xref: /freebsd/sys/kern/subr_smp.c (revision 23f282aa31e9b6fceacd449020e936e98d6f2298)
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_smp.h"
29 #include "opt_cpu.h"
30 #include "opt_user_ldt.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 #include <sys/malloc.h>
44 #include <sys/memrange.h>
45 #ifdef BETTER_CLOCK
46 #include <sys/dkstat.h>
47 #endif
48 #include <sys/cons.h>	/* cngetc() */
49 
50 #include <vm/vm.h>
51 #include <vm/vm_param.h>
52 #include <vm/pmap.h>
53 #include <vm/vm_kern.h>
54 #include <vm/vm_extern.h>
55 #ifdef BETTER_CLOCK
56 #include <sys/lock.h>
57 #include <vm/vm_map.h>
58 #include <sys/user.h>
59 #ifdef GPROF
60 #include <sys/gmon.h>
61 #endif
62 #endif
63 
64 #include <machine/smp.h>
65 #include <machine/apic.h>
66 #include <machine/atomic.h>
67 #include <machine/cpufunc.h>
68 #include <machine/mpapic.h>
69 #include <machine/psl.h>
70 #include <machine/segments.h>
71 #include <machine/smptests.h>	/** TEST_DEFAULT_CONFIG, TEST_TEST1 */
72 #include <machine/tss.h>
73 #include <machine/specialreg.h>
74 #include <machine/globaldata.h>
75 
76 #if defined(APIC_IO)
77 #include <machine/md_var.h>		/* setidt() */
78 #include <i386/isa/icu.h>		/* IPIs */
79 #include <i386/isa/intr_machdep.h>	/* IPIs */
80 #endif	/* APIC_IO */
81 
82 #if defined(TEST_DEFAULT_CONFIG)
83 #define MPFPS_MPFB1	TEST_DEFAULT_CONFIG
84 #else
85 #define MPFPS_MPFB1	mpfps->mpfb1
86 #endif  /* TEST_DEFAULT_CONFIG */
87 
88 #define WARMBOOT_TARGET		0
89 #define WARMBOOT_OFF		(KERNBASE + 0x0467)
90 #define WARMBOOT_SEG		(KERNBASE + 0x0469)
91 
92 #ifdef PC98
93 #define BIOS_BASE		(0xe8000)
94 #define BIOS_SIZE		(0x18000)
95 #else
96 #define BIOS_BASE		(0xf0000)
97 #define BIOS_SIZE		(0x10000)
98 #endif
99 #define BIOS_COUNT		(BIOS_SIZE/4)
100 
101 #define CMOS_REG		(0x70)
102 #define CMOS_DATA		(0x71)
103 #define BIOS_RESET		(0x0f)
104 #define BIOS_WARM		(0x0a)
105 
106 #define PROCENTRY_FLAG_EN	0x01
107 #define PROCENTRY_FLAG_BP	0x02
108 #define IOAPICENTRY_FLAG_EN	0x01
109 
110 
111 /* MP Floating Pointer Structure */
112 typedef struct MPFPS {
113 	char    signature[4];
114 	void   *pap;
115 	u_char  length;
116 	u_char  spec_rev;
117 	u_char  checksum;
118 	u_char  mpfb1;
119 	u_char  mpfb2;
120 	u_char  mpfb3;
121 	u_char  mpfb4;
122 	u_char  mpfb5;
123 }      *mpfps_t;
124 
125 /* MP Configuration Table Header */
126 typedef struct MPCTH {
127 	char    signature[4];
128 	u_short base_table_length;
129 	u_char  spec_rev;
130 	u_char  checksum;
131 	u_char  oem_id[8];
132 	u_char  product_id[12];
133 	void   *oem_table_pointer;
134 	u_short oem_table_size;
135 	u_short entry_count;
136 	void   *apic_address;
137 	u_short extended_table_length;
138 	u_char  extended_table_checksum;
139 	u_char  reserved;
140 }      *mpcth_t;
141 
142 
143 typedef struct PROCENTRY {
144 	u_char  type;
145 	u_char  apic_id;
146 	u_char  apic_version;
147 	u_char  cpu_flags;
148 	u_long  cpu_signature;
149 	u_long  feature_flags;
150 	u_long  reserved1;
151 	u_long  reserved2;
152 }      *proc_entry_ptr;
153 
154 typedef struct BUSENTRY {
155 	u_char  type;
156 	u_char  bus_id;
157 	char    bus_type[6];
158 }      *bus_entry_ptr;
159 
160 typedef struct IOAPICENTRY {
161 	u_char  type;
162 	u_char  apic_id;
163 	u_char  apic_version;
164 	u_char  apic_flags;
165 	void   *apic_address;
166 }      *io_apic_entry_ptr;
167 
168 typedef struct INTENTRY {
169 	u_char  type;
170 	u_char  int_type;
171 	u_short int_flags;
172 	u_char  src_bus_id;
173 	u_char  src_bus_irq;
174 	u_char  dst_apic_id;
175 	u_char  dst_apic_int;
176 }      *int_entry_ptr;
177 
178 /* descriptions of MP basetable entries */
179 typedef struct BASETABLE_ENTRY {
180 	u_char  type;
181 	u_char  length;
182 	char    name[16];
183 }       basetable_entry;
184 
185 /*
186  * this code MUST be enabled here and in mpboot.s.
187  * it follows the very early stages of AP boot by placing values in CMOS ram.
188  * it NORMALLY will never be needed and thus the primitive method for enabling.
189  *
190 #define CHECK_POINTS
191  */
192 
193 #if defined(CHECK_POINTS) && !defined(PC98)
194 #define CHECK_READ(A)	 (outb(CMOS_REG, (A)), inb(CMOS_DATA))
195 #define CHECK_WRITE(A,D) (outb(CMOS_REG, (A)), outb(CMOS_DATA, (D)))
196 
197 #define CHECK_INIT(D);				\
198 	CHECK_WRITE(0x34, (D));			\
199 	CHECK_WRITE(0x35, (D));			\
200 	CHECK_WRITE(0x36, (D));			\
201 	CHECK_WRITE(0x37, (D));			\
202 	CHECK_WRITE(0x38, (D));			\
203 	CHECK_WRITE(0x39, (D));
204 
205 #define CHECK_PRINT(S);				\
206 	printf("%s: %d, %d, %d, %d, %d, %d\n",	\
207 	   (S),					\
208 	   CHECK_READ(0x34),			\
209 	   CHECK_READ(0x35),			\
210 	   CHECK_READ(0x36),			\
211 	   CHECK_READ(0x37),			\
212 	   CHECK_READ(0x38),			\
213 	   CHECK_READ(0x39));
214 
215 #else				/* CHECK_POINTS */
216 
217 #define CHECK_INIT(D)
218 #define CHECK_PRINT(S)
219 
220 #endif				/* CHECK_POINTS */
221 
222 /*
223  * Values to send to the POST hardware.
224  */
225 #define MP_BOOTADDRESS_POST	0x10
226 #define MP_PROBE_POST		0x11
227 #define MPTABLE_PASS1_POST	0x12
228 
229 #define MP_START_POST		0x13
230 #define MP_ENABLE_POST		0x14
231 #define MPTABLE_PASS2_POST	0x15
232 
233 #define START_ALL_APS_POST	0x16
234 #define INSTALL_AP_TRAMP_POST	0x17
235 #define START_AP_POST		0x18
236 
237 #define MP_ANNOUNCE_POST	0x19
238 
239 
240 /** XXX FIXME: where does this really belong, isa.h/isa.c perhaps? */
241 int	current_postcode;
242 
243 /** XXX FIXME: what system files declare these??? */
244 extern struct region_descriptor r_gdt, r_idt;
245 
246 int	bsp_apic_ready = 0;	/* flags useability of BSP apic */
247 int	mp_ncpus;		/* # of CPUs, including BSP */
248 int	mp_naps;		/* # of Applications processors */
249 int	mp_nbusses;		/* # of busses */
250 int	mp_napics;		/* # of IO APICs */
251 int	boot_cpu_id;		/* designated BSP */
252 vm_offset_t cpu_apic_address;
253 vm_offset_t io_apic_address[NAPICID];	/* NAPICID is more than enough */
254 extern	int nkpt;
255 
256 u_int32_t cpu_apic_versions[NCPU];
257 u_int32_t io_apic_versions[NAPIC];
258 
259 #ifdef APIC_INTR_DIAGNOSTIC
260 int apic_itrace_enter[32];
261 int apic_itrace_tryisrlock[32];
262 int apic_itrace_gotisrlock[32];
263 int apic_itrace_active[32];
264 int apic_itrace_masked[32];
265 int apic_itrace_noisrlock[32];
266 int apic_itrace_masked2[32];
267 int apic_itrace_unmask[32];
268 int apic_itrace_noforward[32];
269 int apic_itrace_leave[32];
270 int apic_itrace_enter2[32];
271 int apic_itrace_doreti[32];
272 int apic_itrace_splz[32];
273 int apic_itrace_eoi[32];
274 #ifdef APIC_INTR_DIAGNOSTIC_IRQ
275 unsigned short apic_itrace_debugbuffer[32768];
276 int apic_itrace_debugbuffer_idx;
277 struct simplelock apic_itrace_debuglock;
278 #endif
279 #endif
280 
281 #ifdef APIC_INTR_REORDER
282 struct {
283 	volatile int *location;
284 	int bit;
285 } apic_isrbit_location[32];
286 #endif
287 
288 struct apic_intmapinfo	int_to_apicintpin[APIC_INTMAPSIZE];
289 
290 /*
291  * APIC ID logical/physical mapping structures.
292  * We oversize these to simplify boot-time config.
293  */
294 int     cpu_num_to_apic_id[NAPICID];
295 int     io_num_to_apic_id[NAPICID];
296 int     apic_id_to_logical[NAPICID];
297 
298 
299 /* Bitmap of all available CPUs */
300 u_int	all_cpus;
301 
302 /* AP uses this during bootstrap.  Do not staticize.  */
303 char *bootSTK;
304 static int bootAP;
305 
306 /* Hotwire a 0->4MB V==P mapping */
307 extern pt_entry_t *KPTphys;
308 
309 /* SMP page table page */
310 extern pt_entry_t *SMPpt;
311 
312 struct pcb stoppcbs[NCPU];
313 
314 int smp_started;		/* has the system started? */
315 
316 /*
317  * Local data and functions.
318  */
319 
320 static int	mp_capable;
321 static u_int	boot_address;
322 static u_int	base_memory;
323 
324 static int	picmode;		/* 0: virtual wire mode, 1: PIC mode */
325 static mpfps_t	mpfps;
326 static int	search_for_sig(u_int32_t target, int count);
327 static void	mp_enable(u_int boot_addr);
328 
329 static int	mptable_pass1(void);
330 static int	mptable_pass2(void);
331 static void	default_mp_table(int type);
332 static void	fix_mp_table(void);
333 static void	setup_apic_irq_mapping(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 static int	apic_int_is_bus_type(int intr, int bus_type);
339 
340 /*
341  * Calculate usable address in base memory for AP trampoline code.
342  */
343 u_int
344 mp_bootaddress(u_int basemem)
345 {
346 	POSTCODE(MP_BOOTADDRESS_POST);
347 
348 	base_memory = basemem * 1024;	/* convert to bytes */
349 
350 	boot_address = base_memory & ~0xfff;	/* round down to 4k boundary */
351 	if ((base_memory - boot_address) < bootMP_size)
352 		boot_address -= 4096;	/* not enough, lower by 4k */
353 
354 	return boot_address;
355 }
356 
357 
358 /*
359  * Look for an Intel MP spec table (ie, SMP capable hardware).
360  */
361 int
362 mp_probe(void)
363 {
364 	int     x;
365 	u_long  segment;
366 	u_int32_t target;
367 
368 	POSTCODE(MP_PROBE_POST);
369 
370 	/* see if EBDA exists */
371 	if ((segment = (u_long) * (u_short *) (KERNBASE + 0x40e)) != 0) {
372 		/* search first 1K of EBDA */
373 		target = (u_int32_t) (segment << 4);
374 		if ((x = search_for_sig(target, 1024 / 4)) >= 0)
375 			goto found;
376 	} else {
377 		/* last 1K of base memory, effective 'top of base' passed in */
378 		target = (u_int32_t) (base_memory - 0x400);
379 		if ((x = search_for_sig(target, 1024 / 4)) >= 0)
380 			goto found;
381 	}
382 
383 	/* search the BIOS */
384 	target = (u_int32_t) BIOS_BASE;
385 	if ((x = search_for_sig(target, BIOS_COUNT)) >= 0)
386 		goto found;
387 
388 	/* nothing found */
389 	mpfps = (mpfps_t)0;
390 	mp_capable = 0;
391 	return 0;
392 
393 found:
394 	/* calculate needed resources */
395 	mpfps = (mpfps_t)x;
396 	if (mptable_pass1())
397 		panic("you must reconfigure your kernel");
398 
399 	/* flag fact that we are running multiple processors */
400 	mp_capable = 1;
401 	return 1;
402 }
403 
404 
405 /*
406  * Startup the SMP processors.
407  */
408 void
409 mp_start(void)
410 {
411 	POSTCODE(MP_START_POST);
412 
413 	/* look for MP capable motherboard */
414 	if (mp_capable)
415 		mp_enable(boot_address);
416 	else
417 		panic("MP hardware not found!");
418 }
419 
420 
421 /*
422  * Print various information about the SMP system hardware and setup.
423  */
424 void
425 mp_announce(void)
426 {
427 	int     x;
428 
429 	POSTCODE(MP_ANNOUNCE_POST);
430 
431 	printf("FreeBSD/SMP: Multiprocessor motherboard\n");
432 	printf(" cpu0 (BSP): apic id: %2d", CPU_TO_ID(0));
433 	printf(", version: 0x%08x", cpu_apic_versions[0]);
434 	printf(", at 0x%08x\n", cpu_apic_address);
435 	for (x = 1; x <= mp_naps; ++x) {
436 		printf(" cpu%d (AP):  apic id: %2d", x, CPU_TO_ID(x));
437 		printf(", version: 0x%08x", cpu_apic_versions[x]);
438 		printf(", at 0x%08x\n", cpu_apic_address);
439 	}
440 
441 #if defined(APIC_IO)
442 	for (x = 0; x < mp_napics; ++x) {
443 		printf(" io%d (APIC): apic id: %2d", x, IO_TO_ID(x));
444 		printf(", version: 0x%08x", io_apic_versions[x]);
445 		printf(", at 0x%08x\n", io_apic_address[x]);
446 	}
447 #else
448 	printf(" Warning: APIC I/O disabled\n");
449 #endif	/* APIC_IO */
450 }
451 
452 /*
453  * AP cpu's call this to sync up protected mode.
454  */
455 void
456 init_secondary(void)
457 {
458 	int	gsel_tss;
459 	int	x, myid = bootAP;
460 
461 	gdt_segs[GPRIV_SEL].ssd_base = (int) &SMP_prvspace[myid];
462 	gdt_segs[GPROC0_SEL].ssd_base =
463 		(int) &SMP_prvspace[myid].globaldata.gd_common_tss;
464 	SMP_prvspace[myid].globaldata.gd_prvspace = &SMP_prvspace[myid];
465 
466 	for (x = 0; x < NGDT; x++) {
467 		ssdtosd(&gdt_segs[x], &gdt[myid * NGDT + x].sd);
468 	}
469 
470 	r_gdt.rd_limit = NGDT * sizeof(gdt[0]) - 1;
471 	r_gdt.rd_base = (int) &gdt[myid * NGDT];
472 	lgdt(&r_gdt);			/* does magic intra-segment return */
473 
474 	lidt(&r_idt);
475 
476 	lldt(_default_ldt);
477 #ifdef USER_LDT
478 	currentldt = _default_ldt;
479 #endif
480 
481 	gsel_tss = GSEL(GPROC0_SEL, SEL_KPL);
482 	gdt[myid * NGDT + GPROC0_SEL].sd.sd_type = SDT_SYS386TSS;
483 	common_tss.tss_esp0 = 0;	/* not used until after switch */
484 	common_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
485 	common_tss.tss_ioopt = (sizeof common_tss) << 16;
486 	tss_gdt = &gdt[myid * NGDT + GPROC0_SEL].sd;
487 	common_tssd = *tss_gdt;
488 	ltr(gsel_tss);
489 
490 	load_cr0(0x8005003b);		/* XXX! */
491 
492 	pmap_set_opt();
493 }
494 
495 
496 #if defined(APIC_IO)
497 /*
498  * Final configuration of the BSP's local APIC:
499  *  - disable 'pic mode'.
500  *  - disable 'virtual wire mode'.
501  *  - enable NMI.
502  */
503 void
504 bsp_apic_configure(void)
505 {
506 	u_char		byte;
507 	u_int32_t	temp;
508 
509 	/* leave 'pic mode' if necessary */
510 	if (picmode) {
511 		outb(0x22, 0x70);	/* select IMCR */
512 		byte = inb(0x23);	/* current contents */
513 		byte |= 0x01;		/* mask external INTR */
514 		outb(0x23, byte);	/* disconnect 8259s/NMI */
515 	}
516 
517 	/* mask lint0 (the 8259 'virtual wire' connection) */
518 	temp = lapic.lvt_lint0;
519 	temp |= APIC_LVT_M;		/* set the mask */
520 	lapic.lvt_lint0 = temp;
521 
522         /* setup lint1 to handle NMI */
523         temp = lapic.lvt_lint1;
524         temp &= ~APIC_LVT_M;		/* clear the mask */
525         lapic.lvt_lint1 = temp;
526 
527 	if (bootverbose)
528 		apic_dump("bsp_apic_configure()");
529 }
530 #endif  /* APIC_IO */
531 
532 
533 /*******************************************************************
534  * local functions and data
535  */
536 
537 /*
538  * start the SMP system
539  */
540 static void
541 mp_enable(u_int boot_addr)
542 {
543 	int     x;
544 #if defined(APIC_IO)
545 	int     apic;
546 	u_int   ux;
547 #endif	/* APIC_IO */
548 
549 	POSTCODE(MP_ENABLE_POST);
550 
551 	/* turn on 4MB of V == P addressing so we can get to MP table */
552 	*(int *)PTD = PG_V | PG_RW | ((uintptr_t)(void *)KPTphys & PG_FRAME);
553 	invltlb();
554 
555 	/* examine the MP table for needed info, uses physical addresses */
556 	x = mptable_pass2();
557 
558 	*(int *)PTD = 0;
559 	invltlb();
560 
561 	/* can't process default configs till the CPU APIC is pmapped */
562 	if (x)
563 		default_mp_table(x);
564 
565 	/* post scan cleanup */
566 	fix_mp_table();
567 	setup_apic_irq_mapping();
568 
569 #if defined(APIC_IO)
570 
571 	/* fill the LOGICAL io_apic_versions table */
572 	for (apic = 0; apic < mp_napics; ++apic) {
573 		ux = io_apic_read(apic, IOAPIC_VER);
574 		io_apic_versions[apic] = ux;
575 	}
576 
577 	/* program each IO APIC in the system */
578 	for (apic = 0; apic < mp_napics; ++apic)
579 		if (io_apic_setup(apic) < 0)
580 			panic("IO APIC setup failure");
581 
582 	/* install a 'Spurious INTerrupt' vector */
583 	setidt(XSPURIOUSINT_OFFSET, Xspuriousint,
584 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
585 
586 	/* install an inter-CPU IPI for TLB invalidation */
587 	setidt(XINVLTLB_OFFSET, Xinvltlb,
588 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
589 
590 #ifdef BETTER_CLOCK
591 	/* install an inter-CPU IPI for reading processor state */
592 	setidt(XCPUCHECKSTATE_OFFSET, Xcpucheckstate,
593 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
594 #endif
595 
596 	/* install an inter-CPU IPI for all-CPU rendezvous */
597 	setidt(XRENDEZVOUS_OFFSET, Xrendezvous,
598 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
599 
600 	/* install an inter-CPU IPI for forcing an additional software trap */
601 	setidt(XCPUAST_OFFSET, Xcpuast,
602 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
603 
604 	/* install an inter-CPU IPI for interrupt forwarding */
605 	setidt(XFORWARD_IRQ_OFFSET, Xforward_irq,
606 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
607 
608 	/* install an inter-CPU IPI for CPU stop/restart */
609 	setidt(XCPUSTOP_OFFSET, Xcpustop,
610 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
611 
612 #if defined(TEST_TEST1)
613 	/* install a "fake hardware INTerrupt" vector */
614 	setidt(XTEST1_OFFSET, Xtest1,
615 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
616 #endif  /** TEST_TEST1 */
617 
618 #endif	/* APIC_IO */
619 
620 	/* initialize all SMP locks */
621 	init_locks();
622 
623 	/* start each Application Processor */
624 	start_all_aps(boot_addr);
625 
626 	/*
627 	 * The init process might be started on a different CPU now,
628 	 * and the boot CPU might not call prepare_usermode to get
629 	 * cr0 correctly configured. Thus we initialize cr0 here.
630 	 */
631 	load_cr0(rcr0() | CR0_WP | CR0_AM);
632 }
633 
634 
635 /*
636  * look for the MP spec signature
637  */
638 
639 /* string defined by the Intel MP Spec as identifying the MP table */
640 #define MP_SIG		0x5f504d5f	/* _MP_ */
641 #define NEXT(X)		((X) += 4)
642 static int
643 search_for_sig(u_int32_t target, int count)
644 {
645 	int     x;
646 	u_int32_t *addr = (u_int32_t *) (KERNBASE + target);
647 
648 	for (x = 0; x < count; NEXT(x))
649 		if (addr[x] == MP_SIG)
650 			/* make array index a byte index */
651 			return (target + (x * sizeof(u_int32_t)));
652 
653 	return -1;
654 }
655 
656 
657 static basetable_entry basetable_entry_types[] =
658 {
659 	{0, 20, "Processor"},
660 	{1, 8, "Bus"},
661 	{2, 8, "I/O APIC"},
662 	{3, 8, "I/O INT"},
663 	{4, 8, "Local INT"}
664 };
665 
666 typedef struct BUSDATA {
667 	u_char  bus_id;
668 	enum busTypes bus_type;
669 }       bus_datum;
670 
671 typedef struct INTDATA {
672 	u_char  int_type;
673 	u_short int_flags;
674 	u_char  src_bus_id;
675 	u_char  src_bus_irq;
676 	u_char  dst_apic_id;
677 	u_char  dst_apic_int;
678 	u_char	int_vector;
679 }       io_int, local_int;
680 
681 typedef struct BUSTYPENAME {
682 	u_char  type;
683 	char    name[7];
684 }       bus_type_name;
685 
686 static bus_type_name bus_type_table[] =
687 {
688 	{CBUS, "CBUS"},
689 	{CBUSII, "CBUSII"},
690 	{EISA, "EISA"},
691 	{MCA, "MCA"},
692 	{UNKNOWN_BUSTYPE, "---"},
693 	{ISA, "ISA"},
694 	{MCA, "MCA"},
695 	{UNKNOWN_BUSTYPE, "---"},
696 	{UNKNOWN_BUSTYPE, "---"},
697 	{UNKNOWN_BUSTYPE, "---"},
698 	{UNKNOWN_BUSTYPE, "---"},
699 	{UNKNOWN_BUSTYPE, "---"},
700 	{PCI, "PCI"},
701 	{UNKNOWN_BUSTYPE, "---"},
702 	{UNKNOWN_BUSTYPE, "---"},
703 	{UNKNOWN_BUSTYPE, "---"},
704 	{UNKNOWN_BUSTYPE, "---"},
705 	{XPRESS, "XPRESS"},
706 	{UNKNOWN_BUSTYPE, "---"}
707 };
708 /* from MP spec v1.4, table 5-1 */
709 static int default_data[7][5] =
710 {
711 /*   nbus, id0, type0, id1, type1 */
712 	{1, 0, ISA, 255, 255},
713 	{1, 0, EISA, 255, 255},
714 	{1, 0, EISA, 255, 255},
715 	{1, 0, MCA, 255, 255},
716 	{2, 0, ISA, 1, PCI},
717 	{2, 0, EISA, 1, PCI},
718 	{2, 0, MCA, 1, PCI}
719 };
720 
721 
722 /* the bus data */
723 static bus_datum bus_data[NBUS];
724 
725 /* the IO INT data, one entry per possible APIC INTerrupt */
726 static io_int  io_apic_ints[NINTR];
727 
728 static int nintrs;
729 
730 static int processor_entry	__P((proc_entry_ptr entry, int cpu));
731 static int bus_entry		__P((bus_entry_ptr entry, int bus));
732 static int io_apic_entry	__P((io_apic_entry_ptr entry, int apic));
733 static int int_entry		__P((int_entry_ptr entry, int intr));
734 static int lookup_bus_type	__P((char *name));
735 
736 
737 /*
738  * 1st pass on motherboard's Intel MP specification table.
739  *
740  * initializes:
741  *	mp_ncpus = 1
742  *
743  * determines:
744  *	cpu_apic_address (common to all CPUs)
745  *	io_apic_address[N]
746  *	mp_naps
747  *	mp_nbusses
748  *	mp_napics
749  *	nintrs
750  */
751 static int
752 mptable_pass1(void)
753 {
754 	int	x;
755 	mpcth_t	cth;
756 	int	totalSize;
757 	void*	position;
758 	int	count;
759 	int	type;
760 	int	mustpanic;
761 
762 	POSTCODE(MPTABLE_PASS1_POST);
763 
764 	mustpanic = 0;
765 
766 	/* clear various tables */
767 	for (x = 0; x < NAPICID; ++x) {
768 		io_apic_address[x] = ~0;	/* IO APIC address table */
769 	}
770 
771 	/* init everything to empty */
772 	mp_naps = 0;
773 	mp_nbusses = 0;
774 	mp_napics = 0;
775 	nintrs = 0;
776 
777 	/* check for use of 'default' configuration */
778 	if (MPFPS_MPFB1 != 0) {
779 		/* use default addresses */
780 		cpu_apic_address = DEFAULT_APIC_BASE;
781 		io_apic_address[0] = DEFAULT_IO_APIC_BASE;
782 
783 		/* fill in with defaults */
784 		mp_naps = 2;		/* includes BSP */
785 		mp_nbusses = default_data[MPFPS_MPFB1 - 1][0];
786 #if defined(APIC_IO)
787 		mp_napics = 1;
788 		nintrs = 16;
789 #endif	/* APIC_IO */
790 	}
791 	else {
792 		if ((cth = mpfps->pap) == 0)
793 			panic("MP Configuration Table Header MISSING!");
794 
795 		cpu_apic_address = (vm_offset_t) cth->apic_address;
796 
797 		/* walk the table, recording info of interest */
798 		totalSize = cth->base_table_length - sizeof(struct MPCTH);
799 		position = (u_char *) cth + sizeof(struct MPCTH);
800 		count = cth->entry_count;
801 
802 		while (count--) {
803 			switch (type = *(u_char *) position) {
804 			case 0: /* processor_entry */
805 				if (((proc_entry_ptr)position)->cpu_flags
806 					& PROCENTRY_FLAG_EN)
807 					++mp_naps;
808 				break;
809 			case 1: /* bus_entry */
810 				++mp_nbusses;
811 				break;
812 			case 2: /* io_apic_entry */
813 				if (((io_apic_entry_ptr)position)->apic_flags
814 					& IOAPICENTRY_FLAG_EN)
815 					io_apic_address[mp_napics++] =
816 					    (vm_offset_t)((io_apic_entry_ptr)
817 						position)->apic_address;
818 				break;
819 			case 3: /* int_entry */
820 				++nintrs;
821 				break;
822 			case 4:	/* int_entry */
823 				break;
824 			default:
825 				panic("mpfps Base Table HOSED!");
826 				/* NOTREACHED */
827 			}
828 
829 			totalSize -= basetable_entry_types[type].length;
830 			(u_char*)position += basetable_entry_types[type].length;
831 		}
832 	}
833 
834 	/* qualify the numbers */
835 	if (mp_naps > NCPU) {
836 		printf("Warning: only using %d of %d available CPUs!\n",
837 			NCPU, mp_naps);
838 		mp_naps = NCPU;
839 	}
840 	if (mp_nbusses > NBUS) {
841 		printf("found %d busses, increase NBUS\n", mp_nbusses);
842 		mustpanic = 1;
843 	}
844 	if (mp_napics > NAPIC) {
845 		printf("found %d apics, increase NAPIC\n", mp_napics);
846 		mustpanic = 1;
847 	}
848 	if (nintrs > NINTR) {
849 		printf("found %d intrs, increase NINTR\n", nintrs);
850 		mustpanic = 1;
851 	}
852 
853 	/*
854 	 * Count the BSP.
855 	 * This is also used as a counter while starting the APs.
856 	 */
857 	mp_ncpus = 1;
858 
859 	--mp_naps;	/* subtract the BSP */
860 
861 	return mustpanic;
862 }
863 
864 
865 /*
866  * 2nd pass on motherboard's Intel MP specification table.
867  *
868  * sets:
869  *	boot_cpu_id
870  *	ID_TO_IO(N), phy APIC ID to log CPU/IO table
871  *	CPU_TO_ID(N), logical CPU to APIC ID table
872  *	IO_TO_ID(N), logical IO to APIC ID table
873  *	bus_data[N]
874  *	io_apic_ints[N]
875  */
876 static int
877 mptable_pass2(void)
878 {
879 	int     x;
880 	mpcth_t cth;
881 	int     totalSize;
882 	void*   position;
883 	int     count;
884 	int     type;
885 	int     apic, bus, cpu, intr;
886 
887 	POSTCODE(MPTABLE_PASS2_POST);
888 
889 	/* clear various tables */
890 	for (x = 0; x < NAPICID; ++x) {
891 		ID_TO_IO(x) = -1;	/* phy APIC ID to log CPU/IO table */
892 		CPU_TO_ID(x) = -1;	/* logical CPU to APIC ID table */
893 		IO_TO_ID(x) = -1;	/* logical IO to APIC ID table */
894 	}
895 
896 	/* clear bus data table */
897 	for (x = 0; x < NBUS; ++x)
898 		bus_data[x].bus_id = 0xff;
899 
900 	/* clear IO APIC INT table */
901 	for (x = 0; x < NINTR; ++x) {
902 		io_apic_ints[x].int_type = 0xff;
903 		io_apic_ints[x].int_vector = 0xff;
904 	}
905 
906 	/* setup the cpu/apic mapping arrays */
907 	boot_cpu_id = -1;
908 
909 	/* record whether PIC or virtual-wire mode */
910 	picmode = (mpfps->mpfb2 & 0x80) ? 1 : 0;
911 
912 	/* check for use of 'default' configuration */
913 	if (MPFPS_MPFB1 != 0)
914 		return MPFPS_MPFB1;	/* return default configuration type */
915 
916 	if ((cth = mpfps->pap) == 0)
917 		panic("MP Configuration Table Header MISSING!");
918 
919 	/* walk the table, recording info of interest */
920 	totalSize = cth->base_table_length - sizeof(struct MPCTH);
921 	position = (u_char *) cth + sizeof(struct MPCTH);
922 	count = cth->entry_count;
923 	apic = bus = intr = 0;
924 	cpu = 1;				/* pre-count the BSP */
925 
926 	while (count--) {
927 		switch (type = *(u_char *) position) {
928 		case 0:
929 			if (processor_entry(position, cpu))
930 				++cpu;
931 			break;
932 		case 1:
933 			if (bus_entry(position, bus))
934 				++bus;
935 			break;
936 		case 2:
937 			if (io_apic_entry(position, apic))
938 				++apic;
939 			break;
940 		case 3:
941 			if (int_entry(position, intr))
942 				++intr;
943 			break;
944 		case 4:
945 			/* int_entry(position); */
946 			break;
947 		default:
948 			panic("mpfps Base Table HOSED!");
949 			/* NOTREACHED */
950 		}
951 
952 		totalSize -= basetable_entry_types[type].length;
953 		(u_char *) position += basetable_entry_types[type].length;
954 	}
955 
956 	if (boot_cpu_id == -1)
957 		panic("NO BSP found!");
958 
959 	/* report fact that its NOT a default configuration */
960 	return 0;
961 }
962 
963 
964 void
965 assign_apic_irq(int apic, int intpin, int irq)
966 {
967 	int x;
968 
969 	if (int_to_apicintpin[irq].ioapic != -1)
970 		panic("assign_apic_irq: inconsistent table");
971 
972 	int_to_apicintpin[irq].ioapic = apic;
973 	int_to_apicintpin[irq].int_pin = intpin;
974 	int_to_apicintpin[irq].apic_address = ioapic[apic];
975 	int_to_apicintpin[irq].redirindex = IOAPIC_REDTBL + 2 * intpin;
976 
977 	for (x = 0; x < nintrs; x++) {
978 		if ((io_apic_ints[x].int_type == 0 ||
979 		     io_apic_ints[x].int_type == 3) &&
980 		    io_apic_ints[x].int_vector == 0xff &&
981 		    io_apic_ints[x].dst_apic_id == IO_TO_ID(apic) &&
982 		    io_apic_ints[x].dst_apic_int == intpin)
983 			io_apic_ints[x].int_vector = irq;
984 	}
985 }
986 
987 void
988 revoke_apic_irq(int irq)
989 {
990 	int x;
991 	int oldapic;
992 	int oldintpin;
993 
994 	if (int_to_apicintpin[irq].ioapic == -1)
995 		panic("assign_apic_irq: inconsistent table");
996 
997 	oldapic = int_to_apicintpin[irq].ioapic;
998 	oldintpin = int_to_apicintpin[irq].int_pin;
999 
1000 	int_to_apicintpin[irq].ioapic = -1;
1001 	int_to_apicintpin[irq].int_pin = 0;
1002 	int_to_apicintpin[irq].apic_address = NULL;
1003 	int_to_apicintpin[irq].redirindex = 0;
1004 
1005 	for (x = 0; x < nintrs; x++) {
1006 		if ((io_apic_ints[x].int_type == 0 ||
1007 		     io_apic_ints[x].int_type == 3) &&
1008 		    io_apic_ints[x].int_vector == 0xff &&
1009 		    io_apic_ints[x].dst_apic_id == IO_TO_ID(oldapic) &&
1010 		    io_apic_ints[x].dst_apic_int == oldintpin)
1011 			io_apic_ints[x].int_vector = 0xff;
1012 	}
1013 }
1014 
1015 /*
1016  * parse an Intel MP specification table
1017  */
1018 static void
1019 fix_mp_table(void)
1020 {
1021 	int	x;
1022 	int	id;
1023 	int	bus_0 = 0;	/* Stop GCC warning */
1024 	int	bus_pci = 0;	/* Stop GCC warning */
1025 	int	num_pci_bus;
1026 
1027 	/*
1028 	 * Fix mis-numbering of the PCI bus and its INT entries if the BIOS
1029 	 * did it wrong.  The MP spec says that when more than 1 PCI bus
1030 	 * exists the BIOS must begin with bus entries for the PCI bus and use
1031 	 * actual PCI bus numbering.  This implies that when only 1 PCI bus
1032 	 * exists the BIOS can choose to ignore this ordering, and indeed many
1033 	 * MP motherboards do ignore it.  This causes a problem when the PCI
1034 	 * sub-system makes requests of the MP sub-system based on PCI bus
1035 	 * numbers.	So here we look for the situation and renumber the
1036 	 * busses and associated INTs in an effort to "make it right".
1037 	 */
1038 
1039 	/* find bus 0, PCI bus, count the number of PCI busses */
1040 	for (num_pci_bus = 0, x = 0; x < mp_nbusses; ++x) {
1041 		if (bus_data[x].bus_id == 0) {
1042 			bus_0 = x;
1043 		}
1044 		if (bus_data[x].bus_type == PCI) {
1045 			++num_pci_bus;
1046 			bus_pci = x;
1047 		}
1048 	}
1049 	/*
1050 	 * bus_0 == slot of bus with ID of 0
1051 	 * bus_pci == slot of last PCI bus encountered
1052 	 */
1053 
1054 	/* check the 1 PCI bus case for sanity */
1055 	if (num_pci_bus == 1) {
1056 
1057 		/* if it is number 0 all is well */
1058 		if (bus_data[bus_pci].bus_id == 0)
1059 			return;
1060 
1061 		/* mis-numbered, swap with whichever bus uses slot 0 */
1062 
1063 		/* swap the bus entry types */
1064 		bus_data[bus_pci].bus_type = bus_data[bus_0].bus_type;
1065 		bus_data[bus_0].bus_type = PCI;
1066 
1067 		/* swap each relavant INTerrupt entry */
1068 		id = bus_data[bus_pci].bus_id;
1069 		for (x = 0; x < nintrs; ++x) {
1070 			if (io_apic_ints[x].src_bus_id == id) {
1071 				io_apic_ints[x].src_bus_id = 0;
1072 			}
1073 			else if (io_apic_ints[x].src_bus_id == 0) {
1074 				io_apic_ints[x].src_bus_id = id;
1075 			}
1076 		}
1077 	}
1078 }
1079 
1080 
1081 /* Assign low level interrupt handlers */
1082 static void
1083 setup_apic_irq_mapping(void)
1084 {
1085 	int	x;
1086 	int	int_vector;
1087 
1088 	/* Clear array */
1089 	for (x = 0; x < APIC_INTMAPSIZE; x++) {
1090 		int_to_apicintpin[x].ioapic = -1;
1091 		int_to_apicintpin[x].int_pin = 0;
1092 		int_to_apicintpin[x].apic_address = NULL;
1093 		int_to_apicintpin[x].redirindex = 0;
1094 	}
1095 
1096 	/* First assign ISA/EISA interrupts */
1097 	for (x = 0; x < nintrs; x++) {
1098 		int_vector = io_apic_ints[x].src_bus_irq;
1099 		if (int_vector < APIC_INTMAPSIZE &&
1100 		    io_apic_ints[x].int_vector == 0xff &&
1101 		    int_to_apicintpin[int_vector].ioapic == -1 &&
1102 		    (apic_int_is_bus_type(x, ISA) ||
1103 		     apic_int_is_bus_type(x, EISA)) &&
1104 		    io_apic_ints[x].int_type == 0) {
1105 			assign_apic_irq(ID_TO_IO(io_apic_ints[x].dst_apic_id),
1106 					io_apic_ints[x].dst_apic_int,
1107 					int_vector);
1108 		}
1109 	}
1110 
1111 	/* Assign interrupts on first 24 intpins on IOAPIC #0 */
1112 	for (x = 0; x < nintrs; x++) {
1113 		int_vector = io_apic_ints[x].dst_apic_int;
1114 		if (int_vector < APIC_INTMAPSIZE &&
1115 		    io_apic_ints[x].dst_apic_id == IO_TO_ID(0) &&
1116 		    io_apic_ints[x].int_vector == 0xff &&
1117 		    int_to_apicintpin[int_vector].ioapic == -1 &&
1118 		    (io_apic_ints[x].int_type == 0 ||
1119 		     io_apic_ints[x].int_type == 3)) {
1120 			assign_apic_irq(0,
1121 					io_apic_ints[x].dst_apic_int,
1122 					int_vector);
1123 		}
1124 	}
1125 	/*
1126 	 * Assign interrupts for remaining intpins.
1127 	 * Skip IOAPIC #0 intpin 0 if the type is ExtInt, since this indicates
1128 	 * that an entry for ISA/EISA irq 0 exist, and a fallback to mixed mode
1129 	 * due to 8254 interrupts not being delivered can reuse that low level
1130 	 * interrupt handler.
1131 	 */
1132 	int_vector = 0;
1133 	while (int_vector < APIC_INTMAPSIZE &&
1134 	       int_to_apicintpin[int_vector].ioapic != -1)
1135 		int_vector++;
1136 	for (x = 0; x < nintrs && int_vector < APIC_INTMAPSIZE; x++) {
1137 		if ((io_apic_ints[x].int_type == 0 ||
1138 		     (io_apic_ints[x].int_type == 3 &&
1139 		      (io_apic_ints[x].dst_apic_id != IO_TO_ID(0) ||
1140 		       io_apic_ints[x].dst_apic_int != 0))) &&
1141 		    io_apic_ints[x].int_vector == 0xff) {
1142 			assign_apic_irq(ID_TO_IO(io_apic_ints[x].dst_apic_id),
1143 					io_apic_ints[x].dst_apic_int,
1144 					int_vector);
1145 			int_vector++;
1146 			while (int_vector < APIC_INTMAPSIZE &&
1147 			       int_to_apicintpin[int_vector].ioapic != -1)
1148 				int_vector++;
1149 		}
1150 	}
1151 }
1152 
1153 
1154 static int
1155 processor_entry(proc_entry_ptr entry, int cpu)
1156 {
1157 	/* check for usability */
1158 	if (!(entry->cpu_flags & PROCENTRY_FLAG_EN))
1159 		return 0;
1160 
1161 	/* check for BSP flag */
1162 	if (entry->cpu_flags & PROCENTRY_FLAG_BP) {
1163 		boot_cpu_id = entry->apic_id;
1164 		CPU_TO_ID(0) = entry->apic_id;
1165 		ID_TO_CPU(entry->apic_id) = 0;
1166 		return 0;	/* its already been counted */
1167 	}
1168 
1169 	/* add another AP to list, if less than max number of CPUs */
1170 	else if (cpu < NCPU) {
1171 		CPU_TO_ID(cpu) = entry->apic_id;
1172 		ID_TO_CPU(entry->apic_id) = cpu;
1173 		return 1;
1174 	}
1175 
1176 	return 0;
1177 }
1178 
1179 
1180 static int
1181 bus_entry(bus_entry_ptr entry, int bus)
1182 {
1183 	int     x;
1184 	char    c, name[8];
1185 
1186 	/* encode the name into an index */
1187 	for (x = 0; x < 6; ++x) {
1188 		if ((c = entry->bus_type[x]) == ' ')
1189 			break;
1190 		name[x] = c;
1191 	}
1192 	name[x] = '\0';
1193 
1194 	if ((x = lookup_bus_type(name)) == UNKNOWN_BUSTYPE)
1195 		panic("unknown bus type: '%s'", name);
1196 
1197 	bus_data[bus].bus_id = entry->bus_id;
1198 	bus_data[bus].bus_type = x;
1199 
1200 	return 1;
1201 }
1202 
1203 
1204 static int
1205 io_apic_entry(io_apic_entry_ptr entry, int apic)
1206 {
1207 	if (!(entry->apic_flags & IOAPICENTRY_FLAG_EN))
1208 		return 0;
1209 
1210 	IO_TO_ID(apic) = entry->apic_id;
1211 	ID_TO_IO(entry->apic_id) = apic;
1212 
1213 	return 1;
1214 }
1215 
1216 
1217 static int
1218 lookup_bus_type(char *name)
1219 {
1220 	int     x;
1221 
1222 	for (x = 0; x < MAX_BUSTYPE; ++x)
1223 		if (strcmp(bus_type_table[x].name, name) == 0)
1224 			return bus_type_table[x].type;
1225 
1226 	return UNKNOWN_BUSTYPE;
1227 }
1228 
1229 
1230 static int
1231 int_entry(int_entry_ptr entry, int intr)
1232 {
1233 	int apic;
1234 
1235 	io_apic_ints[intr].int_type = entry->int_type;
1236 	io_apic_ints[intr].int_flags = entry->int_flags;
1237 	io_apic_ints[intr].src_bus_id = entry->src_bus_id;
1238 	io_apic_ints[intr].src_bus_irq = entry->src_bus_irq;
1239 	if (entry->dst_apic_id == 255) {
1240 		/* This signal goes to all IO APICS.  Select an IO APIC
1241 		   with sufficient number of interrupt pins */
1242 		for (apic = 0; apic < mp_napics; apic++)
1243 			if (((io_apic_read(apic, IOAPIC_VER) &
1244 			      IOART_VER_MAXREDIR) >> MAXREDIRSHIFT) >=
1245 			    entry->dst_apic_int)
1246 				break;
1247 		if (apic < mp_napics)
1248 			io_apic_ints[intr].dst_apic_id = IO_TO_ID(apic);
1249 		else
1250 			io_apic_ints[intr].dst_apic_id = entry->dst_apic_id;
1251 	} else
1252 		io_apic_ints[intr].dst_apic_id = entry->dst_apic_id;
1253 	io_apic_ints[intr].dst_apic_int = entry->dst_apic_int;
1254 
1255 	return 1;
1256 }
1257 
1258 
1259 static int
1260 apic_int_is_bus_type(int intr, int bus_type)
1261 {
1262 	int     bus;
1263 
1264 	for (bus = 0; bus < mp_nbusses; ++bus)
1265 		if ((bus_data[bus].bus_id == io_apic_ints[intr].src_bus_id)
1266 		    && ((int) bus_data[bus].bus_type == bus_type))
1267 			return 1;
1268 
1269 	return 0;
1270 }
1271 
1272 
1273 /*
1274  * Given a traditional ISA INT mask, return an APIC mask.
1275  */
1276 u_int
1277 isa_apic_mask(u_int isa_mask)
1278 {
1279 	int isa_irq;
1280 	int apic_pin;
1281 
1282 #if defined(SKIP_IRQ15_REDIRECT)
1283 	if (isa_mask == (1 << 15)) {
1284 		printf("skipping ISA IRQ15 redirect\n");
1285 		return isa_mask;
1286 	}
1287 #endif  /* SKIP_IRQ15_REDIRECT */
1288 
1289 	isa_irq = ffs(isa_mask);		/* find its bit position */
1290 	if (isa_irq == 0)			/* doesn't exist */
1291 		return 0;
1292 	--isa_irq;				/* make it zero based */
1293 
1294 	apic_pin = isa_apic_irq(isa_irq);	/* look for APIC connection */
1295 	if (apic_pin == -1)
1296 		return 0;
1297 
1298 	return (1 << apic_pin);			/* convert pin# to a mask */
1299 }
1300 
1301 
1302 /*
1303  * Determine which APIC pin an ISA/EISA INT is attached to.
1304  */
1305 #define INTTYPE(I)	(io_apic_ints[(I)].int_type)
1306 #define INTPIN(I)	(io_apic_ints[(I)].dst_apic_int)
1307 #define INTIRQ(I)	(io_apic_ints[(I)].int_vector)
1308 #define INTAPIC(I)	(ID_TO_IO(io_apic_ints[(I)].dst_apic_id))
1309 
1310 #define SRCBUSIRQ(I)	(io_apic_ints[(I)].src_bus_irq)
1311 int
1312 isa_apic_irq(int isa_irq)
1313 {
1314 	int     intr;
1315 
1316 	for (intr = 0; intr < nintrs; ++intr) {		/* check each record */
1317 		if (INTTYPE(intr) == 0) {		/* standard INT */
1318 			if (SRCBUSIRQ(intr) == isa_irq) {
1319 				if (apic_int_is_bus_type(intr, ISA) ||
1320 			            apic_int_is_bus_type(intr, EISA))
1321 					return INTIRQ(intr);	/* found */
1322 			}
1323 		}
1324 	}
1325 	return -1;					/* NOT found */
1326 }
1327 
1328 
1329 /*
1330  * Determine which APIC pin a PCI INT is attached to.
1331  */
1332 #define SRCBUSID(I)	(io_apic_ints[(I)].src_bus_id)
1333 #define SRCBUSDEVICE(I)	((io_apic_ints[(I)].src_bus_irq >> 2) & 0x1f)
1334 #define SRCBUSLINE(I)	(io_apic_ints[(I)].src_bus_irq & 0x03)
1335 int
1336 pci_apic_irq(int pciBus, int pciDevice, int pciInt)
1337 {
1338 	int     intr;
1339 
1340 	--pciInt;					/* zero based */
1341 
1342 	for (intr = 0; intr < nintrs; ++intr)		/* check each record */
1343 		if ((INTTYPE(intr) == 0)		/* standard INT */
1344 		    && (SRCBUSID(intr) == pciBus)
1345 		    && (SRCBUSDEVICE(intr) == pciDevice)
1346 		    && (SRCBUSLINE(intr) == pciInt))	/* a candidate IRQ */
1347 			if (apic_int_is_bus_type(intr, PCI))
1348 				return INTIRQ(intr);	/* exact match */
1349 
1350 	return -1;					/* NOT found */
1351 }
1352 
1353 int
1354 next_apic_irq(int irq)
1355 {
1356 	int intr, ointr;
1357 	int bus, bustype;
1358 
1359 	bus = 0;
1360 	bustype = 0;
1361 	for (intr = 0; intr < nintrs; intr++) {
1362 		if (INTIRQ(intr) != irq || INTTYPE(intr) != 0)
1363 			continue;
1364 		bus = SRCBUSID(intr);
1365 		bustype = apic_bus_type(bus);
1366 		if (bustype != ISA &&
1367 		    bustype != EISA &&
1368 		    bustype != PCI)
1369 			continue;
1370 		break;
1371 	}
1372 	if (intr >= nintrs) {
1373 		return -1;
1374 	}
1375 	for (ointr = intr + 1; ointr < nintrs; ointr++) {
1376 		if (INTTYPE(ointr) != 0)
1377 			continue;
1378 		if (bus != SRCBUSID(ointr))
1379 			continue;
1380 		if (bustype == PCI) {
1381 			if (SRCBUSDEVICE(intr) != SRCBUSDEVICE(ointr))
1382 				continue;
1383 			if (SRCBUSLINE(intr) != SRCBUSLINE(ointr))
1384 				continue;
1385 		}
1386 		if (bustype == ISA || bustype == EISA) {
1387 			if (SRCBUSIRQ(intr) != SRCBUSIRQ(ointr))
1388 				continue;
1389 		}
1390 		if (INTPIN(intr) == INTPIN(ointr))
1391 			continue;
1392 		break;
1393 	}
1394 	if (ointr >= nintrs) {
1395 		return -1;
1396 	}
1397 	return INTIRQ(ointr);
1398 }
1399 #undef SRCBUSLINE
1400 #undef SRCBUSDEVICE
1401 #undef SRCBUSID
1402 #undef SRCBUSIRQ
1403 
1404 #undef INTPIN
1405 #undef INTIRQ
1406 #undef INTAPIC
1407 #undef INTTYPE
1408 
1409 
1410 /*
1411  * Reprogram the MB chipset to NOT redirect an ISA INTerrupt.
1412  *
1413  * XXX FIXME:
1414  *  Exactly what this means is unclear at this point.  It is a solution
1415  *  for motherboards that redirect the MBIRQ0 pin.  Generically a motherboard
1416  *  could route any of the ISA INTs to upper (>15) IRQ values.  But most would
1417  *  NOT be redirected via MBIRQ0, thus "undirect()ing" them would NOT be an
1418  *  option.
1419  */
1420 int
1421 undirect_isa_irq(int rirq)
1422 {
1423 #if defined(READY)
1424 	if (bootverbose)
1425 	    printf("Freeing redirected ISA irq %d.\n", rirq);
1426 	/** FIXME: tickle the MB redirector chip */
1427 	return ???;
1428 #else
1429 	if (bootverbose)
1430 	    printf("Freeing (NOT implemented) redirected ISA irq %d.\n", rirq);
1431 	return 0;
1432 #endif  /* READY */
1433 }
1434 
1435 
1436 /*
1437  * Reprogram the MB chipset to NOT redirect a PCI INTerrupt
1438  */
1439 int
1440 undirect_pci_irq(int rirq)
1441 {
1442 #if defined(READY)
1443 	if (bootverbose)
1444 		printf("Freeing redirected PCI irq %d.\n", rirq);
1445 
1446 	/** FIXME: tickle the MB redirector chip */
1447 	return ???;
1448 #else
1449 	if (bootverbose)
1450 		printf("Freeing (NOT implemented) redirected PCI irq %d.\n",
1451 		       rirq);
1452 	return 0;
1453 #endif  /* READY */
1454 }
1455 
1456 
1457 /*
1458  * given a bus ID, return:
1459  *  the bus type if found
1460  *  -1 if NOT found
1461  */
1462 int
1463 apic_bus_type(int id)
1464 {
1465 	int     x;
1466 
1467 	for (x = 0; x < mp_nbusses; ++x)
1468 		if (bus_data[x].bus_id == id)
1469 			return bus_data[x].bus_type;
1470 
1471 	return -1;
1472 }
1473 
1474 
1475 /*
1476  * given a LOGICAL APIC# and pin#, return:
1477  *  the associated src bus ID if found
1478  *  -1 if NOT found
1479  */
1480 int
1481 apic_src_bus_id(int apic, int pin)
1482 {
1483 	int     x;
1484 
1485 	/* search each of the possible INTerrupt sources */
1486 	for (x = 0; x < nintrs; ++x)
1487 		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1488 		    (pin == io_apic_ints[x].dst_apic_int))
1489 			return (io_apic_ints[x].src_bus_id);
1490 
1491 	return -1;		/* NOT found */
1492 }
1493 
1494 
1495 /*
1496  * given a LOGICAL APIC# and pin#, return:
1497  *  the associated src bus IRQ if found
1498  *  -1 if NOT found
1499  */
1500 int
1501 apic_src_bus_irq(int apic, int pin)
1502 {
1503 	int     x;
1504 
1505 	for (x = 0; x < nintrs; x++)
1506 		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1507 		    (pin == io_apic_ints[x].dst_apic_int))
1508 			return (io_apic_ints[x].src_bus_irq);
1509 
1510 	return -1;		/* NOT found */
1511 }
1512 
1513 
1514 /*
1515  * given a LOGICAL APIC# and pin#, return:
1516  *  the associated INTerrupt type if found
1517  *  -1 if NOT found
1518  */
1519 int
1520 apic_int_type(int apic, int pin)
1521 {
1522 	int     x;
1523 
1524 	/* search each of the possible INTerrupt sources */
1525 	for (x = 0; x < nintrs; ++x)
1526 		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1527 		    (pin == io_apic_ints[x].dst_apic_int))
1528 			return (io_apic_ints[x].int_type);
1529 
1530 	return -1;		/* NOT found */
1531 }
1532 
1533 int
1534 apic_irq(int apic, int pin)
1535 {
1536 	int x;
1537 	int res;
1538 
1539 	for (x = 0; x < nintrs; ++x)
1540 		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1541 		    (pin == io_apic_ints[x].dst_apic_int)) {
1542 			res = io_apic_ints[x].int_vector;
1543 			if (res == 0xff)
1544 				return -1;
1545 			if (apic != int_to_apicintpin[res].ioapic)
1546 				panic("apic_irq: inconsistent table");
1547 			if (pin != int_to_apicintpin[res].int_pin)
1548 				panic("apic_irq inconsistent table (2)");
1549 			return res;
1550 		}
1551 	return -1;
1552 }
1553 
1554 
1555 /*
1556  * given a LOGICAL APIC# and pin#, return:
1557  *  the associated trigger mode if found
1558  *  -1 if NOT found
1559  */
1560 int
1561 apic_trigger(int apic, int pin)
1562 {
1563 	int     x;
1564 
1565 	/* search each of the possible INTerrupt sources */
1566 	for (x = 0; x < nintrs; ++x)
1567 		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1568 		    (pin == io_apic_ints[x].dst_apic_int))
1569 			return ((io_apic_ints[x].int_flags >> 2) & 0x03);
1570 
1571 	return -1;		/* NOT found */
1572 }
1573 
1574 
1575 /*
1576  * given a LOGICAL APIC# and pin#, return:
1577  *  the associated 'active' level if found
1578  *  -1 if NOT found
1579  */
1580 int
1581 apic_polarity(int apic, int pin)
1582 {
1583 	int     x;
1584 
1585 	/* search each of the possible INTerrupt sources */
1586 	for (x = 0; x < nintrs; ++x)
1587 		if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1588 		    (pin == io_apic_ints[x].dst_apic_int))
1589 			return (io_apic_ints[x].int_flags & 0x03);
1590 
1591 	return -1;		/* NOT found */
1592 }
1593 
1594 
1595 /*
1596  * set data according to MP defaults
1597  * FIXME: probably not complete yet...
1598  */
1599 static void
1600 default_mp_table(int type)
1601 {
1602 	int     ap_cpu_id;
1603 #if defined(APIC_IO)
1604 	u_int32_t ux;
1605 	int     io_apic_id;
1606 	int     pin;
1607 #endif	/* APIC_IO */
1608 
1609 #if 0
1610 	printf("  MP default config type: %d\n", type);
1611 	switch (type) {
1612 	case 1:
1613 		printf("   bus: ISA, APIC: 82489DX\n");
1614 		break;
1615 	case 2:
1616 		printf("   bus: EISA, APIC: 82489DX\n");
1617 		break;
1618 	case 3:
1619 		printf("   bus: EISA, APIC: 82489DX\n");
1620 		break;
1621 	case 4:
1622 		printf("   bus: MCA, APIC: 82489DX\n");
1623 		break;
1624 	case 5:
1625 		printf("   bus: ISA+PCI, APIC: Integrated\n");
1626 		break;
1627 	case 6:
1628 		printf("   bus: EISA+PCI, APIC: Integrated\n");
1629 		break;
1630 	case 7:
1631 		printf("   bus: MCA+PCI, APIC: Integrated\n");
1632 		break;
1633 	default:
1634 		printf("   future type\n");
1635 		break;
1636 		/* NOTREACHED */
1637 	}
1638 #endif	/* 0 */
1639 
1640 	boot_cpu_id = (lapic.id & APIC_ID_MASK) >> 24;
1641 	ap_cpu_id = (boot_cpu_id == 0) ? 1 : 0;
1642 
1643 	/* BSP */
1644 	CPU_TO_ID(0) = boot_cpu_id;
1645 	ID_TO_CPU(boot_cpu_id) = 0;
1646 
1647 	/* one and only AP */
1648 	CPU_TO_ID(1) = ap_cpu_id;
1649 	ID_TO_CPU(ap_cpu_id) = 1;
1650 
1651 #if defined(APIC_IO)
1652 	/* one and only IO APIC */
1653 	io_apic_id = (io_apic_read(0, IOAPIC_ID) & APIC_ID_MASK) >> 24;
1654 
1655 	/*
1656 	 * sanity check, refer to MP spec section 3.6.6, last paragraph
1657 	 * necessary as some hardware isn't properly setting up the IO APIC
1658 	 */
1659 #if defined(REALLY_ANAL_IOAPICID_VALUE)
1660 	if (io_apic_id != 2) {
1661 #else
1662 	if ((io_apic_id == 0) || (io_apic_id == 1) || (io_apic_id == 15)) {
1663 #endif	/* REALLY_ANAL_IOAPICID_VALUE */
1664 		ux = io_apic_read(0, IOAPIC_ID);	/* get current contents */
1665 		ux &= ~APIC_ID_MASK;	/* clear the ID field */
1666 		ux |= 0x02000000;	/* set it to '2' */
1667 		io_apic_write(0, IOAPIC_ID, ux);	/* write new value */
1668 		ux = io_apic_read(0, IOAPIC_ID);	/* re-read && test */
1669 		if ((ux & APIC_ID_MASK) != 0x02000000)
1670 			panic("can't control IO APIC ID, reg: 0x%08x", ux);
1671 		io_apic_id = 2;
1672 	}
1673 	IO_TO_ID(0) = io_apic_id;
1674 	ID_TO_IO(io_apic_id) = 0;
1675 #endif	/* APIC_IO */
1676 
1677 	/* fill out bus entries */
1678 	switch (type) {
1679 	case 1:
1680 	case 2:
1681 	case 3:
1682 	case 4:
1683 	case 5:
1684 	case 6:
1685 	case 7:
1686 		bus_data[0].bus_id = default_data[type - 1][1];
1687 		bus_data[0].bus_type = default_data[type - 1][2];
1688 		bus_data[1].bus_id = default_data[type - 1][3];
1689 		bus_data[1].bus_type = default_data[type - 1][4];
1690 		break;
1691 
1692 	/* case 4: case 7:		   MCA NOT supported */
1693 	default:		/* illegal/reserved */
1694 		panic("BAD default MP config: %d", type);
1695 		/* NOTREACHED */
1696 	}
1697 
1698 #if defined(APIC_IO)
1699 	/* general cases from MP v1.4, table 5-2 */
1700 	for (pin = 0; pin < 16; ++pin) {
1701 		io_apic_ints[pin].int_type = 0;
1702 		io_apic_ints[pin].int_flags = 0x05;	/* edge/active-hi */
1703 		io_apic_ints[pin].src_bus_id = 0;
1704 		io_apic_ints[pin].src_bus_irq = pin;	/* IRQ2 caught below */
1705 		io_apic_ints[pin].dst_apic_id = io_apic_id;
1706 		io_apic_ints[pin].dst_apic_int = pin;	/* 1-to-1 */
1707 	}
1708 
1709 	/* special cases from MP v1.4, table 5-2 */
1710 	if (type == 2) {
1711 		io_apic_ints[2].int_type = 0xff;	/* N/C */
1712 		io_apic_ints[13].int_type = 0xff;	/* N/C */
1713 #if !defined(APIC_MIXED_MODE)
1714 		/** FIXME: ??? */
1715 		panic("sorry, can't support type 2 default yet");
1716 #endif	/* APIC_MIXED_MODE */
1717 	}
1718 	else
1719 		io_apic_ints[2].src_bus_irq = 0;	/* ISA IRQ0 is on APIC INT 2 */
1720 
1721 	if (type == 7)
1722 		io_apic_ints[0].int_type = 0xff;	/* N/C */
1723 	else
1724 		io_apic_ints[0].int_type = 3;	/* vectored 8259 */
1725 #endif	/* APIC_IO */
1726 }
1727 
1728 
1729 /*
1730  * initialize all the SMP locks
1731  */
1732 
1733 /* critical region around IO APIC, apic_imen */
1734 struct simplelock	imen_lock;
1735 
1736 /* critical region around splxx(), cpl, cml, cil, ipending */
1737 struct simplelock	cpl_lock;
1738 
1739 /* Make FAST_INTR() routines sequential */
1740 struct simplelock	fast_intr_lock;
1741 
1742 /* critical region around INTR() routines */
1743 struct simplelock	intr_lock;
1744 
1745 /* lock regions protected in UP kernel via cli/sti */
1746 struct simplelock	mpintr_lock;
1747 
1748 /* lock region used by kernel profiling */
1749 struct simplelock	mcount_lock;
1750 
1751 #ifdef USE_COMLOCK
1752 /* locks com (tty) data/hardware accesses: a FASTINTR() */
1753 struct simplelock	com_lock;
1754 #endif /* USE_COMLOCK */
1755 
1756 #ifdef USE_CLOCKLOCK
1757 /* lock regions around the clock hardware */
1758 struct simplelock	clock_lock;
1759 #endif /* USE_CLOCKLOCK */
1760 
1761 /* lock around the MP rendezvous */
1762 static struct simplelock smp_rv_lock;
1763 
1764 static void
1765 init_locks(void)
1766 {
1767 	/*
1768 	 * Get the initial mp_lock with a count of 1 for the BSP.
1769 	 * This uses a LOGICAL cpu ID, ie BSP == 0.
1770 	 */
1771 	mp_lock = 0x00000001;
1772 
1773 #if 0
1774 	/* ISR uses its own "giant lock" */
1775 	isr_lock = FREE_LOCK;
1776 #endif
1777 
1778 #if defined(APIC_INTR_DIAGNOSTIC) && defined(APIC_INTR_DIAGNOSTIC_IRQ)
1779 	s_lock_init((struct simplelock*)&apic_itrace_debuglock);
1780 #endif
1781 
1782 	s_lock_init((struct simplelock*)&mpintr_lock);
1783 
1784 	s_lock_init((struct simplelock*)&mcount_lock);
1785 
1786 	s_lock_init((struct simplelock*)&fast_intr_lock);
1787 	s_lock_init((struct simplelock*)&intr_lock);
1788 	s_lock_init((struct simplelock*)&imen_lock);
1789 	s_lock_init((struct simplelock*)&cpl_lock);
1790 	s_lock_init(&smp_rv_lock);
1791 
1792 #ifdef USE_COMLOCK
1793 	s_lock_init((struct simplelock*)&com_lock);
1794 #endif /* USE_COMLOCK */
1795 #ifdef USE_CLOCKLOCK
1796 	s_lock_init((struct simplelock*)&clock_lock);
1797 #endif /* USE_CLOCKLOCK */
1798 }
1799 
1800 
1801 /* Wait for all APs to be fully initialized */
1802 extern int wait_ap(unsigned int);
1803 
1804 /*
1805  * start each AP in our list
1806  */
1807 static int
1808 start_all_aps(u_int boot_addr)
1809 {
1810 	int     x, i, pg;
1811 	u_char  mpbiosreason;
1812 	u_long  mpbioswarmvec;
1813 	struct globaldata *gd;
1814 	char *stack;
1815 
1816 	POSTCODE(START_ALL_APS_POST);
1817 
1818 	/* initialize BSP's local APIC */
1819 	apic_initialize();
1820 	bsp_apic_ready = 1;
1821 
1822 	/* install the AP 1st level boot code */
1823 	install_ap_tramp(boot_addr);
1824 
1825 
1826 	/* save the current value of the warm-start vector */
1827 	mpbioswarmvec = *((u_long *) WARMBOOT_OFF);
1828 #ifndef PC98
1829 	outb(CMOS_REG, BIOS_RESET);
1830 	mpbiosreason = inb(CMOS_DATA);
1831 #endif
1832 
1833 	/* record BSP in CPU map */
1834 	all_cpus = 1;
1835 
1836 	/* set up 0 -> 4MB P==V mapping for AP boot */
1837 	*(int *)PTD = PG_V | PG_RW | ((uintptr_t)(void *)KPTphys & PG_FRAME);
1838 	invltlb();
1839 
1840 	/* start each AP */
1841 	for (x = 1; x <= mp_naps; ++x) {
1842 
1843 		/* This is a bit verbose, it will go away soon.  */
1844 
1845 		/* first page of AP's private space */
1846 		pg = x * i386_btop(sizeof(struct privatespace));
1847 
1848 		/* allocate a new private data page */
1849 		gd = (struct globaldata *)kmem_alloc(kernel_map, PAGE_SIZE);
1850 
1851 		/* wire it into the private page table page */
1852 		SMPpt[pg] = (pt_entry_t)(PG_V | PG_RW | vtophys(gd));
1853 
1854 		/* allocate and set up an idle stack data page */
1855 		stack = (char *)kmem_alloc(kernel_map, UPAGES*PAGE_SIZE);
1856 		for (i = 0; i < UPAGES; i++)
1857 			SMPpt[pg + 5 + i] = (pt_entry_t)
1858 			    (PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack));
1859 
1860 		SMPpt[pg + 1] = 0;		/* *prv_CMAP1 */
1861 		SMPpt[pg + 2] = 0;		/* *prv_CMAP2 */
1862 		SMPpt[pg + 3] = 0;		/* *prv_CMAP3 */
1863 		SMPpt[pg + 4] = 0;		/* *prv_PMAP1 */
1864 
1865 		/* prime data page for it to use */
1866 		gd->gd_cpuid = x;
1867 		gd->gd_cpu_lockid = x << 24;
1868 		gd->gd_prv_CMAP1 = &SMPpt[pg + 1];
1869 		gd->gd_prv_CMAP2 = &SMPpt[pg + 2];
1870 		gd->gd_prv_CMAP3 = &SMPpt[pg + 3];
1871 		gd->gd_prv_PMAP1 = &SMPpt[pg + 4];
1872 		gd->gd_prv_CADDR1 = SMP_prvspace[x].CPAGE1;
1873 		gd->gd_prv_CADDR2 = SMP_prvspace[x].CPAGE2;
1874 		gd->gd_prv_CADDR3 = SMP_prvspace[x].CPAGE3;
1875 		gd->gd_prv_PADDR1 = (unsigned *)SMP_prvspace[x].PPAGE1;
1876 
1877 		/* setup a vector to our boot code */
1878 		*((volatile u_short *) WARMBOOT_OFF) = WARMBOOT_TARGET;
1879 		*((volatile u_short *) WARMBOOT_SEG) = (boot_addr >> 4);
1880 #ifndef PC98
1881 		outb(CMOS_REG, BIOS_RESET);
1882 		outb(CMOS_DATA, BIOS_WARM);	/* 'warm-start' */
1883 #endif
1884 
1885 		bootSTK = &SMP_prvspace[x].idlestack[UPAGES*PAGE_SIZE];
1886 		bootAP = x;
1887 
1888 		/* attempt to start the Application Processor */
1889 		CHECK_INIT(99);	/* setup checkpoints */
1890 		if (!start_ap(x, boot_addr)) {
1891 			printf("AP #%d (PHY# %d) failed!\n", x, CPU_TO_ID(x));
1892 			CHECK_PRINT("trace");	/* show checkpoints */
1893 			/* better panic as the AP may be running loose */
1894 			printf("panic y/n? [y] ");
1895 			if (cngetc() != 'n')
1896 				panic("bye-bye");
1897 		}
1898 		CHECK_PRINT("trace");		/* show checkpoints */
1899 
1900 		/* record its version info */
1901 		cpu_apic_versions[x] = cpu_apic_versions[0];
1902 
1903 		all_cpus |= (1 << x);		/* record AP in CPU map */
1904 	}
1905 
1906 	/* build our map of 'other' CPUs */
1907 	other_cpus = all_cpus & ~(1 << cpuid);
1908 
1909 	/* fill in our (BSP) APIC version */
1910 	cpu_apic_versions[0] = lapic.version;
1911 
1912 	/* restore the warmstart vector */
1913 	*(u_long *) WARMBOOT_OFF = mpbioswarmvec;
1914 #ifndef PC98
1915 	outb(CMOS_REG, BIOS_RESET);
1916 	outb(CMOS_DATA, mpbiosreason);
1917 #endif
1918 
1919 	/*
1920 	 * Set up the idle context for the BSP.  Similar to above except
1921 	 * that some was done by locore, some by pmap.c and some is implicit
1922 	 * because the BSP is cpu#0 and the page is initially zero, and also
1923 	 * because we can refer to variables by name on the BSP..
1924 	 */
1925 
1926 	/* Allocate and setup BSP idle stack */
1927 	stack = (char *)kmem_alloc(kernel_map, UPAGES * PAGE_SIZE);
1928 	for (i = 0; i < UPAGES; i++)
1929 		SMPpt[5 + i] = (pt_entry_t)
1930 		    (PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack));
1931 
1932 	*(int *)PTD = 0;
1933 	pmap_set_opt();
1934 
1935 	/* number of APs actually started */
1936 	return mp_ncpus - 1;
1937 }
1938 
1939 
1940 /*
1941  * load the 1st level AP boot code into base memory.
1942  */
1943 
1944 /* targets for relocation */
1945 extern void bigJump(void);
1946 extern void bootCodeSeg(void);
1947 extern void bootDataSeg(void);
1948 extern void MPentry(void);
1949 extern u_int MP_GDT;
1950 extern u_int mp_gdtbase;
1951 
1952 static void
1953 install_ap_tramp(u_int boot_addr)
1954 {
1955 	int     x;
1956 	int     size = *(int *) ((u_long) & bootMP_size);
1957 	u_char *src = (u_char *) ((u_long) bootMP);
1958 	u_char *dst = (u_char *) boot_addr + KERNBASE;
1959 	u_int   boot_base = (u_int) bootMP;
1960 	u_int8_t *dst8;
1961 	u_int16_t *dst16;
1962 	u_int32_t *dst32;
1963 
1964 	POSTCODE(INSTALL_AP_TRAMP_POST);
1965 
1966 	for (x = 0; x < size; ++x)
1967 		*dst++ = *src++;
1968 
1969 	/*
1970 	 * modify addresses in code we just moved to basemem. unfortunately we
1971 	 * need fairly detailed info about mpboot.s for this to work.  changes
1972 	 * to mpboot.s might require changes here.
1973 	 */
1974 
1975 	/* boot code is located in KERNEL space */
1976 	dst = (u_char *) boot_addr + KERNBASE;
1977 
1978 	/* modify the lgdt arg */
1979 	dst32 = (u_int32_t *) (dst + ((u_int) & mp_gdtbase - boot_base));
1980 	*dst32 = boot_addr + ((u_int) & MP_GDT - boot_base);
1981 
1982 	/* modify the ljmp target for MPentry() */
1983 	dst32 = (u_int32_t *) (dst + ((u_int) bigJump - boot_base) + 1);
1984 	*dst32 = ((u_int) MPentry - KERNBASE);
1985 
1986 	/* modify the target for boot code segment */
1987 	dst16 = (u_int16_t *) (dst + ((u_int) bootCodeSeg - boot_base));
1988 	dst8 = (u_int8_t *) (dst16 + 1);
1989 	*dst16 = (u_int) boot_addr & 0xffff;
1990 	*dst8 = ((u_int) boot_addr >> 16) & 0xff;
1991 
1992 	/* modify the target for boot data segment */
1993 	dst16 = (u_int16_t *) (dst + ((u_int) bootDataSeg - boot_base));
1994 	dst8 = (u_int8_t *) (dst16 + 1);
1995 	*dst16 = (u_int) boot_addr & 0xffff;
1996 	*dst8 = ((u_int) boot_addr >> 16) & 0xff;
1997 }
1998 
1999 
2000 /*
2001  * this function starts the AP (application processor) identified
2002  * by the APIC ID 'physicalCpu'.  It does quite a "song and dance"
2003  * to accomplish this.  This is necessary because of the nuances
2004  * of the different hardware we might encounter.  It ain't pretty,
2005  * but it seems to work.
2006  */
2007 static int
2008 start_ap(int logical_cpu, u_int boot_addr)
2009 {
2010 	int     physical_cpu;
2011 	int     vector;
2012 	int     cpus;
2013 	u_long  icr_lo, icr_hi;
2014 
2015 	POSTCODE(START_AP_POST);
2016 
2017 	/* get the PHYSICAL APIC ID# */
2018 	physical_cpu = CPU_TO_ID(logical_cpu);
2019 
2020 	/* calculate the vector */
2021 	vector = (boot_addr >> 12) & 0xff;
2022 
2023 	/* used as a watchpoint to signal AP startup */
2024 	cpus = mp_ncpus;
2025 
2026 	/*
2027 	 * first we do an INIT/RESET IPI this INIT IPI might be run, reseting
2028 	 * and running the target CPU. OR this INIT IPI might be latched (P5
2029 	 * bug), CPU waiting for STARTUP IPI. OR this INIT IPI might be
2030 	 * ignored.
2031 	 */
2032 
2033 	/* setup the address for the target AP */
2034 	icr_hi = lapic.icr_hi & ~APIC_ID_MASK;
2035 	icr_hi |= (physical_cpu << 24);
2036 	lapic.icr_hi = icr_hi;
2037 
2038 	/* do an INIT IPI: assert RESET */
2039 	icr_lo = lapic.icr_lo & 0xfff00000;
2040 	lapic.icr_lo = icr_lo | 0x0000c500;
2041 
2042 	/* wait for pending status end */
2043 	while (lapic.icr_lo & APIC_DELSTAT_MASK)
2044 		 /* spin */ ;
2045 
2046 	/* do an INIT IPI: deassert RESET */
2047 	lapic.icr_lo = icr_lo | 0x00008500;
2048 
2049 	/* wait for pending status end */
2050 	u_sleep(10000);		/* wait ~10mS */
2051 	while (lapic.icr_lo & APIC_DELSTAT_MASK)
2052 		 /* spin */ ;
2053 
2054 	/*
2055 	 * next we do a STARTUP IPI: the previous INIT IPI might still be
2056 	 * latched, (P5 bug) this 1st STARTUP would then terminate
2057 	 * immediately, and the previously started INIT IPI would continue. OR
2058 	 * the previous INIT IPI has already run. and this STARTUP IPI will
2059 	 * run. OR the previous INIT IPI was ignored. and this STARTUP IPI
2060 	 * will run.
2061 	 */
2062 
2063 	/* do a STARTUP IPI */
2064 	lapic.icr_lo = icr_lo | 0x00000600 | vector;
2065 	while (lapic.icr_lo & APIC_DELSTAT_MASK)
2066 		 /* spin */ ;
2067 	u_sleep(200);		/* wait ~200uS */
2068 
2069 	/*
2070 	 * finally we do a 2nd STARTUP IPI: this 2nd STARTUP IPI should run IF
2071 	 * the previous STARTUP IPI was cancelled by a latched INIT IPI. OR
2072 	 * this STARTUP IPI will be ignored, as only ONE STARTUP IPI is
2073 	 * recognized after hardware RESET or INIT IPI.
2074 	 */
2075 
2076 	lapic.icr_lo = icr_lo | 0x00000600 | vector;
2077 	while (lapic.icr_lo & APIC_DELSTAT_MASK)
2078 		 /* spin */ ;
2079 	u_sleep(200);		/* wait ~200uS */
2080 
2081 	/* wait for it to start */
2082 	set_apic_timer(5000000);/* == 5 seconds */
2083 	while (read_apic_timer())
2084 		if (mp_ncpus > cpus)
2085 			return 1;	/* return SUCCESS */
2086 
2087 	return 0;		/* return FAILURE */
2088 }
2089 
2090 
2091 /*
2092  * Flush the TLB on all other CPU's
2093  *
2094  * XXX: Needs to handshake and wait for completion before proceding.
2095  */
2096 void
2097 smp_invltlb(void)
2098 {
2099 #if defined(APIC_IO)
2100 	if (smp_started && invltlb_ok)
2101 		all_but_self_ipi(XINVLTLB_OFFSET);
2102 #endif  /* APIC_IO */
2103 }
2104 
2105 void
2106 invlpg(u_int addr)
2107 {
2108 	__asm   __volatile("invlpg (%0)"::"r"(addr):"memory");
2109 
2110 	/* send a message to the other CPUs */
2111 	smp_invltlb();
2112 }
2113 
2114 void
2115 invltlb(void)
2116 {
2117 	u_long  temp;
2118 
2119 	/*
2120 	 * This should be implemented as load_cr3(rcr3()) when load_cr3() is
2121 	 * inlined.
2122 	 */
2123 	__asm __volatile("movl %%cr3, %0; movl %0, %%cr3":"=r"(temp) :: "memory");
2124 
2125 	/* send a message to the other CPUs */
2126 	smp_invltlb();
2127 }
2128 
2129 
2130 /*
2131  * When called the executing CPU will send an IPI to all other CPUs
2132  *  requesting that they halt execution.
2133  *
2134  * Usually (but not necessarily) called with 'other_cpus' as its arg.
2135  *
2136  *  - Signals all CPUs in map to stop.
2137  *  - Waits for each to stop.
2138  *
2139  * Returns:
2140  *  -1: error
2141  *   0: NA
2142  *   1: ok
2143  *
2144  * XXX FIXME: this is not MP-safe, needs a lock to prevent multiple CPUs
2145  *            from executing at same time.
2146  */
2147 int
2148 stop_cpus(u_int map)
2149 {
2150 	if (!smp_started)
2151 		return 0;
2152 
2153 	/* send the Xcpustop IPI to all CPUs in map */
2154 	selected_apic_ipi(map, XCPUSTOP_OFFSET, APIC_DELMODE_FIXED);
2155 
2156 	while ((stopped_cpus & map) != map)
2157 		/* spin */ ;
2158 
2159 	return 1;
2160 }
2161 
2162 
2163 /*
2164  * Called by a CPU to restart stopped CPUs.
2165  *
2166  * Usually (but not necessarily) called with 'stopped_cpus' as its arg.
2167  *
2168  *  - Signals all CPUs in map to restart.
2169  *  - Waits for each to restart.
2170  *
2171  * Returns:
2172  *  -1: error
2173  *   0: NA
2174  *   1: ok
2175  */
2176 int
2177 restart_cpus(u_int map)
2178 {
2179 	if (!smp_started)
2180 		return 0;
2181 
2182 	started_cpus = map;		/* signal other cpus to restart */
2183 
2184 	while ((stopped_cpus & map) != 0) /* wait for each to clear its bit */
2185 		/* spin */ ;
2186 
2187 	return 1;
2188 }
2189 
2190 int smp_active = 0;	/* are the APs allowed to run? */
2191 SYSCTL_INT(_machdep, OID_AUTO, smp_active, CTLFLAG_RW, &smp_active, 0, "");
2192 
2193 /* XXX maybe should be hw.ncpu */
2194 static int smp_cpus = 1;	/* how many cpu's running */
2195 SYSCTL_INT(_machdep, OID_AUTO, smp_cpus, CTLFLAG_RD, &smp_cpus, 0, "");
2196 
2197 int invltlb_ok = 0;	/* throttle smp_invltlb() till safe */
2198 SYSCTL_INT(_machdep, OID_AUTO, invltlb_ok, CTLFLAG_RW, &invltlb_ok, 0, "");
2199 
2200 /* Warning: Do not staticize.  Used from swtch.s */
2201 int do_page_zero_idle = 1; /* bzero pages for fun and profit in idleloop */
2202 SYSCTL_INT(_machdep, OID_AUTO, do_page_zero_idle, CTLFLAG_RW,
2203 	   &do_page_zero_idle, 0, "");
2204 
2205 /* Is forwarding of a interrupt to the CPU holding the ISR lock enabled ? */
2206 int forward_irq_enabled = 1;
2207 SYSCTL_INT(_machdep, OID_AUTO, forward_irq_enabled, CTLFLAG_RW,
2208 	   &forward_irq_enabled, 0, "");
2209 
2210 /* Enable forwarding of a signal to a process running on a different CPU */
2211 static int forward_signal_enabled = 1;
2212 SYSCTL_INT(_machdep, OID_AUTO, forward_signal_enabled, CTLFLAG_RW,
2213 	   &forward_signal_enabled, 0, "");
2214 
2215 /* Enable forwarding of roundrobin to all other cpus */
2216 static int forward_roundrobin_enabled = 1;
2217 SYSCTL_INT(_machdep, OID_AUTO, forward_roundrobin_enabled, CTLFLAG_RW,
2218 	   &forward_roundrobin_enabled, 0, "");
2219 
2220 /*
2221  * This is called once the rest of the system is up and running and we're
2222  * ready to let the AP's out of the pen.
2223  */
2224 void ap_init(void);
2225 
2226 void
2227 ap_init()
2228 {
2229 	u_int	apic_id;
2230 
2231 	/* BSP may have changed PTD while we're waiting for the lock */
2232 	cpu_invltlb();
2233 
2234 	smp_cpus++;
2235 
2236 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
2237 	lidt(&r_idt);
2238 #endif
2239 
2240 	/* Build our map of 'other' CPUs. */
2241 	other_cpus = all_cpus & ~(1 << cpuid);
2242 
2243 	printf("SMP: AP CPU #%d Launched!\n", cpuid);
2244 
2245 	/* XXX FIXME: i386 specific, and redundant: Setup the FPU. */
2246 	load_cr0((rcr0() & ~CR0_EM) | CR0_MP | CR0_NE | CR0_TS);
2247 
2248 	/* set up FPU state on the AP */
2249 	npxinit(__INITIAL_NPXCW__);
2250 
2251 	/* A quick check from sanity claus */
2252 	apic_id = (apic_id_to_logical[(lapic.id & 0x0f000000) >> 24]);
2253 	if (cpuid != apic_id) {
2254 		printf("SMP: cpuid = %d\n", cpuid);
2255 		printf("SMP: apic_id = %d\n", apic_id);
2256 		printf("PTD[MPPTDI] = %p\n", (void *)PTD[MPPTDI]);
2257 		panic("cpuid mismatch! boom!!");
2258 	}
2259 
2260 	/* Init local apic for irq's */
2261 	apic_initialize();
2262 
2263 	/* Set memory range attributes for this CPU to match the BSP */
2264 	mem_range_AP_init();
2265 
2266 	/*
2267 	 * Activate smp_invltlb, although strictly speaking, this isn't
2268 	 * quite correct yet.  We should have a bitfield for cpus willing
2269 	 * to accept TLB flush IPI's or something and sync them.
2270 	 */
2271 	if (smp_cpus == mp_ncpus) {
2272 		invltlb_ok = 1;
2273 		smp_started = 1; /* enable IPI's, tlb shootdown, freezes etc */
2274 		smp_active = 1;	 /* historic */
2275 	}
2276 }
2277 
2278 #ifdef BETTER_CLOCK
2279 
2280 #define CHECKSTATE_USER	0
2281 #define CHECKSTATE_SYS	1
2282 #define CHECKSTATE_INTR	2
2283 
2284 /* Do not staticize.  Used from apic_vector.s */
2285 struct proc*	checkstate_curproc[NCPU];
2286 int		checkstate_cpustate[NCPU];
2287 u_long		checkstate_pc[NCPU];
2288 
2289 extern long	cp_time[CPUSTATES];
2290 
2291 #define PC_TO_INDEX(pc, prof)				\
2292         ((int)(((u_quad_t)((pc) - (prof)->pr_off) *	\
2293             (u_quad_t)((prof)->pr_scale)) >> 16) & ~1)
2294 
2295 static void
2296 addupc_intr_forwarded(struct proc *p, int id, int *astmap)
2297 {
2298 	int i;
2299 	struct uprof *prof;
2300 	u_long pc;
2301 
2302 	pc = checkstate_pc[id];
2303 	prof = &p->p_stats->p_prof;
2304 	if (pc >= prof->pr_off &&
2305 	    (i = PC_TO_INDEX(pc, prof)) < prof->pr_size) {
2306 		if ((p->p_flag & P_OWEUPC) == 0) {
2307 			prof->pr_addr = pc;
2308 			prof->pr_ticks = 1;
2309 			p->p_flag |= P_OWEUPC;
2310 		}
2311 		*astmap |= (1 << id);
2312 	}
2313 }
2314 
2315 static void
2316 forwarded_statclock(int id, int pscnt, int *astmap)
2317 {
2318 	struct pstats *pstats;
2319 	long rss;
2320 	struct rusage *ru;
2321 	struct vmspace *vm;
2322 	int cpustate;
2323 	struct proc *p;
2324 #ifdef GPROF
2325 	register struct gmonparam *g;
2326 	int i;
2327 #endif
2328 
2329 	p = checkstate_curproc[id];
2330 	cpustate = checkstate_cpustate[id];
2331 
2332 	switch (cpustate) {
2333 	case CHECKSTATE_USER:
2334 		if (p->p_flag & P_PROFIL)
2335 			addupc_intr_forwarded(p, id, astmap);
2336 		if (pscnt > 1)
2337 			return;
2338 		p->p_uticks++;
2339 		if (p->p_nice > NZERO)
2340 			cp_time[CP_NICE]++;
2341 		else
2342 			cp_time[CP_USER]++;
2343 		break;
2344 	case CHECKSTATE_SYS:
2345 #ifdef GPROF
2346 		/*
2347 		 * Kernel statistics are just like addupc_intr, only easier.
2348 		 */
2349 		g = &_gmonparam;
2350 		if (g->state == GMON_PROF_ON) {
2351 			i = checkstate_pc[id] - g->lowpc;
2352 			if (i < g->textsize) {
2353 				i /= HISTFRACTION * sizeof(*g->kcount);
2354 				g->kcount[i]++;
2355 			}
2356 		}
2357 #endif
2358 		if (pscnt > 1)
2359 			return;
2360 
2361 		if (!p)
2362 			cp_time[CP_IDLE]++;
2363 		else {
2364 			p->p_sticks++;
2365 			cp_time[CP_SYS]++;
2366 		}
2367 		break;
2368 	case CHECKSTATE_INTR:
2369 	default:
2370 #ifdef GPROF
2371 		/*
2372 		 * Kernel statistics are just like addupc_intr, only easier.
2373 		 */
2374 		g = &_gmonparam;
2375 		if (g->state == GMON_PROF_ON) {
2376 			i = checkstate_pc[id] - g->lowpc;
2377 			if (i < g->textsize) {
2378 				i /= HISTFRACTION * sizeof(*g->kcount);
2379 				g->kcount[i]++;
2380 			}
2381 		}
2382 #endif
2383 		if (pscnt > 1)
2384 			return;
2385 		if (p)
2386 			p->p_iticks++;
2387 		cp_time[CP_INTR]++;
2388 	}
2389 	if (p != NULL) {
2390 		schedclock(p);
2391 
2392 		/* Update resource usage integrals and maximums. */
2393 		if ((pstats = p->p_stats) != NULL &&
2394 		    (ru = &pstats->p_ru) != NULL &&
2395 		    (vm = p->p_vmspace) != NULL) {
2396 			ru->ru_ixrss += pgtok(vm->vm_tsize);
2397 			ru->ru_idrss += pgtok(vm->vm_dsize);
2398 			ru->ru_isrss += pgtok(vm->vm_ssize);
2399 			rss = pgtok(vmspace_resident_count(vm));
2400 			if (ru->ru_maxrss < rss)
2401 				ru->ru_maxrss = rss;
2402         	}
2403 	}
2404 }
2405 
2406 void
2407 forward_statclock(int pscnt)
2408 {
2409 	int map;
2410 	int id;
2411 	int i;
2412 
2413 	/* Kludge. We don't yet have separate locks for the interrupts
2414 	 * and the kernel. This means that we cannot let the other processors
2415 	 * handle complex interrupts while inhibiting them from entering
2416 	 * the kernel in a non-interrupt context.
2417 	 *
2418 	 * What we can do, without changing the locking mechanisms yet,
2419 	 * is letting the other processors handle a very simple interrupt
2420 	 * (wich determines the processor states), and do the main
2421 	 * work ourself.
2422 	 */
2423 
2424 	if (!smp_started || !invltlb_ok || cold || panicstr)
2425 		return;
2426 
2427 	/* Step 1: Probe state   (user, cpu, interrupt, spinlock, idle ) */
2428 
2429 	map = other_cpus & ~stopped_cpus ;
2430 	checkstate_probed_cpus = 0;
2431 	if (map != 0)
2432 		selected_apic_ipi(map,
2433 				  XCPUCHECKSTATE_OFFSET, APIC_DELMODE_FIXED);
2434 
2435 	i = 0;
2436 	while (checkstate_probed_cpus != map) {
2437 		/* spin */
2438 		i++;
2439 		if (i == 100000) {
2440 #ifdef BETTER_CLOCK_DIAGNOSTIC
2441 			printf("forward_statclock: checkstate %x\n",
2442 			       checkstate_probed_cpus);
2443 #endif
2444 			break;
2445 		}
2446 	}
2447 
2448 	/*
2449 	 * Step 2: walk through other processors processes, update ticks and
2450 	 * profiling info.
2451 	 */
2452 
2453 	map = 0;
2454 	for (id = 0; id < mp_ncpus; id++) {
2455 		if (id == cpuid)
2456 			continue;
2457 		if (((1 << id) & checkstate_probed_cpus) == 0)
2458 			continue;
2459 		forwarded_statclock(id, pscnt, &map);
2460 	}
2461 	if (map != 0) {
2462 		checkstate_need_ast |= map;
2463 		selected_apic_ipi(map, XCPUAST_OFFSET, APIC_DELMODE_FIXED);
2464 		i = 0;
2465 		while ((checkstate_need_ast & map) != 0) {
2466 			/* spin */
2467 			i++;
2468 			if (i > 100000) {
2469 #ifdef BETTER_CLOCK_DIAGNOSTIC
2470 				printf("forward_statclock: dropped ast 0x%x\n",
2471 				       checkstate_need_ast & map);
2472 #endif
2473 				break;
2474 			}
2475 		}
2476 	}
2477 }
2478 
2479 void
2480 forward_hardclock(int pscnt)
2481 {
2482 	int map;
2483 	int id;
2484 	struct proc *p;
2485 	struct pstats *pstats;
2486 	int i;
2487 
2488 	/* Kludge. We don't yet have separate locks for the interrupts
2489 	 * and the kernel. This means that we cannot let the other processors
2490 	 * handle complex interrupts while inhibiting them from entering
2491 	 * the kernel in a non-interrupt context.
2492 	 *
2493 	 * What we can do, without changing the locking mechanisms yet,
2494 	 * is letting the other processors handle a very simple interrupt
2495 	 * (wich determines the processor states), and do the main
2496 	 * work ourself.
2497 	 */
2498 
2499 	if (!smp_started || !invltlb_ok || cold || panicstr)
2500 		return;
2501 
2502 	/* Step 1: Probe state   (user, cpu, interrupt, spinlock, idle) */
2503 
2504 	map = other_cpus & ~stopped_cpus ;
2505 	checkstate_probed_cpus = 0;
2506 	if (map != 0)
2507 		selected_apic_ipi(map,
2508 				  XCPUCHECKSTATE_OFFSET, APIC_DELMODE_FIXED);
2509 
2510 	i = 0;
2511 	while (checkstate_probed_cpus != map) {
2512 		/* spin */
2513 		i++;
2514 		if (i == 100000) {
2515 #ifdef BETTER_CLOCK_DIAGNOSTIC
2516 			printf("forward_hardclock: checkstate %x\n",
2517 			       checkstate_probed_cpus);
2518 #endif
2519 			break;
2520 		}
2521 	}
2522 
2523 	/*
2524 	 * Step 2: walk through other processors processes, update virtual
2525 	 * timer and profiling timer. If stathz == 0, also update ticks and
2526 	 * profiling info.
2527 	 */
2528 
2529 	map = 0;
2530 	for (id = 0; id < mp_ncpus; id++) {
2531 		if (id == cpuid)
2532 			continue;
2533 		if (((1 << id) & checkstate_probed_cpus) == 0)
2534 			continue;
2535 		p = checkstate_curproc[id];
2536 		if (p) {
2537 			pstats = p->p_stats;
2538 			if (checkstate_cpustate[id] == CHECKSTATE_USER &&
2539 			    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) &&
2540 			    itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) {
2541 				psignal(p, SIGVTALRM);
2542 				map |= (1 << id);
2543 			}
2544 			if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) &&
2545 			    itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) {
2546 				psignal(p, SIGPROF);
2547 				map |= (1 << id);
2548 			}
2549 		}
2550 		if (stathz == 0) {
2551 			forwarded_statclock( id, pscnt, &map);
2552 		}
2553 	}
2554 	if (map != 0) {
2555 		checkstate_need_ast |= map;
2556 		selected_apic_ipi(map, XCPUAST_OFFSET, APIC_DELMODE_FIXED);
2557 		i = 0;
2558 		while ((checkstate_need_ast & map) != 0) {
2559 			/* spin */
2560 			i++;
2561 			if (i > 100000) {
2562 #ifdef BETTER_CLOCK_DIAGNOSTIC
2563 				printf("forward_hardclock: dropped ast 0x%x\n",
2564 				       checkstate_need_ast & map);
2565 #endif
2566 				break;
2567 			}
2568 		}
2569 	}
2570 }
2571 
2572 #endif /* BETTER_CLOCK */
2573 
2574 void
2575 forward_signal(struct proc *p)
2576 {
2577 	int map;
2578 	int id;
2579 	int i;
2580 
2581 	/* Kludge. We don't yet have separate locks for the interrupts
2582 	 * and the kernel. This means that we cannot let the other processors
2583 	 * handle complex interrupts while inhibiting them from entering
2584 	 * the kernel in a non-interrupt context.
2585 	 *
2586 	 * What we can do, without changing the locking mechanisms yet,
2587 	 * is letting the other processors handle a very simple interrupt
2588 	 * (wich determines the processor states), and do the main
2589 	 * work ourself.
2590 	 */
2591 
2592 	if (!smp_started || !invltlb_ok || cold || panicstr)
2593 		return;
2594 	if (!forward_signal_enabled)
2595 		return;
2596 	while (1) {
2597 		if (p->p_stat != SRUN)
2598 			return;
2599 		id = p->p_oncpu;
2600 		if (id == 0xff)
2601 			return;
2602 		map = (1<<id);
2603 		checkstate_need_ast |= map;
2604 		selected_apic_ipi(map, XCPUAST_OFFSET, APIC_DELMODE_FIXED);
2605 		i = 0;
2606 		while ((checkstate_need_ast & map) != 0) {
2607 			/* spin */
2608 			i++;
2609 			if (i > 100000) {
2610 #if 0
2611 				printf("forward_signal: dropped ast 0x%x\n",
2612 				       checkstate_need_ast & map);
2613 #endif
2614 				break;
2615 			}
2616 		}
2617 		if (id == p->p_oncpu)
2618 			return;
2619 	}
2620 }
2621 
2622 void
2623 forward_roundrobin(void)
2624 {
2625 	u_int map;
2626 	int i;
2627 
2628 	if (!smp_started || !invltlb_ok || cold || panicstr)
2629 		return;
2630 	if (!forward_roundrobin_enabled)
2631 		return;
2632 	resched_cpus |= other_cpus;
2633 	map = other_cpus & ~stopped_cpus ;
2634 #if 1
2635 	selected_apic_ipi(map, XCPUAST_OFFSET, APIC_DELMODE_FIXED);
2636 #else
2637 	(void) all_but_self_ipi(XCPUAST_OFFSET);
2638 #endif
2639 	i = 0;
2640 	while ((checkstate_need_ast & map) != 0) {
2641 		/* spin */
2642 		i++;
2643 		if (i > 100000) {
2644 #if 0
2645 			printf("forward_roundrobin: dropped ast 0x%x\n",
2646 			       checkstate_need_ast & map);
2647 #endif
2648 			break;
2649 		}
2650 	}
2651 }
2652 
2653 
2654 #ifdef APIC_INTR_REORDER
2655 /*
2656  *	Maintain mapping from softintr vector to isr bit in local apic.
2657  */
2658 void
2659 set_lapic_isrloc(int intr, int vector)
2660 {
2661 	if (intr < 0 || intr > 32)
2662 		panic("set_apic_isrloc: bad intr argument: %d",intr);
2663 	if (vector < ICU_OFFSET || vector > 255)
2664 		panic("set_apic_isrloc: bad vector argument: %d",vector);
2665 	apic_isrbit_location[intr].location = &lapic.isr0 + ((vector>>5)<<2);
2666 	apic_isrbit_location[intr].bit = (1<<(vector & 31));
2667 }
2668 #endif
2669 
2670 /*
2671  * All-CPU rendezvous.  CPUs are signalled, all execute the setup function
2672  * (if specified), rendezvous, execute the action function (if specified),
2673  * rendezvous again, execute the teardown function (if specified), and then
2674  * resume.
2675  *
2676  * Note that the supplied external functions _must_ be reentrant and aware
2677  * that they are running in parallel and in an unknown lock context.
2678  */
2679 static void (*smp_rv_setup_func)(void *arg);
2680 static void (*smp_rv_action_func)(void *arg);
2681 static void (*smp_rv_teardown_func)(void *arg);
2682 static void *smp_rv_func_arg;
2683 static volatile int smp_rv_waiters[2];
2684 
2685 void
2686 smp_rendezvous_action(void)
2687 {
2688 	/* setup function */
2689 	if (smp_rv_setup_func != NULL)
2690 		smp_rv_setup_func(smp_rv_func_arg);
2691 	/* spin on entry rendezvous */
2692 	atomic_add_int(&smp_rv_waiters[0], 1);
2693 	while (smp_rv_waiters[0] < mp_ncpus)
2694 		;
2695 	/* action function */
2696 	if (smp_rv_action_func != NULL)
2697 		smp_rv_action_func(smp_rv_func_arg);
2698 	/* spin on exit rendezvous */
2699 	atomic_add_int(&smp_rv_waiters[1], 1);
2700 	while (smp_rv_waiters[1] < mp_ncpus)
2701 		;
2702 	/* teardown function */
2703 	if (smp_rv_teardown_func != NULL)
2704 		smp_rv_teardown_func(smp_rv_func_arg);
2705 }
2706 
2707 void
2708 smp_rendezvous(void (* setup_func)(void *),
2709 	       void (* action_func)(void *),
2710 	       void (* teardown_func)(void *),
2711 	       void *arg)
2712 {
2713 	u_int	efl;
2714 
2715 	/* obtain rendezvous lock */
2716 	s_lock(&smp_rv_lock);		/* XXX sleep here? NOWAIT flag? */
2717 
2718 	/* set static function pointers */
2719 	smp_rv_setup_func = setup_func;
2720 	smp_rv_action_func = action_func;
2721 	smp_rv_teardown_func = teardown_func;
2722 	smp_rv_func_arg = arg;
2723 	smp_rv_waiters[0] = 0;
2724 	smp_rv_waiters[1] = 0;
2725 
2726 	/* disable interrupts on this CPU, save interrupt status */
2727 	efl = read_eflags();
2728 	write_eflags(efl & ~PSL_I);
2729 
2730 	/* signal other processors, which will enter the IPI with interrupts off */
2731 	all_but_self_ipi(XRENDEZVOUS_OFFSET);
2732 
2733 	/* call executor function */
2734 	smp_rendezvous_action();
2735 
2736 	/* restore interrupt flag */
2737 	write_eflags(efl);
2738 
2739 	/* release lock */
2740 	s_unlock(&smp_rv_lock);
2741 }
2742