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 1.59 2008/01/05 14:02:41 swildner 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 {
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];
277 * Local data and functions.
280 static int mp_capable;
281 static u_int boot_address;
282 static u_int base_memory;
283 static int mp_finish;
285 static mpfps_t mpfps;
286 static int search_for_sig(u_int32_t target, int count);
287 static void mp_enable(u_int boot_addr);
289 static void mptable_hyperthread_fixup(u_int id_mask);
290 static void mptable_pass1(void);
291 static int mptable_pass2(void);
292 static void default_mp_table(int type);
293 static void fix_mp_table(void);
295 static void setup_apic_irq_mapping(void);
296 static int apic_int_is_bus_type(int intr, int bus_type);
298 static int start_all_aps(u_int boot_addr);
299 static void install_ap_tramp(u_int boot_addr);
300 static int start_ap(struct mdglobaldata *gd, u_int boot_addr);
302 static cpumask_t smp_startup_mask = 1; /* which cpus have been started */
303 cpumask_t smp_active_mask = 1; /* which cpus are ready for IPIs etc? */
304 SYSCTL_INT(_machdep, OID_AUTO, smp_active, CTLFLAG_RD, &smp_active_mask, 0, "");
307 * Calculate usable address in base memory for AP trampoline code.
310 mp_bootaddress(u_int basemem)
312 POSTCODE(MP_BOOTADDRESS_POST);
314 base_memory = basemem * 1024; /* convert to bytes */
316 boot_address = base_memory & ~0xfff; /* round down to 4k boundary */
317 if ((base_memory - boot_address) < bootMP_size)
318 boot_address -= 4096; /* not enough, lower by 4k */
325 * Look for an Intel MP spec table (ie, SMP capable hardware).
335 * Make sure our SMPpt[] page table is big enough to hold all the
338 KKASSERT(IO_MAPPING_START_INDEX < NPTEPG - 2);
340 POSTCODE(MP_PROBE_POST);
342 /* see if EBDA exists */
343 if ((segment = (u_long) * (u_short *) (KERNBASE + 0x40e)) != 0) {
344 /* search first 1K of EBDA */
345 target = (u_int32_t) (segment << 4);
346 if ((x = search_for_sig(target, 1024 / 4)) >= 0)
349 /* last 1K of base memory, effective 'top of base' passed in */
350 target = (u_int32_t) (base_memory - 0x400);
351 if ((x = search_for_sig(target, 1024 / 4)) >= 0)
355 /* search the BIOS */
356 target = (u_int32_t) BIOS_BASE;
357 if ((x = search_for_sig(target, BIOS_COUNT)) >= 0)
367 * Calculate needed resources. We can safely map physical
368 * memory into SMPpt after mptable_pass1() completes.
373 /* flag fact that we are running multiple processors */
380 * Startup the SMP processors.
385 POSTCODE(MP_START_POST);
387 /* look for MP capable motherboard */
389 mp_enable(boot_address);
391 panic("MP hardware not found!");
396 * Print various information about the SMP system hardware and setup.
403 POSTCODE(MP_ANNOUNCE_POST);
405 kprintf("DragonFly/MP: Multiprocessor motherboard\n");
406 kprintf(" cpu0 (BSP): apic id: %2d", CPU_TO_ID(0));
407 kprintf(", version: 0x%08x", cpu_apic_versions[0]);
408 kprintf(", at 0x%08x\n", cpu_apic_address);
409 for (x = 1; x <= mp_naps; ++x) {
410 kprintf(" cpu%d (AP): apic id: %2d", x, CPU_TO_ID(x));
411 kprintf(", version: 0x%08x", cpu_apic_versions[x]);
412 kprintf(", at 0x%08x\n", cpu_apic_address);
416 for (x = 0; x < mp_napics; ++x) {
417 kprintf(" io%d (APIC): apic id: %2d", x, IO_TO_ID(x));
418 kprintf(", version: 0x%08x", io_apic_versions[x]);
419 kprintf(", at 0x%08x\n", io_apic_address[x]);
422 kprintf(" Warning: APIC I/O disabled\n");
427 * AP cpu's call this to sync up protected mode.
429 * WARNING! We must ensure that the cpu is sufficiently initialized to
430 * be able to use to the FP for our optimized bzero/bcopy code before
431 * we enter more mainstream C code.
433 * WARNING! %fs is not set up on entry. This routine sets up %fs.
439 int x, myid = bootAP;
441 struct mdglobaldata *md;
442 struct privatespace *ps;
444 ps = &CPU_prvspace[myid];
446 gdt_segs[GPRIV_SEL].ssd_base = (int)ps;
447 gdt_segs[GPROC0_SEL].ssd_base =
448 (int) &ps->mdglobaldata.gd_common_tss;
449 ps->mdglobaldata.mi.gd_prvspace = ps;
451 for (x = 0; x < NGDT; x++) {
452 ssdtosd(&gdt_segs[x], &gdt[myid * NGDT + x].sd);
455 r_gdt.rd_limit = NGDT * sizeof(gdt[0]) - 1;
456 r_gdt.rd_base = (int) &gdt[myid * NGDT];
457 lgdt(&r_gdt); /* does magic intra-segment return */
462 mdcpu->gd_currentldt = _default_ldt;
464 gsel_tss = GSEL(GPROC0_SEL, SEL_KPL);
465 gdt[myid * NGDT + GPROC0_SEL].sd.sd_type = SDT_SYS386TSS;
467 md = mdcpu; /* loaded through %fs:0 (mdglobaldata.mi.gd_prvspace)*/
469 md->gd_common_tss.tss_esp0 = 0; /* not used until after switch */
470 md->gd_common_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
471 md->gd_common_tss.tss_ioopt = (sizeof md->gd_common_tss) << 16;
472 md->gd_tss_gdt = &gdt[myid * NGDT + GPROC0_SEL].sd;
473 md->gd_common_tssd = *md->gd_tss_gdt;
477 * Set to a known state:
478 * Set by mpboot.s: CR0_PG, CR0_PE
479 * Set by cpu_setregs: CR0_NE, CR0_MP, CR0_TS, CR0_WP, CR0_AM
482 cr0 &= ~(CR0_CD | CR0_NW | CR0_EM);
484 pmap_set_opt(); /* PSE/4MB pages, etc */
486 /* set up CPU registers and state */
489 /* set up FPU state on the AP */
490 npxinit(__INITIAL_NPXCW__);
492 /* set up SSE registers */
496 /*******************************************************************
497 * local functions and data
501 * start the SMP system
504 mp_enable(u_int boot_addr)
512 POSTCODE(MP_ENABLE_POST);
514 /* turn on 4MB of V == P addressing so we can get to MP table */
515 *(int *)PTD = PG_V | PG_RW | ((uintptr_t)(void *)KPTphys & PG_FRAME);
518 /* examine the MP table for needed info, uses physical addresses */
524 /* can't process default configs till the CPU APIC is pmapped */
528 /* post scan cleanup */
533 setup_apic_irq_mapping();
535 /* fill the LOGICAL io_apic_versions table */
536 for (apic = 0; apic < mp_napics; ++apic) {
537 ux = io_apic_read(apic, IOAPIC_VER);
538 io_apic_versions[apic] = ux;
539 io_apic_set_id(apic, IO_TO_ID(apic));
542 /* program each IO APIC in the system */
543 for (apic = 0; apic < mp_napics; ++apic)
544 if (io_apic_setup(apic) < 0)
545 panic("IO APIC setup failure");
550 * These are required for SMP operation
553 /* install a 'Spurious INTerrupt' vector */
554 setidt(XSPURIOUSINT_OFFSET, Xspuriousint,
555 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
557 /* install an inter-CPU IPI for TLB invalidation */
558 setidt(XINVLTLB_OFFSET, Xinvltlb,
559 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
561 /* install an inter-CPU IPI for IPIQ messaging */
562 setidt(XIPIQ_OFFSET, Xipiq,
563 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
565 /* install an inter-CPU IPI for CPU stop/restart */
566 setidt(XCPUSTOP_OFFSET, Xcpustop,
567 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
569 /* start each Application Processor */
570 start_all_aps(boot_addr);
575 * look for the MP spec signature
578 /* string defined by the Intel MP Spec as identifying the MP table */
579 #define MP_SIG 0x5f504d5f /* _MP_ */
580 #define NEXT(X) ((X) += 4)
582 search_for_sig(u_int32_t target, int count)
585 u_int32_t *addr = (u_int32_t *) (KERNBASE + target);
587 for (x = 0; x < count; NEXT(x))
588 if (addr[x] == MP_SIG)
589 /* make array index a byte index */
590 return (target + (x * sizeof(u_int32_t)));
596 static basetable_entry basetable_entry_types[] =
598 {0, 20, "Processor"},
605 typedef struct BUSDATA {
607 enum busTypes bus_type;
610 typedef struct INTDATA {
620 typedef struct BUSTYPENAME {
625 static bus_type_name bus_type_table[] =
631 {UNKNOWN_BUSTYPE, "---"},
634 {UNKNOWN_BUSTYPE, "---"},
635 {UNKNOWN_BUSTYPE, "---"},
636 {UNKNOWN_BUSTYPE, "---"},
637 {UNKNOWN_BUSTYPE, "---"},
638 {UNKNOWN_BUSTYPE, "---"},
640 {UNKNOWN_BUSTYPE, "---"},
641 {UNKNOWN_BUSTYPE, "---"},
642 {UNKNOWN_BUSTYPE, "---"},
643 {UNKNOWN_BUSTYPE, "---"},
645 {UNKNOWN_BUSTYPE, "---"}
647 /* from MP spec v1.4, table 5-1 */
648 static int default_data[7][5] =
650 /* nbus, id0, type0, id1, type1 */
651 {1, 0, ISA, 255, 255},
652 {1, 0, EISA, 255, 255},
653 {1, 0, EISA, 255, 255},
654 {1, 0, MCA, 255, 255},
656 {2, 0, EISA, 1, PCI},
662 static bus_datum *bus_data;
665 /* the IO INT data, one entry per possible APIC INTerrupt */
666 static io_int *io_apic_ints;
670 static int processor_entry (proc_entry_ptr entry, int cpu);
671 static int bus_entry (bus_entry_ptr entry, int bus);
673 static int io_apic_entry (io_apic_entry_ptr entry, int apic);
674 static int int_entry (int_entry_ptr entry, int intr);
676 static int lookup_bus_type (char *name);
680 * 1st pass on motherboard's Intel MP specification table.
686 * cpu_apic_address (common to all CPUs)
706 POSTCODE(MPTABLE_PASS1_POST);
709 /* clear various tables */
710 for (x = 0; x < NAPICID; ++x) {
711 io_apic_address[x] = ~0; /* IO APIC address table */
715 /* init everything to empty */
724 /* check for use of 'default' configuration */
725 if (mpfps->mpfb1 != 0) {
726 /* use default addresses */
727 cpu_apic_address = DEFAULT_APIC_BASE;
729 io_apic_address[0] = DEFAULT_IO_APIC_BASE;
732 /* fill in with defaults */
733 mp_naps = 2; /* includes BSP */
734 mp_nbusses = default_data[mpfps->mpfb1 - 1][0];
741 if ((cth = mpfps->pap) == 0)
742 panic("MP Configuration Table Header MISSING!");
744 cpu_apic_address = (vm_offset_t) cth->apic_address;
746 /* walk the table, recording info of interest */
747 totalSize = cth->base_table_length - sizeof(struct MPCTH);
748 position = (u_char *) cth + sizeof(struct MPCTH);
749 count = cth->entry_count;
752 switch (type = *(u_char *) position) {
753 case 0: /* processor_entry */
754 if (((proc_entry_ptr)position)->cpu_flags
755 & PROCENTRY_FLAG_EN) {
758 ((proc_entry_ptr)position)->apic_id;
761 case 1: /* bus_entry */
764 case 2: /* io_apic_entry */
766 if (((io_apic_entry_ptr)position)->apic_flags
767 & IOAPICENTRY_FLAG_EN)
768 io_apic_address[mp_napics++] =
769 (vm_offset_t)((io_apic_entry_ptr)
770 position)->apic_address;
773 case 3: /* int_entry */
778 case 4: /* int_entry */
781 panic("mpfps Base Table HOSED!");
785 totalSize -= basetable_entry_types[type].length;
786 position = (uint8_t *)position +
787 basetable_entry_types[type].length;
791 /* qualify the numbers */
792 if (mp_naps > MAXCPU) {
793 kprintf("Warning: only using %d of %d available CPUs!\n",
798 /* See if we need to fixup HT logical CPUs. */
799 mptable_hyperthread_fixup(id_mask);
803 * This is also used as a counter while starting the APs.
807 --mp_naps; /* subtract the BSP */
812 * 2nd pass on motherboard's Intel MP specification table.
816 * ID_TO_IO(N), phy APIC ID to log CPU/IO table
817 * CPU_TO_ID(N), logical CPU to APIC ID table
818 * IO_TO_ID(N), logical IO to APIC ID table
825 struct PROCENTRY proc;
832 int apic, bus, cpu, intr;
835 POSTCODE(MPTABLE_PASS2_POST);
837 /* Initialize fake proc entry for use with HT fixup. */
838 bzero(&proc, sizeof(proc));
840 proc.cpu_flags = PROCENTRY_FLAG_EN;
843 MALLOC(io_apic_versions, u_int32_t *, sizeof(u_int32_t) * mp_napics,
845 MALLOC(ioapic, volatile ioapic_t **, sizeof(ioapic_t *) * mp_napics,
846 M_DEVBUF, M_WAITOK | M_ZERO);
847 MALLOC(io_apic_ints, io_int *, sizeof(io_int) * (nintrs + FIXUP_EXTRA_APIC_INTS),
850 MALLOC(bus_data, bus_datum *, sizeof(bus_datum) * mp_nbusses,
854 for (i = 0; i < mp_napics; i++) {
855 ioapic[i] = permanent_io_mapping(io_apic_address[i]);
859 /* clear various tables */
860 for (x = 0; x < NAPICID; ++x) {
861 CPU_TO_ID(x) = -1; /* logical CPU to APIC ID table */
863 ID_TO_IO(x) = -1; /* phy APIC ID to log CPU/IO table */
864 IO_TO_ID(x) = -1; /* logical IO to APIC ID table */
868 /* clear bus data table */
869 for (x = 0; x < mp_nbusses; ++x)
870 bus_data[x].bus_id = 0xff;
873 /* clear IO APIC INT table */
874 for (x = 0; x < (nintrs + 1); ++x) {
875 io_apic_ints[x].int_type = 0xff;
876 io_apic_ints[x].int_vector = 0xff;
880 /* setup the cpu/apic mapping arrays */
883 /* record whether PIC or virtual-wire mode */
884 machintr_setvar_simple(MACHINTR_VAR_IMCR_PRESENT, mpfps->mpfb2 & 0x80);
886 /* check for use of 'default' configuration */
887 if (mpfps->mpfb1 != 0)
888 return mpfps->mpfb1; /* return default configuration type */
890 if ((cth = mpfps->pap) == 0)
891 panic("MP Configuration Table Header MISSING!");
893 /* walk the table, recording info of interest */
894 totalSize = cth->base_table_length - sizeof(struct MPCTH);
895 position = (u_char *) cth + sizeof(struct MPCTH);
896 count = cth->entry_count;
897 apic = bus = intr = 0;
898 cpu = 1; /* pre-count the BSP */
901 switch (type = *(u_char *) position) {
903 if (processor_entry(position, cpu))
906 if (need_hyperthreading_fixup) {
908 * Create fake mptable processor entries
909 * and feed them to processor_entry() to
910 * enumerate the logical CPUs.
912 proc.apic_id = ((proc_entry_ptr)position)->apic_id;
913 for (i = 1; i < logical_cpus; i++) {
915 processor_entry(&proc, cpu);
916 logical_cpus_mask |= (1 << cpu);
922 if (bus_entry(position, bus))
927 if (io_apic_entry(position, apic))
933 if (int_entry(position, intr))
938 /* int_entry(position); */
941 panic("mpfps Base Table HOSED!");
945 totalSize -= basetable_entry_types[type].length;
946 position = (uint8_t *)position + basetable_entry_types[type].length;
949 if (boot_cpu_id == -1)
950 panic("NO BSP found!");
952 /* report fact that its NOT a default configuration */
957 * Check if we should perform a hyperthreading "fix-up" to
958 * enumerate any logical CPU's that aren't already listed
961 * XXX: We assume that all of the physical CPUs in the
962 * system have the same number of logical CPUs.
964 * XXX: We assume that APIC ID's are allocated such that
965 * the APIC ID's for a physical processor are aligned
966 * with the number of logical CPU's in the processor.
969 mptable_hyperthread_fixup(u_int id_mask)
973 /* Nothing to do if there is no HTT support. */
974 if ((cpu_feature & CPUID_HTT) == 0)
976 logical_cpus = (cpu_procinfo & CPUID_HTT_CORES) >> 16;
977 if (logical_cpus <= 1)
981 * For each APIC ID of a CPU that is set in the mask,
982 * scan the other candidate APIC ID's for this
983 * physical processor. If any of those ID's are
984 * already in the table, then kill the fixup.
986 for (id = 0; id <= MAXCPU; id++) {
987 if ((id_mask & 1 << id) == 0)
989 /* First, make sure we are on a logical_cpus boundary. */
990 if (id % logical_cpus != 0)
992 for (i = id + 1; i < id + logical_cpus; i++)
993 if ((id_mask & 1 << i) != 0)
998 * Ok, the ID's checked out, so enable the fixup. We have to fixup
1001 need_hyperthreading_fixup = 1;
1002 mp_naps *= logical_cpus;
1008 assign_apic_irq(int apic, int intpin, int irq)
1012 if (int_to_apicintpin[irq].ioapic != -1)
1013 panic("assign_apic_irq: inconsistent table");
1015 int_to_apicintpin[irq].ioapic = apic;
1016 int_to_apicintpin[irq].int_pin = intpin;
1017 int_to_apicintpin[irq].apic_address = ioapic[apic];
1018 int_to_apicintpin[irq].redirindex = IOAPIC_REDTBL + 2 * intpin;
1020 for (x = 0; x < nintrs; x++) {
1021 if ((io_apic_ints[x].int_type == 0 ||
1022 io_apic_ints[x].int_type == 3) &&
1023 io_apic_ints[x].int_vector == 0xff &&
1024 io_apic_ints[x].dst_apic_id == IO_TO_ID(apic) &&
1025 io_apic_ints[x].dst_apic_int == intpin)
1026 io_apic_ints[x].int_vector = irq;
1031 revoke_apic_irq(int irq)
1037 if (int_to_apicintpin[irq].ioapic == -1)
1038 panic("revoke_apic_irq: inconsistent table");
1040 oldapic = int_to_apicintpin[irq].ioapic;
1041 oldintpin = int_to_apicintpin[irq].int_pin;
1043 int_to_apicintpin[irq].ioapic = -1;
1044 int_to_apicintpin[irq].int_pin = 0;
1045 int_to_apicintpin[irq].apic_address = NULL;
1046 int_to_apicintpin[irq].redirindex = 0;
1048 for (x = 0; x < nintrs; x++) {
1049 if ((io_apic_ints[x].int_type == 0 ||
1050 io_apic_ints[x].int_type == 3) &&
1051 io_apic_ints[x].int_vector != 0xff &&
1052 io_apic_ints[x].dst_apic_id == IO_TO_ID(oldapic) &&
1053 io_apic_ints[x].dst_apic_int == oldintpin)
1054 io_apic_ints[x].int_vector = 0xff;
1062 allocate_apic_irq(int intr)
1068 if (io_apic_ints[intr].int_vector != 0xff)
1069 return; /* Interrupt handler already assigned */
1071 if (io_apic_ints[intr].int_type != 0 &&
1072 (io_apic_ints[intr].int_type != 3 ||
1073 (io_apic_ints[intr].dst_apic_id == IO_TO_ID(0) &&
1074 io_apic_ints[intr].dst_apic_int == 0)))
1075 return; /* Not INT or ExtInt on != (0, 0) */
1078 while (irq < APIC_INTMAPSIZE &&
1079 int_to_apicintpin[irq].ioapic != -1)
1082 if (irq >= APIC_INTMAPSIZE)
1083 return; /* No free interrupt handlers */
1085 apic = ID_TO_IO(io_apic_ints[intr].dst_apic_id);
1086 intpin = io_apic_ints[intr].dst_apic_int;
1088 assign_apic_irq(apic, intpin, irq);
1089 io_apic_setup_intpin(apic, intpin);
1094 swap_apic_id(int apic, int oldid, int newid)
1101 return; /* Nothing to do */
1103 kprintf("Changing APIC ID for IO APIC #%d from %d to %d in MP table\n",
1104 apic, oldid, newid);
1106 /* Swap physical APIC IDs in interrupt entries */
1107 for (x = 0; x < nintrs; x++) {
1108 if (io_apic_ints[x].dst_apic_id == oldid)
1109 io_apic_ints[x].dst_apic_id = newid;
1110 else if (io_apic_ints[x].dst_apic_id == newid)
1111 io_apic_ints[x].dst_apic_id = oldid;
1114 /* Swap physical APIC IDs in IO_TO_ID mappings */
1115 for (oapic = 0; oapic < mp_napics; oapic++)
1116 if (IO_TO_ID(oapic) == newid)
1119 if (oapic < mp_napics) {
1120 kprintf("Changing APIC ID for IO APIC #%d from "
1121 "%d to %d in MP table\n",
1122 oapic, newid, oldid);
1123 IO_TO_ID(oapic) = oldid;
1125 IO_TO_ID(apic) = newid;
1130 fix_id_to_io_mapping(void)
1134 for (x = 0; x < NAPICID; x++)
1137 for (x = 0; x <= mp_naps; x++)
1138 if (CPU_TO_ID(x) < NAPICID)
1139 ID_TO_IO(CPU_TO_ID(x)) = x;
1141 for (x = 0; x < mp_napics; x++)
1142 if (IO_TO_ID(x) < NAPICID)
1143 ID_TO_IO(IO_TO_ID(x)) = x;
1148 first_free_apic_id(void)
1152 for (freeid = 0; freeid < NAPICID; freeid++) {
1153 for (x = 0; x <= mp_naps; x++)
1154 if (CPU_TO_ID(x) == freeid)
1158 for (x = 0; x < mp_napics; x++)
1159 if (IO_TO_ID(x) == freeid)
1170 io_apic_id_acceptable(int apic, int id)
1172 int cpu; /* Logical CPU number */
1173 int oapic; /* Logical IO APIC number for other IO APIC */
1176 return 0; /* Out of range */
1178 for (cpu = 0; cpu <= mp_naps; cpu++)
1179 if (CPU_TO_ID(cpu) == id)
1180 return 0; /* Conflict with CPU */
1182 for (oapic = 0; oapic < mp_napics && oapic < apic; oapic++)
1183 if (IO_TO_ID(oapic) == id)
1184 return 0; /* Conflict with other APIC */
1186 return 1; /* ID is acceptable for IO APIC */
1191 io_apic_find_int_entry(int apic, int pin)
1195 /* search each of the possible INTerrupt sources */
1196 for (x = 0; x < nintrs; ++x) {
1197 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1198 (pin == io_apic_ints[x].dst_apic_int))
1199 return (&io_apic_ints[x]);
1207 * parse an Intel MP specification table
1215 int apic; /* IO APIC unit number */
1216 int freeid; /* Free physical APIC ID */
1217 int physid; /* Current physical IO APIC ID */
1220 int bus_0 = 0; /* Stop GCC warning */
1221 int bus_pci = 0; /* Stop GCC warning */
1225 * Fix mis-numbering of the PCI bus and its INT entries if the BIOS
1226 * did it wrong. The MP spec says that when more than 1 PCI bus
1227 * exists the BIOS must begin with bus entries for the PCI bus and use
1228 * actual PCI bus numbering. This implies that when only 1 PCI bus
1229 * exists the BIOS can choose to ignore this ordering, and indeed many
1230 * MP motherboards do ignore it. This causes a problem when the PCI
1231 * sub-system makes requests of the MP sub-system based on PCI bus
1232 * numbers. So here we look for the situation and renumber the
1233 * busses and associated INTs in an effort to "make it right".
1236 /* find bus 0, PCI bus, count the number of PCI busses */
1237 for (num_pci_bus = 0, x = 0; x < mp_nbusses; ++x) {
1238 if (bus_data[x].bus_id == 0) {
1241 if (bus_data[x].bus_type == PCI) {
1247 * bus_0 == slot of bus with ID of 0
1248 * bus_pci == slot of last PCI bus encountered
1251 /* check the 1 PCI bus case for sanity */
1252 /* if it is number 0 all is well */
1253 if (num_pci_bus == 1 &&
1254 bus_data[bus_pci].bus_id != 0) {
1256 /* mis-numbered, swap with whichever bus uses slot 0 */
1258 /* swap the bus entry types */
1259 bus_data[bus_pci].bus_type = bus_data[bus_0].bus_type;
1260 bus_data[bus_0].bus_type = PCI;
1263 /* swap each relavant INTerrupt entry */
1264 id = bus_data[bus_pci].bus_id;
1265 for (x = 0; x < nintrs; ++x) {
1266 if (io_apic_ints[x].src_bus_id == id) {
1267 io_apic_ints[x].src_bus_id = 0;
1269 else if (io_apic_ints[x].src_bus_id == 0) {
1270 io_apic_ints[x].src_bus_id = id;
1277 /* Assign IO APIC IDs.
1279 * First try the existing ID. If a conflict is detected, try
1280 * the ID in the MP table. If a conflict is still detected, find
1283 * We cannot use the ID_TO_IO table before all conflicts has been
1284 * resolved and the table has been corrected.
1286 for (apic = 0; apic < mp_napics; ++apic) { /* For all IO APICs */
1288 /* First try to use the value set by the BIOS */
1289 physid = io_apic_get_id(apic);
1290 if (io_apic_id_acceptable(apic, physid)) {
1291 if (IO_TO_ID(apic) != physid)
1292 swap_apic_id(apic, IO_TO_ID(apic), physid);
1296 /* Then check if the value in the MP table is acceptable */
1297 if (io_apic_id_acceptable(apic, IO_TO_ID(apic)))
1300 /* Last resort, find a free APIC ID and use it */
1301 freeid = first_free_apic_id();
1302 if (freeid >= NAPICID)
1303 panic("No free physical APIC IDs found");
1305 if (io_apic_id_acceptable(apic, freeid)) {
1306 swap_apic_id(apic, IO_TO_ID(apic), freeid);
1309 panic("Free physical APIC ID not usable");
1311 fix_id_to_io_mapping();
1315 /* detect and fix broken Compaq MP table */
1316 if (apic_int_type(0, 0) == -1) {
1317 kprintf("APIC_IO: MP table broken: 8259->APIC entry missing!\n");
1318 io_apic_ints[nintrs].int_type = 3; /* ExtInt */
1319 io_apic_ints[nintrs].int_vector = 0xff; /* Unassigned */
1320 /* XXX fixme, set src bus id etc, but it doesn't seem to hurt */
1321 io_apic_ints[nintrs].dst_apic_id = IO_TO_ID(0);
1322 io_apic_ints[nintrs].dst_apic_int = 0; /* Pin 0 */
1324 } else if (apic_int_type(0, 0) == 0) {
1325 kprintf("APIC_IO: MP table broken: ExtINT entry corrupt!\n");
1326 for (x = 0; x < nintrs; ++x)
1327 if ((0 == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1328 (0 == io_apic_ints[x].dst_apic_int)) {
1329 io_apic_ints[x].int_type = 3;
1330 io_apic_ints[x].int_vector = 0xff;
1336 * Fix missing IRQ 15 when IRQ 14 is an ISA interrupt. IDE
1337 * controllers universally come in pairs. If IRQ 14 is specified
1338 * as an ISA interrupt, then IRQ 15 had better be too.
1340 * [ Shuttle XPC / AMD Athlon X2 ]
1341 * The MPTable is missing an entry for IRQ 15. Note that the
1342 * ACPI table has an entry for both 14 and 15.
1344 if (apic_int_type(0, 14) == 0 && apic_int_type(0, 15) == -1) {
1345 kprintf("APIC_IO: MP table broken: IRQ 15 not ISA when IRQ 14 is!\n");
1346 io14 = io_apic_find_int_entry(0, 14);
1347 io_apic_ints[nintrs] = *io14;
1348 io_apic_ints[nintrs].src_bus_irq = 15;
1349 io_apic_ints[nintrs].dst_apic_int = 15;
1357 /* Assign low level interrupt handlers */
1359 setup_apic_irq_mapping(void)
1365 for (x = 0; x < APIC_INTMAPSIZE; x++) {
1366 int_to_apicintpin[x].ioapic = -1;
1367 int_to_apicintpin[x].int_pin = 0;
1368 int_to_apicintpin[x].apic_address = NULL;
1369 int_to_apicintpin[x].redirindex = 0;
1372 /* First assign ISA/EISA interrupts */
1373 for (x = 0; x < nintrs; x++) {
1374 int_vector = io_apic_ints[x].src_bus_irq;
1375 if (int_vector < APIC_INTMAPSIZE &&
1376 io_apic_ints[x].int_vector == 0xff &&
1377 int_to_apicintpin[int_vector].ioapic == -1 &&
1378 (apic_int_is_bus_type(x, ISA) ||
1379 apic_int_is_bus_type(x, EISA)) &&
1380 io_apic_ints[x].int_type == 0) {
1381 assign_apic_irq(ID_TO_IO(io_apic_ints[x].dst_apic_id),
1382 io_apic_ints[x].dst_apic_int,
1387 /* Assign ExtInt entry if no ISA/EISA interrupt 0 entry */
1388 for (x = 0; x < nintrs; x++) {
1389 if (io_apic_ints[x].dst_apic_int == 0 &&
1390 io_apic_ints[x].dst_apic_id == IO_TO_ID(0) &&
1391 io_apic_ints[x].int_vector == 0xff &&
1392 int_to_apicintpin[0].ioapic == -1 &&
1393 io_apic_ints[x].int_type == 3) {
1394 assign_apic_irq(0, 0, 0);
1398 /* PCI interrupt assignment is deferred */
1404 processor_entry(proc_entry_ptr entry, int cpu)
1406 /* check for usability */
1407 if (!(entry->cpu_flags & PROCENTRY_FLAG_EN))
1410 if(entry->apic_id >= NAPICID)
1411 panic("CPU APIC ID out of range (0..%d)", NAPICID - 1);
1412 /* check for BSP flag */
1413 if (entry->cpu_flags & PROCENTRY_FLAG_BP) {
1414 boot_cpu_id = entry->apic_id;
1415 CPU_TO_ID(0) = entry->apic_id;
1416 ID_TO_CPU(entry->apic_id) = 0;
1417 return 0; /* its already been counted */
1420 /* add another AP to list, if less than max number of CPUs */
1421 else if (cpu < MAXCPU) {
1422 CPU_TO_ID(cpu) = entry->apic_id;
1423 ID_TO_CPU(entry->apic_id) = cpu;
1432 bus_entry(bus_entry_ptr entry, int bus)
1437 /* encode the name into an index */
1438 for (x = 0; x < 6; ++x) {
1439 if ((c = entry->bus_type[x]) == ' ')
1445 if ((x = lookup_bus_type(name)) == UNKNOWN_BUSTYPE)
1446 panic("unknown bus type: '%s'", name);
1448 bus_data[bus].bus_id = entry->bus_id;
1449 bus_data[bus].bus_type = x;
1457 io_apic_entry(io_apic_entry_ptr entry, int apic)
1459 if (!(entry->apic_flags & IOAPICENTRY_FLAG_EN))
1462 IO_TO_ID(apic) = entry->apic_id;
1463 if (entry->apic_id < NAPICID)
1464 ID_TO_IO(entry->apic_id) = apic;
1472 lookup_bus_type(char *name)
1476 for (x = 0; x < MAX_BUSTYPE; ++x)
1477 if (strcmp(bus_type_table[x].name, name) == 0)
1478 return bus_type_table[x].type;
1480 return UNKNOWN_BUSTYPE;
1486 int_entry(int_entry_ptr entry, int intr)
1490 io_apic_ints[intr].int_type = entry->int_type;
1491 io_apic_ints[intr].int_flags = entry->int_flags;
1492 io_apic_ints[intr].src_bus_id = entry->src_bus_id;
1493 io_apic_ints[intr].src_bus_irq = entry->src_bus_irq;
1494 if (entry->dst_apic_id == 255) {
1495 /* This signal goes to all IO APICS. Select an IO APIC
1496 with sufficient number of interrupt pins */
1497 for (apic = 0; apic < mp_napics; apic++)
1498 if (((io_apic_read(apic, IOAPIC_VER) &
1499 IOART_VER_MAXREDIR) >> MAXREDIRSHIFT) >=
1500 entry->dst_apic_int)
1502 if (apic < mp_napics)
1503 io_apic_ints[intr].dst_apic_id = IO_TO_ID(apic);
1505 io_apic_ints[intr].dst_apic_id = entry->dst_apic_id;
1507 io_apic_ints[intr].dst_apic_id = entry->dst_apic_id;
1508 io_apic_ints[intr].dst_apic_int = entry->dst_apic_int;
1514 apic_int_is_bus_type(int intr, int bus_type)
1518 for (bus = 0; bus < mp_nbusses; ++bus)
1519 if ((bus_data[bus].bus_id == io_apic_ints[intr].src_bus_id)
1520 && ((int) bus_data[bus].bus_type == bus_type))
1527 * Given a traditional ISA INT mask, return an APIC mask.
1530 isa_apic_mask(u_int isa_mask)
1535 #if defined(SKIP_IRQ15_REDIRECT)
1536 if (isa_mask == (1 << 15)) {
1537 kprintf("skipping ISA IRQ15 redirect\n");
1540 #endif /* SKIP_IRQ15_REDIRECT */
1542 isa_irq = ffs(isa_mask); /* find its bit position */
1543 if (isa_irq == 0) /* doesn't exist */
1545 --isa_irq; /* make it zero based */
1547 apic_pin = isa_apic_irq(isa_irq); /* look for APIC connection */
1551 return (1 << apic_pin); /* convert pin# to a mask */
1555 * Determine which APIC pin an ISA/EISA INT is attached to.
1557 #define INTTYPE(I) (io_apic_ints[(I)].int_type)
1558 #define INTPIN(I) (io_apic_ints[(I)].dst_apic_int)
1559 #define INTIRQ(I) (io_apic_ints[(I)].int_vector)
1560 #define INTAPIC(I) (ID_TO_IO(io_apic_ints[(I)].dst_apic_id))
1562 #define SRCBUSIRQ(I) (io_apic_ints[(I)].src_bus_irq)
1564 isa_apic_irq(int isa_irq)
1568 for (intr = 0; intr < nintrs; ++intr) { /* check each record */
1569 if (INTTYPE(intr) == 0) { /* standard INT */
1570 if (SRCBUSIRQ(intr) == isa_irq) {
1571 if (apic_int_is_bus_type(intr, ISA) ||
1572 apic_int_is_bus_type(intr, EISA)) {
1573 if (INTIRQ(intr) == 0xff)
1574 return -1; /* unassigned */
1575 return INTIRQ(intr); /* found */
1580 return -1; /* NOT found */
1585 * Determine which APIC pin a PCI INT is attached to.
1587 #define SRCBUSID(I) (io_apic_ints[(I)].src_bus_id)
1588 #define SRCBUSDEVICE(I) ((io_apic_ints[(I)].src_bus_irq >> 2) & 0x1f)
1589 #define SRCBUSLINE(I) (io_apic_ints[(I)].src_bus_irq & 0x03)
1591 pci_apic_irq(int pciBus, int pciDevice, int pciInt)
1595 --pciInt; /* zero based */
1597 for (intr = 0; intr < nintrs; ++intr) { /* check each record */
1598 if ((INTTYPE(intr) == 0) /* standard INT */
1599 && (SRCBUSID(intr) == pciBus)
1600 && (SRCBUSDEVICE(intr) == pciDevice)
1601 && (SRCBUSLINE(intr) == pciInt)) { /* a candidate IRQ */
1602 if (apic_int_is_bus_type(intr, PCI)) {
1603 if (INTIRQ(intr) == 0xff)
1604 allocate_apic_irq(intr);
1605 if (INTIRQ(intr) == 0xff)
1606 return -1; /* unassigned */
1607 return INTIRQ(intr); /* exact match */
1612 return -1; /* NOT found */
1616 next_apic_irq(int irq)
1623 for (intr = 0; intr < nintrs; intr++) {
1624 if (INTIRQ(intr) != irq || INTTYPE(intr) != 0)
1626 bus = SRCBUSID(intr);
1627 bustype = apic_bus_type(bus);
1628 if (bustype != ISA &&
1634 if (intr >= nintrs) {
1637 for (ointr = intr + 1; ointr < nintrs; ointr++) {
1638 if (INTTYPE(ointr) != 0)
1640 if (bus != SRCBUSID(ointr))
1642 if (bustype == PCI) {
1643 if (SRCBUSDEVICE(intr) != SRCBUSDEVICE(ointr))
1645 if (SRCBUSLINE(intr) != SRCBUSLINE(ointr))
1648 if (bustype == ISA || bustype == EISA) {
1649 if (SRCBUSIRQ(intr) != SRCBUSIRQ(ointr))
1652 if (INTPIN(intr) == INTPIN(ointr))
1656 if (ointr >= nintrs) {
1659 return INTIRQ(ointr);
1674 * Reprogram the MB chipset to NOT redirect an ISA INTerrupt.
1677 * Exactly what this means is unclear at this point. It is a solution
1678 * for motherboards that redirect the MBIRQ0 pin. Generically a motherboard
1679 * could route any of the ISA INTs to upper (>15) IRQ values. But most would
1680 * NOT be redirected via MBIRQ0, thus "undirect()ing" them would NOT be an
1684 undirect_isa_irq(int rirq)
1688 kprintf("Freeing redirected ISA irq %d.\n", rirq);
1689 /** FIXME: tickle the MB redirector chip */
1693 kprintf("Freeing (NOT implemented) redirected ISA irq %d.\n", rirq);
1700 * Reprogram the MB chipset to NOT redirect a PCI INTerrupt
1703 undirect_pci_irq(int rirq)
1707 kprintf("Freeing redirected PCI irq %d.\n", rirq);
1709 /** FIXME: tickle the MB redirector chip */
1713 kprintf("Freeing (NOT implemented) redirected PCI irq %d.\n",
1721 * given a bus ID, return:
1722 * the bus type if found
1726 apic_bus_type(int id)
1730 for (x = 0; x < mp_nbusses; ++x)
1731 if (bus_data[x].bus_id == id)
1732 return bus_data[x].bus_type;
1740 * given a LOGICAL APIC# and pin#, return:
1741 * the associated src bus ID if found
1745 apic_src_bus_id(int apic, int pin)
1749 /* search each of the possible INTerrupt sources */
1750 for (x = 0; x < nintrs; ++x)
1751 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1752 (pin == io_apic_ints[x].dst_apic_int))
1753 return (io_apic_ints[x].src_bus_id);
1755 return -1; /* NOT found */
1759 * given a LOGICAL APIC# and pin#, return:
1760 * the associated src bus IRQ if found
1764 apic_src_bus_irq(int apic, int pin)
1768 for (x = 0; x < nintrs; x++)
1769 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1770 (pin == io_apic_ints[x].dst_apic_int))
1771 return (io_apic_ints[x].src_bus_irq);
1773 return -1; /* NOT found */
1778 * given a LOGICAL APIC# and pin#, return:
1779 * the associated INTerrupt type if found
1783 apic_int_type(int apic, int pin)
1787 /* search each of the possible INTerrupt sources */
1788 for (x = 0; x < nintrs; ++x) {
1789 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1790 (pin == io_apic_ints[x].dst_apic_int))
1791 return (io_apic_ints[x].int_type);
1793 return -1; /* NOT found */
1797 * Return the IRQ associated with an APIC pin
1800 apic_irq(int apic, int pin)
1805 for (x = 0; x < nintrs; ++x) {
1806 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1807 (pin == io_apic_ints[x].dst_apic_int)) {
1808 res = io_apic_ints[x].int_vector;
1811 if (apic != int_to_apicintpin[res].ioapic)
1812 panic("apic_irq: inconsistent table %d/%d", apic, int_to_apicintpin[res].ioapic);
1813 if (pin != int_to_apicintpin[res].int_pin)
1814 panic("apic_irq inconsistent table (2)");
1823 * given a LOGICAL APIC# and pin#, return:
1824 * the associated trigger mode if found
1828 apic_trigger(int apic, int pin)
1832 /* search each of the possible INTerrupt sources */
1833 for (x = 0; x < nintrs; ++x)
1834 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1835 (pin == io_apic_ints[x].dst_apic_int))
1836 return ((io_apic_ints[x].int_flags >> 2) & 0x03);
1838 return -1; /* NOT found */
1843 * given a LOGICAL APIC# and pin#, return:
1844 * the associated 'active' level if found
1848 apic_polarity(int apic, int pin)
1852 /* search each of the possible INTerrupt sources */
1853 for (x = 0; x < nintrs; ++x)
1854 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) &&
1855 (pin == io_apic_ints[x].dst_apic_int))
1856 return (io_apic_ints[x].int_flags & 0x03);
1858 return -1; /* NOT found */
1864 * set data according to MP defaults
1865 * FIXME: probably not complete yet...
1868 default_mp_table(int type)
1871 #if defined(APIC_IO)
1874 #endif /* APIC_IO */
1877 kprintf(" MP default config type: %d\n", type);
1880 kprintf(" bus: ISA, APIC: 82489DX\n");
1883 kprintf(" bus: EISA, APIC: 82489DX\n");
1886 kprintf(" bus: EISA, APIC: 82489DX\n");
1889 kprintf(" bus: MCA, APIC: 82489DX\n");
1892 kprintf(" bus: ISA+PCI, APIC: Integrated\n");
1895 kprintf(" bus: EISA+PCI, APIC: Integrated\n");
1898 kprintf(" bus: MCA+PCI, APIC: Integrated\n");
1901 kprintf(" future type\n");
1907 boot_cpu_id = (lapic.id & APIC_ID_MASK) >> 24;
1908 ap_cpu_id = (boot_cpu_id == 0) ? 1 : 0;
1911 CPU_TO_ID(0) = boot_cpu_id;
1912 ID_TO_CPU(boot_cpu_id) = 0;
1914 /* one and only AP */
1915 CPU_TO_ID(1) = ap_cpu_id;
1916 ID_TO_CPU(ap_cpu_id) = 1;
1918 #if defined(APIC_IO)
1919 /* one and only IO APIC */
1920 io_apic_id = (io_apic_read(0, IOAPIC_ID) & APIC_ID_MASK) >> 24;
1923 * sanity check, refer to MP spec section 3.6.6, last paragraph
1924 * necessary as some hardware isn't properly setting up the IO APIC
1926 #if defined(REALLY_ANAL_IOAPICID_VALUE)
1927 if (io_apic_id != 2) {
1929 if ((io_apic_id == 0) || (io_apic_id == 1) || (io_apic_id == 15)) {
1930 #endif /* REALLY_ANAL_IOAPICID_VALUE */
1931 io_apic_set_id(0, 2);
1934 IO_TO_ID(0) = io_apic_id;
1935 ID_TO_IO(io_apic_id) = 0;
1936 #endif /* APIC_IO */
1938 /* fill out bus entries */
1947 bus_data[0].bus_id = default_data[type - 1][1];
1948 bus_data[0].bus_type = default_data[type - 1][2];
1949 bus_data[1].bus_id = default_data[type - 1][3];
1950 bus_data[1].bus_type = default_data[type - 1][4];
1953 /* case 4: case 7: MCA NOT supported */
1954 default: /* illegal/reserved */
1955 panic("BAD default MP config: %d", type);
1959 #if defined(APIC_IO)
1960 /* general cases from MP v1.4, table 5-2 */
1961 for (pin = 0; pin < 16; ++pin) {
1962 io_apic_ints[pin].int_type = 0;
1963 io_apic_ints[pin].int_flags = 0x05; /* edge/active-hi */
1964 io_apic_ints[pin].src_bus_id = 0;
1965 io_apic_ints[pin].src_bus_irq = pin; /* IRQ2 caught below */
1966 io_apic_ints[pin].dst_apic_id = io_apic_id;
1967 io_apic_ints[pin].dst_apic_int = pin; /* 1-to-1 */
1970 /* special cases from MP v1.4, table 5-2 */
1972 io_apic_ints[2].int_type = 0xff; /* N/C */
1973 io_apic_ints[13].int_type = 0xff; /* N/C */
1974 #if !defined(APIC_MIXED_MODE)
1976 panic("sorry, can't support type 2 default yet");
1977 #endif /* APIC_MIXED_MODE */
1980 io_apic_ints[2].src_bus_irq = 0; /* ISA IRQ0 is on APIC INT 2 */
1983 io_apic_ints[0].int_type = 0xff; /* N/C */
1985 io_apic_ints[0].int_type = 3; /* vectored 8259 */
1986 #endif /* APIC_IO */
1990 * Map a physical memory address representing I/O into KVA. The I/O
1991 * block is assumed not to cross a page boundary.
1994 permanent_io_mapping(vm_paddr_t pa)
2000 KKASSERT(pa < 0x100000000LL);
2002 pgeflag = 0; /* not used for SMP yet */
2005 * If the requested physical address has already been incidently
2006 * mapped, just use the existing mapping. Otherwise create a new
2009 for (i = IO_MAPPING_START_INDEX; i < SMPpt_alloc_index; ++i) {
2010 if (((vm_offset_t)SMPpt[i] & PG_FRAME) ==
2011 ((vm_offset_t)pa & PG_FRAME)) {
2015 if (i == SMPpt_alloc_index) {
2016 if (i == NPTEPG - 2) {
2017 panic("permanent_io_mapping: We ran out of space"
2020 SMPpt[i] = (pt_entry_t)(PG_V | PG_RW | pgeflag |
2021 ((vm_offset_t)pa & PG_FRAME));
2022 ++SMPpt_alloc_index;
2024 vaddr = (vm_offset_t)CPU_prvspace + (i * PAGE_SIZE) +
2025 ((vm_offset_t)pa & PAGE_MASK);
2026 return ((void *)vaddr);
2030 * start each AP in our list
2033 start_all_aps(u_int boot_addr)
2037 u_char mpbiosreason;
2038 u_long mpbioswarmvec;
2039 struct mdglobaldata *gd;
2040 struct privatespace *ps;
2044 POSTCODE(START_ALL_APS_POST);
2046 /* initialize BSP's local APIC */
2050 /* install the AP 1st level boot code */
2051 install_ap_tramp(boot_addr);
2054 /* save the current value of the warm-start vector */
2055 mpbioswarmvec = *((u_long *) WARMBOOT_OFF);
2056 outb(CMOS_REG, BIOS_RESET);
2057 mpbiosreason = inb(CMOS_DATA);
2059 /* set up temporary P==V mapping for AP boot */
2060 /* XXX this is a hack, we should boot the AP on its own stack/PTD */
2061 kptbase = (uintptr_t)(void *)KPTphys;
2062 for (x = 0; x < NKPT; x++) {
2063 PTD[x] = (pd_entry_t)(PG_V | PG_RW |
2064 ((kptbase + x * PAGE_SIZE) & PG_FRAME));
2069 for (x = 1; x <= mp_naps; ++x) {
2071 /* This is a bit verbose, it will go away soon. */
2073 /* first page of AP's private space */
2074 pg = x * i386_btop(sizeof(struct privatespace));
2076 /* allocate new private data page(s) */
2077 gd = (struct mdglobaldata *)kmem_alloc(&kernel_map,
2078 MDGLOBALDATA_BASEALLOC_SIZE);
2079 /* wire it into the private page table page */
2080 for (i = 0; i < MDGLOBALDATA_BASEALLOC_SIZE; i += PAGE_SIZE) {
2081 SMPpt[pg + i / PAGE_SIZE] = (pt_entry_t)
2082 (PG_V | PG_RW | vtophys_pte((char *)gd + i));
2084 pg += MDGLOBALDATA_BASEALLOC_PAGES;
2086 SMPpt[pg + 0] = 0; /* *gd_CMAP1 */
2087 SMPpt[pg + 1] = 0; /* *gd_CMAP2 */
2088 SMPpt[pg + 2] = 0; /* *gd_CMAP3 */
2089 SMPpt[pg + 3] = 0; /* *gd_PMAP1 */
2091 /* allocate and set up an idle stack data page */
2092 stack = (char *)kmem_alloc(&kernel_map, UPAGES*PAGE_SIZE);
2093 for (i = 0; i < UPAGES; i++) {
2094 SMPpt[pg + 4 + i] = (pt_entry_t)
2095 (PG_V | PG_RW | vtophys_pte(PAGE_SIZE * i + stack));
2098 gd = &CPU_prvspace[x].mdglobaldata; /* official location */
2099 bzero(gd, sizeof(*gd));
2100 gd->mi.gd_prvspace = ps = &CPU_prvspace[x];
2102 /* prime data page for it to use */
2103 mi_gdinit(&gd->mi, x);
2105 gd->gd_CMAP1 = &SMPpt[pg + 0];
2106 gd->gd_CMAP2 = &SMPpt[pg + 1];
2107 gd->gd_CMAP3 = &SMPpt[pg + 2];
2108 gd->gd_PMAP1 = &SMPpt[pg + 3];
2109 gd->gd_CADDR1 = ps->CPAGE1;
2110 gd->gd_CADDR2 = ps->CPAGE2;
2111 gd->gd_CADDR3 = ps->CPAGE3;
2112 gd->gd_PADDR1 = (unsigned *)ps->PPAGE1;
2113 gd->mi.gd_ipiq = (void *)kmem_alloc(&kernel_map, sizeof(lwkt_ipiq) * (mp_naps + 1));
2114 bzero(gd->mi.gd_ipiq, sizeof(lwkt_ipiq) * (mp_naps + 1));
2116 /* setup a vector to our boot code */
2117 *((volatile u_short *) WARMBOOT_OFF) = WARMBOOT_TARGET;
2118 *((volatile u_short *) WARMBOOT_SEG) = (boot_addr >> 4);
2119 outb(CMOS_REG, BIOS_RESET);
2120 outb(CMOS_DATA, BIOS_WARM); /* 'warm-start' */
2123 * Setup the AP boot stack
2125 bootSTK = &ps->idlestack[UPAGES*PAGE_SIZE/2];
2128 /* attempt to start the Application Processor */
2129 CHECK_INIT(99); /* setup checkpoints */
2130 if (!start_ap(gd, boot_addr)) {
2131 kprintf("AP #%d (PHY# %d) failed!\n", x, CPU_TO_ID(x));
2132 CHECK_PRINT("trace"); /* show checkpoints */
2133 /* better panic as the AP may be running loose */
2134 kprintf("panic y/n? [y] ");
2135 if (cngetc() != 'n')
2138 CHECK_PRINT("trace"); /* show checkpoints */
2140 /* record its version info */
2141 cpu_apic_versions[x] = cpu_apic_versions[0];
2144 /* set ncpus to 1 + highest logical cpu. Not all may have come up */
2147 /* ncpus2 -- ncpus rounded down to the nearest power of 2 */
2148 for (shift = 0; (1 << shift) <= ncpus; ++shift)
2151 ncpus2_shift = shift;
2152 ncpus2 = 1 << shift;
2153 ncpus2_mask = ncpus2 - 1;
2155 /* ncpus_fit -- ncpus rounded up to the nearest power of 2 */
2156 if ((1 << shift) < ncpus)
2158 ncpus_fit = 1 << shift;
2159 ncpus_fit_mask = ncpus_fit - 1;
2161 /* build our map of 'other' CPUs */
2162 mycpu->gd_other_cpus = smp_startup_mask & ~(1 << mycpu->gd_cpuid);
2163 mycpu->gd_ipiq = (void *)kmem_alloc(&kernel_map, sizeof(lwkt_ipiq) * ncpus);
2164 bzero(mycpu->gd_ipiq, sizeof(lwkt_ipiq) * ncpus);
2166 /* fill in our (BSP) APIC version */
2167 cpu_apic_versions[0] = lapic.version;
2169 /* restore the warmstart vector */
2170 *(u_long *) WARMBOOT_OFF = mpbioswarmvec;
2171 outb(CMOS_REG, BIOS_RESET);
2172 outb(CMOS_DATA, mpbiosreason);
2175 * NOTE! The idlestack for the BSP was setup by locore. Finish
2176 * up, clean out the P==V mapping we did earlier.
2178 for (x = 0; x < NKPT; x++)
2182 /* number of APs actually started */
2188 * load the 1st level AP boot code into base memory.
2191 /* targets for relocation */
2192 extern void bigJump(void);
2193 extern void bootCodeSeg(void);
2194 extern void bootDataSeg(void);
2195 extern void MPentry(void);
2196 extern u_int MP_GDT;
2197 extern u_int mp_gdtbase;
2200 install_ap_tramp(u_int boot_addr)
2203 int size = *(int *) ((u_long) & bootMP_size);
2204 u_char *src = (u_char *) ((u_long) bootMP);
2205 u_char *dst = (u_char *) boot_addr + KERNBASE;
2206 u_int boot_base = (u_int) bootMP;
2211 POSTCODE(INSTALL_AP_TRAMP_POST);
2213 for (x = 0; x < size; ++x)
2217 * modify addresses in code we just moved to basemem. unfortunately we
2218 * need fairly detailed info about mpboot.s for this to work. changes
2219 * to mpboot.s might require changes here.
2222 /* boot code is located in KERNEL space */
2223 dst = (u_char *) boot_addr + KERNBASE;
2225 /* modify the lgdt arg */
2226 dst32 = (u_int32_t *) (dst + ((u_int) & mp_gdtbase - boot_base));
2227 *dst32 = boot_addr + ((u_int) & MP_GDT - boot_base);
2229 /* modify the ljmp target for MPentry() */
2230 dst32 = (u_int32_t *) (dst + ((u_int) bigJump - boot_base) + 1);
2231 *dst32 = ((u_int) MPentry - KERNBASE);
2233 /* modify the target for boot code segment */
2234 dst16 = (u_int16_t *) (dst + ((u_int) bootCodeSeg - boot_base));
2235 dst8 = (u_int8_t *) (dst16 + 1);
2236 *dst16 = (u_int) boot_addr & 0xffff;
2237 *dst8 = ((u_int) boot_addr >> 16) & 0xff;
2239 /* modify the target for boot data segment */
2240 dst16 = (u_int16_t *) (dst + ((u_int) bootDataSeg - boot_base));
2241 dst8 = (u_int8_t *) (dst16 + 1);
2242 *dst16 = (u_int) boot_addr & 0xffff;
2243 *dst8 = ((u_int) boot_addr >> 16) & 0xff;
2248 * this function starts the AP (application processor) identified
2249 * by the APIC ID 'physicalCpu'. It does quite a "song and dance"
2250 * to accomplish this. This is necessary because of the nuances
2251 * of the different hardware we might encounter. It ain't pretty,
2252 * but it seems to work.
2254 * NOTE: eventually an AP gets to ap_init(), which is called just
2255 * before the AP goes into the LWKT scheduler's idle loop.
2258 start_ap(struct mdglobaldata *gd, u_int boot_addr)
2262 u_long icr_lo, icr_hi;
2264 POSTCODE(START_AP_POST);
2266 /* get the PHYSICAL APIC ID# */
2267 physical_cpu = CPU_TO_ID(gd->mi.gd_cpuid);
2269 /* calculate the vector */
2270 vector = (boot_addr >> 12) & 0xff;
2272 /* Make sure the target cpu sees everything */
2276 * first we do an INIT/RESET IPI this INIT IPI might be run, reseting
2277 * and running the target CPU. OR this INIT IPI might be latched (P5
2278 * bug), CPU waiting for STARTUP IPI. OR this INIT IPI might be
2282 /* setup the address for the target AP */
2283 icr_hi = lapic.icr_hi & ~APIC_ID_MASK;
2284 icr_hi |= (physical_cpu << 24);
2285 lapic.icr_hi = icr_hi;
2287 /* do an INIT IPI: assert RESET */
2288 icr_lo = lapic.icr_lo & 0xfff00000;
2289 lapic.icr_lo = icr_lo | 0x0000c500;
2291 /* wait for pending status end */
2292 while (lapic.icr_lo & APIC_DELSTAT_MASK)
2295 /* do an INIT IPI: deassert RESET */
2296 lapic.icr_lo = icr_lo | 0x00008500;
2298 /* wait for pending status end */
2299 u_sleep(10000); /* wait ~10mS */
2300 while (lapic.icr_lo & APIC_DELSTAT_MASK)
2304 * next we do a STARTUP IPI: the previous INIT IPI might still be
2305 * latched, (P5 bug) this 1st STARTUP would then terminate
2306 * immediately, and the previously started INIT IPI would continue. OR
2307 * the previous INIT IPI has already run. and this STARTUP IPI will
2308 * run. OR the previous INIT IPI was ignored. and this STARTUP IPI
2312 /* do a STARTUP IPI */
2313 lapic.icr_lo = icr_lo | 0x00000600 | vector;
2314 while (lapic.icr_lo & APIC_DELSTAT_MASK)
2316 u_sleep(200); /* wait ~200uS */
2319 * finally we do a 2nd STARTUP IPI: this 2nd STARTUP IPI should run IF
2320 * the previous STARTUP IPI was cancelled by a latched INIT IPI. OR
2321 * this STARTUP IPI will be ignored, as only ONE STARTUP IPI is
2322 * recognized after hardware RESET or INIT IPI.
2325 lapic.icr_lo = icr_lo | 0x00000600 | vector;
2326 while (lapic.icr_lo & APIC_DELSTAT_MASK)
2328 u_sleep(200); /* wait ~200uS */
2330 /* wait for it to start, see ap_init() */
2331 set_apic_timer(5000000);/* == 5 seconds */
2332 while (read_apic_timer()) {
2333 if (smp_startup_mask & (1 << gd->mi.gd_cpuid))
2334 return 1; /* return SUCCESS */
2336 return 0; /* return FAILURE */
2341 * Lazy flush the TLB on all other CPU's. DEPRECATED.
2343 * If for some reason we were unable to start all cpus we cannot safely
2344 * use broadcast IPIs.
2350 if (smp_startup_mask == smp_active_mask) {
2351 all_but_self_ipi(XINVLTLB_OFFSET);
2353 selected_apic_ipi(smp_active_mask, XINVLTLB_OFFSET,
2354 APIC_DELMODE_FIXED);
2360 * When called the executing CPU will send an IPI to all other CPUs
2361 * requesting that they halt execution.
2363 * Usually (but not necessarily) called with 'other_cpus' as its arg.
2365 * - Signals all CPUs in map to stop.
2366 * - Waits for each to stop.
2373 * XXX FIXME: this is not MP-safe, needs a lock to prevent multiple CPUs
2374 * from executing at same time.
2377 stop_cpus(u_int map)
2379 map &= smp_active_mask;
2381 /* send the Xcpustop IPI to all CPUs in map */
2382 selected_apic_ipi(map, XCPUSTOP_OFFSET, APIC_DELMODE_FIXED);
2384 while ((stopped_cpus & map) != map)
2392 * Called by a CPU to restart stopped CPUs.
2394 * Usually (but not necessarily) called with 'stopped_cpus' as its arg.
2396 * - Signals all CPUs in map to restart.
2397 * - Waits for each to restart.
2405 restart_cpus(u_int map)
2407 /* signal other cpus to restart */
2408 started_cpus = map & smp_active_mask;
2410 while ((stopped_cpus & map) != 0) /* wait for each to clear its bit */
2417 * This is called once the mpboot code has gotten us properly relocated
2418 * and the MMU turned on, etc. ap_init() is actually the idle thread,
2419 * and when it returns the scheduler will call the real cpu_idle() main
2420 * loop for the idlethread. Interrupts are disabled on entry and should
2421 * remain disabled at return.
2429 * Adjust smp_startup_mask to signal the BSP that we have started
2430 * up successfully. Note that we do not yet hold the BGL. The BSP
2431 * is waiting for our signal.
2433 * We can't set our bit in smp_active_mask yet because we are holding
2434 * interrupts physically disabled and remote cpus could deadlock
2435 * trying to send us an IPI.
2437 smp_startup_mask |= 1 << mycpu->gd_cpuid;
2441 * Interlock for finalization. Wait until mp_finish is non-zero,
2442 * then get the MP lock.
2444 * Note: We are in a critical section.
2446 * Note: We have to synchronize td_mpcount to our desired MP state
2447 * before calling cpu_try_mplock().
2449 * Note: we are the idle thread, we can only spin.
2451 * Note: The load fence is memory volatile and prevents the compiler
2452 * from improperly caching mp_finish, and the cpu from improperly
2455 while (mp_finish == 0)
2457 ++curthread->td_mpcount;
2458 while (cpu_try_mplock() == 0)
2461 if (cpu_feature & CPUID_TSC) {
2463 * The BSP is constantly updating tsc0_offset, figure out the
2464 * relative difference to synchronize ktrdump.
2466 tsc_offsets[mycpu->gd_cpuid] = rdtsc() - tsc0_offset;
2469 /* BSP may have changed PTD while we're waiting for the lock */
2472 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
2476 /* Build our map of 'other' CPUs. */
2477 mycpu->gd_other_cpus = smp_startup_mask & ~(1 << mycpu->gd_cpuid);
2479 kprintf("SMP: AP CPU #%d Launched!\n", mycpu->gd_cpuid);
2481 /* A quick check from sanity claus */
2482 apic_id = (apic_id_to_logical[(lapic.id & 0x0f000000) >> 24]);
2483 if (mycpu->gd_cpuid != apic_id) {
2484 kprintf("SMP: cpuid = %d\n", mycpu->gd_cpuid);
2485 kprintf("SMP: apic_id = %d\n", apic_id);
2486 kprintf("PTD[MPPTDI] = %p\n", (void *)PTD[MPPTDI]);
2487 panic("cpuid mismatch! boom!!");
2490 /* Init local apic for irq's */
2493 /* Set memory range attributes for this CPU to match the BSP */
2494 mem_range_AP_init();
2497 * Once we go active we must process any IPIQ messages that may
2498 * have been queued, because no actual IPI will occur until we
2499 * set our bit in the smp_active_mask. If we don't the IPI
2500 * message interlock could be left set which would also prevent
2503 * The idle loop doesn't expect the BGL to be held and while
2504 * lwkt_switch() normally cleans things up this is a special case
2505 * because we returning almost directly into the idle loop.
2507 * The idle thread is never placed on the runq, make sure
2508 * nothing we've done put it there.
2510 KKASSERT(curthread->td_mpcount == 1);
2511 smp_active_mask |= 1 << mycpu->gd_cpuid;
2514 * Enable interrupts here. idle_restore will also do it, but
2515 * doing it here lets us clean up any strays that got posted to
2516 * the CPU during the AP boot while we are still in a critical
2519 __asm __volatile("sti; pause; pause"::);
2520 mdcpu->gd_fpending = 0;
2521 mdcpu->gd_ipending = 0;
2523 initclocks_pcpu(); /* clock interrupts (via IPIs) */
2524 lwkt_process_ipiq();
2527 * Releasing the mp lock lets the BSP finish up the SMP init
2530 KKASSERT((curthread->td_flags & TDF_RUNQ) == 0);
2534 * Get SMP fully working before we start initializing devices.
2542 kprintf("Finish MP startup\n");
2543 if (cpu_feature & CPUID_TSC)
2544 tsc0_offset = rdtsc();
2547 while (smp_active_mask != smp_startup_mask) {
2549 if (cpu_feature & CPUID_TSC)
2550 tsc0_offset = rdtsc();
2552 while (try_mplock() == 0)
2555 kprintf("Active CPU Mask: %08x\n", smp_active_mask);
2558 SYSINIT(finishsmp, SI_BOOT2_FINISH_SMP, SI_ORDER_FIRST, ap_finish, NULL)
2561 cpu_send_ipiq(int dcpu)
2563 if ((1 << dcpu) & smp_active_mask)
2564 single_apic_ipi(dcpu, XIPIQ_OFFSET, APIC_DELMODE_FIXED);
2567 #if 0 /* single_apic_ipi_passive() not working yet */
2569 * Returns 0 on failure, 1 on success
2572 cpu_send_ipiq_passive(int dcpu)
2575 if ((1 << dcpu) & smp_active_mask) {
2576 r = single_apic_ipi_passive(dcpu, XIPIQ_OFFSET,
2577 APIC_DELMODE_FIXED);