2 * Copyright (c) 1996, by Steve Passe
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
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 $
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>
41 #include <vm/vm_param.h>
43 #include <vm/vm_kern.h>
44 #include <vm/vm_extern.h>
46 #include <vm/vm_map.h>
52 #include <machine/smp.h>
53 #include <machine_base/apic/apicreg.h>
54 #include <machine/atomic.h>
55 #include <machine/cpufunc.h>
56 #include <machine_base/apic/mpapic.h>
57 #include <machine/psl.h>
58 #include <machine/segments.h>
59 #include <machine/tss.h>
60 #include <machine/specialreg.h>
61 #include <machine/globaldata.h>
63 #include <machine/md_var.h> /* setidt() */
64 #include <machine_base/icu/icu.h> /* IPIs */
65 #include <machine_base/isa/intr_machdep.h> /* IPIs */
67 #define FIXUP_EXTRA_APIC_INTS 8 /* additional entries we may create */
69 #define WARMBOOT_TARGET 0
70 #define WARMBOOT_OFF (KERNBASE + 0x0467)
71 #define WARMBOOT_SEG (KERNBASE + 0x0469)
73 #define BIOS_BASE (0xf0000)
74 #define BIOS_SIZE (0x10000)
75 #define BIOS_COUNT (BIOS_SIZE/4)
77 #define CMOS_REG (0x70)
78 #define CMOS_DATA (0x71)
79 #define BIOS_RESET (0x0f)
80 #define BIOS_WARM (0x0a)
82 #define PROCENTRY_FLAG_EN 0x01
83 #define PROCENTRY_FLAG_BP 0x02
84 #define IOAPICENTRY_FLAG_EN 0x01
87 /* MP Floating Pointer Structure */
88 typedef struct MPFPS
{
101 /* MP Configuration Table Header */
102 typedef struct MPCTH
{
104 u_short base_table_length
;
108 u_char product_id
[12];
109 u_int32_t oem_table_pointer
;
110 u_short oem_table_size
;
112 u_int32_t apic_address
;
113 u_short extended_table_length
;
114 u_char extended_table_checksum
;
119 typedef struct PROCENTRY
{
124 u_int32_t cpu_signature
;
125 u_int32_t feature_flags
;
130 typedef struct BUSENTRY
{
136 typedef struct IOAPICENTRY
{
141 u_int32_t apic_address
;
142 } *io_apic_entry_ptr
;
144 typedef struct INTENTRY
{
154 /* descriptions of MP basetable entries */
155 typedef struct BASETABLE_ENTRY
{
162 * this code MUST be enabled here and in mpboot.s.
163 * it follows the very early stages of AP boot by placing values in CMOS ram.
164 * it NORMALLY will never be needed and thus the primitive method for enabling.
167 #if defined(CHECK_POINTS)
168 #define CHECK_READ(A) (outb(CMOS_REG, (A)), inb(CMOS_DATA))
169 #define CHECK_WRITE(A,D) (outb(CMOS_REG, (A)), outb(CMOS_DATA, (D)))
171 #define CHECK_INIT(D); \
172 CHECK_WRITE(0x34, (D)); \
173 CHECK_WRITE(0x35, (D)); \
174 CHECK_WRITE(0x36, (D)); \
175 CHECK_WRITE(0x37, (D)); \
176 CHECK_WRITE(0x38, (D)); \
177 CHECK_WRITE(0x39, (D));
179 #define CHECK_PRINT(S); \
180 kprintf("%s: %d, %d, %d, %d, %d, %d\n", \
189 #else /* CHECK_POINTS */
191 #define CHECK_INIT(D)
192 #define CHECK_PRINT(S)
194 #endif /* CHECK_POINTS */
197 * Values to send to the POST hardware.
199 #define MP_BOOTADDRESS_POST 0x10
200 #define MP_PROBE_POST 0x11
201 #define MPTABLE_PASS1_POST 0x12
203 #define MP_START_POST 0x13
204 #define MP_ENABLE_POST 0x14
205 #define MPTABLE_PASS2_POST 0x15
207 #define START_ALL_APS_POST 0x16
208 #define INSTALL_AP_TRAMP_POST 0x17
209 #define START_AP_POST 0x18
211 #define MP_ANNOUNCE_POST 0x19
213 static int need_hyperthreading_fixup
;
214 static u_int logical_cpus
;
215 u_int logical_cpus_mask
;
217 /** XXX FIXME: where does this really belong, isa.h/isa.c perhaps? */
218 int current_postcode
;
220 /** XXX FIXME: what system files declare these??? */
221 extern struct region_descriptor r_gdt
, r_idt
;
223 int bsp_apic_ready
= 0; /* flags useability of BSP apic */
224 int mp_naps
; /* # of Applications processors */
225 int mp_nbusses
; /* # of busses */
227 int mp_napics
; /* # of IO APICs */
229 int boot_cpu_id
; /* designated BSP */
230 vm_offset_t cpu_apic_address
;
232 vm_offset_t io_apic_address
[NAPICID
]; /* NAPICID is more than enough */
233 u_int32_t
*io_apic_versions
;
237 u_int32_t cpu_apic_versions
[MAXCPU
];
239 extern int64_t tsc_offsets
[];
242 struct apic_intmapinfo int_to_apicintpin
[APIC_INTMAPSIZE
];
246 * APIC ID logical/physical mapping structures.
247 * We oversize these to simplify boot-time config.
249 int cpu_num_to_apic_id
[NAPICID
];
251 int io_num_to_apic_id
[NAPICID
];
253 int apic_id_to_logical
[NAPICID
];
255 /* AP uses this during bootstrap. Do not staticize. */
259 /* Hotwire a 0->4MB V==P mapping */
260 extern pt_entry_t
*KPTphys
;
263 * SMP page table page. Setup by locore to point to a page table
264 * page from which we allocate per-cpu privatespace areas io_apics,
268 #define IO_MAPPING_START_INDEX \
269 (SMP_MAXCPU * sizeof(struct privatespace) / PAGE_SIZE)
271 extern pt_entry_t
*SMPpt
;
272 static int SMPpt_alloc_index
= IO_MAPPING_START_INDEX
;
274 struct pcb stoppcbs
[MAXCPU
];
276 extern inthand_t
IDTVEC(fast_syscall
), IDTVEC(fast_syscall32
);
278 extern void initializecpu(void);
281 * Local data and functions.
284 static int mp_capable
;
285 static u_int boot_address
;
286 static u_int base_memory
;
287 static int mp_finish
;
289 static mpfps_t mpfps
;
290 static long search_for_sig(u_int32_t target
, int count
);
291 static void mp_enable(u_int boot_addr
);
293 static void mptable_hyperthread_fixup(u_int id_mask
);
294 static void mptable_pass1(void);
295 static int mptable_pass2(void);
296 static void default_mp_table(int type
);
297 static void fix_mp_table(void);
299 static void setup_apic_irq_mapping(void);
300 static int apic_int_is_bus_type(int intr
, int bus_type
);
302 static int start_all_aps(u_int boot_addr
);
303 static void install_ap_tramp(u_int boot_addr
);
304 static int start_ap(struct mdglobaldata
*gd
, u_int boot_addr
);
306 static cpumask_t smp_startup_mask
= 1; /* which cpus have been started */
307 cpumask_t smp_active_mask
= 1; /* which cpus are ready for IPIs etc? */
308 SYSCTL_INT(_machdep
, OID_AUTO
, smp_active
, CTLFLAG_RD
, &smp_active_mask
, 0, "");
309 static u_int bootMP_size
;
312 * Calculate usable address in base memory for AP trampoline code.
315 mp_bootaddress(u_int basemem
)
317 POSTCODE(MP_BOOTADDRESS_POST
);
319 base_memory
= basemem
;
321 bootMP_size
= mptramp_end
- mptramp_start
;
322 boot_address
= trunc_page(basemem
* 1024); /* round down to 4k boundary */
323 if (((basemem
* 1024) - boot_address
) < bootMP_size
)
324 boot_address
-= PAGE_SIZE
; /* not enough, lower by 4k */
325 /* 3 levels of page table pages */
326 mptramp_pagetables
= boot_address
- (PAGE_SIZE
* 3);
328 return mptramp_pagetables
;
333 * Look for an Intel MP spec table (ie, SMP capable hardware).
343 * Make sure our SMPpt[] page table is big enough to hold all the
346 KKASSERT(IO_MAPPING_START_INDEX
< NPTEPG
- 2);
348 POSTCODE(MP_PROBE_POST
);
350 /* see if EBDA exists */
351 if ((segment
= (u_long
) * (u_short
*) (KERNBASE
+ 0x40e)) != 0) {
352 /* search first 1K of EBDA */
353 target
= (u_int32_t
) (segment
<< 4);
354 if ((x
= search_for_sig(target
, 1024 / 4)) != -1L)
357 /* last 1K of base memory, effective 'top of base' passed in */
358 target
= (u_int32_t
) (base_memory
- 0x400);
359 if ((x
= search_for_sig(target
, 1024 / 4)) != -1L)
363 /* search the BIOS */
364 target
= (u_int32_t
) BIOS_BASE
;
365 if ((x
= search_for_sig(target
, BIOS_COUNT
)) != -1L)
375 * Calculate needed resources. We can safely map physical
376 * memory into SMPpt after mptable_pass1() completes.
381 /* flag fact that we are running multiple processors */
388 * Startup the SMP processors.
393 POSTCODE(MP_START_POST
);
395 /* look for MP capable motherboard */
397 mp_enable(boot_address
);
399 panic("MP hardware not found!");
404 * Print various information about the SMP system hardware and setup.
411 POSTCODE(MP_ANNOUNCE_POST
);
413 kprintf("DragonFly/MP: Multiprocessor motherboard\n");
414 kprintf(" cpu0 (BSP): apic id: %2d", CPU_TO_ID(0));
415 kprintf(", version: 0x%08x", cpu_apic_versions
[0]);
416 kprintf(", at 0x%08x\n", cpu_apic_address
);
417 for (x
= 1; x
<= mp_naps
; ++x
) {
418 kprintf(" cpu%d (AP): apic id: %2d", x
, CPU_TO_ID(x
));
419 kprintf(", version: 0x%08x", cpu_apic_versions
[x
]);
420 kprintf(", at 0x%08x\n", cpu_apic_address
);
424 for (x
= 0; x
< mp_napics
; ++x
) {
425 kprintf(" io%d (APIC): apic id: %2d", x
, IO_TO_ID(x
));
426 kprintf(", version: 0x%08x", io_apic_versions
[x
]);
427 kprintf(", at 0x%08x\n", io_apic_address
[x
]);
430 kprintf(" Warning: APIC I/O disabled\n");
435 * AP cpu's call this to sync up protected mode.
437 * WARNING! %gs is not set up on entry. This routine sets up %gs.
443 int x
, myid
= bootAP
;
445 struct mdglobaldata
*md
;
446 struct privatespace
*ps
;
448 ps
= &CPU_prvspace
[myid
];
450 gdt_segs
[GPROC0_SEL
].ssd_base
=
451 (long) &ps
->mdglobaldata
.gd_common_tss
;
452 ps
->mdglobaldata
.mi
.gd_prvspace
= ps
;
454 /* We fill the 32-bit segment descriptors */
455 for (x
= 0; x
< NGDT
; x
++) {
456 if (x
!= GPROC0_SEL
&& x
!= (GPROC0_SEL
+ 1))
457 ssdtosd(&gdt_segs
[x
], &gdt
[myid
* NGDT
+ x
]);
459 /* And now a 64-bit one */
460 ssdtosyssd(&gdt_segs
[GPROC0_SEL
],
461 (struct system_segment_descriptor
*)&gdt
[myid
* NGDT
+ GPROC0_SEL
]);
463 r_gdt
.rd_limit
= NGDT
* sizeof(gdt
[0]) - 1;
464 r_gdt
.rd_base
= (long) &gdt
[myid
* NGDT
];
465 lgdt(&r_gdt
); /* does magic intra-segment return */
467 /* lgdt() destroys the GSBASE value, so we load GSBASE after lgdt() */
468 wrmsr(MSR_FSBASE
, 0); /* User value */
469 wrmsr(MSR_GSBASE
, (u_int64_t
)ps
);
470 wrmsr(MSR_KGSBASE
, 0); /* XXX User value while we're in the kernel */
476 mdcpu
->gd_currentldt
= _default_ldt
;
479 gsel_tss
= GSEL(GPROC0_SEL
, SEL_KPL
);
480 gdt
[myid
* NGDT
+ GPROC0_SEL
].sd_type
= SDT_SYSTSS
;
482 md
= mdcpu
; /* loaded through %gs:0 (mdglobaldata.mi.gd_prvspace)*/
484 md
->gd_common_tss
.tss_rsp0
= 0; /* not used until after switch */
486 md
->gd_common_tss
.tss_ioopt
= (sizeof md
->gd_common_tss
) << 16;
488 md
->gd_tss_gdt
= &gdt
[myid
* NGDT
+ GPROC0_SEL
];
489 md
->gd_common_tssd
= *md
->gd_tss_gdt
;
491 md
->gd_common_tss
.tss_ist1
= (long)&doublefault_stack
[PAGE_SIZE
];
496 * Set to a known state:
497 * Set by mpboot.s: CR0_PG, CR0_PE
498 * Set by cpu_setregs: CR0_NE, CR0_MP, CR0_TS, CR0_WP, CR0_AM
501 cr0
&= ~(CR0_CD
| CR0_NW
| CR0_EM
);
504 /* Set up the fast syscall stuff */
505 msr
= rdmsr(MSR_EFER
) | EFER_SCE
;
506 wrmsr(MSR_EFER
, msr
);
507 wrmsr(MSR_LSTAR
, (u_int64_t
)IDTVEC(fast_syscall
));
508 wrmsr(MSR_CSTAR
, (u_int64_t
)IDTVEC(fast_syscall32
));
509 msr
= ((u_int64_t
)GSEL(GCODE_SEL
, SEL_KPL
) << 32) |
510 ((u_int64_t
)GSEL(GUCODE32_SEL
, SEL_UPL
) << 48);
511 wrmsr(MSR_STAR
, msr
);
512 wrmsr(MSR_SF_MASK
, PSL_NT
|PSL_T
|PSL_I
|PSL_C
|PSL_D
);
514 pmap_set_opt(); /* PSE/4MB pages, etc */
516 /* Initialize the PAT MSR. */
520 /* set up CPU registers and state */
523 /* set up SSE/NX registers */
526 /* set up FPU state on the AP */
527 npxinit(__INITIAL_NPXCW__
);
529 /* disable the APIC, just to be SURE */
530 lapic
->svr
&= ~APIC_SVR_ENABLE
;
532 /* data returned to BSP */
533 cpu_apic_versions
[0] = lapic
->version
;
536 /*******************************************************************
537 * local functions and data
541 * start the SMP system
544 mp_enable(u_int boot_addr
)
552 POSTCODE(MP_ENABLE_POST
);
555 /* turn on 4MB of V == P addressing so we can get to MP table */
556 *(int *)PTD
= PG_V
| PG_RW
| ((uintptr_t)(void *)KPTphys
& PG_FRAME
);
560 /* examine the MP table for needed info, uses physical addresses */
568 /* can't process default configs till the CPU APIC is pmapped */
572 /* post scan cleanup */
577 setup_apic_irq_mapping();
579 /* fill the LOGICAL io_apic_versions table */
580 for (apic
= 0; apic
< mp_napics
; ++apic
) {
581 ux
= io_apic_read(apic
, IOAPIC_VER
);
582 io_apic_versions
[apic
] = ux
;
583 io_apic_set_id(apic
, IO_TO_ID(apic
));
586 /* program each IO APIC in the system */
587 for (apic
= 0; apic
< mp_napics
; ++apic
)
588 if (io_apic_setup(apic
) < 0)
589 panic("IO APIC setup failure");
594 * These are required for SMP operation
597 /* install a 'Spurious INTerrupt' vector */
598 setidt(XSPURIOUSINT_OFFSET
, Xspuriousint
,
599 SDT_SYSIGT
, SEL_KPL
, 0);
601 /* install an inter-CPU IPI for TLB invalidation */
602 setidt(XINVLTLB_OFFSET
, Xinvltlb
,
603 SDT_SYSIGT
, SEL_KPL
, 0);
605 /* install an inter-CPU IPI for IPIQ messaging */
606 setidt(XIPIQ_OFFSET
, Xipiq
,
607 SDT_SYSIGT
, SEL_KPL
, 0);
609 /* install a timer vector */
610 setidt(XTIMER_OFFSET
, Xtimer
,
611 SDT_SYSIGT
, SEL_KPL
, 0);
613 /* install an inter-CPU IPI for CPU stop/restart */
614 setidt(XCPUSTOP_OFFSET
, Xcpustop
,
615 SDT_SYSIGT
, SEL_KPL
, 0);
617 /* start each Application Processor */
618 start_all_aps(boot_addr
);
623 * look for the MP spec signature
626 /* string defined by the Intel MP Spec as identifying the MP table */
627 #define MP_SIG 0x5f504d5f /* _MP_ */
628 #define NEXT(X) ((X) += 4)
630 search_for_sig(u_int32_t target
, int count
)
633 u_int32_t
*addr
= (u_int32_t
*) (KERNBASE
+ target
);
635 for (x
= 0; x
< count
; NEXT(x
))
636 if (addr
[x
] == MP_SIG
)
637 /* make array index a byte index */
638 return (long)(&addr
[x
]);
644 static basetable_entry basetable_entry_types
[] =
646 {0, 20, "Processor"},
653 typedef struct BUSDATA
{
655 enum busTypes bus_type
;
658 typedef struct INTDATA
{
668 typedef struct BUSTYPENAME
{
673 static bus_type_name bus_type_table
[] =
679 {UNKNOWN_BUSTYPE
, "---"},
682 {UNKNOWN_BUSTYPE
, "---"},
683 {UNKNOWN_BUSTYPE
, "---"},
684 {UNKNOWN_BUSTYPE
, "---"},
685 {UNKNOWN_BUSTYPE
, "---"},
686 {UNKNOWN_BUSTYPE
, "---"},
688 {UNKNOWN_BUSTYPE
, "---"},
689 {UNKNOWN_BUSTYPE
, "---"},
690 {UNKNOWN_BUSTYPE
, "---"},
691 {UNKNOWN_BUSTYPE
, "---"},
693 {UNKNOWN_BUSTYPE
, "---"}
695 /* from MP spec v1.4, table 5-1 */
696 static int default_data
[7][5] =
698 /* nbus, id0, type0, id1, type1 */
699 {1, 0, ISA
, 255, 255},
700 {1, 0, EISA
, 255, 255},
701 {1, 0, EISA
, 255, 255},
702 {1, 0, MCA
, 255, 255},
704 {2, 0, EISA
, 1, PCI
},
710 static bus_datum
*bus_data
;
713 /* the IO INT data, one entry per possible APIC INTerrupt */
714 static io_int
*io_apic_ints
;
718 static int processor_entry (proc_entry_ptr entry
, int cpu
);
719 static int bus_entry (bus_entry_ptr entry
, int bus
);
721 static int io_apic_entry (io_apic_entry_ptr entry
, int apic
);
722 static int int_entry (int_entry_ptr entry
, int intr
);
724 static int lookup_bus_type (char *name
);
728 * 1st pass on motherboard's Intel MP specification table.
734 * cpu_apic_address (common to all CPUs)
754 POSTCODE(MPTABLE_PASS1_POST
);
757 /* clear various tables */
758 for (x
= 0; x
< NAPICID
; ++x
) {
759 io_apic_address
[x
] = ~0; /* IO APIC address table */
763 /* init everything to empty */
772 /* check for use of 'default' configuration */
773 if (mpfps
->mpfb1
!= 0) {
774 /* use default addresses */
775 cpu_apic_address
= DEFAULT_APIC_BASE
;
777 io_apic_address
[0] = DEFAULT_IO_APIC_BASE
;
780 /* fill in with defaults */
781 mp_naps
= 2; /* includes BSP */
782 mp_nbusses
= default_data
[mpfps
->mpfb1
- 1][0];
789 if ((cth
= mpfps
->pap
) == 0)
790 panic("MP Configuration Table Header MISSING!");
792 cpu_apic_address
= (vm_offset_t
) cth
->apic_address
;
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
;
800 switch (type
= *(u_char
*) position
) {
801 case 0: /* processor_entry */
802 if (((proc_entry_ptr
)position
)->cpu_flags
803 & PROCENTRY_FLAG_EN
) {
806 ((proc_entry_ptr
)position
)->apic_id
;
809 case 1: /* bus_entry */
812 case 2: /* io_apic_entry */
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
;
821 case 3: /* int_entry */
826 case 4: /* int_entry */
829 panic("mpfps Base Table HOSED!");
833 totalSize
-= basetable_entry_types
[type
].length
;
834 position
= (uint8_t *)position
+
835 basetable_entry_types
[type
].length
;
839 /* qualify the numbers */
840 if (mp_naps
> MAXCPU
) {
841 kprintf("Warning: only using %d of %d available CPUs!\n",
846 /* See if we need to fixup HT logical CPUs. */
847 mptable_hyperthread_fixup(id_mask
);
851 * This is also used as a counter while starting the APs.
855 --mp_naps
; /* subtract the BSP */
860 * 2nd pass on motherboard's Intel MP specification table.
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
873 struct PROCENTRY proc
;
880 int apic
, bus
, cpu
, intr
;
883 POSTCODE(MPTABLE_PASS2_POST
);
885 /* Initialize fake proc entry for use with HT fixup. */
886 bzero(&proc
, sizeof(proc
));
888 proc
.cpu_flags
= PROCENTRY_FLAG_EN
;
891 MALLOC(io_apic_versions
, u_int32_t
*, sizeof(u_int32_t
) * mp_napics
,
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
),
898 MALLOC(bus_data
, bus_datum
*, sizeof(bus_datum
) * mp_nbusses
,
902 for (i
= 0; i
< mp_napics
; i
++) {
903 ioapic
[i
] = permanent_io_mapping(io_apic_address
[i
]);
907 /* clear various tables */
908 for (x
= 0; x
< NAPICID
; ++x
) {
909 CPU_TO_ID(x
) = -1; /* logical CPU to APIC ID table */
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 */
916 /* clear bus data table */
917 for (x
= 0; x
< mp_nbusses
; ++x
)
918 bus_data
[x
].bus_id
= 0xff;
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;
928 /* setup the cpu/apic mapping arrays */
931 /* record whether PIC or virtual-wire mode */
932 machintr_setvar_simple(MACHINTR_VAR_IMCR_PRESENT
, mpfps
->mpfb2
& 0x80);
934 /* check for use of 'default' configuration */
935 if (mpfps
->mpfb1
!= 0)
936 return mpfps
->mpfb1
; /* return default configuration type */
938 if ((cth
= mpfps
->pap
) == 0)
939 panic("MP Configuration Table Header MISSING!");
941 cth
= PHYS_TO_DMAP(mpfps
->pap
);
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 */
950 switch (type
= *(u_char
*) position
) {
952 if (processor_entry(position
, cpu
))
955 if (need_hyperthreading_fixup
) {
957 * Create fake mptable processor entries
958 * and feed them to processor_entry() to
959 * enumerate the logical CPUs.
961 proc
.apic_id
= ((proc_entry_ptr
)position
)->apic_id
;
962 for (i
= 1; i
< logical_cpus
; i
++) {
964 processor_entry(&proc
, cpu
);
965 logical_cpus_mask
|= (1 << cpu
);
971 if (bus_entry(position
, bus
))
976 if (io_apic_entry(position
, apic
))
982 if (int_entry(position
, intr
))
987 /* int_entry(position); */
990 panic("mpfps Base Table HOSED!");
994 totalSize
-= basetable_entry_types
[type
].length
;
995 position
= (uint8_t *)position
+ basetable_entry_types
[type
].length
;
998 if (boot_cpu_id
== -1)
999 panic("NO BSP found!");
1001 /* report fact that its NOT a default configuration */
1006 * Check if we should perform a hyperthreading "fix-up" to
1007 * enumerate any logical CPU's that aren't already listed
1010 * XXX: We assume that all of the physical CPUs in the
1011 * system have the same number of logical CPUs.
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.
1018 mptable_hyperthread_fixup(u_int id_mask
)
1022 /* Nothing to do if there is no HTT support. */
1023 if ((cpu_feature
& CPUID_HTT
) == 0)
1025 logical_cpus
= (cpu_procinfo
& CPUID_HTT_CORES
) >> 16;
1026 if (logical_cpus
<= 1)
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.
1035 for (id
= 0; id
<= MAXCPU
; id
++) {
1036 if ((id_mask
& 1 << id
) == 0)
1038 /* First, make sure we are on a logical_cpus boundary. */
1039 if (id
% logical_cpus
!= 0)
1041 for (i
= id
+ 1; i
< id
+ logical_cpus
; i
++)
1042 if ((id_mask
& 1 << i
) != 0)
1047 * Ok, the ID's checked out, so enable the fixup. We have to fixup
1048 * mp_naps right now.
1050 need_hyperthreading_fixup
= 1;
1051 mp_naps
*= logical_cpus
;
1057 assign_apic_irq(int apic
, int intpin
, int irq
)
1061 if (int_to_apicintpin
[irq
].ioapic
!= -1)
1062 panic("assign_apic_irq: inconsistent table");
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
;
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
;
1080 revoke_apic_irq(int irq
)
1086 if (int_to_apicintpin
[irq
].ioapic
== -1)
1087 panic("revoke_apic_irq: inconsistent table");
1089 oldapic
= int_to_apicintpin
[irq
].ioapic
;
1090 oldintpin
= int_to_apicintpin
[irq
].int_pin
;
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;
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;
1111 allocate_apic_irq(int intr
)
1117 if (io_apic_ints
[intr
].int_vector
!= 0xff)
1118 return; /* Interrupt handler already assigned */
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) */
1127 while (irq
< APIC_INTMAPSIZE
&&
1128 int_to_apicintpin
[irq
].ioapic
!= -1)
1131 if (irq
>= APIC_INTMAPSIZE
)
1132 return; /* No free interrupt handlers */
1134 apic
= ID_TO_IO(io_apic_ints
[intr
].dst_apic_id
);
1135 intpin
= io_apic_ints
[intr
].dst_apic_int
;
1137 assign_apic_irq(apic
, intpin
, irq
);
1142 swap_apic_id(int apic
, int oldid
, int newid
)
1149 return; /* Nothing to do */
1151 kprintf("Changing APIC ID for IO APIC #%d from %d to %d in MP table\n",
1152 apic
, oldid
, newid
);
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
;
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
)
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
;
1173 IO_TO_ID(apic
) = newid
;
1178 fix_id_to_io_mapping(void)
1182 for (x
= 0; x
< NAPICID
; x
++)
1185 for (x
= 0; x
<= mp_naps
; x
++)
1186 if (CPU_TO_ID(x
) < NAPICID
)
1187 ID_TO_IO(CPU_TO_ID(x
)) = x
;
1189 for (x
= 0; x
< mp_napics
; x
++)
1190 if (IO_TO_ID(x
) < NAPICID
)
1191 ID_TO_IO(IO_TO_ID(x
)) = x
;
1196 first_free_apic_id(void)
1200 for (freeid
= 0; freeid
< NAPICID
; freeid
++) {
1201 for (x
= 0; x
<= mp_naps
; x
++)
1202 if (CPU_TO_ID(x
) == freeid
)
1206 for (x
= 0; x
< mp_napics
; x
++)
1207 if (IO_TO_ID(x
) == freeid
)
1218 io_apic_id_acceptable(int apic
, int id
)
1220 int cpu
; /* Logical CPU number */
1221 int oapic
; /* Logical IO APIC number for other IO APIC */
1224 return 0; /* Out of range */
1226 for (cpu
= 0; cpu
<= mp_naps
; cpu
++)
1227 if (CPU_TO_ID(cpu
) == id
)
1228 return 0; /* Conflict with CPU */
1230 for (oapic
= 0; oapic
< mp_napics
&& oapic
< apic
; oapic
++)
1231 if (IO_TO_ID(oapic
) == id
)
1232 return 0; /* Conflict with other APIC */
1234 return 1; /* ID is acceptable for IO APIC */
1239 io_apic_find_int_entry(int apic
, int pin
)
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
]);
1255 * parse an Intel MP specification table
1263 int apic
; /* IO APIC unit number */
1264 int freeid
; /* Free physical APIC ID */
1265 int physid
; /* Current physical IO APIC ID */
1268 int bus_0
= 0; /* Stop GCC warning */
1269 int bus_pci
= 0; /* Stop GCC warning */
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".
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) {
1289 if (bus_data
[x
].bus_type
== PCI
) {
1295 * bus_0 == slot of bus with ID of 0
1296 * bus_pci == slot of last PCI bus encountered
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) {
1304 /* mis-numbered, swap with whichever bus uses slot 0 */
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
;
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;
1317 else if (io_apic_ints
[x
].src_bus_id
== 0) {
1318 io_apic_ints
[x
].src_bus_id
= id
;
1325 /* Assign IO APIC IDs.
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
1331 * We cannot use the ID_TO_IO table before all conflicts has been
1332 * resolved and the table has been corrected.
1334 for (apic
= 0; apic
< mp_napics
; ++apic
) { /* For all IO APICs */
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
);
1344 /* Then check if the value in the MP table is acceptable */
1345 if (io_apic_id_acceptable(apic
, IO_TO_ID(apic
)))
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");
1353 if (io_apic_id_acceptable(apic
, freeid
)) {
1354 swap_apic_id(apic
, IO_TO_ID(apic
), freeid
);
1357 panic("Free physical APIC ID not usable");
1359 fix_id_to_io_mapping();
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 */
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;
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.
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.
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;
1405 /* Assign low level interrupt handlers */
1407 setup_apic_irq_mapping(void)
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;
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
,
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);
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
);
1459 processor_entry(proc_entry_ptr entry
, int cpu
)
1461 /* check for usability */
1462 if (!(entry
->cpu_flags
& PROCENTRY_FLAG_EN
))
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 */
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
;
1487 bus_entry(bus_entry_ptr entry
, int bus
)
1492 /* encode the name into an index */
1493 for (x
= 0; x
< 6; ++x
) {
1494 if ((c
= entry
->bus_type
[x
]) == ' ')
1500 if ((x
= lookup_bus_type(name
)) == UNKNOWN_BUSTYPE
)
1501 panic("unknown bus type: '%s'", name
);
1503 bus_data
[bus
].bus_id
= entry
->bus_id
;
1504 bus_data
[bus
].bus_type
= x
;
1512 io_apic_entry(io_apic_entry_ptr entry
, int apic
)
1514 if (!(entry
->apic_flags
& IOAPICENTRY_FLAG_EN
))
1517 IO_TO_ID(apic
) = entry
->apic_id
;
1518 if (entry
->apic_id
< NAPICID
)
1519 ID_TO_IO(entry
->apic_id
) = apic
;
1527 lookup_bus_type(char *name
)
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
;
1535 return UNKNOWN_BUSTYPE
;
1541 int_entry(int_entry_ptr entry
, int intr
)
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
)
1557 if (apic
< mp_napics
)
1558 io_apic_ints
[intr
].dst_apic_id
= IO_TO_ID(apic
);
1560 io_apic_ints
[intr
].dst_apic_id
= entry
->dst_apic_id
;
1562 io_apic_ints
[intr
].dst_apic_id
= entry
->dst_apic_id
;
1563 io_apic_ints
[intr
].dst_apic_int
= entry
->dst_apic_int
;
1569 apic_int_is_bus_type(int intr
, int bus_type
)
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
))
1582 * Given a traditional ISA INT mask, return an APIC mask.
1585 isa_apic_mask(u_int isa_mask
)
1590 #if defined(SKIP_IRQ15_REDIRECT)
1591 if (isa_mask
== (1 << 15)) {
1592 kprintf("skipping ISA IRQ15 redirect\n");
1595 #endif /* SKIP_IRQ15_REDIRECT */
1597 isa_irq
= ffs(isa_mask
); /* find its bit position */
1598 if (isa_irq
== 0) /* doesn't exist */
1600 --isa_irq
; /* make it zero based */
1602 apic_pin
= isa_apic_irq(isa_irq
); /* look for APIC connection */
1606 return (1 << apic_pin
); /* convert pin# to a mask */
1610 * Determine which APIC pin an ISA/EISA INT is attached to.
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))
1617 #define SRCBUSIRQ(I) (io_apic_ints[(I)].src_bus_irq)
1619 isa_apic_irq(int isa_irq
)
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 */
1635 return -1; /* NOT found */
1640 * Determine which APIC pin a PCI INT is attached to.
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)
1646 pci_apic_irq(int pciBus
, int pciDevice
, int pciInt
)
1650 --pciInt
; /* zero based */
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
)) {
1658 if (INTIRQ(intr
) == 0xff) {
1659 kprintf("IOAPIC: pci_apic_irq() "
1661 return -1; /* unassigned */
1663 return INTIRQ(intr
); /* exact match */
1668 return -1; /* NOT found */
1672 next_apic_irq(int irq
)
1679 for (intr
= 0; intr
< nintrs
; intr
++) {
1680 if (INTIRQ(intr
) != irq
|| INTTYPE(intr
) != 0)
1682 bus
= SRCBUSID(intr
);
1683 bustype
= apic_bus_type(bus
);
1684 if (bustype
!= ISA
&&
1690 if (intr
>= nintrs
) {
1693 for (ointr
= intr
+ 1; ointr
< nintrs
; ointr
++) {
1694 if (INTTYPE(ointr
) != 0)
1696 if (bus
!= SRCBUSID(ointr
))
1698 if (bustype
== PCI
) {
1699 if (SRCBUSDEVICE(intr
) != SRCBUSDEVICE(ointr
))
1701 if (SRCBUSLINE(intr
) != SRCBUSLINE(ointr
))
1704 if (bustype
== ISA
|| bustype
== EISA
) {
1705 if (SRCBUSIRQ(intr
) != SRCBUSIRQ(ointr
))
1708 if (INTPIN(intr
) == INTPIN(ointr
))
1712 if (ointr
>= nintrs
) {
1715 return INTIRQ(ointr
);
1730 * Reprogram the MB chipset to NOT redirect an ISA INTerrupt.
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
1740 undirect_isa_irq(int rirq
)
1744 kprintf("Freeing redirected ISA irq %d.\n", rirq
);
1745 /** FIXME: tickle the MB redirector chip */
1749 kprintf("Freeing (NOT implemented) redirected ISA irq %d.\n", rirq
);
1756 * Reprogram the MB chipset to NOT redirect a PCI INTerrupt
1759 undirect_pci_irq(int rirq
)
1763 kprintf("Freeing redirected PCI irq %d.\n", rirq
);
1765 /** FIXME: tickle the MB redirector chip */
1769 kprintf("Freeing (NOT implemented) redirected PCI irq %d.\n",
1777 * given a bus ID, return:
1778 * the bus type if found
1782 apic_bus_type(int id
)
1786 for (x
= 0; x
< mp_nbusses
; ++x
)
1787 if (bus_data
[x
].bus_id
== id
)
1788 return bus_data
[x
].bus_type
;
1796 * given a LOGICAL APIC# and pin#, return:
1797 * the associated src bus ID if found
1801 apic_src_bus_id(int apic
, int pin
)
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
);
1811 return -1; /* NOT found */
1815 * given a LOGICAL APIC# and pin#, return:
1816 * the associated src bus IRQ if found
1820 apic_src_bus_irq(int apic
, int pin
)
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
);
1829 return -1; /* NOT found */
1834 * given a LOGICAL APIC# and pin#, return:
1835 * the associated INTerrupt type if found
1839 apic_int_type(int apic
, int pin
)
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
);
1849 return -1; /* NOT found */
1853 * Return the IRQ associated with an APIC pin
1856 apic_irq(int apic
, int pin
)
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
;
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)");
1879 * given a LOGICAL APIC# and pin#, return:
1880 * the associated trigger mode if found
1884 apic_trigger(int apic
, int pin
)
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);
1894 return -1; /* NOT found */
1899 * given a LOGICAL APIC# and pin#, return:
1900 * the associated 'active' level if found
1904 apic_polarity(int apic
, int pin
)
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);
1914 return -1; /* NOT found */
1920 * set data according to MP defaults
1921 * FIXME: probably not complete yet...
1924 default_mp_table(int type
)
1927 #if defined(APIC_IO)
1930 #endif /* APIC_IO */
1933 kprintf(" MP default config type: %d\n", type
);
1936 kprintf(" bus: ISA, APIC: 82489DX\n");
1939 kprintf(" bus: EISA, APIC: 82489DX\n");
1942 kprintf(" bus: EISA, APIC: 82489DX\n");
1945 kprintf(" bus: MCA, APIC: 82489DX\n");
1948 kprintf(" bus: ISA+PCI, APIC: Integrated\n");
1951 kprintf(" bus: EISA+PCI, APIC: Integrated\n");
1954 kprintf(" bus: MCA+PCI, APIC: Integrated\n");
1957 kprintf(" future type\n");
1963 boot_cpu_id
= (lapic
->id
& APIC_ID_MASK
) >> 24;
1964 ap_cpu_id
= (boot_cpu_id
== 0) ? 1 : 0;
1967 CPU_TO_ID(0) = boot_cpu_id
;
1968 ID_TO_CPU(boot_cpu_id
) = 0;
1970 /* one and only AP */
1971 CPU_TO_ID(1) = ap_cpu_id
;
1972 ID_TO_CPU(ap_cpu_id
) = 1;
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;
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
1982 #if defined(REALLY_ANAL_IOAPICID_VALUE)
1983 if (io_apic_id
!= 2) {
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);
1990 IO_TO_ID(0) = io_apic_id
;
1991 ID_TO_IO(io_apic_id
) = 0;
1992 #endif /* APIC_IO */
1994 /* fill out bus entries */
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];
2009 /* case 4: case 7: MCA NOT supported */
2010 default: /* illegal/reserved */
2011 panic("BAD default MP config: %d", type
);
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 */
2026 /* special cases from MP v1.4, table 5-2 */
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)
2032 panic("sorry, can't support type 2 default yet");
2033 #endif /* APIC_MIXED_MODE */
2036 io_apic_ints
[2].src_bus_irq
= 0; /* ISA IRQ0 is on APIC INT 2 */
2039 io_apic_ints
[0].int_type
= 0xff; /* N/C */
2041 io_apic_ints
[0].int_type
= 3; /* vectored 8259 */
2042 #endif /* APIC_IO */
2046 * Map a physical memory address representing I/O into KVA. The I/O
2047 * block is assumed not to cross a page boundary.
2050 permanent_io_mapping(vm_paddr_t pa
)
2052 KKASSERT(pa
< 0x100000000LL
);
2054 return pmap_mapdev_uncacheable(pa
, PAGE_SIZE
);
2058 * start each AP in our list
2061 start_all_aps(u_int boot_addr
)
2063 vm_offset_t va
= boot_address
+ KERNBASE
;
2064 u_int64_t
*pt4
, *pt3
, *pt2
;
2067 u_char mpbiosreason
;
2068 u_long mpbioswarmvec
;
2069 struct mdglobaldata
*gd
;
2070 struct privatespace
*ps
;
2074 POSTCODE(START_ALL_APS_POST
);
2076 /* Initialize BSP's local APIC */
2077 apic_initialize(TRUE
);
2080 /* install the AP 1st level boot code */
2081 pmap_kenter(va
, boot_address
);
2082 cpu_invlpg(va
); /* JG XXX */
2083 bcopy(mptramp_start
, (void *)va
, bootMP_size
);
2085 /* Locate the page tables, they'll be below the trampoline */
2086 pt4
= (u_int64_t
*)(uintptr_t)(mptramp_pagetables
+ KERNBASE
);
2087 pt3
= pt4
+ (PAGE_SIZE
) / sizeof(u_int64_t
);
2088 pt2
= pt3
+ (PAGE_SIZE
) / sizeof(u_int64_t
);
2090 /* Create the initial 1GB replicated page tables */
2091 for (i
= 0; i
< 512; i
++) {
2092 /* Each slot of the level 4 pages points to the same level 3 page */
2093 pt4
[i
] = (u_int64_t
)(uintptr_t)(mptramp_pagetables
+ PAGE_SIZE
);
2094 pt4
[i
] |= PG_V
| PG_RW
| PG_U
;
2096 /* Each slot of the level 3 pages points to the same level 2 page */
2097 pt3
[i
] = (u_int64_t
)(uintptr_t)(mptramp_pagetables
+ (2 * PAGE_SIZE
));
2098 pt3
[i
] |= PG_V
| PG_RW
| PG_U
;
2100 /* The level 2 page slots are mapped with 2MB pages for 1GB. */
2101 pt2
[i
] = i
* (2 * 1024 * 1024);
2102 pt2
[i
] |= PG_V
| PG_RW
| PG_PS
| PG_U
;
2105 /* save the current value of the warm-start vector */
2106 mpbioswarmvec
= *((u_int32_t
*) WARMBOOT_OFF
);
2107 outb(CMOS_REG
, BIOS_RESET
);
2108 mpbiosreason
= inb(CMOS_DATA
);
2110 /* setup a vector to our boot code */
2111 *((volatile u_short
*) WARMBOOT_OFF
) = WARMBOOT_TARGET
;
2112 *((volatile u_short
*) WARMBOOT_SEG
) = (boot_address
>> 4);
2113 outb(CMOS_REG
, BIOS_RESET
);
2114 outb(CMOS_DATA
, BIOS_WARM
); /* 'warm-start' */
2117 for (x
= 1; x
<= mp_naps
; ++x
) {
2119 /* This is a bit verbose, it will go away soon. */
2121 /* first page of AP's private space */
2122 pg
= x
* amd64_btop(sizeof(struct privatespace
));
2124 /* allocate new private data page(s) */
2125 gd
= (struct mdglobaldata
*)kmem_alloc(&kernel_map
,
2126 MDGLOBALDATA_BASEALLOC_SIZE
);
2128 /* wire it into the private page table page */
2129 for (i
= 0; i
< MDGLOBALDATA_BASEALLOC_SIZE
; i
+= PAGE_SIZE
) {
2130 SMPpt
[pg
+ i
/ PAGE_SIZE
] = (pt_entry_t
)
2131 (PG_V
| PG_RW
| vtophys_pte((char *)gd
+ i
));
2133 pg
+= MDGLOBALDATA_BASEALLOC_PAGES
;
2135 SMPpt
[pg
+ 0] = 0; /* *gd_CMAP1 */
2136 SMPpt
[pg
+ 1] = 0; /* *gd_CMAP2 */
2137 SMPpt
[pg
+ 2] = 0; /* *gd_CMAP3 */
2138 SMPpt
[pg
+ 3] = 0; /* *gd_PMAP1 */
2140 /* allocate and set up an idle stack data page */
2141 stack
= (char *)kmem_alloc(&kernel_map
, UPAGES
*PAGE_SIZE
);
2142 for (i
= 0; i
< UPAGES
; i
++) {
2143 SMPpt
[pg
+ 4 + i
] = (pt_entry_t
)
2144 (PG_V
| PG_RW
| vtophys_pte(PAGE_SIZE
* i
+ stack
));
2148 gd
= &CPU_prvspace
[x
].mdglobaldata
; /* official location */
2149 bzero(gd
, sizeof(*gd
));
2150 gd
->mi
.gd_prvspace
= ps
= &CPU_prvspace
[x
];
2152 /* prime data page for it to use */
2153 mi_gdinit(&gd
->mi
, x
);
2155 gd
->gd_CMAP1
= &SMPpt
[pg
+ 0];
2156 gd
->gd_CMAP2
= &SMPpt
[pg
+ 1];
2157 gd
->gd_CMAP3
= &SMPpt
[pg
+ 2];
2158 gd
->gd_PMAP1
= &SMPpt
[pg
+ 3];
2159 gd
->gd_CADDR1
= ps
->CPAGE1
;
2160 gd
->gd_CADDR2
= ps
->CPAGE2
;
2161 gd
->gd_CADDR3
= ps
->CPAGE3
;
2162 gd
->gd_PADDR1
= (unsigned *)ps
->PPAGE1
;
2163 gd
->mi
.gd_ipiq
= (void *)kmem_alloc(&kernel_map
, sizeof(lwkt_ipiq
) * (mp_naps
+ 1));
2164 bzero(gd
->mi
.gd_ipiq
, sizeof(lwkt_ipiq
) * (mp_naps
+ 1));
2166 /* setup a vector to our boot code */
2167 *((volatile u_short
*) WARMBOOT_OFF
) = WARMBOOT_TARGET
;
2168 *((volatile u_short
*) WARMBOOT_SEG
) = (boot_addr
>> 4);
2169 outb(CMOS_REG
, BIOS_RESET
);
2170 outb(CMOS_DATA
, BIOS_WARM
); /* 'warm-start' */
2173 * Setup the AP boot stack
2175 bootSTK
= &ps
->idlestack
[UPAGES
*PAGE_SIZE
/2];
2178 /* attempt to start the Application Processor */
2179 CHECK_INIT(99); /* setup checkpoints */
2180 if (!start_ap(gd
, boot_addr
)) {
2181 kprintf("AP #%d (PHY# %d) failed!\n", x
, CPU_TO_ID(x
));
2182 CHECK_PRINT("trace"); /* show checkpoints */
2183 /* better panic as the AP may be running loose */
2184 kprintf("panic y/n? [y] ");
2185 if (cngetc() != 'n')
2188 CHECK_PRINT("trace"); /* show checkpoints */
2190 /* record its version info */
2191 cpu_apic_versions
[x
] = cpu_apic_versions
[0];
2194 /* set ncpus to 1 + highest logical cpu. Not all may have come up */
2197 /* ncpus2 -- ncpus rounded down to the nearest power of 2 */
2198 for (shift
= 0; (1 << shift
) <= ncpus
; ++shift
)
2201 ncpus2_shift
= shift
;
2202 ncpus2
= 1 << shift
;
2203 ncpus2_mask
= ncpus2
- 1;
2205 /* ncpus_fit -- ncpus rounded up to the nearest power of 2 */
2206 if ((1 << shift
) < ncpus
)
2208 ncpus_fit
= 1 << shift
;
2209 ncpus_fit_mask
= ncpus_fit
- 1;
2211 /* build our map of 'other' CPUs */
2212 mycpu
->gd_other_cpus
= smp_startup_mask
& ~(1 << mycpu
->gd_cpuid
);
2213 mycpu
->gd_ipiq
= (void *)kmem_alloc(&kernel_map
, sizeof(lwkt_ipiq
) * ncpus
);
2214 bzero(mycpu
->gd_ipiq
, sizeof(lwkt_ipiq
) * ncpus
);
2216 /* fill in our (BSP) APIC version */
2217 cpu_apic_versions
[0] = lapic
->version
;
2219 /* restore the warmstart vector */
2220 *(u_long
*) WARMBOOT_OFF
= mpbioswarmvec
;
2221 outb(CMOS_REG
, BIOS_RESET
);
2222 outb(CMOS_DATA
, mpbiosreason
);
2225 * NOTE! The idlestack for the BSP was setup by locore. Finish
2226 * up, clean out the P==V mapping we did earlier.
2229 for (x
= 0; x
< NKPT
; x
++)
2234 /* number of APs actually started */
2240 * load the 1st level AP boot code into base memory.
2243 /* targets for relocation */
2244 extern void bigJump(void);
2245 extern void bootCodeSeg(void);
2246 extern void bootDataSeg(void);
2247 extern void MPentry(void);
2248 extern u_int MP_GDT
;
2249 extern u_int mp_gdtbase
;
2252 install_ap_tramp(u_int boot_addr
)
2255 int size
= *(int *) ((u_long
) & bootMP_size
);
2256 u_char
*src
= (u_char
*) ((u_long
) bootMP
);
2257 u_char
*dst
= (u_char
*) boot_addr
+ KERNBASE
;
2258 u_int boot_base
= (u_int
) bootMP
;
2263 POSTCODE(INSTALL_AP_TRAMP_POST
);
2265 for (x
= 0; x
< size
; ++x
)
2269 * modify addresses in code we just moved to basemem. unfortunately we
2270 * need fairly detailed info about mpboot.s for this to work. changes
2271 * to mpboot.s might require changes here.
2274 /* boot code is located in KERNEL space */
2275 dst
= (u_char
*) boot_addr
+ KERNBASE
;
2277 /* modify the lgdt arg */
2278 dst32
= (u_int32_t
*) (dst
+ ((u_int
) & mp_gdtbase
- boot_base
));
2279 *dst32
= boot_addr
+ ((u_int
) & MP_GDT
- boot_base
);
2281 /* modify the ljmp target for MPentry() */
2282 dst32
= (u_int32_t
*) (dst
+ ((u_int
) bigJump
- boot_base
) + 1);
2283 *dst32
= ((u_int
) MPentry
- KERNBASE
);
2285 /* modify the target for boot code segment */
2286 dst16
= (u_int16_t
*) (dst
+ ((u_int
) bootCodeSeg
- boot_base
));
2287 dst8
= (u_int8_t
*) (dst16
+ 1);
2288 *dst16
= (u_int
) boot_addr
& 0xffff;
2289 *dst8
= ((u_int
) boot_addr
>> 16) & 0xff;
2291 /* modify the target for boot data segment */
2292 dst16
= (u_int16_t
*) (dst
+ ((u_int
) bootDataSeg
- boot_base
));
2293 dst8
= (u_int8_t
*) (dst16
+ 1);
2294 *dst16
= (u_int
) boot_addr
& 0xffff;
2295 *dst8
= ((u_int
) boot_addr
>> 16) & 0xff;
2300 * this function starts the AP (application processor) identified
2301 * by the APIC ID 'physicalCpu'. It does quite a "song and dance"
2302 * to accomplish this. This is necessary because of the nuances
2303 * of the different hardware we might encounter. It ain't pretty,
2304 * but it seems to work.
2306 * NOTE: eventually an AP gets to ap_init(), which is called just
2307 * before the AP goes into the LWKT scheduler's idle loop.
2310 start_ap(struct mdglobaldata
*gd
, u_int boot_addr
)
2314 u_long icr_lo
, icr_hi
;
2316 POSTCODE(START_AP_POST
);
2318 /* get the PHYSICAL APIC ID# */
2319 physical_cpu
= CPU_TO_ID(gd
->mi
.gd_cpuid
);
2321 /* calculate the vector */
2322 vector
= (boot_addr
>> 12) & 0xff;
2324 /* Make sure the target cpu sees everything */
2328 * first we do an INIT/RESET IPI this INIT IPI might be run, reseting
2329 * and running the target CPU. OR this INIT IPI might be latched (P5
2330 * bug), CPU waiting for STARTUP IPI. OR this INIT IPI might be
2334 /* setup the address for the target AP */
2335 icr_hi
= lapic
->icr_hi
& ~APIC_ID_MASK
;
2336 icr_hi
|= (physical_cpu
<< 24);
2337 lapic
->icr_hi
= icr_hi
;
2339 /* do an INIT IPI: assert RESET */
2340 icr_lo
= lapic
->icr_lo
& 0xfff00000;
2341 lapic
->icr_lo
= icr_lo
| 0x0000c500;
2343 /* wait for pending status end */
2344 while (lapic
->icr_lo
& APIC_DELSTAT_MASK
)
2347 /* do an INIT IPI: deassert RESET */
2348 lapic
->icr_lo
= icr_lo
| 0x00008500;
2350 /* wait for pending status end */
2351 u_sleep(10000); /* wait ~10mS */
2352 while (lapic
->icr_lo
& APIC_DELSTAT_MASK
)
2356 * next we do a STARTUP IPI: the previous INIT IPI might still be
2357 * latched, (P5 bug) this 1st STARTUP would then terminate
2358 * immediately, and the previously started INIT IPI would continue. OR
2359 * the previous INIT IPI has already run. and this STARTUP IPI will
2360 * run. OR the previous INIT IPI was ignored. and this STARTUP IPI
2364 /* do a STARTUP IPI */
2365 lapic
->icr_lo
= icr_lo
| 0x00000600 | vector
;
2366 while (lapic
->icr_lo
& APIC_DELSTAT_MASK
)
2368 u_sleep(200); /* wait ~200uS */
2371 * finally we do a 2nd STARTUP IPI: this 2nd STARTUP IPI should run IF
2372 * the previous STARTUP IPI was cancelled by a latched INIT IPI. OR
2373 * this STARTUP IPI will be ignored, as only ONE STARTUP IPI is
2374 * recognized after hardware RESET or INIT IPI.
2377 lapic
->icr_lo
= icr_lo
| 0x00000600 | vector
;
2378 while (lapic
->icr_lo
& APIC_DELSTAT_MASK
)
2380 u_sleep(200); /* wait ~200uS */
2382 /* wait for it to start, see ap_init() */
2383 set_apic_timer(5000000);/* == 5 seconds */
2384 while (read_apic_timer()) {
2385 if (smp_startup_mask
& (1 << gd
->mi
.gd_cpuid
))
2386 return 1; /* return SUCCESS */
2388 return 0; /* return FAILURE */
2393 * Lazy flush the TLB on all other CPU's. DEPRECATED.
2395 * If for some reason we were unable to start all cpus we cannot safely
2396 * use broadcast IPIs.
2402 if (smp_startup_mask
== smp_active_mask
) {
2403 all_but_self_ipi(XINVLTLB_OFFSET
);
2405 selected_apic_ipi(smp_active_mask
, XINVLTLB_OFFSET
,
2406 APIC_DELMODE_FIXED
);
2412 * When called the executing CPU will send an IPI to all other CPUs
2413 * requesting that they halt execution.
2415 * Usually (but not necessarily) called with 'other_cpus' as its arg.
2417 * - Signals all CPUs in map to stop.
2418 * - Waits for each to stop.
2425 * XXX FIXME: this is not MP-safe, needs a lock to prevent multiple CPUs
2426 * from executing at same time.
2429 stop_cpus(u_int map
)
2431 map
&= smp_active_mask
;
2433 /* send the Xcpustop IPI to all CPUs in map */
2434 selected_apic_ipi(map
, XCPUSTOP_OFFSET
, APIC_DELMODE_FIXED
);
2436 while ((stopped_cpus
& map
) != map
)
2444 * Called by a CPU to restart stopped CPUs.
2446 * Usually (but not necessarily) called with 'stopped_cpus' as its arg.
2448 * - Signals all CPUs in map to restart.
2449 * - Waits for each to restart.
2457 restart_cpus(u_int map
)
2459 /* signal other cpus to restart */
2460 started_cpus
= map
& smp_active_mask
;
2462 while ((stopped_cpus
& map
) != 0) /* wait for each to clear its bit */
2469 * This is called once the mpboot code has gotten us properly relocated
2470 * and the MMU turned on, etc. ap_init() is actually the idle thread,
2471 * and when it returns the scheduler will call the real cpu_idle() main
2472 * loop for the idlethread. Interrupts are disabled on entry and should
2473 * remain disabled at return.
2481 * Adjust smp_startup_mask to signal the BSP that we have started
2482 * up successfully. Note that we do not yet hold the BGL. The BSP
2483 * is waiting for our signal.
2485 * We can't set our bit in smp_active_mask yet because we are holding
2486 * interrupts physically disabled and remote cpus could deadlock
2487 * trying to send us an IPI.
2489 smp_startup_mask
|= 1 << mycpu
->gd_cpuid
;
2493 * Interlock for finalization. Wait until mp_finish is non-zero,
2494 * then get the MP lock.
2496 * Note: We are in a critical section.
2498 * Note: We have to synchronize td_mpcount to our desired MP state
2499 * before calling cpu_try_mplock().
2501 * Note: we are the idle thread, we can only spin.
2503 * Note: The load fence is memory volatile and prevents the compiler
2504 * from improperly caching mp_finish, and the cpu from improperly
2507 while (mp_finish
== 0)
2509 ++curthread
->td_mpcount
;
2510 while (cpu_try_mplock() == 0)
2513 if (cpu_feature
& CPUID_TSC
) {
2515 * The BSP is constantly updating tsc0_offset, figure out the
2516 * relative difference to synchronize ktrdump.
2518 tsc_offsets
[mycpu
->gd_cpuid
] = rdtsc() - tsc0_offset
;
2521 /* BSP may have changed PTD while we're waiting for the lock */
2524 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
2528 /* Build our map of 'other' CPUs. */
2529 mycpu
->gd_other_cpus
= smp_startup_mask
& ~(1 << mycpu
->gd_cpuid
);
2531 kprintf("SMP: AP CPU #%d Launched!\n", mycpu
->gd_cpuid
);
2533 /* A quick check from sanity claus */
2534 apic_id
= (apic_id_to_logical
[(lapic
->id
& 0x0f000000) >> 24]);
2535 if (mycpu
->gd_cpuid
!= apic_id
) {
2536 kprintf("SMP: cpuid = %d\n", mycpu
->gd_cpuid
);
2537 kprintf("SMP: apic_id = %d\n", apic_id
);
2539 kprintf("PTD[MPPTDI] = %p\n", (void *)PTD
[MPPTDI
]);
2541 panic("cpuid mismatch! boom!!");
2544 /* Initialize AP's local APIC for irq's */
2545 apic_initialize(FALSE
);
2547 /* Set memory range attributes for this CPU to match the BSP */
2548 mem_range_AP_init();
2551 * Once we go active we must process any IPIQ messages that may
2552 * have been queued, because no actual IPI will occur until we
2553 * set our bit in the smp_active_mask. If we don't the IPI
2554 * message interlock could be left set which would also prevent
2557 * The idle loop doesn't expect the BGL to be held and while
2558 * lwkt_switch() normally cleans things up this is a special case
2559 * because we returning almost directly into the idle loop.
2561 * The idle thread is never placed on the runq, make sure
2562 * nothing we've done put it there.
2564 KKASSERT(curthread
->td_mpcount
== 1);
2565 smp_active_mask
|= 1 << mycpu
->gd_cpuid
;
2568 * Enable interrupts here. idle_restore will also do it, but
2569 * doing it here lets us clean up any strays that got posted to
2570 * the CPU during the AP boot while we are still in a critical
2573 __asm
__volatile("sti; pause; pause"::);
2574 mdcpu
->gd_fpending
= 0;
2576 initclocks_pcpu(); /* clock interrupts (via IPIs) */
2577 lwkt_process_ipiq();
2580 * Releasing the mp lock lets the BSP finish up the SMP init
2583 KKASSERT((curthread
->td_flags
& TDF_RUNQ
) == 0);
2587 * Get SMP fully working before we start initializing devices.
2595 kprintf("Finish MP startup\n");
2596 if (cpu_feature
& CPUID_TSC
)
2597 tsc0_offset
= rdtsc();
2600 while (smp_active_mask
!= smp_startup_mask
) {
2602 if (cpu_feature
& CPUID_TSC
)
2603 tsc0_offset
= rdtsc();
2605 while (try_mplock() == 0)
2608 kprintf("Active CPU Mask: %08x\n", smp_active_mask
);
2611 SYSINIT(finishsmp
, SI_BOOT2_FINISH_SMP
, SI_ORDER_FIRST
, ap_finish
, NULL
)
2614 cpu_send_ipiq(int dcpu
)
2616 if ((1 << dcpu
) & smp_active_mask
)
2617 single_apic_ipi(dcpu
, XIPIQ_OFFSET
, APIC_DELMODE_FIXED
);
2620 #if 0 /* single_apic_ipi_passive() not working yet */
2622 * Returns 0 on failure, 1 on success
2625 cpu_send_ipiq_passive(int dcpu
)
2628 if ((1 << dcpu
) & smp_active_mask
) {
2629 r
= single_apic_ipi_passive(dcpu
, XIPIQ_OFFSET
,
2630 APIC_DELMODE_FIXED
);