kernel - SMP - "Fix AP #%d (PHY# %d) failed" issues
[dragonfly.git] / sys / platform / pc64 / x86_64 / mp_machdep.c
CommitLineData
46d4e165
JG
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: src/sys/i386/i386/mp_machdep.c,v 1.115.2.15 2003/03/14 21:22:35 jhb Exp $
26 * $DragonFly: src/sys/platform/pc32/i386/mp_machdep.c,v 1.60 2008/06/07 12:03:52 mneumann Exp $
27 */
28
29#include "opt_cpu.h"
30
31#include <sys/param.h>
32#include <sys/systm.h>
33#include <sys/kernel.h>
34#include <sys/sysctl.h>
35#include <sys/malloc.h>
36#include <sys/memrange.h>
37#include <sys/cons.h> /* cngetc() */
38#include <sys/machintr.h>
39
684a93c4
MD
40#include <sys/mplock2.h>
41
46d4e165
JG
42#include <vm/vm.h>
43#include <vm/vm_param.h>
44#include <vm/pmap.h>
45#include <vm/vm_kern.h>
46#include <vm/vm_extern.h>
47#include <sys/lock.h>
48#include <vm/vm_map.h>
49#include <sys/user.h>
50#ifdef GPROF
51#include <sys/gmon.h>
52#endif
53
54#include <machine/smp.h>
55#include <machine_base/apic/apicreg.h>
56#include <machine/atomic.h>
57#include <machine/cpufunc.h>
58#include <machine_base/apic/mpapic.h>
59#include <machine/psl.h>
60#include <machine/segments.h>
61#include <machine/tss.h>
62#include <machine/specialreg.h>
63#include <machine/globaldata.h>
64
65#include <machine/md_var.h> /* setidt() */
66#include <machine_base/icu/icu.h> /* IPIs */
67#include <machine_base/isa/intr_machdep.h> /* IPIs */
68
69#define FIXUP_EXTRA_APIC_INTS 8 /* additional entries we may create */
70
71#define WARMBOOT_TARGET 0
72#define WARMBOOT_OFF (KERNBASE + 0x0467)
73#define WARMBOOT_SEG (KERNBASE + 0x0469)
74
75#define BIOS_BASE (0xf0000)
76#define BIOS_SIZE (0x10000)
77#define BIOS_COUNT (BIOS_SIZE/4)
78
79#define CMOS_REG (0x70)
80#define CMOS_DATA (0x71)
81#define BIOS_RESET (0x0f)
82#define BIOS_WARM (0x0a)
83
84#define PROCENTRY_FLAG_EN 0x01
85#define PROCENTRY_FLAG_BP 0x02
86#define IOAPICENTRY_FLAG_EN 0x01
87
88
89/* MP Floating Pointer Structure */
90typedef struct MPFPS {
91 char signature[4];
92 u_int32_t pap;
93 u_char length;
94 u_char spec_rev;
95 u_char checksum;
96 u_char mpfb1;
97 u_char mpfb2;
98 u_char mpfb3;
99 u_char mpfb4;
100 u_char mpfb5;
101} *mpfps_t;
102
103/* MP Configuration Table Header */
104typedef struct MPCTH {
105 char signature[4];
106 u_short base_table_length;
107 u_char spec_rev;
108 u_char checksum;
109 u_char oem_id[8];
110 u_char product_id[12];
a5f51ef3 111 u_int32_t oem_table_pointer;
46d4e165
JG
112 u_short oem_table_size;
113 u_short entry_count;
a5f51ef3 114 u_int32_t apic_address;
46d4e165
JG
115 u_short extended_table_length;
116 u_char extended_table_checksum;
117 u_char reserved;
118} *mpcth_t;
119
120
121typedef struct PROCENTRY {
122 u_char type;
123 u_char apic_id;
124 u_char apic_version;
125 u_char cpu_flags;
a5f51ef3
JG
126 u_int32_t cpu_signature;
127 u_int32_t feature_flags;
128 u_int32_t reserved1;
129 u_int32_t reserved2;
46d4e165
JG
130} *proc_entry_ptr;
131
132typedef struct BUSENTRY {
133 u_char type;
134 u_char bus_id;
135 char bus_type[6];
136} *bus_entry_ptr;
137
138typedef struct IOAPICENTRY {
139 u_char type;
140 u_char apic_id;
141 u_char apic_version;
142 u_char apic_flags;
a5f51ef3 143 u_int32_t apic_address;
46d4e165
JG
144} *io_apic_entry_ptr;
145
146typedef struct INTENTRY {
147 u_char type;
148 u_char int_type;
149 u_short int_flags;
150 u_char src_bus_id;
151 u_char src_bus_irq;
152 u_char dst_apic_id;
153 u_char dst_apic_int;
154} *int_entry_ptr;
155
156/* descriptions of MP basetable entries */
157typedef struct BASETABLE_ENTRY {
158 u_char type;
159 u_char length;
160 char name[16];
161} basetable_entry;
162
163/*
164 * this code MUST be enabled here and in mpboot.s.
165 * it follows the very early stages of AP boot by placing values in CMOS ram.
166 * it NORMALLY will never be needed and thus the primitive method for enabling.
167 *
168 */
169#if defined(CHECK_POINTS)
170#define CHECK_READ(A) (outb(CMOS_REG, (A)), inb(CMOS_DATA))
171#define CHECK_WRITE(A,D) (outb(CMOS_REG, (A)), outb(CMOS_DATA, (D)))
172
173#define CHECK_INIT(D); \
174 CHECK_WRITE(0x34, (D)); \
175 CHECK_WRITE(0x35, (D)); \
176 CHECK_WRITE(0x36, (D)); \
177 CHECK_WRITE(0x37, (D)); \
178 CHECK_WRITE(0x38, (D)); \
179 CHECK_WRITE(0x39, (D));
180
181#define CHECK_PRINT(S); \
182 kprintf("%s: %d, %d, %d, %d, %d, %d\n", \
183 (S), \
184 CHECK_READ(0x34), \
185 CHECK_READ(0x35), \
186 CHECK_READ(0x36), \
187 CHECK_READ(0x37), \
188 CHECK_READ(0x38), \
189 CHECK_READ(0x39));
190
191#else /* CHECK_POINTS */
192
193#define CHECK_INIT(D)
194#define CHECK_PRINT(S)
195
196#endif /* CHECK_POINTS */
197
198/*
199 * Values to send to the POST hardware.
200 */
201#define MP_BOOTADDRESS_POST 0x10
202#define MP_PROBE_POST 0x11
203#define MPTABLE_PASS1_POST 0x12
204
205#define MP_START_POST 0x13
206#define MP_ENABLE_POST 0x14
207#define MPTABLE_PASS2_POST 0x15
208
209#define START_ALL_APS_POST 0x16
210#define INSTALL_AP_TRAMP_POST 0x17
211#define START_AP_POST 0x18
212
213#define MP_ANNOUNCE_POST 0x19
214
215static int need_hyperthreading_fixup;
216static u_int logical_cpus;
217u_int logical_cpus_mask;
218
219/** XXX FIXME: where does this really belong, isa.h/isa.c perhaps? */
220int current_postcode;
221
222/** XXX FIXME: what system files declare these??? */
223extern struct region_descriptor r_gdt, r_idt;
224
225int bsp_apic_ready = 0; /* flags useability of BSP apic */
226int mp_naps; /* # of Applications processors */
227int mp_nbusses; /* # of busses */
228#ifdef APIC_IO
229int mp_napics; /* # of IO APICs */
230#endif
231int boot_cpu_id; /* designated BSP */
232vm_offset_t cpu_apic_address;
233#ifdef APIC_IO
234vm_offset_t io_apic_address[NAPICID]; /* NAPICID is more than enough */
235u_int32_t *io_apic_versions;
236#endif
237extern int nkpt;
238
239u_int32_t cpu_apic_versions[MAXCPU];
240int64_t tsc0_offset;
241extern int64_t tsc_offsets[];
242
243#ifdef APIC_IO
244struct apic_intmapinfo int_to_apicintpin[APIC_INTMAPSIZE];
245#endif
246
247/*
248 * APIC ID logical/physical mapping structures.
249 * We oversize these to simplify boot-time config.
250 */
251int cpu_num_to_apic_id[NAPICID];
252#ifdef APIC_IO
253int io_num_to_apic_id[NAPICID];
254#endif
255int apic_id_to_logical[NAPICID];
256
257/* AP uses this during bootstrap. Do not staticize. */
258char *bootSTK;
259static int bootAP;
260
46d4e165
JG
261/*
262 * SMP page table page. Setup by locore to point to a page table
263 * page from which we allocate per-cpu privatespace areas io_apics,
264 * and so forth.
265 */
266
267#define IO_MAPPING_START_INDEX \
268 (SMP_MAXCPU * sizeof(struct privatespace) / PAGE_SIZE)
269
270extern pt_entry_t *SMPpt;
46d4e165
JG
271
272struct pcb stoppcbs[MAXCPU];
273
274extern inthand_t IDTVEC(fast_syscall), IDTVEC(fast_syscall32);
275
46d4e165
JG
276/*
277 * Local data and functions.
278 */
279
280static int mp_capable;
281static u_int boot_address;
282static u_int base_memory;
283static int mp_finish;
284
285static mpfps_t mpfps;
3a918cfd 286static long search_for_sig(u_int32_t target, int count);
46d4e165
JG
287static void mp_enable(u_int boot_addr);
288
289static void mptable_hyperthread_fixup(u_int id_mask);
290static void mptable_pass1(void);
291static int mptable_pass2(void);
292static void default_mp_table(int type);
293static void fix_mp_table(void);
294#ifdef APIC_IO
295static void setup_apic_irq_mapping(void);
296static int apic_int_is_bus_type(int intr, int bus_type);
297#endif
298static int start_all_aps(u_int boot_addr);
bfc09ba0 299#if 0
46d4e165 300static void install_ap_tramp(u_int boot_addr);
bfc09ba0 301#endif
bb467734
MD
302static int start_ap(struct mdglobaldata *gd, u_int boot_addr, int smibest);
303static int smitest(void);
46d4e165
JG
304
305static cpumask_t smp_startup_mask = 1; /* which cpus have been started */
306cpumask_t smp_active_mask = 1; /* which cpus are ready for IPIs etc? */
307SYSCTL_INT(_machdep, OID_AUTO, smp_active, CTLFLAG_RD, &smp_active_mask, 0, "");
308static u_int bootMP_size;
309
310/*
311 * Calculate usable address in base memory for AP trampoline code.
312 */
313u_int
314mp_bootaddress(u_int basemem)
315{
316 POSTCODE(MP_BOOTADDRESS_POST);
317
46d4e165
JG
318 base_memory = basemem;
319
c855ebba
JG
320 bootMP_size = mptramp_end - mptramp_start;
321 boot_address = trunc_page(basemem * 1024); /* round down to 4k boundary */
322 if (((basemem * 1024) - boot_address) < bootMP_size)
323 boot_address -= PAGE_SIZE; /* not enough, lower by 4k */
46d4e165
JG
324 /* 3 levels of page table pages */
325 mptramp_pagetables = boot_address - (PAGE_SIZE * 3);
326
327 return mptramp_pagetables;
328}
329
330
331/*
332 * Look for an Intel MP spec table (ie, SMP capable hardware).
333 */
334int
335mp_probe(void)
336{
3a918cfd 337 long x;
46d4e165
JG
338 u_long segment;
339 u_int32_t target;
340
341 /*
342 * Make sure our SMPpt[] page table is big enough to hold all the
343 * mappings we need.
344 */
345 KKASSERT(IO_MAPPING_START_INDEX < NPTEPG - 2);
346
347 POSTCODE(MP_PROBE_POST);
348
349 /* see if EBDA exists */
350 if ((segment = (u_long) * (u_short *) (KERNBASE + 0x40e)) != 0) {
351 /* search first 1K of EBDA */
352 target = (u_int32_t) (segment << 4);
a9e9c21a 353 if ((x = search_for_sig(target, 1024 / 4)) != -1L)
46d4e165
JG
354 goto found;
355 } else {
356 /* last 1K of base memory, effective 'top of base' passed in */
357 target = (u_int32_t) (base_memory - 0x400);
a9e9c21a 358 if ((x = search_for_sig(target, 1024 / 4)) != -1L)
46d4e165
JG
359 goto found;
360 }
361
362 /* search the BIOS */
363 target = (u_int32_t) BIOS_BASE;
a9e9c21a 364 if ((x = search_for_sig(target, BIOS_COUNT)) != -1L)
46d4e165
JG
365 goto found;
366
367 /* nothing found */
368 mpfps = (mpfps_t)0;
369 mp_capable = 0;
370 return 0;
371
372found:
373 /*
374 * Calculate needed resources. We can safely map physical
375 * memory into SMPpt after mptable_pass1() completes.
376 */
377 mpfps = (mpfps_t)x;
378 mptable_pass1();
379
380 /* flag fact that we are running multiple processors */
381 mp_capable = 1;
382 return 1;
383}
384
385
386/*
387 * Startup the SMP processors.
388 */
389void
390mp_start(void)
391{
392 POSTCODE(MP_START_POST);
393
394 /* look for MP capable motherboard */
395 if (mp_capable)
396 mp_enable(boot_address);
397 else
398 panic("MP hardware not found!");
399}
400
401
402/*
403 * Print various information about the SMP system hardware and setup.
404 */
405void
406mp_announce(void)
407{
408 int x;
409
410 POSTCODE(MP_ANNOUNCE_POST);
411
412 kprintf("DragonFly/MP: Multiprocessor motherboard\n");
413 kprintf(" cpu0 (BSP): apic id: %2d", CPU_TO_ID(0));
414 kprintf(", version: 0x%08x", cpu_apic_versions[0]);
bfc09ba0 415 kprintf(", at 0x%08jx\n", (intmax_t)cpu_apic_address);
46d4e165
JG
416 for (x = 1; x <= mp_naps; ++x) {
417 kprintf(" cpu%d (AP): apic id: %2d", x, CPU_TO_ID(x));
418 kprintf(", version: 0x%08x", cpu_apic_versions[x]);
bfc09ba0 419 kprintf(", at 0x%08jx\n", (intmax_t)cpu_apic_address);
46d4e165
JG
420 }
421
422#if defined(APIC_IO)
423 for (x = 0; x < mp_napics; ++x) {
424 kprintf(" io%d (APIC): apic id: %2d", x, IO_TO_ID(x));
425 kprintf(", version: 0x%08x", io_apic_versions[x]);
bcdeeecd 426 kprintf(", at 0x%08lx\n", io_apic_address[x]);
46d4e165
JG
427 }
428#else
429 kprintf(" Warning: APIC I/O disabled\n");
430#endif /* APIC_IO */
431}
432
433/*
434 * AP cpu's call this to sync up protected mode.
435 *
ec073ddc 436 * WARNING! %gs is not set up on entry. This routine sets up %gs.
46d4e165
JG
437 */
438void
439init_secondary(void)
440{
441 int gsel_tss;
442 int x, myid = bootAP;
443 u_int64_t msr, cr0;
444 struct mdglobaldata *md;
445 struct privatespace *ps;
446
447 ps = &CPU_prvspace[myid];
448
449 gdt_segs[GPROC0_SEL].ssd_base =
450 (long) &ps->mdglobaldata.gd_common_tss;
451 ps->mdglobaldata.mi.gd_prvspace = ps;
452
453 /* We fill the 32-bit segment descriptors */
454 for (x = 0; x < NGDT; x++) {
455 if (x != GPROC0_SEL && x != (GPROC0_SEL + 1))
456 ssdtosd(&gdt_segs[x], &gdt[myid * NGDT + x]);
457 }
458 /* And now a 64-bit one */
459 ssdtosyssd(&gdt_segs[GPROC0_SEL],
460 (struct system_segment_descriptor *)&gdt[myid * NGDT + GPROC0_SEL]);
461
462 r_gdt.rd_limit = NGDT * sizeof(gdt[0]) - 1;
463 r_gdt.rd_base = (long) &gdt[myid * NGDT];
464 lgdt(&r_gdt); /* does magic intra-segment return */
465
ec073ddc
JG
466 /* lgdt() destroys the GSBASE value, so we load GSBASE after lgdt() */
467 wrmsr(MSR_FSBASE, 0); /* User value */
468 wrmsr(MSR_GSBASE, (u_int64_t)ps);
469 wrmsr(MSR_KGSBASE, 0); /* XXX User value while we're in the kernel */
470
46d4e165
JG
471 lidt(&r_idt);
472
473#if 0
474 lldt(_default_ldt);
475 mdcpu->gd_currentldt = _default_ldt;
476#endif
477
478 gsel_tss = GSEL(GPROC0_SEL, SEL_KPL);
479 gdt[myid * NGDT + GPROC0_SEL].sd_type = SDT_SYSTSS;
480
481 md = mdcpu; /* loaded through %gs:0 (mdglobaldata.mi.gd_prvspace)*/
482
483 md->gd_common_tss.tss_rsp0 = 0; /* not used until after switch */
484#if 0 /* JG XXX */
485 md->gd_common_tss.tss_ioopt = (sizeof md->gd_common_tss) << 16;
486#endif
487 md->gd_tss_gdt = &gdt[myid * NGDT + GPROC0_SEL];
488 md->gd_common_tssd = *md->gd_tss_gdt;
489#if 0 /* JG XXX */
490 md->gd_common_tss.tss_ist1 = (long)&doublefault_stack[PAGE_SIZE];
491#endif
492 ltr(gsel_tss);
493
46d4e165
JG
494 /*
495 * Set to a known state:
496 * Set by mpboot.s: CR0_PG, CR0_PE
497 * Set by cpu_setregs: CR0_NE, CR0_MP, CR0_TS, CR0_WP, CR0_AM
498 */
499 cr0 = rcr0();
500 cr0 &= ~(CR0_CD | CR0_NW | CR0_EM);
501 load_cr0(cr0);
502
503 /* Set up the fast syscall stuff */
504 msr = rdmsr(MSR_EFER) | EFER_SCE;
505 wrmsr(MSR_EFER, msr);
506 wrmsr(MSR_LSTAR, (u_int64_t)IDTVEC(fast_syscall));
507 wrmsr(MSR_CSTAR, (u_int64_t)IDTVEC(fast_syscall32));
508 msr = ((u_int64_t)GSEL(GCODE_SEL, SEL_KPL) << 32) |
509 ((u_int64_t)GSEL(GUCODE32_SEL, SEL_UPL) << 48);
510 wrmsr(MSR_STAR, msr);
511 wrmsr(MSR_SF_MASK, PSL_NT|PSL_T|PSL_I|PSL_C|PSL_D);
512
513 pmap_set_opt(); /* PSE/4MB pages, etc */
514#if JGXXX
515 /* Initialize the PAT MSR. */
516 pmap_init_pat();
517#endif
518
519 /* set up CPU registers and state */
520 cpu_setregs();
521
522 /* set up SSE/NX registers */
523 initializecpu();
524
525 /* set up FPU state on the AP */
526 npxinit(__INITIAL_NPXCW__);
ec073ddc
JG
527
528 /* disable the APIC, just to be SURE */
529 lapic->svr &= ~APIC_SVR_ENABLE;
530
531 /* data returned to BSP */
532 cpu_apic_versions[0] = lapic->version;
46d4e165
JG
533}
534
535/*******************************************************************
536 * local functions and data
537 */
538
539/*
540 * start the SMP system
541 */
542static void
543mp_enable(u_int boot_addr)
544{
545 int x;
546#if defined(APIC_IO)
547 int apic;
548 u_int ux;
549#endif /* APIC_IO */
550
551 POSTCODE(MP_ENABLE_POST);
552
553#if 0 /* JGXXX */
554 /* turn on 4MB of V == P addressing so we can get to MP table */
555 *(int *)PTD = PG_V | PG_RW | ((uintptr_t)(void *)KPTphys & PG_FRAME);
556 cpu_invltlb();
2ea1c65d 557#endif
46d4e165
JG
558
559 /* examine the MP table for needed info, uses physical addresses */
560 x = mptable_pass2();
561
2ea1c65d 562#if 0 /* JGXXX */
46d4e165
JG
563 *(int *)PTD = 0;
564 cpu_invltlb();
565#endif /* 0 JGXXX */
566
567 /* can't process default configs till the CPU APIC is pmapped */
568 if (x)
569 default_mp_table(x);
570
571 /* post scan cleanup */
572 fix_mp_table();
573
574#if defined(APIC_IO)
575
576 setup_apic_irq_mapping();
577
578 /* fill the LOGICAL io_apic_versions table */
579 for (apic = 0; apic < mp_napics; ++apic) {
580 ux = io_apic_read(apic, IOAPIC_VER);
581 io_apic_versions[apic] = ux;
582 io_apic_set_id(apic, IO_TO_ID(apic));
583 }
584
585 /* program each IO APIC in the system */
586 for (apic = 0; apic < mp_napics; ++apic)
587 if (io_apic_setup(apic) < 0)
588 panic("IO APIC setup failure");
589
590#endif /* APIC_IO */
591
592 /*
593 * These are required for SMP operation
594 */
595
596 /* install a 'Spurious INTerrupt' vector */
597 setidt(XSPURIOUSINT_OFFSET, Xspuriousint,
a4f18b62 598 SDT_SYSIGT, SEL_KPL, 0);
46d4e165
JG
599
600 /* install an inter-CPU IPI for TLB invalidation */
601 setidt(XINVLTLB_OFFSET, Xinvltlb,
a4f18b62 602 SDT_SYSIGT, SEL_KPL, 0);
46d4e165
JG
603
604 /* install an inter-CPU IPI for IPIQ messaging */
605 setidt(XIPIQ_OFFSET, Xipiq,
a4f18b62 606 SDT_SYSIGT, SEL_KPL, 0);
46d4e165
JG
607
608 /* install a timer vector */
609 setidt(XTIMER_OFFSET, Xtimer,
a4f18b62 610 SDT_SYSIGT, SEL_KPL, 0);
46d4e165
JG
611
612 /* install an inter-CPU IPI for CPU stop/restart */
613 setidt(XCPUSTOP_OFFSET, Xcpustop,
a4f18b62 614 SDT_SYSIGT, SEL_KPL, 0);
46d4e165
JG
615
616 /* start each Application Processor */
617 start_all_aps(boot_addr);
618}
619
620
621/*
622 * look for the MP spec signature
623 */
624
625/* string defined by the Intel MP Spec as identifying the MP table */
626#define MP_SIG 0x5f504d5f /* _MP_ */
627#define NEXT(X) ((X) += 4)
3a918cfd 628static long
46d4e165
JG
629search_for_sig(u_int32_t target, int count)
630{
631 int x;
632 u_int32_t *addr = (u_int32_t *) (KERNBASE + target);
633
634 for (x = 0; x < count; NEXT(x))
635 if (addr[x] == MP_SIG)
636 /* make array index a byte index */
a9e9c21a 637 return (long)(&addr[x]);
46d4e165
JG
638
639 return -1;
640}
641
642
643static basetable_entry basetable_entry_types[] =
644{
645 {0, 20, "Processor"},
646 {1, 8, "Bus"},
647 {2, 8, "I/O APIC"},
648 {3, 8, "I/O INT"},
649 {4, 8, "Local INT"}
650};
651
652typedef struct BUSDATA {
653 u_char bus_id;
654 enum busTypes bus_type;
655} bus_datum;
656
657typedef struct INTDATA {
658 u_char int_type;
659 u_short int_flags;
660 u_char src_bus_id;
661 u_char src_bus_irq;
662 u_char dst_apic_id;
663 u_char dst_apic_int;
664 u_char int_vector;
665} io_int, local_int;
666
667typedef struct BUSTYPENAME {
668 u_char type;
669 char name[7];
670} bus_type_name;
671
672static bus_type_name bus_type_table[] =
673{
674 {CBUS, "CBUS"},
675 {CBUSII, "CBUSII"},
676 {EISA, "EISA"},
677 {MCA, "MCA"},
678 {UNKNOWN_BUSTYPE, "---"},
679 {ISA, "ISA"},
680 {MCA, "MCA"},
681 {UNKNOWN_BUSTYPE, "---"},
682 {UNKNOWN_BUSTYPE, "---"},
683 {UNKNOWN_BUSTYPE, "---"},
684 {UNKNOWN_BUSTYPE, "---"},
685 {UNKNOWN_BUSTYPE, "---"},
686 {PCI, "PCI"},
687 {UNKNOWN_BUSTYPE, "---"},
688 {UNKNOWN_BUSTYPE, "---"},
689 {UNKNOWN_BUSTYPE, "---"},
690 {UNKNOWN_BUSTYPE, "---"},
691 {XPRESS, "XPRESS"},
692 {UNKNOWN_BUSTYPE, "---"}
693};
694/* from MP spec v1.4, table 5-1 */
695static int default_data[7][5] =
696{
697/* nbus, id0, type0, id1, type1 */
698 {1, 0, ISA, 255, 255},
699 {1, 0, EISA, 255, 255},
700 {1, 0, EISA, 255, 255},
701 {1, 0, MCA, 255, 255},
702 {2, 0, ISA, 1, PCI},
703 {2, 0, EISA, 1, PCI},
704 {2, 0, MCA, 1, PCI}
705};
706
707
708/* the bus data */
709static bus_datum *bus_data;
710
711#ifdef APIC_IO
712/* the IO INT data, one entry per possible APIC INTerrupt */
713static io_int *io_apic_ints;
714static int nintrs;
715#endif
716
717static int processor_entry (proc_entry_ptr entry, int cpu);
718static int bus_entry (bus_entry_ptr entry, int bus);
719#ifdef APIC_IO
720static int io_apic_entry (io_apic_entry_ptr entry, int apic);
721static int int_entry (int_entry_ptr entry, int intr);
722#endif
723static int lookup_bus_type (char *name);
724
725
726/*
727 * 1st pass on motherboard's Intel MP specification table.
728 *
729 * initializes:
730 * ncpus = 1
731 *
732 * determines:
733 * cpu_apic_address (common to all CPUs)
734 * io_apic_address[N]
735 * mp_naps
736 * mp_nbusses
737 * mp_napics
738 * nintrs
739 */
740static void
741mptable_pass1(void)
742{
743#ifdef APIC_IO
744 int x;
745#endif
746 mpcth_t cth;
747 int totalSize;
748 void* position;
749 int count;
750 int type;
751 u_int id_mask;
752
753 POSTCODE(MPTABLE_PASS1_POST);
754
755#ifdef APIC_IO
756 /* clear various tables */
757 for (x = 0; x < NAPICID; ++x) {
758 io_apic_address[x] = ~0; /* IO APIC address table */
759 }
760#endif
761
762 /* init everything to empty */
763 mp_naps = 0;
764 mp_nbusses = 0;
765#ifdef APIC_IO
766 mp_napics = 0;
767 nintrs = 0;
768#endif
769 id_mask = 0;
770
771 /* check for use of 'default' configuration */
772 if (mpfps->mpfb1 != 0) {
773 /* use default addresses */
774 cpu_apic_address = DEFAULT_APIC_BASE;
775#ifdef APIC_IO
776 io_apic_address[0] = DEFAULT_IO_APIC_BASE;
777#endif
778
779 /* fill in with defaults */
780 mp_naps = 2; /* includes BSP */
781 mp_nbusses = default_data[mpfps->mpfb1 - 1][0];
782#if defined(APIC_IO)
783 mp_napics = 1;
784 nintrs = 16;
785#endif /* APIC_IO */
786 }
787 else {
bfc09ba0 788 if (mpfps->pap == 0)
46d4e165 789 panic("MP Configuration Table Header MISSING!");
1f71eeb1 790 cth = (void *)PHYS_TO_DMAP(mpfps->pap);
46d4e165
JG
791
792 cpu_apic_address = (vm_offset_t) cth->apic_address;
793
794 /* walk the table, recording info of interest */
795 totalSize = cth->base_table_length - sizeof(struct MPCTH);
796 position = (u_char *) cth + sizeof(struct MPCTH);
797 count = cth->entry_count;
798
799 while (count--) {
800 switch (type = *(u_char *) position) {
801 case 0: /* processor_entry */
802 if (((proc_entry_ptr)position)->cpu_flags
803 & PROCENTRY_FLAG_EN) {
804 ++mp_naps;
805 id_mask |= 1 <<
806 ((proc_entry_ptr)position)->apic_id;
807 }
808 break;
809 case 1: /* bus_entry */
810 ++mp_nbusses;
811 break;
812 case 2: /* io_apic_entry */
813#ifdef APIC_IO
814 if (((io_apic_entry_ptr)position)->apic_flags
815 & IOAPICENTRY_FLAG_EN)
816 io_apic_address[mp_napics++] =
817 (vm_offset_t)((io_apic_entry_ptr)
818 position)->apic_address;
819#endif
820 break;
821 case 3: /* int_entry */
822#ifdef APIC_IO
823 ++nintrs;
824#endif
825 break;
826 case 4: /* int_entry */
827 break;
828 default:
829 panic("mpfps Base Table HOSED!");
830 /* NOTREACHED */
831 }
832
833 totalSize -= basetable_entry_types[type].length;
834 position = (uint8_t *)position +
835 basetable_entry_types[type].length;
836 }
837 }
838
839 /* qualify the numbers */
840 if (mp_naps > MAXCPU) {
841 kprintf("Warning: only using %d of %d available CPUs!\n",
842 MAXCPU, mp_naps);
843 mp_naps = MAXCPU;
844 }
845
846 /* See if we need to fixup HT logical CPUs. */
847 mptable_hyperthread_fixup(id_mask);
848
849 /*
850 * Count the BSP.
851 * This is also used as a counter while starting the APs.
852 */
853 ncpus = 1;
854
855 --mp_naps; /* subtract the BSP */
856}
857
858
859/*
860 * 2nd pass on motherboard's Intel MP specification table.
861 *
862 * sets:
863 * boot_cpu_id
864 * ID_TO_IO(N), phy APIC ID to log CPU/IO table
865 * CPU_TO_ID(N), logical CPU to APIC ID table
866 * IO_TO_ID(N), logical IO to APIC ID table
867 * bus_data[N]
868 * io_apic_ints[N]
869 */
870static int
871mptable_pass2(void)
872{
873 struct PROCENTRY proc;
874 int x;
875 mpcth_t cth;
876 int totalSize;
877 void* position;
878 int count;
879 int type;
880 int apic, bus, cpu, intr;
881 int i;
882
883 POSTCODE(MPTABLE_PASS2_POST);
884
885 /* Initialize fake proc entry for use with HT fixup. */
886 bzero(&proc, sizeof(proc));
887 proc.type = 0;
888 proc.cpu_flags = PROCENTRY_FLAG_EN;
889
890#ifdef APIC_IO
891 MALLOC(io_apic_versions, u_int32_t *, sizeof(u_int32_t) * mp_napics,
892 M_DEVBUF, M_WAITOK);
893 MALLOC(ioapic, volatile ioapic_t **, sizeof(ioapic_t *) * mp_napics,
894 M_DEVBUF, M_WAITOK | M_ZERO);
895 MALLOC(io_apic_ints, io_int *, sizeof(io_int) * (nintrs + FIXUP_EXTRA_APIC_INTS),
896 M_DEVBUF, M_WAITOK);
897#endif
898 MALLOC(bus_data, bus_datum *, sizeof(bus_datum) * mp_nbusses,
899 M_DEVBUF, M_WAITOK);
900
901#ifdef APIC_IO
902 for (i = 0; i < mp_napics; i++) {
903 ioapic[i] = permanent_io_mapping(io_apic_address[i]);
904 }
905#endif
906
907 /* clear various tables */
908 for (x = 0; x < NAPICID; ++x) {
909 CPU_TO_ID(x) = -1; /* logical CPU to APIC ID table */
910#ifdef APIC_IO
911 ID_TO_IO(x) = -1; /* phy APIC ID to log CPU/IO table */
912 IO_TO_ID(x) = -1; /* logical IO to APIC ID table */
913#endif
914 }
915
916 /* clear bus data table */
917 for (x = 0; x < mp_nbusses; ++x)
918 bus_data[x].bus_id = 0xff;
919
920#ifdef APIC_IO
921 /* clear IO APIC INT table */
922 for (x = 0; x < (nintrs + 1); ++x) {
923 io_apic_ints[x].int_type = 0xff;
924 io_apic_ints[x].int_vector = 0xff;
925 }
926#endif
927
928 /* setup the cpu/apic mapping arrays */
929 boot_cpu_id = -1;
930
931 /* record whether PIC or virtual-wire mode */
932 machintr_setvar_simple(MACHINTR_VAR_IMCR_PRESENT, mpfps->mpfb2 & 0x80);
933
934 /* check for use of 'default' configuration */
935 if (mpfps->mpfb1 != 0)
936 return mpfps->mpfb1; /* return default configuration type */
937
bfc09ba0 938 if (mpfps->pap == 0)
46d4e165
JG
939 panic("MP Configuration Table Header MISSING!");
940
bfc09ba0 941 cth = (void *)PHYS_TO_DMAP(mpfps->pap);
46d4e165
JG
942 /* walk the table, recording info of interest */
943 totalSize = cth->base_table_length - sizeof(struct MPCTH);
944 position = (u_char *) cth + sizeof(struct MPCTH);
945 count = cth->entry_count;
946 apic = bus = intr = 0;
947 cpu = 1; /* pre-count the BSP */
948
949 while (count--) {
950 switch (type = *(u_char *) position) {
951 case 0:
952 if (processor_entry(position, cpu))
953 ++cpu;
954
955 if (need_hyperthreading_fixup) {
956 /*
957 * Create fake mptable processor entries
958 * and feed them to processor_entry() to
959 * enumerate the logical CPUs.
960 */
961 proc.apic_id = ((proc_entry_ptr)position)->apic_id;
962 for (i = 1; i < logical_cpus; i++) {
963 proc.apic_id++;
964 processor_entry(&proc, cpu);
965 logical_cpus_mask |= (1 << cpu);
966 cpu++;
967 }
968 }
969 break;
970 case 1:
971 if (bus_entry(position, bus))
972 ++bus;
973 break;
974 case 2:
975#ifdef APIC_IO
976 if (io_apic_entry(position, apic))
977 ++apic;
978#endif
979 break;
980 case 3:
981#ifdef APIC_IO
982 if (int_entry(position, intr))
983 ++intr;
984#endif
985 break;
986 case 4:
987 /* int_entry(position); */
988 break;
989 default:
990 panic("mpfps Base Table HOSED!");
991 /* NOTREACHED */
992 }
993
994 totalSize -= basetable_entry_types[type].length;
995 position = (uint8_t *)position + basetable_entry_types[type].length;
996 }
997
998 if (boot_cpu_id == -1)
999 panic("NO BSP found!");
1000
1001 /* report fact that its NOT a default configuration */
1002 return 0;
1003}
1004
1005/*
1006 * Check if we should perform a hyperthreading "fix-up" to
1007 * enumerate any logical CPU's that aren't already listed
1008 * in the table.
1009 *
1010 * XXX: We assume that all of the physical CPUs in the
1011 * system have the same number of logical CPUs.
1012 *
1013 * XXX: We assume that APIC ID's are allocated such that
1014 * the APIC ID's for a physical processor are aligned
1015 * with the number of logical CPU's in the processor.
1016 */
1017static void
1018mptable_hyperthread_fixup(u_int id_mask)
1019{
1020 u_int i, id;
1021
1022 /* Nothing to do if there is no HTT support. */
1023 if ((cpu_feature & CPUID_HTT) == 0)
1024 return;
1025 logical_cpus = (cpu_procinfo & CPUID_HTT_CORES) >> 16;
1026 if (logical_cpus <= 1)
1027 return;
1028
1029 /*
1030 * For each APIC ID of a CPU that is set in the mask,
1031 * scan the other candidate APIC ID's for this
1032 * physical processor. If any of those ID's are
1033 * already in the table, then kill the fixup.
1034 */
1035 for (id = 0; id <= MAXCPU; id++) {
1036 if ((id_mask & 1 << id) == 0)
1037 continue;
1038 /* First, make sure we are on a logical_cpus boundary. */
1039 if (id % logical_cpus != 0)
1040 return;
1041 for (i = id + 1; i < id + logical_cpus; i++)
1042 if ((id_mask & 1 << i) != 0)
1043 return;
1044 }
1045
1046 /*
1047 * Ok, the ID's checked out, so enable the fixup. We have to fixup
1048 * mp_naps right now.
1049 */
1050 need_hyperthreading_fixup = 1;
1051 mp_naps *= logical_cpus;
1052}
1053
1054#ifdef APIC_IO
1055
1056void
1057assign_apic_irq(int apic, int intpin, int irq)
1058{
1059 int x;
1060
1061 if (int_to_apicintpin[irq].ioapic != -1)
1062 panic("assign_apic_irq: inconsistent table");
1063
1064 int_to_apicintpin[irq].ioapic = apic;
1065 int_to_apicintpin[irq].int_pin = intpin;
1066 int_to_apicintpin[irq].apic_address = ioapic[apic];
1067 int_to_apicintpin[irq].redirindex = IOAPIC_REDTBL + 2 * intpin;
1068
1069 for (x = 0; x < nintrs; x++) {
1070 if ((io_apic_ints[x].int_type == 0 ||
1071 io_apic_ints[x].int_type == 3) &&
1072 io_apic_ints[x].int_vector == 0xff &&
1073 io_apic_ints[x].dst_apic_id == IO_TO_ID(apic) &&
1074 io_apic_ints[x].dst_apic_int == intpin)
1075 io_apic_ints[x].int_vector = irq;
1076 }
1077}
1078
1079void
1080revoke_apic_irq(int irq)
1081{
1082 int x;
1083 int oldapic;
1084 int oldintpin;
1085
1086 if (int_to_apicintpin[irq].ioapic == -1)
1087 panic("revoke_apic_irq: inconsistent table");
1088
1089 oldapic = int_to_apicintpin[irq].ioapic;
1090 oldintpin = int_to_apicintpin[irq].int_pin;
1091
1092 int_to_apicintpin[irq].ioapic = -1;
1093 int_to_apicintpin[irq].int_pin = 0;
1094 int_to_apicintpin[irq].apic_address = NULL;
1095 int_to_apicintpin[irq].redirindex = 0;
1096
1097 for (x = 0; x < nintrs; x++) {
1098 if ((io_apic_ints[x].int_type == 0 ||
1099 io_apic_ints[x].int_type == 3) &&
1100 io_apic_ints[x].int_vector != 0xff &&
1101 io_apic_ints[x].dst_apic_id == IO_TO_ID(oldapic) &&
1102 io_apic_ints[x].dst_apic_int == oldintpin)
1103 io_apic_ints[x].int_vector = 0xff;
1104 }
1105}
1106
1107/*
1108 * Allocate an IRQ
1109 */
1110static void
1111allocate_apic_irq(int intr)
1112{
1113 int apic;
1114 int intpin;
1115 int irq;
1116
1117 if (io_apic_ints[intr].int_vector != 0xff)
1118 return; /* Interrupt handler already assigned */
1119
1120 if (io_apic_ints[intr].int_type != 0 &&
1121 (io_apic_ints[intr].int_type != 3 ||
1122 (io_apic_ints[intr].dst_apic_id == IO_TO_ID(0) &&
1123 io_apic_ints[intr].dst_apic_int == 0)))
1124 return; /* Not INT or ExtInt on != (0, 0) */
1125
1126 irq = 0;
1127 while (irq < APIC_INTMAPSIZE &&
1128 int_to_apicintpin[irq].ioapic != -1)
1129 irq++;
1130
1131 if (irq >= APIC_INTMAPSIZE)
1132 return; /* No free interrupt handlers */
1133
1134 apic = ID_TO_IO(io_apic_ints[intr].dst_apic_id);
1135 intpin = io_apic_ints[intr].dst_apic_int;
1136
1137 assign_apic_irq(apic, intpin, irq);
46d4e165
JG
1138}
1139
1140
1141static void
1142swap_apic_id(int apic, int oldid, int newid)
1143{
1144 int x;
1145 int oapic;
1146
1147
1148 if (oldid == newid)
1149 return; /* Nothing to do */
1150
1151 kprintf("Changing APIC ID for IO APIC #%d from %d to %d in MP table\n",
1152 apic, oldid, newid);
1153
1154 /* Swap physical APIC IDs in interrupt entries */
1155 for (x = 0; x < nintrs; x++) {
1156 if (io_apic_ints[x].dst_apic_id == oldid)
1157 io_apic_ints[x].dst_apic_id = newid;
1158 else if (io_apic_ints[x].dst_apic_id == newid)
1159 io_apic_ints[x].dst_apic_id = oldid;
1160 }
1161
1162 /* Swap physical APIC IDs in IO_TO_ID mappings */
1163 for (oapic = 0; oapic < mp_napics; oapic++)
1164 if (IO_TO_ID(oapic) == newid)
1165 break;
1166
1167 if (oapic < mp_napics) {
1168 kprintf("Changing APIC ID for IO APIC #%d from "
1169 "%d to %d in MP table\n",
1170 oapic, newid, oldid);
1171 IO_TO_ID(oapic) = oldid;
1172 }
1173 IO_TO_ID(apic) = newid;
1174}
1175
1176
1177static void
1178fix_id_to_io_mapping(void)
1179{
1180 int x;
1181
1182 for (x = 0; x < NAPICID; x++)
1183 ID_TO_IO(x) = -1;
1184
1185 for (x = 0; x <= mp_naps; x++)
1186 if (CPU_TO_ID(x) < NAPICID)
1187 ID_TO_IO(CPU_TO_ID(x)) = x;
1188
1189 for (x = 0; x < mp_napics; x++)
1190 if (IO_TO_ID(x) < NAPICID)
1191 ID_TO_IO(IO_TO_ID(x)) = x;
1192}
1193
1194
1195static int
1196first_free_apic_id(void)
1197{
1198 int freeid, x;
1199
1200 for (freeid = 0; freeid < NAPICID; freeid++) {
1201 for (x = 0; x <= mp_naps; x++)
1202 if (CPU_TO_ID(x) == freeid)
1203 break;
1204 if (x <= mp_naps)
1205 continue;
1206 for (x = 0; x < mp_napics; x++)
1207 if (IO_TO_ID(x) == freeid)
1208 break;
1209 if (x < mp_napics)
1210 continue;
1211 return freeid;
1212 }
1213 return freeid;
1214}
1215
1216
1217static int
1218io_apic_id_acceptable(int apic, int id)
1219{
1220 int cpu; /* Logical CPU number */
1221 int oapic; /* Logical IO APIC number for other IO APIC */
1222
1223 if (id >= NAPICID)
1224 return 0; /* Out of range */
1225
1226 for (cpu = 0; cpu <= mp_naps; cpu++)
1227 if (CPU_TO_ID(cpu) == id)
1228 return 0; /* Conflict with CPU */
1229
1230 for (oapic = 0; oapic < mp_napics && oapic < apic; oapic++)
1231 if (IO_TO_ID(oapic) == id)
1232 return 0; /* Conflict with other APIC */
1233
1234 return 1; /* ID is acceptable for IO APIC */
1235}
1236
1237static
1238io_int *
1239io_apic_find_int_entry(int apic, int pin)
1240{
1241 int x;
1242
1243 /* search each of the possible INTerrupt sources */
1244 for (x = 0; x < nintrs; ++x) {
1245 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1246 (pin == io_apic_ints[x].dst_apic_int))
1247 return (&io_apic_ints[x]);
1248 }
1249 return NULL;
1250}
1251
1252#endif
1253
1254/*
1255 * parse an Intel MP specification table
1256 */
1257static void
1258fix_mp_table(void)
1259{
1260 int x;
1261#ifdef APIC_IO
1262 int id;
1263 int apic; /* IO APIC unit number */
1264 int freeid; /* Free physical APIC ID */
1265 int physid; /* Current physical IO APIC ID */
1266 io_int *io14;
1267#endif
1268 int bus_0 = 0; /* Stop GCC warning */
1269 int bus_pci = 0; /* Stop GCC warning */
1270 int num_pci_bus;
1271
1272 /*
1273 * Fix mis-numbering of the PCI bus and its INT entries if the BIOS
1274 * did it wrong. The MP spec says that when more than 1 PCI bus
1275 * exists the BIOS must begin with bus entries for the PCI bus and use
1276 * actual PCI bus numbering. This implies that when only 1 PCI bus
1277 * exists the BIOS can choose to ignore this ordering, and indeed many
1278 * MP motherboards do ignore it. This causes a problem when the PCI
1279 * sub-system makes requests of the MP sub-system based on PCI bus
1280 * numbers. So here we look for the situation and renumber the
1281 * busses and associated INTs in an effort to "make it right".
1282 */
1283
1284 /* find bus 0, PCI bus, count the number of PCI busses */
1285 for (num_pci_bus = 0, x = 0; x < mp_nbusses; ++x) {
1286 if (bus_data[x].bus_id == 0) {
1287 bus_0 = x;
1288 }
1289 if (bus_data[x].bus_type == PCI) {
1290 ++num_pci_bus;
1291 bus_pci = x;
1292 }
1293 }
1294 /*
1295 * bus_0 == slot of bus with ID of 0
1296 * bus_pci == slot of last PCI bus encountered
1297 */
1298
1299 /* check the 1 PCI bus case for sanity */
1300 /* if it is number 0 all is well */
1301 if (num_pci_bus == 1 &&
1302 bus_data[bus_pci].bus_id != 0) {
1303
1304 /* mis-numbered, swap with whichever bus uses slot 0 */
1305
1306 /* swap the bus entry types */
1307 bus_data[bus_pci].bus_type = bus_data[bus_0].bus_type;
1308 bus_data[bus_0].bus_type = PCI;
1309
1310#ifdef APIC_IO
1311 /* swap each relavant INTerrupt entry */
1312 id = bus_data[bus_pci].bus_id;
1313 for (x = 0; x < nintrs; ++x) {
1314 if (io_apic_ints[x].src_bus_id == id) {
1315 io_apic_ints[x].src_bus_id = 0;
1316 }
1317 else if (io_apic_ints[x].src_bus_id == 0) {
1318 io_apic_ints[x].src_bus_id = id;
1319 }
1320 }
1321#endif
1322 }
1323
1324#ifdef APIC_IO
1325 /* Assign IO APIC IDs.
1326 *
1327 * First try the existing ID. If a conflict is detected, try
1328 * the ID in the MP table. If a conflict is still detected, find
1329 * a free id.
1330 *
1331 * We cannot use the ID_TO_IO table before all conflicts has been
1332 * resolved and the table has been corrected.
1333 */
1334 for (apic = 0; apic < mp_napics; ++apic) { /* For all IO APICs */
1335
1336 /* First try to use the value set by the BIOS */
1337 physid = io_apic_get_id(apic);
1338 if (io_apic_id_acceptable(apic, physid)) {
1339 if (IO_TO_ID(apic) != physid)
1340 swap_apic_id(apic, IO_TO_ID(apic), physid);
1341 continue;
1342 }
1343
1344 /* Then check if the value in the MP table is acceptable */
1345 if (io_apic_id_acceptable(apic, IO_TO_ID(apic)))
1346 continue;
1347
1348 /* Last resort, find a free APIC ID and use it */
1349 freeid = first_free_apic_id();
1350 if (freeid >= NAPICID)
1351 panic("No free physical APIC IDs found");
1352
1353 if (io_apic_id_acceptable(apic, freeid)) {
1354 swap_apic_id(apic, IO_TO_ID(apic), freeid);
1355 continue;
1356 }
1357 panic("Free physical APIC ID not usable");
1358 }
1359 fix_id_to_io_mapping();
1360#endif
1361
1362#ifdef APIC_IO
1363 /* detect and fix broken Compaq MP table */
1364 if (apic_int_type(0, 0) == -1) {
1365 kprintf("APIC_IO: MP table broken: 8259->APIC entry missing!\n");
1366 io_apic_ints[nintrs].int_type = 3; /* ExtInt */
1367 io_apic_ints[nintrs].int_vector = 0xff; /* Unassigned */
1368 /* XXX fixme, set src bus id etc, but it doesn't seem to hurt */
1369 io_apic_ints[nintrs].dst_apic_id = IO_TO_ID(0);
1370 io_apic_ints[nintrs].dst_apic_int = 0; /* Pin 0 */
1371 nintrs++;
1372 } else if (apic_int_type(0, 0) == 0) {
1373 kprintf("APIC_IO: MP table broken: ExtINT entry corrupt!\n");
1374 for (x = 0; x < nintrs; ++x)
1375 if ((0 == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1376 (0 == io_apic_ints[x].dst_apic_int)) {
1377 io_apic_ints[x].int_type = 3;
1378 io_apic_ints[x].int_vector = 0xff;
1379 break;
1380 }
1381 }
1382
1383 /*
1384 * Fix missing IRQ 15 when IRQ 14 is an ISA interrupt. IDE
1385 * controllers universally come in pairs. If IRQ 14 is specified
1386 * as an ISA interrupt, then IRQ 15 had better be too.
1387 *
1388 * [ Shuttle XPC / AMD Athlon X2 ]
1389 * The MPTable is missing an entry for IRQ 15. Note that the
1390 * ACPI table has an entry for both 14 and 15.
1391 */
1392 if (apic_int_type(0, 14) == 0 && apic_int_type(0, 15) == -1) {
1393 kprintf("APIC_IO: MP table broken: IRQ 15 not ISA when IRQ 14 is!\n");
1394 io14 = io_apic_find_int_entry(0, 14);
1395 io_apic_ints[nintrs] = *io14;
1396 io_apic_ints[nintrs].src_bus_irq = 15;
1397 io_apic_ints[nintrs].dst_apic_int = 15;
1398 nintrs++;
1399 }
1400#endif
1401}
1402
1403#ifdef APIC_IO
1404
1405/* Assign low level interrupt handlers */
1406static void
1407setup_apic_irq_mapping(void)
1408{
1409 int x;
1410 int int_vector;
1411
1412 /* Clear array */
1413 for (x = 0; x < APIC_INTMAPSIZE; x++) {
1414 int_to_apicintpin[x].ioapic = -1;
1415 int_to_apicintpin[x].int_pin = 0;
1416 int_to_apicintpin[x].apic_address = NULL;
1417 int_to_apicintpin[x].redirindex = 0;
1418 }
1419
1420 /* First assign ISA/EISA interrupts */
1421 for (x = 0; x < nintrs; x++) {
1422 int_vector = io_apic_ints[x].src_bus_irq;
1423 if (int_vector < APIC_INTMAPSIZE &&
1424 io_apic_ints[x].int_vector == 0xff &&
1425 int_to_apicintpin[int_vector].ioapic == -1 &&
1426 (apic_int_is_bus_type(x, ISA) ||
1427 apic_int_is_bus_type(x, EISA)) &&
1428 io_apic_ints[x].int_type == 0) {
1429 assign_apic_irq(ID_TO_IO(io_apic_ints[x].dst_apic_id),
1430 io_apic_ints[x].dst_apic_int,
1431 int_vector);
1432 }
1433 }
1434
1435 /* Assign ExtInt entry if no ISA/EISA interrupt 0 entry */
1436 for (x = 0; x < nintrs; x++) {
1437 if (io_apic_ints[x].dst_apic_int == 0 &&
1438 io_apic_ints[x].dst_apic_id == IO_TO_ID(0) &&
1439 io_apic_ints[x].int_vector == 0xff &&
1440 int_to_apicintpin[0].ioapic == -1 &&
1441 io_apic_ints[x].int_type == 3) {
1442 assign_apic_irq(0, 0, 0);
1443 break;
1444 }
1445 }
662f60ef
SZ
1446
1447 /* Assign PCI interrupts */
1448 for (x = 0; x < nintrs; ++x) {
1449 if (io_apic_ints[x].int_type == 0 &&
1450 io_apic_ints[x].int_vector == 0xff &&
1451 apic_int_is_bus_type(x, PCI))
1452 allocate_apic_irq(x);
1453 }
46d4e165
JG
1454}
1455
1456#endif
1457
1458static int
1459processor_entry(proc_entry_ptr entry, int cpu)
1460{
1461 /* check for usability */
1462 if (!(entry->cpu_flags & PROCENTRY_FLAG_EN))
1463 return 0;
1464
1465 if(entry->apic_id >= NAPICID)
1466 panic("CPU APIC ID out of range (0..%d)", NAPICID - 1);
1467 /* check for BSP flag */
1468 if (entry->cpu_flags & PROCENTRY_FLAG_BP) {
1469 boot_cpu_id = entry->apic_id;
1470 CPU_TO_ID(0) = entry->apic_id;
1471 ID_TO_CPU(entry->apic_id) = 0;
1472 return 0; /* its already been counted */
1473 }
1474
1475 /* add another AP to list, if less than max number of CPUs */
1476 else if (cpu < MAXCPU) {
1477 CPU_TO_ID(cpu) = entry->apic_id;
1478 ID_TO_CPU(entry->apic_id) = cpu;
1479 return 1;
1480 }
1481
1482 return 0;
1483}
1484
1485
1486static int
1487bus_entry(bus_entry_ptr entry, int bus)
1488{
1489 int x;
1490 char c, name[8];
1491
1492 /* encode the name into an index */
1493 for (x = 0; x < 6; ++x) {
1494 if ((c = entry->bus_type[x]) == ' ')
1495 break;
1496 name[x] = c;
1497 }
1498 name[x] = '\0';
1499
1500 if ((x = lookup_bus_type(name)) == UNKNOWN_BUSTYPE)
1501 panic("unknown bus type: '%s'", name);
1502
1503 bus_data[bus].bus_id = entry->bus_id;
1504 bus_data[bus].bus_type = x;
1505
1506 return 1;
1507}
1508
1509#ifdef APIC_IO
1510
1511static int
1512io_apic_entry(io_apic_entry_ptr entry, int apic)
1513{
1514 if (!(entry->apic_flags & IOAPICENTRY_FLAG_EN))
1515 return 0;
1516
1517 IO_TO_ID(apic) = entry->apic_id;
1518 if (entry->apic_id < NAPICID)
1519 ID_TO_IO(entry->apic_id) = apic;
1520
1521 return 1;
1522}
1523
1524#endif
1525
1526static int
1527lookup_bus_type(char *name)
1528{
1529 int x;
1530
1531 for (x = 0; x < MAX_BUSTYPE; ++x)
1532 if (strcmp(bus_type_table[x].name, name) == 0)
1533 return bus_type_table[x].type;
1534
1535 return UNKNOWN_BUSTYPE;
1536}
1537
1538#ifdef APIC_IO
1539
1540static int
1541int_entry(int_entry_ptr entry, int intr)
1542{
1543 int apic;
1544
1545 io_apic_ints[intr].int_type = entry->int_type;
1546 io_apic_ints[intr].int_flags = entry->int_flags;
1547 io_apic_ints[intr].src_bus_id = entry->src_bus_id;
1548 io_apic_ints[intr].src_bus_irq = entry->src_bus_irq;
1549 if (entry->dst_apic_id == 255) {
1550 /* This signal goes to all IO APICS. Select an IO APIC
1551 with sufficient number of interrupt pins */
1552 for (apic = 0; apic < mp_napics; apic++)
1553 if (((io_apic_read(apic, IOAPIC_VER) &
1554 IOART_VER_MAXREDIR) >> MAXREDIRSHIFT) >=
1555 entry->dst_apic_int)
1556 break;
1557 if (apic < mp_napics)
1558 io_apic_ints[intr].dst_apic_id = IO_TO_ID(apic);
1559 else
1560 io_apic_ints[intr].dst_apic_id = entry->dst_apic_id;
1561 } else
1562 io_apic_ints[intr].dst_apic_id = entry->dst_apic_id;
1563 io_apic_ints[intr].dst_apic_int = entry->dst_apic_int;
1564
1565 return 1;
1566}
1567
1568static int
1569apic_int_is_bus_type(int intr, int bus_type)
1570{
1571 int bus;
1572
1573 for (bus = 0; bus < mp_nbusses; ++bus)
1574 if ((bus_data[bus].bus_id == io_apic_ints[intr].src_bus_id)
1575 && ((int) bus_data[bus].bus_type == bus_type))
1576 return 1;
1577
1578 return 0;
1579}
1580
1581/*
1582 * Given a traditional ISA INT mask, return an APIC mask.
1583 */
1584u_int
1585isa_apic_mask(u_int isa_mask)
1586{
1587 int isa_irq;
1588 int apic_pin;
1589
1590#if defined(SKIP_IRQ15_REDIRECT)
1591 if (isa_mask == (1 << 15)) {
1592 kprintf("skipping ISA IRQ15 redirect\n");
1593 return isa_mask;
1594 }
1595#endif /* SKIP_IRQ15_REDIRECT */
1596
1597 isa_irq = ffs(isa_mask); /* find its bit position */
1598 if (isa_irq == 0) /* doesn't exist */
1599 return 0;
1600 --isa_irq; /* make it zero based */
1601
1602 apic_pin = isa_apic_irq(isa_irq); /* look for APIC connection */
1603 if (apic_pin == -1)
1604 return 0;
1605
1606 return (1 << apic_pin); /* convert pin# to a mask */
1607}
1608
1609/*
1610 * Determine which APIC pin an ISA/EISA INT is attached to.
1611 */
1612#define INTTYPE(I) (io_apic_ints[(I)].int_type)
1613#define INTPIN(I) (io_apic_ints[(I)].dst_apic_int)
1614#define INTIRQ(I) (io_apic_ints[(I)].int_vector)
1615#define INTAPIC(I) (ID_TO_IO(io_apic_ints[(I)].dst_apic_id))
1616
1617#define SRCBUSIRQ(I) (io_apic_ints[(I)].src_bus_irq)
1618int
1619isa_apic_irq(int isa_irq)
1620{
1621 int intr;
1622
1623 for (intr = 0; intr < nintrs; ++intr) { /* check each record */
1624 if (INTTYPE(intr) == 0) { /* standard INT */
1625 if (SRCBUSIRQ(intr) == isa_irq) {
1626 if (apic_int_is_bus_type(intr, ISA) ||
1627 apic_int_is_bus_type(intr, EISA)) {
1628 if (INTIRQ(intr) == 0xff)
1629 return -1; /* unassigned */
1630 return INTIRQ(intr); /* found */
1631 }
1632 }
1633 }
1634 }
1635 return -1; /* NOT found */
1636}
1637
1638
1639/*
1640 * Determine which APIC pin a PCI INT is attached to.
1641 */
1642#define SRCBUSID(I) (io_apic_ints[(I)].src_bus_id)
1643#define SRCBUSDEVICE(I) ((io_apic_ints[(I)].src_bus_irq >> 2) & 0x1f)
1644#define SRCBUSLINE(I) (io_apic_ints[(I)].src_bus_irq & 0x03)
1645int
1646pci_apic_irq(int pciBus, int pciDevice, int pciInt)
1647{
1648 int intr;
1649
1650 --pciInt; /* zero based */
1651
1652 for (intr = 0; intr < nintrs; ++intr) { /* check each record */
1653 if ((INTTYPE(intr) == 0) /* standard INT */
1654 && (SRCBUSID(intr) == pciBus)
1655 && (SRCBUSDEVICE(intr) == pciDevice)
1656 && (SRCBUSLINE(intr) == pciInt)) { /* a candidate IRQ */
1657 if (apic_int_is_bus_type(intr, PCI)) {
662f60ef
SZ
1658 if (INTIRQ(intr) == 0xff) {
1659 kprintf("IOAPIC: pci_apic_irq() "
1660 "failed\n");
46d4e165 1661 return -1; /* unassigned */
662f60ef 1662 }
46d4e165
JG
1663 return INTIRQ(intr); /* exact match */
1664 }
1665 }
1666 }
1667
1668 return -1; /* NOT found */
1669}
1670
1671int
1672next_apic_irq(int irq)
1673{
1674 int intr, ointr;
1675 int bus, bustype;
1676
1677 bus = 0;
1678 bustype = 0;
1679 for (intr = 0; intr < nintrs; intr++) {
1680 if (INTIRQ(intr) != irq || INTTYPE(intr) != 0)
1681 continue;
1682 bus = SRCBUSID(intr);
1683 bustype = apic_bus_type(bus);
1684 if (bustype != ISA &&
1685 bustype != EISA &&
1686 bustype != PCI)
1687 continue;
1688 break;
1689 }
1690 if (intr >= nintrs) {
1691 return -1;
1692 }
1693 for (ointr = intr + 1; ointr < nintrs; ointr++) {
1694 if (INTTYPE(ointr) != 0)
1695 continue;
1696 if (bus != SRCBUSID(ointr))
1697 continue;
1698 if (bustype == PCI) {
1699 if (SRCBUSDEVICE(intr) != SRCBUSDEVICE(ointr))
1700 continue;
1701 if (SRCBUSLINE(intr) != SRCBUSLINE(ointr))
1702 continue;
1703 }
1704 if (bustype == ISA || bustype == EISA) {
1705 if (SRCBUSIRQ(intr) != SRCBUSIRQ(ointr))
1706 continue;
1707 }
1708 if (INTPIN(intr) == INTPIN(ointr))
1709 continue;
1710 break;
1711 }
1712 if (ointr >= nintrs) {
1713 return -1;
1714 }
1715 return INTIRQ(ointr);
1716}
1717#undef SRCBUSLINE
1718#undef SRCBUSDEVICE
1719#undef SRCBUSID
1720#undef SRCBUSIRQ
1721
1722#undef INTPIN
1723#undef INTIRQ
1724#undef INTAPIC
1725#undef INTTYPE
1726
1727#endif
1728
1729/*
1730 * Reprogram the MB chipset to NOT redirect an ISA INTerrupt.
1731 *
1732 * XXX FIXME:
1733 * Exactly what this means is unclear at this point. It is a solution
1734 * for motherboards that redirect the MBIRQ0 pin. Generically a motherboard
1735 * could route any of the ISA INTs to upper (>15) IRQ values. But most would
1736 * NOT be redirected via MBIRQ0, thus "undirect()ing" them would NOT be an
1737 * option.
1738 */
1739int
1740undirect_isa_irq(int rirq)
1741{
1742#if defined(READY)
1743 if (bootverbose)
1744 kprintf("Freeing redirected ISA irq %d.\n", rirq);
1745 /** FIXME: tickle the MB redirector chip */
1746 return /* XXX */;
1747#else
1748 if (bootverbose)
1749 kprintf("Freeing (NOT implemented) redirected ISA irq %d.\n", rirq);
1750 return 0;
1751#endif /* READY */
1752}
1753
1754
1755/*
1756 * Reprogram the MB chipset to NOT redirect a PCI INTerrupt
1757 */
1758int
1759undirect_pci_irq(int rirq)
1760{
1761#if defined(READY)
1762 if (bootverbose)
1763 kprintf("Freeing redirected PCI irq %d.\n", rirq);
1764
1765 /** FIXME: tickle the MB redirector chip */
1766 return /* XXX */;
1767#else
1768 if (bootverbose)
1769 kprintf("Freeing (NOT implemented) redirected PCI irq %d.\n",
1770 rirq);
1771 return 0;
1772#endif /* READY */
1773}
1774
1775
1776/*
1777 * given a bus ID, return:
1778 * the bus type if found
1779 * -1 if NOT found
1780 */
1781int
1782apic_bus_type(int id)
1783{
1784 int x;
1785
1786 for (x = 0; x < mp_nbusses; ++x)
1787 if (bus_data[x].bus_id == id)
1788 return bus_data[x].bus_type;
1789
1790 return -1;
1791}
1792
1793#ifdef APIC_IO
1794
1795/*
1796 * given a LOGICAL APIC# and pin#, return:
1797 * the associated src bus ID if found
1798 * -1 if NOT found
1799 */
1800int
1801apic_src_bus_id(int apic, int pin)
1802{
1803 int x;
1804
1805 /* search each of the possible INTerrupt sources */
1806 for (x = 0; x < nintrs; ++x)
1807 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1808 (pin == io_apic_ints[x].dst_apic_int))
1809 return (io_apic_ints[x].src_bus_id);
1810
1811 return -1; /* NOT found */
1812}
1813
1814/*
1815 * given a LOGICAL APIC# and pin#, return:
1816 * the associated src bus IRQ if found
1817 * -1 if NOT found
1818 */
1819int
1820apic_src_bus_irq(int apic, int pin)
1821{
1822 int x;
1823
1824 for (x = 0; x < nintrs; x++)
1825 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1826 (pin == io_apic_ints[x].dst_apic_int))
1827 return (io_apic_ints[x].src_bus_irq);
1828
1829 return -1; /* NOT found */
1830}
1831
1832
1833/*
1834 * given a LOGICAL APIC# and pin#, return:
1835 * the associated INTerrupt type if found
1836 * -1 if NOT found
1837 */
1838int
1839apic_int_type(int apic, int pin)
1840{
1841 int x;
1842
1843 /* search each of the possible INTerrupt sources */
1844 for (x = 0; x < nintrs; ++x) {
1845 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1846 (pin == io_apic_ints[x].dst_apic_int))
1847 return (io_apic_ints[x].int_type);
1848 }
1849 return -1; /* NOT found */
1850}
1851
1852/*
1853 * Return the IRQ associated with an APIC pin
1854 */
1855int
1856apic_irq(int apic, int pin)
1857{
1858 int x;
1859 int res;
1860
1861 for (x = 0; x < nintrs; ++x) {
1862 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1863 (pin == io_apic_ints[x].dst_apic_int)) {
1864 res = io_apic_ints[x].int_vector;
1865 if (res == 0xff)
1866 return -1;
1867 if (apic != int_to_apicintpin[res].ioapic)
1868 panic("apic_irq: inconsistent table %d/%d", apic, int_to_apicintpin[res].ioapic);
1869 if (pin != int_to_apicintpin[res].int_pin)
1870 panic("apic_irq inconsistent table (2)");
1871 return res;
1872 }
1873 }
1874 return -1;
1875}
1876
1877
1878/*
1879 * given a LOGICAL APIC# and pin#, return:
1880 * the associated trigger mode if found
1881 * -1 if NOT found
1882 */
1883int
1884apic_trigger(int apic, int pin)
1885{
1886 int x;
1887
1888 /* search each of the possible INTerrupt sources */
1889 for (x = 0; x < nintrs; ++x)
1890 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1891 (pin == io_apic_ints[x].dst_apic_int))
1892 return ((io_apic_ints[x].int_flags >> 2) & 0x03);
1893
1894 return -1; /* NOT found */
1895}
1896
1897
1898/*
1899 * given a LOGICAL APIC# and pin#, return:
1900 * the associated 'active' level if found
1901 * -1 if NOT found
1902 */
1903int
1904apic_polarity(int apic, int pin)
1905{
1906 int x;
1907
1908 /* search each of the possible INTerrupt sources */
1909 for (x = 0; x < nintrs; ++x)
1910 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1911 (pin == io_apic_ints[x].dst_apic_int))
1912 return (io_apic_ints[x].int_flags & 0x03);
1913
1914 return -1; /* NOT found */
1915}
1916
1917#endif
1918
1919/*
1920 * set data according to MP defaults
1921 * FIXME: probably not complete yet...
1922 */
1923static void
1924default_mp_table(int type)
1925{
1926 int ap_cpu_id;
1927#if defined(APIC_IO)
1928 int io_apic_id;
1929 int pin;
1930#endif /* APIC_IO */
1931
1932#if 0
1933 kprintf(" MP default config type: %d\n", type);
1934 switch (type) {
1935 case 1:
1936 kprintf(" bus: ISA, APIC: 82489DX\n");
1937 break;
1938 case 2:
1939 kprintf(" bus: EISA, APIC: 82489DX\n");
1940 break;
1941 case 3:
1942 kprintf(" bus: EISA, APIC: 82489DX\n");
1943 break;
1944 case 4:
1945 kprintf(" bus: MCA, APIC: 82489DX\n");
1946 break;
1947 case 5:
1948 kprintf(" bus: ISA+PCI, APIC: Integrated\n");
1949 break;
1950 case 6:
1951 kprintf(" bus: EISA+PCI, APIC: Integrated\n");
1952 break;
1953 case 7:
1954 kprintf(" bus: MCA+PCI, APIC: Integrated\n");
1955 break;
1956 default:
1957 kprintf(" future type\n");
1958 break;
1959 /* NOTREACHED */
1960 }
1961#endif /* 0 */
1962
1963 boot_cpu_id = (lapic->id & APIC_ID_MASK) >> 24;
1964 ap_cpu_id = (boot_cpu_id == 0) ? 1 : 0;
1965
1966 /* BSP */
1967 CPU_TO_ID(0) = boot_cpu_id;
1968 ID_TO_CPU(boot_cpu_id) = 0;
1969
1970 /* one and only AP */
1971 CPU_TO_ID(1) = ap_cpu_id;
1972 ID_TO_CPU(ap_cpu_id) = 1;
1973
1974#if defined(APIC_IO)
1975 /* one and only IO APIC */
1976 io_apic_id = (io_apic_read(0, IOAPIC_ID) & APIC_ID_MASK) >> 24;
1977
1978 /*
1979 * sanity check, refer to MP spec section 3.6.6, last paragraph
1980 * necessary as some hardware isn't properly setting up the IO APIC
1981 */
1982#if defined(REALLY_ANAL_IOAPICID_VALUE)
1983 if (io_apic_id != 2) {
1984#else
1985 if ((io_apic_id == 0) || (io_apic_id == 1) || (io_apic_id == 15)) {
1986#endif /* REALLY_ANAL_IOAPICID_VALUE */
1987 io_apic_set_id(0, 2);
1988 io_apic_id = 2;
1989 }
1990 IO_TO_ID(0) = io_apic_id;
1991 ID_TO_IO(io_apic_id) = 0;
1992#endif /* APIC_IO */
1993
1994 /* fill out bus entries */
1995 switch (type) {
1996 case 1:
1997 case 2:
1998 case 3:
1999 case 4:
2000 case 5:
2001 case 6:
2002 case 7:
2003 bus_data[0].bus_id = default_data[type - 1][1];
2004 bus_data[0].bus_type = default_data[type - 1][2];
2005 bus_data[1].bus_id = default_data[type - 1][3];
2006 bus_data[1].bus_type = default_data[type - 1][4];
2007 break;
2008
2009 /* case 4: case 7: MCA NOT supported */
2010 default: /* illegal/reserved */
2011 panic("BAD default MP config: %d", type);
2012 /* NOTREACHED */
2013 }
2014
2015#if defined(APIC_IO)
2016 /* general cases from MP v1.4, table 5-2 */
2017 for (pin = 0; pin < 16; ++pin) {
2018 io_apic_ints[pin].int_type = 0;
2019 io_apic_ints[pin].int_flags = 0x05; /* edge/active-hi */
2020 io_apic_ints[pin].src_bus_id = 0;
2021 io_apic_ints[pin].src_bus_irq = pin; /* IRQ2 caught below */
2022 io_apic_ints[pin].dst_apic_id = io_apic_id;
2023 io_apic_ints[pin].dst_apic_int = pin; /* 1-to-1 */
2024 }
2025
2026 /* special cases from MP v1.4, table 5-2 */
2027 if (type == 2) {
2028 io_apic_ints[2].int_type = 0xff; /* N/C */
2029 io_apic_ints[13].int_type = 0xff; /* N/C */
2030#if !defined(APIC_MIXED_MODE)
2031 /** FIXME: ??? */
2032 panic("sorry, can't support type 2 default yet");
2033#endif /* APIC_MIXED_MODE */
2034 }
2035 else
2036 io_apic_ints[2].src_bus_irq = 0; /* ISA IRQ0 is on APIC INT 2 */
2037
2038 if (type == 7)
2039 io_apic_ints[0].int_type = 0xff; /* N/C */
2040 else
2041 io_apic_ints[0].int_type = 3; /* vectored 8259 */
2042#endif /* APIC_IO */
2043}
2044
2045/*
2046 * Map a physical memory address representing I/O into KVA. The I/O
2047 * block is assumed not to cross a page boundary.
2048 */
2049void *
2050permanent_io_mapping(vm_paddr_t pa)
2051{
46d4e165
JG
2052 KKASSERT(pa < 0x100000000LL);
2053
403c36ea 2054 return pmap_mapdev_uncacheable(pa, PAGE_SIZE);
46d4e165
JG
2055}
2056
2057/*
2058 * start each AP in our list
2059 */
2060static int
2061start_all_aps(u_int boot_addr)
2062{
2063 vm_offset_t va = boot_address + KERNBASE;
2064 u_int64_t *pt4, *pt3, *pt2;
2065 int x, i, pg;
2066 int shift;
bb467734
MD
2067 int smicount;
2068 int smibest;
2069 int smilast;
46d4e165
JG
2070 u_char mpbiosreason;
2071 u_long mpbioswarmvec;
2072 struct mdglobaldata *gd;
2073 struct privatespace *ps;
46d4e165
JG
2074
2075 POSTCODE(START_ALL_APS_POST);
2076
2077 /* Initialize BSP's local APIC */
2078 apic_initialize(TRUE);
2079 bsp_apic_ready = 1;
2080
2081 /* install the AP 1st level boot code */
2082 pmap_kenter(va, boot_address);
bfc09ba0 2083 cpu_invlpg((void *)va); /* JG XXX */
46d4e165
JG
2084 bcopy(mptramp_start, (void *)va, bootMP_size);
2085
2086 /* Locate the page tables, they'll be below the trampoline */
2087 pt4 = (u_int64_t *)(uintptr_t)(mptramp_pagetables + KERNBASE);
2088 pt3 = pt4 + (PAGE_SIZE) / sizeof(u_int64_t);
2089 pt2 = pt3 + (PAGE_SIZE) / sizeof(u_int64_t);
2090
2091 /* Create the initial 1GB replicated page tables */
2092 for (i = 0; i < 512; i++) {
2093 /* Each slot of the level 4 pages points to the same level 3 page */
2094 pt4[i] = (u_int64_t)(uintptr_t)(mptramp_pagetables + PAGE_SIZE);
2095 pt4[i] |= PG_V | PG_RW | PG_U;
2096
2097 /* Each slot of the level 3 pages points to the same level 2 page */
2098 pt3[i] = (u_int64_t)(uintptr_t)(mptramp_pagetables + (2 * PAGE_SIZE));
2099 pt3[i] |= PG_V | PG_RW | PG_U;
2100
2101 /* The level 2 page slots are mapped with 2MB pages for 1GB. */
2102 pt2[i] = i * (2 * 1024 * 1024);
2103 pt2[i] |= PG_V | PG_RW | PG_PS | PG_U;
2104 }
2105
2106 /* save the current value of the warm-start vector */
2107 mpbioswarmvec = *((u_int32_t *) WARMBOOT_OFF);
2108 outb(CMOS_REG, BIOS_RESET);
2109 mpbiosreason = inb(CMOS_DATA);
2110
2111 /* setup a vector to our boot code */
2112 *((volatile u_short *) WARMBOOT_OFF) = WARMBOOT_TARGET;
2113 *((volatile u_short *) WARMBOOT_SEG) = (boot_address >> 4);
2114 outb(CMOS_REG, BIOS_RESET);
2115 outb(CMOS_DATA, BIOS_WARM); /* 'warm-start' */
2116
bb467734
MD
2117 /*
2118 * If we have a TSC we can figure out the SMI interrupt rate.
2119 * The SMI does not necessarily use a constant rate. Spend
2120 * up to 250ms trying to figure it out.
2121 */
2122 smibest = 0;
2123 if (cpu_feature & CPUID_TSC) {
2124 set_apic_timer(275000);
2125 smilast = read_apic_timer();
2126 for (x = 0; x < 20 && read_apic_timer(); ++x) {
2127 smicount = smitest();
2128 if (smibest == 0 || smilast - smicount < smibest)
2129 smibest = smilast - smicount;
2130 smilast = smicount;
2131 }
2132 if (smibest > 250000)
2133 smibest = 0;
2134 if (smibest) {
2135 smibest = smibest * (int64_t)1000000 /
2136 get_apic_timer_frequency();
2137 }
2138 }
2139 if (smibest)
2140 kprintf("SMI Frequency (worst case): %d Hz (%d us)\n",
2141 1000000 / smibest, smibest);
2142
46d4e165
JG
2143 /* start each AP */
2144 for (x = 1; x <= mp_naps; ++x) {
2145
2146 /* This is a bit verbose, it will go away soon. */
2147
2148 /* first page of AP's private space */
b2b3ffcd 2149 pg = x * x86_64_btop(sizeof(struct privatespace));
46d4e165
JG
2150
2151 /* allocate new private data page(s) */
2152 gd = (struct mdglobaldata *)kmem_alloc(&kernel_map,
2153 MDGLOBALDATA_BASEALLOC_SIZE);
46d4e165
JG
2154
2155 gd = &CPU_prvspace[x].mdglobaldata; /* official location */
2156 bzero(gd, sizeof(*gd));
2157 gd->mi.gd_prvspace = ps = &CPU_prvspace[x];
2158
2159 /* prime data page for it to use */
2160 mi_gdinit(&gd->mi, x);
2161 cpu_gdinit(gd, x);
2162 gd->gd_CMAP1 = &SMPpt[pg + 0];
2163 gd->gd_CMAP2 = &SMPpt[pg + 1];
2164 gd->gd_CMAP3 = &SMPpt[pg + 2];
2165 gd->gd_PMAP1 = &SMPpt[pg + 3];
2166 gd->gd_CADDR1 = ps->CPAGE1;
2167 gd->gd_CADDR2 = ps->CPAGE2;
2168 gd->gd_CADDR3 = ps->CPAGE3;
bfc09ba0 2169 gd->gd_PADDR1 = (pt_entry_t *)ps->PPAGE1;
46d4e165
JG
2170 gd->mi.gd_ipiq = (void *)kmem_alloc(&kernel_map, sizeof(lwkt_ipiq) * (mp_naps + 1));
2171 bzero(gd->mi.gd_ipiq, sizeof(lwkt_ipiq) * (mp_naps + 1));
2172
2173 /* setup a vector to our boot code */
2174 *((volatile u_short *) WARMBOOT_OFF) = WARMBOOT_TARGET;
2175 *((volatile u_short *) WARMBOOT_SEG) = (boot_addr >> 4);
2176 outb(CMOS_REG, BIOS_RESET);
2177 outb(CMOS_DATA, BIOS_WARM); /* 'warm-start' */
2178
2179 /*
2180 * Setup the AP boot stack
2181 */
2182 bootSTK = &ps->idlestack[UPAGES*PAGE_SIZE/2];
2183 bootAP = x;
2184
2185 /* attempt to start the Application Processor */
2186 CHECK_INIT(99); /* setup checkpoints */
bb467734 2187 if (!start_ap(gd, boot_addr, smibest)) {
46d4e165
JG
2188 kprintf("AP #%d (PHY# %d) failed!\n", x, CPU_TO_ID(x));
2189 CHECK_PRINT("trace"); /* show checkpoints */
2190 /* better panic as the AP may be running loose */
2191 kprintf("panic y/n? [y] ");
2192 if (cngetc() != 'n')
2193 panic("bye-bye");
2194 }
2195 CHECK_PRINT("trace"); /* show checkpoints */
2196
2197 /* record its version info */
2198 cpu_apic_versions[x] = cpu_apic_versions[0];
2199 }
2200
2201 /* set ncpus to 1 + highest logical cpu. Not all may have come up */
2202 ncpus = x;
2203
2204 /* ncpus2 -- ncpus rounded down to the nearest power of 2 */
2205 for (shift = 0; (1 << shift) <= ncpus; ++shift)
2206 ;
2207 --shift;
2208 ncpus2_shift = shift;
2209 ncpus2 = 1 << shift;
2210 ncpus2_mask = ncpus2 - 1;
2211
2212 /* ncpus_fit -- ncpus rounded up to the nearest power of 2 */
2213 if ((1 << shift) < ncpus)
2214 ++shift;
2215 ncpus_fit = 1 << shift;
2216 ncpus_fit_mask = ncpus_fit - 1;
2217
2218 /* build our map of 'other' CPUs */
2219 mycpu->gd_other_cpus = smp_startup_mask & ~(1 << mycpu->gd_cpuid);
2220 mycpu->gd_ipiq = (void *)kmem_alloc(&kernel_map, sizeof(lwkt_ipiq) * ncpus);
2221 bzero(mycpu->gd_ipiq, sizeof(lwkt_ipiq) * ncpus);
2222
2223 /* fill in our (BSP) APIC version */
2224 cpu_apic_versions[0] = lapic->version;
2225
2226 /* restore the warmstart vector */
2227 *(u_long *) WARMBOOT_OFF = mpbioswarmvec;
2228 outb(CMOS_REG, BIOS_RESET);
2229 outb(CMOS_DATA, mpbiosreason);
2230
2231 /*
2232 * NOTE! The idlestack for the BSP was setup by locore. Finish
2233 * up, clean out the P==V mapping we did earlier.
2234 */
2235#if JGXXX
2236 for (x = 0; x < NKPT; x++)
2237 PTD[x] = 0;
2238#endif
2239 pmap_set_opt();
2240
2241 /* number of APs actually started */
2242 return ncpus - 1;
2243}
2244
2245
2246/*
2247 * load the 1st level AP boot code into base memory.
2248 */
2249
2250/* targets for relocation */
2251extern void bigJump(void);
2252extern void bootCodeSeg(void);
2253extern void bootDataSeg(void);
2254extern void MPentry(void);
2255extern u_int MP_GDT;
2256extern u_int mp_gdtbase;
2257
bfc09ba0
MD
2258#if 0
2259
46d4e165
JG
2260static void
2261install_ap_tramp(u_int boot_addr)
2262{
2263 int x;
2264 int size = *(int *) ((u_long) & bootMP_size);
2265 u_char *src = (u_char *) ((u_long) bootMP);
2266 u_char *dst = (u_char *) boot_addr + KERNBASE;
2267 u_int boot_base = (u_int) bootMP;
2268 u_int8_t *dst8;
2269 u_int16_t *dst16;
2270 u_int32_t *dst32;
2271
2272 POSTCODE(INSTALL_AP_TRAMP_POST);
2273
2274 for (x = 0; x < size; ++x)
2275 *dst++ = *src++;
2276
2277 /*
2278 * modify addresses in code we just moved to basemem. unfortunately we
2279 * need fairly detailed info about mpboot.s for this to work. changes
2280 * to mpboot.s might require changes here.
2281 */
2282
2283 /* boot code is located in KERNEL space */
2284 dst = (u_char *) boot_addr + KERNBASE;
2285
2286 /* modify the lgdt arg */
2287 dst32 = (u_int32_t *) (dst + ((u_int) & mp_gdtbase - boot_base));
2288 *dst32 = boot_addr + ((u_int) & MP_GDT - boot_base);
2289
2290 /* modify the ljmp target for MPentry() */
2291 dst32 = (u_int32_t *) (dst + ((u_int) bigJump - boot_base) + 1);
2292 *dst32 = ((u_int) MPentry - KERNBASE);
2293
2294 /* modify the target for boot code segment */
2295 dst16 = (u_int16_t *) (dst + ((u_int) bootCodeSeg - boot_base));
2296 dst8 = (u_int8_t *) (dst16 + 1);
2297 *dst16 = (u_int) boot_addr & 0xffff;
2298 *dst8 = ((u_int) boot_addr >> 16) & 0xff;
2299
2300 /* modify the target for boot data segment */
2301 dst16 = (u_int16_t *) (dst + ((u_int) bootDataSeg - boot_base));
2302 dst8 = (u_int8_t *) (dst16 + 1);
2303 *dst16 = (u_int) boot_addr & 0xffff;
2304 *dst8 = ((u_int) boot_addr >> 16) & 0xff;
2305}
2306
bfc09ba0 2307#endif
46d4e165
JG
2308
2309/*
bb467734 2310 * This function starts the AP (application processor) identified
46d4e165
JG
2311 * by the APIC ID 'physicalCpu'. It does quite a "song and dance"
2312 * to accomplish this. This is necessary because of the nuances
2313 * of the different hardware we might encounter. It ain't pretty,
2314 * but it seems to work.
2315 *
2316 * NOTE: eventually an AP gets to ap_init(), which is called just
2317 * before the AP goes into the LWKT scheduler's idle loop.
2318 */
2319static int
bb467734 2320start_ap(struct mdglobaldata *gd, u_int boot_addr, int smibest)
46d4e165
JG
2321{
2322 int physical_cpu;
2323 int vector;
2324 u_long icr_lo, icr_hi;
2325
2326 POSTCODE(START_AP_POST);
2327
2328 /* get the PHYSICAL APIC ID# */
2329 physical_cpu = CPU_TO_ID(gd->mi.gd_cpuid);
2330
2331 /* calculate the vector */
2332 vector = (boot_addr >> 12) & 0xff;
2333
bb467734
MD
2334 /* We don't want anything interfering */
2335 cpu_disable_intr();
2336
46d4e165
JG
2337 /* Make sure the target cpu sees everything */
2338 wbinvd();
2339
2340 /*
bb467734
MD
2341 * Try to detect when a SMI has occurred, wait up to 200ms.
2342 *
2343 * If a SMI occurs during an AP reset but before we issue
2344 * the STARTUP command, the AP may brick. To work around
2345 * this problem we hold off doing the AP startup until
2346 * after we have detected the SMI. Hopefully another SMI
2347 * will not occur before we finish the AP startup.
2348 *
2349 * Retries don't seem to help. SMIs have a window of opportunity
2350 * and if USB->legacy keyboard emulation is enabled in the BIOS
2351 * the interrupt rate can be quite high.
2352 *
2353 * NOTE: Don't worry about the L1 cache load, it might bloat
2354 * ldelta a little but ndelta will be so huge when the SMI
2355 * occurs the detection logic will still work fine.
2356 */
2357 if (smibest) {
2358 set_apic_timer(200000);
2359 smitest();
2360 }
2361
2362 /*
46d4e165
JG
2363 * first we do an INIT/RESET IPI this INIT IPI might be run, reseting
2364 * and running the target CPU. OR this INIT IPI might be latched (P5
2365 * bug), CPU waiting for STARTUP IPI. OR this INIT IPI might be
2366 * ignored.
bb467734
MD
2367 *
2368 * see apic/apicreg.h for icr bit definitions.
2369 *
2370 * TIME CRITICAL CODE, DO NOT DO ANY KPRINTFS IN THE HOT PATH.
46d4e165
JG
2371 */
2372
bb467734
MD
2373 /*
2374 * Setup the address for the target AP. We can setup
2375 * icr_hi once and then just trigger operations with
2376 * icr_lo.
2377 */
46d4e165
JG
2378 icr_hi = lapic->icr_hi & ~APIC_ID_MASK;
2379 icr_hi |= (physical_cpu << 24);
46d4e165 2380 icr_lo = lapic->icr_lo & 0xfff00000;
bb467734 2381 lapic->icr_hi = icr_hi;
46d4e165 2382
bb467734
MD
2383 /*
2384 * Do an INIT IPI: assert RESET
2385 *
2386 * Use edge triggered mode to assert INIT
2387 */
2388 lapic->icr_lo = icr_lo | 0x00004500;
46d4e165
JG
2389 while (lapic->icr_lo & APIC_DELSTAT_MASK)
2390 /* spin */ ;
2391
bb467734
MD
2392 /*
2393 * The spec calls for a 10ms delay but we may have to use a
2394 * MUCH lower delay to avoid bricking an AP due to a fast SMI
2395 * interrupt. We have other loops here too and dividing by 2
2396 * doesn't seem to be enough even after subtracting 350us,
2397 * so we divide by 4.
2398 *
2399 * Our minimum delay is 150uS, maximum is 10ms. If no SMI
2400 * interrupt was detected we use the full 10ms.
2401 */
2402 if (smibest == 0)
2403 u_sleep(10000);
2404 else if (smibest < 150 * 4 + 350)
2405 u_sleep(150);
2406 else if ((smibest - 350) / 4 < 10000)
2407 u_sleep((smibest - 350) / 4);
2408 else
2409 u_sleep(10000);
46d4e165 2410
bb467734
MD
2411 /*
2412 * Do an INIT IPI: deassert RESET
2413 *
2414 * Use level triggered mode to deassert. It is unclear
2415 * why we need to do this.
2416 */
2417 lapic->icr_lo = icr_lo | 0x00008500;
46d4e165
JG
2418 while (lapic->icr_lo & APIC_DELSTAT_MASK)
2419 /* spin */ ;
bb467734 2420 u_sleep(150); /* wait 150us */
46d4e165
JG
2421
2422 /*
bb467734 2423 * Next we do a STARTUP IPI: the previous INIT IPI might still be
46d4e165
JG
2424 * latched, (P5 bug) this 1st STARTUP would then terminate
2425 * immediately, and the previously started INIT IPI would continue. OR
2426 * the previous INIT IPI has already run. and this STARTUP IPI will
2427 * run. OR the previous INIT IPI was ignored. and this STARTUP IPI
2428 * will run.
2429 */
46d4e165
JG
2430 lapic->icr_lo = icr_lo | 0x00000600 | vector;
2431 while (lapic->icr_lo & APIC_DELSTAT_MASK)
2432 /* spin */ ;
2433 u_sleep(200); /* wait ~200uS */
2434
2435 /*
bb467734 2436 * Finally we do a 2nd STARTUP IPI: this 2nd STARTUP IPI should run IF
46d4e165
JG
2437 * the previous STARTUP IPI was cancelled by a latched INIT IPI. OR
2438 * this STARTUP IPI will be ignored, as only ONE STARTUP IPI is
2439 * recognized after hardware RESET or INIT IPI.
2440 */
46d4e165
JG
2441 lapic->icr_lo = icr_lo | 0x00000600 | vector;
2442 while (lapic->icr_lo & APIC_DELSTAT_MASK)
2443 /* spin */ ;
bb467734
MD
2444
2445 /* Resume normal operation */
2446 cpu_enable_intr();
46d4e165
JG
2447
2448 /* wait for it to start, see ap_init() */
2449 set_apic_timer(5000000);/* == 5 seconds */
2450 while (read_apic_timer()) {
2451 if (smp_startup_mask & (1 << gd->mi.gd_cpuid))
2452 return 1; /* return SUCCESS */
2453 }
bb467734 2454
46d4e165
JG
2455 return 0; /* return FAILURE */
2456}
2457
bb467734
MD
2458static
2459int
2460smitest(void)
2461{
2462 int64_t ltsc;
2463 int64_t ntsc;
2464 int64_t ldelta;
2465 int64_t ndelta;
2466 int count;
2467
2468 ldelta = 0;
2469 ndelta = 0;
2470 while (read_apic_timer()) {
2471 ltsc = rdtsc();
2472 for (count = 0; count < 100; ++count)
2473 ntsc = rdtsc(); /* force loop to occur */
2474 if (ldelta) {
2475 ndelta = ntsc - ltsc;
2476 if (ldelta > ndelta)
2477 ldelta = ndelta;
2478 if (ndelta > ldelta * 2)
2479 break;
2480 } else {
2481 ldelta = ntsc - ltsc;
2482 }
2483 }
2484 return(read_apic_timer());
2485}
46d4e165
JG
2486
2487/*
2488 * Lazy flush the TLB on all other CPU's. DEPRECATED.
2489 *
2490 * If for some reason we were unable to start all cpus we cannot safely
2491 * use broadcast IPIs.
2492 */
2493void
2494smp_invltlb(void)
2495{
2496#ifdef SMP
2497 if (smp_startup_mask == smp_active_mask) {
2498 all_but_self_ipi(XINVLTLB_OFFSET);
2499 } else {
2500 selected_apic_ipi(smp_active_mask, XINVLTLB_OFFSET,
2501 APIC_DELMODE_FIXED);
2502 }
2503#endif
2504}
2505
2506/*
2507 * When called the executing CPU will send an IPI to all other CPUs
2508 * requesting that they halt execution.
2509 *
2510 * Usually (but not necessarily) called with 'other_cpus' as its arg.
2511 *
2512 * - Signals all CPUs in map to stop.
2513 * - Waits for each to stop.
2514 *
2515 * Returns:
2516 * -1: error
2517 * 0: NA
2518 * 1: ok
2519 *
2520 * XXX FIXME: this is not MP-safe, needs a lock to prevent multiple CPUs
2521 * from executing at same time.
2522 */
2523int
2524stop_cpus(u_int map)
2525{
2526 map &= smp_active_mask;
2527
2528 /* send the Xcpustop IPI to all CPUs in map */
2529 selected_apic_ipi(map, XCPUSTOP_OFFSET, APIC_DELMODE_FIXED);
2530
2531 while ((stopped_cpus & map) != map)
2532 /* spin */ ;
2533
2534 return 1;
2535}
2536
2537
2538/*
2539 * Called by a CPU to restart stopped CPUs.
2540 *
2541 * Usually (but not necessarily) called with 'stopped_cpus' as its arg.
2542 *
2543 * - Signals all CPUs in map to restart.
2544 * - Waits for each to restart.
2545 *
2546 * Returns:
2547 * -1: error
2548 * 0: NA
2549 * 1: ok
2550 */
2551int
2552restart_cpus(u_int map)
2553{
2554 /* signal other cpus to restart */
2555 started_cpus = map & smp_active_mask;
2556
2557 while ((stopped_cpus & map) != 0) /* wait for each to clear its bit */
2558 /* spin */ ;
2559
2560 return 1;
2561}
2562
2563/*
2564 * This is called once the mpboot code has gotten us properly relocated
2565 * and the MMU turned on, etc. ap_init() is actually the idle thread,
2566 * and when it returns the scheduler will call the real cpu_idle() main
2567 * loop for the idlethread. Interrupts are disabled on entry and should
2568 * remain disabled at return.
2569 */
2570void
2571ap_init(void)
2572{
2573 u_int apic_id;
2574
2575 /*
2576 * Adjust smp_startup_mask to signal the BSP that we have started
2577 * up successfully. Note that we do not yet hold the BGL. The BSP
2578 * is waiting for our signal.
2579 *
2580 * We can't set our bit in smp_active_mask yet because we are holding
2581 * interrupts physically disabled and remote cpus could deadlock
2582 * trying to send us an IPI.
2583 */
2584 smp_startup_mask |= 1 << mycpu->gd_cpuid;
2585 cpu_mfence();
2586
2587 /*
2588 * Interlock for finalization. Wait until mp_finish is non-zero,
2589 * then get the MP lock.
2590 *
2591 * Note: We are in a critical section.
2592 *
2593 * Note: We have to synchronize td_mpcount to our desired MP state
2594 * before calling cpu_try_mplock().
2595 *
2596 * Note: we are the idle thread, we can only spin.
2597 *
2598 * Note: The load fence is memory volatile and prevents the compiler
2599 * from improperly caching mp_finish, and the cpu from improperly
2600 * caching it.
2601 */
2602 while (mp_finish == 0)
2603 cpu_lfence();
2604 ++curthread->td_mpcount;
2605 while (cpu_try_mplock() == 0)
2606 ;
2607
2608 if (cpu_feature & CPUID_TSC) {
2609 /*
2610 * The BSP is constantly updating tsc0_offset, figure out the
2611 * relative difference to synchronize ktrdump.
2612 */
2613 tsc_offsets[mycpu->gd_cpuid] = rdtsc() - tsc0_offset;
2614 }
2615
2616 /* BSP may have changed PTD while we're waiting for the lock */
2617 cpu_invltlb();
2618
2619#if defined(I586_CPU) && !defined(NO_F00F_HACK)
2620 lidt(&r_idt);
2621#endif
2622
2623 /* Build our map of 'other' CPUs. */
2624 mycpu->gd_other_cpus = smp_startup_mask & ~(1 << mycpu->gd_cpuid);
2625
2626 kprintf("SMP: AP CPU #%d Launched!\n", mycpu->gd_cpuid);
2627
2628 /* A quick check from sanity claus */
2629 apic_id = (apic_id_to_logical[(lapic->id & 0x0f000000) >> 24]);
2630 if (mycpu->gd_cpuid != apic_id) {
2631 kprintf("SMP: cpuid = %d\n", mycpu->gd_cpuid);
2632 kprintf("SMP: apic_id = %d\n", apic_id);
2633#if JGXXX
2634 kprintf("PTD[MPPTDI] = %p\n", (void *)PTD[MPPTDI]);
2635#endif
2636 panic("cpuid mismatch! boom!!");
2637 }
2638
2639 /* Initialize AP's local APIC for irq's */
2640 apic_initialize(FALSE);
2641
2642 /* Set memory range attributes for this CPU to match the BSP */
2643 mem_range_AP_init();
2644
2645 /*
2646 * Once we go active we must process any IPIQ messages that may
2647 * have been queued, because no actual IPI will occur until we
2648 * set our bit in the smp_active_mask. If we don't the IPI
2649 * message interlock could be left set which would also prevent
2650 * further IPIs.
2651 *
2652 * The idle loop doesn't expect the BGL to be held and while
2653 * lwkt_switch() normally cleans things up this is a special case
2654 * because we returning almost directly into the idle loop.
2655 *
2656 * The idle thread is never placed on the runq, make sure
2657 * nothing we've done put it there.
2658 */
2659 KKASSERT(curthread->td_mpcount == 1);
2660 smp_active_mask |= 1 << mycpu->gd_cpuid;
2661
2662 /*
2663 * Enable interrupts here. idle_restore will also do it, but
2664 * doing it here lets us clean up any strays that got posted to
2665 * the CPU during the AP boot while we are still in a critical
2666 * section.
2667 */
2668 __asm __volatile("sti; pause; pause"::);
2669 mdcpu->gd_fpending = 0;
46d4e165
JG
2670
2671 initclocks_pcpu(); /* clock interrupts (via IPIs) */
2672 lwkt_process_ipiq();
2673
2674 /*
2675 * Releasing the mp lock lets the BSP finish up the SMP init
2676 */
2677 rel_mplock();
2678 KKASSERT((curthread->td_flags & TDF_RUNQ) == 0);
2679}
2680
2681/*
2682 * Get SMP fully working before we start initializing devices.
2683 */
2684static
2685void
2686ap_finish(void)
2687{
2688 mp_finish = 1;
2689 if (bootverbose)
2690 kprintf("Finish MP startup\n");
2691 if (cpu_feature & CPUID_TSC)
2692 tsc0_offset = rdtsc();
2693 tsc_offsets[0] = 0;
2694 rel_mplock();
2695 while (smp_active_mask != smp_startup_mask) {
2696 cpu_lfence();
2697 if (cpu_feature & CPUID_TSC)
2698 tsc0_offset = rdtsc();
2699 }
2700 while (try_mplock() == 0)
2701 ;
2702 if (bootverbose)
2703 kprintf("Active CPU Mask: %08x\n", smp_active_mask);
2704}
2705
2706SYSINIT(finishsmp, SI_BOOT2_FINISH_SMP, SI_ORDER_FIRST, ap_finish, NULL)
2707
2708void
2709cpu_send_ipiq(int dcpu)
2710{
2711 if ((1 << dcpu) & smp_active_mask)
2712 single_apic_ipi(dcpu, XIPIQ_OFFSET, APIC_DELMODE_FIXED);
2713}
2714
2715#if 0 /* single_apic_ipi_passive() not working yet */
2716/*
2717 * Returns 0 on failure, 1 on success
2718 */
2719int
2720cpu_send_ipiq_passive(int dcpu)
2721{
2722 int r = 0;
2723 if ((1 << dcpu) & smp_active_mask) {
2724 r = single_apic_ipi_passive(dcpu, XIPIQ_OFFSET,
2725 APIC_DELMODE_FIXED);
2726 }
2727 return(r);
2728}
2729#endif
2730