xref: /freebsd/sys/kern/subr_smp.c (revision 5931a9c24e798d31c4389671b8c8431911b18c5d)
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  *	$Id: mp_machdep.c,v 1.74 1998/05/11 01:06:06 dyson Exp $
26  */
27 
28 #include "opt_smp.h"
29 #include "opt_vm86.h"
30 #include "opt_cpu.h"
31 
32 #ifdef SMP
33 #include <machine/smptests.h>
34 #else
35 #error
36 #endif
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/kernel.h>
41 #include <sys/proc.h>
42 #include <sys/sysctl.h>
43 #ifdef BETTER_CLOCK
44 #include <sys/dkstat.h>
45 #endif
46 
47 #include <vm/vm.h>
48 #include <vm/vm_param.h>
49 #include <vm/pmap.h>
50 #include <vm/vm_kern.h>
51 #include <vm/vm_extern.h>
52 #ifdef BETTER_CLOCK
53 #include <sys/lock.h>
54 #include <vm/vm_map.h>
55 #include <sys/user.h>
56 #ifdef GPROF
57 #include <sys/gmon.h>
58 #endif
59 #endif
60 
61 #include <machine/smp.h>
62 #include <machine/apic.h>
63 #include <machine/mpapic.h>
64 #include <machine/segments.h>
65 #include <machine/smptests.h>	/** TEST_DEFAULT_CONFIG, TEST_TEST1 */
66 #include <machine/tss.h>
67 #include <machine/specialreg.h>
68 #include <machine/cputypes.h>
69 #include <machine/globaldata.h>
70 
71 #include <i386/i386/cons.h>	/* cngetc() */
72 
73 #if defined(APIC_IO)
74 #include <machine/md_var.h>		/* setidt() */
75 #include <i386/isa/icu.h>		/* IPIs */
76 #include <i386/isa/intr_machdep.h>	/* IPIs */
77 #endif	/* APIC_IO */
78 
79 #if defined(TEST_DEFAULT_CONFIG)
80 #define MPFPS_MPFB1	TEST_DEFAULT_CONFIG
81 #else
82 #define MPFPS_MPFB1	mpfps->mpfb1
83 #endif  /* TEST_DEFAULT_CONFIG */
84 
85 #define WARMBOOT_TARGET		0
86 #define WARMBOOT_OFF		(KERNBASE + 0x0467)
87 #define WARMBOOT_SEG		(KERNBASE + 0x0469)
88 
89 #define BIOS_BASE		(0xf0000)
90 #define BIOS_SIZE		(0x10000)
91 #define BIOS_COUNT		(BIOS_SIZE/4)
92 
93 #define CMOS_REG		(0x70)
94 #define CMOS_DATA		(0x71)
95 #define BIOS_RESET		(0x0f)
96 #define BIOS_WARM		(0x0a)
97 
98 #define PROCENTRY_FLAG_EN	0x01
99 #define PROCENTRY_FLAG_BP	0x02
100 #define IOAPICENTRY_FLAG_EN	0x01
101 
102 
103 /* MP Floating Pointer Structure */
104 typedef struct MPFPS {
105 	char    signature[4];
106 	void   *pap;
107 	u_char  length;
108 	u_char  spec_rev;
109 	u_char  checksum;
110 	u_char  mpfb1;
111 	u_char  mpfb2;
112 	u_char  mpfb3;
113 	u_char  mpfb4;
114 	u_char  mpfb5;
115 }      *mpfps_t;
116 
117 /* MP Configuration Table Header */
118 typedef struct MPCTH {
119 	char    signature[4];
120 	u_short base_table_length;
121 	u_char  spec_rev;
122 	u_char  checksum;
123 	u_char  oem_id[8];
124 	u_char  product_id[12];
125 	void   *oem_table_pointer;
126 	u_short oem_table_size;
127 	u_short entry_count;
128 	void   *apic_address;
129 	u_short extended_table_length;
130 	u_char  extended_table_checksum;
131 	u_char  reserved;
132 }      *mpcth_t;
133 
134 
135 typedef struct PROCENTRY {
136 	u_char  type;
137 	u_char  apic_id;
138 	u_char  apic_version;
139 	u_char  cpu_flags;
140 	u_long  cpu_signature;
141 	u_long  feature_flags;
142 	u_long  reserved1;
143 	u_long  reserved2;
144 }      *proc_entry_ptr;
145 
146 typedef struct BUSENTRY {
147 	u_char  type;
148 	u_char  bus_id;
149 	char    bus_type[6];
150 }      *bus_entry_ptr;
151 
152 typedef struct IOAPICENTRY {
153 	u_char  type;
154 	u_char  apic_id;
155 	u_char  apic_version;
156 	u_char  apic_flags;
157 	void   *apic_address;
158 }      *io_apic_entry_ptr;
159 
160 typedef struct INTENTRY {
161 	u_char  type;
162 	u_char  int_type;
163 	u_short int_flags;
164 	u_char  src_bus_id;
165 	u_char  src_bus_irq;
166 	u_char  dst_apic_id;
167 	u_char  dst_apic_int;
168 }      *int_entry_ptr;
169 
170 /* descriptions of MP basetable entries */
171 typedef struct BASETABLE_ENTRY {
172 	u_char  type;
173 	u_char  length;
174 	char    name[16];
175 }       basetable_entry;
176 
177 /*
178  * this code MUST be enabled here and in mpboot.s.
179  * it follows the very early stages of AP boot by placing values in CMOS ram.
180  * it NORMALLY will never be needed and thus the primitive method for enabling.
181  *
182 #define CHECK_POINTS
183  */
184 
185 #if defined(CHECK_POINTS)
186 #define CHECK_READ(A)	 (outb(CMOS_REG, (A)), inb(CMOS_DATA))
187 #define CHECK_WRITE(A,D) (outb(CMOS_REG, (A)), outb(CMOS_DATA, (D)))
188 
189 #define CHECK_INIT(D);				\
190 	CHECK_WRITE(0x34, (D));			\
191 	CHECK_WRITE(0x35, (D));			\
192 	CHECK_WRITE(0x36, (D));			\
193 	CHECK_WRITE(0x37, (D));			\
194 	CHECK_WRITE(0x38, (D));			\
195 	CHECK_WRITE(0x39, (D));
196 
197 #define CHECK_PRINT(S);				\
198 	printf("%s: %d, %d, %d, %d, %d, %d\n",	\
199 	   (S),					\
200 	   CHECK_READ(0x34),			\
201 	   CHECK_READ(0x35),			\
202 	   CHECK_READ(0x36),			\
203 	   CHECK_READ(0x37),			\
204 	   CHECK_READ(0x38),			\
205 	   CHECK_READ(0x39));
206 
207 #else				/* CHECK_POINTS */
208 
209 #define CHECK_INIT(D)
210 #define CHECK_PRINT(S)
211 
212 #endif				/* CHECK_POINTS */
213 
214 /*
215  * Values to send to the POST hardware.
216  */
217 #define MP_BOOTADDRESS_POST	0x10
218 #define MP_PROBE_POST		0x11
219 #define MPTABLE_PASS1_POST	0x12
220 
221 #define MP_START_POST		0x13
222 #define MP_ENABLE_POST		0x14
223 #define MPTABLE_PASS2_POST	0x15
224 
225 #define START_ALL_APS_POST	0x16
226 #define INSTALL_AP_TRAMP_POST	0x17
227 #define START_AP_POST		0x18
228 
229 #define MP_ANNOUNCE_POST	0x19
230 
231 
232 /** XXX FIXME: where does this really belong, isa.h/isa.c perhaps? */
233 int	current_postcode;
234 
235 /** XXX FIXME: what system files declare these??? */
236 extern struct region_descriptor r_gdt, r_idt;
237 
238 int	bsp_apic_ready = 0;	/* flags useability of BSP apic */
239 int	mp_ncpus;		/* # of CPUs, including BSP */
240 int	mp_naps;		/* # of Applications processors */
241 int	mp_nbusses;		/* # of busses */
242 int	mp_napics;		/* # of IO APICs */
243 int	boot_cpu_id;		/* designated BSP */
244 vm_offset_t cpu_apic_address;
245 vm_offset_t io_apic_address[NAPICID];	/* NAPICID is more than enough */
246 extern	int nkpt;
247 
248 u_int32_t cpu_apic_versions[NCPU];
249 u_int32_t io_apic_versions[NAPIC];
250 
251 #ifdef APIC_INTR_DIAGNOSTIC
252 int apic_itrace_enter[32];
253 int apic_itrace_tryisrlock[32];
254 int apic_itrace_gotisrlock[32];
255 int apic_itrace_active[32];
256 int apic_itrace_masked[32];
257 int apic_itrace_noisrlock[32];
258 int apic_itrace_masked2[32];
259 int apic_itrace_unmask[32];
260 int apic_itrace_noforward[32];
261 int apic_itrace_leave[32];
262 int apic_itrace_enter2[32];
263 int apic_itrace_doreti[32];
264 int apic_itrace_splz[32];
265 int apic_itrace_eoi[32];
266 #ifdef APIC_INTR_DIAGNOSTIC_IRQ
267 unsigned short apic_itrace_debugbuffer[32768];
268 int apic_itrace_debugbuffer_idx;
269 struct simplelock apic_itrace_debuglock;
270 #endif
271 #endif
272 
273 #ifdef APIC_INTR_REORDER
274 struct {
275 	volatile int *location;
276 	int bit;
277 } apic_isrbit_location[32];
278 #endif
279 
280 /*
281  * APIC ID logical/physical mapping structures.
282  * We oversize these to simplify boot-time config.
283  */
284 int     cpu_num_to_apic_id[NAPICID];
285 int     io_num_to_apic_id[NAPICID];
286 int     apic_id_to_logical[NAPICID];
287 
288 
289 /* Bitmap of all available CPUs */
290 u_int	all_cpus;
291 
292 /* AP uses this PTD during bootstrap.  Do not staticize.  */
293 pd_entry_t *bootPTD;
294 
295 /* Hotwire a 0->4MB V==P mapping */
296 extern pt_entry_t *KPTphys;
297 
298 /* Virtual address of per-cpu common_tss */
299 extern struct i386tss common_tss;
300 #ifdef VM86
301 extern struct segment_descriptor common_tssd;
302 extern u_int private_tss;		/* flag indicating private tss */
303 extern u_int my_tr;
304 #endif /* VM86 */
305 
306 /* IdlePTD per cpu */
307 pd_entry_t *IdlePTDS[NCPU];
308 
309 /* "my" private page table page, for BSP init */
310 extern pt_entry_t SMP_prvpt[];
311 
312 /* Private page pointer to curcpu's PTD, used during BSP init */
313 extern pd_entry_t *my_idlePTD;
314 
315 static int smp_started;		/* has the system started? */
316 
317 /*
318  * Local data and functions.
319  */
320 
321 static int	mp_capable;
322 static u_int	boot_address;
323 static u_int	base_memory;
324 
325 static int	picmode;		/* 0: virtual wire mode, 1: PIC mode */
326 static mpfps_t	mpfps;
327 static int	search_for_sig(u_int32_t target, int count);
328 static void	mp_enable(u_int boot_addr);
329 
330 static int	mptable_pass1(void);
331 static int	mptable_pass2(void);
332 static void	default_mp_table(int type);
333 static void	fix_mp_table(void);
334 static void	init_locks(void);
335 static int	start_all_aps(u_int boot_addr);
336 static void	install_ap_tramp(u_int boot_addr);
337 static int	start_ap(int logicalCpu, u_int boot_addr);
338 
339 /*
340  * Calculate usable address in base memory for AP trampoline code.
341  */
342 u_int
343 mp_bootaddress(u_int basemem)
344 {
345 	POSTCODE(MP_BOOTADDRESS_POST);
346 
347 	base_memory = basemem * 1024;	/* convert to bytes */
348 
349 	boot_address = base_memory & ~0xfff;	/* round down to 4k boundary */
350 	if ((base_memory - boot_address) < bootMP_size)
351 		boot_address -= 4096;	/* not enough, lower by 4k */
352 
353 	return boot_address;
354 }
355 
356 
357 /*
358  * Look for an Intel MP spec table (ie, SMP capable hardware).
359  */
360 int
361 mp_probe(void)
362 {
363 	int     x;
364 	u_long  segment;
365 	u_int32_t target;
366 
367 	POSTCODE(MP_PROBE_POST);
368 
369 	/* see if EBDA exists */
370 	if (segment = (u_long) * (u_short *) (KERNBASE + 0x40e)) {
371 		/* search first 1K of EBDA */
372 		target = (u_int32_t) (segment << 4);
373 		if ((x = search_for_sig(target, 1024 / 4)) >= 0)
374 			goto found;
375 	} else {
376 		/* last 1K of base memory, effective 'top of base' passed in */
377 		target = (u_int32_t) (base_memory - 0x400);
378 		if ((x = search_for_sig(target, 1024 / 4)) >= 0)
379 			goto found;
380 	}
381 
382 	/* search the BIOS */
383 	target = (u_int32_t) BIOS_BASE;
384 	if ((x = search_for_sig(target, BIOS_COUNT)) >= 0)
385 		goto found;
386 
387 	/* nothing found */
388 	mpfps = (mpfps_t)0;
389 	mp_capable = 0;
390 	return 0;
391 
392 found:
393 	/* calculate needed resources */
394 	mpfps = (mpfps_t)x;
395 	if (mptable_pass1())
396 		panic("you must reconfigure your kernel");
397 
398 	/* flag fact that we are running multiple processors */
399 	mp_capable = 1;
400 	return 1;
401 }
402 
403 
404 /*
405  * Startup the SMP processors.
406  */
407 void
408 mp_start(void)
409 {
410 	POSTCODE(MP_START_POST);
411 
412 	/* look for MP capable motherboard */
413 	if (mp_capable)
414 		mp_enable(boot_address);
415 	else
416 		panic("MP hardware not found!");
417 }
418 
419 
420 /*
421  * Print various information about the SMP system hardware and setup.
422  */
423 void
424 mp_announce(void)
425 {
426 	int     x;
427 
428 	POSTCODE(MP_ANNOUNCE_POST);
429 
430 	printf("FreeBSD/SMP: Multiprocessor motherboard\n");
431 	printf(" cpu0 (BSP): apic id: %2d", CPU_TO_ID(0));
432 	printf(", version: 0x%08x", cpu_apic_versions[0]);
433 	printf(", at 0x%08x\n", cpu_apic_address);
434 	for (x = 1; x <= mp_naps; ++x) {
435 		printf(" cpu%d (AP):  apic id: %2d", x, CPU_TO_ID(x));
436 		printf(", version: 0x%08x", cpu_apic_versions[x]);
437 		printf(", at 0x%08x\n", cpu_apic_address);
438 	}
439 
440 #if defined(APIC_IO)
441 	for (x = 0; x < mp_napics; ++x) {
442 		printf(" io%d (APIC): apic id: %2d", x, IO_TO_ID(x));
443 		printf(", version: 0x%08x", io_apic_versions[x]);
444 		printf(", at 0x%08x\n", io_apic_address[x]);
445 	}
446 #else
447 	printf(" Warning: APIC I/O disabled\n");
448 #endif	/* APIC_IO */
449 }
450 
451 /*
452  * AP cpu's call this to sync up protected mode.
453  */
454 void
455 init_secondary(void)
456 {
457 	int	gsel_tss;
458 #ifndef VM86
459 	u_int	my_tr;
460 #endif
461 
462 	r_gdt.rd_limit = sizeof(gdt[0]) * (NGDT + NCPU) - 1;
463 	r_gdt.rd_base = (int) gdt;
464 	lgdt(&r_gdt);			/* does magic intra-segment return */
465 	lidt(&r_idt);
466 	lldt(_default_ldt);
467 
468 	my_tr = NGDT + cpuid;
469 	gsel_tss = GSEL(my_tr, SEL_KPL);
470 	gdt[my_tr].sd.sd_type = SDT_SYS386TSS;
471 	common_tss.tss_esp0 = 0;	/* not used until after switch */
472 	common_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
473 	common_tss.tss_ioopt = (sizeof common_tss) << 16;
474 #ifdef VM86
475 	common_tssd = gdt[my_tr].sd;
476 	private_tss = 0;
477 #endif /* VM86 */
478 	ltr(gsel_tss);
479 
480 	load_cr0(0x8005003b);		/* XXX! */
481 
482 	PTD[0] = 0;
483 	pmap_set_opt((unsigned *)PTD);
484 
485 	putmtrr();
486 	pmap_setvidram();
487 
488 	invltlb();
489 }
490 
491 
492 #if defined(APIC_IO)
493 /*
494  * Final configuration of the BSP's local APIC:
495  *  - disable 'pic mode'.
496  *  - disable 'virtual wire mode'.
497  *  - enable NMI.
498  */
499 void
500 bsp_apic_configure(void)
501 {
502 	u_char		byte;
503 	u_int32_t	temp;
504 
505 	/* leave 'pic mode' if necessary */
506 	if (picmode) {
507 		outb(0x22, 0x70);	/* select IMCR */
508 		byte = inb(0x23);	/* current contents */
509 		byte |= 0x01;		/* mask external INTR */
510 		outb(0x23, byte);	/* disconnect 8259s/NMI */
511 	}
512 
513 	/* mask lint0 (the 8259 'virtual wire' connection) */
514 	temp = lapic.lvt_lint0;
515 	temp |= APIC_LVT_M;		/* set the mask */
516 	lapic.lvt_lint0 = temp;
517 
518         /* setup lint1 to handle NMI */
519         temp = lapic.lvt_lint1;
520         temp &= ~APIC_LVT_M;		/* clear the mask */
521         lapic.lvt_lint1 = temp;
522 
523 	if (bootverbose)
524 		apic_dump("bsp_apic_configure()");
525 }
526 #endif  /* APIC_IO */
527 
528 
529 /*******************************************************************
530  * local functions and data
531  */
532 
533 /*
534  * start the SMP system
535  */
536 static void
537 mp_enable(u_int boot_addr)
538 {
539 	int     x;
540 #if defined(APIC_IO)
541 	int     apic;
542 	u_int   ux;
543 #endif	/* APIC_IO */
544 
545 	getmtrr();
546 	pmap_setvidram();
547 
548 	POSTCODE(MP_ENABLE_POST);
549 
550 	/* turn on 4MB of V == P addressing so we can get to MP table */
551 	*(int *)PTD = PG_V | PG_RW | ((u_long)KPTphys & PG_FRAME);
552 	invltlb();
553 
554 	/* examine the MP table for needed info, uses physical addresses */
555 	x = mptable_pass2();
556 
557 	*(int *)PTD = 0;
558 	invltlb();
559 
560 	/* can't process default configs till the CPU APIC is pmapped */
561 	if (x)
562 		default_mp_table(x);
563 
564 	/* post scan cleanup */
565 	fix_mp_table();
566 
567 #if defined(APIC_IO)
568 
569 	/* fill the LOGICAL io_apic_versions table */
570 	for (apic = 0; apic < mp_napics; ++apic) {
571 		ux = io_apic_read(apic, IOAPIC_VER);
572 		io_apic_versions[apic] = ux;
573 	}
574 
575 	/* program each IO APIC in the system */
576 	for (apic = 0; apic < mp_napics; ++apic)
577 		if (io_apic_setup(apic) < 0)
578 			panic("IO APIC setup failure");
579 
580 	/* install a 'Spurious INTerrupt' vector */
581 	setidt(XSPURIOUSINT_OFFSET, Xspuriousint,
582 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
583 
584 	/* install an inter-CPU IPI for TLB invalidation */
585 	setidt(XINVLTLB_OFFSET, Xinvltlb,
586 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
587 
588 #ifdef BETTER_CLOCK
589 	/* install an inter-CPU IPI for reading processor state */
590 	setidt(XCPUCHECKSTATE_OFFSET, Xcpucheckstate,
591 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
592 #endif
593 
594 	/* install an inter-CPU IPI for forcing an additional software trap */
595 	setidt(XCPUAST_OFFSET, Xcpuast,
596 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
597 
598 	/* install an inter-CPU IPI for interrupt forwarding */
599 	setidt(XFORWARD_IRQ_OFFSET, Xforward_irq,
600 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
601 
602 	/* install an inter-CPU IPI for CPU stop/restart */
603 	setidt(XCPUSTOP_OFFSET, Xcpustop,
604 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
605 
606 #if defined(TEST_TEST1)
607 	/* install a "fake hardware INTerrupt" vector */
608 	setidt(XTEST1_OFFSET, Xtest1,
609 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
610 #endif  /** TEST_TEST1 */
611 
612 #endif	/* APIC_IO */
613 
614 	/* initialize all SMP locks */
615 	init_locks();
616 
617 	/* start each Application Processor */
618 	start_all_aps(boot_addr);
619 
620 	/*
621 	 * The init process might be started on a different CPU now,
622 	 * and the boot CPU might not call prepare_usermode to get
623 	 * cr0 correctly configured. Thus we initialize cr0 here.
624 	 */
625 	load_cr0(rcr0() | CR0_WP | CR0_AM);
626 }
627 
628 
629 /*
630  * look for the MP spec signature
631  */
632 
633 /* string defined by the Intel MP Spec as identifying the MP table */
634 #define MP_SIG		0x5f504d5f	/* _MP_ */
635 #define NEXT(X)		((X) += 4)
636 static int
637 search_for_sig(u_int32_t target, int count)
638 {
639 	int     x;
640 	u_int32_t *addr = (u_int32_t *) (KERNBASE + target);
641 
642 	for (x = 0; x < count; NEXT(x))
643 		if (addr[x] == MP_SIG)
644 			/* make array index a byte index */
645 			return (target + (x * sizeof(u_int32_t)));
646 
647 	return -1;
648 }
649 
650 
651 static basetable_entry basetable_entry_types[] =
652 {
653 	{0, 20, "Processor"},
654 	{1, 8, "Bus"},
655 	{2, 8, "I/O APIC"},
656 	{3, 8, "I/O INT"},
657 	{4, 8, "Local INT"}
658 };
659 
660 typedef struct BUSDATA {
661 	u_char  bus_id;
662 	enum busTypes bus_type;
663 }       bus_datum;
664 
665 typedef struct INTDATA {
666 	u_char  int_type;
667 	u_short int_flags;
668 	u_char  src_bus_id;
669 	u_char  src_bus_irq;
670 	u_char  dst_apic_id;
671 	u_char  dst_apic_int;
672 }       io_int, local_int;
673 
674 typedef struct BUSTYPENAME {
675 	u_char  type;
676 	char    name[7];
677 }       bus_type_name;
678 
679 static bus_type_name bus_type_table[] =
680 {
681 	{CBUS, "CBUS"},
682 	{CBUSII, "CBUSII"},
683 	{EISA, "EISA"},
684 	{UNKNOWN_BUSTYPE, "---"},
685 	{UNKNOWN_BUSTYPE, "---"},
686 	{ISA, "ISA"},
687 	{UNKNOWN_BUSTYPE, "---"},
688 	{UNKNOWN_BUSTYPE, "---"},
689 	{UNKNOWN_BUSTYPE, "---"},
690 	{UNKNOWN_BUSTYPE, "---"},
691 	{UNKNOWN_BUSTYPE, "---"},
692 	{UNKNOWN_BUSTYPE, "---"},
693 	{PCI, "PCI"},
694 	{UNKNOWN_BUSTYPE, "---"},
695 	{UNKNOWN_BUSTYPE, "---"},
696 	{UNKNOWN_BUSTYPE, "---"},
697 	{UNKNOWN_BUSTYPE, "---"},
698 	{XPRESS, "XPRESS"},
699 	{UNKNOWN_BUSTYPE, "---"}
700 };
701 /* from MP spec v1.4, table 5-1 */
702 static int default_data[7][5] =
703 {
704 /*   nbus, id0, type0, id1, type1 */
705 	{1, 0, ISA, 255, 255},
706 	{1, 0, EISA, 255, 255},
707 	{1, 0, EISA, 255, 255},
708 	{0, 255, 255, 255, 255},/* MCA not supported */
709 	{2, 0, ISA, 1, PCI},
710 	{2, 0, EISA, 1, PCI},
711 	{0, 255, 255, 255, 255}	/* MCA not supported */
712 };
713 
714 
715 /* the bus data */
716 static bus_datum bus_data[NBUS];
717 
718 /* the IO INT data, one entry per possible APIC INTerrupt */
719 static io_int  io_apic_ints[NINTR];
720 
721 static int nintrs;
722 
723 static int processor_entry	__P((proc_entry_ptr entry, int cpu));
724 static int bus_entry		__P((bus_entry_ptr entry, int bus));
725 static int io_apic_entry	__P((io_apic_entry_ptr entry, int apic));
726 static int int_entry		__P((int_entry_ptr entry, int intr));
727 static int lookup_bus_type	__P((char *name));
728 
729 
730 /*
731  * 1st pass on motherboard's Intel MP specification table.
732  *
733  * initializes:
734  *	mp_ncpus = 1
735  *
736  * determines:
737  *	cpu_apic_address (common to all CPUs)
738  *	io_apic_address[N]
739  *	mp_naps
740  *	mp_nbusses
741  *	mp_napics
742  *	nintrs
743  */
744 static int
745 mptable_pass1(void)
746 {
747 	int	x;
748 	mpcth_t	cth;
749 	int	totalSize;
750 	void*	position;
751 	int	count;
752 	int	type;
753 	int	mustpanic;
754 
755 	POSTCODE(MPTABLE_PASS1_POST);
756 
757 	mustpanic = 0;
758 
759 	/* clear various tables */
760 	for (x = 0; x < NAPICID; ++x) {
761 		io_apic_address[x] = ~0;	/* IO APIC address table */
762 	}
763 
764 	/* init everything to empty */
765 	mp_naps = 0;
766 	mp_nbusses = 0;
767 	mp_napics = 0;
768 	nintrs = 0;
769 
770 	/* check for use of 'default' configuration */
771 	if (MPFPS_MPFB1 != 0) {
772 		/* use default addresses */
773 		cpu_apic_address = DEFAULT_APIC_BASE;
774 		io_apic_address[0] = DEFAULT_IO_APIC_BASE;
775 
776 		/* fill in with defaults */
777 		mp_naps = 2;		/* includes BSP */
778 		mp_nbusses = default_data[MPFPS_MPFB1 - 1][0];
779 #if defined(APIC_IO)
780 		mp_napics = 1;
781 		nintrs = 16;
782 #endif	/* APIC_IO */
783 	}
784 	else {
785 		if ((cth = mpfps->pap) == 0)
786 			panic("MP Configuration Table Header MISSING!");
787 
788 		cpu_apic_address = (vm_offset_t) cth->apic_address;
789 
790 		/* walk the table, recording info of interest */
791 		totalSize = cth->base_table_length - sizeof(struct MPCTH);
792 		position = (u_char *) cth + sizeof(struct MPCTH);
793 		count = cth->entry_count;
794 
795 		while (count--) {
796 			switch (type = *(u_char *) position) {
797 			case 0: /* processor_entry */
798 				if (((proc_entry_ptr)position)->cpu_flags
799 					& PROCENTRY_FLAG_EN)
800 					++mp_naps;
801 				break;
802 			case 1: /* bus_entry */
803 				++mp_nbusses;
804 				break;
805 			case 2: /* io_apic_entry */
806 				if (((io_apic_entry_ptr)position)->apic_flags
807 					& IOAPICENTRY_FLAG_EN)
808 					io_apic_address[mp_napics++] =
809 					    (vm_offset_t)((io_apic_entry_ptr)
810 						position)->apic_address;
811 				break;
812 			case 3: /* int_entry */
813 				++nintrs;
814 				break;
815 			case 4:	/* int_entry */
816 				break;
817 			default:
818 				panic("mpfps Base Table HOSED!");
819 				/* NOTREACHED */
820 			}
821 
822 			totalSize -= basetable_entry_types[type].length;
823 			(u_char*)position += basetable_entry_types[type].length;
824 		}
825 	}
826 
827 	/* qualify the numbers */
828 	if (mp_naps > NCPU)
829 #if 0 /* XXX FIXME: kern/4255 */
830 		printf("Warning: only using %d of %d available CPUs!\n",
831 			NCPU, mp_naps);
832 #else
833 	{
834 		printf("NCPU cannot be different than actual CPU count.\n");
835 		printf(" add 'options NCPU=%d' to your kernel config file,\n",
836 			mp_naps);
837 		printf(" then rerun config & rebuild your SMP kernel\n");
838 		mustpanic = 1;
839 	}
840 #endif /* XXX FIXME: kern/4255 */
841 	if (mp_nbusses > NBUS) {
842 		printf("found %d busses, increase NBUS\n", mp_nbusses);
843 		mustpanic = 1;
844 	}
845 	if (mp_napics > NAPIC) {
846 		printf("found %d apics, increase NAPIC\n", mp_napics);
847 		mustpanic = 1;
848 	}
849 	if (nintrs > NINTR) {
850 		printf("found %d intrs, increase NINTR\n", nintrs);
851 		mustpanic = 1;
852 	}
853 
854 	/*
855 	 * Count the BSP.
856 	 * This is also used as a counter while starting the APs.
857 	 */
858 	mp_ncpus = 1;
859 
860 	--mp_naps;	/* subtract the BSP */
861 
862 	return mustpanic;
863 }
864 
865 
866 /*
867  * 2nd pass on motherboard's Intel MP specification table.
868  *
869  * sets:
870  *	boot_cpu_id
871  *	ID_TO_IO(N), phy APIC ID to log CPU/IO table
872  *	CPU_TO_ID(N), logical CPU to APIC ID table
873  *	IO_TO_ID(N), logical IO to APIC ID table
874  *	bus_data[N]
875  *	io_apic_ints[N]
876  */
877 static int
878 mptable_pass2(void)
879 {
880 	int     x;
881 	mpcth_t cth;
882 	int     totalSize;
883 	void*   position;
884 	int     count;
885 	int     type;
886 	int     apic, bus, cpu, intr;
887 
888 	POSTCODE(MPTABLE_PASS2_POST);
889 
890 	/* clear various tables */
891 	for (x = 0; x < NAPICID; ++x) {
892 		ID_TO_IO(x) = -1;	/* phy APIC ID to log CPU/IO table */
893 		CPU_TO_ID(x) = -1;	/* logical CPU to APIC ID table */
894 		IO_TO_ID(x) = -1;	/* logical IO to APIC ID table */
895 	}
896 
897 	/* clear bus data table */
898 	for (x = 0; x < NBUS; ++x)
899 		bus_data[x].bus_id = 0xff;
900 
901 	/* clear IO APIC INT table */
902 	for (x = 0; x < NINTR; ++x)
903 		io_apic_ints[x].int_type = 0xff;
904 
905 	/* setup the cpu/apic mapping arrays */
906 	boot_cpu_id = -1;
907 
908 	/* record whether PIC or virtual-wire mode */
909 	picmode = (mpfps->mpfb2 & 0x80) ? 1 : 0;
910 
911 	/* check for use of 'default' configuration */
912 	if (MPFPS_MPFB1 != 0)
913 		return MPFPS_MPFB1;	/* return default configuration type */
914 
915 	if ((cth = mpfps->pap) == 0)
916 		panic("MP Configuration Table Header MISSING!");
917 
918 	/* walk the table, recording info of interest */
919 	totalSize = cth->base_table_length - sizeof(struct MPCTH);
920 	position = (u_char *) cth + sizeof(struct MPCTH);
921 	count = cth->entry_count;
922 	apic = bus = intr = 0;
923 	cpu = 1;				/* pre-count the BSP */
924 
925 	while (count--) {
926 		switch (type = *(u_char *) position) {
927 		case 0:
928 			if (processor_entry(position, cpu))
929 				++cpu;
930 			break;
931 		case 1:
932 			if (bus_entry(position, bus))
933 				++bus;
934 			break;
935 		case 2:
936 			if (io_apic_entry(position, apic))
937 				++apic;
938 			break;
939 		case 3:
940 			if (int_entry(position, intr))
941 				++intr;
942 			break;
943 		case 4:
944 			/* int_entry(position); */
945 			break;
946 		default:
947 			panic("mpfps Base Table HOSED!");
948 			/* NOTREACHED */
949 		}
950 
951 		totalSize -= basetable_entry_types[type].length;
952 		(u_char *) position += basetable_entry_types[type].length;
953 	}
954 
955 	if (boot_cpu_id == -1)
956 		panic("NO BSP found!");
957 
958 	/* report fact that its NOT a default configuration */
959 	return 0;
960 }
961 
962 
963 /*
964  * parse an Intel MP specification table
965  */
966 static void
967 fix_mp_table(void)
968 {
969 	int	x;
970 	int	id;
971 	int	bus_0;
972 	int	bus_pci;
973 	int	num_pci_bus;
974 
975 	/*
976 	 * Fix mis-numbering of the PCI bus and its INT entries if the BIOS
977 	 * did it wrong.  The MP spec says that when more than 1 PCI bus
978 	 * exists the BIOS must begin with bus entries for the PCI bus and use
979 	 * actual PCI bus numbering.  This implies that when only 1 PCI bus
980 	 * exists the BIOS can choose to ignore this ordering, and indeed many
981 	 * MP motherboards do ignore it.  This causes a problem when the PCI
982 	 * sub-system makes requests of the MP sub-system based on PCI bus
983 	 * numbers.	So here we look for the situation and renumber the
984 	 * busses and associated INTs in an effort to "make it right".
985 	 */
986 
987 	/* find bus 0, PCI bus, count the number of PCI busses */
988 	for (num_pci_bus = 0, x = 0; x < mp_nbusses; ++x) {
989 		if (bus_data[x].bus_id == 0) {
990 			bus_0 = x;
991 		}
992 		if (bus_data[x].bus_type == PCI) {
993 			++num_pci_bus;
994 			bus_pci = x;
995 		}
996 	}
997 	/*
998 	 * bus_0 == slot of bus with ID of 0
999 	 * bus_pci == slot of last PCI bus encountered
1000 	 */
1001 
1002 	/* check the 1 PCI bus case for sanity */
1003 	if (num_pci_bus == 1) {
1004 
1005 		/* if it is number 0 all is well */
1006 		if (bus_data[bus_pci].bus_id == 0)
1007 			return;
1008 
1009 		/* mis-numbered, swap with whichever bus uses slot 0 */
1010 
1011 		/* swap the bus entry types */
1012 		bus_data[bus_pci].bus_type = bus_data[bus_0].bus_type;
1013 		bus_data[bus_0].bus_type = PCI;
1014 
1015 		/* swap each relavant INTerrupt entry */
1016 		id = bus_data[bus_pci].bus_id;
1017 		for (x = 0; x < nintrs; ++x) {
1018 			if (io_apic_ints[x].src_bus_id == id) {
1019 				io_apic_ints[x].src_bus_id = 0;
1020 			}
1021 			else if (io_apic_ints[x].src_bus_id == 0) {
1022 				io_apic_ints[x].src_bus_id = id;
1023 			}
1024 		}
1025 	}
1026 	/* sanity check if more than 1 PCI bus */
1027 	else if (num_pci_bus > 1) {
1028 		for (x = 0; x < mp_nbusses; ++x) {
1029 			if (bus_data[x].bus_type != PCI)
1030 				continue;
1031 			if (bus_data[x].bus_id >= num_pci_bus)
1032 				panic("bad PCI bus numbering");
1033 		}
1034 	}
1035 }
1036 
1037 
1038 static int
1039 processor_entry(proc_entry_ptr entry, int cpu)
1040 {
1041 	/* check for usability */
1042 	if ((cpu >= NCPU) || !(entry->cpu_flags & PROCENTRY_FLAG_EN))
1043 		return 0;
1044 
1045 	/* check for BSP flag */
1046 	if (entry->cpu_flags & PROCENTRY_FLAG_BP) {
1047 		boot_cpu_id = entry->apic_id;
1048 		CPU_TO_ID(0) = entry->apic_id;
1049 		ID_TO_CPU(entry->apic_id) = 0;
1050 		return 0;	/* its already been counted */
1051 	}
1052 
1053 	/* add another AP to list, if less than max number of CPUs */
1054 	else {
1055 		CPU_TO_ID(cpu) = entry->apic_id;
1056 		ID_TO_CPU(entry->apic_id) = cpu;
1057 		return 1;
1058 	}
1059 }
1060 
1061 
1062 static int
1063 bus_entry(bus_entry_ptr entry, int bus)
1064 {
1065 	int     x;
1066 	char    c, name[8];
1067 
1068 	/* encode the name into an index */
1069 	for (x = 0; x < 6; ++x) {
1070 		if ((c = entry->bus_type[x]) == ' ')
1071 			break;
1072 		name[x] = c;
1073 	}
1074 	name[x] = '\0';
1075 
1076 	if ((x = lookup_bus_type(name)) == UNKNOWN_BUSTYPE)
1077 		panic("unknown bus type: '%s'", name);
1078 
1079 	bus_data[bus].bus_id = entry->bus_id;
1080 	bus_data[bus].bus_type = x;
1081 
1082 	return 1;
1083 }
1084 
1085 
1086 static int
1087 io_apic_entry(io_apic_entry_ptr entry, int apic)
1088 {
1089 	if (!(entry->apic_flags & IOAPICENTRY_FLAG_EN))
1090 		return 0;
1091 
1092 	IO_TO_ID(apic) = entry->apic_id;
1093 	ID_TO_IO(entry->apic_id) = apic;
1094 
1095 	return 1;
1096 }
1097 
1098 
1099 static int
1100 lookup_bus_type(char *name)
1101 {
1102 	int     x;
1103 
1104 	for (x = 0; x < MAX_BUSTYPE; ++x)
1105 		if (strcmp(bus_type_table[x].name, name) == 0)
1106 			return bus_type_table[x].type;
1107 
1108 	return UNKNOWN_BUSTYPE;
1109 }
1110 
1111 
1112 static int
1113 int_entry(int_entry_ptr entry, int intr)
1114 {
1115 	io_apic_ints[intr].int_type = entry->int_type;
1116 	io_apic_ints[intr].int_flags = entry->int_flags;
1117 	io_apic_ints[intr].src_bus_id = entry->src_bus_id;
1118 	io_apic_ints[intr].src_bus_irq = entry->src_bus_irq;
1119 	io_apic_ints[intr].dst_apic_id = entry->dst_apic_id;
1120 	io_apic_ints[intr].dst_apic_int = entry->dst_apic_int;
1121 
1122 	return 1;
1123 }
1124 
1125 
1126 static int
1127 apic_int_is_bus_type(int intr, int bus_type)
1128 {
1129 	int     bus;
1130 
1131 	for (bus = 0; bus < mp_nbusses; ++bus)
1132 		if ((bus_data[bus].bus_id == io_apic_ints[intr].src_bus_id)
1133 		    && ((int) bus_data[bus].bus_type == bus_type))
1134 			return 1;
1135 
1136 	return 0;
1137 }
1138 
1139 
1140 /*
1141  * Given a traditional ISA INT mask, return an APIC mask.
1142  */
1143 u_int
1144 isa_apic_mask(u_int isa_mask)
1145 {
1146 	int isa_irq;
1147 	int apic_pin;
1148 
1149 #if defined(SKIP_IRQ15_REDIRECT)
1150 	if (isa_mask == (1 << 15)) {
1151 		printf("skipping ISA IRQ15 redirect\n");
1152 		return isa_mask;
1153 	}
1154 #endif  /* SKIP_IRQ15_REDIRECT */
1155 
1156 	isa_irq = ffs(isa_mask);		/* find its bit position */
1157 	if (isa_irq == 0)			/* doesn't exist */
1158 		return 0;
1159 	--isa_irq;				/* make it zero based */
1160 
1161 	apic_pin = isa_apic_pin(isa_irq);	/* look for APIC connection */
1162 	if (apic_pin == -1)
1163 		return 0;
1164 
1165 	return (1 << apic_pin);			/* convert pin# to a mask */
1166 }
1167 
1168 
1169 /*
1170  * Determine which APIC pin an ISA/EISA INT is attached to.
1171  */
1172 #define INTTYPE(I)	(io_apic_ints[(I)].int_type)
1173 #define INTPIN(I)	(io_apic_ints[(I)].dst_apic_int)
1174 
1175 #define SRCBUSIRQ(I)	(io_apic_ints[(I)].src_bus_irq)
1176 int
1177 isa_apic_pin(int isa_irq)
1178 {
1179 	int     intr;
1180 
1181 	for (intr = 0; intr < nintrs; ++intr) {		/* check each record */
1182 		if (INTTYPE(intr) == 0) {		/* standard INT */
1183 			if (SRCBUSIRQ(intr) == isa_irq) {
1184 				if (apic_int_is_bus_type(intr, ISA) ||
1185 			            apic_int_is_bus_type(intr, EISA))
1186 					return INTPIN(intr);	/* found */
1187 			}
1188 		}
1189 	}
1190 	return -1;					/* NOT found */
1191 }
1192 
1193 
1194 /*
1195  * Determine which APIC pin a PCI INT is attached to.
1196  */
1197 #define SRCBUSID(I)	(io_apic_ints[(I)].src_bus_id)
1198 #define SRCBUSDEVICE(I)	((io_apic_ints[(I)].src_bus_irq >> 2) & 0x1f)
1199 #define SRCBUSLINE(I)	(io_apic_ints[(I)].src_bus_irq & 0x03)
1200 int
1201 pci_apic_pin(int pciBus, int pciDevice, int pciInt)
1202 {
1203 	int     intr;
1204 
1205 	--pciInt;					/* zero based */
1206 
1207 	for (intr = 0; intr < nintrs; ++intr)		/* check each record */
1208 		if ((INTTYPE(intr) == 0)		/* standard INT */
1209 		    && (SRCBUSID(intr) == pciBus)
1210 		    && (SRCBUSDEVICE(intr) == pciDevice)
1211 		    && (SRCBUSLINE(intr) == pciInt))	/* a candidate IRQ */
1212 			if (apic_int_is_bus_type(intr, PCI))
1213 				return INTPIN(intr);	/* exact match */
1214 
1215 	return -1;					/* NOT found */
1216 }
1217 
1218 int
1219 next_apic_pin(int pin)
1220 {
1221 	int intr, ointr;
1222 	int bus, bustype;
1223 
1224 	bus = 0;
1225 	bustype = 0;
1226 	for (intr = 0; intr < nintrs; intr++) {
1227 		if (INTPIN(intr) != pin || INTTYPE(intr) != 0)
1228 			continue;
1229 		bus = SRCBUSID(intr);
1230 		bustype = apic_bus_type(bus);
1231 		if (bustype != ISA &&
1232 		    bustype != EISA &&
1233 		    bustype != PCI)
1234 			continue;
1235 		break;
1236 	}
1237 	if (intr >= nintrs) {
1238 		return -1;
1239 	}
1240 	for (ointr = intr + 1; ointr < nintrs; ointr++) {
1241 		if (INTTYPE(ointr) != 0)
1242 			continue;
1243 		if (bus != SRCBUSID(ointr))
1244 			continue;
1245 		if (bustype == PCI) {
1246 			if (SRCBUSDEVICE(intr) != SRCBUSDEVICE(ointr))
1247 				continue;
1248 			if (SRCBUSLINE(intr) != SRCBUSLINE(ointr))
1249 				continue;
1250 		}
1251 		if (bustype == ISA || bustype == EISA) {
1252 			if (SRCBUSIRQ(intr) != SRCBUSIRQ(ointr))
1253 				continue;
1254 		}
1255 		if (INTPIN(intr) == INTPIN(ointr))
1256 			continue;
1257 		break;
1258 	}
1259 	if (ointr >= nintrs) {
1260 		return -1;
1261 	}
1262 	return INTPIN(ointr);
1263 }
1264 #undef SRCBUSLINE
1265 #undef SRCBUSDEVICE
1266 #undef SRCBUSID
1267 #undef SRCBUSIRQ
1268 
1269 #undef INTPIN
1270 #undef INTTYPE
1271 
1272 
1273 /*
1274  * Reprogram the MB chipset to NOT redirect an ISA INTerrupt.
1275  *
1276  * XXX FIXME:
1277  *  Exactly what this means is unclear at this point.  It is a solution
1278  *  for motherboards that redirect the MBIRQ0 pin.  Generically a motherboard
1279  *  could route any of the ISA INTs to upper (>15) IRQ values.  But most would
1280  *  NOT be redirected via MBIRQ0, thus "undirect()ing" them would NOT be an
1281  *  option.
1282  */
1283 int
1284 undirect_isa_irq(int rirq)
1285 {
1286 #if defined(READY)
1287 	printf("Freeing redirected ISA irq %d.\n", rirq);
1288 	/** FIXME: tickle the MB redirector chip */
1289 	return ???;
1290 #else
1291 	printf("Freeing (NOT implemented) redirected ISA irq %d.\n", rirq);
1292 	return 0;
1293 #endif  /* READY */
1294 }
1295 
1296 
1297 /*
1298  * Reprogram the MB chipset to NOT redirect a PCI INTerrupt
1299  */
1300 int
1301 undirect_pci_irq(int rirq)
1302 {
1303 #if defined(READY)
1304 	if (bootverbose)
1305 		printf("Freeing redirected PCI irq %d.\n", rirq);
1306 
1307 	/** FIXME: tickle the MB redirector chip */
1308 	return ???;
1309 #else
1310 	if (bootverbose)
1311 		printf("Freeing (NOT implemented) redirected PCI irq %d.\n",
1312 		       rirq);
1313 	return 0;
1314 #endif  /* READY */
1315 }
1316 
1317 
1318 /*
1319  * given a bus ID, return:
1320  *  the bus type if found
1321  *  -1 if NOT found
1322  */
1323 int
1324 apic_bus_type(int id)
1325 {
1326 	int     x;
1327 
1328 	for (x = 0; x < mp_nbusses; ++x)
1329 		if (bus_data[x].bus_id == id)
1330 			return bus_data[x].bus_type;
1331 
1332 	return -1;
1333 }
1334 
1335 
1336 /*
1337  * given a LOGICAL APIC# and pin#, return:
1338  *  the associated src bus ID if found
1339  *  -1 if NOT found
1340  */
1341 int
1342 apic_src_bus_id(int apic, int pin)
1343 {
1344 	int     x;
1345 
1346 	/* search each of the possible INTerrupt sources */
1347 	for (x = 0; x < nintrs; ++x)
1348 		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1349 		    (pin == io_apic_ints[x].dst_apic_int))
1350 			return (io_apic_ints[x].src_bus_id);
1351 
1352 	return -1;		/* NOT found */
1353 }
1354 
1355 
1356 /*
1357  * given a LOGICAL APIC# and pin#, return:
1358  *  the associated src bus IRQ if found
1359  *  -1 if NOT found
1360  */
1361 int
1362 apic_src_bus_irq(int apic, int pin)
1363 {
1364 	int     x;
1365 
1366 	for (x = 0; x < nintrs; x++)
1367 		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1368 		    (pin == io_apic_ints[x].dst_apic_int))
1369 			return (io_apic_ints[x].src_bus_irq);
1370 
1371 	return -1;		/* NOT found */
1372 }
1373 
1374 
1375 /*
1376  * given a LOGICAL APIC# and pin#, return:
1377  *  the associated INTerrupt type if found
1378  *  -1 if NOT found
1379  */
1380 int
1381 apic_int_type(int apic, int pin)
1382 {
1383 	int     x;
1384 
1385 	/* search each of the possible INTerrupt sources */
1386 	for (x = 0; x < nintrs; ++x)
1387 		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1388 		    (pin == io_apic_ints[x].dst_apic_int))
1389 			return (io_apic_ints[x].int_type);
1390 
1391 	return -1;		/* NOT found */
1392 }
1393 
1394 
1395 /*
1396  * given a LOGICAL APIC# and pin#, return:
1397  *  the associated trigger mode if found
1398  *  -1 if NOT found
1399  */
1400 int
1401 apic_trigger(int apic, int pin)
1402 {
1403 	int     x;
1404 
1405 	/* search each of the possible INTerrupt sources */
1406 	for (x = 0; x < nintrs; ++x)
1407 		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1408 		    (pin == io_apic_ints[x].dst_apic_int))
1409 			return ((io_apic_ints[x].int_flags >> 2) & 0x03);
1410 
1411 	return -1;		/* NOT found */
1412 }
1413 
1414 
1415 /*
1416  * given a LOGICAL APIC# and pin#, return:
1417  *  the associated 'active' level if found
1418  *  -1 if NOT found
1419  */
1420 int
1421 apic_polarity(int apic, int pin)
1422 {
1423 	int     x;
1424 
1425 	/* search each of the possible INTerrupt sources */
1426 	for (x = 0; x < nintrs; ++x)
1427 		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1428 		    (pin == io_apic_ints[x].dst_apic_int))
1429 			return (io_apic_ints[x].int_flags & 0x03);
1430 
1431 	return -1;		/* NOT found */
1432 }
1433 
1434 
1435 /*
1436  * set data according to MP defaults
1437  * FIXME: probably not complete yet...
1438  */
1439 static void
1440 default_mp_table(int type)
1441 {
1442 	int     ap_cpu_id;
1443 #if defined(APIC_IO)
1444 	u_int32_t ux;
1445 	int     io_apic_id;
1446 	int     pin;
1447 #endif	/* APIC_IO */
1448 
1449 #if 0
1450 	printf("  MP default config type: %d\n", type);
1451 	switch (type) {
1452 	case 1:
1453 		printf("   bus: ISA, APIC: 82489DX\n");
1454 		break;
1455 	case 2:
1456 		printf("   bus: EISA, APIC: 82489DX\n");
1457 		break;
1458 	case 3:
1459 		printf("   bus: EISA, APIC: 82489DX\n");
1460 		break;
1461 	case 4:
1462 		printf("   bus: MCA, APIC: 82489DX\n");
1463 		break;
1464 	case 5:
1465 		printf("   bus: ISA+PCI, APIC: Integrated\n");
1466 		break;
1467 	case 6:
1468 		printf("   bus: EISA+PCI, APIC: Integrated\n");
1469 		break;
1470 	case 7:
1471 		printf("   bus: MCA+PCI, APIC: Integrated\n");
1472 		break;
1473 	default:
1474 		printf("   future type\n");
1475 		break;
1476 		/* NOTREACHED */
1477 	}
1478 #endif	/* 0 */
1479 
1480 	boot_cpu_id = (lapic.id & APIC_ID_MASK) >> 24;
1481 	ap_cpu_id = (boot_cpu_id == 0) ? 1 : 0;
1482 
1483 	/* BSP */
1484 	CPU_TO_ID(0) = boot_cpu_id;
1485 	ID_TO_CPU(boot_cpu_id) = 0;
1486 
1487 	/* one and only AP */
1488 	CPU_TO_ID(1) = ap_cpu_id;
1489 	ID_TO_CPU(ap_cpu_id) = 1;
1490 
1491 #if defined(APIC_IO)
1492 	/* one and only IO APIC */
1493 	io_apic_id = (io_apic_read(0, IOAPIC_ID) & APIC_ID_MASK) >> 24;
1494 
1495 	/*
1496 	 * sanity check, refer to MP spec section 3.6.6, last paragraph
1497 	 * necessary as some hardware isn't properly setting up the IO APIC
1498 	 */
1499 #if defined(REALLY_ANAL_IOAPICID_VALUE)
1500 	if (io_apic_id != 2) {
1501 #else
1502 	if ((io_apic_id == 0) || (io_apic_id == 1) || (io_apic_id == 15)) {
1503 #endif	/* REALLY_ANAL_IOAPICID_VALUE */
1504 		ux = io_apic_read(0, IOAPIC_ID);	/* get current contents */
1505 		ux &= ~APIC_ID_MASK;	/* clear the ID field */
1506 		ux |= 0x02000000;	/* set it to '2' */
1507 		io_apic_write(0, IOAPIC_ID, ux);	/* write new value */
1508 		ux = io_apic_read(0, IOAPIC_ID);	/* re-read && test */
1509 		if ((ux & APIC_ID_MASK) != 0x02000000)
1510 			panic("can't control IO APIC ID, reg: 0x%08x", ux);
1511 		io_apic_id = 2;
1512 	}
1513 	IO_TO_ID(0) = io_apic_id;
1514 	ID_TO_IO(io_apic_id) = 0;
1515 #endif	/* APIC_IO */
1516 
1517 	/* fill out bus entries */
1518 	switch (type) {
1519 	case 1:
1520 	case 2:
1521 	case 3:
1522 	case 5:
1523 	case 6:
1524 		bus_data[0].bus_id = default_data[type - 1][1];
1525 		bus_data[0].bus_type = default_data[type - 1][2];
1526 		bus_data[1].bus_id = default_data[type - 1][3];
1527 		bus_data[1].bus_type = default_data[type - 1][4];
1528 		break;
1529 
1530 	/* case 4: case 7:		   MCA NOT supported */
1531 	default:		/* illegal/reserved */
1532 		panic("BAD default MP config: %d", type);
1533 		/* NOTREACHED */
1534 	}
1535 
1536 #if defined(APIC_IO)
1537 	/* general cases from MP v1.4, table 5-2 */
1538 	for (pin = 0; pin < 16; ++pin) {
1539 		io_apic_ints[pin].int_type = 0;
1540 		io_apic_ints[pin].int_flags = 0x05;	/* edge/active-hi */
1541 		io_apic_ints[pin].src_bus_id = 0;
1542 		io_apic_ints[pin].src_bus_irq = pin;	/* IRQ2 caught below */
1543 		io_apic_ints[pin].dst_apic_id = io_apic_id;
1544 		io_apic_ints[pin].dst_apic_int = pin;	/* 1-to-1 */
1545 	}
1546 
1547 	/* special cases from MP v1.4, table 5-2 */
1548 	if (type == 2) {
1549 		io_apic_ints[2].int_type = 0xff;	/* N/C */
1550 		io_apic_ints[13].int_type = 0xff;	/* N/C */
1551 #if !defined(APIC_MIXED_MODE)
1552 		/** FIXME: ??? */
1553 		panic("sorry, can't support type 2 default yet");
1554 #endif	/* APIC_MIXED_MODE */
1555 	}
1556 	else
1557 		io_apic_ints[2].src_bus_irq = 0;	/* ISA IRQ0 is on APIC INT 2 */
1558 
1559 	if (type == 7)
1560 		io_apic_ints[0].int_type = 0xff;	/* N/C */
1561 	else
1562 		io_apic_ints[0].int_type = 3;	/* vectored 8259 */
1563 #endif	/* APIC_IO */
1564 }
1565 
1566 
1567 /*
1568  * initialize all the SMP locks
1569  */
1570 
1571 /* critical region around IO APIC, apic_imen */
1572 struct simplelock	imen_lock;
1573 
1574 /* critical region around splxx(), cpl, cml, cil, ipending */
1575 struct simplelock	cpl_lock;
1576 
1577 /* Make FAST_INTR() routines sequential */
1578 struct simplelock	fast_intr_lock;
1579 
1580 /* critical region around INTR() routines */
1581 struct simplelock	intr_lock;
1582 
1583 /* lock regions protected in UP kernel via cli/sti */
1584 struct simplelock	mpintr_lock;
1585 
1586 /* lock region used by kernel profiling */
1587 struct simplelock	mcount_lock;
1588 
1589 #ifdef USE_COMLOCK
1590 /* locks com (tty) data/hardware accesses: a FASTINTR() */
1591 struct simplelock	com_lock;
1592 #endif /* USE_COMLOCK */
1593 
1594 #ifdef USE_CLOCKLOCK
1595 /* lock regions around the clock hardware */
1596 struct simplelock	clock_lock;
1597 #endif /* USE_CLOCKLOCK */
1598 
1599 static void
1600 init_locks(void)
1601 {
1602 	/*
1603 	 * Get the initial mp_lock with a count of 1 for the BSP.
1604 	 * This uses a LOGICAL cpu ID, ie BSP == 0.
1605 	 */
1606 	mp_lock = 0x00000001;
1607 
1608 	/* ISR uses its own "giant lock" */
1609 	isr_lock = FREE_LOCK;
1610 
1611 #if defined(APIC_INTR_DIAGNOSTIC) && defined(APIC_INTR_DIAGNOSTIC_IRQ)
1612 	s_lock_init((struct simplelock*)&apic_itrace_debuglock);
1613 #endif
1614 
1615 	s_lock_init((struct simplelock*)&mpintr_lock);
1616 
1617 	s_lock_init((struct simplelock*)&mcount_lock);
1618 
1619 	s_lock_init((struct simplelock*)&fast_intr_lock);
1620 	s_lock_init((struct simplelock*)&intr_lock);
1621 	s_lock_init((struct simplelock*)&imen_lock);
1622 	s_lock_init((struct simplelock*)&cpl_lock);
1623 
1624 #ifdef USE_COMLOCK
1625 	s_lock_init((struct simplelock*)&com_lock);
1626 #endif /* USE_COMLOCK */
1627 #ifdef USE_CLOCKLOCK
1628 	s_lock_init((struct simplelock*)&clock_lock);
1629 #endif /* USE_CLOCKLOCK */
1630 }
1631 
1632 
1633 /*
1634  * start each AP in our list
1635  */
1636 static int
1637 start_all_aps(u_int boot_addr)
1638 {
1639 	int     x, i;
1640 	u_char  mpbiosreason;
1641 	u_long  mpbioswarmvec;
1642 	pd_entry_t *newptd;
1643 	pt_entry_t *newpt;
1644 	struct globaldata *gd;
1645 	char *stack;
1646 	pd_entry_t	*myPTD;
1647 
1648 	POSTCODE(START_ALL_APS_POST);
1649 
1650 	/* initialize BSP's local APIC */
1651 	apic_initialize();
1652 	bsp_apic_ready = 1;
1653 
1654 	/* install the AP 1st level boot code */
1655 	install_ap_tramp(boot_addr);
1656 
1657 
1658 	/* save the current value of the warm-start vector */
1659 	mpbioswarmvec = *((u_long *) WARMBOOT_OFF);
1660 	outb(CMOS_REG, BIOS_RESET);
1661 	mpbiosreason = inb(CMOS_DATA);
1662 
1663 	/* record BSP in CPU map */
1664 	all_cpus = 1;
1665 
1666 	/* start each AP */
1667 	for (x = 1; x <= mp_naps; ++x) {
1668 
1669 		/* This is a bit verbose, it will go away soon.  */
1670 
1671 		/* alloc new page table directory */
1672 		newptd = (pd_entry_t *)(kmem_alloc(kernel_map, PAGE_SIZE));
1673 
1674 		/* Store the virtual PTD address for this CPU */
1675 		IdlePTDS[x] = newptd;
1676 
1677 		/* clone currently active one (ie: IdlePTD) */
1678 		bcopy(PTD, newptd, PAGE_SIZE);	/* inc prv page pde */
1679 
1680 		/* set up 0 -> 4MB P==V mapping for AP boot */
1681 		newptd[0] = (pd_entry_t) (PG_V | PG_RW |
1682 						((u_long)KPTphys & PG_FRAME));
1683 
1684 		/* store PTD for this AP's boot sequence */
1685 		myPTD = (pd_entry_t *)vtophys(newptd);
1686 
1687 		/* alloc new page table page */
1688 		newpt = (pt_entry_t *)(kmem_alloc(kernel_map, PAGE_SIZE));
1689 
1690 		/* set the new PTD's private page to point there */
1691 		newptd[MPPTDI] = (pt_entry_t)(PG_V | PG_RW | vtophys(newpt));
1692 
1693 		/* install self referential entry */
1694 		newptd[PTDPTDI] = (pd_entry_t)(PG_V | PG_RW | vtophys(newptd));
1695 
1696 		/* allocate a new private data page */
1697 		gd = (struct globaldata *)kmem_alloc(kernel_map, PAGE_SIZE);
1698 
1699 		/* wire it into the private page table page */
1700 		newpt[0] = (pt_entry_t)(PG_V | PG_RW | vtophys(gd));
1701 
1702 		/* wire the ptp into itself for access */
1703 		newpt[1] = (pt_entry_t)(PG_V | PG_RW | vtophys(newpt));
1704 
1705 		/* copy in the pointer to the local apic */
1706 		newpt[2] = SMP_prvpt[2];
1707 
1708 		/* and the IO apic mapping[s] */
1709 		for (i = 16; i < 32; i++)
1710 			newpt[i] = SMP_prvpt[i];
1711 
1712 		/* allocate and set up an idle stack data page */
1713 		stack = (char *)kmem_alloc(kernel_map, UPAGES*PAGE_SIZE);
1714 		for (i = 0; i < UPAGES; i++)
1715 			newpt[i + 3] = (pt_entry_t)(PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack));
1716 
1717 		newpt[3 + UPAGES] = 0;		/* *prv_CMAP1 */
1718 		newpt[4 + UPAGES] = 0;		/* *prv_CMAP2 */
1719 		newpt[5 + UPAGES] = 0;		/* *prv_CMAP3 */
1720 		newpt[6 + UPAGES] = 0;		/* *prv_PMAP1 */
1721 
1722 		/* prime data page for it to use */
1723 		gd->cpuid = x;
1724 		gd->cpu_lockid = x << 24;
1725 		gd->my_idlePTD = myPTD;
1726 		gd->prv_CMAP1 = &newpt[3 + UPAGES];
1727 		gd->prv_CMAP2 = &newpt[4 + UPAGES];
1728 		gd->prv_CMAP3 = &newpt[5 + UPAGES];
1729 		gd->prv_PMAP1 = &newpt[6 + UPAGES];
1730 
1731 		/* setup a vector to our boot code */
1732 		*((volatile u_short *) WARMBOOT_OFF) = WARMBOOT_TARGET;
1733 		*((volatile u_short *) WARMBOOT_SEG) = (boot_addr >> 4);
1734 		outb(CMOS_REG, BIOS_RESET);
1735 		outb(CMOS_DATA, BIOS_WARM);	/* 'warm-start' */
1736 
1737 		bootPTD = myPTD;
1738 		/* attempt to start the Application Processor */
1739 		CHECK_INIT(99);	/* setup checkpoints */
1740 		if (!start_ap(x, boot_addr)) {
1741 			printf("AP #%d (PHY# %d) failed!\n", x, CPU_TO_ID(x));
1742 			CHECK_PRINT("trace");	/* show checkpoints */
1743 			/* better panic as the AP may be running loose */
1744 			printf("panic y/n? [y] ");
1745 			if (cngetc() != 'n')
1746 				panic("bye-bye");
1747 		}
1748 		CHECK_PRINT("trace");		/* show checkpoints */
1749 
1750 		/* record its version info */
1751 		cpu_apic_versions[x] = cpu_apic_versions[0];
1752 
1753 		all_cpus |= (1 << x);		/* record AP in CPU map */
1754 	}
1755 
1756 	/* build our map of 'other' CPUs */
1757 	other_cpus = all_cpus & ~(1 << cpuid);
1758 
1759 	/* fill in our (BSP) APIC version */
1760 	cpu_apic_versions[0] = lapic.version;
1761 
1762 	/* restore the warmstart vector */
1763 	*(u_long *) WARMBOOT_OFF = mpbioswarmvec;
1764 	outb(CMOS_REG, BIOS_RESET);
1765 	outb(CMOS_DATA, mpbiosreason);
1766 
1767 	/*
1768 	 * Set up the idle context for the BSP.  Similar to above except
1769 	 * that some was done by locore, some by pmap.c and some is implicit
1770 	 * because the BSP is cpu#0 and the page is initially zero, and also
1771 	 * because we can refer to variables by name on the BSP..
1772 	 */
1773 	newptd = (pd_entry_t *)(kmem_alloc(kernel_map, PAGE_SIZE));
1774 
1775 	bcopy(PTD, newptd, PAGE_SIZE);	/* inc prv page pde */
1776 	IdlePTDS[0] = newptd;
1777 
1778 	/* Point PTD[] to this page instead of IdlePTD's physical page */
1779 	newptd[PTDPTDI] = (pd_entry_t)(PG_V | PG_RW | vtophys(newptd));
1780 
1781 	my_idlePTD = (pd_entry_t *)vtophys(newptd);
1782 
1783 	/* Allocate and setup BSP idle stack */
1784 	stack = (char *)kmem_alloc(kernel_map, UPAGES * PAGE_SIZE);
1785 	for (i = 0; i < UPAGES; i++)
1786 		SMP_prvpt[i + 3] = (pt_entry_t)(PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack));
1787 
1788 	pmap_set_opt_bsp();
1789 
1790 	for (i = 0; i < mp_ncpus; i++) {
1791 		bcopy( (int *) PTD + KPTDI, (int *) IdlePTDS[i] + KPTDI, NKPDE * sizeof (int));
1792 	}
1793 
1794 	/* number of APs actually started */
1795 	return mp_ncpus - 1;
1796 }
1797 
1798 
1799 /*
1800  * load the 1st level AP boot code into base memory.
1801  */
1802 
1803 /* targets for relocation */
1804 extern void bigJump(void);
1805 extern void bootCodeSeg(void);
1806 extern void bootDataSeg(void);
1807 extern void MPentry(void);
1808 extern u_int MP_GDT;
1809 extern u_int mp_gdtbase;
1810 
1811 static void
1812 install_ap_tramp(u_int boot_addr)
1813 {
1814 	int     x;
1815 	int     size = *(int *) ((u_long) & bootMP_size);
1816 	u_char *src = (u_char *) ((u_long) bootMP);
1817 	u_char *dst = (u_char *) boot_addr + KERNBASE;
1818 	u_int   boot_base = (u_int) bootMP;
1819 	u_int8_t *dst8;
1820 	u_int16_t *dst16;
1821 	u_int32_t *dst32;
1822 
1823 	POSTCODE(INSTALL_AP_TRAMP_POST);
1824 
1825 	for (x = 0; x < size; ++x)
1826 		*dst++ = *src++;
1827 
1828 	/*
1829 	 * modify addresses in code we just moved to basemem. unfortunately we
1830 	 * need fairly detailed info about mpboot.s for this to work.  changes
1831 	 * to mpboot.s might require changes here.
1832 	 */
1833 
1834 	/* boot code is located in KERNEL space */
1835 	dst = (u_char *) boot_addr + KERNBASE;
1836 
1837 	/* modify the lgdt arg */
1838 	dst32 = (u_int32_t *) (dst + ((u_int) & mp_gdtbase - boot_base));
1839 	*dst32 = boot_addr + ((u_int) & MP_GDT - boot_base);
1840 
1841 	/* modify the ljmp target for MPentry() */
1842 	dst32 = (u_int32_t *) (dst + ((u_int) bigJump - boot_base) + 1);
1843 	*dst32 = ((u_int) MPentry - KERNBASE);
1844 
1845 	/* modify the target for boot code segment */
1846 	dst16 = (u_int16_t *) (dst + ((u_int) bootCodeSeg - boot_base));
1847 	dst8 = (u_int8_t *) (dst16 + 1);
1848 	*dst16 = (u_int) boot_addr & 0xffff;
1849 	*dst8 = ((u_int) boot_addr >> 16) & 0xff;
1850 
1851 	/* modify the target for boot data segment */
1852 	dst16 = (u_int16_t *) (dst + ((u_int) bootDataSeg - boot_base));
1853 	dst8 = (u_int8_t *) (dst16 + 1);
1854 	*dst16 = (u_int) boot_addr & 0xffff;
1855 	*dst8 = ((u_int) boot_addr >> 16) & 0xff;
1856 }
1857 
1858 
1859 /*
1860  * this function starts the AP (application processor) identified
1861  * by the APIC ID 'physicalCpu'.  It does quite a "song and dance"
1862  * to accomplish this.  This is necessary because of the nuances
1863  * of the different hardware we might encounter.  It ain't pretty,
1864  * but it seems to work.
1865  */
1866 static int
1867 start_ap(int logical_cpu, u_int boot_addr)
1868 {
1869 	int     physical_cpu;
1870 	int     vector;
1871 	int     cpus;
1872 	u_long  icr_lo, icr_hi;
1873 
1874 	POSTCODE(START_AP_POST);
1875 
1876 	/* get the PHYSICAL APIC ID# */
1877 	physical_cpu = CPU_TO_ID(logical_cpu);
1878 
1879 	/* calculate the vector */
1880 	vector = (boot_addr >> 12) & 0xff;
1881 
1882 	/* used as a watchpoint to signal AP startup */
1883 	cpus = mp_ncpus;
1884 
1885 	/*
1886 	 * first we do an INIT/RESET IPI this INIT IPI might be run, reseting
1887 	 * and running the target CPU. OR this INIT IPI might be latched (P5
1888 	 * bug), CPU waiting for STARTUP IPI. OR this INIT IPI might be
1889 	 * ignored.
1890 	 */
1891 
1892 	/* setup the address for the target AP */
1893 	icr_hi = lapic.icr_hi & ~APIC_ID_MASK;
1894 	icr_hi |= (physical_cpu << 24);
1895 	lapic.icr_hi = icr_hi;
1896 
1897 	/* do an INIT IPI: assert RESET */
1898 	icr_lo = lapic.icr_lo & 0xfff00000;
1899 	lapic.icr_lo = icr_lo | 0x0000c500;
1900 
1901 	/* wait for pending status end */
1902 	while (lapic.icr_lo & APIC_DELSTAT_MASK)
1903 		 /* spin */ ;
1904 
1905 	/* do an INIT IPI: deassert RESET */
1906 	lapic.icr_lo = icr_lo | 0x00008500;
1907 
1908 	/* wait for pending status end */
1909 	u_sleep(10000);		/* wait ~10mS */
1910 	while (lapic.icr_lo & APIC_DELSTAT_MASK)
1911 		 /* spin */ ;
1912 
1913 	/*
1914 	 * next we do a STARTUP IPI: the previous INIT IPI might still be
1915 	 * latched, (P5 bug) this 1st STARTUP would then terminate
1916 	 * immediately, and the previously started INIT IPI would continue. OR
1917 	 * the previous INIT IPI has already run. and this STARTUP IPI will
1918 	 * run. OR the previous INIT IPI was ignored. and this STARTUP IPI
1919 	 * will run.
1920 	 */
1921 
1922 	/* do a STARTUP IPI */
1923 	lapic.icr_lo = icr_lo | 0x00000600 | vector;
1924 	while (lapic.icr_lo & APIC_DELSTAT_MASK)
1925 		 /* spin */ ;
1926 	u_sleep(200);		/* wait ~200uS */
1927 
1928 	/*
1929 	 * finally we do a 2nd STARTUP IPI: this 2nd STARTUP IPI should run IF
1930 	 * the previous STARTUP IPI was cancelled by a latched INIT IPI. OR
1931 	 * this STARTUP IPI will be ignored, as only ONE STARTUP IPI is
1932 	 * recognized after hardware RESET or INIT IPI.
1933 	 */
1934 
1935 	lapic.icr_lo = icr_lo | 0x00000600 | vector;
1936 	while (lapic.icr_lo & APIC_DELSTAT_MASK)
1937 		 /* spin */ ;
1938 	u_sleep(200);		/* wait ~200uS */
1939 
1940 	/* wait for it to start */
1941 	set_apic_timer(5000000);/* == 5 seconds */
1942 	while (read_apic_timer())
1943 		if (mp_ncpus > cpus)
1944 			return 1;	/* return SUCCESS */
1945 
1946 	return 0;		/* return FAILURE */
1947 }
1948 
1949 
1950 /*
1951  * Flush the TLB on all other CPU's
1952  *
1953  * XXX: Needs to handshake and wait for completion before proceding.
1954  */
1955 void
1956 smp_invltlb(void)
1957 {
1958 #if defined(APIC_IO)
1959 	if (smp_started && invltlb_ok)
1960 		all_but_self_ipi(XINVLTLB_OFFSET);
1961 #endif  /* APIC_IO */
1962 }
1963 
1964 void
1965 invlpg(u_int addr)
1966 {
1967 	__asm   __volatile("invlpg (%0)"::"r"(addr):"memory");
1968 
1969 	/* send a message to the other CPUs */
1970 	smp_invltlb();
1971 }
1972 
1973 void
1974 invltlb(void)
1975 {
1976 	u_long  temp;
1977 
1978 	/*
1979 	 * This should be implemented as load_cr3(rcr3()) when load_cr3() is
1980 	 * inlined.
1981 	 */
1982 	__asm __volatile("movl %%cr3, %0; movl %0, %%cr3":"=r"(temp) :: "memory");
1983 
1984 	/* send a message to the other CPUs */
1985 	smp_invltlb();
1986 }
1987 
1988 
1989 /*
1990  * When called the executing CPU will send an IPI to all other CPUs
1991  *  requesting that they halt execution.
1992  *
1993  * Usually (but not necessarily) called with 'other_cpus' as its arg.
1994  *
1995  *  - Signals all CPUs in map to stop.
1996  *  - Waits for each to stop.
1997  *
1998  * Returns:
1999  *  -1: error
2000  *   0: NA
2001  *   1: ok
2002  *
2003  * XXX FIXME: this is not MP-safe, needs a lock to prevent multiple CPUs
2004  *            from executing at same time.
2005  */
2006 int
2007 stop_cpus(u_int map)
2008 {
2009 	if (!smp_started)
2010 		return 0;
2011 
2012 	/* send IPI to all CPUs in map */
2013 	stopped_cpus = 0;
2014 
2015 	/* send the Xcpustop IPI to all CPUs in map */
2016 	selected_apic_ipi(map, XCPUSTOP_OFFSET, APIC_DELMODE_FIXED);
2017 
2018 	while (stopped_cpus != map)
2019 		/* spin */ ;
2020 
2021 	return 1;
2022 }
2023 
2024 
2025 /*
2026  * Called by a CPU to restart stopped CPUs.
2027  *
2028  * Usually (but not necessarily) called with 'stopped_cpus' as its arg.
2029  *
2030  *  - Signals all CPUs in map to restart.
2031  *  - Waits for each to restart.
2032  *
2033  * Returns:
2034  *  -1: error
2035  *   0: NA
2036  *   1: ok
2037  */
2038 int
2039 restart_cpus(u_int map)
2040 {
2041 	if (!smp_started)
2042 		return 0;
2043 
2044 	started_cpus = map;		/* signal other cpus to restart */
2045 
2046 	while (started_cpus)		/* wait for each to clear its bit */
2047 		/* spin */ ;
2048 	stopped_cpus = 0;
2049 
2050 	return 1;
2051 }
2052 
2053 int smp_active = 0;	/* are the APs allowed to run? */
2054 SYSCTL_INT(_machdep, OID_AUTO, smp_active, CTLFLAG_RW, &smp_active, 0, "");
2055 
2056 /* XXX maybe should be hw.ncpu */
2057 static int smp_cpus = 1;	/* how many cpu's running */
2058 SYSCTL_INT(_machdep, OID_AUTO, smp_cpus, CTLFLAG_RD, &smp_cpus, 0, "");
2059 
2060 int invltlb_ok = 0;	/* throttle smp_invltlb() till safe */
2061 SYSCTL_INT(_machdep, OID_AUTO, invltlb_ok, CTLFLAG_RW, &invltlb_ok, 0, "");
2062 
2063 /* Warning: Do not staticize.  Used from swtch.s */
2064 int do_page_zero_idle = 1; /* bzero pages for fun and profit in idleloop */
2065 SYSCTL_INT(_machdep, OID_AUTO, do_page_zero_idle, CTLFLAG_RW,
2066 	   &do_page_zero_idle, 0, "");
2067 
2068 /* Is forwarding of a interrupt to the CPU holding the ISR lock enabled ? */
2069 int forward_irq_enabled = 1;
2070 SYSCTL_INT(_machdep, OID_AUTO, forward_irq_enabled, CTLFLAG_RW,
2071 	   &forward_irq_enabled, 0, "");
2072 
2073 /* Enable forwarding of a signal to a process running on a different CPU */
2074 int forward_signal_enabled = 1;
2075 SYSCTL_INT(_machdep, OID_AUTO, forward_signal_enabled, CTLFLAG_RW,
2076 	   &forward_signal_enabled, 0, "");
2077 
2078 /*
2079  * This is called once the rest of the system is up and running and we're
2080  * ready to let the AP's out of the pen.
2081  */
2082 void ap_init(void);
2083 
2084 void
2085 ap_init()
2086 {
2087 	u_int   temp;
2088 	u_int	apic_id;
2089 
2090 	smp_cpus++;
2091 
2092 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
2093 	lidt(&r_idt);
2094 #endif
2095 
2096 	/* Build our map of 'other' CPUs. */
2097 	other_cpus = all_cpus & ~(1 << cpuid);
2098 
2099 	printf("SMP: AP CPU #%d Launched!\n", cpuid);
2100 
2101 	/* XXX FIXME: i386 specific, and redundant: Setup the FPU. */
2102 	load_cr0((rcr0() & ~CR0_EM) | CR0_MP | CR0_NE | CR0_TS);
2103 
2104 	/* A quick check from sanity claus */
2105 	apic_id = (apic_id_to_logical[(lapic.id & 0x0f000000) >> 24]);
2106 	if (cpuid != apic_id) {
2107 		printf("SMP: cpuid = %d\n", cpuid);
2108 		printf("SMP: apic_id = %d\n", apic_id);
2109 		printf("PTD[MPPTDI] = %08x\n", PTD[MPPTDI]);
2110 		panic("cpuid mismatch! boom!!");
2111 	}
2112 
2113 	getmtrr();
2114 
2115 	/* Init local apic for irq's */
2116 	apic_initialize();
2117 
2118 	/*
2119 	 * Activate smp_invltlb, although strictly speaking, this isn't
2120 	 * quite correct yet.  We should have a bitfield for cpus willing
2121 	 * to accept TLB flush IPI's or something and sync them.
2122 	 */
2123 	invltlb_ok = 1;
2124 	smp_started = 1;	/* enable IPI's, tlb shootdown, freezes etc */
2125 	smp_active = 1;		/* historic */
2126 
2127 	curproc = NULL;		/* make sure */
2128 }
2129 
2130 #ifdef BETTER_CLOCK
2131 
2132 #define CHECKSTATE_USER	0
2133 #define CHECKSTATE_SYS	1
2134 #define CHECKSTATE_INTR	2
2135 
2136 /* Do not staticize.  Used from apic_vector.s */
2137 struct proc*	checkstate_curproc[NCPU];
2138 int		checkstate_cpustate[NCPU];
2139 u_long		checkstate_pc[NCPU];
2140 
2141 extern long	cp_time[CPUSTATES];
2142 
2143 #define PC_TO_INDEX(pc, prof)				\
2144         ((int)(((u_quad_t)((pc) - (prof)->pr_off) *	\
2145             (u_quad_t)((prof)->pr_scale)) >> 16) & ~1)
2146 
2147 static void
2148 addupc_intr_forwarded(struct proc *p, int id, int *astmap)
2149 {
2150 	int i;
2151 	struct uprof *prof;
2152 	u_long pc;
2153 
2154 	pc = checkstate_pc[id];
2155 	prof = &p->p_stats->p_prof;
2156 	if (pc >= prof->pr_off &&
2157 	    (i = PC_TO_INDEX(pc, prof)) < prof->pr_size) {
2158 		if ((p->p_flag & P_OWEUPC) == 0) {
2159 			prof->pr_addr = pc;
2160 			prof->pr_ticks = 1;
2161 			p->p_flag |= P_OWEUPC;
2162 		}
2163 		*astmap |= (1 << id);
2164 	}
2165 }
2166 
2167 static void
2168 forwarded_statclock(int id, int pscnt, int *astmap)
2169 {
2170 	struct pstats *pstats;
2171 	long rss;
2172 	struct rusage *ru;
2173 	struct vmspace *vm;
2174 	int cpustate;
2175 	struct proc *p;
2176 #ifdef GPROF
2177 	register struct gmonparam *g;
2178 	int i;
2179 #endif
2180 
2181 	p = checkstate_curproc[id];
2182 	cpustate = checkstate_cpustate[id];
2183 
2184 	switch (cpustate) {
2185 	case CHECKSTATE_USER:
2186 		if (p->p_flag & P_PROFIL)
2187 			addupc_intr_forwarded(p, id, astmap);
2188 		if (pscnt > 1)
2189 			return;
2190 		p->p_uticks++;
2191 		if (p->p_nice > NZERO)
2192 			cp_time[CP_NICE]++;
2193 		else
2194 			cp_time[CP_USER]++;
2195 		break;
2196 	case CHECKSTATE_SYS:
2197 #ifdef GPROF
2198 		/*
2199 		 * Kernel statistics are just like addupc_intr, only easier.
2200 		 */
2201 		g = &_gmonparam;
2202 		if (g->state == GMON_PROF_ON) {
2203 			i = checkstate_pc[id] - g->lowpc;
2204 			if (i < g->textsize) {
2205 				i /= HISTFRACTION * sizeof(*g->kcount);
2206 				g->kcount[i]++;
2207 			}
2208 		}
2209 #endif
2210 		if (pscnt > 1)
2211 			return;
2212 
2213 		if (!p)
2214 			cp_time[CP_IDLE]++;
2215 		else {
2216 			p->p_sticks++;
2217 			cp_time[CP_SYS]++;
2218 		}
2219 		break;
2220 	case CHECKSTATE_INTR:
2221 	default:
2222 #ifdef GPROF
2223 		/*
2224 		 * Kernel statistics are just like addupc_intr, only easier.
2225 		 */
2226 		g = &_gmonparam;
2227 		if (g->state == GMON_PROF_ON) {
2228 			i = checkstate_pc[id] - g->lowpc;
2229 			if (i < g->textsize) {
2230 				i /= HISTFRACTION * sizeof(*g->kcount);
2231 				g->kcount[i]++;
2232 			}
2233 		}
2234 #endif
2235 		if (pscnt > 1)
2236 			return;
2237 		if (p)
2238 			p->p_iticks++;
2239 		cp_time[CP_INTR]++;
2240 	}
2241 	if (p != NULL) {
2242 		p->p_cpticks++;
2243 		if (++p->p_estcpu == 0)
2244 			p->p_estcpu--;
2245 		if ((p->p_estcpu & 3) == 0) {
2246 			resetpriority(p);
2247 			if (p->p_priority >= PUSER)
2248 				p->p_priority = p->p_usrpri;
2249 		}
2250 
2251 		/* Update resource usage integrals and maximums. */
2252 		if ((pstats = p->p_stats) != NULL &&
2253 		    (ru = &pstats->p_ru) != NULL &&
2254 		    (vm = p->p_vmspace) != NULL) {
2255 			ru->ru_ixrss += vm->vm_tsize * PAGE_SIZE / 1024;
2256 			ru->ru_idrss += vm->vm_dsize * PAGE_SIZE / 1024;
2257 			ru->ru_isrss += vm->vm_ssize * PAGE_SIZE / 1024;
2258 			rss = vm->vm_pmap.pm_stats.resident_count *
2259 				PAGE_SIZE / 1024;
2260 			if (ru->ru_maxrss < rss)
2261 				ru->ru_maxrss = rss;
2262         	}
2263 	}
2264 }
2265 
2266 void
2267 forward_statclock(int pscnt)
2268 {
2269 	int map;
2270 	int id;
2271 	int i;
2272 
2273 	/* Kludge. We don't yet have separate locks for the interrupts
2274 	 * and the kernel. This means that we cannot let the other processors
2275 	 * handle complex interrupts while inhibiting them from entering
2276 	 * the kernel in a non-interrupt context.
2277 	 *
2278 	 * What we can do, without changing the locking mechanisms yet,
2279 	 * is letting the other processors handle a very simple interrupt
2280 	 * (wich determines the processor states), and do the main
2281 	 * work ourself.
2282 	 */
2283 
2284 	if (!smp_started || !invltlb_ok || cold || panicstr)
2285 		return;
2286 
2287 	/* Step 1: Probe state   (user, cpu, interrupt, spinlock, idle ) */
2288 
2289 	map = other_cpus & ~stopped_cpus ;
2290 	checkstate_probed_cpus = 0;
2291 	if (map != 0)
2292 		selected_apic_ipi(map,
2293 				  XCPUCHECKSTATE_OFFSET, APIC_DELMODE_FIXED);
2294 
2295 	i = 0;
2296 	while (checkstate_probed_cpus != map) {
2297 		/* spin */
2298 		i++;
2299 		if (i == 1000000) {
2300 			printf("forward_statclock: checkstate %x\n",
2301 			       checkstate_probed_cpus);
2302 			break;
2303 		}
2304 	}
2305 
2306 	/*
2307 	 * Step 2: walk through other processors processes, update ticks and
2308 	 * profiling info.
2309 	 */
2310 
2311 	map = 0;
2312 	for (id = 0; id < mp_ncpus; id++) {
2313 		if (id == cpuid)
2314 			continue;
2315 		if (((1 << id) & checkstate_probed_cpus) == 0)
2316 			continue;
2317 		forwarded_statclock(id, pscnt, &map);
2318 	}
2319 	if (map != 0) {
2320 		checkstate_need_ast |= map;
2321 		selected_apic_ipi(map, XCPUAST_OFFSET, APIC_DELMODE_FIXED);
2322 		i = 0;
2323 		while ((checkstate_need_ast & map) != 0) {
2324 			/* spin */
2325 			i++;
2326 			if (i > 100000) {
2327 #ifdef BETTER_CLOCK_DIAGNOSTIC
2328 				printf("forward_statclock: dropped ast 0x%x\n",
2329 				       checkstate_need_ast & map);
2330 #endif
2331 				break;
2332 			}
2333 		}
2334 	}
2335 }
2336 
2337 void
2338 forward_hardclock(int pscnt)
2339 {
2340 	int map;
2341 	int id;
2342 	struct proc *p;
2343 	struct pstats *pstats;
2344 	int i;
2345 
2346 	/* Kludge. We don't yet have separate locks for the interrupts
2347 	 * and the kernel. This means that we cannot let the other processors
2348 	 * handle complex interrupts while inhibiting them from entering
2349 	 * the kernel in a non-interrupt context.
2350 	 *
2351 	 * What we can do, without changing the locking mechanisms yet,
2352 	 * is letting the other processors handle a very simple interrupt
2353 	 * (wich determines the processor states), and do the main
2354 	 * work ourself.
2355 	 */
2356 
2357 	if (!smp_started || !invltlb_ok || cold || panicstr)
2358 		return;
2359 
2360 	/* Step 1: Probe state   (user, cpu, interrupt, spinlock, idle) */
2361 
2362 	map = other_cpus & ~stopped_cpus ;
2363 	checkstate_probed_cpus = 0;
2364 	if (map != 0)
2365 		selected_apic_ipi(map,
2366 				  XCPUCHECKSTATE_OFFSET, APIC_DELMODE_FIXED);
2367 
2368 	i = 0;
2369 	while (checkstate_probed_cpus != map) {
2370 		/* spin */
2371 		i++;
2372 		if (i == 1000000) {
2373 			printf("forward_hardclock: checkstate %x\n",
2374 			       checkstate_probed_cpus);
2375 			break;
2376 		}
2377 	}
2378 
2379 	/*
2380 	 * Step 2: walk through other processors processes, update virtual
2381 	 * timer and profiling timer. If stathz == 0, also update ticks and
2382 	 * profiling info.
2383 	 */
2384 
2385 	map = 0;
2386 	for (id = 0; id < mp_ncpus; id++) {
2387 		if (id == cpuid)
2388 			continue;
2389 		if (((1 << id) & checkstate_probed_cpus) == 0)
2390 			continue;
2391 		p = checkstate_curproc[id];
2392 		if (p) {
2393 			pstats = p->p_stats;
2394 			if (checkstate_cpustate[id] == CHECKSTATE_USER &&
2395 			    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) &&
2396 			    itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) {
2397 				psignal(p, SIGVTALRM);
2398 				map |= (1 << id);
2399 			}
2400 			if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) &&
2401 			    itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) {
2402 				psignal(p, SIGPROF);
2403 				map |= (1 << id);
2404 			}
2405 		}
2406 		if (stathz == 0) {
2407 			forwarded_statclock( id, pscnt, &map);
2408 		}
2409 	}
2410 	if (map != 0) {
2411 		checkstate_need_ast |= map;
2412 		selected_apic_ipi(map, XCPUAST_OFFSET, APIC_DELMODE_FIXED);
2413 		i = 0;
2414 		while ((checkstate_need_ast & map) != 0) {
2415 			/* spin */
2416 			i++;
2417 			if (i > 100000) {
2418 #ifdef BETTER_CLOCK_DIAGNOSTIC
2419 				printf("forward_hardclock: dropped ast 0x%x\n",
2420 				       checkstate_need_ast & map);
2421 #endif
2422 				break;
2423 			}
2424 		}
2425 	}
2426 }
2427 
2428 #endif /* BETTER_CLOCK */
2429 
2430 void
2431 forward_signal(struct proc *p)
2432 {
2433 	int map;
2434 	int id;
2435 	int i;
2436 
2437 	/* Kludge. We don't yet have separate locks for the interrupts
2438 	 * and the kernel. This means that we cannot let the other processors
2439 	 * handle complex interrupts while inhibiting them from entering
2440 	 * the kernel in a non-interrupt context.
2441 	 *
2442 	 * What we can do, without changing the locking mechanisms yet,
2443 	 * is letting the other processors handle a very simple interrupt
2444 	 * (wich determines the processor states), and do the main
2445 	 * work ourself.
2446 	 */
2447 
2448 	if (!smp_started || !invltlb_ok || cold || panicstr)
2449 		return;
2450 	if (!forward_signal_enabled)
2451 		return;
2452 	while (1) {
2453 		if (p->p_stat != SRUN)
2454 			return;
2455 		id = (u_char) p->p_oncpu;
2456 		if (id == 0xff)
2457 			return;
2458 		map = (1<<id);
2459 		checkstate_need_ast |= map;
2460 		selected_apic_ipi(map, XCPUAST_OFFSET, APIC_DELMODE_FIXED);
2461 		i = 0;
2462 		while ((checkstate_need_ast & map) != 0) {
2463 			/* spin */
2464 			i++;
2465 			if (i > 100000) {
2466 #if 0
2467 				printf("forward_signal: dropped ast 0x%x\n",
2468 				       checkstate_need_ast & map);
2469 #endif
2470 				break;
2471 			}
2472 		}
2473 		if (id == (u_char) p->p_oncpu)
2474 			return;
2475 	}
2476 }
2477 
2478 
2479 #ifdef APIC_INTR_REORDER
2480 /*
2481  *	Maintain mapping from softintr vector to isr bit in local apic.
2482  */
2483 void
2484 set_lapic_isrloc(int intr, int vector)
2485 {
2486 	if (intr < 0 || intr > 32)
2487 		panic("set_apic_isrloc: bad intr argument: %d",intr);
2488 	if (vector < ICU_OFFSET || vector > 255)
2489 		panic("set_apic_isrloc: bad vector argument: %d",vector);
2490 	apic_isrbit_location[intr].location = &lapic.isr0 + ((vector>>5)<<2);
2491 	apic_isrbit_location[intr].bit = (1<<(vector & 31));
2492 }
2493 #endif
2494