10
#define MP_PROBE_POST		0x11
#define MPTABLE_PASS1_POST	0x12

#define MP_START_POST		0x13
#define MP_ENABLE_POST		0x14
#define MPTABLE_PASS2_POST	0x15

#define START_ALL_APS_POST	0x16
#define INSTALL_AP_TRAMP_POST	0x17
#define START_AP_POST		0x18

#define MP_ANNOUNCE_POST	0x19

/** XXX FIXME: where does this really belong, isa.h/isa.c perhaps? */
int	current_postcode;

/** XXX FIXME: what system files declare these??? */
extern struct region_descriptor r_gdt, r_idt;

int	mp_naps;		/* # of Applications processors */
#ifdef APIC_IO
static int	mp_nbusses;	/* # of busses */
int	mp_napics;		/* # of IO APICs */
#endif
#ifdef APIC_IO
vm_offset_t io_apic_address[NAPICID];	/* NAPICID is more than enough */
u_int32_t *io_apic_versions;
#endif
extern	int nkpt;

u_int32_t cpu_apic_versions[MAXCPU];
int64_t tsc0_offset;
extern int64_t tsc_offsets[];

extern u_long ebda_addr;

#ifdef APIC_IO
struct apic_intmapinfo	int_to_apicintpin[APIC_INTMAPSIZE];
#endif

/*
 * APIC ID logical/physical mapping structures.
 * We oversize these to simplify boot-time config.
 */
int     cpu_num_to_apic_id[NAPICID];
#ifdef APIC_IO
int     io_num_to_apic_id[NAPICID];
#endif
int     apic_id_to_logical[NAPICID];

/* AP uses this during bootstrap.  Do not staticize.  */
char *bootSTK;
static int bootAP;

/* Hotwire a 0->4MB V==P mapping */
extern pt_entry_t *KPTphys;

/*
 * SMP page table page.  Setup by locore to point to a page table
 * page from which we allocate per-cpu privatespace areas io_apics,
 * and so forth.
 */

#define IO_MAPPING_START_INDEX	\
		(SMP_MAXCPU * sizeof(struct privatespace) / PAGE_SIZE)

extern pt_entry_t *SMPpt;
static int SMPpt_alloc_index = IO_MAPPING_START_INDEX;

struct pcb stoppcbs[MAXCPU];

static basetable_entry basetable_entry_types[] =
{
	{0, 20, "Processor"},
	{1, 8, "Bus"},
	{2, 8, "I/O APIC"},
	{3, 8, "I/O INT"},
	{4, 8, "Local INT"}
};

/*
 * Local data and functions.
 */

static u_int	boot_address;
static u_int	base_memory;
static int	mp_finish;

static void	mp_enable(u_int boot_addr);

static int	mptable_iterate_entries(const mpcth_t,
		    mptable_iter_func, void *);
static int	mptable_probe(void);
static int	mptable_search(void);
static int	mptable_check(vm_paddr_t);
static int	mptable_search_sig(u_int32_t target, int count);
static int	mptable_hyperthread_fixup(u_int, int);
#ifdef APIC_IO
static void	mptable_pass1(struct mptable_pos *);
static void	mptable_pass2(struct mptable_pos *);
static void	mptable_default(int type);
static void	mptable_fix(void);
#endif
static int	mptable_map(struct mptable_pos *, vm_paddr_t);
static void	mptable_unmap(struct mptable_pos *);
static void	mptable_imcr(struct mptable_pos *);

static int	mptable_lapic_probe(struct lapic_enumerator *);
static void	mptable_lapic_enumerate(struct lapic_enumerator *);
static void	mptable_lapic_default(void);

#ifdef APIC_IO
static void	setup_apic_irq_mapping(void);
static int	apic_int_is_bus_type(int intr, int bus_type);
#endif
static int	start_all_aps(u_int boot_addr);
static void	install_ap_tramp(u_int boot_addr);
static int	start_ap(struct mdglobaldata *gd, u_int boot_addr, int smibest);
static int	smitest(void);

static cpumask_t smp_startup_mask = 1;	/* which cpus have been started */
cpumask_t smp_active_mask = 1;	/* which cpus are ready for IPIs etc? */
SYSCTL_INT(_machdep, OID_AUTO, smp_active, CTLFLAG_RD, &smp_active_mask, 0, "");

/*
 * Calculate usable address in base memory for AP trampoline code.
 */
u_int
mp_bootaddress(u_int basemem)
{
	POSTCODE(MP_BOOTADDRESS_POST);

	base_memory = basemem;

	boot_address = base_memory & ~0xfff;	/* round down to 4k boundary */
	if ((base_memory - boot_address) < bootMP_size)
		boot_address -= 4096;	/* not enough, lower by 4k */

	return boot_address;
}


static int
mptable_probe(void)
{
	int mpfps_paddr;

	mpfps_paddr = mptable_search();
	if (mptable_check(mpfps_paddr))
		return 0;

	return mpfps_paddr;
}

/*
 * Look for an Intel MP spec table (ie, SMP capable hardware).
 */
static int
mptable_search(void)
{
	int     x;
	u_int32_t target;
 
	/*
	 * Make sure our SMPpt[] page table is big enough to hold all the
	 * mappings we need.
	 */
	KKASSERT(IO_MAPPING_START_INDEX < NPTEPG - 2);

	POSTCODE(MP_PROBE_POST);

	/* see if EBDA exists */
	if (ebda_addr != 0) {
		/* search first 1K of EBDA */
		target = (u_int32_t)ebda_addr;
		if ((x = mptable_search_sig(target, 1024 / 4)) > 0)
			return x;
	} else {
		/* last 1K of base memory, effective 'top of base' passed in */
		target = (u_int32_t)(base_memory - 0x400);
		if ((x = mptable_search_sig(target, 1024 / 4)) > 0)
			return x;
	}

	/* search the BIOS */
	target = (u_int32_t)BIOS_BASE;
	if ((x = mptable_search_sig(target, BIOS_COUNT)) > 0)
		return x;

	/* search the extended BIOS */
	target = (u_int32_t)BIOS_BASE2;
	if ((x = mptable_search_sig(target, BIOS_COUNT)) > 0)
		return x;

	/* nothing found */
	return 0;
}

struct mptable_check_cbarg {
	int	cpu_count;
	int	found_bsp;
};

static int
mptable_check_callback(void *xarg, const void *pos, int type)
{
	const struct PROCENTRY *ent;
	struct mptable_check_cbarg *arg = xarg;

	if (type != 0)
		return 0;
	ent = pos;

	if ((ent->cpu_flags & PROCENTRY_FLAG_EN) == 0)
		return 0;
	arg->cpu_count++;

	if (ent->cpu_flags & PROCENTRY_FLAG_BP) {
		if (arg->found_bsp) {
			kprintf("more than one BSP in base MP table\n");
			return EINVAL;
		}
		arg->found_bsp = 1;
	}
	return 0;
}

static int
mptable_check(vm_paddr_t mpfps_paddr)
{
	struct mptable_pos mpt;
	struct mptable_check_cbarg arg;
	mpcth_t cth;
	int error;

	if (mpfps_paddr == 0)
		return EOPNOTSUPP;

	error = mptable_map(&mpt, mpfps_paddr);
	if (error)
		return error;

	if (mpt.mp_fps->mpfb1 != 0)
		goto done;

	error = EINVAL;

	cth = mpt.mp_cth;
	if (cth == NULL)
		goto done;
	if (cth->apic_address == 0)
		goto done;

	bzero(&arg, sizeof(arg));
	error = mptable_iterate_entries(cth, mptable_check_callback, &arg);
	if (!error) {
		if (arg.cpu_count == 0) {
			kprintf("MP table contains no processor entries\n");
			error = EINVAL;
		} else if (!arg.found_bsp) {
			kprintf("MP table does not contains BSP entry\n");
			error = EINVAL;
		}
	}
done:
	mptable_unmap(&mpt);
	return error;
}

static int
mptable_iterate_entries(const mpcth_t cth, mptable_iter_func func, void *arg)
{
	int count, total_size;
	const void *position;

	KKASSERT(cth->base_table_length >= sizeof(struct MPCTH));
	total_size = cth->base_table_length - sizeof(struct MPCTH);
	position = (const uint8_t *)cth + sizeof(struct MPCTH);
	count = cth->entry_count;

	while (count--) {
		int type, error;

		KKASSERT(total_size >= 0);
		if (total_size == 0) {
			kprintf("invalid base MP table, "
				"entry count and length mismatch\n");
			return EINVAL;
		}

		type = *(const uint8_t *)position;
		switch (type) {
		case 0: /* processor_entry */
		case 1: /* bus_entry */
		case 2: /* io_apic_entry */
		case 3: /* int_entry */
		case 4:	/* int_entry */
			break;
		default:
			kprintf("unknown base MP table entry type %d\n", type);
			return EINVAL;
		}

		if (total_size < basetable_entry_types[type].length) {
			kprintf("invalid base MP table length, "
				"does not contain all entries\n");
			return EINVAL;
		}
		total_size -= basetable_entry_types[type].length;

		error = func(arg, position, type);
		if (error)
			return error;

		position = (const uint8_t *)position +
		    basetable_entry_types[type].length;
	}
	return 0;
}


/*
 * Startup the SMP processors.
 */
void
mp_start(void)
{
	POSTCODE(MP_START_POST);
	mp_enable(boot_address);
}


/*
 * Print various information about the SMP system hardware and setup.
 */
void
mp_announce(void)
{
	int     x;

	POSTCODE(MP_ANNOUNCE_POST);

	kprintf("DragonFly/MP: Multiprocessor motherboard\n");
	kprintf(" cpu0 (BSP): apic id: %2d", CPU_TO_ID(0));
	kprintf(", version: 0x%08x\n", cpu_apic_versions[0]);
	for (x = 1; x <= mp_naps; ++x) {
		kprintf(" cpu%d (AP):  apic id: %2d", x, CPU_TO_ID(x));
		kprintf(", version: 0x%08x\n", cpu_apic_versions[x]);
	}

#if defined(APIC_IO)
	for (x = 0; x < mp_napics; ++x) {
		kprintf(" io%d (APIC): apic id: %2d", x, IO_TO_ID(x));
		kprintf(", version: 0x%08x", io_apic_versions[x]);
		kprintf(", at 0x%08lx\n", io_apic_address[x]);
	}
#else
	kprintf(" Warning: APIC I/O disabled\n");
#endif	/* APIC_IO */
}

/*
 * AP cpu's call this to sync up protected mode.
 *
 * WARNING!  We must ensure that the cpu is sufficiently initialized to
 * be able to use to the FP for our optimized bzero/bcopy code before
 * we enter more mainstream C code.
 *
 * WARNING! %fs is not set up on entry.  This routine sets up %fs.
 */
void
init_secondary(void)
{
	int	gsel_tss;
	int	x, myid = bootAP;
	u_int	cr0;
	struct mdglobaldata *md;
	struct privatespace *ps;

	ps = &CPU_prvspace[myid];

	gdt_segs[GPRIV_SEL].ssd_base = (int)ps;
	gdt_segs[GPROC0_SEL].ssd_base =
		(int) &ps->mdglobaldata.gd_common_tss;
	ps->mdglobaldata.mi.gd_prvspace = ps;

	for (x = 0; x < NGDT; x++) {
		ssdtosd(&gdt_segs[x], &gdt[myid * NGDT + x].sd);
	}

	r_gdt.rd_limit = NGDT * sizeof(gdt[0]) - 1;
	r_gdt.rd_base = (int) &gdt[myid * NGDT];
	lgdt(&r_gdt);			/* does magic intra-segment return */

	lidt(&r_idt);

	lldt(_default_ldt);
	mdcpu->gd_currentldt = _default_ldt;

	gsel_tss = GSEL(GPROC0_SEL, SEL_KPL);
	gdt[myid * NGDT + GPROC0_SEL].sd.sd_type = SDT_SYS386TSS;

	md = mdcpu;	/* loaded through %fs:0 (mdglobaldata.mi.gd_prvspace)*/

	md->gd_common_tss.tss_esp0 = 0;	/* not used until after switch */
	md->gd_common_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
	md->gd_common_tss.tss_ioopt = (sizeof md->gd_common_tss) << 16;
	md->gd_tss_gdt = &gdt[myid * NGDT + GPROC0_SEL].sd;
	md->gd_common_tssd = *md->gd_tss_gdt;
	ltr(gsel_tss);

	/*
	 * Set to a known state:
	 * Set by mpboot.s: CR0_PG, CR0_PE
	 * Set by cpu_setregs: CR0_NE, CR0_MP, CR0_TS, CR0_WP, CR0_AM
	 */
	cr0 = rcr0();
	cr0 &= ~(CR0_CD | CR0_NW | CR0_EM);
	load_cr0(cr0);
	pmap_set_opt();		/* PSE/4MB pages, etc */

	/* set up CPU registers and state */
	cpu_setregs();

	/* set up FPU state on the AP */
	npxinit(__INITIAL_NPXCW__);

	/* set up SSE registers */
	enable_sse();
}

/*******************************************************************
 * local functions and data
 */

/*
 * start the SMP system
 */
static void
mp_enable(u_int boot_addr)
{
#if defined(APIC_IO)
	int     apic;
	u_int   ux;
#endif	/* APIC_IO */
	vm_paddr_t mpfps_paddr;
	struct mptable_pos mpt;

	POSTCODE(MP_ENABLE_POST);

	lapic_config();

	mpfps_paddr = mptable_probe();
	if (mpfps_paddr) {
		mptable_map(&mpt, mpfps_paddr);
		mptable_imcr(&mpt);
		mptable_unmap(&mpt);
	}
#if defined(APIC_IO)

	if (!mpfps_paddr)
		panic("no MP table, disable APIC_IO!\n");

	mptable_map(&mpt, mpfps_paddr);

	/*
	 * Examine the MP table for needed info
	 */
	mptable_pass1(&mpt);
	mptable_pass2(&mpt);

	mptable_unmap(&mpt);

	/* Post scan cleanup */
	mptable_fix();

	setup_apic_irq_mapping();

	/* fill the LOGICAL io_apic_versions table */
	for (apic = 0; apic < mp_napics; ++apic) {
		ux = io_apic_read(apic, IOAPIC_VER);
		io_apic_versions[apic] = ux;
		io_apic_set_id(apic, IO_TO_ID(apic));
	}

	/* program each IO APIC in the system */
	for (apic = 0; apic < mp_napics; ++apic)
		if (io_apic_setup(apic) < 0)
			panic("IO APIC setup failure");

#endif	/* APIC_IO */

	/*
	 * These are required for SMP operation
	 */

	/* install a 'Spurious INTerrupt' vector */
	setidt(XSPURIOUSINT_OFFSET, Xspuriousint,
	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));

	/* install an inter-CPU IPI for TLB invalidation */
	setidt(XINVLTLB_OFFSET, Xinvltlb,
	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));

	/* install an inter-CPU IPI for IPIQ messaging */
	setidt(XIPIQ_OFFSET, Xipiq,
	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));

	/* install a timer vector */
	setidt(XTIMER_OFFSET, Xtimer,
	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
	
	/* install an inter-CPU IPI for CPU stop/restart */
	setidt(XCPUSTOP_OFFSET, Xcpustop,
	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));

	/* start each Application Processor */
	start_all_aps(boot_addr);
}


/*
 * look for the MP spec signature
 */

/* string defined by the Intel MP Spec as identifying the MP table */
#define MP_SIG		0x5f504d5f	/* _MP_ */
#define NEXT(X)		((X) += 4)
static int
mptable_search_sig(u_int32_t target, int count)
{
	vm_size_t map_size;
	u_int32_t *addr;
	int x, ret;

	KKASSERT(target != 0);

	map_size = count * sizeof(u_int32_t);
	addr = pmap_mapdev((vm_paddr_t)target, map_size);

	ret = 0;
	for (x = 0; x < count; NEXT(x)) {
		if (addr[x] == MP_SIG) {
			/* make array index a byte index */
			ret = target + (x * sizeof(u_int32_t));
			break;
		}
	}

	pmap_unmapdev((vm_offset_t)addr, map_size);
	return ret;
}


typedef struct BUSDATA {
	u_char  bus_id;
	enum busTypes bus_type;
}       bus_datum;

typedef struct INTDATA {
	u_char  int_type;
	u_short int_flags;
	u_char  src_bus_id;
	u_char  src_bus_irq;
	u_char  dst_apic_id;
	u_char  dst_apic_int;
	u_char	int_vector;
}       io_int, local_int;

typedef struct BUSTYPENAME {
	u_char  type;
	char    name[7];
}       bus_type_name;

#ifdef APIC_IO

static bus_type_name bus_type_table[] =
{
	{CBUS, "CBUS"},
	{CBUSII, "CBUSII"},
	{EISA, "EISA"},
	{MCA, "MCA"},
	{UNKNOWN_BUSTYPE, "---"},
	{ISA, "ISA"},
	{MCA, "MCA"},
	{UNKNOWN_BUSTYPE, "---"},
	{UNKNOWN_BUSTYPE, "---"},
	{UNKNOWN_BUSTYPE, "---"},
	{UNKNOWN_BUSTYPE, "---"},
	{UNKNOWN_BUSTYPE, "---"},
	{PCI, "PCI"},
	{UNKNOWN_BUSTYPE, "---"},
	{UNKNOWN_BUSTYPE, "---"},
	{UNKNOWN_BUSTYPE, "---"},
	{UNKNOWN_BUSTYPE, "---"},
	{XPRESS, "XPRESS"},
	{UNKNOWN_BUSTYPE, "---"}
};
/* from MP spec v1.4, table 5-1 */
static int default_data[7][5] =
{
/*   nbus, id0, type0, id1, type1 */
	{1, 0, ISA, 255, 255},
	{1, 0, EISA, 255, 255},
	{1, 0, EISA, 255, 255},
	{1, 0, MCA, 255, 255},
	{2, 0, ISA, 1, PCI},
	{2, 0, EISA, 1, PCI},
	{2, 0, MCA, 1, PCI}
};


/* the bus data */
static bus_datum *bus_data;

/* the IO INT data, one entry per possible APIC INTerrupt */
static io_int  *io_apic_ints;
static int nintrs;

#endif

static int processor_entry	(const struct PROCENTRY *entry, int cpu);
#ifdef APIC_IO
static int bus_entry		(const struct BUSENTRY *entry, int bus);
static int io_apic_entry	(const struct IOAPICENTRY *entry, int apic);
static int int_entry		(const struct INTENTRY *entry, int intr);
static int lookup_bus_type	(char *name);
#endif

#ifdef APIC_IO

static int
mptable_ioapic_pass1_callback(void *xarg, const void *pos, int type)
{
	const struct IOAPICENTRY *ioapic_ent;

	switch (type) {
	case 1: /* bus_entry */
		++mp_nbusses;
		break;

	case 2: /* io_apic_entry */
		ioapic_ent = pos;
		if (ioapic_ent->apic_flags & IOAPICENTRY_FLAG_EN) {
			io_apic_address[mp_napics++] =
			    (vm_offset_t)ioapic_ent->apic_address;
		}
		break;

	case 3: /* int_entry */
		++nintrs;
		break;
	}
	return 0;
}

/*
 * 1st pass on motherboard's Intel MP specification table.
 *
 * determines:
 *	io_apic_address[N]
 *	mp_nbusses
 *	mp_napics
 *	nintrs
 */
static void
mptable_pass1(struct mptable_pos *mpt)
{
	mpfps_t fps;
	int x;

	POSTCODE(MPTABLE_PASS1_POST);

	fps = mpt->mp_fps;
	KKASSERT(fps != NULL);

	/* clear various tables */
	for (x = 0; x < NAPICID; ++x)
		io_apic_address[x] = ~0;	/* IO APIC address table */

	mp_nbusses = 0;
	mp_napics = 0;
	nintrs = 0;

	/* check for use of 'default' configuration */
	if (fps->mpfb1 != 0) {
		io_apic_address[0] = DEFAULT_IO_APIC_BASE;
		mp_nbusses = default_data[fps->mpfb1 - 1][0];
		mp_napics = 1;
		nintrs = 16;
	} else {
		int error;

		error = mptable_iterate_entries(mpt->mp_cth,
			    mptable_ioapic_pass1_callback, NULL);
		if (error)
			panic("mptable_iterate_entries(ioapic_pass1) failed\n");
	}
}

struct mptable_ioapic2_cbarg {
	int	bus;
	int	apic;
	int	intr;
};

static int
mptable_ioapic_pass2_callback(void *xarg, const void *pos, int type)
{
	struct mptable_ioapic2_cbarg *arg = xarg;

	switch (type) {
	case 1:
		if (bus_entry(pos, arg->bus))
			++arg->bus;
		break;

	case 2:
		if (io_apic_entry(pos, arg->apic))
			++arg->apic;
		break;

	case 3:
		if (int_entry(pos, arg->intr))
			++arg->intr;
		break;
	}
	return 0;
}

/*
 * 2nd pass on motherboard's Intel MP specification table.
 *
 * sets:
 *	ID_TO_IO(N), phy APIC ID to log CPU/IO table
 *	IO_TO_ID(N), logical IO to APIC ID table
 *	bus_data[N]
 *	io_apic_ints[N]
 */
static void
mptable_pass2(struct mptable_pos *mpt)
{
	struct mptable_ioapic2_cbarg arg;
	mpfps_t fps;
	int error, x;

	POSTCODE(MPTABLE_PASS2_POST);

	fps = mpt->mp_fps;
	KKASSERT(fps != NULL);

	MALLOC(io_apic_versions, u_int32_t *, sizeof(u_int32_t) * mp_napics,
	    M_DEVBUF, M_WAITOK);
	MALLOC(ioapic, volatile ioapic_t **, sizeof(ioapic_t *) * mp_napics,
	    M_DEVBUF, M_WAITOK | M_ZERO);
	MALLOC(io_apic_ints, io_int *, sizeof(io_int) * (nintrs + FIXUP_EXTRA_APIC_INTS),
	    M_DEVBUF, M_WAITOK);
	MALLOC(bus_data, bus_datum *, sizeof(bus_datum) * mp_nbusses,
	    M_DEVBUF, M_WAITOK);

	for (x = 0; x < mp_napics; x++)
		ioapic[x] = permanent_io_mapping(io_apic_address[x]);

	/* clear various tables */
	for (x = 0; x < NAPICID; ++x) {
		ID_TO_IO(x) = -1;	/* phy APIC ID to log CPU/IO table */
		IO_TO_ID(x) = -1;	/* logical IO to APIC ID table */
	}

	/* clear bus data table */
	for (x = 0; x < mp_nbusses; ++x)
		bus_data[x].bus_id = 0xff;

	/* clear IO APIC INT table */
	for (x = 0; x < (nintrs + 1); ++x) {
		io_apic_ints[x].int_type = 0xff;
		io_apic_ints[x].int_vector = 0xff;
	}

	/* check for use of 'default' configuration */
	if (fps->mpfb1 != 0) {
		mptable_default(fps->mpfb1);
		return;
	}

	bzero(&arg, sizeof(arg));
	error = mptable_iterate_entries(mpt->mp_cth,
		    mptable_ioapic_pass2_callback, &arg);
	if (error)
		panic("mptable_iterate_entries(ioapic_pass2) failed\n");
}

#endif	/* APIC_IO */

/*
 * Check if we should perform a hyperthreading "fix-up" to
 * enumerate any logical CPU's that aren't already listed
 * in the table.
 *
 * XXX: We assume that all of the physical CPUs in the
 * system have the same number of logical CPUs.
 *
 * XXX: We assume that APIC ID's are allocated such that
 * the APIC ID's for a physical processor are aligned
 * with the number of logical CPU's in the processor.
 */
static int
mptable_hyperthread_fixup(u_int id_mask, int cpu_count)
{
	int i, id, lcpus_max, logical_cpus;

	if ((cpu_feature & CPUID_HTT) == 0)
		return 0;

	lcpus_max = (cpu_procinfo & CPUID_HTT_CORES) >> 16;
	if (lcpus_max <= 1)
		return 0;

	if (cpu_vendor_id == CPU_VENDOR_INTEL) {
		/*
		 * INSTRUCTION SET REFERENCE, A-M (#253666)
		 * Page 3-181, Table 3-20
		 * "The nearest power-of-2 integer that is not smaller
		 *  than EBX[23:16] is the number of unique initial APIC
		 *  IDs reserved for addressing different logical
		 *  processors in a physical package."
		 */
		for (i = 0; ; ++i) {
			if ((1 << i) >= lcpus_max) {
				lcpus_max = 1 << i;
				break;
			}
		}
	}

	KKASSERT(cpu_count != 0);
	if (cpu_count == lcpus_max) {
		/* We have nothing to fix */
		return 0;
	} else if (cpu_count == 1) {
		/* XXX this may be incorrect */
		logical_cpus = lcpus_max;
	} else {
		int cur, prev, dist;

		/*
		 * Calculate the distances between two nearest
		 * APIC IDs.  If all such distances are same,
		 * then it is the number of missing cpus that
		 * we are going to fill later.
		 */
		dist = cur = prev = -1;
		for (id = 0; id < MAXCPU; ++id) {
			if ((id_mask & 1 << id) == 0)
				continue;

			cur = id;
			if (prev >= 0) {
				int new_dist = cur - prev;

				if (dist < 0)
					dist = new_dist;

				/*
				 * Make sure that all distances
				 * between two nearest APIC IDs
				 * are same.
				 */
				if (dist != new_dist)
					return 0;
			}
			prev = cur;
		}
		if (dist == 1)
			return 0;

		/* Must be power of 2 */
		if (dist & (dist - 1))
			return 0;

		/* Can't exceed CPU package capacity */
		if (dist > lcpus_max)
			logical_cpus = lcpus_max;
		else
			logical_cpus = dist;
	}

	/*
	 * For each APIC ID of a CPU that is set in the mask,
	 * scan the other candidate APIC ID's for this
	 * physical processor.  If any of those ID's are
	 * already in the table, then kill the fixup.
	 */
	for (id = 0; id < MAXCPU; id++) {
		if ((id_mask & 1 << id) == 0)
			continue;
		/* First, make sure we are on a logical_cpus boundary. */
		if (id % logical_cpus != 0)
			return 0;
		for (i = id + 1; i < id + logical_cpus; i++)
			if ((id_mask & 1 << i) != 0)
				return 0;
	}
	return logical_cpus;
}

static int
mptable_map(struct mptable_pos *mpt, vm_paddr_t mpfps_paddr)
{
	mpfps_t fps = NULL;
	mpcth_t cth = NULL;
	vm_size_t cth_mapsz = 0;

	bzero(mpt, sizeof(*mpt));

	fps = pmap_mapdev(mpfps_paddr, sizeof(*fps));
	if (fps->pap != 0) {
		/*
		 * Map configuration table header to get
		 * the base table size
		 */
		cth = pmap_mapdev(fps->pap, sizeof(*cth));
		cth_mapsz = cth->base_table_length;
		pmap_unmapdev((vm_offset_t)cth, sizeof(*cth));

		if (cth_mapsz < sizeof(*cth)) {
			kprintf("invalid base MP table length %d\n",
				(int)cth_mapsz);
			pmap_unmapdev((vm_offset_t)fps, sizeof(*fps));
			return EINVAL;
		}

		/*
		 * Map the base table
		 */
		cth = pmap_mapdev(fps->pap, cth_mapsz);
	}

	mpt->mp_fps = fps;
	mpt->mp_cth = cth;
	mpt->mp_cth_mapsz = cth_mapsz;

	return 0;
}

static void
mptable_unmap(struct mptable_pos *mpt)
{
	if (mpt->mp_cth != NULL) {
		pmap_unmapdev((vm_offset_t)mpt->mp_cth, mpt->mp_cth_mapsz);
		mpt->mp_cth = NULL;
		mpt->mp_cth_mapsz = 0;
	}
	if (mpt->mp_fps != NULL) {
		pmap_unmapdev((vm_offset_t)mpt->mp_fps, sizeof(*mpt->mp_fps));
		mpt->mp_fps = NULL;
	}
}

#ifdef APIC_IO

void
assign_apic_irq(int apic, int intpin, int irq)
{
	int x;
	
	if (int_to_apicintpin[irq].ioapic != -1)
		panic("assign_apic_irq: inconsistent table");
	
	int_to_apicintpin[irq].ioapic = apic;
	int_to_apicintpin[irq].int_pin = intpin;
	int_to_apicintpin[irq].apic_address = ioapic[apic];
	int_to_apicintpin[irq].redirindex = IOAPIC_REDTBL + 2 * intpin;
	
	for (x = 0; x < nintrs; x++) {
		if ((io_apic_ints[x].int_type == 0 || 
		     io_apic_ints[x].int_type == 3) &&
		    io_apic_ints[x].int_vector == 0xff &&
		    io_apic_ints[x].dst_apic_id == IO_TO_ID(apic) &&
		    io_apic_ints[x].dst_apic_int == intpin)
			io_apic_ints[x].int_vector = irq;
	}
}

void
revoke_apic_irq(int irq)
{
	int x;
	int oldapic;
	int oldintpin;
	
	if (int_to_apicintpin[irq].ioapic == -1)
		panic("revoke_apic_irq: inconsistent table");
	
	oldapic = int_to_apicintpin[irq].ioapic;
	oldintpin = int_to_apicintpin[irq].int_pin;

	int_to_apicintpin[irq].ioapic = -1;
	int_to_apicintpin[irq].int_pin = 0;
	int_to_apicintpin[irq].apic_address = NULL;
	int_to_apicintpin[irq].redirindex = 0;
	
	for (x = 0; x < nintrs; x++) {
		if ((io_apic_ints[x].int_type == 0 || 
		     io_apic_ints[x].int_type == 3) &&
		    io_apic_ints[x].int_vector != 0xff &&
		    io_apic_ints[x].dst_apic_id == IO_TO_ID(oldapic) &&
		    io_apic_ints[x].dst_apic_int == oldintpin)
			io_apic_ints[x].int_vector = 0xff;
	}
}

/*
 * Allocate an IRQ 
 */
static void
allocate_apic_irq(int intr)
{
	int apic;
	int intpin;
	int irq;
	
	if (io_apic_ints[intr].int_vector != 0xff)
		return;		/* Interrupt handler already assigned */
	
	if (io_apic_ints[intr].int_type != 0 &&
	    (io_apic_ints[intr].int_type != 3 ||
	     (io_apic_ints[intr].dst_apic_id == IO_TO_ID(0) &&
	      io_apic_ints[intr].dst_apic_int == 0)))
		return;		/* Not INT or ExtInt on != (0, 0) */
	
	irq = 0;
	while (irq < APIC_INTMAPSIZE &&
	       int_to_apicintpin[irq].ioapic != -1)
		irq++;
	
	if (irq >= APIC_INTMAPSIZE)
		return;		/* No free interrupt handlers */
	
	apic = ID_TO_IO(io_apic_ints[intr].dst_apic_id);
	intpin = io_apic_ints[intr].dst_apic_int;
	
	assign_apic_irq(apic, intpin, irq);
	io_apic_setup_intpin(apic, intpin);
}


static void
swap_apic_id(int apic, int oldid, int newid)
{
	int x;
	int oapic;
	

	if (oldid == newid)
		return;			/* Nothing to do */
	
	kprintf("Changing APIC ID for IO APIC #%d from %d to %d in MP table\n",
	       apic, oldid, newid);
	
	/* Swap physical APIC IDs in interrupt entries */
	for (x = 0; x < nintrs; x++) {
		if (io_apic_ints[x].dst_apic_id == oldid)
			io_apic_ints[x].dst_apic_id = newid;
		else if (io_apic_ints[x].dst_apic_id == newid)
			io_apic_ints[x].dst_apic_id = oldid;
	}
	
	/* Swap physical APIC IDs in IO_TO_ID mappings */
	for (oapic = 0; oapic < mp_napics; oapic++)
		if (IO_TO_ID(oapic) == newid)
			break;
	
	if (oapic < mp_napics) {
		kprintf("Changing APIC ID for IO APIC #%d from "
		       "%d to %d in MP table\n",
		       oapic, newid, oldid);
		IO_TO_ID(oapic) = oldid;
	}
	IO_TO_ID(apic) = newid;
}


static void
fix_id_to_io_mapping(void)
{
	int x;

	for (x = 0; x < NAPICID; x++)
		ID_TO_IO(x) = -1;
	
	for (x = 0; x <= mp_naps; x++)
		if (CPU_TO_ID(x) < NAPICID)
			ID_TO_IO(CPU_TO_ID(x)) = x;
	
	for (x = 0; x < mp_napics; x++)
		if (IO_TO_ID(x) < NAPICID)
			ID_TO_IO(IO_TO_ID(x)) = x;
}


static int
first_free_apic_id(void)
{
	int freeid, x;
	
	for (freeid = 0; freeid < NAPICID; freeid++) {
		for (x = 0; x <= mp_naps; x++)
			if (CPU_TO_ID(x) == freeid)
				break;
		if (x <= mp_naps)
			continue;
		for (x = 0; x < mp_napics; x++)
			if (IO_TO_ID(x) == freeid)
				break;
		if (x < mp_napics)
			continue;
		return freeid;
	}
	return freeid;
}


static int
io_apic_id_acceptable(int apic, int id)
{
	int cpu;		/* Logical CPU number */
	int oapic;		/* Logical IO APIC number for other IO APIC */

	if (id >= NAPICID)
		return 0;	/* Out of range */
	
	for (cpu = 0; cpu <= mp_naps; cpu++)
		if (CPU_TO_ID(cpu) == id)
			return 0;	/* Conflict with CPU */
	
	for (oapic = 0; oapic < mp_napics && oapic < apic; oapic++)
		if (IO_TO_ID(oapic) == id)
			return 0;	/* Conflict with other APIC */
	
	return 1;		/* ID is acceptable for IO APIC */
}

static
io_int *
io_apic_find_int_entry(int apic, int pin)
{
	int     x;

	/* search each of the possible INTerrupt sources */
	for (x = 0; x < nintrs; ++x) {
		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
		    (pin == io_apic_ints[x].dst_apic_int))
			return (&io_apic_ints[x]);
	}
	return NULL;
}

/*
 * parse an Intel MP specification table
 */
static void
mptable_fix(void)
{
	int	x;
	int	id;
	int	apic;		/* IO APIC unit number */
	int     freeid;		/* Free physical APIC ID */
	int	physid;		/* Current physical IO APIC ID */
	io_int *io14;
	int	bus_0 = 0;	/* Stop GCC warning */
	int	bus_pci = 0;	/* Stop GCC warning */
	int	num_pci_bus;

	/*
	 * Fix mis-numbering of the PCI bus and its INT entries if the BIOS
	 * did it wrong.  The MP spec says that when more than 1 PCI bus
	 * exists the BIOS must begin with bus entries for the PCI bus and use
	 * actual PCI bus numbering.  This implies that when only 1 PCI bus
	 * exists the BIOS can choose to ignore this ordering, and indeed many
	 * MP motherboards do ignore it.  This causes a problem when the PCI
	 * sub-system makes requests of the MP sub-system based on PCI bus
	 * numbers.	So here we look for the situation and renumber the
	 * busses and associated INTs in an effort to "make it right".
	 */

	/* find bus 0, PCI bus, count the number of PCI busses */
	for (num_pci_bus = 0, x = 0; x < mp_nbusses; ++x) {
		if (bus_data[x].bus_id == 0) {
			bus_0 = x;
		}
		if (bus_data[x].bus_type == PCI) {
			++num_pci_bus;
			bus_pci = x;
		}
	}
	/*
	 * bus_0 == slot of bus with ID of 0
	 * bus_pci == slot of last PCI bus encountered
	 */

	/* check the 1 PCI bus case for sanity */
	/* if it is number 0 all is well */
	if (num_pci_bus == 1 &&
	    bus_data[bus_pci].bus_id != 0) {
		
		/* mis-numbered, swap with whichever bus uses slot 0 */

		/* swap the bus entry types */
		bus_data[bus_pci].bus_type = bus_data[bus_0].bus_type;
		bus_data[bus_0].bus_type = PCI;

		/* swap each relavant INTerrupt entry */
		id = bus_data[bus_pci].bus_id;
		for (x = 0; x < nintrs; ++x) {
			if (io_apic_ints[x].src_bus_id == id) {
				io_apic_ints[x].src_bus_id = 0;
			}
			else if (io_apic_ints[x].src_bus_id == 0) {
				io_apic_ints[x].src_bus_id = id;
			}
		}
	}

	/* Assign IO APIC IDs.
	 * 
	 * First try the existing ID. If a conflict is detected, try
	 * the ID in the MP table.  If a conflict is still detected, find
	 * a free id.
	 *
	 * We cannot use the ID_TO_IO table before all conflicts has been
	 * resolved and the table has been corrected.
	 */
	for (apic = 0; apic < mp_napics; ++apic) { /* For all IO APICs */
		
		/* First try to use the value set by the BIOS */
		physid = io_apic_get_id(apic);
		if (io_apic_id_acceptable(apic, physid)) {
			if (IO_TO_ID(apic) != physid)
				swap_apic_id(apic, IO_TO_ID(apic), physid);
			continue;
		}

		/* Then check if the value in the MP table is acceptable */
		if (io_apic_id_acceptable(apic, IO_TO_ID(apic)))
			continue;

		/* Last resort, find a free APIC ID and use it */
		freeid = first_free_apic_id();
		if (freeid >= NAPICID)
			panic("No free physical APIC IDs found");
		
		if (io_apic_id_acceptable(apic, freeid)) {
			swap_apic_id(apic, IO_TO_ID(apic), freeid);
			continue;
		}
		panic("Free physical APIC ID not usable");
	}
	fix_id_to_io_mapping();

	/* detect and fix broken Compaq MP table */
	if (apic_int_type(0, 0) == -1) {
		kprintf("APIC_IO: MP table broken: 8259->APIC entry missing!\n");
		io_apic_ints[nintrs].int_type = 3;	/* ExtInt */
		io_apic_ints[nintrs].int_vector = 0xff;	/* Unassigned */
		/* XXX fixme, set src bus id etc, but it doesn't seem to hurt */
		io_apic_ints[nintrs].dst_apic_id = IO_TO_ID(0);
		io_apic_ints[nintrs].dst_apic_int = 0;	/* Pin 0 */
		nintrs++;
	} else if (apic_int_type(0, 0) == 0) {
		kprintf("APIC_IO: MP table broken: ExtINT entry corrupt!\n");
		for (x = 0; x < nintrs; ++x)
			if ((0 == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
			    (0 == io_apic_ints[x].dst_apic_int)) {
				io_apic_ints[x].int_type = 3;
				io_apic_ints[x].int_vector = 0xff;
				break;
			}
	}

	/*
	 * Fix missing IRQ 15 when IRQ 14 is an ISA interrupt.  IDE
	 * controllers universally come in pairs.  If IRQ 14 is specified
	 * as an ISA interrupt, then IRQ 15 had better be too.
	 *
	 * [ Shuttle XPC / AMD Athlon X2 ]
	 *	The MPTable is missing an entry for IRQ 15.  Note that the
	 *	ACPI table has an entry for both 14 and 15.
	 */
	if (apic_int_type(0, 14) == 0 && apic_int_type(0, 15) == -1) {
		kprintf("APIC_IO: MP table broken: IRQ 15 not ISA when IRQ 14 is!\n");
		io14 = io_apic_find_int_entry(0, 14);
		io_apic_ints[nintrs] = *io14;
		io_apic_ints[nintrs].src_bus_irq = 15;
		io_apic_ints[nintrs].dst_apic_int = 15;
		nintrs++;
	}
}

/* Assign low level interrupt handlers */
static void
setup_apic_irq_mapping(void)
{
	int	x;
	int	int_vector;

	/* Clear array */
	for (x = 0; x < APIC_INTMAPSIZE; x++) {
		int_to_apicintpin[x].ioapic = -1;
		int_to_apicintpin[x].int_pin = 0;
		int_to_apicintpin[x].apic_address = NULL;
		int_to_apicintpin[x].redirindex = 0;

		/* Default to masked */
		int_to_apicintpin[x].flags = IOAPIC_IM_FLAG_MASKED;
	}

	/* First assign ISA/EISA interrupts */
	for (x = 0; x < nintrs; x++) {
		int_vector = io_apic_ints[x].src_bus_irq;
		if (int_vector < APIC_INTMAPSIZE &&
		    io_apic_ints[x].int_vector == 0xff && 
		    int_to_apicintpin[int_vector].ioapic == -1 &&
		    (apic_int_is_bus_type(x, ISA) ||
		     apic_int_is_bus_type(x, EISA)) &&
		    io_apic_ints[x].int_type == 0) {
			assign_apic_irq(ID_TO_IO(io_apic_ints[x].dst_apic_id), 
					io_apic_ints[x].dst_apic_int,
					int_vector);
		}
	}

	/* Assign ExtInt entry if no ISA/EISA interrupt 0 entry */
	for (x = 0; x < nintrs; x++) {
		if (io_apic_ints[x].dst_apic_int == 0 &&
		    io_apic_ints[x].dst_apic_id == IO_TO_ID(0) &&
		    io_apic_ints[x].int_vector == 0xff && 
		    int_to_apicintpin[0].ioapic == -1 &&
		    io_apic_ints[x].int_type == 3) {
			assign_apic_irq(0, 0, 0);
			break;
		}
	}
	/* PCI interrupt assignment is deferred */
}

#endif

void
mp_set_cpuids(int cpu_id, int apic_id)
{
	CPU_TO_ID(cpu_id) = apic_id;
	ID_TO_CPU(apic_id) = cpu_id;
}

static int
processor_entry(const struct PROCENTRY *entry, int cpu)
{
	KKASSERT(cpu > 0);

	/* check for usability */
	if (!(entry->cpu_flags & PROCENTRY_FLAG_EN))
		return 0;

	/* check for BSP flag */
	if (entry->cpu_flags & PROCENTRY_FLAG_BP) {
		mp_set_cpuids(0, entry->apic_id);
		return 0;	/* its already been counted */
	}

	/* add another AP to list, if less than max number of CPUs */
	else if (cpu < MAXCPU) {
		mp_set_cpuids(cpu, entry->apic_id);
		return 1;
	}

	return 0;
}

#ifdef APIC_IO

static int
bus_entry(const struct BUSENTRY *entry, int bus)
{
	int     x;
	char    c, name[8];

	/* encode the name into an index */
	for (x = 0; x < 6; ++x) {
		if ((c = entry->bus_type[x]) == ' ')
			break;
		name[x] = c;
	}
	name[x] = '\0';

	if ((x = lookup_bus_type(name)) == UNKNOWN_BUSTYPE)
		panic("unknown bus type: '%s'", name);

	bus_data[bus].bus_id = entry->bus_id;
	bus_data[bus].bus_type = x;

	return 1;
}

static int
io_apic_entry(const struct IOAPICENTRY *entry, int apic)
{
	if (!(entry->apic_flags & IOAPICENTRY_FLAG_EN))
		return 0;

	IO_TO_ID(apic) = entry->apic_id;
	ID_TO_IO(entry->apic_id) = apic;

	return 1;
}

static int
lookup_bus_type(char *name)
{
	int     x;

	for (x = 0; x < MAX_BUSTYPE; ++x)
		if (strcmp(bus_type_table[x].name, name) == 0)
			return bus_type_table[x].type;

	return UNKNOWN_BUSTYPE;
}

static int
int_entry(const struct INTENTRY *entry, int intr)
{
	int apic;

	io_apic_ints[intr].int_type = entry->int_type;
	io_apic_ints[intr].int_flags = entry->int_flags;
	io_apic_ints[intr].src_bus_id = entry->src_bus_id;
	io_apic_ints[intr].src_bus_irq = entry->src_bus_irq;
	if (entry->dst_apic_id == 255) {
		/* This signal goes to all IO APICS.  Select an IO APIC
		   with sufficient number of interrupt pins */
		for (apic = 0; apic < mp_napics; apic++)
			if (((io_apic_read(apic, IOAPIC_VER) & 
			      IOART_VER_MAXREDIR) >> MAXREDIRSHIFT) >= 
			    entry->dst_apic_int)
				break;
		if (apic < mp_napics)
			io_apic_ints[intr].dst_apic_id = IO_TO_ID(apic);
		else
			io_apic_ints[intr].dst_apic_id = entry->dst_apic_id;
	} else
		io_apic_ints[intr].dst_apic_id = entry->dst_apic_id;
	io_apic_ints[intr].dst_apic_int = entry->dst_apic_int;

	return 1;
}

static int
apic_int_is_bus_type(int intr, int bus_type)
{
	int     bus;

	for (bus = 0; bus < mp_nbusses; ++bus)
		if ((bus_data[bus].bus_id == io_apic_ints[intr].src_bus_id)
		    && ((int) bus_data[bus].bus_type == bus_type))
			return 1;

	return 0;
}

/*
 * Given a traditional ISA INT mask, return an APIC mask.
 */
u_int
isa_apic_mask(u_int isa_mask)
{
	int isa_irq;
	int apic_pin;

#if defined(SKIP_IRQ15_REDIRECT)
	if (isa_mask == (1 << 15)) {
		kprintf("skipping ISA IRQ15 redirect\n");
		return isa_mask;
	}
#endif  /* SKIP_IRQ15_REDIRECT */

	isa_irq = ffs(isa_mask);		/* find its bit position */
	if (isa_irq == 0)			/* doesn't exist */
		return 0;
	--isa_irq;				/* make it zero based */

	apic_pin = isa_apic_irq(isa_irq);	/* look for APIC connection */
	if (apic_pin == -1)
		return 0;

	return (1 << apic_pin);			/* convert pin# to a mask */
}

/*
 * Determine which APIC pin an ISA/EISA INT is attached to.
 */
#define INTTYPE(I)	(io_apic_ints[(I)].int_type)
#define INTPIN(I)	(io_apic_ints[(I)].dst_apic_int)
#define INTIRQ(I)	(io_apic_ints[(I)].int_vector)
#define INTAPIC(I)	(ID_TO_IO(io_apic_ints[(I)].dst_apic_id))

#define SRCBUSIRQ(I)	(io_apic_ints[(I)].src_bus_irq)
int
isa_apic_irq(int isa_irq)
{
	int     intr;

	for (intr = 0; intr < nintrs; ++intr) {		/* check each record */
		if (INTTYPE(intr) == 0) {		/* standard INT */
			if (SRCBUSIRQ(intr) == isa_irq) {
				if (apic_int_is_bus_type(intr, ISA) ||
			            apic_int_is_bus_type(intr, EISA)) {
					if (INTIRQ(intr) == 0xff)
						return -1; /* unassigned */
					return INTIRQ(intr);	/* found */
				}
			}
		}
	}
	return -1;					/* NOT found */
}


/*
 * Determine which APIC pin a PCI INT is attached to.
 */
#define SRCBUSID(I)	(io_apic_ints[(I)].src_bus_id)
#define SRCBUSDEVICE(I)	((io_apic_ints[(I)].src_bus_irq >> 2) & 0x1f)
#define SRCBUSLINE(I)	(io_apic_ints[(I)].src_bus_irq & 0x03)
int
pci_apic_irq(int pciBus, int pciDevice, int pciInt)
{
	int     intr;

	--pciInt;					/* zero based */

	for (intr = 0; intr < nintrs; ++intr) {		/* check each record */
		if ((INTTYPE(intr) == 0)		/* standard INT */
		    && (SRCBUSID(intr) == pciBus)
		    && (SRCBUSDEVICE(intr) == pciDevice)
		    && (SRCBUSLINE(intr) == pciInt)) {	/* a candidate IRQ */
			if (apic_int_is_bus_type(intr, PCI)) {
				if (INTIRQ(intr) == 0xff)
					allocate_apic_irq(intr);
				if (INTIRQ(intr) == 0xff)
					return -1;	/* unassigned */
				return INTIRQ(intr);	/* exact match */
			}
		}
	}

	return -1;					/* NOT found */
}

int
next_apic_irq(int irq) 
{
	int intr, ointr;
	int bus, bustype;

	bus = 0;
	bustype = 0;
	for (intr = 0; intr < nintrs; intr++) {
		if (INTIRQ(intr) != irq || INTTYPE(intr) != 0)
			continue;
		bus = SRCBUSID(intr);
		bustype = apic_bus_type(bus);
		if (bustype != ISA &&
		    bustype != EISA &&
		    bustype != PCI)
			continue;
		break;
	}
	if (intr >= nintrs) {
		return -1;
	}
	for (ointr = intr + 1; ointr < nintrs; ointr++) {
		if (INTTYPE(ointr) != 0)
			continue;
		if (bus != SRCBUSID(ointr))
			continue;
		if (bustype == PCI) {
			if (SRCBUSDEVICE(intr) != SRCBUSDEVICE(ointr))
				continue;
			if (SRCBUSLINE(intr) != SRCBUSLINE(ointr))
				continue;
		}
		if (bustype == ISA || bustype == EISA) {
			if (SRCBUSIRQ(intr) != SRCBUSIRQ(ointr))
				continue;
		}
		if (INTPIN(intr) == INTPIN(ointr))
			continue;
		break;
	}
	if (ointr >= nintrs) {
		return -1;
	}
	return INTIRQ(ointr);
}
#undef SRCBUSLINE
#undef SRCBUSDEVICE
#undef SRCBUSID
#undef SRCBUSIRQ

#undef INTPIN
#undef INTIRQ
#undef INTAPIC
#undef INTTYPE

#endif

/*
 * Reprogram the MB chipset to NOT redirect an ISA INTerrupt.
 *
 * XXX FIXME:
 *  Exactly what this means is unclear at this point.  It is a solution
 *  for motherboards that redirect the MBIRQ0 pin.  Generically a motherboard
 *  could route any of the ISA INTs to upper (>15) IRQ values.  But most would
 *  NOT be redirected via MBIRQ0, thus "undirect()ing" them would NOT be an
 *  option.
 */
int
undirect_isa_irq(int rirq)
{
#if defined(READY)
	if (bootverbose)
	    kprintf("Freeing redirected ISA irq %d.\n", rirq);
	/** FIXME: tickle the MB redirector chip */
	return /* XXX */;
#else
	if (bootverbose)
	    kprintf("Freeing (NOT implemented) redirected ISA irq %d.\n", rirq);
	return 0;
#endif  /* READY */
}


/*
 * Reprogram the MB chipset to NOT redirect a PCI INTerrupt
 */
int
undirect_pci_irq(int rirq)
{
#if defined(READY)
	if (bootverbose)
		kprintf("Freeing redirected PCI irq %d.\n", rirq);

	/** FIXME: tickle the MB redirector chip */
	return /* XXX */;
#else
	if (bootverbose)
		kprintf("Freeing (NOT implemented) redirected PCI irq %d.\n",
		       rirq);
	return 0;
#endif  /* READY */
}


#ifdef APIC_IO

/*
 * given a bus ID, return:
 *  the bus type if found
 *  -1 if NOT found
 */
int
apic_bus_type(int id)
{
	int     x;

	for (x = 0; x < mp_nbusses; ++x)
		if (bus_data[x].bus_id == id)
			return bus_data[x].bus_type;

	return -1;
}

/*
 * given a LOGICAL APIC# and pin#, return:
 *  the associated src bus ID if found
 *  -1 if NOT found
 */
int
apic_src_bus_id(int apic, int pin)
{
	int     x;

	/* search each of the possible INTerrupt sources */
	for (x = 0; x < nintrs; ++x)
		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
		    (pin == io_apic_ints[x].dst_apic_int))
			return (io_apic_ints[x].src_bus_id);

	return -1;		/* NOT found */
}

/*
 * given a LOGICAL APIC# and pin#, return:
 *  the associated src bus IRQ if found
 *  -1 if NOT found
 */
int
apic_src_bus_irq(int apic, int pin)
{
	int     x;

	for (x = 0; x < nintrs; x++)
		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
		    (pin == io_apic_ints[x].dst_apic_int))
			return (io_apic_ints[x].src_bus_irq);

	return -1;		/* NOT found */
}


/*
 * given a LOGICAL APIC# and pin#, return:
 *  the associated INTerrupt type if found
 *  -1 if NOT found
 */
int
apic_int_type(int apic, int pin)
{
	int     x;

	/* search each of the possible INTerrupt sources */
	for (x = 0; x < nintrs; ++x) {
		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
		    (pin == io_apic_ints[x].dst_apic_int))
			return (io_apic_ints[x].int_type);
	}
	return -1;		/* NOT found */
}

/*
 * Return the IRQ associated with an APIC pin
 */
int 
apic_irq(int apic, int pin)
{
	int x;
	int res;

	for (x = 0; x < nintrs; ++x) {
		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
		    (pin == io_apic_ints[x].dst_apic_int)) {
			res = io_apic_ints[x].int_vector;
			if (res == 0xff)
				return -1;
			if (apic != int_to_apicintpin[res].ioapic)
				panic("apic_irq: inconsistent table %d/%d", apic, int_to_apicintpin[res].ioapic);
			if (pin != int_to_apicintpin[res].int_pin)
				panic("apic_irq inconsistent table (2)");
			return res;
		}
	}
	return -1;
}


/*
 * given a LOGICAL APIC# and pin#, return:
 *  the associated trigger mode if found
 *  -1 if NOT found
 */
int
apic_trigger(int apic, int pin)
{
	int     x;

	/* search each of the possible INTerrupt sources */
	for (x = 0; x < nintrs; ++x)
		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
		    (pin == io_apic_ints[x].dst_apic_int))
			return ((io_apic_ints[x].int_flags >> 2) & 0x03);

	return -1;		/* NOT found */
}


/*
 * given a LOGICAL APIC# and pin#, return:
 *  the associated 'active' level if found
 *  -1 if NOT found
 */
int
apic_polarity(int apic, int pin)
{
	int     x;

	/* search each of the possible INTerrupt sources */
	for (x = 0; x < nintrs; ++x)
		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
		    (pin == io_apic_ints[x].dst_apic_int))
			return (io_apic_ints[x].int_flags & 0x03);

	return -1;		/* NOT found */
}

/*
 * set data according to MP defaults
 * FIXME: probably not complete yet...
 */
static void
mptable_default(int type)
{
	int     io_apic_id;
	int     pin;

#if 0
	kprintf("  MP default config type: %d\n", type);
	switch (type) {
	case 1:
		kprintf("   bus: ISA, APIC: 82489DX\n");
		break;
	case 2:
		kprintf("   bus: EISA, APIC: 82489DX\n");
		break;
	case 3:
		kprintf("   bus: EISA, APIC: 82489DX\n");
		break;
	case 4:
		kprintf("   bus: MCA, APIC: 82489DX\n");
		break;
	case 5:
		kprintf("   bus: ISA+PCI, APIC: Integrated\n");
		break;
	case 6:
		kprintf("   bus: EISA+PCI, APIC: Integrated\n");
		break;
	case 7:
		kprintf("   bus: MCA+PCI, APIC: Integrated\n");
		break;
	default:
		kprintf("   future type\n");
		break;
		/* NOTREACHED */
	}
#endif	/* 0 */

	/* one and only IO APIC */
	io_apic_id = (io_apic_read(0, IOAPIC_ID) & APIC_ID_MASK) >> 24;

	/*
	 * sanity check, refer to MP spec section 3.6.6, last paragraph
	 * necessary as some hardware isn't properly setting up the IO APIC
	 */
#if defined(REALLY_ANAL_IOAPICID_VALUE)
	if (io_apic_id != 2) {
#else
	if ((io_apic_id == 0) || (io_apic_id == 1) || (io_apic_id == 15)) {
#endif	/* REALLY_ANAL_IOAPICID_VALUE */
		io_apic_set_id(0, 2);
		io_apic_id = 2;
	}
	IO_TO_ID(0) = io_apic_id;
	ID_TO_IO(io_apic_id) = 0;

	/* fill out bus entries */
	switch (type) {
	case 1:
	case 2:
	case 3:
	case 4:
	case 5:
	case 6:
	case 7:
		bus_data[0].bus_id = default_data[type - 1][1];
		bus_data[0].bus_type = default_data[type - 1][2];
		bus_data[1].bus_id = default_data[type - 1][3];
		bus_data[1].bus_type = default_data[type - 1][4];
		break;

	/* case 4: case 7:		   MCA NOT supported */
	default:		/* illegal/reserved */
		panic("BAD default MP config: %d", type);
		/* NOTREACHED */
	}

	/* general cases from MP v1.4, table 5-2 */
	for (pin = 0; pin < 16; ++pin) {
		io_apic_ints[pin].int_type = 0;
		io_apic_ints[pin].int_flags = 0x05;	/* edge/active-hi */
		io_apic_ints[pin].src_bus_id = 0;
		io_apic_ints[pin].src_bus_irq = pin;	/* IRQ2 caught below */
		io_apic_ints[pin].dst_apic_id = io_apic_id;
		io_apic_ints[pin].dst_apic_int = pin;	/* 1-to-1 */
	}

	/* special cases from MP v1.4, table 5-2 */
	if (type == 2) {
		io_apic_ints[2].int_type = 0xff;	/* N/C */
		io_apic_ints[13].int_type = 0xff;	/* N/C */
#if !defined(APIC_MIXED_MODE)
		/** FIXME: ??? */
		panic("sorry, can't support type 2 default yet");
#endif	/* APIC_MIXED_MODE */
	}
	else
		io_apic_ints[2].src_bus_irq = 0;	/* ISA IRQ0 is on APIC INT 2 */

	if (type == 7)
		io_apic_ints[0].int_type = 0xff;	/* N/C */
	else
		io_apic_ints[0].int_type = 3;	/* vectored 8259 */
}

#endif	/* APIC_IO */

/*
 * Map a physical memory address representing I/O into KVA.  The I/O
 * block is assumed not to cross a page boundary.
 */
void *
permanent_io_mapping(vm_paddr_t pa)
{
	vm_offset_t vaddr;
	int pgeflag;
	int i;

	KKASSERT(pa < 0x100000000LL);

	pgeflag = 0;	/* not used for SMP yet */

	/*
	 * If the requested physical address has already been incidently
	 * mapped, just use the existing mapping.  Otherwise create a new
	 * mapping.
	 */
	for (i = IO_MAPPING_START_INDEX; i < SMPpt_alloc_index; ++i) {
		if (((vm_offset_t)SMPpt[i] & PG_FRAME) ==
		    ((vm_offset_t)pa & PG_FRAME)) {
			break;
		}
	}
	if (i == SMPpt_alloc_index) {
		if (i == NPTEPG - 2) {
			panic("permanent_io_mapping: We ran out of space"
			      " in SMPpt[]!");
		}
		SMPpt[i] = (pt_entry_t)(PG_V | PG_RW | PG_N | pgeflag |
			   ((vm_offset_t)pa & PG_FRAME));
		++SMPpt_alloc_index;
	}
	vaddr = (vm_offset_t)CPU_prvspace + (i * PAGE_SIZE) +
		((vm_offset_t)pa & PAGE_MASK);
	return ((void *)vaddr);
}

/*
 * start each AP in our list
 */
static int
start_all_aps(u_int boot_addr)
{
	int     x, i, pg;
	int	shift;
	int	smicount;
	int	smibest;
	int	smilast;
	u_char  mpbiosreason;
	u_long  mpbioswarmvec;
	struct mdglobaldata *gd;
	struct privatespace *ps;
	char *stack;
	uintptr_t kptbase;

	POSTCODE(START_ALL_APS_POST);

	/* Initialize BSP's local APIC */
	apic_initialize(TRUE);

	/* install the AP 1st level boot code */
	install_ap_tramp(boot_addr);


	/* save the current value of the warm-start vector */
	mpbioswarmvec = *((u_long *) WARMBOOT_OFF);
	outb(CMOS_REG, BIOS_RESET);
	mpbiosreason = inb(CMOS_DATA);

	/* setup a vector to our boot code */
	*((volatile u_short *) WARMBOOT_OFF) = WARMBOOT_TARGET;
	*((volatile u_short *) WARMBOOT_SEG) = (boot_addr >> 4);
	outb(CMOS_REG, BIOS_RESET);
	outb(CMOS_DATA, BIOS_WARM);	/* 'warm-start' */

	/*
	 * If we have a TSC we can figure out the SMI interrupt rate.
	 * The SMI does not necessarily use a constant rate.  Spend
	 * up to 250ms trying to figure it out.
	 */
	smibest = 0;
	if (cpu_feature & CPUID_TSC) {
		set_apic_timer(275000);
		smilast = read_apic_timer();
		for (x = 0; x < 20 && read_apic_timer(); ++x) {
			smicount = smitest();
			if (smibest == 0 || smilast - smicount < smibest)
				smibest = smilast - smicount;
			smilast = smicount;
		}
		if (smibest > 250000)
			smibest = 0;
		if (smibest) {
			smibest = smibest * (int64_t)1000000 /
				  get_apic_timer_frequency();
		}
	}
	if (smibest)
		kprintf("SMI Frequency (worst case): %d Hz (%d us)\n",
			1000000 / smibest, smibest);


	/* set up temporary P==V mapping for AP boot */
	/* XXX this is a hack, we should boot the AP on its own stack/PTD */
	kptbase = (uintptr_t)(void *)KPTphys;
	for (x = 0; x < NKPT; x++) {
		PTD[x] = (pd_entry_t)(PG_V | PG_RW |
		    ((kptbase + x * PAGE_SIZE) & PG_FRAME));
	}
	cpu_invltlb();

	/* start each AP */
	for (x = 1; x <= mp_naps; ++x) {

		/* This is a bit verbose, it will go away soon.  */

		/* first page of AP's private space */
		pg = x * i386_btop(sizeof(struct privatespace));

		/* allocate new private data page(s) */
		gd = (struct mdglobaldata *)kmem_alloc(&kernel_map, 
				MDGLOBALDATA_BASEALLOC_SIZE);
		/* wire it into the private page table page */
		for (i = 0; i < MDGLOBALDATA_BASEALLOC_SIZE; i += PAGE_SIZE) {
			SMPpt[pg + i / PAGE_SIZE] = (pt_entry_t)
			    (PG_V | PG_RW | vtophys_pte((char *)gd + i));
		}
		pg += MDGLOBALDATA_BASEALLOC_PAGES;

		SMPpt[pg + 0] = 0;		/* *gd_CMAP1 */
		SMPpt[pg + 1] = 0;		/* *gd_CMAP2 */
		SMPpt[pg + 2] = 0;		/* *gd_CMAP3 */
		SMPpt[pg + 3] = 0;		/* *gd_PMAP1 */

		/* allocate and set up an idle stack data page */
		stack = (char *)kmem_alloc(&kernel_map, UPAGES*PAGE_SIZE);
		for (i = 0; i < UPAGES; i++) {
			SMPpt[pg + 4 + i] = (pt_entry_t)
			    (PG_V | PG_RW | vtophys_pte(PAGE_SIZE * i + stack));
		}

		gd = &CPU_prvspace[x].mdglobaldata;	/* official location */
		bzero(gd, sizeof(*gd));
		gd->mi.gd_prvspace = ps = &CPU_prvspace[x];

		/* prime data page for it to use */
		mi_gdinit(&gd->mi, x);
		cpu_gdinit(gd, x);
		gd->gd_CMAP1 = &SMPpt[pg + 0];
		gd->gd_CMAP2 = &SMPpt[pg + 1];
		gd->gd_CMAP3 = &SMPpt[pg + 2];
		gd->gd_PMAP1 = &SMPpt[pg + 3];
		gd->gd_CADDR1 = ps->CPAGE1;
		gd->gd_CADDR2 = ps->CPAGE2;
		gd->gd_CADDR3 = ps->CPAGE3;
		gd->gd_PADDR1 = (unsigned *)ps->PPAGE1;

		/*
		 * Per-cpu pmap for get_ptbase().
		 */
		gd->gd_GDADDR1= (unsigned *)
			kmem_alloc_nofault(&kernel_map, SEG_SIZE, SEG_SIZE);
		gd->gd_GDMAP1 = &PTD[(vm_offset_t)gd->gd_GDADDR1 >> PDRSHIFT];

		gd->mi.gd_ipiq = (void *)kmem_alloc(&kernel_map, sizeof(lwkt_ipiq) * (mp_naps + 1));
		bzero(gd->mi.gd_ipiq, sizeof(lwkt_ipiq) * (mp_naps + 1));

		/*
		 * Setup the AP boot stack
		 */
		bootSTK = &ps->idlestack[UPAGES*PAGE_SIZE/2];
		bootAP = x;

		/* attempt to start the Application Processor */
		CHECK_INIT(99);	/* setup checkpoints */
		if (!start_ap(gd, boot_addr, smibest)) {
			kprintf("AP #%d (PHY# %d) failed!\n", x, CPU_TO_ID(x));
			CHECK_PRINT("trace");	/* show checkpoints */
			/* better panic as the AP may be running loose */
			kprintf("panic y/n? [y] ");
			if (cngetc() != 'n')
				panic("bye-bye");
		}
		CHECK_PRINT("trace");		/* show checkpoints */

		/* record its version info */
		cpu_apic_versions[x] = cpu_apic_versions[0];
	}

	/* set ncpus to 1 + highest logical cpu.  Not all may have come up */
	ncpus = x;

	/* ncpus2 -- ncpus rounded down to the nearest power of 2 */
	for (shift = 0; (1 << shift) <= ncpus; ++shift)
		;
	--shift;
	ncpus2_shift = shift;
	ncpus2 = 1 << shift;
	ncpus2_mask = ncpus2 - 1;

	/* ncpus_fit -- ncpus rounded up to the nearest power of 2 */
	if ((1 << shift) < ncpus)
		++shift;
	ncpus_fit = 1 << shift;
	ncpus_fit_mask = ncpus_fit - 1;

	/* build our map of 'other' CPUs */
	mycpu->gd_other_cpus = smp_startup_mask & ~(1 << mycpu->gd_cpuid);
	mycpu->gd_ipiq = (void *)kmem_alloc(&kernel_map, sizeof(lwkt_ipiq) * ncpus);
	bzero(mycpu->gd_ipiq, sizeof(lwkt_ipiq) * ncpus);

	/* fill in our (BSP) APIC version */
	cpu_apic_versions[0] = lapic.version;

	/* restore the warmstart vector */
	*(u_long *) WARMBOOT_OFF = mpbioswarmvec;
	outb(CMOS_REG, BIOS_RESET);
	outb(CMOS_DATA, mpbiosreason);

	/*
	 * NOTE!  The idlestack for the BSP was setup by locore.  Finish
	 * up, clean out the P==V mapping we did earlier.
	 */
	for (x = 0; x < NKPT; x++)
		PTD[x] = 0;
	pmap_set_opt();

	/* number of APs actually started */
	return ncpus - 1;
}

/*
 * load the 1st level AP boot code into base memory.
 */

/* targets for relocation */
extern void bigJump(void);
extern void bootCodeSeg(void);
extern void bootDataSeg(void);
extern void MPentry(void);
extern u_int MP_GDT;
extern u_int mp_gdtbase;

static void
install_ap_tramp(u_int boot_addr)
{
	int     x;
	int     size = *(int *) ((u_long) & bootMP_size);
	u_char *src = (u_char *) ((u_long) bootMP);
	u_char *dst = (u_char *) boot_addr + KERNBASE;
	u_int   boot_base = (u_int) bootMP;
	u_int8_t *dst8;
	u_int16_t *dst16;
	u_int32_t *dst32;

	POSTCODE(INSTALL_AP_TRAMP_POST);

	for (x = 0; x < size; ++x)
		*dst++ = *src++;

	/*
	 * modify addresses in code we just moved to basemem. unfortunately we
	 * need fairly detailed info about mpboot.s for this to work.  changes
	 * to mpboot.s might require changes here.
	 */

	/* boot code is located in KERNEL space */
	dst = (u_char *) boot_addr + KERNBASE;

	/* modify the lgdt arg */
	dst32 = (u_int32_t *) (dst + ((u_int) & mp_gdtbase - boot_base));
	*dst32 = boot_addr + ((u_int) & MP_GDT - boot_base);

	/* modify the ljmp target for MPentry() */
	dst32 = (u_int32_t *) (dst + ((u_int) bigJump - boot_base) + 1);
	*dst32 = ((u_int) MPentry - KERNBASE);

	/* modify the target for boot code segment */
	dst16 = (u_int16_t *) (dst + ((u_int) bootCodeSeg - boot_base));
	dst8 = (u_int8_t *) (dst16 + 1);
	*dst16 = (u_int) boot_addr & 0xffff;
	*dst8 = ((u_int) boot_addr >> 16) & 0xff;

	/* modify the target for boot data segment */
	dst16 = (u_int16_t *) (dst + ((u_int) bootDataSeg - boot_base));
	dst8 = (u_int8_t *) (dst16 + 1);
	*dst16 = (u_int) boot_addr & 0xffff;
	*dst8 = ((u_int) boot_addr >> 16) & 0xff;
}


/*
 * This function starts the AP (application processor) identified
 * by the APIC ID 'physicalCpu'.  It does quite a "song and dance"
 * to accomplish this.  This is necessary because of the nuances
 * of the different hardware we might encounter.  It ain't pretty,
 * but it seems to work.
 *
 * NOTE: eventually an AP gets to ap_init(), which is called just 
 * before the AP goes into the LWKT scheduler's idle loop.
 */
static int
start_ap(struct mdglobaldata *gd, u_int boot_addr, int smibest)
{
	int     physical_cpu;
	int     vector;
	u_long  icr_lo, icr_hi;

	POSTCODE(START_AP_POST);

	/* get the PHYSICAL APIC ID# */
	physical_cpu = CPU_TO_ID(gd->mi.gd_cpuid);

	/* calculate the vector */
	vector = (boot_addr >> 12) & 0xff;

	/* We don't want anything interfering */
	cpu_disable_intr();

	/* Make sure the target cpu sees everything */
	wbinvd();

	/*
	 * Try to detect when a SMI has occurred, wait up to 200ms.
	 *
	 * If a SMI occurs during an AP reset but before we issue
	 * the STARTUP command, the AP may brick.  To work around
	 * this problem we hold off doing the AP startup until
	 * after we have detected the SMI.  Hopefully another SMI
	 * will not occur before we finish the AP startup.
	 *
	 * Retries don't seem to help.  SMIs have a window of opportunity
	 * and if USB->legacy keyboard emulation is enabled in the BIOS
	 * the interrupt rate can be quite high.
	 *
	 * NOTE: Don't worry about the L1 cache load, it might bloat
	 *       ldelta a little but ndelta will be so huge when the SMI
	 *       occurs the detection logic will still work fine.
	 */
	if (smibest) {
		set_apic_timer(200000);
		smitest();
	}

	/*
	 * first we do an INIT/RESET IPI this INIT IPI might be run, reseting
	 * and running the target CPU. OR this INIT IPI might be latched (P5
	 * bug), CPU waiting for STARTUP IPI. OR this INIT IPI might be
	 * ignored.
	 *
	 * see apic/apicreg.h for icr bit definitions.
	 *
	 * TIME CRITICAL CODE, DO NOT DO ANY KPRINTFS IN THE HOT PATH.
	 */

	/*
	 * Setup the address for the target AP.  We can setup
	 * icr_hi once and then just trigger operations with
	 * icr_lo.
	 */
	icr_hi = lapic.icr_hi & ~APIC_ID_MASK;
	icr_hi |= (physical_cpu << 24);
	icr_lo = lapic.icr_lo & 0xfff00000;
	lapic.icr_hi = icr_hi;

	/*
	 * Do an INIT IPI: assert RESET
	 *
	 * Use edge triggered mode to assert INIT
	 */
	lapic.icr_lo = icr_lo | 0x0000c500;
	while (lapic.icr_lo & APIC_DELSTAT_MASK)
		 /* spin */ ;

	/*
	 * The spec calls for a 10ms delay but we may have to use a
	 * MUCH lower delay to avoid bricking an AP due to a fast SMI
	 * interrupt.  We have other loops here too and dividing by 2
	 * doesn't seem to be enough even after subtracting 350us,
	 * so we divide by 4.
	 *
	 * Our minimum delay is 150uS, maximum is 10ms.  If no SMI
	 * interrupt was detected we use the full 10ms.
	 */
	if (smibest == 0)
		u_sleep(10000);
	else if (smibest < 150 * 4 + 350)
		u_sleep(150);
	else if ((smibest - 350) / 4 < 10000)
		u_sleep((smibest - 350) / 4);
	else
		u_sleep(10000);

	/*
	 * Do an INIT IPI: deassert RESET
	 *
	 * Use level triggered mode to deassert.  It is unclear
	 * why we need to do this.
	 */
	lapic.icr_lo = icr_lo | 0x00008500;
	while (lapic.icr_lo & APIC_DELSTAT_MASK)
		 /* spin */ ;
	u_sleep(150);				/* wait 150us */

	/*
	 * Next we do a STARTUP IPI: the previous INIT IPI might still be
	 * latched, (P5 bug) this 1st STARTUP would then terminate
	 * immediately, and the previously started INIT IPI would continue. OR
	 * the previous INIT IPI has already run. and this STARTUP IPI will
	 * run. OR the previous INIT IPI was ignored. and this STARTUP IPI
	 * will run.
	 */
	lapic.icr_lo = icr_lo | 0x00000600 | vector;
	while (lapic.icr_lo & APIC_DELSTAT_MASK)
		 /* spin */ ;
	u_sleep(200);		/* wait ~200uS */

	/*
	 * Finally we do a 2nd STARTUP IPI: this 2nd STARTUP IPI should run IF
	 * the previous STARTUP IPI was cancelled by a latched INIT IPI. OR
	 * this STARTUP IPI will be ignored, as only ONE STARTUP IPI is
	 * recognized after hardware RESET or INIT IPI.
	 */
	lapic.icr_lo = icr_lo | 0x00000600 | vector;
	while (lapic.icr_lo & APIC_DELSTAT_MASK)
		 /* spin */ ;

	/* Resume normal operation */
	cpu_enable_intr();

	/* wait for it to start, see ap_init() */
	set_apic_timer(5000000);/* == 5 seconds */
	while (read_apic_timer()) {
		if (smp_startup_mask & (1 << gd->mi.gd_cpuid))
			return 1;	/* return SUCCESS */
	}

	return 0;		/* return FAILURE */
}

static
int
smitest(void)
{
	int64_t	ltsc;
	int64_t	ntsc;
	int64_t	ldelta;
	int64_t	ndelta;
	int count;

	ldelta = 0;
	ndelta = 0;
	while (read_apic_timer()) {
		ltsc = rdtsc();
		for (count = 0; count < 100; ++count)
			ntsc = rdtsc();	/* force loop to occur */
		if (ldelta) {
			ndelta = ntsc - ltsc;
			if (ldelta > ndelta)
				ldelta = ndelta;
			if (ndelta > ldelta * 2)
				break;
		} else {
			ldelta = ntsc - ltsc;
		}
	}
	return(read_apic_timer());
}

/*
 * Lazy flush the TLB on all other CPU's.  DEPRECATED.
 *
 * If for some reason we were unable to start all cpus we cannot safely
 * use broadcast IPIs.
 */

static cpumask_t smp_invltlb_req;
#define SMP_INVLTLB_DEBUG

void
smp_invltlb(void)
{
#ifdef SMP
	struct mdglobaldata *md = mdcpu;
#ifdef SMP_INVLTLB_DEBUG
	long count = 0;
	long xcount = 0;
#endif

	crit_enter_gd(&md->mi);
	md->gd_invltlb_ret = 0;
	++md->mi.gd_cnt.v_smpinvltlb;
	atomic_set_int(&smp_invltlb_req, md->mi.gd_cpumask);
#ifdef SMP_INVLTLB_DEBUG
again:
#endif
	if (smp_startup_mask == smp_active_mask) {
		all_but_self_ipi(XINVLTLB_OFFSET);
	} else {
		selected_apic_ipi(smp_active_mask & ~md->mi.gd_cpumask,
				  XINVLTLB_OFFSET, APIC_DELMODE_FIXED);
	}

#ifdef SMP_INVLTLB_DEBUG
	if (xcount)
		kprintf("smp_invltlb: ipi sent\n");
#endif
	while ((md->gd_invltlb_ret & smp_active_mask & ~md->mi.gd_cpumask) !=
	       (smp_active_mask & ~md->mi.gd_cpumask)) {
		cpu_mfence();
		cpu_pause();
#ifdef SMP_INVLTLB_DEBUG
		/* DEBUGGING */
		if (++count == 400000000) {
			print_backtrace(-1);
			kprintf("smp_invltlb: endless loop %08lx %08lx, "
				"rflags %016lx retry",
				(long)md->gd_invltlb_ret,
				(long)smp_invltlb_req,
				(long)read_eflags());
			__asm __volatile ("sti");
			++xcount;
			if (xcount > 2)
				lwkt_process_ipiq();
			if (xcount > 3) {
				int bcpu = bsfl(~md->gd_invltlb_ret & ~md->mi.gd_cpumask & smp_active_mask);
				globaldata_t xgd;
				kprintf("bcpu %d\n", bcpu);
				xgd = globaldata_find(bcpu);
				kprintf("thread %p %s\n", xgd->gd_curthread, xgd->gd_curthread->td_comm);
			}
			if (xcount > 5)
				panic("giving up");
			count = 0;
			goto again;
		}
#endif
	}
	atomic_clear_int(&smp_invltlb_req, md->mi.gd_cpumask);
	crit_exit_gd(&md->mi);
#endif
}

#ifdef SMP

/*
 * Called from Xinvltlb assembly with interrupts disabled.  We didn't
 * bother to bump the critical section count or nested interrupt count
 * so only do very low level operations here.
 */
void
smp_invltlb_intr(void)
{
	struct mdglobaldata *md = mdcpu;
	struct mdglobaldata *omd;
	cpumask_t mask;
	int cpu;

	mask = smp_invltlb_req;
	cpu_mfence();
	cpu_invltlb();
	while (mask) {
		cpu = bsfl(mask);
		mask &= ~(1 << cpu);
		omd = (struct mdglobaldata *)globaldata_find(cpu);
		atomic_set_int(&omd->gd_invltlb_ret, md->mi.gd_cpumask);
	}
}

#endif

/*
 * When called the executing CPU will send an IPI to all other CPUs
 *  requesting that they halt execution.
 *
 * Usually (but not necessarily) called with 'other_cpus' as its arg.
 *
 *  - Signals all CPUs in map to stop.
 *  - Waits for each to stop.
 *
 * Returns:
 *  -1: error
 *   0: NA
 *   1: ok
 *
 * XXX FIXME: this is not MP-safe, needs a lock to prevent multiple CPUs
 *            from executing at same time.
 */
int
stop_cpus(u_int map)
{
	map &= smp_active_mask;

	/* send the Xcpustop IPI to all CPUs in map */
	selected_apic_ipi(map, XCPUSTOP_OFFSET, APIC_DELMODE_FIXED);
	
	while ((stopped_cpus & map) != map)
		/* spin */ ;

	return 1;
}


/*
 * Called by a CPU to restart stopped CPUs. 
 *
 * Usually (but not necessarily) called with 'stopped_cpus' as its arg.
 *
 *  - Signals all CPUs in map to restart.
 *  - Waits for each to restart.
 *
 * Returns:
 *  -1: error
 *   0: NA
 *   1: ok
 */
int
restart_cpus(u_int map)
{
	/* signal other cpus to restart */
	started_cpus = map & smp_active_mask;

	while ((stopped_cpus & map) != 0) /* wait for each to clear its bit */
		/* spin */ ;

	return 1;
}

/*
 * This is called once the mpboot code has gotten us properly relocated
 * and the MMU turned on, etc.   ap_init() is actually the idle thread,
 * and when it returns the scheduler will call the real cpu_idle() main
 * loop for the idlethread.  Interrupts are disabled on entry and should
 * remain disabled at return.
 */
void
ap_init(void)
{
	u_int	apic_id;

	/*
	 * Adjust smp_startup_mask to signal the BSP that we have started
	 * up successfully.  Note that we do not yet hold the BGL.  The BSP
	 * is waiting for our signal.
	 *
	 * We can't set our bit in smp_active_mask yet because we are holding
	 * interrupts physically disabled and remote cpus could deadlock
	 * trying to send us an IPI.
	 */
	smp_startup_mask |= 1 << mycpu->gd_cpuid;
	cpu_mfence();

	/*
	 * Interlock for finalization.  Wait until mp_finish is non-zero,
	 * then get the MP lock.
	 *
	 * Note: We are in a critical section.
	 *
	 * Note: We have to synchronize td_mpcount to our desired MP state
	 * before calling cpu_try_mplock().
	 *
	 * Note: we are the idle thread, we can only spin.
	 *
	 * Note: The load fence is memory volatile and prevents the compiler
	 * from improperly caching mp_finish, and the cpu from improperly
	 * caching it.
	 */
	while (mp_finish == 0)
	    cpu_lfence();
	++curthread->td_mpcount;
	while (cpu_try_mplock() == 0)
	    ;

	if (cpu_feature & CPUID_TSC) {
	    /*
	     * The BSP is constantly updating tsc0_offset, figure out the
	     * relative difference to synchronize ktrdump.
	     */
	    tsc_offsets[mycpu->gd_cpuid] = rdtsc() - tsc0_offset;
	}

	/* BSP may have changed PTD while we're waiting for the lock */
	cpu_invltlb();

#if defined(I586_CPU) && !defined(NO_F00F_HACK)
	lidt(&r_idt);
#endif

	/* Build our map of 'other' CPUs. */
	mycpu->gd_other_cpus = smp_startup_mask & ~(1 << mycpu->gd_cpuid);

	kprintf("SMP: AP CPU #%d Launched!\n", mycpu->gd_cpuid);

	/* A quick check from sanity claus */
	apic_id = (apic_id_to_logical[(lapic.id & 0x0f000000) >> 24]);
	if (mycpu->gd_cpuid != apic_id) {
		kprintf("SMP: cpuid = %d\n", mycpu->gd_cpuid);
		kprintf("SMP: apic_id = %d\n", apic_id);
		kprintf("PTD[MPPTDI] = %p\n", (void *)PTD[MPPTDI]);
		panic("cpuid mismatch! boom!!");
	}

	/* Initialize AP's local APIC for irq's */
	apic_initialize(FALSE);

	/* Set memory range attributes for this CPU to match the BSP */
	mem_range_AP_init();

	/*
	 * Once we go active we must process any IPIQ messages that may
	 * have been queued, because no actual IPI will occur until we
	 * set our bit in the smp_active_mask.  If we don't the IPI
	 * message interlock could be left set which would also prevent
	 * further IPIs.
	 *
	 * The idle loop doesn't expect the BGL to be held and while
	 * lwkt_switch() normally cleans things up this is a special case
	 * because we returning almost directly into the idle loop.
	 *
	 * The idle thread is never placed on the runq, make sure
	 * nothing we've done put it there.
	 */
	KKASSERT(curthread->td_mpcount == 1);
	smp_active_mask |= 1 << mycpu->gd_cpuid;

	/*
	 * Enable interrupts here.  idle_restore will also do it, but
	 * doing it here lets us clean up any strays that got posted to
	 * the CPU during the AP boot while we are still in a critical
	 * section.
	 */
	__asm __volatile("sti; pause; pause"::);
	mdcpu->gd_fpending = 0;

	initclocks_pcpu();	/* clock interrupts (via IPIs) */
	lwkt_process_ipiq();

	/*
	 * Releasing the mp lock lets the BSP finish up the SMP init
	 */
	rel_mplock();
	KKASSERT((curthread->td_flags & TDF_RUNQ) == 0);
}

/*
 * Get SMP fully working before we start initializing devices.
 */
static
void
ap_finish(void)
{
	mp_finish = 1;
	if (bootverbose)
		kprintf("Finish MP startup\n");
	if (cpu_feature & CPUID_TSC)
		tsc0_offset = rdtsc();
	tsc_offsets[0] = 0;
	rel_mplock();
	while (smp_active_mask != smp_startup_mask) {
		cpu_lfence();
		if (cpu_feature & CPUID_TSC)
			tsc0_offset = rdtsc();
	}
	while (try_mplock() == 0)
		;
	if (bootverbose)
		kprintf("Active CPU Mask: %08x\n", smp_active_mask);
}

SYSINIT(finishsmp, SI_BOOT2_FINISH_SMP, SI_ORDER_FIRST, ap_finish, NULL)

void
cpu_send_ipiq(int dcpu)
{
        if ((1 << dcpu) & smp_active_mask)
                single_apic_ipi(dcpu, XIPIQ_OFFSET, APIC_DELMODE_FIXED);
}

#if 0	/* single_apic_ipi_passive() not working yet */
/*
 * Returns 0 on failure, 1 on success
 */
int
cpu_send_ipiq_passive(int dcpu)
{
        int r = 0;
        if ((1 << dcpu) & smp_active_mask) {
                r = single_apic_ipi_passive(dcpu, XIPIQ_OFFSET,
                                        APIC_DELMODE_FIXED);
        }
	return(r);
}
#endif

struct mptable_lapic_cbarg1 {
	int	cpu_count;
	int	ht_fixup;
	u_int	ht_apicid_mask;
};

static int
mptable_lapic_pass1_callback(void *xarg, const void *pos, int type)
{
	const struct PROCENTRY *ent;
	struct mptable_lapic_cbarg1 *arg = xarg;

	if (type != 0)
		return 0;
	ent = pos;

	if ((ent->cpu_flags & PROCENTRY_FLAG_EN) == 0)
		return 0;

	arg->cpu_count++;
	if (ent->apic_id < 32) {
		arg->ht_apicid_mask |= 1 << ent->apic_id;
	} else if (arg->ht_fixup) {
		kprintf("MPTABLE: lapic id > 32, disable HTT fixup\n");
		arg->ht_fixup = 0;
	}
	return 0;
}

struct mptable_lapic_cbarg2 {
	int	cpu;
	int	logical_cpus;
	int	found_bsp;
};

static int
mptable_lapic_pass2_callback(void *xarg, const void *pos, int type)
{
	const struct PROCENTRY *ent;
	struct mptable_lapic_cbarg2 *arg = xarg;

	if (type != 0)
		return 0;
	ent = pos;

	if (ent->cpu_flags & PROCENTRY_FLAG_BP) {
		KKASSERT(!arg->found_bsp);
		arg->found_bsp = 1;
	}

	if (processor_entry(ent, arg->cpu))
		arg->cpu++;

	if (arg->logical_cpus) {
		struct PROCENTRY proc;
		int i;

		/*
		 * Create fake mptable processor entries
		 * and feed them to processor_entry() to
		 * enumerate the logical CPUs.
		 */
		bzero(&proc, sizeof(proc));
		proc.type = 0;
		proc.cpu_flags = PROCENTRY_FLAG_EN;
		proc.apic_id = ent->apic_id;

		for (i = 1; i < arg->logical_cpus; i++) {
			proc.apic_id++;
			processor_entry(&proc, arg->cpu);
			arg->cpu++;
		}
	}
	return 0;
}

static void
mptable_imcr(struct mptable_pos *mpt)
{
	/* record whether PIC or virtual-wire mode */
	machintr_setvar_simple(MACHINTR_VAR_IMCR_PRESENT,
			       mpt->mp_fps->mpfb2 & 0x80);
}

struct mptable_lapic_enumerator {
	struct lapic_enumerator	enumerator;
	vm_paddr_t		mpfps_paddr;
};

static void
mptable_lapic_default(void)
{
	int ap_apicid, bsp_apicid;

	mp_naps = 1; /* exclude BSP */

	/* Map local apic before the id field is accessed */
	lapic_map(DEFAULT_APIC_BASE);

	bsp_apicid = APIC_ID(lapic.id);
	ap_apicid = (bsp_apicid == 0) ? 1 : 0;

	/* BSP */
	mp_set_cpuids(0, bsp_apicid);
	/* one and only AP */
	mp_set_cpuids(1, ap_apicid);
}

/*
 * Configure:
 *     mp_naps
 *     ID_TO_CPU(N), APIC ID to logical CPU table
 *     CPU_TO_ID(N), logical CPU to APIC ID table
 */
static void
mptable_lapic_enumerate(struct lapic_enumerator *e)
{
	struct mptable_pos mpt;
	struct mptable_lapic_cbarg1 arg1;
	struct mptable_lapic_cbarg2 arg2;
	mpcth_t	cth;
	int error, logical_cpus = 0;
	vm_offset_t lapic_addr;
	vm_paddr_t mpfps_paddr;

	mpfps_paddr = ((struct mptable_lapic_enumerator *)e)->mpfps_paddr;
	KKASSERT(mpfps_paddr != 0);

	error = mptable_map(&mpt, mpfps_paddr);
	if (error)
		panic("mptable_lapic_enumerate mptable_map failed\n");

	KKASSERT(mpt.mp_fps != NULL);

	/*
	 * Check for use of 'default' configuration
	 */
	if (mpt.mp_fps->mpfb1 != 0) {
		mptable_lapic_default();
		mptable_unmap(&mpt);
		return;
	}

	cth = mpt.mp_cth;
	KKASSERT(cth != NULL);

	/* Save local apic address */
	lapic_addr = (vm_offset_t)cth->apic_address;
	KKASSERT(lapic_addr != 0);

	/*
	 * Find out how many CPUs do we have
	 */
	bzero(&arg1, sizeof(arg1));
	arg1.ht_fixup = 1; /* Apply ht fixup by default */

	error = mptable_iterate_entries(cth,
		    mptable_lapic_pass1_callback, &arg1);
	if (error)
		panic("mptable_iterate_entries(lapic_pass1) failed\n");
	KKASSERT(arg1.cpu_count != 0);

	/* See if we need to fixup HT logical CPUs. */
	if (arg1.ht_fixup) {
		logical_cpus = mptable_hyperthread_fixup(arg1.ht_apicid_mask,
							 arg1.cpu_count);
		if (logical_cpus != 0)
			arg1.cpu_count *= logical_cpus;
	}
	mp_naps = arg1.cpu_count;

	/* Qualify the numbers again, after possible HT fixup */
	if (mp_naps > MAXCPU) {
		kprintf("Warning: only using %d of %d available CPUs!\n",
			MAXCPU, mp_naps);
		mp_naps = MAXCPU;
	}

	--mp_naps;	/* subtract the BSP */

	/*
	 * Link logical CPU id to local apic id
	 */
	bzero(&arg2, sizeof(arg2));
	arg2.cpu = 1;
	arg2.logical_cpus = logical_cpus;

	error = mptable_iterate_entries(cth,
		    mptable_lapic_pass2_callback, &arg2);
	if (error)
		panic("mptable_iterate_entries(lapic_pass2) failed\n");
	KKASSERT(arg2.found_bsp);

	/* Map local apic */
	lapic_map(lapic_addr);

	mptable_unmap(&mpt);
}

static int
mptable_lapic_probe(struct lapic_enumerator *e)
{
	vm_paddr_t mpfps_paddr;

	mpfps_paddr = mptable_probe();
	if (mpfps_paddr == 0)
		return ENXIO;

	((struct mptable_lapic_enumerator *)e)->mpfps_paddr = mpfps_paddr;
	return 0;
}

static struct mptable_lapic_enumerator	mptable_lapic_enumerator = {
	.enumerator = {
		.lapic_prio = LAPIC_ENUM_PRIO_MPTABLE,
		.lapic_probe = mptable_lapic_probe,
		.lapic_enumerate = mptable_lapic_enumerate
	}
};

static void
mptable_apic_register(void)
{
	lapic_enumerator_register(&mptable_lapic_enumerator.enumerator);
}
SYSINIT(madt, SI_BOOT2_PRESMP, SI_ORDER_ANY, mptable_apic_register, 0);