xref: /freebsd/sys/kern/subr_smp.c (revision 77a0943ded95b9e6438f7db70c4a28e4d93946d4)
1 /*
2  * Copyright (c) 1996, by Steve Passe
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. The name of the developer may NOT be used to endorse or promote products
11  *    derived from this software without specific prior written permission.
12  *
13  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
14  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
17  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
23  * SUCH DAMAGE.
24  *
25  * $FreeBSD$
26  */
27 
28 #include "opt_cpu.h"
29 #include "opt_user_ldt.h"
30 
31 #ifdef SMP
32 #include <machine/smptests.h>
33 #else
34 #error
35 #endif
36 
37 #include <sys/param.h>
38 #include <sys/bus.h>
39 #include <sys/systm.h>
40 #include <sys/kernel.h>
41 #include <sys/proc.h>
42 #include <sys/sysctl.h>
43 #include <sys/malloc.h>
44 #include <sys/memrange.h>
45 #include <sys/mutex.h>
46 #ifdef BETTER_CLOCK
47 #include <sys/dkstat.h>
48 #endif
49 #include <sys/cons.h>	/* cngetc() */
50 
51 #include <vm/vm.h>
52 #include <vm/vm_param.h>
53 #include <vm/pmap.h>
54 #include <vm/vm_kern.h>
55 #include <vm/vm_extern.h>
56 #ifdef BETTER_CLOCK
57 #include <sys/lock.h>
58 #include <vm/vm_map.h>
59 #include <sys/user.h>
60 #ifdef GPROF
61 #include <sys/gmon.h>
62 #endif
63 #endif
64 
65 #include <machine/smp.h>
66 #include <machine/apic.h>
67 #include <machine/atomic.h>
68 #include <machine/cpufunc.h>
69 #include <machine/mpapic.h>
70 #include <machine/psl.h>
71 #include <machine/segments.h>
72 #include <machine/smptests.h>	/** TEST_DEFAULT_CONFIG, TEST_TEST1 */
73 #include <machine/tss.h>
74 #include <machine/specialreg.h>
75 #include <machine/globaldata.h>
76 
77 #if defined(APIC_IO)
78 #include <machine/md_var.h>		/* setidt() */
79 #include <i386/isa/icu.h>		/* IPIs */
80 #include <i386/isa/intr_machdep.h>	/* IPIs */
81 #endif	/* APIC_IO */
82 
83 #if defined(TEST_DEFAULT_CONFIG)
84 #define MPFPS_MPFB1	TEST_DEFAULT_CONFIG
85 #else
86 #define MPFPS_MPFB1	mpfps->mpfb1
87 #endif  /* TEST_DEFAULT_CONFIG */
88 
89 #define WARMBOOT_TARGET		0
90 #define WARMBOOT_OFF		(KERNBASE + 0x0467)
91 #define WARMBOOT_SEG		(KERNBASE + 0x0469)
92 
93 #ifdef PC98
94 #define BIOS_BASE		(0xe8000)
95 #define BIOS_SIZE		(0x18000)
96 #else
97 #define BIOS_BASE		(0xf0000)
98 #define BIOS_SIZE		(0x10000)
99 #endif
100 #define BIOS_COUNT		(BIOS_SIZE/4)
101 
102 #define CMOS_REG		(0x70)
103 #define CMOS_DATA		(0x71)
104 #define BIOS_RESET		(0x0f)
105 #define BIOS_WARM		(0x0a)
106 
107 #define PROCENTRY_FLAG_EN	0x01
108 #define PROCENTRY_FLAG_BP	0x02
109 #define IOAPICENTRY_FLAG_EN	0x01
110 
111 
112 /* MP Floating Pointer Structure */
113 typedef struct MPFPS {
114 	char    signature[4];
115 	void   *pap;
116 	u_char  length;
117 	u_char  spec_rev;
118 	u_char  checksum;
119 	u_char  mpfb1;
120 	u_char  mpfb2;
121 	u_char  mpfb3;
122 	u_char  mpfb4;
123 	u_char  mpfb5;
124 }      *mpfps_t;
125 
126 /* MP Configuration Table Header */
127 typedef struct MPCTH {
128 	char    signature[4];
129 	u_short base_table_length;
130 	u_char  spec_rev;
131 	u_char  checksum;
132 	u_char  oem_id[8];
133 	u_char  product_id[12];
134 	void   *oem_table_pointer;
135 	u_short oem_table_size;
136 	u_short entry_count;
137 	void   *apic_address;
138 	u_short extended_table_length;
139 	u_char  extended_table_checksum;
140 	u_char  reserved;
141 }      *mpcth_t;
142 
143 
144 typedef struct PROCENTRY {
145 	u_char  type;
146 	u_char  apic_id;
147 	u_char  apic_version;
148 	u_char  cpu_flags;
149 	u_long  cpu_signature;
150 	u_long  feature_flags;
151 	u_long  reserved1;
152 	u_long  reserved2;
153 }      *proc_entry_ptr;
154 
155 typedef struct BUSENTRY {
156 	u_char  type;
157 	u_char  bus_id;
158 	char    bus_type[6];
159 }      *bus_entry_ptr;
160 
161 typedef struct IOAPICENTRY {
162 	u_char  type;
163 	u_char  apic_id;
164 	u_char  apic_version;
165 	u_char  apic_flags;
166 	void   *apic_address;
167 }      *io_apic_entry_ptr;
168 
169 typedef struct INTENTRY {
170 	u_char  type;
171 	u_char  int_type;
172 	u_short int_flags;
173 	u_char  src_bus_id;
174 	u_char  src_bus_irq;
175 	u_char  dst_apic_id;
176 	u_char  dst_apic_int;
177 }      *int_entry_ptr;
178 
179 /* descriptions of MP basetable entries */
180 typedef struct BASETABLE_ENTRY {
181 	u_char  type;
182 	u_char  length;
183 	char    name[16];
184 }       basetable_entry;
185 
186 /*
187  * this code MUST be enabled here and in mpboot.s.
188  * it follows the very early stages of AP boot by placing values in CMOS ram.
189  * it NORMALLY will never be needed and thus the primitive method for enabling.
190  *
191 #define CHECK_POINTS
192  */
193 
194 #if defined(CHECK_POINTS) && !defined(PC98)
195 #define CHECK_READ(A)	 (outb(CMOS_REG, (A)), inb(CMOS_DATA))
196 #define CHECK_WRITE(A,D) (outb(CMOS_REG, (A)), outb(CMOS_DATA, (D)))
197 
198 #define CHECK_INIT(D);				\
199 	CHECK_WRITE(0x34, (D));			\
200 	CHECK_WRITE(0x35, (D));			\
201 	CHECK_WRITE(0x36, (D));			\
202 	CHECK_WRITE(0x37, (D));			\
203 	CHECK_WRITE(0x38, (D));			\
204 	CHECK_WRITE(0x39, (D));
205 
206 #define CHECK_PRINT(S);				\
207 	printf("%s: %d, %d, %d, %d, %d, %d\n",	\
208 	   (S),					\
209 	   CHECK_READ(0x34),			\
210 	   CHECK_READ(0x35),			\
211 	   CHECK_READ(0x36),			\
212 	   CHECK_READ(0x37),			\
213 	   CHECK_READ(0x38),			\
214 	   CHECK_READ(0x39));
215 
216 #else				/* CHECK_POINTS */
217 
218 #define CHECK_INIT(D)
219 #define CHECK_PRINT(S)
220 
221 #endif				/* CHECK_POINTS */
222 
223 /*
224  * Values to send to the POST hardware.
225  */
226 #define MP_BOOTADDRESS_POST	0x10
227 #define MP_PROBE_POST		0x11
228 #define MPTABLE_PASS1_POST	0x12
229 
230 #define MP_START_POST		0x13
231 #define MP_ENABLE_POST		0x14
232 #define MPTABLE_PASS2_POST	0x15
233 
234 #define START_ALL_APS_POST	0x16
235 #define INSTALL_AP_TRAMP_POST	0x17
236 #define START_AP_POST		0x18
237 
238 #define MP_ANNOUNCE_POST	0x19
239 
240 /* used to hold the AP's until we are ready to release them */
241 struct simplelock	ap_boot_lock;
242 
243 /** XXX FIXME: where does this really belong, isa.h/isa.c perhaps? */
244 int	current_postcode;
245 
246 /** XXX FIXME: what system files declare these??? */
247 extern struct region_descriptor r_gdt, r_idt;
248 
249 int	bsp_apic_ready = 0;	/* flags useability of BSP apic */
250 int	mp_ncpus;		/* # of CPUs, including BSP */
251 int	mp_naps;		/* # of Applications processors */
252 int	mp_nbusses;		/* # of busses */
253 int	mp_napics;		/* # of IO APICs */
254 int	boot_cpu_id;		/* designated BSP */
255 vm_offset_t cpu_apic_address;
256 vm_offset_t io_apic_address[NAPICID];	/* NAPICID is more than enough */
257 extern	int nkpt;
258 
259 u_int32_t cpu_apic_versions[MAXCPU];
260 u_int32_t *io_apic_versions;
261 
262 #ifdef APIC_INTR_DIAGNOSTIC
263 int apic_itrace_enter[32];
264 int apic_itrace_tryisrlock[32];
265 int apic_itrace_gotisrlock[32];
266 int apic_itrace_active[32];
267 int apic_itrace_masked[32];
268 int apic_itrace_noisrlock[32];
269 int apic_itrace_masked2[32];
270 int apic_itrace_unmask[32];
271 int apic_itrace_noforward[32];
272 int apic_itrace_leave[32];
273 int apic_itrace_enter2[32];
274 int apic_itrace_doreti[32];
275 int apic_itrace_splz[32];
276 int apic_itrace_eoi[32];
277 #ifdef APIC_INTR_DIAGNOSTIC_IRQ
278 unsigned short apic_itrace_debugbuffer[32768];
279 int apic_itrace_debugbuffer_idx;
280 struct simplelock apic_itrace_debuglock;
281 #endif
282 #endif
283 
284 #ifdef APIC_INTR_REORDER
285 struct {
286 	volatile int *location;
287 	int bit;
288 } apic_isrbit_location[32];
289 #endif
290 
291 struct apic_intmapinfo	int_to_apicintpin[APIC_INTMAPSIZE];
292 
293 /*
294  * APIC ID logical/physical mapping structures.
295  * We oversize these to simplify boot-time config.
296  */
297 int     cpu_num_to_apic_id[NAPICID];
298 int     io_num_to_apic_id[NAPICID];
299 int     apic_id_to_logical[NAPICID];
300 
301 
302 /* Bitmap of all available CPUs */
303 u_int	all_cpus;
304 
305 /* AP uses this during bootstrap.  Do not staticize.  */
306 char *bootSTK;
307 static int bootAP;
308 
309 /* Hotwire a 0->4MB V==P mapping */
310 extern pt_entry_t *KPTphys;
311 
312 /* SMP page table page */
313 extern pt_entry_t *SMPpt;
314 
315 struct pcb stoppcbs[MAXCPU];
316 
317 int smp_started;		/* has the system started? */
318 
319 /*
320  * Local data and functions.
321  */
322 
323 static int	mp_capable;
324 static u_int	boot_address;
325 static u_int	base_memory;
326 
327 static int	picmode;		/* 0: virtual wire mode, 1: PIC mode */
328 static mpfps_t	mpfps;
329 static int	search_for_sig(u_int32_t target, int count);
330 static void	mp_enable(u_int boot_addr);
331 
332 static void	mptable_pass1(void);
333 static int	mptable_pass2(void);
334 static void	default_mp_table(int type);
335 static void	fix_mp_table(void);
336 static void	setup_apic_irq_mapping(void);
337 static void	init_locks(void);
338 static int	start_all_aps(u_int boot_addr);
339 static void	install_ap_tramp(u_int boot_addr);
340 static int	start_ap(int logicalCpu, u_int boot_addr);
341 static int	apic_int_is_bus_type(int intr, int bus_type);
342 static void	release_aps(void *dummy);
343 
344 /*
345  * Calculate usable address in base memory for AP trampoline code.
346  */
347 u_int
348 mp_bootaddress(u_int basemem)
349 {
350 	POSTCODE(MP_BOOTADDRESS_POST);
351 
352 	base_memory = basemem * 1024;	/* convert to bytes */
353 
354 	boot_address = base_memory & ~0xfff;	/* round down to 4k boundary */
355 	if ((base_memory - boot_address) < bootMP_size)
356 		boot_address -= 4096;	/* not enough, lower by 4k */
357 
358 	return boot_address;
359 }
360 
361 
362 /*
363  * Look for an Intel MP spec table (ie, SMP capable hardware).
364  */
365 int
366 mp_probe(void)
367 {
368 	int     x;
369 	u_long  segment;
370 	u_int32_t target;
371 
372 	POSTCODE(MP_PROBE_POST);
373 
374 	/* see if EBDA exists */
375 	if ((segment = (u_long) * (u_short *) (KERNBASE + 0x40e)) != 0) {
376 		/* search first 1K of EBDA */
377 		target = (u_int32_t) (segment << 4);
378 		if ((x = search_for_sig(target, 1024 / 4)) >= 0)
379 			goto found;
380 	} else {
381 		/* last 1K of base memory, effective 'top of base' passed in */
382 		target = (u_int32_t) (base_memory - 0x400);
383 		if ((x = search_for_sig(target, 1024 / 4)) >= 0)
384 			goto found;
385 	}
386 
387 	/* search the BIOS */
388 	target = (u_int32_t) BIOS_BASE;
389 	if ((x = search_for_sig(target, BIOS_COUNT)) >= 0)
390 		goto found;
391 
392 	/* nothing found */
393 	mpfps = (mpfps_t)0;
394 	mp_capable = 0;
395 	return 0;
396 
397 found:
398 	/* calculate needed resources */
399 	mpfps = (mpfps_t)x;
400 	mptable_pass1();
401 
402 	/* flag fact that we are running multiple processors */
403 	mp_capable = 1;
404 	return 1;
405 }
406 
407 
408 /*
409  * Initialize the SMP hardware and the APIC and start up the AP's.
410  */
411 void
412 mp_start(void)
413 {
414 	POSTCODE(MP_START_POST);
415 
416 	/* look for MP capable motherboard */
417 	if (mp_capable)
418 		mp_enable(boot_address);
419 	else
420 		panic("MP hardware not found!");
421 }
422 
423 
424 /*
425  * Print various information about the SMP system hardware and setup.
426  */
427 void
428 mp_announce(void)
429 {
430 	int     x;
431 
432 	POSTCODE(MP_ANNOUNCE_POST);
433 
434 	printf("FreeBSD/SMP: Multiprocessor motherboard\n");
435 	printf(" cpu0 (BSP): apic id: %2d", CPU_TO_ID(0));
436 	printf(", version: 0x%08x", cpu_apic_versions[0]);
437 	printf(", at 0x%08x\n", cpu_apic_address);
438 	for (x = 1; x <= mp_naps; ++x) {
439 		printf(" cpu%d (AP):  apic id: %2d", x, CPU_TO_ID(x));
440 		printf(", version: 0x%08x", cpu_apic_versions[x]);
441 		printf(", at 0x%08x\n", cpu_apic_address);
442 	}
443 
444 #if defined(APIC_IO)
445 	for (x = 0; x < mp_napics; ++x) {
446 		printf(" io%d (APIC): apic id: %2d", x, IO_TO_ID(x));
447 		printf(", version: 0x%08x", io_apic_versions[x]);
448 		printf(", at 0x%08x\n", io_apic_address[x]);
449 	}
450 #else
451 	printf(" Warning: APIC I/O disabled\n");
452 #endif	/* APIC_IO */
453 }
454 
455 /*
456  * AP cpu's call this to sync up protected mode.
457  */
458 void
459 init_secondary(void)
460 {
461 	int	gsel_tss;
462 	int	x, myid = bootAP;
463 
464 	gdt_segs[GPRIV_SEL].ssd_base = (int) &SMP_prvspace[myid];
465 	gdt_segs[GPROC0_SEL].ssd_base =
466 		(int) &SMP_prvspace[myid].globaldata.gd_common_tss;
467 	SMP_prvspace[myid].globaldata.gd_prvspace = &SMP_prvspace[myid];
468 
469 	for (x = 0; x < NGDT; x++) {
470 		ssdtosd(&gdt_segs[x], &gdt[myid * NGDT + x].sd);
471 	}
472 
473 	r_gdt.rd_limit = NGDT * sizeof(gdt[0]) - 1;
474 	r_gdt.rd_base = (int) &gdt[myid * NGDT];
475 	lgdt(&r_gdt);			/* does magic intra-segment return */
476 
477 	lidt(&r_idt);
478 
479 	lldt(_default_ldt);
480 #ifdef USER_LDT
481 	PCPU_SET(currentldt, _default_ldt);
482 #endif
483 
484 	gsel_tss = GSEL(GPROC0_SEL, SEL_KPL);
485 	gdt[myid * NGDT + GPROC0_SEL].sd.sd_type = SDT_SYS386TSS;
486 	common_tss.tss_esp0 = 0;	/* not used until after switch */
487 	common_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
488 	common_tss.tss_ioopt = (sizeof common_tss) << 16;
489 	tss_gdt = &gdt[myid * NGDT + GPROC0_SEL].sd;
490 	common_tssd = *tss_gdt;
491 	ltr(gsel_tss);
492 
493 	pmap_set_opt();
494 }
495 
496 
497 #if defined(APIC_IO)
498 /*
499  * Final configuration of the BSP's local APIC:
500  *  - disable 'pic mode'.
501  *  - disable 'virtual wire mode'.
502  *  - enable NMI.
503  */
504 void
505 bsp_apic_configure(void)
506 {
507 	u_char		byte;
508 	u_int32_t	temp;
509 
510 	/* leave 'pic mode' if necessary */
511 	if (picmode) {
512 		outb(0x22, 0x70);	/* select IMCR */
513 		byte = inb(0x23);	/* current contents */
514 		byte |= 0x01;		/* mask external INTR */
515 		outb(0x23, byte);	/* disconnect 8259s/NMI */
516 	}
517 
518 	/* mask lint0 (the 8259 'virtual wire' connection) */
519 	temp = lapic.lvt_lint0;
520 	temp |= APIC_LVT_M;		/* set the mask */
521 	lapic.lvt_lint0 = temp;
522 
523         /* setup lint1 to handle NMI */
524         temp = lapic.lvt_lint1;
525         temp &= ~APIC_LVT_M;		/* clear the mask */
526         lapic.lvt_lint1 = temp;
527 
528 	if (bootverbose)
529 		apic_dump("bsp_apic_configure()");
530 }
531 #endif  /* APIC_IO */
532 
533 
534 /*******************************************************************
535  * local functions and data
536  */
537 
538 /*
539  * start the SMP system
540  */
541 static void
542 mp_enable(u_int boot_addr)
543 {
544 	int     x;
545 #if defined(APIC_IO)
546 	int     apic;
547 	u_int   ux;
548 #endif	/* APIC_IO */
549 
550 	POSTCODE(MP_ENABLE_POST);
551 
552 	/* turn on 4MB of V == P addressing so we can get to MP table */
553 	*(int *)PTD = PG_V | PG_RW | ((uintptr_t)(void *)KPTphys & PG_FRAME);
554 	invltlb();
555 
556 	/* examine the MP table for needed info, uses physical addresses */
557 	x = mptable_pass2();
558 
559 	*(int *)PTD = 0;
560 	invltlb();
561 
562 	/* can't process default configs till the CPU APIC is pmapped */
563 	if (x)
564 		default_mp_table(x);
565 
566 	/* post scan cleanup */
567 	fix_mp_table();
568 	setup_apic_irq_mapping();
569 
570 #if defined(APIC_IO)
571 
572 	/* fill the LOGICAL io_apic_versions table */
573 	for (apic = 0; apic < mp_napics; ++apic) {
574 		ux = io_apic_read(apic, IOAPIC_VER);
575 		io_apic_versions[apic] = ux;
576 		io_apic_set_id(apic, IO_TO_ID(apic));
577 	}
578 
579 	/* program each IO APIC in the system */
580 	for (apic = 0; apic < mp_napics; ++apic)
581 		if (io_apic_setup(apic) < 0)
582 			panic("IO APIC setup failure");
583 
584 	/* install a 'Spurious INTerrupt' vector */
585 	setidt(XSPURIOUSINT_OFFSET, Xspuriousint,
586 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
587 
588 	/* install an inter-CPU IPI for TLB invalidation */
589 	setidt(XINVLTLB_OFFSET, Xinvltlb,
590 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
591 
592 #ifdef BETTER_CLOCK
593 	/* install an inter-CPU IPI for reading processor state */
594 	setidt(XCPUCHECKSTATE_OFFSET, Xcpucheckstate,
595 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
596 #endif
597 
598 	/* install an inter-CPU IPI for all-CPU rendezvous */
599 	setidt(XRENDEZVOUS_OFFSET, Xrendezvous,
600 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
601 
602 	/* install an inter-CPU IPI for forcing an additional software trap */
603 	setidt(XCPUAST_OFFSET, Xcpuast,
604 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
605 
606 	/* install an inter-CPU IPI for interrupt forwarding */
607 	setidt(XFORWARD_IRQ_OFFSET, Xforward_irq,
608 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
609 
610 	/* install an inter-CPU IPI for CPU stop/restart */
611 	setidt(XCPUSTOP_OFFSET, Xcpustop,
612 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
613 
614 #if defined(TEST_TEST1)
615 	/* install a "fake hardware INTerrupt" vector */
616 	setidt(XTEST1_OFFSET, Xtest1,
617 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
618 #endif  /** TEST_TEST1 */
619 
620 #endif	/* APIC_IO */
621 
622 	/* initialize all SMP locks */
623 	init_locks();
624 
625 	/* obtain the ap_boot_lock */
626 	s_lock(&ap_boot_lock);
627 
628 	/* start each Application Processor */
629 	start_all_aps(boot_addr);
630 }
631 
632 
633 /*
634  * look for the MP spec signature
635  */
636 
637 /* string defined by the Intel MP Spec as identifying the MP table */
638 #define MP_SIG		0x5f504d5f	/* _MP_ */
639 #define NEXT(X)		((X) += 4)
640 static int
641 search_for_sig(u_int32_t target, int count)
642 {
643 	int     x;
644 	u_int32_t *addr = (u_int32_t *) (KERNBASE + target);
645 
646 	for (x = 0; x < count; NEXT(x))
647 		if (addr[x] == MP_SIG)
648 			/* make array index a byte index */
649 			return (target + (x * sizeof(u_int32_t)));
650 
651 	return -1;
652 }
653 
654 
655 static basetable_entry basetable_entry_types[] =
656 {
657 	{0, 20, "Processor"},
658 	{1, 8, "Bus"},
659 	{2, 8, "I/O APIC"},
660 	{3, 8, "I/O INT"},
661 	{4, 8, "Local INT"}
662 };
663 
664 typedef struct BUSDATA {
665 	u_char  bus_id;
666 	enum busTypes bus_type;
667 }       bus_datum;
668 
669 typedef struct INTDATA {
670 	u_char  int_type;
671 	u_short int_flags;
672 	u_char  src_bus_id;
673 	u_char  src_bus_irq;
674 	u_char  dst_apic_id;
675 	u_char  dst_apic_int;
676 	u_char	int_vector;
677 }       io_int, local_int;
678 
679 typedef struct BUSTYPENAME {
680 	u_char  type;
681 	char    name[7];
682 }       bus_type_name;
683 
684 static bus_type_name bus_type_table[] =
685 {
686 	{CBUS, "CBUS"},
687 	{CBUSII, "CBUSII"},
688 	{EISA, "EISA"},
689 	{MCA, "MCA"},
690 	{UNKNOWN_BUSTYPE, "---"},
691 	{ISA, "ISA"},
692 	{MCA, "MCA"},
693 	{UNKNOWN_BUSTYPE, "---"},
694 	{UNKNOWN_BUSTYPE, "---"},
695 	{UNKNOWN_BUSTYPE, "---"},
696 	{UNKNOWN_BUSTYPE, "---"},
697 	{UNKNOWN_BUSTYPE, "---"},
698 	{PCI, "PCI"},
699 	{UNKNOWN_BUSTYPE, "---"},
700 	{UNKNOWN_BUSTYPE, "---"},
701 	{UNKNOWN_BUSTYPE, "---"},
702 	{UNKNOWN_BUSTYPE, "---"},
703 	{XPRESS, "XPRESS"},
704 	{UNKNOWN_BUSTYPE, "---"}
705 };
706 /* from MP spec v1.4, table 5-1 */
707 static int default_data[7][5] =
708 {
709 /*   nbus, id0, type0, id1, type1 */
710 	{1, 0, ISA, 255, 255},
711 	{1, 0, EISA, 255, 255},
712 	{1, 0, EISA, 255, 255},
713 	{1, 0, MCA, 255, 255},
714 	{2, 0, ISA, 1, PCI},
715 	{2, 0, EISA, 1, PCI},
716 	{2, 0, MCA, 1, PCI}
717 };
718 
719 
720 /* the bus data */
721 static bus_datum *bus_data;
722 
723 /* the IO INT data, one entry per possible APIC INTerrupt */
724 static io_int  *io_apic_ints;
725 
726 static int nintrs;
727 
728 static int processor_entry	__P((proc_entry_ptr entry, int cpu));
729 static int bus_entry		__P((bus_entry_ptr entry, int bus));
730 static int io_apic_entry	__P((io_apic_entry_ptr entry, int apic));
731 static int int_entry		__P((int_entry_ptr entry, int intr));
732 static int lookup_bus_type	__P((char *name));
733 
734 
735 /*
736  * 1st pass on motherboard's Intel MP specification table.
737  *
738  * initializes:
739  *	mp_ncpus = 1
740  *
741  * determines:
742  *	cpu_apic_address (common to all CPUs)
743  *	io_apic_address[N]
744  *	mp_naps
745  *	mp_nbusses
746  *	mp_napics
747  *	nintrs
748  */
749 static void
750 mptable_pass1(void)
751 {
752 	int	x;
753 	mpcth_t	cth;
754 	int	totalSize;
755 	void*	position;
756 	int	count;
757 	int	type;
758 
759 	POSTCODE(MPTABLE_PASS1_POST);
760 
761 	/* clear various tables */
762 	for (x = 0; x < NAPICID; ++x) {
763 		io_apic_address[x] = ~0;	/* IO APIC address table */
764 	}
765 
766 	/* init everything to empty */
767 	mp_naps = 0;
768 	mp_nbusses = 0;
769 	mp_napics = 0;
770 	nintrs = 0;
771 
772 	/* check for use of 'default' configuration */
773 	if (MPFPS_MPFB1 != 0) {
774 		/* use default addresses */
775 		cpu_apic_address = DEFAULT_APIC_BASE;
776 		io_apic_address[0] = DEFAULT_IO_APIC_BASE;
777 
778 		/* fill in with defaults */
779 		mp_naps = 2;		/* includes BSP */
780 		mp_nbusses = default_data[MPFPS_MPFB1 - 1][0];
781 #if defined(APIC_IO)
782 		mp_napics = 1;
783 		nintrs = 16;
784 #endif	/* APIC_IO */
785 	}
786 	else {
787 		if ((cth = mpfps->pap) == 0)
788 			panic("MP Configuration Table Header MISSING!");
789 
790 		cpu_apic_address = (vm_offset_t) cth->apic_address;
791 
792 		/* walk the table, recording info of interest */
793 		totalSize = cth->base_table_length - sizeof(struct MPCTH);
794 		position = (u_char *) cth + sizeof(struct MPCTH);
795 		count = cth->entry_count;
796 
797 		while (count--) {
798 			switch (type = *(u_char *) position) {
799 			case 0: /* processor_entry */
800 				if (((proc_entry_ptr)position)->cpu_flags
801 					& PROCENTRY_FLAG_EN)
802 					++mp_naps;
803 				break;
804 			case 1: /* bus_entry */
805 				++mp_nbusses;
806 				break;
807 			case 2: /* io_apic_entry */
808 				if (((io_apic_entry_ptr)position)->apic_flags
809 					& IOAPICENTRY_FLAG_EN)
810 					io_apic_address[mp_napics++] =
811 					    (vm_offset_t)((io_apic_entry_ptr)
812 						position)->apic_address;
813 				break;
814 			case 3: /* int_entry */
815 				++nintrs;
816 				break;
817 			case 4:	/* int_entry */
818 				break;
819 			default:
820 				panic("mpfps Base Table HOSED!");
821 				/* NOTREACHED */
822 			}
823 
824 			totalSize -= basetable_entry_types[type].length;
825 			(u_char*)position += basetable_entry_types[type].length;
826 		}
827 	}
828 
829 	/* qualify the numbers */
830 	if (mp_naps > MAXCPU) {
831 		printf("Warning: only using %d of %d available CPUs!\n",
832 			MAXCPU, mp_naps);
833 		mp_naps = MAXCPU;
834 	}
835 
836 	/*
837 	 * Count the BSP.
838 	 * This is also used as a counter while starting the APs.
839 	 */
840 	mp_ncpus = 1;
841 
842 	--mp_naps;	/* subtract the BSP */
843 }
844 
845 
846 /*
847  * 2nd pass on motherboard's Intel MP specification table.
848  *
849  * sets:
850  *	boot_cpu_id
851  *	ID_TO_IO(N), phy APIC ID to log CPU/IO table
852  *	CPU_TO_ID(N), logical CPU to APIC ID table
853  *	IO_TO_ID(N), logical IO to APIC ID table
854  *	bus_data[N]
855  *	io_apic_ints[N]
856  */
857 static int
858 mptable_pass2(void)
859 {
860 	int     x;
861 	mpcth_t cth;
862 	int     totalSize;
863 	void*   position;
864 	int     count;
865 	int     type;
866 	int     apic, bus, cpu, intr;
867 	int	i, j;
868 	int	pgeflag;
869 
870 	POSTCODE(MPTABLE_PASS2_POST);
871 
872 	pgeflag = 0;		/* XXX - Not used under SMP yet.  */
873 
874 	MALLOC(io_apic_versions, u_int32_t *, sizeof(u_int32_t) * mp_napics,
875 	    M_DEVBUF, M_WAITOK);
876 	MALLOC(ioapic, volatile ioapic_t **, sizeof(ioapic_t *) * mp_napics,
877 	    M_DEVBUF, M_WAITOK);
878 	MALLOC(io_apic_ints, io_int *, sizeof(io_int) * nintrs,
879 	    M_DEVBUF, M_WAITOK);
880 	MALLOC(bus_data, bus_datum *, sizeof(bus_datum) * mp_nbusses,
881 	    M_DEVBUF, M_WAITOK);
882 
883 	bzero(ioapic, sizeof(ioapic_t *) * mp_napics);
884 
885 	for (i = 0; i < mp_napics; i++) {
886 		for (j = 0; j < mp_napics; j++) {
887 			/* same page frame as a previous IO apic? */
888 			if (((vm_offset_t)SMPpt[NPTEPG-2-j] & PG_FRAME) ==
889 			    (io_apic_address[i] & PG_FRAME)) {
890 				ioapic[i] = (ioapic_t *)((u_int)SMP_prvspace
891 					+ (NPTEPG-2-j) * PAGE_SIZE
892 					+ (io_apic_address[i] & PAGE_MASK));
893 				break;
894 			}
895 			/* use this slot if available */
896 			if (((vm_offset_t)SMPpt[NPTEPG-2-j] & PG_FRAME) == 0) {
897 				SMPpt[NPTEPG-2-j] = (pt_entry_t)(PG_V | PG_RW |
898 				    pgeflag | (io_apic_address[i] & PG_FRAME));
899 				ioapic[i] = (ioapic_t *)((u_int)SMP_prvspace
900 					+ (NPTEPG-2-j) * PAGE_SIZE
901 					+ (io_apic_address[i] & PAGE_MASK));
902 				break;
903 			}
904 		}
905 	}
906 
907 	/* clear various tables */
908 	for (x = 0; x < NAPICID; ++x) {
909 		ID_TO_IO(x) = -1;	/* phy APIC ID to log CPU/IO table */
910 		CPU_TO_ID(x) = -1;	/* logical CPU to APIC ID table */
911 		IO_TO_ID(x) = -1;	/* logical IO to APIC ID table */
912 	}
913 
914 	/* clear bus data table */
915 	for (x = 0; x < mp_nbusses; ++x)
916 		bus_data[x].bus_id = 0xff;
917 
918 	/* clear IO APIC INT table */
919 	for (x = 0; x < nintrs; ++x) {
920 		io_apic_ints[x].int_type = 0xff;
921 		io_apic_ints[x].int_vector = 0xff;
922 	}
923 
924 	/* setup the cpu/apic mapping arrays */
925 	boot_cpu_id = -1;
926 
927 	/* record whether PIC or virtual-wire mode */
928 	picmode = (mpfps->mpfb2 & 0x80) ? 1 : 0;
929 
930 	/* check for use of 'default' configuration */
931 	if (MPFPS_MPFB1 != 0)
932 		return MPFPS_MPFB1;	/* return default configuration type */
933 
934 	if ((cth = mpfps->pap) == 0)
935 		panic("MP Configuration Table Header MISSING!");
936 
937 	/* walk the table, recording info of interest */
938 	totalSize = cth->base_table_length - sizeof(struct MPCTH);
939 	position = (u_char *) cth + sizeof(struct MPCTH);
940 	count = cth->entry_count;
941 	apic = bus = intr = 0;
942 	cpu = 1;				/* pre-count the BSP */
943 
944 	while (count--) {
945 		switch (type = *(u_char *) position) {
946 		case 0:
947 			if (processor_entry(position, cpu))
948 				++cpu;
949 			break;
950 		case 1:
951 			if (bus_entry(position, bus))
952 				++bus;
953 			break;
954 		case 2:
955 			if (io_apic_entry(position, apic))
956 				++apic;
957 			break;
958 		case 3:
959 			if (int_entry(position, intr))
960 				++intr;
961 			break;
962 		case 4:
963 			/* int_entry(position); */
964 			break;
965 		default:
966 			panic("mpfps Base Table HOSED!");
967 			/* NOTREACHED */
968 		}
969 
970 		totalSize -= basetable_entry_types[type].length;
971 		(u_char *) position += basetable_entry_types[type].length;
972 	}
973 
974 	if (boot_cpu_id == -1)
975 		panic("NO BSP found!");
976 
977 	/* report fact that its NOT a default configuration */
978 	return 0;
979 }
980 
981 
982 void
983 assign_apic_irq(int apic, int intpin, int irq)
984 {
985 	int x;
986 
987 	if (int_to_apicintpin[irq].ioapic != -1)
988 		panic("assign_apic_irq: inconsistent table");
989 
990 	int_to_apicintpin[irq].ioapic = apic;
991 	int_to_apicintpin[irq].int_pin = intpin;
992 	int_to_apicintpin[irq].apic_address = ioapic[apic];
993 	int_to_apicintpin[irq].redirindex = IOAPIC_REDTBL + 2 * intpin;
994 
995 	for (x = 0; x < nintrs; x++) {
996 		if ((io_apic_ints[x].int_type == 0 ||
997 		     io_apic_ints[x].int_type == 3) &&
998 		    io_apic_ints[x].int_vector == 0xff &&
999 		    io_apic_ints[x].dst_apic_id == IO_TO_ID(apic) &&
1000 		    io_apic_ints[x].dst_apic_int == intpin)
1001 			io_apic_ints[x].int_vector = irq;
1002 	}
1003 }
1004 
1005 void
1006 revoke_apic_irq(int irq)
1007 {
1008 	int x;
1009 	int oldapic;
1010 	int oldintpin;
1011 
1012 	if (int_to_apicintpin[irq].ioapic == -1)
1013 		panic("assign_apic_irq: inconsistent table");
1014 
1015 	oldapic = int_to_apicintpin[irq].ioapic;
1016 	oldintpin = int_to_apicintpin[irq].int_pin;
1017 
1018 	int_to_apicintpin[irq].ioapic = -1;
1019 	int_to_apicintpin[irq].int_pin = 0;
1020 	int_to_apicintpin[irq].apic_address = NULL;
1021 	int_to_apicintpin[irq].redirindex = 0;
1022 
1023 	for (x = 0; x < nintrs; x++) {
1024 		if ((io_apic_ints[x].int_type == 0 ||
1025 		     io_apic_ints[x].int_type == 3) &&
1026 		    io_apic_ints[x].int_vector == 0xff &&
1027 		    io_apic_ints[x].dst_apic_id == IO_TO_ID(oldapic) &&
1028 		    io_apic_ints[x].dst_apic_int == oldintpin)
1029 			io_apic_ints[x].int_vector = 0xff;
1030 	}
1031 }
1032 
1033 
1034 
1035 static void
1036 swap_apic_id(int apic, int oldid, int newid)
1037 {
1038 	int x;
1039 	int oapic;
1040 
1041 
1042 	if (oldid == newid)
1043 		return;			/* Nothing to do */
1044 
1045 	printf("Changing APIC ID for IO APIC #%d from %d to %d in MP table\n",
1046 	       apic, oldid, newid);
1047 
1048 	/* Swap physical APIC IDs in interrupt entries */
1049 	for (x = 0; x < nintrs; x++) {
1050 		if (io_apic_ints[x].dst_apic_id == oldid)
1051 			io_apic_ints[x].dst_apic_id = newid;
1052 		else if (io_apic_ints[x].dst_apic_id == newid)
1053 			io_apic_ints[x].dst_apic_id = oldid;
1054 	}
1055 
1056 	/* Swap physical APIC IDs in IO_TO_ID mappings */
1057 	for (oapic = 0; oapic < mp_napics; oapic++)
1058 		if (IO_TO_ID(oapic) == newid)
1059 			break;
1060 
1061 	if (oapic < mp_napics) {
1062 		printf("Changing APIC ID for IO APIC #%d from "
1063 		       "%d to %d in MP table\n",
1064 		       oapic, newid, oldid);
1065 		IO_TO_ID(oapic) = oldid;
1066 	}
1067 	IO_TO_ID(apic) = newid;
1068 }
1069 
1070 
1071 static void
1072 fix_id_to_io_mapping(void)
1073 {
1074 	int x;
1075 
1076 	for (x = 0; x < NAPICID; x++)
1077 		ID_TO_IO(x) = -1;
1078 
1079 	for (x = 0; x <= mp_naps; x++)
1080 		if (CPU_TO_ID(x) < NAPICID)
1081 			ID_TO_IO(CPU_TO_ID(x)) = x;
1082 
1083 	for (x = 0; x < mp_napics; x++)
1084 		if (IO_TO_ID(x) < NAPICID)
1085 			ID_TO_IO(IO_TO_ID(x)) = x;
1086 }
1087 
1088 
1089 static int
1090 first_free_apic_id(void)
1091 {
1092 	int freeid, x;
1093 
1094 	for (freeid = 0; freeid < NAPICID; freeid++) {
1095 		for (x = 0; x <= mp_naps; x++)
1096 			if (CPU_TO_ID(x) == freeid)
1097 				break;
1098 		if (x <= mp_naps)
1099 			continue;
1100 		for (x = 0; x < mp_napics; x++)
1101 			if (IO_TO_ID(x) == freeid)
1102 				break;
1103 		if (x < mp_napics)
1104 			continue;
1105 		return freeid;
1106 	}
1107 	return freeid;
1108 }
1109 
1110 
1111 static int
1112 io_apic_id_acceptable(int apic, int id)
1113 {
1114 	int cpu;		/* Logical CPU number */
1115 	int oapic;		/* Logical IO APIC number for other IO APIC */
1116 
1117 	if (id >= NAPICID)
1118 		return 0;	/* Out of range */
1119 
1120 	for (cpu = 0; cpu <= mp_naps; cpu++)
1121 		if (CPU_TO_ID(cpu) == id)
1122 			return 0;	/* Conflict with CPU */
1123 
1124 	for (oapic = 0; oapic < mp_napics && oapic < apic; oapic++)
1125 		if (IO_TO_ID(oapic) == id)
1126 			return 0;	/* Conflict with other APIC */
1127 
1128 	return 1;		/* ID is acceptable for IO APIC */
1129 }
1130 
1131 
1132 /*
1133  * parse an Intel MP specification table
1134  */
1135 static void
1136 fix_mp_table(void)
1137 {
1138 	int	x;
1139 	int	id;
1140 	int	bus_0 = 0;	/* Stop GCC warning */
1141 	int	bus_pci = 0;	/* Stop GCC warning */
1142 	int	num_pci_bus;
1143 	int	apic;		/* IO APIC unit number */
1144 	int     freeid;		/* Free physical APIC ID */
1145 	int	physid;		/* Current physical IO APIC ID */
1146 
1147 	/*
1148 	 * Fix mis-numbering of the PCI bus and its INT entries if the BIOS
1149 	 * did it wrong.  The MP spec says that when more than 1 PCI bus
1150 	 * exists the BIOS must begin with bus entries for the PCI bus and use
1151 	 * actual PCI bus numbering.  This implies that when only 1 PCI bus
1152 	 * exists the BIOS can choose to ignore this ordering, and indeed many
1153 	 * MP motherboards do ignore it.  This causes a problem when the PCI
1154 	 * sub-system makes requests of the MP sub-system based on PCI bus
1155 	 * numbers.	So here we look for the situation and renumber the
1156 	 * busses and associated INTs in an effort to "make it right".
1157 	 */
1158 
1159 	/* find bus 0, PCI bus, count the number of PCI busses */
1160 	for (num_pci_bus = 0, x = 0; x < mp_nbusses; ++x) {
1161 		if (bus_data[x].bus_id == 0) {
1162 			bus_0 = x;
1163 		}
1164 		if (bus_data[x].bus_type == PCI) {
1165 			++num_pci_bus;
1166 			bus_pci = x;
1167 		}
1168 	}
1169 	/*
1170 	 * bus_0 == slot of bus with ID of 0
1171 	 * bus_pci == slot of last PCI bus encountered
1172 	 */
1173 
1174 	/* check the 1 PCI bus case for sanity */
1175 	/* if it is number 0 all is well */
1176 	if (num_pci_bus == 1 &&
1177 	    bus_data[bus_pci].bus_id != 0) {
1178 
1179 		/* mis-numbered, swap with whichever bus uses slot 0 */
1180 
1181 		/* swap the bus entry types */
1182 		bus_data[bus_pci].bus_type = bus_data[bus_0].bus_type;
1183 		bus_data[bus_0].bus_type = PCI;
1184 
1185 		/* swap each relavant INTerrupt entry */
1186 		id = bus_data[bus_pci].bus_id;
1187 		for (x = 0; x < nintrs; ++x) {
1188 			if (io_apic_ints[x].src_bus_id == id) {
1189 				io_apic_ints[x].src_bus_id = 0;
1190 			}
1191 			else if (io_apic_ints[x].src_bus_id == 0) {
1192 				io_apic_ints[x].src_bus_id = id;
1193 			}
1194 		}
1195 	}
1196 
1197 	/* Assign IO APIC IDs.
1198 	 *
1199 	 * First try the existing ID. If a conflict is detected, try
1200 	 * the ID in the MP table.  If a conflict is still detected, find
1201 	 * a free id.
1202 	 *
1203 	 * We cannot use the ID_TO_IO table before all conflicts has been
1204 	 * resolved and the table has been corrected.
1205 	 */
1206 	for (apic = 0; apic < mp_napics; ++apic) { /* For all IO APICs */
1207 
1208 		/* First try to use the value set by the BIOS */
1209 		physid = io_apic_get_id(apic);
1210 		if (io_apic_id_acceptable(apic, physid)) {
1211 			if (IO_TO_ID(apic) != physid)
1212 				swap_apic_id(apic, IO_TO_ID(apic), physid);
1213 			continue;
1214 		}
1215 
1216 		/* Then check if the value in the MP table is acceptable */
1217 		if (io_apic_id_acceptable(apic, IO_TO_ID(apic)))
1218 			continue;
1219 
1220 		/* Last resort, find a free APIC ID and use it */
1221 		freeid = first_free_apic_id();
1222 		if (freeid >= NAPICID)
1223 			panic("No free physical APIC IDs found");
1224 
1225 		if (io_apic_id_acceptable(apic, freeid)) {
1226 			swap_apic_id(apic, IO_TO_ID(apic), freeid);
1227 			continue;
1228 		}
1229 		panic("Free physical APIC ID not usable");
1230 	}
1231 	fix_id_to_io_mapping();
1232 }
1233 
1234 
1235 /* Assign low level interrupt handlers */
1236 static void
1237 setup_apic_irq_mapping(void)
1238 {
1239 	int	x;
1240 	int	int_vector;
1241 
1242 	/* Clear array */
1243 	for (x = 0; x < APIC_INTMAPSIZE; x++) {
1244 		int_to_apicintpin[x].ioapic = -1;
1245 		int_to_apicintpin[x].int_pin = 0;
1246 		int_to_apicintpin[x].apic_address = NULL;
1247 		int_to_apicintpin[x].redirindex = 0;
1248 	}
1249 
1250 	/* First assign ISA/EISA interrupts */
1251 	for (x = 0; x < nintrs; x++) {
1252 		int_vector = io_apic_ints[x].src_bus_irq;
1253 		if (int_vector < APIC_INTMAPSIZE &&
1254 		    io_apic_ints[x].int_vector == 0xff &&
1255 		    int_to_apicintpin[int_vector].ioapic == -1 &&
1256 		    (apic_int_is_bus_type(x, ISA) ||
1257 		     apic_int_is_bus_type(x, EISA)) &&
1258 		    io_apic_ints[x].int_type == 0) {
1259 			assign_apic_irq(ID_TO_IO(io_apic_ints[x].dst_apic_id),
1260 					io_apic_ints[x].dst_apic_int,
1261 					int_vector);
1262 		}
1263 	}
1264 
1265 	/* Assign interrupts on first 24 intpins on IOAPIC #0 */
1266 	for (x = 0; x < nintrs; x++) {
1267 		int_vector = io_apic_ints[x].dst_apic_int;
1268 		if (int_vector < APIC_INTMAPSIZE &&
1269 		    io_apic_ints[x].dst_apic_id == IO_TO_ID(0) &&
1270 		    io_apic_ints[x].int_vector == 0xff &&
1271 		    int_to_apicintpin[int_vector].ioapic == -1 &&
1272 		    (io_apic_ints[x].int_type == 0 ||
1273 		     io_apic_ints[x].int_type == 3)) {
1274 			assign_apic_irq(0,
1275 					io_apic_ints[x].dst_apic_int,
1276 					int_vector);
1277 		}
1278 	}
1279 	/*
1280 	 * Assign interrupts for remaining intpins.
1281 	 * Skip IOAPIC #0 intpin 0 if the type is ExtInt, since this indicates
1282 	 * that an entry for ISA/EISA irq 0 exist, and a fallback to mixed mode
1283 	 * due to 8254 interrupts not being delivered can reuse that low level
1284 	 * interrupt handler.
1285 	 */
1286 	int_vector = 0;
1287 	while (int_vector < APIC_INTMAPSIZE &&
1288 	       int_to_apicintpin[int_vector].ioapic != -1)
1289 		int_vector++;
1290 	for (x = 0; x < nintrs && int_vector < APIC_INTMAPSIZE; x++) {
1291 		if ((io_apic_ints[x].int_type == 0 ||
1292 		     (io_apic_ints[x].int_type == 3 &&
1293 		      (io_apic_ints[x].dst_apic_id != IO_TO_ID(0) ||
1294 		       io_apic_ints[x].dst_apic_int != 0))) &&
1295 		    io_apic_ints[x].int_vector == 0xff) {
1296 			assign_apic_irq(ID_TO_IO(io_apic_ints[x].dst_apic_id),
1297 					io_apic_ints[x].dst_apic_int,
1298 					int_vector);
1299 			int_vector++;
1300 			while (int_vector < APIC_INTMAPSIZE &&
1301 			       int_to_apicintpin[int_vector].ioapic != -1)
1302 				int_vector++;
1303 		}
1304 	}
1305 }
1306 
1307 
1308 static int
1309 processor_entry(proc_entry_ptr entry, int cpu)
1310 {
1311 	/* check for usability */
1312 	if (!(entry->cpu_flags & PROCENTRY_FLAG_EN))
1313 		return 0;
1314 
1315 	if(entry->apic_id >= NAPICID)
1316 		panic("CPU APIC ID out of range (0..%d)", NAPICID - 1);
1317 	/* check for BSP flag */
1318 	if (entry->cpu_flags & PROCENTRY_FLAG_BP) {
1319 		boot_cpu_id = entry->apic_id;
1320 		CPU_TO_ID(0) = entry->apic_id;
1321 		ID_TO_CPU(entry->apic_id) = 0;
1322 		return 0;	/* its already been counted */
1323 	}
1324 
1325 	/* add another AP to list, if less than max number of CPUs */
1326 	else if (cpu < MAXCPU) {
1327 		CPU_TO_ID(cpu) = entry->apic_id;
1328 		ID_TO_CPU(entry->apic_id) = cpu;
1329 		return 1;
1330 	}
1331 
1332 	return 0;
1333 }
1334 
1335 
1336 static int
1337 bus_entry(bus_entry_ptr entry, int bus)
1338 {
1339 	int     x;
1340 	char    c, name[8];
1341 
1342 	/* encode the name into an index */
1343 	for (x = 0; x < 6; ++x) {
1344 		if ((c = entry->bus_type[x]) == ' ')
1345 			break;
1346 		name[x] = c;
1347 	}
1348 	name[x] = '\0';
1349 
1350 	if ((x = lookup_bus_type(name)) == UNKNOWN_BUSTYPE)
1351 		panic("unknown bus type: '%s'", name);
1352 
1353 	bus_data[bus].bus_id = entry->bus_id;
1354 	bus_data[bus].bus_type = x;
1355 
1356 	return 1;
1357 }
1358 
1359 
1360 static int
1361 io_apic_entry(io_apic_entry_ptr entry, int apic)
1362 {
1363 	if (!(entry->apic_flags & IOAPICENTRY_FLAG_EN))
1364 		return 0;
1365 
1366 	IO_TO_ID(apic) = entry->apic_id;
1367 	if (entry->apic_id < NAPICID)
1368 		ID_TO_IO(entry->apic_id) = apic;
1369 
1370 	return 1;
1371 }
1372 
1373 
1374 static int
1375 lookup_bus_type(char *name)
1376 {
1377 	int     x;
1378 
1379 	for (x = 0; x < MAX_BUSTYPE; ++x)
1380 		if (strcmp(bus_type_table[x].name, name) == 0)
1381 			return bus_type_table[x].type;
1382 
1383 	return UNKNOWN_BUSTYPE;
1384 }
1385 
1386 
1387 static int
1388 int_entry(int_entry_ptr entry, int intr)
1389 {
1390 	int apic;
1391 
1392 	io_apic_ints[intr].int_type = entry->int_type;
1393 	io_apic_ints[intr].int_flags = entry->int_flags;
1394 	io_apic_ints[intr].src_bus_id = entry->src_bus_id;
1395 	io_apic_ints[intr].src_bus_irq = entry->src_bus_irq;
1396 	if (entry->dst_apic_id == 255) {
1397 		/* This signal goes to all IO APICS.  Select an IO APIC
1398 		   with sufficient number of interrupt pins */
1399 		for (apic = 0; apic < mp_napics; apic++)
1400 			if (((io_apic_read(apic, IOAPIC_VER) &
1401 			      IOART_VER_MAXREDIR) >> MAXREDIRSHIFT) >=
1402 			    entry->dst_apic_int)
1403 				break;
1404 		if (apic < mp_napics)
1405 			io_apic_ints[intr].dst_apic_id = IO_TO_ID(apic);
1406 		else
1407 			io_apic_ints[intr].dst_apic_id = entry->dst_apic_id;
1408 	} else
1409 		io_apic_ints[intr].dst_apic_id = entry->dst_apic_id;
1410 	io_apic_ints[intr].dst_apic_int = entry->dst_apic_int;
1411 
1412 	return 1;
1413 }
1414 
1415 
1416 static int
1417 apic_int_is_bus_type(int intr, int bus_type)
1418 {
1419 	int     bus;
1420 
1421 	for (bus = 0; bus < mp_nbusses; ++bus)
1422 		if ((bus_data[bus].bus_id == io_apic_ints[intr].src_bus_id)
1423 		    && ((int) bus_data[bus].bus_type == bus_type))
1424 			return 1;
1425 
1426 	return 0;
1427 }
1428 
1429 
1430 /*
1431  * Given a traditional ISA INT mask, return an APIC mask.
1432  */
1433 u_int
1434 isa_apic_mask(u_int isa_mask)
1435 {
1436 	int isa_irq;
1437 	int apic_pin;
1438 
1439 #if defined(SKIP_IRQ15_REDIRECT)
1440 	if (isa_mask == (1 << 15)) {
1441 		printf("skipping ISA IRQ15 redirect\n");
1442 		return isa_mask;
1443 	}
1444 #endif  /* SKIP_IRQ15_REDIRECT */
1445 
1446 	isa_irq = ffs(isa_mask);		/* find its bit position */
1447 	if (isa_irq == 0)			/* doesn't exist */
1448 		return 0;
1449 	--isa_irq;				/* make it zero based */
1450 
1451 	apic_pin = isa_apic_irq(isa_irq);	/* look for APIC connection */
1452 	if (apic_pin == -1)
1453 		return 0;
1454 
1455 	return (1 << apic_pin);			/* convert pin# to a mask */
1456 }
1457 
1458 
1459 /*
1460  * Determine which APIC pin an ISA/EISA INT is attached to.
1461  */
1462 #define INTTYPE(I)	(io_apic_ints[(I)].int_type)
1463 #define INTPIN(I)	(io_apic_ints[(I)].dst_apic_int)
1464 #define INTIRQ(I)	(io_apic_ints[(I)].int_vector)
1465 #define INTAPIC(I)	(ID_TO_IO(io_apic_ints[(I)].dst_apic_id))
1466 
1467 #define SRCBUSIRQ(I)	(io_apic_ints[(I)].src_bus_irq)
1468 int
1469 isa_apic_irq(int isa_irq)
1470 {
1471 	int     intr;
1472 
1473 	for (intr = 0; intr < nintrs; ++intr) {		/* check each record */
1474 		if (INTTYPE(intr) == 0) {		/* standard INT */
1475 			if (SRCBUSIRQ(intr) == isa_irq) {
1476 				if (apic_int_is_bus_type(intr, ISA) ||
1477 			            apic_int_is_bus_type(intr, EISA))
1478 					return INTIRQ(intr);	/* found */
1479 			}
1480 		}
1481 	}
1482 	return -1;					/* NOT found */
1483 }
1484 
1485 
1486 /*
1487  * Determine which APIC pin a PCI INT is attached to.
1488  */
1489 #define SRCBUSID(I)	(io_apic_ints[(I)].src_bus_id)
1490 #define SRCBUSDEVICE(I)	((io_apic_ints[(I)].src_bus_irq >> 2) & 0x1f)
1491 #define SRCBUSLINE(I)	(io_apic_ints[(I)].src_bus_irq & 0x03)
1492 int
1493 pci_apic_irq(int pciBus, int pciDevice, int pciInt)
1494 {
1495 	int     intr;
1496 
1497 	--pciInt;					/* zero based */
1498 
1499 	for (intr = 0; intr < nintrs; ++intr)		/* check each record */
1500 		if ((INTTYPE(intr) == 0)		/* standard INT */
1501 		    && (SRCBUSID(intr) == pciBus)
1502 		    && (SRCBUSDEVICE(intr) == pciDevice)
1503 		    && (SRCBUSLINE(intr) == pciInt))	/* a candidate IRQ */
1504 			if (apic_int_is_bus_type(intr, PCI))
1505 				return INTIRQ(intr);	/* exact match */
1506 
1507 	return -1;					/* NOT found */
1508 }
1509 
1510 int
1511 next_apic_irq(int irq)
1512 {
1513 	int intr, ointr;
1514 	int bus, bustype;
1515 
1516 	bus = 0;
1517 	bustype = 0;
1518 	for (intr = 0; intr < nintrs; intr++) {
1519 		if (INTIRQ(intr) != irq || INTTYPE(intr) != 0)
1520 			continue;
1521 		bus = SRCBUSID(intr);
1522 		bustype = apic_bus_type(bus);
1523 		if (bustype != ISA &&
1524 		    bustype != EISA &&
1525 		    bustype != PCI)
1526 			continue;
1527 		break;
1528 	}
1529 	if (intr >= nintrs) {
1530 		return -1;
1531 	}
1532 	for (ointr = intr + 1; ointr < nintrs; ointr++) {
1533 		if (INTTYPE(ointr) != 0)
1534 			continue;
1535 		if (bus != SRCBUSID(ointr))
1536 			continue;
1537 		if (bustype == PCI) {
1538 			if (SRCBUSDEVICE(intr) != SRCBUSDEVICE(ointr))
1539 				continue;
1540 			if (SRCBUSLINE(intr) != SRCBUSLINE(ointr))
1541 				continue;
1542 		}
1543 		if (bustype == ISA || bustype == EISA) {
1544 			if (SRCBUSIRQ(intr) != SRCBUSIRQ(ointr))
1545 				continue;
1546 		}
1547 		if (INTPIN(intr) == INTPIN(ointr))
1548 			continue;
1549 		break;
1550 	}
1551 	if (ointr >= nintrs) {
1552 		return -1;
1553 	}
1554 	return INTIRQ(ointr);
1555 }
1556 #undef SRCBUSLINE
1557 #undef SRCBUSDEVICE
1558 #undef SRCBUSID
1559 #undef SRCBUSIRQ
1560 
1561 #undef INTPIN
1562 #undef INTIRQ
1563 #undef INTAPIC
1564 #undef INTTYPE
1565 
1566 
1567 /*
1568  * Reprogram the MB chipset to NOT redirect an ISA INTerrupt.
1569  *
1570  * XXX FIXME:
1571  *  Exactly what this means is unclear at this point.  It is a solution
1572  *  for motherboards that redirect the MBIRQ0 pin.  Generically a motherboard
1573  *  could route any of the ISA INTs to upper (>15) IRQ values.  But most would
1574  *  NOT be redirected via MBIRQ0, thus "undirect()ing" them would NOT be an
1575  *  option.
1576  */
1577 int
1578 undirect_isa_irq(int rirq)
1579 {
1580 #if defined(READY)
1581 	if (bootverbose)
1582 	    printf("Freeing redirected ISA irq %d.\n", rirq);
1583 	/** FIXME: tickle the MB redirector chip */
1584 	return ???;
1585 #else
1586 	if (bootverbose)
1587 	    printf("Freeing (NOT implemented) redirected ISA irq %d.\n", rirq);
1588 	return 0;
1589 #endif  /* READY */
1590 }
1591 
1592 
1593 /*
1594  * Reprogram the MB chipset to NOT redirect a PCI INTerrupt
1595  */
1596 int
1597 undirect_pci_irq(int rirq)
1598 {
1599 #if defined(READY)
1600 	if (bootverbose)
1601 		printf("Freeing redirected PCI irq %d.\n", rirq);
1602 
1603 	/** FIXME: tickle the MB redirector chip */
1604 	return ???;
1605 #else
1606 	if (bootverbose)
1607 		printf("Freeing (NOT implemented) redirected PCI irq %d.\n",
1608 		       rirq);
1609 	return 0;
1610 #endif  /* READY */
1611 }
1612 
1613 
1614 /*
1615  * given a bus ID, return:
1616  *  the bus type if found
1617  *  -1 if NOT found
1618  */
1619 int
1620 apic_bus_type(int id)
1621 {
1622 	int     x;
1623 
1624 	for (x = 0; x < mp_nbusses; ++x)
1625 		if (bus_data[x].bus_id == id)
1626 			return bus_data[x].bus_type;
1627 
1628 	return -1;
1629 }
1630 
1631 
1632 /*
1633  * given a LOGICAL APIC# and pin#, return:
1634  *  the associated src bus ID if found
1635  *  -1 if NOT found
1636  */
1637 int
1638 apic_src_bus_id(int apic, int pin)
1639 {
1640 	int     x;
1641 
1642 	/* search each of the possible INTerrupt sources */
1643 	for (x = 0; x < nintrs; ++x)
1644 		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1645 		    (pin == io_apic_ints[x].dst_apic_int))
1646 			return (io_apic_ints[x].src_bus_id);
1647 
1648 	return -1;		/* NOT found */
1649 }
1650 
1651 
1652 /*
1653  * given a LOGICAL APIC# and pin#, return:
1654  *  the associated src bus IRQ if found
1655  *  -1 if NOT found
1656  */
1657 int
1658 apic_src_bus_irq(int apic, int pin)
1659 {
1660 	int     x;
1661 
1662 	for (x = 0; x < nintrs; x++)
1663 		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1664 		    (pin == io_apic_ints[x].dst_apic_int))
1665 			return (io_apic_ints[x].src_bus_irq);
1666 
1667 	return -1;		/* NOT found */
1668 }
1669 
1670 
1671 /*
1672  * given a LOGICAL APIC# and pin#, return:
1673  *  the associated INTerrupt type if found
1674  *  -1 if NOT found
1675  */
1676 int
1677 apic_int_type(int apic, int pin)
1678 {
1679 	int     x;
1680 
1681 	/* search each of the possible INTerrupt sources */
1682 	for (x = 0; x < nintrs; ++x)
1683 		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1684 		    (pin == io_apic_ints[x].dst_apic_int))
1685 			return (io_apic_ints[x].int_type);
1686 
1687 	return -1;		/* NOT found */
1688 }
1689 
1690 int
1691 apic_irq(int apic, int pin)
1692 {
1693 	int x;
1694 	int res;
1695 
1696 	for (x = 0; x < nintrs; ++x)
1697 		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1698 		    (pin == io_apic_ints[x].dst_apic_int)) {
1699 			res = io_apic_ints[x].int_vector;
1700 			if (res == 0xff)
1701 				return -1;
1702 			if (apic != int_to_apicintpin[res].ioapic)
1703 				panic("apic_irq: inconsistent table");
1704 			if (pin != int_to_apicintpin[res].int_pin)
1705 				panic("apic_irq inconsistent table (2)");
1706 			return res;
1707 		}
1708 	return -1;
1709 }
1710 
1711 
1712 /*
1713  * given a LOGICAL APIC# and pin#, return:
1714  *  the associated trigger mode if found
1715  *  -1 if NOT found
1716  */
1717 int
1718 apic_trigger(int apic, int pin)
1719 {
1720 	int     x;
1721 
1722 	/* search each of the possible INTerrupt sources */
1723 	for (x = 0; x < nintrs; ++x)
1724 		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1725 		    (pin == io_apic_ints[x].dst_apic_int))
1726 			return ((io_apic_ints[x].int_flags >> 2) & 0x03);
1727 
1728 	return -1;		/* NOT found */
1729 }
1730 
1731 
1732 /*
1733  * given a LOGICAL APIC# and pin#, return:
1734  *  the associated 'active' level if found
1735  *  -1 if NOT found
1736  */
1737 int
1738 apic_polarity(int apic, int pin)
1739 {
1740 	int     x;
1741 
1742 	/* search each of the possible INTerrupt sources */
1743 	for (x = 0; x < nintrs; ++x)
1744 		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1745 		    (pin == io_apic_ints[x].dst_apic_int))
1746 			return (io_apic_ints[x].int_flags & 0x03);
1747 
1748 	return -1;		/* NOT found */
1749 }
1750 
1751 
1752 /*
1753  * set data according to MP defaults
1754  * FIXME: probably not complete yet...
1755  */
1756 static void
1757 default_mp_table(int type)
1758 {
1759 	int     ap_cpu_id;
1760 #if defined(APIC_IO)
1761 	int     io_apic_id;
1762 	int     pin;
1763 #endif	/* APIC_IO */
1764 
1765 #if 0
1766 	printf("  MP default config type: %d\n", type);
1767 	switch (type) {
1768 	case 1:
1769 		printf("   bus: ISA, APIC: 82489DX\n");
1770 		break;
1771 	case 2:
1772 		printf("   bus: EISA, APIC: 82489DX\n");
1773 		break;
1774 	case 3:
1775 		printf("   bus: EISA, APIC: 82489DX\n");
1776 		break;
1777 	case 4:
1778 		printf("   bus: MCA, APIC: 82489DX\n");
1779 		break;
1780 	case 5:
1781 		printf("   bus: ISA+PCI, APIC: Integrated\n");
1782 		break;
1783 	case 6:
1784 		printf("   bus: EISA+PCI, APIC: Integrated\n");
1785 		break;
1786 	case 7:
1787 		printf("   bus: MCA+PCI, APIC: Integrated\n");
1788 		break;
1789 	default:
1790 		printf("   future type\n");
1791 		break;
1792 		/* NOTREACHED */
1793 	}
1794 #endif	/* 0 */
1795 
1796 	boot_cpu_id = (lapic.id & APIC_ID_MASK) >> 24;
1797 	ap_cpu_id = (boot_cpu_id == 0) ? 1 : 0;
1798 
1799 	/* BSP */
1800 	CPU_TO_ID(0) = boot_cpu_id;
1801 	ID_TO_CPU(boot_cpu_id) = 0;
1802 
1803 	/* one and only AP */
1804 	CPU_TO_ID(1) = ap_cpu_id;
1805 	ID_TO_CPU(ap_cpu_id) = 1;
1806 
1807 #if defined(APIC_IO)
1808 	/* one and only IO APIC */
1809 	io_apic_id = (io_apic_read(0, IOAPIC_ID) & APIC_ID_MASK) >> 24;
1810 
1811 	/*
1812 	 * sanity check, refer to MP spec section 3.6.6, last paragraph
1813 	 * necessary as some hardware isn't properly setting up the IO APIC
1814 	 */
1815 #if defined(REALLY_ANAL_IOAPICID_VALUE)
1816 	if (io_apic_id != 2) {
1817 #else
1818 	if ((io_apic_id == 0) || (io_apic_id == 1) || (io_apic_id == 15)) {
1819 #endif	/* REALLY_ANAL_IOAPICID_VALUE */
1820 		io_apic_set_id(0, 2);
1821 		io_apic_id = 2;
1822 	}
1823 	IO_TO_ID(0) = io_apic_id;
1824 	ID_TO_IO(io_apic_id) = 0;
1825 #endif	/* APIC_IO */
1826 
1827 	/* fill out bus entries */
1828 	switch (type) {
1829 	case 1:
1830 	case 2:
1831 	case 3:
1832 	case 4:
1833 	case 5:
1834 	case 6:
1835 	case 7:
1836 		bus_data[0].bus_id = default_data[type - 1][1];
1837 		bus_data[0].bus_type = default_data[type - 1][2];
1838 		bus_data[1].bus_id = default_data[type - 1][3];
1839 		bus_data[1].bus_type = default_data[type - 1][4];
1840 		break;
1841 
1842 	/* case 4: case 7:		   MCA NOT supported */
1843 	default:		/* illegal/reserved */
1844 		panic("BAD default MP config: %d", type);
1845 		/* NOTREACHED */
1846 	}
1847 
1848 #if defined(APIC_IO)
1849 	/* general cases from MP v1.4, table 5-2 */
1850 	for (pin = 0; pin < 16; ++pin) {
1851 		io_apic_ints[pin].int_type = 0;
1852 		io_apic_ints[pin].int_flags = 0x05;	/* edge/active-hi */
1853 		io_apic_ints[pin].src_bus_id = 0;
1854 		io_apic_ints[pin].src_bus_irq = pin;	/* IRQ2 caught below */
1855 		io_apic_ints[pin].dst_apic_id = io_apic_id;
1856 		io_apic_ints[pin].dst_apic_int = pin;	/* 1-to-1 */
1857 	}
1858 
1859 	/* special cases from MP v1.4, table 5-2 */
1860 	if (type == 2) {
1861 		io_apic_ints[2].int_type = 0xff;	/* N/C */
1862 		io_apic_ints[13].int_type = 0xff;	/* N/C */
1863 #if !defined(APIC_MIXED_MODE)
1864 		/** FIXME: ??? */
1865 		panic("sorry, can't support type 2 default yet");
1866 #endif	/* APIC_MIXED_MODE */
1867 	}
1868 	else
1869 		io_apic_ints[2].src_bus_irq = 0;	/* ISA IRQ0 is on APIC INT 2 */
1870 
1871 	if (type == 7)
1872 		io_apic_ints[0].int_type = 0xff;	/* N/C */
1873 	else
1874 		io_apic_ints[0].int_type = 3;	/* vectored 8259 */
1875 #endif	/* APIC_IO */
1876 }
1877 
1878 
1879 /*
1880  * initialize all the SMP locks
1881  */
1882 
1883 /* critical region around IO APIC, apic_imen */
1884 struct simplelock	imen_lock;
1885 
1886 /* critical region around splxx(), cpl, cml, cil, ipending */
1887 struct simplelock	cpl_lock;
1888 
1889 /* Make FAST_INTR() routines sequential */
1890 struct simplelock	fast_intr_lock;
1891 
1892 /* critical region around INTR() routines */
1893 struct simplelock	intr_lock;
1894 
1895 /* lock region used by kernel profiling */
1896 struct simplelock	mcount_lock;
1897 
1898 #ifdef USE_COMLOCK
1899 /* locks com (tty) data/hardware accesses: a FASTINTR() */
1900 struct simplelock	com_lock;
1901 #endif /* USE_COMLOCK */
1902 
1903 /* lock around the MP rendezvous */
1904 static struct simplelock smp_rv_lock;
1905 
1906 /* only 1 CPU can panic at a time :) */
1907 struct simplelock	panic_lock;
1908 
1909 static void
1910 init_locks(void)
1911 {
1912 #if defined(APIC_INTR_DIAGNOSTIC) && defined(APIC_INTR_DIAGNOSTIC_IRQ)
1913 	s_lock_init((struct simplelock*)&apic_itrace_debuglock);
1914 #endif
1915 
1916 	s_lock_init((struct simplelock*)&mcount_lock);
1917 
1918 	s_lock_init((struct simplelock*)&fast_intr_lock);
1919 	s_lock_init((struct simplelock*)&intr_lock);
1920 	s_lock_init((struct simplelock*)&imen_lock);
1921 	s_lock_init((struct simplelock*)&cpl_lock);
1922 	s_lock_init(&smp_rv_lock);
1923 	s_lock_init(&panic_lock);
1924 
1925 #ifdef USE_COMLOCK
1926 	s_lock_init((struct simplelock*)&com_lock);
1927 #endif /* USE_COMLOCK */
1928 
1929 	s_lock_init(&ap_boot_lock);
1930 }
1931 
1932 /*
1933  * start each AP in our list
1934  */
1935 static int
1936 start_all_aps(u_int boot_addr)
1937 {
1938 	int     x, i, pg;
1939 	u_char  mpbiosreason;
1940 	u_long  mpbioswarmvec;
1941 	struct globaldata *gd;
1942 	char *stack;
1943 
1944 	POSTCODE(START_ALL_APS_POST);
1945 
1946 	/* initialize BSP's local APIC */
1947 	apic_initialize();
1948 	bsp_apic_ready = 1;
1949 
1950 	/* install the AP 1st level boot code */
1951 	install_ap_tramp(boot_addr);
1952 
1953 
1954 	/* save the current value of the warm-start vector */
1955 	mpbioswarmvec = *((u_long *) WARMBOOT_OFF);
1956 #ifndef PC98
1957 	outb(CMOS_REG, BIOS_RESET);
1958 	mpbiosreason = inb(CMOS_DATA);
1959 #endif
1960 
1961 	/* record BSP in CPU map */
1962 	all_cpus = 1;
1963 
1964 	/* set up 0 -> 4MB P==V mapping for AP boot */
1965 	*(int *)PTD = PG_V | PG_RW | ((uintptr_t)(void *)KPTphys & PG_FRAME);
1966 	invltlb();
1967 
1968 	/* start each AP */
1969 	for (x = 1; x <= mp_naps; ++x) {
1970 
1971 		/* This is a bit verbose, it will go away soon.  */
1972 
1973 		/* first page of AP's private space */
1974 		pg = x * i386_btop(sizeof(struct privatespace));
1975 
1976 		/* allocate a new private data page */
1977 		gd = (struct globaldata *)kmem_alloc(kernel_map, PAGE_SIZE);
1978 
1979 		/* wire it into the private page table page */
1980 		SMPpt[pg] = (pt_entry_t)(PG_V | PG_RW | vtophys(gd));
1981 
1982 		/* allocate and set up an idle stack data page */
1983 		stack = (char *)kmem_alloc(kernel_map, UPAGES*PAGE_SIZE);
1984 		for (i = 0; i < UPAGES; i++)
1985 			SMPpt[pg + 5 + i] = (pt_entry_t)
1986 			    (PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack));
1987 
1988 		SMPpt[pg + 1] = 0;		/* *prv_CMAP1 */
1989 		SMPpt[pg + 2] = 0;		/* *prv_CMAP2 */
1990 		SMPpt[pg + 3] = 0;		/* *prv_CMAP3 */
1991 		SMPpt[pg + 4] = 0;		/* *prv_PMAP1 */
1992 
1993 		/* prime data page for it to use */
1994 		SLIST_INSERT_HEAD(&cpuhead, gd, gd_allcpu);
1995 		gd->gd_cpuid = x;
1996 		gd->gd_cpu_lockid = x << 24;
1997 		gd->gd_prv_CMAP1 = &SMPpt[pg + 1];
1998 		gd->gd_prv_CMAP2 = &SMPpt[pg + 2];
1999 		gd->gd_prv_CMAP3 = &SMPpt[pg + 3];
2000 		gd->gd_prv_PMAP1 = &SMPpt[pg + 4];
2001 		gd->gd_prv_CADDR1 = SMP_prvspace[x].CPAGE1;
2002 		gd->gd_prv_CADDR2 = SMP_prvspace[x].CPAGE2;
2003 		gd->gd_prv_CADDR3 = SMP_prvspace[x].CPAGE3;
2004 		gd->gd_prv_PADDR1 = (unsigned *)SMP_prvspace[x].PPAGE1;
2005 
2006 		/* setup a vector to our boot code */
2007 		*((volatile u_short *) WARMBOOT_OFF) = WARMBOOT_TARGET;
2008 		*((volatile u_short *) WARMBOOT_SEG) = (boot_addr >> 4);
2009 #ifndef PC98
2010 		outb(CMOS_REG, BIOS_RESET);
2011 		outb(CMOS_DATA, BIOS_WARM);	/* 'warm-start' */
2012 #endif
2013 
2014 		bootSTK = &SMP_prvspace[x].idlestack[UPAGES*PAGE_SIZE];
2015 		bootAP = x;
2016 
2017 		/* attempt to start the Application Processor */
2018 		CHECK_INIT(99);	/* setup checkpoints */
2019 		if (!start_ap(x, boot_addr)) {
2020 			printf("AP #%d (PHY# %d) failed!\n", x, CPU_TO_ID(x));
2021 			CHECK_PRINT("trace");	/* show checkpoints */
2022 			/* better panic as the AP may be running loose */
2023 			printf("panic y/n? [y] ");
2024 			if (cngetc() != 'n')
2025 				panic("bye-bye");
2026 		}
2027 		CHECK_PRINT("trace");		/* show checkpoints */
2028 
2029 		/* record its version info */
2030 		cpu_apic_versions[x] = cpu_apic_versions[0];
2031 
2032 		all_cpus |= (1 << x);		/* record AP in CPU map */
2033 	}
2034 
2035 	/* build our map of 'other' CPUs */
2036 	other_cpus = all_cpus & ~(1 << cpuid);
2037 
2038 	/* fill in our (BSP) APIC version */
2039 	cpu_apic_versions[0] = lapic.version;
2040 
2041 	/* restore the warmstart vector */
2042 	*(u_long *) WARMBOOT_OFF = mpbioswarmvec;
2043 #ifndef PC98
2044 	outb(CMOS_REG, BIOS_RESET);
2045 	outb(CMOS_DATA, mpbiosreason);
2046 #endif
2047 
2048 	/*
2049 	 * Set up the idle context for the BSP.  Similar to above except
2050 	 * that some was done by locore, some by pmap.c and some is implicit
2051 	 * because the BSP is cpu#0 and the page is initially zero, and also
2052 	 * because we can refer to variables by name on the BSP..
2053 	 */
2054 
2055 	/* Allocate and setup BSP idle stack */
2056 	stack = (char *)kmem_alloc(kernel_map, UPAGES * PAGE_SIZE);
2057 	for (i = 0; i < UPAGES; i++)
2058 		SMPpt[5 + i] = (pt_entry_t)
2059 		    (PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack));
2060 
2061 	*(int *)PTD = 0;
2062 	pmap_set_opt();
2063 
2064 	/* number of APs actually started */
2065 	return mp_ncpus - 1;
2066 }
2067 
2068 
2069 /*
2070  * load the 1st level AP boot code into base memory.
2071  */
2072 
2073 /* targets for relocation */
2074 extern void bigJump(void);
2075 extern void bootCodeSeg(void);
2076 extern void bootDataSeg(void);
2077 extern void MPentry(void);
2078 extern u_int MP_GDT;
2079 extern u_int mp_gdtbase;
2080 
2081 static void
2082 install_ap_tramp(u_int boot_addr)
2083 {
2084 	int     x;
2085 	int     size = *(int *) ((u_long) & bootMP_size);
2086 	u_char *src = (u_char *) ((u_long) bootMP);
2087 	u_char *dst = (u_char *) boot_addr + KERNBASE;
2088 	u_int   boot_base = (u_int) bootMP;
2089 	u_int8_t *dst8;
2090 	u_int16_t *dst16;
2091 	u_int32_t *dst32;
2092 
2093 	POSTCODE(INSTALL_AP_TRAMP_POST);
2094 
2095 	for (x = 0; x < size; ++x)
2096 		*dst++ = *src++;
2097 
2098 	/*
2099 	 * modify addresses in code we just moved to basemem. unfortunately we
2100 	 * need fairly detailed info about mpboot.s for this to work.  changes
2101 	 * to mpboot.s might require changes here.
2102 	 */
2103 
2104 	/* boot code is located in KERNEL space */
2105 	dst = (u_char *) boot_addr + KERNBASE;
2106 
2107 	/* modify the lgdt arg */
2108 	dst32 = (u_int32_t *) (dst + ((u_int) & mp_gdtbase - boot_base));
2109 	*dst32 = boot_addr + ((u_int) & MP_GDT - boot_base);
2110 
2111 	/* modify the ljmp target for MPentry() */
2112 	dst32 = (u_int32_t *) (dst + ((u_int) bigJump - boot_base) + 1);
2113 	*dst32 = ((u_int) MPentry - KERNBASE);
2114 
2115 	/* modify the target for boot code segment */
2116 	dst16 = (u_int16_t *) (dst + ((u_int) bootCodeSeg - boot_base));
2117 	dst8 = (u_int8_t *) (dst16 + 1);
2118 	*dst16 = (u_int) boot_addr & 0xffff;
2119 	*dst8 = ((u_int) boot_addr >> 16) & 0xff;
2120 
2121 	/* modify the target for boot data segment */
2122 	dst16 = (u_int16_t *) (dst + ((u_int) bootDataSeg - boot_base));
2123 	dst8 = (u_int8_t *) (dst16 + 1);
2124 	*dst16 = (u_int) boot_addr & 0xffff;
2125 	*dst8 = ((u_int) boot_addr >> 16) & 0xff;
2126 }
2127 
2128 
2129 /*
2130  * this function starts the AP (application processor) identified
2131  * by the APIC ID 'physicalCpu'.  It does quite a "song and dance"
2132  * to accomplish this.  This is necessary because of the nuances
2133  * of the different hardware we might encounter.  It ain't pretty,
2134  * but it seems to work.
2135  */
2136 static int
2137 start_ap(int logical_cpu, u_int boot_addr)
2138 {
2139 	int     physical_cpu;
2140 	int     vector;
2141 	int     cpus;
2142 	u_long  icr_lo, icr_hi;
2143 
2144 	POSTCODE(START_AP_POST);
2145 
2146 	/* get the PHYSICAL APIC ID# */
2147 	physical_cpu = CPU_TO_ID(logical_cpu);
2148 
2149 	/* calculate the vector */
2150 	vector = (boot_addr >> 12) & 0xff;
2151 
2152 	/* used as a watchpoint to signal AP startup */
2153 	cpus = mp_ncpus;
2154 
2155 	/*
2156 	 * first we do an INIT/RESET IPI this INIT IPI might be run, reseting
2157 	 * and running the target CPU. OR this INIT IPI might be latched (P5
2158 	 * bug), CPU waiting for STARTUP IPI. OR this INIT IPI might be
2159 	 * ignored.
2160 	 */
2161 
2162 	/* setup the address for the target AP */
2163 	icr_hi = lapic.icr_hi & ~APIC_ID_MASK;
2164 	icr_hi |= (physical_cpu << 24);
2165 	lapic.icr_hi = icr_hi;
2166 
2167 	/* do an INIT IPI: assert RESET */
2168 	icr_lo = lapic.icr_lo & 0xfff00000;
2169 	lapic.icr_lo = icr_lo | 0x0000c500;
2170 
2171 	/* wait for pending status end */
2172 	while (lapic.icr_lo & APIC_DELSTAT_MASK)
2173 		 /* spin */ ;
2174 
2175 	/* do an INIT IPI: deassert RESET */
2176 	lapic.icr_lo = icr_lo | 0x00008500;
2177 
2178 	/* wait for pending status end */
2179 	u_sleep(10000);		/* wait ~10mS */
2180 	while (lapic.icr_lo & APIC_DELSTAT_MASK)
2181 		 /* spin */ ;
2182 
2183 	/*
2184 	 * next we do a STARTUP IPI: the previous INIT IPI might still be
2185 	 * latched, (P5 bug) this 1st STARTUP would then terminate
2186 	 * immediately, and the previously started INIT IPI would continue. OR
2187 	 * the previous INIT IPI has already run. and this STARTUP IPI will
2188 	 * run. OR the previous INIT IPI was ignored. and this STARTUP IPI
2189 	 * will run.
2190 	 */
2191 
2192 	/* do a STARTUP IPI */
2193 	lapic.icr_lo = icr_lo | 0x00000600 | vector;
2194 	while (lapic.icr_lo & APIC_DELSTAT_MASK)
2195 		 /* spin */ ;
2196 	u_sleep(200);		/* wait ~200uS */
2197 
2198 	/*
2199 	 * finally we do a 2nd STARTUP IPI: this 2nd STARTUP IPI should run IF
2200 	 * the previous STARTUP IPI was cancelled by a latched INIT IPI. OR
2201 	 * this STARTUP IPI will be ignored, as only ONE STARTUP IPI is
2202 	 * recognized after hardware RESET or INIT IPI.
2203 	 */
2204 
2205 	lapic.icr_lo = icr_lo | 0x00000600 | vector;
2206 	while (lapic.icr_lo & APIC_DELSTAT_MASK)
2207 		 /* spin */ ;
2208 	u_sleep(200);		/* wait ~200uS */
2209 
2210 	/* wait for it to start */
2211 	set_apic_timer(5000000);/* == 5 seconds */
2212 	while (read_apic_timer())
2213 		if (mp_ncpus > cpus)
2214 			return 1;	/* return SUCCESS */
2215 
2216 	return 0;		/* return FAILURE */
2217 }
2218 
2219 /*
2220  * Flush the TLB on all other CPU's
2221  *
2222  * XXX: Needs to handshake and wait for completion before proceding.
2223  */
2224 void
2225 smp_invltlb(void)
2226 {
2227 #if defined(APIC_IO)
2228 	if (smp_started && invltlb_ok)
2229 		all_but_self_ipi(XINVLTLB_OFFSET);
2230 #endif  /* APIC_IO */
2231 }
2232 
2233 void
2234 invlpg(u_int addr)
2235 {
2236 	__asm   __volatile("invlpg (%0)"::"r"(addr):"memory");
2237 
2238 	/* send a message to the other CPUs */
2239 	smp_invltlb();
2240 }
2241 
2242 void
2243 invltlb(void)
2244 {
2245 	u_long  temp;
2246 
2247 	/*
2248 	 * This should be implemented as load_cr3(rcr3()) when load_cr3() is
2249 	 * inlined.
2250 	 */
2251 	__asm __volatile("movl %%cr3, %0; movl %0, %%cr3":"=r"(temp) :: "memory");
2252 
2253 	/* send a message to the other CPUs */
2254 	smp_invltlb();
2255 }
2256 
2257 
2258 /*
2259  * When called the executing CPU will send an IPI to all other CPUs
2260  *  requesting that they halt execution.
2261  *
2262  * Usually (but not necessarily) called with 'other_cpus' as its arg.
2263  *
2264  *  - Signals all CPUs in map to stop.
2265  *  - Waits for each to stop.
2266  *
2267  * Returns:
2268  *  -1: error
2269  *   0: NA
2270  *   1: ok
2271  *
2272  * XXX FIXME: this is not MP-safe, needs a lock to prevent multiple CPUs
2273  *            from executing at same time.
2274  */
2275 int
2276 stop_cpus(u_int map)
2277 {
2278 	if (!smp_started)
2279 		return 0;
2280 
2281 	/* send the Xcpustop IPI to all CPUs in map */
2282 	selected_apic_ipi(map, XCPUSTOP_OFFSET, APIC_DELMODE_FIXED);
2283 
2284 	while ((stopped_cpus & map) != map)
2285 		/* spin */ ;
2286 
2287 	return 1;
2288 }
2289 
2290 
2291 /*
2292  * Called by a CPU to restart stopped CPUs.
2293  *
2294  * Usually (but not necessarily) called with 'stopped_cpus' as its arg.
2295  *
2296  *  - Signals all CPUs in map to restart.
2297  *  - Waits for each to restart.
2298  *
2299  * Returns:
2300  *  -1: error
2301  *   0: NA
2302  *   1: ok
2303  */
2304 int
2305 restart_cpus(u_int map)
2306 {
2307 	if (!smp_started)
2308 		return 0;
2309 
2310 	started_cpus = map;		/* signal other cpus to restart */
2311 
2312 	while ((stopped_cpus & map) != 0) /* wait for each to clear its bit */
2313 		/* spin */ ;
2314 
2315 	return 1;
2316 }
2317 
2318 int smp_active = 0;	/* are the APs allowed to run? */
2319 SYSCTL_INT(_machdep, OID_AUTO, smp_active, CTLFLAG_RW, &smp_active, 0, "");
2320 
2321 /* XXX maybe should be hw.ncpu */
2322 static int smp_cpus = 1;	/* how many cpu's running */
2323 SYSCTL_INT(_machdep, OID_AUTO, smp_cpus, CTLFLAG_RD, &smp_cpus, 0, "");
2324 
2325 int invltlb_ok = 0;	/* throttle smp_invltlb() till safe */
2326 SYSCTL_INT(_machdep, OID_AUTO, invltlb_ok, CTLFLAG_RW, &invltlb_ok, 0, "");
2327 
2328 /* Warning: Do not staticize.  Used from swtch.s */
2329 int do_page_zero_idle = 1; /* bzero pages for fun and profit in idleloop */
2330 SYSCTL_INT(_machdep, OID_AUTO, do_page_zero_idle, CTLFLAG_RW,
2331 	   &do_page_zero_idle, 0, "");
2332 
2333 /* Is forwarding of a interrupt to the CPU holding the ISR lock enabled ? */
2334 int forward_irq_enabled = 1;
2335 SYSCTL_INT(_machdep, OID_AUTO, forward_irq_enabled, CTLFLAG_RW,
2336 	   &forward_irq_enabled, 0, "");
2337 
2338 /* Enable forwarding of a signal to a process running on a different CPU */
2339 static int forward_signal_enabled = 1;
2340 SYSCTL_INT(_machdep, OID_AUTO, forward_signal_enabled, CTLFLAG_RW,
2341 	   &forward_signal_enabled, 0, "");
2342 
2343 /* Enable forwarding of roundrobin to all other cpus */
2344 static int forward_roundrobin_enabled = 1;
2345 SYSCTL_INT(_machdep, OID_AUTO, forward_roundrobin_enabled, CTLFLAG_RW,
2346 	   &forward_roundrobin_enabled, 0, "");
2347 
2348 /*
2349  * This is called once the rest of the system is up and running and we're
2350  * ready to let the AP's out of the pen.
2351  */
2352 void ap_init(void);
2353 
2354 void
2355 ap_init(void)
2356 {
2357 	u_int	apic_id;
2358 
2359 	/* lock against other AP's that are waking up */
2360 	s_lock(&ap_boot_lock);
2361 
2362 	/* BSP may have changed PTD while we're waiting for the lock */
2363 	cpu_invltlb();
2364 
2365 	smp_cpus++;
2366 
2367 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
2368 	lidt(&r_idt);
2369 #endif
2370 
2371 	/* Build our map of 'other' CPUs. */
2372 	other_cpus = all_cpus & ~(1 << cpuid);
2373 
2374 	printf("SMP: AP CPU #%d Launched!\n", cpuid);
2375 
2376 	/* set up CPU registers and state */
2377 	cpu_setregs();
2378 
2379 	/* set up FPU state on the AP */
2380 	npxinit(__INITIAL_NPXCW__);
2381 
2382 	/* A quick check from sanity claus */
2383 	apic_id = (apic_id_to_logical[(lapic.id & 0x0f000000) >> 24]);
2384 	if (cpuid != apic_id) {
2385 		printf("SMP: cpuid = %d\n", cpuid);
2386 		printf("SMP: apic_id = %d\n", apic_id);
2387 		printf("PTD[MPPTDI] = %p\n", (void *)PTD[MPPTDI]);
2388 		panic("cpuid mismatch! boom!!");
2389 	}
2390 
2391 	/* Init local apic for irq's */
2392 	apic_initialize();
2393 
2394 	/* Set memory range attributes for this CPU to match the BSP */
2395 	mem_range_AP_init();
2396 
2397 	/*
2398 	 * Activate smp_invltlb, although strictly speaking, this isn't
2399 	 * quite correct yet.  We should have a bitfield for cpus willing
2400 	 * to accept TLB flush IPI's or something and sync them.
2401 	 */
2402 	if (smp_cpus == mp_ncpus) {
2403 		invltlb_ok = 1;
2404 		smp_started = 1; /* enable IPI's, tlb shootdown, freezes etc */
2405 		smp_active = 1;	 /* historic */
2406 	}
2407 
2408 	/* let other AP's wake up now */
2409 	s_unlock(&ap_boot_lock);
2410 
2411 	/* wait until all the AP's are up */
2412 	while (smp_started == 0)
2413 		; /* nothing */
2414 
2415 	/*
2416 	 * Set curproc to our per-cpu idleproc so that mutexes have
2417 	 * something unique to lock with.
2418 	 */
2419 	PCPU_SET(curproc,idleproc);
2420 
2421 	microuptime(&switchtime);
2422 	switchticks = ticks;
2423 
2424 	/* ok, now grab sched_lock and enter the scheduler */
2425 	enable_intr();
2426 	mtx_enter(&sched_lock, MTX_SPIN);
2427 	cpu_throw();	/* doesn't return */
2428 
2429 	panic("scheduler returned us to ap_init");
2430 }
2431 
2432 #ifdef BETTER_CLOCK
2433 
2434 #define CHECKSTATE_USER	0
2435 #define CHECKSTATE_SYS	1
2436 #define CHECKSTATE_INTR	2
2437 
2438 /* Do not staticize.  Used from apic_vector.s */
2439 struct proc*	checkstate_curproc[MAXCPU];
2440 int		checkstate_cpustate[MAXCPU];
2441 u_long		checkstate_pc[MAXCPU];
2442 
2443 #define PC_TO_INDEX(pc, prof)				\
2444         ((int)(((u_quad_t)((pc) - (prof)->pr_off) *	\
2445             (u_quad_t)((prof)->pr_scale)) >> 16) & ~1)
2446 
2447 static void
2448 addupc_intr_forwarded(struct proc *p, int id, int *astmap)
2449 {
2450 	int i;
2451 	struct uprof *prof;
2452 	u_long pc;
2453 
2454 	pc = checkstate_pc[id];
2455 	prof = &p->p_stats->p_prof;
2456 	if (pc >= prof->pr_off &&
2457 	    (i = PC_TO_INDEX(pc, prof)) < prof->pr_size) {
2458 		if ((p->p_flag & P_OWEUPC) == 0) {
2459 			prof->pr_addr = pc;
2460 			prof->pr_ticks = 1;
2461 			p->p_flag |= P_OWEUPC;
2462 		}
2463 		*astmap |= (1 << id);
2464 	}
2465 }
2466 
2467 static void
2468 forwarded_statclock(int id, int pscnt, int *astmap)
2469 {
2470 	struct pstats *pstats;
2471 	long rss;
2472 	struct rusage *ru;
2473 	struct vmspace *vm;
2474 	int cpustate;
2475 	struct proc *p;
2476 #ifdef GPROF
2477 	register struct gmonparam *g;
2478 	int i;
2479 #endif
2480 
2481 	p = checkstate_curproc[id];
2482 	cpustate = checkstate_cpustate[id];
2483 
2484 	/* XXX */
2485 	if (p->p_ithd)
2486 		cpustate = CHECKSTATE_INTR;
2487 	else if (p == SMP_prvspace[id].globaldata.gd_idleproc)
2488 		cpustate = CHECKSTATE_SYS;
2489 
2490 	switch (cpustate) {
2491 	case CHECKSTATE_USER:
2492 		if (p->p_flag & P_PROFIL)
2493 			addupc_intr_forwarded(p, id, astmap);
2494 		if (pscnt > 1)
2495 			return;
2496 		p->p_uticks++;
2497 		if (p->p_nice > NZERO)
2498 			cp_time[CP_NICE]++;
2499 		else
2500 			cp_time[CP_USER]++;
2501 		break;
2502 	case CHECKSTATE_SYS:
2503 #ifdef GPROF
2504 		/*
2505 		 * Kernel statistics are just like addupc_intr, only easier.
2506 		 */
2507 		g = &_gmonparam;
2508 		if (g->state == GMON_PROF_ON) {
2509 			i = checkstate_pc[id] - g->lowpc;
2510 			if (i < g->textsize) {
2511 				i /= HISTFRACTION * sizeof(*g->kcount);
2512 				g->kcount[i]++;
2513 			}
2514 		}
2515 #endif
2516 		if (pscnt > 1)
2517 			return;
2518 
2519 		p->p_sticks++;
2520 		if (p == SMP_prvspace[id].globaldata.gd_idleproc)
2521 			cp_time[CP_IDLE]++;
2522 		else
2523 			cp_time[CP_SYS]++;
2524 		break;
2525 	case CHECKSTATE_INTR:
2526 	default:
2527 #ifdef GPROF
2528 		/*
2529 		 * Kernel statistics are just like addupc_intr, only easier.
2530 		 */
2531 		g = &_gmonparam;
2532 		if (g->state == GMON_PROF_ON) {
2533 			i = checkstate_pc[id] - g->lowpc;
2534 			if (i < g->textsize) {
2535 				i /= HISTFRACTION * sizeof(*g->kcount);
2536 				g->kcount[i]++;
2537 			}
2538 		}
2539 #endif
2540 		if (pscnt > 1)
2541 			return;
2542 		if (p)
2543 			p->p_iticks++;
2544 		cp_time[CP_INTR]++;
2545 	}
2546 	schedclock(p);
2547 
2548 	/* Update resource usage integrals and maximums. */
2549 	if ((pstats = p->p_stats) != NULL &&
2550 	    (ru = &pstats->p_ru) != NULL &&
2551 	    (vm = p->p_vmspace) != NULL) {
2552 		ru->ru_ixrss += pgtok(vm->vm_tsize);
2553 		ru->ru_idrss += pgtok(vm->vm_dsize);
2554 		ru->ru_isrss += pgtok(vm->vm_ssize);
2555 		rss = pgtok(vmspace_resident_count(vm));
2556 		if (ru->ru_maxrss < rss)
2557 			ru->ru_maxrss = rss;
2558 	}
2559 }
2560 
2561 void
2562 forward_statclock(int pscnt)
2563 {
2564 	int map;
2565 	int id;
2566 	int i;
2567 
2568 	/* Kludge. We don't yet have separate locks for the interrupts
2569 	 * and the kernel. This means that we cannot let the other processors
2570 	 * handle complex interrupts while inhibiting them from entering
2571 	 * the kernel in a non-interrupt context.
2572 	 *
2573 	 * What we can do, without changing the locking mechanisms yet,
2574 	 * is letting the other processors handle a very simple interrupt
2575 	 * (wich determines the processor states), and do the main
2576 	 * work ourself.
2577 	 */
2578 
2579 	if (!smp_started || !invltlb_ok || cold || panicstr)
2580 		return;
2581 
2582 	/* Step 1: Probe state   (user, cpu, interrupt, spinlock, idle ) */
2583 
2584 	map = other_cpus & ~stopped_cpus ;
2585 	checkstate_probed_cpus = 0;
2586 	if (map != 0)
2587 		selected_apic_ipi(map,
2588 				  XCPUCHECKSTATE_OFFSET, APIC_DELMODE_FIXED);
2589 
2590 	i = 0;
2591 	while (checkstate_probed_cpus != map) {
2592 		/* spin */
2593 		i++;
2594 		if (i == 100000) {
2595 #ifdef BETTER_CLOCK_DIAGNOSTIC
2596 			printf("forward_statclock: checkstate %x\n",
2597 			       checkstate_probed_cpus);
2598 #endif
2599 			break;
2600 		}
2601 	}
2602 
2603 	/*
2604 	 * Step 2: walk through other processors processes, update ticks and
2605 	 * profiling info.
2606 	 */
2607 
2608 	map = 0;
2609 	for (id = 0; id < mp_ncpus; id++) {
2610 		if (id == cpuid)
2611 			continue;
2612 		if (((1 << id) & checkstate_probed_cpus) == 0)
2613 			continue;
2614 		forwarded_statclock(id, pscnt, &map);
2615 	}
2616 	if (map != 0) {
2617 		checkstate_need_ast |= map;
2618 		selected_apic_ipi(map, XCPUAST_OFFSET, APIC_DELMODE_FIXED);
2619 		i = 0;
2620 		while ((checkstate_need_ast & map) != 0) {
2621 			/* spin */
2622 			i++;
2623 			if (i > 100000) {
2624 #ifdef BETTER_CLOCK_DIAGNOSTIC
2625 				printf("forward_statclock: dropped ast 0x%x\n",
2626 				       checkstate_need_ast & map);
2627 #endif
2628 				break;
2629 			}
2630 		}
2631 	}
2632 }
2633 
2634 void
2635 forward_hardclock(int pscnt)
2636 {
2637 	int map;
2638 	int id;
2639 	struct proc *p;
2640 	struct pstats *pstats;
2641 	int i;
2642 
2643 	/* Kludge. We don't yet have separate locks for the interrupts
2644 	 * and the kernel. This means that we cannot let the other processors
2645 	 * handle complex interrupts while inhibiting them from entering
2646 	 * the kernel in a non-interrupt context.
2647 	 *
2648 	 * What we can do, without changing the locking mechanisms yet,
2649 	 * is letting the other processors handle a very simple interrupt
2650 	 * (wich determines the processor states), and do the main
2651 	 * work ourself.
2652 	 */
2653 
2654 	if (!smp_started || !invltlb_ok || cold || panicstr)
2655 		return;
2656 
2657 	/* Step 1: Probe state   (user, cpu, interrupt, spinlock, idle) */
2658 
2659 	map = other_cpus & ~stopped_cpus ;
2660 	checkstate_probed_cpus = 0;
2661 	if (map != 0)
2662 		selected_apic_ipi(map,
2663 				  XCPUCHECKSTATE_OFFSET, APIC_DELMODE_FIXED);
2664 
2665 	i = 0;
2666 	while (checkstate_probed_cpus != map) {
2667 		/* spin */
2668 		i++;
2669 		if (i == 100000) {
2670 #ifdef BETTER_CLOCK_DIAGNOSTIC
2671 			printf("forward_hardclock: checkstate %x\n",
2672 			       checkstate_probed_cpus);
2673 #endif
2674 			break;
2675 		}
2676 	}
2677 
2678 	/*
2679 	 * Step 2: walk through other processors processes, update virtual
2680 	 * timer and profiling timer. If stathz == 0, also update ticks and
2681 	 * profiling info.
2682 	 */
2683 
2684 	map = 0;
2685 	for (id = 0; id < mp_ncpus; id++) {
2686 		if (id == cpuid)
2687 			continue;
2688 		if (((1 << id) & checkstate_probed_cpus) == 0)
2689 			continue;
2690 		p = checkstate_curproc[id];
2691 		if (p) {
2692 			pstats = p->p_stats;
2693 			if (checkstate_cpustate[id] == CHECKSTATE_USER &&
2694 			    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) &&
2695 			    itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) {
2696 				psignal(p, SIGVTALRM);
2697 				map |= (1 << id);
2698 			}
2699 			if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) &&
2700 			    itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) {
2701 				psignal(p, SIGPROF);
2702 				map |= (1 << id);
2703 			}
2704 		}
2705 		if (stathz == 0) {
2706 			forwarded_statclock( id, pscnt, &map);
2707 		}
2708 	}
2709 	if (map != 0) {
2710 		checkstate_need_ast |= map;
2711 		selected_apic_ipi(map, XCPUAST_OFFSET, APIC_DELMODE_FIXED);
2712 		i = 0;
2713 		while ((checkstate_need_ast & map) != 0) {
2714 			/* spin */
2715 			i++;
2716 			if (i > 100000) {
2717 #ifdef BETTER_CLOCK_DIAGNOSTIC
2718 				printf("forward_hardclock: dropped ast 0x%x\n",
2719 				       checkstate_need_ast & map);
2720 #endif
2721 				break;
2722 			}
2723 		}
2724 	}
2725 }
2726 
2727 #endif /* BETTER_CLOCK */
2728 
2729 void
2730 forward_signal(struct proc *p)
2731 {
2732 	int map;
2733 	int id;
2734 	int i;
2735 
2736 	/* Kludge. We don't yet have separate locks for the interrupts
2737 	 * and the kernel. This means that we cannot let the other processors
2738 	 * handle complex interrupts while inhibiting them from entering
2739 	 * the kernel in a non-interrupt context.
2740 	 *
2741 	 * What we can do, without changing the locking mechanisms yet,
2742 	 * is letting the other processors handle a very simple interrupt
2743 	 * (wich determines the processor states), and do the main
2744 	 * work ourself.
2745 	 */
2746 
2747 	if (!smp_started || !invltlb_ok || cold || panicstr)
2748 		return;
2749 	if (!forward_signal_enabled)
2750 		return;
2751 	while (1) {
2752 		if (p->p_stat != SRUN)
2753 			return;
2754 		id = p->p_oncpu;
2755 		if (id == 0xff)
2756 			return;
2757 		map = (1<<id);
2758 		checkstate_need_ast |= map;
2759 		selected_apic_ipi(map, XCPUAST_OFFSET, APIC_DELMODE_FIXED);
2760 		i = 0;
2761 		while ((checkstate_need_ast & map) != 0) {
2762 			/* spin */
2763 			i++;
2764 			if (i > 100000) {
2765 #if 0
2766 				printf("forward_signal: dropped ast 0x%x\n",
2767 				       checkstate_need_ast & map);
2768 #endif
2769 				break;
2770 			}
2771 		}
2772 		if (id == p->p_oncpu)
2773 			return;
2774 	}
2775 }
2776 
2777 void
2778 forward_roundrobin(void)
2779 {
2780 	u_int map;
2781 	int i;
2782 
2783 	if (!smp_started || !invltlb_ok || cold || panicstr)
2784 		return;
2785 	if (!forward_roundrobin_enabled)
2786 		return;
2787 	resched_cpus |= other_cpus;
2788 	map = other_cpus & ~stopped_cpus ;
2789 #if 1
2790 	selected_apic_ipi(map, XCPUAST_OFFSET, APIC_DELMODE_FIXED);
2791 #else
2792 	(void) all_but_self_ipi(XCPUAST_OFFSET);
2793 #endif
2794 	i = 0;
2795 	while ((checkstate_need_ast & map) != 0) {
2796 		/* spin */
2797 		i++;
2798 		if (i > 100000) {
2799 #if 0
2800 			printf("forward_roundrobin: dropped ast 0x%x\n",
2801 			       checkstate_need_ast & map);
2802 #endif
2803 			break;
2804 		}
2805 	}
2806 }
2807 
2808 
2809 #ifdef APIC_INTR_REORDER
2810 /*
2811  *	Maintain mapping from softintr vector to isr bit in local apic.
2812  */
2813 void
2814 set_lapic_isrloc(int intr, int vector)
2815 {
2816 	if (intr < 0 || intr > 32)
2817 		panic("set_apic_isrloc: bad intr argument: %d",intr);
2818 	if (vector < ICU_OFFSET || vector > 255)
2819 		panic("set_apic_isrloc: bad vector argument: %d",vector);
2820 	apic_isrbit_location[intr].location = &lapic.isr0 + ((vector>>5)<<2);
2821 	apic_isrbit_location[intr].bit = (1<<(vector & 31));
2822 }
2823 #endif
2824 
2825 /*
2826  * All-CPU rendezvous.  CPUs are signalled, all execute the setup function
2827  * (if specified), rendezvous, execute the action function (if specified),
2828  * rendezvous again, execute the teardown function (if specified), and then
2829  * resume.
2830  *
2831  * Note that the supplied external functions _must_ be reentrant and aware
2832  * that they are running in parallel and in an unknown lock context.
2833  */
2834 static void (*smp_rv_setup_func)(void *arg);
2835 static void (*smp_rv_action_func)(void *arg);
2836 static void (*smp_rv_teardown_func)(void *arg);
2837 static void *smp_rv_func_arg;
2838 static volatile int smp_rv_waiters[2];
2839 
2840 void
2841 smp_rendezvous_action(void)
2842 {
2843 	/* setup function */
2844 	if (smp_rv_setup_func != NULL)
2845 		smp_rv_setup_func(smp_rv_func_arg);
2846 	/* spin on entry rendezvous */
2847 	atomic_add_int(&smp_rv_waiters[0], 1);
2848 	while (smp_rv_waiters[0] < mp_ncpus)
2849 		;
2850 	/* action function */
2851 	if (smp_rv_action_func != NULL)
2852 		smp_rv_action_func(smp_rv_func_arg);
2853 	/* spin on exit rendezvous */
2854 	atomic_add_int(&smp_rv_waiters[1], 1);
2855 	while (smp_rv_waiters[1] < mp_ncpus)
2856 		;
2857 	/* teardown function */
2858 	if (smp_rv_teardown_func != NULL)
2859 		smp_rv_teardown_func(smp_rv_func_arg);
2860 }
2861 
2862 void
2863 smp_rendezvous(void (* setup_func)(void *),
2864 	       void (* action_func)(void *),
2865 	       void (* teardown_func)(void *),
2866 	       void *arg)
2867 {
2868 	u_int	efl;
2869 
2870 	/* obtain rendezvous lock */
2871 	s_lock(&smp_rv_lock);		/* XXX sleep here? NOWAIT flag? */
2872 
2873 	/* set static function pointers */
2874 	smp_rv_setup_func = setup_func;
2875 	smp_rv_action_func = action_func;
2876 	smp_rv_teardown_func = teardown_func;
2877 	smp_rv_func_arg = arg;
2878 	smp_rv_waiters[0] = 0;
2879 	smp_rv_waiters[1] = 0;
2880 
2881 	/* disable interrupts on this CPU, save interrupt status */
2882 	efl = read_eflags();
2883 	write_eflags(efl & ~PSL_I);
2884 
2885 	/* signal other processors, which will enter the IPI with interrupts off */
2886 	all_but_self_ipi(XRENDEZVOUS_OFFSET);
2887 
2888 	/* call executor function */
2889 	smp_rendezvous_action();
2890 
2891 	/* restore interrupt flag */
2892 	write_eflags(efl);
2893 
2894 	/* release lock */
2895 	s_unlock(&smp_rv_lock);
2896 }
2897 
2898 void
2899 release_aps(void *dummy __unused)
2900 {
2901 	s_unlock(&ap_boot_lock);
2902 }
2903 
2904 SYSINIT(start_aps, SI_SUB_SMP, SI_ORDER_FIRST, release_aps, NULL);
2905