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 void *oem_table_pointer
;
110 u_short oem_table_size
;
113 u_short extended_table_length
;
114 u_char extended_table_checksum
;
119 typedef struct PROCENTRY
{
124 u_long cpu_signature
;
125 u_long feature_flags
;
130 typedef struct BUSENTRY
{
136 typedef struct IOAPICENTRY
{
142 } *io_apic_entry_ptr
;
144 typedef struct INTENTRY
{
154 /* descriptions of MP basetable entries */
155 typedef struct BASETABLE_ENTRY
{
164 vm_size_t mp_cth_mapsz
;
168 * this code MUST be enabled here and in mpboot.s.
169 * it follows the very early stages of AP boot by placing values in CMOS ram.
170 * it NORMALLY will never be needed and thus the primitive method for enabling.
173 #if defined(CHECK_POINTS)
174 #define CHECK_READ(A) (outb(CMOS_REG, (A)), inb(CMOS_DATA))
175 #define CHECK_WRITE(A,D) (outb(CMOS_REG, (A)), outb(CMOS_DATA, (D)))
177 #define CHECK_INIT(D); \
178 CHECK_WRITE(0x34, (D)); \
179 CHECK_WRITE(0x35, (D)); \
180 CHECK_WRITE(0x36, (D)); \
181 CHECK_WRITE(0x37, (D)); \
182 CHECK_WRITE(0x38, (D)); \
183 CHECK_WRITE(0x39, (D));
185 #define CHECK_PRINT(S); \
186 kprintf("%s: %d, %d, %d, %d, %d, %d\n", \
195 #else /* CHECK_POINTS */
197 #define CHECK_INIT(D)
198 #define CHECK_PRINT(S)
200 #endif /* CHECK_POINTS */
203 * Values to send to the POST hardware.
205 #define MP_BOOTADDRESS_POST 0x10
206 #define MP_PROBE_POST 0x11
207 #define MPTABLE_PASS1_POST 0x12
209 #define MP_START_POST 0x13
210 #define MP_ENABLE_POST 0x14
211 #define MPTABLE_PASS2_POST 0x15
213 #define START_ALL_APS_POST 0x16
214 #define INSTALL_AP_TRAMP_POST 0x17
215 #define START_AP_POST 0x18
217 #define MP_ANNOUNCE_POST 0x19
219 static int need_hyperthreading_fixup
;
220 static u_int logical_cpus
;
221 u_int logical_cpus_mask
;
223 /** XXX FIXME: where does this really belong, isa.h/isa.c perhaps? */
224 int current_postcode
;
226 /** XXX FIXME: what system files declare these??? */
227 extern struct region_descriptor r_gdt
, r_idt
;
229 int bsp_apic_ready
= 0; /* flags useability of BSP apic */
230 int mp_naps
; /* # of Applications processors */
231 int mp_nbusses
; /* # of busses */
233 int mp_napics
; /* # of IO APICs */
235 int boot_cpu_id
; /* designated BSP */
236 vm_offset_t cpu_apic_address
;
238 vm_offset_t io_apic_address
[NAPICID
]; /* NAPICID is more than enough */
239 u_int32_t
*io_apic_versions
;
243 u_int32_t cpu_apic_versions
[MAXCPU
];
245 extern int64_t tsc_offsets
[];
247 extern u_long ebda_addr
;
250 struct apic_intmapinfo int_to_apicintpin
[APIC_INTMAPSIZE
];
254 * APIC ID logical/physical mapping structures.
255 * We oversize these to simplify boot-time config.
257 int cpu_num_to_apic_id
[NAPICID
];
259 int io_num_to_apic_id
[NAPICID
];
261 int apic_id_to_logical
[NAPICID
];
263 /* AP uses this during bootstrap. Do not staticize. */
267 /* Hotwire a 0->4MB V==P mapping */
268 extern pt_entry_t
*KPTphys
;
271 * SMP page table page. Setup by locore to point to a page table
272 * page from which we allocate per-cpu privatespace areas io_apics,
276 #define IO_MAPPING_START_INDEX \
277 (SMP_MAXCPU * sizeof(struct privatespace) / PAGE_SIZE)
279 extern pt_entry_t
*SMPpt
;
280 static int SMPpt_alloc_index
= IO_MAPPING_START_INDEX
;
282 struct pcb stoppcbs
[MAXCPU
];
285 * Local data and functions.
288 static u_int boot_address
;
289 static u_int base_memory
;
290 static int mp_finish
;
292 static int search_for_sig(u_int32_t target
, int count
);
293 static void mp_enable(u_int boot_addr
);
295 static int mptable_probe(void);
296 static void mptable_hyperthread_fixup(u_int id_mask
);
297 static void mptable_pass1(struct mptable_pos
*);
298 static int mptable_pass2(struct mptable_pos
*);
299 static void default_mp_table(int type
);
300 static void fix_mp_table(void);
301 static void mptable_map(struct mptable_pos
*, vm_paddr_t
);
302 static void mptable_unmap(struct mptable_pos
*);
304 static void setup_apic_irq_mapping(void);
305 static int apic_int_is_bus_type(int intr
, int bus_type
);
307 static int start_all_aps(u_int boot_addr
);
308 static void install_ap_tramp(u_int boot_addr
);
309 static int start_ap(struct mdglobaldata
*gd
, u_int boot_addr
);
311 static cpumask_t smp_startup_mask
= 1; /* which cpus have been started */
312 cpumask_t smp_active_mask
= 1; /* which cpus are ready for IPIs etc? */
313 SYSCTL_INT(_machdep
, OID_AUTO
, smp_active
, CTLFLAG_RD
, &smp_active_mask
, 0, "");
316 * Calculate usable address in base memory for AP trampoline code.
319 mp_bootaddress(u_int basemem
)
321 POSTCODE(MP_BOOTADDRESS_POST
);
323 base_memory
= basemem
;
325 boot_address
= base_memory
& ~0xfff; /* round down to 4k boundary */
326 if ((base_memory
- boot_address
) < bootMP_size
)
327 boot_address
-= 4096; /* not enough, lower by 4k */
334 * 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 (ebda_addr
!= 0) {
352 /* search first 1K of EBDA */
353 target
= (u_int32_t
)ebda_addr
;
354 if ((x
= search_for_sig(target
, 1024 / 4)) > 0)
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)) > 0)
363 /* search the BIOS */
364 target
= (u_int32_t
)BIOS_BASE
;
365 if ((x
= search_for_sig(target
, BIOS_COUNT
)) > 0)
374 * Startup the SMP processors.
379 POSTCODE(MP_START_POST
);
380 mp_enable(boot_address
);
385 * Print various information about the SMP system hardware and setup.
392 POSTCODE(MP_ANNOUNCE_POST
);
394 kprintf("DragonFly/MP: Multiprocessor motherboard\n");
395 kprintf(" cpu0 (BSP): apic id: %2d", CPU_TO_ID(0));
396 kprintf(", version: 0x%08x", cpu_apic_versions
[0]);
397 kprintf(", at 0x%08x\n", cpu_apic_address
);
398 for (x
= 1; x
<= mp_naps
; ++x
) {
399 kprintf(" cpu%d (AP): apic id: %2d", x
, CPU_TO_ID(x
));
400 kprintf(", version: 0x%08x", cpu_apic_versions
[x
]);
401 kprintf(", at 0x%08x\n", cpu_apic_address
);
405 for (x
= 0; x
< mp_napics
; ++x
) {
406 kprintf(" io%d (APIC): apic id: %2d", x
, IO_TO_ID(x
));
407 kprintf(", version: 0x%08x", io_apic_versions
[x
]);
408 kprintf(", at 0x%08x\n", io_apic_address
[x
]);
411 kprintf(" Warning: APIC I/O disabled\n");
416 * AP cpu's call this to sync up protected mode.
418 * WARNING! We must ensure that the cpu is sufficiently initialized to
419 * be able to use to the FP for our optimized bzero/bcopy code before
420 * we enter more mainstream C code.
422 * WARNING! %fs is not set up on entry. This routine sets up %fs.
428 int x
, myid
= bootAP
;
430 struct mdglobaldata
*md
;
431 struct privatespace
*ps
;
433 ps
= &CPU_prvspace
[myid
];
435 gdt_segs
[GPRIV_SEL
].ssd_base
= (int)ps
;
436 gdt_segs
[GPROC0_SEL
].ssd_base
=
437 (int) &ps
->mdglobaldata
.gd_common_tss
;
438 ps
->mdglobaldata
.mi
.gd_prvspace
= ps
;
440 for (x
= 0; x
< NGDT
; x
++) {
441 ssdtosd(&gdt_segs
[x
], &gdt
[myid
* NGDT
+ x
].sd
);
444 r_gdt
.rd_limit
= NGDT
* sizeof(gdt
[0]) - 1;
445 r_gdt
.rd_base
= (int) &gdt
[myid
* NGDT
];
446 lgdt(&r_gdt
); /* does magic intra-segment return */
451 mdcpu
->gd_currentldt
= _default_ldt
;
453 gsel_tss
= GSEL(GPROC0_SEL
, SEL_KPL
);
454 gdt
[myid
* NGDT
+ GPROC0_SEL
].sd
.sd_type
= SDT_SYS386TSS
;
456 md
= mdcpu
; /* loaded through %fs:0 (mdglobaldata.mi.gd_prvspace)*/
458 md
->gd_common_tss
.tss_esp0
= 0; /* not used until after switch */
459 md
->gd_common_tss
.tss_ss0
= GSEL(GDATA_SEL
, SEL_KPL
);
460 md
->gd_common_tss
.tss_ioopt
= (sizeof md
->gd_common_tss
) << 16;
461 md
->gd_tss_gdt
= &gdt
[myid
* NGDT
+ GPROC0_SEL
].sd
;
462 md
->gd_common_tssd
= *md
->gd_tss_gdt
;
466 * Set to a known state:
467 * Set by mpboot.s: CR0_PG, CR0_PE
468 * Set by cpu_setregs: CR0_NE, CR0_MP, CR0_TS, CR0_WP, CR0_AM
471 cr0
&= ~(CR0_CD
| CR0_NW
| CR0_EM
);
473 pmap_set_opt(); /* PSE/4MB pages, etc */
475 /* set up CPU registers and state */
478 /* set up FPU state on the AP */
479 npxinit(__INITIAL_NPXCW__
);
481 /* set up SSE registers */
485 /*******************************************************************
486 * local functions and data
490 * start the SMP system
493 mp_enable(u_int boot_addr
)
500 vm_paddr_t mpfps_paddr
;
501 struct mptable_pos mpt
;
503 POSTCODE(MP_ENABLE_POST
);
505 mpfps_paddr
= mptable_probe();
506 if (mpfps_paddr
== 0)
507 panic("mp_enable: mptable_probe failed\n");
509 mptable_map(&mpt
, mpfps_paddr
);
512 * We can safely map physical memory into SMPpt after
513 * mptable_pass1() completes.
517 if (cpu_apic_address
== 0)
518 panic("mp_enable: no local apic!\n");
520 /* examine the MP table for needed info, uses physical addresses */
521 x
= mptable_pass2(&mpt
);
525 /* local apic is mapped on last page */
526 SMPpt
[NPTEPG
- 1] = (pt_entry_t
)(PG_V
| PG_RW
| PG_N
|
527 pmap_get_pgeflag() | (cpu_apic_address
& PG_FRAME
));
529 /* can't process default configs till the CPU APIC is pmapped */
533 /* post scan cleanup */
538 setup_apic_irq_mapping();
540 /* fill the LOGICAL io_apic_versions table */
541 for (apic
= 0; apic
< mp_napics
; ++apic
) {
542 ux
= io_apic_read(apic
, IOAPIC_VER
);
543 io_apic_versions
[apic
] = ux
;
544 io_apic_set_id(apic
, IO_TO_ID(apic
));
547 /* program each IO APIC in the system */
548 for (apic
= 0; apic
< mp_napics
; ++apic
)
549 if (io_apic_setup(apic
) < 0)
550 panic("IO APIC setup failure");
555 * These are required for SMP operation
558 /* install a 'Spurious INTerrupt' vector */
559 setidt(XSPURIOUSINT_OFFSET
, Xspuriousint
,
560 SDT_SYS386IGT
, SEL_KPL
, GSEL(GCODE_SEL
, SEL_KPL
));
562 /* install an inter-CPU IPI for TLB invalidation */
563 setidt(XINVLTLB_OFFSET
, Xinvltlb
,
564 SDT_SYS386IGT
, SEL_KPL
, GSEL(GCODE_SEL
, SEL_KPL
));
566 /* install an inter-CPU IPI for IPIQ messaging */
567 setidt(XIPIQ_OFFSET
, Xipiq
,
568 SDT_SYS386IGT
, SEL_KPL
, GSEL(GCODE_SEL
, SEL_KPL
));
570 /* install a timer vector */
571 setidt(XTIMER_OFFSET
, Xtimer
,
572 SDT_SYS386IGT
, SEL_KPL
, GSEL(GCODE_SEL
, SEL_KPL
));
574 /* install an inter-CPU IPI for CPU stop/restart */
575 setidt(XCPUSTOP_OFFSET
, Xcpustop
,
576 SDT_SYS386IGT
, SEL_KPL
, GSEL(GCODE_SEL
, SEL_KPL
));
578 /* start each Application Processor */
579 start_all_aps(boot_addr
);
584 * look for the MP spec signature
587 /* string defined by the Intel MP Spec as identifying the MP table */
588 #define MP_SIG 0x5f504d5f /* _MP_ */
589 #define NEXT(X) ((X) += 4)
591 search_for_sig(u_int32_t target
, int count
)
597 KKASSERT(target
!= 0);
599 map_size
= count
* sizeof(u_int32_t
);
600 addr
= pmap_mapdev((vm_paddr_t
)target
, map_size
);
603 for (x
= 0; x
< count
; NEXT(x
)) {
604 if (addr
[x
] == MP_SIG
) {
605 /* make array index a byte index */
606 ret
= target
+ (x
* sizeof(u_int32_t
));
611 pmap_unmapdev((vm_offset_t
)addr
, map_size
);
616 static basetable_entry basetable_entry_types
[] =
618 {0, 20, "Processor"},
625 typedef struct BUSDATA
{
627 enum busTypes bus_type
;
630 typedef struct INTDATA
{
640 typedef struct BUSTYPENAME
{
645 static bus_type_name bus_type_table
[] =
651 {UNKNOWN_BUSTYPE
, "---"},
654 {UNKNOWN_BUSTYPE
, "---"},
655 {UNKNOWN_BUSTYPE
, "---"},
656 {UNKNOWN_BUSTYPE
, "---"},
657 {UNKNOWN_BUSTYPE
, "---"},
658 {UNKNOWN_BUSTYPE
, "---"},
660 {UNKNOWN_BUSTYPE
, "---"},
661 {UNKNOWN_BUSTYPE
, "---"},
662 {UNKNOWN_BUSTYPE
, "---"},
663 {UNKNOWN_BUSTYPE
, "---"},
665 {UNKNOWN_BUSTYPE
, "---"}
667 /* from MP spec v1.4, table 5-1 */
668 static int default_data
[7][5] =
670 /* nbus, id0, type0, id1, type1 */
671 {1, 0, ISA
, 255, 255},
672 {1, 0, EISA
, 255, 255},
673 {1, 0, EISA
, 255, 255},
674 {1, 0, MCA
, 255, 255},
676 {2, 0, EISA
, 1, PCI
},
682 static bus_datum
*bus_data
;
685 /* the IO INT data, one entry per possible APIC INTerrupt */
686 static io_int
*io_apic_ints
;
690 static int processor_entry (proc_entry_ptr entry
, int cpu
);
691 static int bus_entry (bus_entry_ptr entry
, int bus
);
693 static int io_apic_entry (io_apic_entry_ptr entry
, int apic
);
694 static int int_entry (int_entry_ptr entry
, int intr
);
696 static int lookup_bus_type (char *name
);
700 * 1st pass on motherboard's Intel MP specification table.
706 * cpu_apic_address (common to all CPUs)
714 mptable_pass1(struct mptable_pos
*mpt
)
727 POSTCODE(MPTABLE_PASS1_POST
);
730 KKASSERT(fps
!= NULL
);
733 /* clear various tables */
734 for (x
= 0; x
< NAPICID
; ++x
) {
735 io_apic_address
[x
] = ~0; /* IO APIC address table */
739 /* init everything to empty */
748 /* check for use of 'default' configuration */
749 if (fps
->mpfb1
!= 0) {
750 /* use default addresses */
751 cpu_apic_address
= DEFAULT_APIC_BASE
;
753 io_apic_address
[0] = DEFAULT_IO_APIC_BASE
;
756 /* fill in with defaults */
757 mp_naps
= 2; /* includes BSP */
758 mp_nbusses
= default_data
[fps
->mpfb1
- 1][0];
767 panic("MP Configuration Table Header MISSING!");
769 cpu_apic_address
= (vm_offset_t
) cth
->apic_address
;
771 /* walk the table, recording info of interest */
772 totalSize
= cth
->base_table_length
- sizeof(struct MPCTH
);
773 position
= (u_char
*) cth
+ sizeof(struct MPCTH
);
774 count
= cth
->entry_count
;
777 switch (type
= *(u_char
*) position
) {
778 case 0: /* processor_entry */
779 if (((proc_entry_ptr
)position
)->cpu_flags
780 & PROCENTRY_FLAG_EN
) {
783 ((proc_entry_ptr
)position
)->apic_id
;
786 case 1: /* bus_entry */
789 case 2: /* io_apic_entry */
791 if (((io_apic_entry_ptr
)position
)->apic_flags
792 & IOAPICENTRY_FLAG_EN
)
793 io_apic_address
[mp_napics
++] =
794 (vm_offset_t
)((io_apic_entry_ptr
)
795 position
)->apic_address
;
798 case 3: /* int_entry */
803 case 4: /* int_entry */
806 panic("mpfps Base Table HOSED!");
810 totalSize
-= basetable_entry_types
[type
].length
;
811 position
= (uint8_t *)position
+
812 basetable_entry_types
[type
].length
;
816 /* qualify the numbers */
817 if (mp_naps
> MAXCPU
) {
818 kprintf("Warning: only using %d of %d available CPUs!\n",
823 /* See if we need to fixup HT logical CPUs. */
824 mptable_hyperthread_fixup(id_mask
);
826 --mp_naps
; /* subtract the BSP */
831 * 2nd pass on motherboard's Intel MP specification table.
835 * ID_TO_IO(N), phy APIC ID to log CPU/IO table
836 * CPU_TO_ID(N), logical CPU to APIC ID table
837 * IO_TO_ID(N), logical IO to APIC ID table
842 mptable_pass2(struct mptable_pos
*mpt
)
844 struct PROCENTRY proc
;
852 int apic
, bus
, cpu
, intr
;
855 POSTCODE(MPTABLE_PASS2_POST
);
858 KKASSERT(fps
!= NULL
);
860 /* Initialize fake proc entry for use with HT fixup. */
861 bzero(&proc
, sizeof(proc
));
863 proc
.cpu_flags
= PROCENTRY_FLAG_EN
;
866 MALLOC(io_apic_versions
, u_int32_t
*, sizeof(u_int32_t
) * mp_napics
,
868 MALLOC(ioapic
, volatile ioapic_t
**, sizeof(ioapic_t
*) * mp_napics
,
869 M_DEVBUF
, M_WAITOK
| M_ZERO
);
870 MALLOC(io_apic_ints
, io_int
*, sizeof(io_int
) * (nintrs
+ FIXUP_EXTRA_APIC_INTS
),
873 MALLOC(bus_data
, bus_datum
*, sizeof(bus_datum
) * mp_nbusses
,
877 for (i
= 0; i
< mp_napics
; i
++) {
878 ioapic
[i
] = permanent_io_mapping(io_apic_address
[i
]);
882 /* clear various tables */
883 for (x
= 0; x
< NAPICID
; ++x
) {
884 CPU_TO_ID(x
) = -1; /* logical CPU to APIC ID table */
886 ID_TO_IO(x
) = -1; /* phy APIC ID to log CPU/IO table */
887 IO_TO_ID(x
) = -1; /* logical IO to APIC ID table */
891 /* clear bus data table */
892 for (x
= 0; x
< mp_nbusses
; ++x
)
893 bus_data
[x
].bus_id
= 0xff;
896 /* clear IO APIC INT table */
897 for (x
= 0; x
< (nintrs
+ 1); ++x
) {
898 io_apic_ints
[x
].int_type
= 0xff;
899 io_apic_ints
[x
].int_vector
= 0xff;
903 /* setup the cpu/apic mapping arrays */
906 /* record whether PIC or virtual-wire mode */
907 machintr_setvar_simple(MACHINTR_VAR_IMCR_PRESENT
, fps
->mpfb2
& 0x80);
909 /* check for use of 'default' configuration */
911 return fps
->mpfb1
; /* return default configuration type */
915 panic("MP Configuration Table Header MISSING!");
917 /* walk the table, recording info of interest */
918 totalSize
= cth
->base_table_length
- sizeof(struct MPCTH
);
919 position
= (u_char
*) cth
+ sizeof(struct MPCTH
);
920 count
= cth
->entry_count
;
921 apic
= bus
= intr
= 0;
922 cpu
= 1; /* pre-count the BSP */
925 switch (type
= *(u_char
*) position
) {
927 if (processor_entry(position
, cpu
))
930 if (need_hyperthreading_fixup
) {
932 * Create fake mptable processor entries
933 * and feed them to processor_entry() to
934 * enumerate the logical CPUs.
936 proc
.apic_id
= ((proc_entry_ptr
)position
)->apic_id
;
937 for (i
= 1; i
< logical_cpus
; i
++) {
939 processor_entry(&proc
, cpu
);
940 logical_cpus_mask
|= (1 << cpu
);
946 if (bus_entry(position
, bus
))
951 if (io_apic_entry(position
, apic
))
957 if (int_entry(position
, intr
))
962 /* int_entry(position); */
965 panic("mpfps Base Table HOSED!");
969 totalSize
-= basetable_entry_types
[type
].length
;
970 position
= (uint8_t *)position
+ basetable_entry_types
[type
].length
;
973 if (boot_cpu_id
== -1)
974 panic("NO BSP found!");
976 /* report fact that its NOT a default configuration */
981 * Check if we should perform a hyperthreading "fix-up" to
982 * enumerate any logical CPU's that aren't already listed
985 * XXX: We assume that all of the physical CPUs in the
986 * system have the same number of logical CPUs.
988 * XXX: We assume that APIC ID's are allocated such that
989 * the APIC ID's for a physical processor are aligned
990 * with the number of logical CPU's in the processor.
993 mptable_hyperthread_fixup(u_int id_mask
)
997 /* Nothing to do if there is no HTT support. */
998 if ((cpu_feature
& CPUID_HTT
) == 0)
1000 logical_cpus
= (cpu_procinfo
& CPUID_HTT_CORES
) >> 16;
1001 if (logical_cpus
<= 1)
1005 * For each APIC ID of a CPU that is set in the mask,
1006 * scan the other candidate APIC ID's for this
1007 * physical processor. If any of those ID's are
1008 * already in the table, then kill the fixup.
1010 for (id
= 0; id
<= MAXCPU
; id
++) {
1011 if ((id_mask
& 1 << id
) == 0)
1013 /* First, make sure we are on a logical_cpus boundary. */
1014 if (id
% logical_cpus
!= 0)
1016 for (i
= id
+ 1; i
< id
+ logical_cpus
; i
++)
1017 if ((id_mask
& 1 << i
) != 0)
1022 * Ok, the ID's checked out, so enable the fixup. We have to fixup
1023 * mp_naps right now.
1025 need_hyperthreading_fixup
= 1;
1026 mp_naps
*= logical_cpus
;
1030 mptable_map(struct mptable_pos
*mpt
, vm_paddr_t mpfps_paddr
)
1034 vm_size_t cth_mapsz
= 0;
1036 fps
= pmap_mapdev(mpfps_paddr
, sizeof(*fps
));
1037 if (fps
->pap
!= 0) {
1039 * Map configuration table header to get
1040 * the base table size
1042 cth
= pmap_mapdev(fps
->pap
, sizeof(*cth
));
1043 cth_mapsz
= cth
->base_table_length
;
1044 pmap_unmapdev((vm_offset_t
)cth
, sizeof(*cth
));
1047 * Map the base table
1049 cth
= pmap_mapdev(fps
->pap
, cth_mapsz
);
1054 mpt
->mp_cth_mapsz
= cth_mapsz
;
1058 mptable_unmap(struct mptable_pos
*mpt
)
1060 if (mpt
->mp_cth
!= NULL
) {
1061 pmap_unmapdev((vm_offset_t
)mpt
->mp_cth
, mpt
->mp_cth_mapsz
);
1063 mpt
->mp_cth_mapsz
= 0;
1065 if (mpt
->mp_fps
!= NULL
) {
1066 pmap_unmapdev((vm_offset_t
)mpt
->mp_fps
, sizeof(*mpt
->mp_fps
));
1074 assign_apic_irq(int apic
, int intpin
, int irq
)
1078 if (int_to_apicintpin
[irq
].ioapic
!= -1)
1079 panic("assign_apic_irq: inconsistent table");
1081 int_to_apicintpin
[irq
].ioapic
= apic
;
1082 int_to_apicintpin
[irq
].int_pin
= intpin
;
1083 int_to_apicintpin
[irq
].apic_address
= ioapic
[apic
];
1084 int_to_apicintpin
[irq
].redirindex
= IOAPIC_REDTBL
+ 2 * intpin
;
1086 for (x
= 0; x
< nintrs
; x
++) {
1087 if ((io_apic_ints
[x
].int_type
== 0 ||
1088 io_apic_ints
[x
].int_type
== 3) &&
1089 io_apic_ints
[x
].int_vector
== 0xff &&
1090 io_apic_ints
[x
].dst_apic_id
== IO_TO_ID(apic
) &&
1091 io_apic_ints
[x
].dst_apic_int
== intpin
)
1092 io_apic_ints
[x
].int_vector
= irq
;
1097 revoke_apic_irq(int irq
)
1103 if (int_to_apicintpin
[irq
].ioapic
== -1)
1104 panic("revoke_apic_irq: inconsistent table");
1106 oldapic
= int_to_apicintpin
[irq
].ioapic
;
1107 oldintpin
= int_to_apicintpin
[irq
].int_pin
;
1109 int_to_apicintpin
[irq
].ioapic
= -1;
1110 int_to_apicintpin
[irq
].int_pin
= 0;
1111 int_to_apicintpin
[irq
].apic_address
= NULL
;
1112 int_to_apicintpin
[irq
].redirindex
= 0;
1114 for (x
= 0; x
< nintrs
; x
++) {
1115 if ((io_apic_ints
[x
].int_type
== 0 ||
1116 io_apic_ints
[x
].int_type
== 3) &&
1117 io_apic_ints
[x
].int_vector
!= 0xff &&
1118 io_apic_ints
[x
].dst_apic_id
== IO_TO_ID(oldapic
) &&
1119 io_apic_ints
[x
].dst_apic_int
== oldintpin
)
1120 io_apic_ints
[x
].int_vector
= 0xff;
1128 allocate_apic_irq(int intr
)
1134 if (io_apic_ints
[intr
].int_vector
!= 0xff)
1135 return; /* Interrupt handler already assigned */
1137 if (io_apic_ints
[intr
].int_type
!= 0 &&
1138 (io_apic_ints
[intr
].int_type
!= 3 ||
1139 (io_apic_ints
[intr
].dst_apic_id
== IO_TO_ID(0) &&
1140 io_apic_ints
[intr
].dst_apic_int
== 0)))
1141 return; /* Not INT or ExtInt on != (0, 0) */
1144 while (irq
< APIC_INTMAPSIZE
&&
1145 int_to_apicintpin
[irq
].ioapic
!= -1)
1148 if (irq
>= APIC_INTMAPSIZE
)
1149 return; /* No free interrupt handlers */
1151 apic
= ID_TO_IO(io_apic_ints
[intr
].dst_apic_id
);
1152 intpin
= io_apic_ints
[intr
].dst_apic_int
;
1154 assign_apic_irq(apic
, intpin
, irq
);
1155 io_apic_setup_intpin(apic
, intpin
);
1160 swap_apic_id(int apic
, int oldid
, int newid
)
1167 return; /* Nothing to do */
1169 kprintf("Changing APIC ID for IO APIC #%d from %d to %d in MP table\n",
1170 apic
, oldid
, newid
);
1172 /* Swap physical APIC IDs in interrupt entries */
1173 for (x
= 0; x
< nintrs
; x
++) {
1174 if (io_apic_ints
[x
].dst_apic_id
== oldid
)
1175 io_apic_ints
[x
].dst_apic_id
= newid
;
1176 else if (io_apic_ints
[x
].dst_apic_id
== newid
)
1177 io_apic_ints
[x
].dst_apic_id
= oldid
;
1180 /* Swap physical APIC IDs in IO_TO_ID mappings */
1181 for (oapic
= 0; oapic
< mp_napics
; oapic
++)
1182 if (IO_TO_ID(oapic
) == newid
)
1185 if (oapic
< mp_napics
) {
1186 kprintf("Changing APIC ID for IO APIC #%d from "
1187 "%d to %d in MP table\n",
1188 oapic
, newid
, oldid
);
1189 IO_TO_ID(oapic
) = oldid
;
1191 IO_TO_ID(apic
) = newid
;
1196 fix_id_to_io_mapping(void)
1200 for (x
= 0; x
< NAPICID
; x
++)
1203 for (x
= 0; x
<= mp_naps
; x
++)
1204 if (CPU_TO_ID(x
) < NAPICID
)
1205 ID_TO_IO(CPU_TO_ID(x
)) = x
;
1207 for (x
= 0; x
< mp_napics
; x
++)
1208 if (IO_TO_ID(x
) < NAPICID
)
1209 ID_TO_IO(IO_TO_ID(x
)) = x
;
1214 first_free_apic_id(void)
1218 for (freeid
= 0; freeid
< NAPICID
; freeid
++) {
1219 for (x
= 0; x
<= mp_naps
; x
++)
1220 if (CPU_TO_ID(x
) == freeid
)
1224 for (x
= 0; x
< mp_napics
; x
++)
1225 if (IO_TO_ID(x
) == freeid
)
1236 io_apic_id_acceptable(int apic
, int id
)
1238 int cpu
; /* Logical CPU number */
1239 int oapic
; /* Logical IO APIC number for other IO APIC */
1242 return 0; /* Out of range */
1244 for (cpu
= 0; cpu
<= mp_naps
; cpu
++)
1245 if (CPU_TO_ID(cpu
) == id
)
1246 return 0; /* Conflict with CPU */
1248 for (oapic
= 0; oapic
< mp_napics
&& oapic
< apic
; oapic
++)
1249 if (IO_TO_ID(oapic
) == id
)
1250 return 0; /* Conflict with other APIC */
1252 return 1; /* ID is acceptable for IO APIC */
1257 io_apic_find_int_entry(int apic
, int pin
)
1261 /* search each of the possible INTerrupt sources */
1262 for (x
= 0; x
< nintrs
; ++x
) {
1263 if ((apic
== ID_TO_IO(io_apic_ints
[x
].dst_apic_id
)) &&
1264 (pin
== io_apic_ints
[x
].dst_apic_int
))
1265 return (&io_apic_ints
[x
]);
1273 * parse an Intel MP specification table
1281 int apic
; /* IO APIC unit number */
1282 int freeid
; /* Free physical APIC ID */
1283 int physid
; /* Current physical IO APIC ID */
1286 int bus_0
= 0; /* Stop GCC warning */
1287 int bus_pci
= 0; /* Stop GCC warning */
1291 * Fix mis-numbering of the PCI bus and its INT entries if the BIOS
1292 * did it wrong. The MP spec says that when more than 1 PCI bus
1293 * exists the BIOS must begin with bus entries for the PCI bus and use
1294 * actual PCI bus numbering. This implies that when only 1 PCI bus
1295 * exists the BIOS can choose to ignore this ordering, and indeed many
1296 * MP motherboards do ignore it. This causes a problem when the PCI
1297 * sub-system makes requests of the MP sub-system based on PCI bus
1298 * numbers. So here we look for the situation and renumber the
1299 * busses and associated INTs in an effort to "make it right".
1302 /* find bus 0, PCI bus, count the number of PCI busses */
1303 for (num_pci_bus
= 0, x
= 0; x
< mp_nbusses
; ++x
) {
1304 if (bus_data
[x
].bus_id
== 0) {
1307 if (bus_data
[x
].bus_type
== PCI
) {
1313 * bus_0 == slot of bus with ID of 0
1314 * bus_pci == slot of last PCI bus encountered
1317 /* check the 1 PCI bus case for sanity */
1318 /* if it is number 0 all is well */
1319 if (num_pci_bus
== 1 &&
1320 bus_data
[bus_pci
].bus_id
!= 0) {
1322 /* mis-numbered, swap with whichever bus uses slot 0 */
1324 /* swap the bus entry types */
1325 bus_data
[bus_pci
].bus_type
= bus_data
[bus_0
].bus_type
;
1326 bus_data
[bus_0
].bus_type
= PCI
;
1329 /* swap each relavant INTerrupt entry */
1330 id
= bus_data
[bus_pci
].bus_id
;
1331 for (x
= 0; x
< nintrs
; ++x
) {
1332 if (io_apic_ints
[x
].src_bus_id
== id
) {
1333 io_apic_ints
[x
].src_bus_id
= 0;
1335 else if (io_apic_ints
[x
].src_bus_id
== 0) {
1336 io_apic_ints
[x
].src_bus_id
= id
;
1343 /* Assign IO APIC IDs.
1345 * First try the existing ID. If a conflict is detected, try
1346 * the ID in the MP table. If a conflict is still detected, find
1349 * We cannot use the ID_TO_IO table before all conflicts has been
1350 * resolved and the table has been corrected.
1352 for (apic
= 0; apic
< mp_napics
; ++apic
) { /* For all IO APICs */
1354 /* First try to use the value set by the BIOS */
1355 physid
= io_apic_get_id(apic
);
1356 if (io_apic_id_acceptable(apic
, physid
)) {
1357 if (IO_TO_ID(apic
) != physid
)
1358 swap_apic_id(apic
, IO_TO_ID(apic
), physid
);
1362 /* Then check if the value in the MP table is acceptable */
1363 if (io_apic_id_acceptable(apic
, IO_TO_ID(apic
)))
1366 /* Last resort, find a free APIC ID and use it */
1367 freeid
= first_free_apic_id();
1368 if (freeid
>= NAPICID
)
1369 panic("No free physical APIC IDs found");
1371 if (io_apic_id_acceptable(apic
, freeid
)) {
1372 swap_apic_id(apic
, IO_TO_ID(apic
), freeid
);
1375 panic("Free physical APIC ID not usable");
1377 fix_id_to_io_mapping();
1381 /* detect and fix broken Compaq MP table */
1382 if (apic_int_type(0, 0) == -1) {
1383 kprintf("APIC_IO: MP table broken: 8259->APIC entry missing!\n");
1384 io_apic_ints
[nintrs
].int_type
= 3; /* ExtInt */
1385 io_apic_ints
[nintrs
].int_vector
= 0xff; /* Unassigned */
1386 /* XXX fixme, set src bus id etc, but it doesn't seem to hurt */
1387 io_apic_ints
[nintrs
].dst_apic_id
= IO_TO_ID(0);
1388 io_apic_ints
[nintrs
].dst_apic_int
= 0; /* Pin 0 */
1390 } else if (apic_int_type(0, 0) == 0) {
1391 kprintf("APIC_IO: MP table broken: ExtINT entry corrupt!\n");
1392 for (x
= 0; x
< nintrs
; ++x
)
1393 if ((0 == ID_TO_IO(io_apic_ints
[x
].dst_apic_id
)) &&
1394 (0 == io_apic_ints
[x
].dst_apic_int
)) {
1395 io_apic_ints
[x
].int_type
= 3;
1396 io_apic_ints
[x
].int_vector
= 0xff;
1402 * Fix missing IRQ 15 when IRQ 14 is an ISA interrupt. IDE
1403 * controllers universally come in pairs. If IRQ 14 is specified
1404 * as an ISA interrupt, then IRQ 15 had better be too.
1406 * [ Shuttle XPC / AMD Athlon X2 ]
1407 * The MPTable is missing an entry for IRQ 15. Note that the
1408 * ACPI table has an entry for both 14 and 15.
1410 if (apic_int_type(0, 14) == 0 && apic_int_type(0, 15) == -1) {
1411 kprintf("APIC_IO: MP table broken: IRQ 15 not ISA when IRQ 14 is!\n");
1412 io14
= io_apic_find_int_entry(0, 14);
1413 io_apic_ints
[nintrs
] = *io14
;
1414 io_apic_ints
[nintrs
].src_bus_irq
= 15;
1415 io_apic_ints
[nintrs
].dst_apic_int
= 15;
1423 /* Assign low level interrupt handlers */
1425 setup_apic_irq_mapping(void)
1431 for (x
= 0; x
< APIC_INTMAPSIZE
; x
++) {
1432 int_to_apicintpin
[x
].ioapic
= -1;
1433 int_to_apicintpin
[x
].int_pin
= 0;
1434 int_to_apicintpin
[x
].apic_address
= NULL
;
1435 int_to_apicintpin
[x
].redirindex
= 0;
1438 /* First assign ISA/EISA interrupts */
1439 for (x
= 0; x
< nintrs
; x
++) {
1440 int_vector
= io_apic_ints
[x
].src_bus_irq
;
1441 if (int_vector
< APIC_INTMAPSIZE
&&
1442 io_apic_ints
[x
].int_vector
== 0xff &&
1443 int_to_apicintpin
[int_vector
].ioapic
== -1 &&
1444 (apic_int_is_bus_type(x
, ISA
) ||
1445 apic_int_is_bus_type(x
, EISA
)) &&
1446 io_apic_ints
[x
].int_type
== 0) {
1447 assign_apic_irq(ID_TO_IO(io_apic_ints
[x
].dst_apic_id
),
1448 io_apic_ints
[x
].dst_apic_int
,
1453 /* Assign ExtInt entry if no ISA/EISA interrupt 0 entry */
1454 for (x
= 0; x
< nintrs
; x
++) {
1455 if (io_apic_ints
[x
].dst_apic_int
== 0 &&
1456 io_apic_ints
[x
].dst_apic_id
== IO_TO_ID(0) &&
1457 io_apic_ints
[x
].int_vector
== 0xff &&
1458 int_to_apicintpin
[0].ioapic
== -1 &&
1459 io_apic_ints
[x
].int_type
== 3) {
1460 assign_apic_irq(0, 0, 0);
1464 /* PCI interrupt assignment is deferred */
1470 processor_entry(proc_entry_ptr entry
, int cpu
)
1472 /* check for usability */
1473 if (!(entry
->cpu_flags
& PROCENTRY_FLAG_EN
))
1476 if(entry
->apic_id
>= NAPICID
)
1477 panic("CPU APIC ID out of range (0..%d)", NAPICID
- 1);
1478 /* check for BSP flag */
1479 if (entry
->cpu_flags
& PROCENTRY_FLAG_BP
) {
1480 boot_cpu_id
= entry
->apic_id
;
1481 CPU_TO_ID(0) = entry
->apic_id
;
1482 ID_TO_CPU(entry
->apic_id
) = 0;
1483 return 0; /* its already been counted */
1486 /* add another AP to list, if less than max number of CPUs */
1487 else if (cpu
< MAXCPU
) {
1488 CPU_TO_ID(cpu
) = entry
->apic_id
;
1489 ID_TO_CPU(entry
->apic_id
) = cpu
;
1498 bus_entry(bus_entry_ptr entry
, int bus
)
1503 /* encode the name into an index */
1504 for (x
= 0; x
< 6; ++x
) {
1505 if ((c
= entry
->bus_type
[x
]) == ' ')
1511 if ((x
= lookup_bus_type(name
)) == UNKNOWN_BUSTYPE
)
1512 panic("unknown bus type: '%s'", name
);
1514 bus_data
[bus
].bus_id
= entry
->bus_id
;
1515 bus_data
[bus
].bus_type
= x
;
1523 io_apic_entry(io_apic_entry_ptr entry
, int apic
)
1525 if (!(entry
->apic_flags
& IOAPICENTRY_FLAG_EN
))
1528 IO_TO_ID(apic
) = entry
->apic_id
;
1529 if (entry
->apic_id
< NAPICID
)
1530 ID_TO_IO(entry
->apic_id
) = apic
;
1538 lookup_bus_type(char *name
)
1542 for (x
= 0; x
< MAX_BUSTYPE
; ++x
)
1543 if (strcmp(bus_type_table
[x
].name
, name
) == 0)
1544 return bus_type_table
[x
].type
;
1546 return UNKNOWN_BUSTYPE
;
1552 int_entry(int_entry_ptr entry
, int intr
)
1556 io_apic_ints
[intr
].int_type
= entry
->int_type
;
1557 io_apic_ints
[intr
].int_flags
= entry
->int_flags
;
1558 io_apic_ints
[intr
].src_bus_id
= entry
->src_bus_id
;
1559 io_apic_ints
[intr
].src_bus_irq
= entry
->src_bus_irq
;
1560 if (entry
->dst_apic_id
== 255) {
1561 /* This signal goes to all IO APICS. Select an IO APIC
1562 with sufficient number of interrupt pins */
1563 for (apic
= 0; apic
< mp_napics
; apic
++)
1564 if (((io_apic_read(apic
, IOAPIC_VER
) &
1565 IOART_VER_MAXREDIR
) >> MAXREDIRSHIFT
) >=
1566 entry
->dst_apic_int
)
1568 if (apic
< mp_napics
)
1569 io_apic_ints
[intr
].dst_apic_id
= IO_TO_ID(apic
);
1571 io_apic_ints
[intr
].dst_apic_id
= entry
->dst_apic_id
;
1573 io_apic_ints
[intr
].dst_apic_id
= entry
->dst_apic_id
;
1574 io_apic_ints
[intr
].dst_apic_int
= entry
->dst_apic_int
;
1580 apic_int_is_bus_type(int intr
, int bus_type
)
1584 for (bus
= 0; bus
< mp_nbusses
; ++bus
)
1585 if ((bus_data
[bus
].bus_id
== io_apic_ints
[intr
].src_bus_id
)
1586 && ((int) bus_data
[bus
].bus_type
== bus_type
))
1593 * Given a traditional ISA INT mask, return an APIC mask.
1596 isa_apic_mask(u_int isa_mask
)
1601 #if defined(SKIP_IRQ15_REDIRECT)
1602 if (isa_mask
== (1 << 15)) {
1603 kprintf("skipping ISA IRQ15 redirect\n");
1606 #endif /* SKIP_IRQ15_REDIRECT */
1608 isa_irq
= ffs(isa_mask
); /* find its bit position */
1609 if (isa_irq
== 0) /* doesn't exist */
1611 --isa_irq
; /* make it zero based */
1613 apic_pin
= isa_apic_irq(isa_irq
); /* look for APIC connection */
1617 return (1 << apic_pin
); /* convert pin# to a mask */
1621 * Determine which APIC pin an ISA/EISA INT is attached to.
1623 #define INTTYPE(I) (io_apic_ints[(I)].int_type)
1624 #define INTPIN(I) (io_apic_ints[(I)].dst_apic_int)
1625 #define INTIRQ(I) (io_apic_ints[(I)].int_vector)
1626 #define INTAPIC(I) (ID_TO_IO(io_apic_ints[(I)].dst_apic_id))
1628 #define SRCBUSIRQ(I) (io_apic_ints[(I)].src_bus_irq)
1630 isa_apic_irq(int isa_irq
)
1634 for (intr
= 0; intr
< nintrs
; ++intr
) { /* check each record */
1635 if (INTTYPE(intr
) == 0) { /* standard INT */
1636 if (SRCBUSIRQ(intr
) == isa_irq
) {
1637 if (apic_int_is_bus_type(intr
, ISA
) ||
1638 apic_int_is_bus_type(intr
, EISA
)) {
1639 if (INTIRQ(intr
) == 0xff)
1640 return -1; /* unassigned */
1641 return INTIRQ(intr
); /* found */
1646 return -1; /* NOT found */
1651 * Determine which APIC pin a PCI INT is attached to.
1653 #define SRCBUSID(I) (io_apic_ints[(I)].src_bus_id)
1654 #define SRCBUSDEVICE(I) ((io_apic_ints[(I)].src_bus_irq >> 2) & 0x1f)
1655 #define SRCBUSLINE(I) (io_apic_ints[(I)].src_bus_irq & 0x03)
1657 pci_apic_irq(int pciBus
, int pciDevice
, int pciInt
)
1661 --pciInt
; /* zero based */
1663 for (intr
= 0; intr
< nintrs
; ++intr
) { /* check each record */
1664 if ((INTTYPE(intr
) == 0) /* standard INT */
1665 && (SRCBUSID(intr
) == pciBus
)
1666 && (SRCBUSDEVICE(intr
) == pciDevice
)
1667 && (SRCBUSLINE(intr
) == pciInt
)) { /* a candidate IRQ */
1668 if (apic_int_is_bus_type(intr
, PCI
)) {
1669 if (INTIRQ(intr
) == 0xff)
1670 allocate_apic_irq(intr
);
1671 if (INTIRQ(intr
) == 0xff)
1672 return -1; /* unassigned */
1673 return INTIRQ(intr
); /* exact match */
1678 return -1; /* NOT found */
1682 next_apic_irq(int irq
)
1689 for (intr
= 0; intr
< nintrs
; intr
++) {
1690 if (INTIRQ(intr
) != irq
|| INTTYPE(intr
) != 0)
1692 bus
= SRCBUSID(intr
);
1693 bustype
= apic_bus_type(bus
);
1694 if (bustype
!= ISA
&&
1700 if (intr
>= nintrs
) {
1703 for (ointr
= intr
+ 1; ointr
< nintrs
; ointr
++) {
1704 if (INTTYPE(ointr
) != 0)
1706 if (bus
!= SRCBUSID(ointr
))
1708 if (bustype
== PCI
) {
1709 if (SRCBUSDEVICE(intr
) != SRCBUSDEVICE(ointr
))
1711 if (SRCBUSLINE(intr
) != SRCBUSLINE(ointr
))
1714 if (bustype
== ISA
|| bustype
== EISA
) {
1715 if (SRCBUSIRQ(intr
) != SRCBUSIRQ(ointr
))
1718 if (INTPIN(intr
) == INTPIN(ointr
))
1722 if (ointr
>= nintrs
) {
1725 return INTIRQ(ointr
);
1740 * Reprogram the MB chipset to NOT redirect an ISA INTerrupt.
1743 * Exactly what this means is unclear at this point. It is a solution
1744 * for motherboards that redirect the MBIRQ0 pin. Generically a motherboard
1745 * could route any of the ISA INTs to upper (>15) IRQ values. But most would
1746 * NOT be redirected via MBIRQ0, thus "undirect()ing" them would NOT be an
1750 undirect_isa_irq(int rirq
)
1754 kprintf("Freeing redirected ISA irq %d.\n", rirq
);
1755 /** FIXME: tickle the MB redirector chip */
1759 kprintf("Freeing (NOT implemented) redirected ISA irq %d.\n", rirq
);
1766 * Reprogram the MB chipset to NOT redirect a PCI INTerrupt
1769 undirect_pci_irq(int rirq
)
1773 kprintf("Freeing redirected PCI irq %d.\n", rirq
);
1775 /** FIXME: tickle the MB redirector chip */
1779 kprintf("Freeing (NOT implemented) redirected PCI irq %d.\n",
1787 * given a bus ID, return:
1788 * the bus type if found
1792 apic_bus_type(int id
)
1796 for (x
= 0; x
< mp_nbusses
; ++x
)
1797 if (bus_data
[x
].bus_id
== id
)
1798 return bus_data
[x
].bus_type
;
1806 * given a LOGICAL APIC# and pin#, return:
1807 * the associated src bus ID if found
1811 apic_src_bus_id(int apic
, int pin
)
1815 /* search each of the possible INTerrupt sources */
1816 for (x
= 0; x
< nintrs
; ++x
)
1817 if ((apic
== ID_TO_IO(io_apic_ints
[x
].dst_apic_id
)) &&
1818 (pin
== io_apic_ints
[x
].dst_apic_int
))
1819 return (io_apic_ints
[x
].src_bus_id
);
1821 return -1; /* NOT found */
1825 * given a LOGICAL APIC# and pin#, return:
1826 * the associated src bus IRQ if found
1830 apic_src_bus_irq(int apic
, int pin
)
1834 for (x
= 0; x
< nintrs
; x
++)
1835 if ((apic
== ID_TO_IO(io_apic_ints
[x
].dst_apic_id
)) &&
1836 (pin
== io_apic_ints
[x
].dst_apic_int
))
1837 return (io_apic_ints
[x
].src_bus_irq
);
1839 return -1; /* NOT found */
1844 * given a LOGICAL APIC# and pin#, return:
1845 * the associated INTerrupt type if found
1849 apic_int_type(int apic
, int pin
)
1853 /* search each of the possible INTerrupt sources */
1854 for (x
= 0; x
< nintrs
; ++x
) {
1855 if ((apic
== ID_TO_IO(io_apic_ints
[x
].dst_apic_id
)) &&
1856 (pin
== io_apic_ints
[x
].dst_apic_int
))
1857 return (io_apic_ints
[x
].int_type
);
1859 return -1; /* NOT found */
1863 * Return the IRQ associated with an APIC pin
1866 apic_irq(int apic
, int pin
)
1871 for (x
= 0; x
< nintrs
; ++x
) {
1872 if ((apic
== ID_TO_IO(io_apic_ints
[x
].dst_apic_id
)) &&
1873 (pin
== io_apic_ints
[x
].dst_apic_int
)) {
1874 res
= io_apic_ints
[x
].int_vector
;
1877 if (apic
!= int_to_apicintpin
[res
].ioapic
)
1878 panic("apic_irq: inconsistent table %d/%d", apic
, int_to_apicintpin
[res
].ioapic
);
1879 if (pin
!= int_to_apicintpin
[res
].int_pin
)
1880 panic("apic_irq inconsistent table (2)");
1889 * given a LOGICAL APIC# and pin#, return:
1890 * the associated trigger mode if found
1894 apic_trigger(int apic
, int pin
)
1898 /* search each of the possible INTerrupt sources */
1899 for (x
= 0; x
< nintrs
; ++x
)
1900 if ((apic
== ID_TO_IO(io_apic_ints
[x
].dst_apic_id
)) &&
1901 (pin
== io_apic_ints
[x
].dst_apic_int
))
1902 return ((io_apic_ints
[x
].int_flags
>> 2) & 0x03);
1904 return -1; /* NOT found */
1909 * given a LOGICAL APIC# and pin#, return:
1910 * the associated 'active' level if found
1914 apic_polarity(int apic
, int pin
)
1918 /* search each of the possible INTerrupt sources */
1919 for (x
= 0; x
< nintrs
; ++x
)
1920 if ((apic
== ID_TO_IO(io_apic_ints
[x
].dst_apic_id
)) &&
1921 (pin
== io_apic_ints
[x
].dst_apic_int
))
1922 return (io_apic_ints
[x
].int_flags
& 0x03);
1924 return -1; /* NOT found */
1930 * set data according to MP defaults
1931 * FIXME: probably not complete yet...
1934 default_mp_table(int type
)
1937 #if defined(APIC_IO)
1940 #endif /* APIC_IO */
1943 kprintf(" MP default config type: %d\n", type
);
1946 kprintf(" bus: ISA, APIC: 82489DX\n");
1949 kprintf(" bus: EISA, APIC: 82489DX\n");
1952 kprintf(" bus: EISA, APIC: 82489DX\n");
1955 kprintf(" bus: MCA, APIC: 82489DX\n");
1958 kprintf(" bus: ISA+PCI, APIC: Integrated\n");
1961 kprintf(" bus: EISA+PCI, APIC: Integrated\n");
1964 kprintf(" bus: MCA+PCI, APIC: Integrated\n");
1967 kprintf(" future type\n");
1973 boot_cpu_id
= (lapic
.id
& APIC_ID_MASK
) >> 24;
1974 ap_cpu_id
= (boot_cpu_id
== 0) ? 1 : 0;
1977 CPU_TO_ID(0) = boot_cpu_id
;
1978 ID_TO_CPU(boot_cpu_id
) = 0;
1980 /* one and only AP */
1981 CPU_TO_ID(1) = ap_cpu_id
;
1982 ID_TO_CPU(ap_cpu_id
) = 1;
1984 #if defined(APIC_IO)
1985 /* one and only IO APIC */
1986 io_apic_id
= (io_apic_read(0, IOAPIC_ID
) & APIC_ID_MASK
) >> 24;
1989 * sanity check, refer to MP spec section 3.6.6, last paragraph
1990 * necessary as some hardware isn't properly setting up the IO APIC
1992 #if defined(REALLY_ANAL_IOAPICID_VALUE)
1993 if (io_apic_id
!= 2) {
1995 if ((io_apic_id
== 0) || (io_apic_id
== 1) || (io_apic_id
== 15)) {
1996 #endif /* REALLY_ANAL_IOAPICID_VALUE */
1997 io_apic_set_id(0, 2);
2000 IO_TO_ID(0) = io_apic_id
;
2001 ID_TO_IO(io_apic_id
) = 0;
2002 #endif /* APIC_IO */
2004 /* fill out bus entries */
2013 bus_data
[0].bus_id
= default_data
[type
- 1][1];
2014 bus_data
[0].bus_type
= default_data
[type
- 1][2];
2015 bus_data
[1].bus_id
= default_data
[type
- 1][3];
2016 bus_data
[1].bus_type
= default_data
[type
- 1][4];
2019 /* case 4: case 7: MCA NOT supported */
2020 default: /* illegal/reserved */
2021 panic("BAD default MP config: %d", type
);
2025 #if defined(APIC_IO)
2026 /* general cases from MP v1.4, table 5-2 */
2027 for (pin
= 0; pin
< 16; ++pin
) {
2028 io_apic_ints
[pin
].int_type
= 0;
2029 io_apic_ints
[pin
].int_flags
= 0x05; /* edge/active-hi */
2030 io_apic_ints
[pin
].src_bus_id
= 0;
2031 io_apic_ints
[pin
].src_bus_irq
= pin
; /* IRQ2 caught below */
2032 io_apic_ints
[pin
].dst_apic_id
= io_apic_id
;
2033 io_apic_ints
[pin
].dst_apic_int
= pin
; /* 1-to-1 */
2036 /* special cases from MP v1.4, table 5-2 */
2038 io_apic_ints
[2].int_type
= 0xff; /* N/C */
2039 io_apic_ints
[13].int_type
= 0xff; /* N/C */
2040 #if !defined(APIC_MIXED_MODE)
2042 panic("sorry, can't support type 2 default yet");
2043 #endif /* APIC_MIXED_MODE */
2046 io_apic_ints
[2].src_bus_irq
= 0; /* ISA IRQ0 is on APIC INT 2 */
2049 io_apic_ints
[0].int_type
= 0xff; /* N/C */
2051 io_apic_ints
[0].int_type
= 3; /* vectored 8259 */
2052 #endif /* APIC_IO */
2056 * Map a physical memory address representing I/O into KVA. The I/O
2057 * block is assumed not to cross a page boundary.
2060 permanent_io_mapping(vm_paddr_t pa
)
2066 KKASSERT(pa
< 0x100000000LL
);
2068 pgeflag
= 0; /* not used for SMP yet */
2071 * If the requested physical address has already been incidently
2072 * mapped, just use the existing mapping. Otherwise create a new
2075 for (i
= IO_MAPPING_START_INDEX
; i
< SMPpt_alloc_index
; ++i
) {
2076 if (((vm_offset_t
)SMPpt
[i
] & PG_FRAME
) ==
2077 ((vm_offset_t
)pa
& PG_FRAME
)) {
2081 if (i
== SMPpt_alloc_index
) {
2082 if (i
== NPTEPG
- 2) {
2083 panic("permanent_io_mapping: We ran out of space"
2086 SMPpt
[i
] = (pt_entry_t
)(PG_V
| PG_RW
| pgeflag
|
2087 ((vm_offset_t
)pa
& PG_FRAME
));
2088 ++SMPpt_alloc_index
;
2090 vaddr
= (vm_offset_t
)CPU_prvspace
+ (i
* PAGE_SIZE
) +
2091 ((vm_offset_t
)pa
& PAGE_MASK
);
2092 return ((void *)vaddr
);
2096 * start each AP in our list
2099 start_all_aps(u_int boot_addr
)
2103 u_char mpbiosreason
;
2104 u_long mpbioswarmvec
;
2105 struct mdglobaldata
*gd
;
2106 struct privatespace
*ps
;
2110 POSTCODE(START_ALL_APS_POST
);
2112 /* Initialize BSP's local APIC */
2113 apic_initialize(TRUE
);
2116 /* install the AP 1st level boot code */
2117 install_ap_tramp(boot_addr
);
2120 /* save the current value of the warm-start vector */
2121 mpbioswarmvec
= *((u_long
*) WARMBOOT_OFF
);
2122 outb(CMOS_REG
, BIOS_RESET
);
2123 mpbiosreason
= inb(CMOS_DATA
);
2125 /* set up temporary P==V mapping for AP boot */
2126 /* XXX this is a hack, we should boot the AP on its own stack/PTD */
2127 kptbase
= (uintptr_t)(void *)KPTphys
;
2128 for (x
= 0; x
< NKPT
; x
++) {
2129 PTD
[x
] = (pd_entry_t
)(PG_V
| PG_RW
|
2130 ((kptbase
+ x
* PAGE_SIZE
) & PG_FRAME
));
2135 for (x
= 1; x
<= mp_naps
; ++x
) {
2137 /* This is a bit verbose, it will go away soon. */
2139 /* first page of AP's private space */
2140 pg
= x
* i386_btop(sizeof(struct privatespace
));
2142 /* allocate new private data page(s) */
2143 gd
= (struct mdglobaldata
*)kmem_alloc(&kernel_map
,
2144 MDGLOBALDATA_BASEALLOC_SIZE
);
2145 /* wire it into the private page table page */
2146 for (i
= 0; i
< MDGLOBALDATA_BASEALLOC_SIZE
; i
+= PAGE_SIZE
) {
2147 SMPpt
[pg
+ i
/ PAGE_SIZE
] = (pt_entry_t
)
2148 (PG_V
| PG_RW
| vtophys_pte((char *)gd
+ i
));
2150 pg
+= MDGLOBALDATA_BASEALLOC_PAGES
;
2152 SMPpt
[pg
+ 0] = 0; /* *gd_CMAP1 */
2153 SMPpt
[pg
+ 1] = 0; /* *gd_CMAP2 */
2154 SMPpt
[pg
+ 2] = 0; /* *gd_CMAP3 */
2155 SMPpt
[pg
+ 3] = 0; /* *gd_PMAP1 */
2157 /* allocate and set up an idle stack data page */
2158 stack
= (char *)kmem_alloc(&kernel_map
, UPAGES
*PAGE_SIZE
);
2159 for (i
= 0; i
< UPAGES
; i
++) {
2160 SMPpt
[pg
+ 4 + i
] = (pt_entry_t
)
2161 (PG_V
| PG_RW
| vtophys_pte(PAGE_SIZE
* i
+ stack
));
2164 gd
= &CPU_prvspace
[x
].mdglobaldata
; /* official location */
2165 bzero(gd
, sizeof(*gd
));
2166 gd
->mi
.gd_prvspace
= ps
= &CPU_prvspace
[x
];
2168 /* prime data page for it to use */
2169 mi_gdinit(&gd
->mi
, x
);
2171 gd
->gd_CMAP1
= &SMPpt
[pg
+ 0];
2172 gd
->gd_CMAP2
= &SMPpt
[pg
+ 1];
2173 gd
->gd_CMAP3
= &SMPpt
[pg
+ 2];
2174 gd
->gd_PMAP1
= &SMPpt
[pg
+ 3];
2175 gd
->gd_CADDR1
= ps
->CPAGE1
;
2176 gd
->gd_CADDR2
= ps
->CPAGE2
;
2177 gd
->gd_CADDR3
= ps
->CPAGE3
;
2178 gd
->gd_PADDR1
= (unsigned *)ps
->PPAGE1
;
2179 gd
->mi
.gd_ipiq
= (void *)kmem_alloc(&kernel_map
, sizeof(lwkt_ipiq
) * (mp_naps
+ 1));
2180 bzero(gd
->mi
.gd_ipiq
, sizeof(lwkt_ipiq
) * (mp_naps
+ 1));
2182 /* setup a vector to our boot code */
2183 *((volatile u_short
*) WARMBOOT_OFF
) = WARMBOOT_TARGET
;
2184 *((volatile u_short
*) WARMBOOT_SEG
) = (boot_addr
>> 4);
2185 outb(CMOS_REG
, BIOS_RESET
);
2186 outb(CMOS_DATA
, BIOS_WARM
); /* 'warm-start' */
2189 * Setup the AP boot stack
2191 bootSTK
= &ps
->idlestack
[UPAGES
*PAGE_SIZE
/2];
2194 /* attempt to start the Application Processor */
2195 CHECK_INIT(99); /* setup checkpoints */
2196 if (!start_ap(gd
, boot_addr
)) {
2197 kprintf("AP #%d (PHY# %d) failed!\n", x
, CPU_TO_ID(x
));
2198 CHECK_PRINT("trace"); /* show checkpoints */
2199 /* better panic as the AP may be running loose */
2200 kprintf("panic y/n? [y] ");
2201 if (cngetc() != 'n')
2204 CHECK_PRINT("trace"); /* show checkpoints */
2206 /* record its version info */
2207 cpu_apic_versions
[x
] = cpu_apic_versions
[0];
2210 /* set ncpus to 1 + highest logical cpu. Not all may have come up */
2213 /* ncpus2 -- ncpus rounded down to the nearest power of 2 */
2214 for (shift
= 0; (1 << shift
) <= ncpus
; ++shift
)
2217 ncpus2_shift
= shift
;
2218 ncpus2
= 1 << shift
;
2219 ncpus2_mask
= ncpus2
- 1;
2221 /* ncpus_fit -- ncpus rounded up to the nearest power of 2 */
2222 if ((1 << shift
) < ncpus
)
2224 ncpus_fit
= 1 << shift
;
2225 ncpus_fit_mask
= ncpus_fit
- 1;
2227 /* build our map of 'other' CPUs */
2228 mycpu
->gd_other_cpus
= smp_startup_mask
& ~(1 << mycpu
->gd_cpuid
);
2229 mycpu
->gd_ipiq
= (void *)kmem_alloc(&kernel_map
, sizeof(lwkt_ipiq
) * ncpus
);
2230 bzero(mycpu
->gd_ipiq
, sizeof(lwkt_ipiq
) * ncpus
);
2232 /* fill in our (BSP) APIC version */
2233 cpu_apic_versions
[0] = lapic
.version
;
2235 /* restore the warmstart vector */
2236 *(u_long
*) WARMBOOT_OFF
= mpbioswarmvec
;
2237 outb(CMOS_REG
, BIOS_RESET
);
2238 outb(CMOS_DATA
, mpbiosreason
);
2241 * NOTE! The idlestack for the BSP was setup by locore. Finish
2242 * up, clean out the P==V mapping we did earlier.
2244 for (x
= 0; x
< NKPT
; x
++)
2248 /* number of APs actually started */
2254 * load the 1st level AP boot code into base memory.
2257 /* targets for relocation */
2258 extern void bigJump(void);
2259 extern void bootCodeSeg(void);
2260 extern void bootDataSeg(void);
2261 extern void MPentry(void);
2262 extern u_int MP_GDT
;
2263 extern u_int mp_gdtbase
;
2266 install_ap_tramp(u_int boot_addr
)
2269 int size
= *(int *) ((u_long
) & bootMP_size
);
2270 u_char
*src
= (u_char
*) ((u_long
) bootMP
);
2271 u_char
*dst
= (u_char
*) boot_addr
+ KERNBASE
;
2272 u_int boot_base
= (u_int
) bootMP
;
2277 POSTCODE(INSTALL_AP_TRAMP_POST
);
2279 for (x
= 0; x
< size
; ++x
)
2283 * modify addresses in code we just moved to basemem. unfortunately we
2284 * need fairly detailed info about mpboot.s for this to work. changes
2285 * to mpboot.s might require changes here.
2288 /* boot code is located in KERNEL space */
2289 dst
= (u_char
*) boot_addr
+ KERNBASE
;
2291 /* modify the lgdt arg */
2292 dst32
= (u_int32_t
*) (dst
+ ((u_int
) & mp_gdtbase
- boot_base
));
2293 *dst32
= boot_addr
+ ((u_int
) & MP_GDT
- boot_base
);
2295 /* modify the ljmp target for MPentry() */
2296 dst32
= (u_int32_t
*) (dst
+ ((u_int
) bigJump
- boot_base
) + 1);
2297 *dst32
= ((u_int
) MPentry
- KERNBASE
);
2299 /* modify the target for boot code segment */
2300 dst16
= (u_int16_t
*) (dst
+ ((u_int
) bootCodeSeg
- boot_base
));
2301 dst8
= (u_int8_t
*) (dst16
+ 1);
2302 *dst16
= (u_int
) boot_addr
& 0xffff;
2303 *dst8
= ((u_int
) boot_addr
>> 16) & 0xff;
2305 /* modify the target for boot data segment */
2306 dst16
= (u_int16_t
*) (dst
+ ((u_int
) bootDataSeg
- boot_base
));
2307 dst8
= (u_int8_t
*) (dst16
+ 1);
2308 *dst16
= (u_int
) boot_addr
& 0xffff;
2309 *dst8
= ((u_int
) boot_addr
>> 16) & 0xff;
2314 * this function starts the AP (application processor) identified
2315 * by the APIC ID 'physicalCpu'. It does quite a "song and dance"
2316 * to accomplish this. This is necessary because of the nuances
2317 * of the different hardware we might encounter. It ain't pretty,
2318 * but it seems to work.
2320 * NOTE: eventually an AP gets to ap_init(), which is called just
2321 * before the AP goes into the LWKT scheduler's idle loop.
2324 start_ap(struct mdglobaldata
*gd
, u_int boot_addr
)
2328 u_long icr_lo
, icr_hi
;
2330 POSTCODE(START_AP_POST
);
2332 /* get the PHYSICAL APIC ID# */
2333 physical_cpu
= CPU_TO_ID(gd
->mi
.gd_cpuid
);
2335 /* calculate the vector */
2336 vector
= (boot_addr
>> 12) & 0xff;
2338 /* Make sure the target cpu sees everything */
2342 * first we do an INIT/RESET IPI this INIT IPI might be run, reseting
2343 * and running the target CPU. OR this INIT IPI might be latched (P5
2344 * bug), CPU waiting for STARTUP IPI. OR this INIT IPI might be
2348 /* setup the address for the target AP */
2349 icr_hi
= lapic
.icr_hi
& ~APIC_ID_MASK
;
2350 icr_hi
|= (physical_cpu
<< 24);
2351 lapic
.icr_hi
= icr_hi
;
2353 /* do an INIT IPI: assert RESET */
2354 icr_lo
= lapic
.icr_lo
& 0xfff00000;
2355 lapic
.icr_lo
= icr_lo
| 0x0000c500;
2357 /* wait for pending status end */
2358 while (lapic
.icr_lo
& APIC_DELSTAT_MASK
)
2361 /* do an INIT IPI: deassert RESET */
2362 lapic
.icr_lo
= icr_lo
| 0x00008500;
2364 /* wait for pending status end */
2365 u_sleep(10000); /* wait ~10mS */
2366 while (lapic
.icr_lo
& APIC_DELSTAT_MASK
)
2370 * next we do a STARTUP IPI: the previous INIT IPI might still be
2371 * latched, (P5 bug) this 1st STARTUP would then terminate
2372 * immediately, and the previously started INIT IPI would continue. OR
2373 * the previous INIT IPI has already run. and this STARTUP IPI will
2374 * run. OR the previous INIT IPI was ignored. and this STARTUP IPI
2378 /* do a STARTUP IPI */
2379 lapic
.icr_lo
= icr_lo
| 0x00000600 | vector
;
2380 while (lapic
.icr_lo
& APIC_DELSTAT_MASK
)
2382 u_sleep(200); /* wait ~200uS */
2385 * finally we do a 2nd STARTUP IPI: this 2nd STARTUP IPI should run IF
2386 * the previous STARTUP IPI was cancelled by a latched INIT IPI. OR
2387 * this STARTUP IPI will be ignored, as only ONE STARTUP IPI is
2388 * recognized after hardware RESET or INIT IPI.
2391 lapic
.icr_lo
= icr_lo
| 0x00000600 | vector
;
2392 while (lapic
.icr_lo
& APIC_DELSTAT_MASK
)
2394 u_sleep(200); /* wait ~200uS */
2396 /* wait for it to start, see ap_init() */
2397 set_apic_timer(5000000);/* == 5 seconds */
2398 while (read_apic_timer()) {
2399 if (smp_startup_mask
& (1 << gd
->mi
.gd_cpuid
))
2400 return 1; /* return SUCCESS */
2402 return 0; /* return FAILURE */
2407 * Lazy flush the TLB on all other CPU's. DEPRECATED.
2409 * If for some reason we were unable to start all cpus we cannot safely
2410 * use broadcast IPIs.
2416 if (smp_startup_mask
== smp_active_mask
) {
2417 all_but_self_ipi(XINVLTLB_OFFSET
);
2419 selected_apic_ipi(smp_active_mask
, XINVLTLB_OFFSET
,
2420 APIC_DELMODE_FIXED
);
2426 * When called the executing CPU will send an IPI to all other CPUs
2427 * requesting that they halt execution.
2429 * Usually (but not necessarily) called with 'other_cpus' as its arg.
2431 * - Signals all CPUs in map to stop.
2432 * - Waits for each to stop.
2439 * XXX FIXME: this is not MP-safe, needs a lock to prevent multiple CPUs
2440 * from executing at same time.
2443 stop_cpus(u_int map
)
2445 map
&= smp_active_mask
;
2447 /* send the Xcpustop IPI to all CPUs in map */
2448 selected_apic_ipi(map
, XCPUSTOP_OFFSET
, APIC_DELMODE_FIXED
);
2450 while ((stopped_cpus
& map
) != map
)
2458 * Called by a CPU to restart stopped CPUs.
2460 * Usually (but not necessarily) called with 'stopped_cpus' as its arg.
2462 * - Signals all CPUs in map to restart.
2463 * - Waits for each to restart.
2471 restart_cpus(u_int map
)
2473 /* signal other cpus to restart */
2474 started_cpus
= map
& smp_active_mask
;
2476 while ((stopped_cpus
& map
) != 0) /* wait for each to clear its bit */
2483 * This is called once the mpboot code has gotten us properly relocated
2484 * and the MMU turned on, etc. ap_init() is actually the idle thread,
2485 * and when it returns the scheduler will call the real cpu_idle() main
2486 * loop for the idlethread. Interrupts are disabled on entry and should
2487 * remain disabled at return.
2495 * Adjust smp_startup_mask to signal the BSP that we have started
2496 * up successfully. Note that we do not yet hold the BGL. The BSP
2497 * is waiting for our signal.
2499 * We can't set our bit in smp_active_mask yet because we are holding
2500 * interrupts physically disabled and remote cpus could deadlock
2501 * trying to send us an IPI.
2503 smp_startup_mask
|= 1 << mycpu
->gd_cpuid
;
2507 * Interlock for finalization. Wait until mp_finish is non-zero,
2508 * then get the MP lock.
2510 * Note: We are in a critical section.
2512 * Note: We have to synchronize td_mpcount to our desired MP state
2513 * before calling cpu_try_mplock().
2515 * Note: we are the idle thread, we can only spin.
2517 * Note: The load fence is memory volatile and prevents the compiler
2518 * from improperly caching mp_finish, and the cpu from improperly
2521 while (mp_finish
== 0)
2523 ++curthread
->td_mpcount
;
2524 while (cpu_try_mplock() == 0)
2527 if (cpu_feature
& CPUID_TSC
) {
2529 * The BSP is constantly updating tsc0_offset, figure out the
2530 * relative difference to synchronize ktrdump.
2532 tsc_offsets
[mycpu
->gd_cpuid
] = rdtsc() - tsc0_offset
;
2535 /* BSP may have changed PTD while we're waiting for the lock */
2538 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
2542 /* Build our map of 'other' CPUs. */
2543 mycpu
->gd_other_cpus
= smp_startup_mask
& ~(1 << mycpu
->gd_cpuid
);
2545 kprintf("SMP: AP CPU #%d Launched!\n", mycpu
->gd_cpuid
);
2547 /* A quick check from sanity claus */
2548 apic_id
= (apic_id_to_logical
[(lapic
.id
& 0x0f000000) >> 24]);
2549 if (mycpu
->gd_cpuid
!= apic_id
) {
2550 kprintf("SMP: cpuid = %d\n", mycpu
->gd_cpuid
);
2551 kprintf("SMP: apic_id = %d\n", apic_id
);
2552 kprintf("PTD[MPPTDI] = %p\n", (void *)PTD
[MPPTDI
]);
2553 panic("cpuid mismatch! boom!!");
2556 /* Initialize AP's local APIC for irq's */
2557 apic_initialize(FALSE
);
2559 /* Set memory range attributes for this CPU to match the BSP */
2560 mem_range_AP_init();
2563 * Once we go active we must process any IPIQ messages that may
2564 * have been queued, because no actual IPI will occur until we
2565 * set our bit in the smp_active_mask. If we don't the IPI
2566 * message interlock could be left set which would also prevent
2569 * The idle loop doesn't expect the BGL to be held and while
2570 * lwkt_switch() normally cleans things up this is a special case
2571 * because we returning almost directly into the idle loop.
2573 * The idle thread is never placed on the runq, make sure
2574 * nothing we've done put it there.
2576 KKASSERT(curthread
->td_mpcount
== 1);
2577 smp_active_mask
|= 1 << mycpu
->gd_cpuid
;
2580 * Enable interrupts here. idle_restore will also do it, but
2581 * doing it here lets us clean up any strays that got posted to
2582 * the CPU during the AP boot while we are still in a critical
2585 __asm
__volatile("sti; pause; pause"::);
2586 mdcpu
->gd_fpending
= 0;
2587 mdcpu
->gd_ipending
= 0;
2589 initclocks_pcpu(); /* clock interrupts (via IPIs) */
2590 lwkt_process_ipiq();
2593 * Releasing the mp lock lets the BSP finish up the SMP init
2596 KKASSERT((curthread
->td_flags
& TDF_RUNQ
) == 0);
2600 * Get SMP fully working before we start initializing devices.
2608 kprintf("Finish MP startup\n");
2609 if (cpu_feature
& CPUID_TSC
)
2610 tsc0_offset
= rdtsc();
2613 while (smp_active_mask
!= smp_startup_mask
) {
2615 if (cpu_feature
& CPUID_TSC
)
2616 tsc0_offset
= rdtsc();
2618 while (try_mplock() == 0)
2621 kprintf("Active CPU Mask: %08x\n", smp_active_mask
);
2624 SYSINIT(finishsmp
, SI_BOOT2_FINISH_SMP
, SI_ORDER_FIRST
, ap_finish
, NULL
)
2627 cpu_send_ipiq(int dcpu
)
2629 if ((1 << dcpu
) & smp_active_mask
)
2630 single_apic_ipi(dcpu
, XIPIQ_OFFSET
, APIC_DELMODE_FIXED
);
2633 #if 0 /* single_apic_ipi_passive() not working yet */
2635 * Returns 0 on failure, 1 on success
2638 cpu_send_ipiq_passive(int dcpu
)
2641 if ((1 << dcpu
) & smp_active_mask
) {
2642 r
= single_apic_ipi_passive(dcpu
, XIPIQ_OFFSET
,
2643 APIC_DELMODE_FIXED
);