mp_x86.c revision 121754
1/*- 2 * Copyright (c) 1996, by Steve Passe 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. The name of the developer may NOT be used to endorse or promote products 11 * derived from this software without specific prior written permission. 12 * 13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 16 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 23 * SUCH DAMAGE. 24 */ 25 26#include <sys/cdefs.h> 27__FBSDID("$FreeBSD: head/sys/i386/i386/mp_machdep.c 121754 2003-10-30 21:42:17Z jhb $"); 28 29#include "opt_cpu.h" 30#include "opt_kstack_pages.h" 31 32#ifdef SMP 33#include <machine/smptests.h> 34#else 35#if !defined(lint) 36#error 37#endif 38#endif 39 40#include <sys/param.h> 41#include <sys/systm.h> 42#include <sys/bus.h> 43#include <sys/cons.h> /* cngetc() */ 44#ifdef GPROF 45#include <sys/gmon.h> 46#endif 47#include <sys/kernel.h> 48#include <sys/ktr.h> 49#include <sys/lock.h> 50#include <sys/malloc.h> 51#include <sys/memrange.h> 52#include <sys/mutex.h> 53#include <sys/pcpu.h> 54#include <sys/proc.h> 55#include <sys/smp.h> 56#include <sys/sysctl.h> 57#include <sys/user.h> 58 59#include <vm/vm.h> 60#include <vm/vm_param.h> 61#include <vm/pmap.h> 62#include <vm/vm_kern.h> 63#include <vm/vm_extern.h> 64#include <vm/vm_map.h> 65 66#include <machine/apic.h> 67#include <machine/atomic.h> 68#include <machine/clock.h> 69#include <machine/cpu.h> 70#include <machine/cpufunc.h> 71#include <machine/mpapic.h> 72#include <machine/psl.h> 73#include <machine/segments.h> 74#include <machine/smp.h> 75#include <machine/smptests.h> /** TEST_DEFAULT_CONFIG, TEST_TEST1 */ 76#include <machine/tss.h> 77#include <machine/specialreg.h> 78#include <machine/privatespace.h> 79 80#if defined(APIC_IO) 81#include <machine/md_var.h> /* setidt() */ 82#include <i386/isa/icu.h> /* IPIs */ 83#include <i386/isa/intr_machdep.h> /* IPIs */ 84#endif /* APIC_IO */ 85 86#if defined(TEST_DEFAULT_CONFIG) 87#define MPFPS_MPFB1 TEST_DEFAULT_CONFIG 88#else 89#define MPFPS_MPFB1 mpfps->mpfb1 90#endif /* TEST_DEFAULT_CONFIG */ 91 92#define WARMBOOT_TARGET 0 93#define WARMBOOT_OFF (KERNBASE + 0x0467) 94#define WARMBOOT_SEG (KERNBASE + 0x0469) 95 96#ifdef PC98 97#define BIOS_BASE (0xe8000) 98#define BIOS_SIZE (0x18000) 99#else 100#define BIOS_BASE (0xf0000) 101#define BIOS_SIZE (0x10000) 102#endif 103#define BIOS_COUNT (BIOS_SIZE/4) 104 105#define CMOS_REG (0x70) 106#define CMOS_DATA (0x71) 107#define BIOS_RESET (0x0f) 108#define BIOS_WARM (0x0a) 109 110#define PROCENTRY_FLAG_EN 0x01 111#define PROCENTRY_FLAG_BP 0x02 112#define IOAPICENTRY_FLAG_EN 0x01 113 114 115/* MP Floating Pointer Structure */ 116typedef struct MPFPS { 117 char signature[4]; 118 void *pap; 119 u_char length; 120 u_char spec_rev; 121 u_char checksum; 122 u_char mpfb1; 123 u_char mpfb2; 124 u_char mpfb3; 125 u_char mpfb4; 126 u_char mpfb5; 127} *mpfps_t; 128 129/* MP Configuration Table Header */ 130typedef struct MPCTH { 131 char signature[4]; 132 u_short base_table_length; 133 u_char spec_rev; 134 u_char checksum; 135 u_char oem_id[8]; 136 u_char product_id[12]; 137 void *oem_table_pointer; 138 u_short oem_table_size; 139 u_short entry_count; 140 void *apic_address; 141 u_short extended_table_length; 142 u_char extended_table_checksum; 143 u_char reserved; 144} *mpcth_t; 145 146 147typedef struct PROCENTRY { 148 u_char type; 149 u_char apic_id; 150 u_char apic_version; 151 u_char cpu_flags; 152 u_long cpu_signature; 153 u_long feature_flags; 154 u_long reserved1; 155 u_long reserved2; 156} *proc_entry_ptr; 157 158typedef struct BUSENTRY { 159 u_char type; 160 u_char bus_id; 161 char bus_type[6]; 162} *bus_entry_ptr; 163 164typedef struct IOAPICENTRY { 165 u_char type; 166 u_char apic_id; 167 u_char apic_version; 168 u_char apic_flags; 169 void *apic_address; 170} *io_apic_entry_ptr; 171 172typedef struct INTENTRY { 173 u_char type; 174 u_char int_type; 175 u_short int_flags; 176 u_char src_bus_id; 177 u_char src_bus_irq; 178 u_char dst_apic_id; 179 u_char dst_apic_int; 180} *int_entry_ptr; 181 182/* descriptions of MP basetable entries */ 183typedef struct BASETABLE_ENTRY { 184 u_char type; 185 u_char length; 186 char name[16]; 187} basetable_entry; 188 189/* 190 * this code MUST be enabled here and in mpboot.s. 191 * it follows the very early stages of AP boot by placing values in CMOS ram. 192 * it NORMALLY will never be needed and thus the primitive method for enabling. 193 * 194#define CHECK_POINTS 195 */ 196 197#if defined(CHECK_POINTS) && !defined(PC98) 198#define CHECK_READ(A) (outb(CMOS_REG, (A)), inb(CMOS_DATA)) 199#define CHECK_WRITE(A,D) (outb(CMOS_REG, (A)), outb(CMOS_DATA, (D))) 200 201#define CHECK_INIT(D); \ 202 CHECK_WRITE(0x34, (D)); \ 203 CHECK_WRITE(0x35, (D)); \ 204 CHECK_WRITE(0x36, (D)); \ 205 CHECK_WRITE(0x37, (D)); \ 206 CHECK_WRITE(0x38, (D)); \ 207 CHECK_WRITE(0x39, (D)); 208 209#define CHECK_PRINT(S); \ 210 printf("%s: %d, %d, %d, %d, %d, %d\n", \ 211 (S), \ 212 CHECK_READ(0x34), \ 213 CHECK_READ(0x35), \ 214 CHECK_READ(0x36), \ 215 CHECK_READ(0x37), \ 216 CHECK_READ(0x38), \ 217 CHECK_READ(0x39)); 218 219#else /* CHECK_POINTS */ 220 221#define CHECK_INIT(D) 222#define CHECK_PRINT(S) 223 224#endif /* CHECK_POINTS */ 225 226/* 227 * Values to send to the POST hardware. 228 */ 229#define MP_BOOTADDRESS_POST 0x10 230#define MP_PROBE_POST 0x11 231#define MPTABLE_PASS1_POST 0x12 232 233#define MP_START_POST 0x13 234#define MP_ENABLE_POST 0x14 235#define MPTABLE_PASS2_POST 0x15 236 237#define START_ALL_APS_POST 0x16 238#define INSTALL_AP_TRAMP_POST 0x17 239#define START_AP_POST 0x18 240 241#define MP_ANNOUNCE_POST 0x19 242 243static int need_hyperthreading_fixup; 244static u_int logical_cpus; 245static u_int logical_cpus_mask; 246 247/* used to hold the AP's until we are ready to release them */ 248static struct mtx ap_boot_mtx; 249 250/** XXX FIXME: where does this really belong, isa.h/isa.c perhaps? */ 251int current_postcode; 252 253int bsp_apic_ready = 0; /* flags useability of BSP apic */ 254int mp_naps; /* # of Applications processors */ 255int mp_nbusses; /* # of busses */ 256int mp_napics; /* # of IO APICs */ 257int boot_cpu_id; /* designated BSP */ 258vm_offset_t cpu_apic_address; 259vm_offset_t io_apic_address[NAPICID]; /* NAPICID is more than enough */ 260extern int nkpt; 261 262u_int32_t cpu_apic_versions[MAXCPU]; 263u_int32_t *io_apic_versions; 264 265#ifdef APIC_INTR_REORDER 266struct { 267 volatile int *location; 268 int bit; 269} apic_isrbit_location[32]; 270#endif 271 272struct apic_intmapinfo int_to_apicintpin[APIC_INTMAPSIZE]; 273 274/* 275 * APIC ID logical/physical mapping structures. 276 * We oversize these to simplify boot-time config. 277 */ 278int cpu_num_to_apic_id[NAPICID]; 279int io_num_to_apic_id[NAPICID]; 280int apic_id_to_logical[NAPICID]; 281 282/* 283 * CPU topology map datastructures for HTT. 284 */ 285struct cpu_group mp_groups[NAPICID]; 286struct cpu_top mp_top; 287struct cpu_top *smp_topology; 288 289 290/* AP uses this during bootstrap. Do not staticize. */ 291char *bootSTK; 292static int bootAP; 293 294/* Hotwire a 0->4MB V==P mapping */ 295extern pt_entry_t *KPTphys; 296 297/* SMP page table page */ 298extern pt_entry_t *SMPpt; 299 300struct pcb stoppcbs[MAXCPU]; 301 302#ifdef APIC_IO 303/* Variables needed for SMP tlb shootdown. */ 304vm_offset_t smp_tlb_addr1; 305vm_offset_t smp_tlb_addr2; 306volatile int smp_tlb_wait; 307static struct mtx smp_tlb_mtx; 308#endif 309 310/* 311 * Local data and functions. 312 */ 313 314/* Set to 1 once we're ready to let the APs out of the pen. */ 315static volatile int aps_ready = 0; 316 317static int mp_capable; 318static u_int boot_address; 319static u_int base_memory; 320 321static int picmode; /* 0: virtual wire mode, 1: PIC mode */ 322static mpfps_t mpfps; 323static int search_for_sig(u_int32_t target, int count); 324static void mp_enable(u_int boot_addr); 325 326static void mptable_hyperthread_fixup(u_int id_mask); 327static void mptable_pass1(void); 328static int mptable_pass2(void); 329static void default_mp_table(int type); 330static void fix_mp_table(void); 331static void setup_apic_irq_mapping(void); 332static void init_locks(void); 333static int start_all_aps(u_int boot_addr); 334static void install_ap_tramp(u_int boot_addr); 335static int start_ap(int logicalCpu, u_int boot_addr); 336void ap_init(void); 337static int apic_int_is_bus_type(int intr, int bus_type); 338static void release_aps(void *dummy); 339 340static int hlt_cpus_mask; 341static int hlt_logical_cpus = 1; 342static struct sysctl_ctx_list logical_cpu_clist; 343 344/* 345 * initialize all the SMP locks 346 */ 347 348/* lock region used by kernel profiling */ 349int mcount_lock; 350 351#ifdef USE_COMLOCK 352/* locks com (tty) data/hardware accesses: a FASTINTR() */ 353struct mtx com_mtx; 354#endif /* USE_COMLOCK */ 355 356static void 357init_locks(void) 358{ 359 360#ifdef USE_COMLOCK 361 mtx_init(&com_mtx, "com", NULL, MTX_SPIN); 362#endif /* USE_COMLOCK */ 363#ifdef APIC_IO 364 mtx_init(&smp_tlb_mtx, "tlb", NULL, MTX_SPIN); 365#endif 366} 367 368/* 369 * Calculate usable address in base memory for AP trampoline code. 370 */ 371u_int 372mp_bootaddress(u_int basemem) 373{ 374 POSTCODE(MP_BOOTADDRESS_POST); 375 376 base_memory = basemem * 1024; /* convert to bytes */ 377 378 boot_address = base_memory & ~0xfff; /* round down to 4k boundary */ 379 if ((base_memory - boot_address) < bootMP_size) 380 boot_address -= 4096; /* not enough, lower by 4k */ 381 382 return boot_address; 383} 384 385 386/* 387 * Look for an Intel MP spec table (ie, SMP capable hardware). 388 */ 389void 390i386_mp_probe(void) 391{ 392 int x; 393 u_long segment; 394 u_int32_t target; 395 396 POSTCODE(MP_PROBE_POST); 397 398 /* see if EBDA exists */ 399 if ((segment = (u_long) * (u_short *) (KERNBASE + 0x40e)) != 0) { 400 /* search first 1K of EBDA */ 401 target = (u_int32_t) (segment << 4); 402 if ((x = search_for_sig(target, 1024 / 4)) >= 0) 403 goto found; 404 } else { 405 /* last 1K of base memory, effective 'top of base' passed in */ 406 target = (u_int32_t) (base_memory - 0x400); 407 if ((x = search_for_sig(target, 1024 / 4)) >= 0) 408 goto found; 409 } 410 411 /* search the BIOS */ 412 target = (u_int32_t) BIOS_BASE; 413 if ((x = search_for_sig(target, BIOS_COUNT)) >= 0) 414 goto found; 415 416 /* nothing found */ 417 mpfps = (mpfps_t)0; 418 mp_capable = 0; 419 return; 420 421found: 422 /* calculate needed resources */ 423 mpfps = (mpfps_t)x; 424 mptable_pass1(); 425 426 /* flag fact that we are running multiple processors */ 427 mp_capable = 1; 428} 429 430int 431cpu_mp_probe(void) 432{ 433 /* 434 * Record BSP in CPU map 435 * This is done here so that MBUF init code works correctly. 436 */ 437 all_cpus = 1; 438 439 return (mp_capable); 440} 441 442/* 443 * Initialize the SMP hardware and the APIC and start up the AP's. 444 */ 445void 446cpu_mp_start(void) 447{ 448 POSTCODE(MP_START_POST); 449 450 /* look for MP capable motherboard */ 451 if (mp_capable) 452 mp_enable(boot_address); 453 else 454 panic("MP hardware not found!"); 455 456 cpu_setregs(); 457} 458 459 460/* 461 * Print various information about the SMP system hardware and setup. 462 */ 463void 464cpu_mp_announce(void) 465{ 466 int x; 467 468 POSTCODE(MP_ANNOUNCE_POST); 469 470 printf(" cpu0 (BSP): apic id: %2d", CPU_TO_ID(0)); 471 printf(", version: 0x%08x", cpu_apic_versions[0]); 472 printf(", at 0x%08x\n", cpu_apic_address); 473 for (x = 1; x <= mp_naps; ++x) { 474 printf(" cpu%d (AP): apic id: %2d", x, CPU_TO_ID(x)); 475 printf(", version: 0x%08x", cpu_apic_versions[x]); 476 printf(", at 0x%08x\n", cpu_apic_address); 477 } 478 479#if defined(APIC_IO) 480 for (x = 0; x < mp_napics; ++x) { 481 printf(" io%d (APIC): apic id: %2d", x, IO_TO_ID(x)); 482 printf(", version: 0x%08x", io_apic_versions[x]); 483 printf(", at 0x%08x\n", io_apic_address[x]); 484 } 485#else 486 printf(" Warning: APIC I/O disabled\n"); 487#endif /* APIC_IO */ 488} 489 490/* 491 * AP cpu's call this to sync up protected mode. 492 */ 493void 494init_secondary(void) 495{ 496 int gsel_tss; 497 int x, myid = bootAP; 498 u_int cr0; 499 500 gdt_segs[GPRIV_SEL].ssd_base = (int) &SMP_prvspace[myid]; 501 gdt_segs[GPROC0_SEL].ssd_base = 502 (int) &SMP_prvspace[myid].pcpu.pc_common_tss; 503 SMP_prvspace[myid].pcpu.pc_prvspace = 504 &SMP_prvspace[myid].pcpu; 505 506 for (x = 0; x < NGDT; x++) { 507 ssdtosd(&gdt_segs[x], &gdt[myid * NGDT + x].sd); 508 } 509 510 r_gdt.rd_limit = NGDT * sizeof(gdt[0]) - 1; 511 r_gdt.rd_base = (int) &gdt[myid * NGDT]; 512 lgdt(&r_gdt); /* does magic intra-segment return */ 513 514 lidt(&r_idt); 515 516 lldt(_default_ldt); 517 PCPU_SET(currentldt, _default_ldt); 518 519 gsel_tss = GSEL(GPROC0_SEL, SEL_KPL); 520 gdt[myid * NGDT + GPROC0_SEL].sd.sd_type = SDT_SYS386TSS; 521 PCPU_SET(common_tss.tss_esp0, 0); /* not used until after switch */ 522 PCPU_SET(common_tss.tss_ss0, GSEL(GDATA_SEL, SEL_KPL)); 523 PCPU_SET(common_tss.tss_ioopt, (sizeof (struct i386tss)) << 16); 524 PCPU_SET(tss_gdt, &gdt[myid * NGDT + GPROC0_SEL].sd); 525 PCPU_SET(common_tssd, *PCPU_GET(tss_gdt)); 526 ltr(gsel_tss); 527 528 /* 529 * Set to a known state: 530 * Set by mpboot.s: CR0_PG, CR0_PE 531 * Set by cpu_setregs: CR0_NE, CR0_MP, CR0_TS, CR0_WP, CR0_AM 532 */ 533 cr0 = rcr0(); 534 cr0 &= ~(CR0_CD | CR0_NW | CR0_EM); 535 load_cr0(cr0); 536} 537 538 539#if defined(APIC_IO) 540/* 541 * Final configuration of the BSP's local APIC: 542 * - disable 'pic mode'. 543 * - disable 'virtual wire mode'. 544 * - enable NMI. 545 */ 546void 547bsp_apic_configure(void) 548{ 549 u_char byte; 550 u_int32_t temp; 551 552 /* leave 'pic mode' if necessary */ 553 if (picmode) { 554 outb(0x22, 0x70); /* select IMCR */ 555 byte = inb(0x23); /* current contents */ 556 byte |= 0x01; /* mask external INTR */ 557 outb(0x23, byte); /* disconnect 8259s/NMI */ 558 } 559 560 /* mask lint0 (the 8259 'virtual wire' connection) */ 561 temp = lapic.lvt_lint0; 562 temp |= APIC_LVT_M; /* set the mask */ 563 lapic.lvt_lint0 = temp; 564 565 /* setup lint1 to handle NMI */ 566 temp = lapic.lvt_lint1; 567 temp &= ~APIC_LVT_M; /* clear the mask */ 568 lapic.lvt_lint1 = temp; 569 570 if (bootverbose) 571 apic_dump("bsp_apic_configure()"); 572} 573#endif /* APIC_IO */ 574 575 576/******************************************************************* 577 * local functions and data 578 */ 579 580/* 581 * start the SMP system 582 */ 583static void 584mp_enable(u_int boot_addr) 585{ 586 int x; 587#if defined(APIC_IO) 588 int apic; 589 u_int ux; 590#endif /* APIC_IO */ 591 592 POSTCODE(MP_ENABLE_POST); 593 594 /* turn on 4MB of V == P addressing so we can get to MP table */ 595 *(int *)PTD = PG_V | PG_RW | ((uintptr_t)(void *)KPTphys & PG_FRAME); 596 invltlb(); 597 598 /* examine the MP table for needed info, uses physical addresses */ 599 x = mptable_pass2(); 600 601 *(int *)PTD = 0; 602 invltlb(); 603 604 /* can't process default configs till the CPU APIC is pmapped */ 605 if (x) 606 default_mp_table(x); 607 608 /* post scan cleanup */ 609 fix_mp_table(); 610 setup_apic_irq_mapping(); 611 612#if defined(APIC_IO) 613 614 /* fill the LOGICAL io_apic_versions table */ 615 for (apic = 0; apic < mp_napics; ++apic) { 616 ux = io_apic_read(apic, IOAPIC_VER); 617 io_apic_versions[apic] = ux; 618 io_apic_set_id(apic, IO_TO_ID(apic)); 619 } 620 621 /* program each IO APIC in the system */ 622 for (apic = 0; apic < mp_napics; ++apic) 623 if (io_apic_setup(apic) < 0) 624 panic("IO APIC setup failure"); 625 626 /* install a 'Spurious INTerrupt' vector */ 627 setidt(XSPURIOUSINT_OFFSET, Xspuriousint, 628 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 629 630 /* install an inter-CPU IPI for TLB invalidation */ 631 setidt(XINVLTLB_OFFSET, Xinvltlb, 632 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 633 setidt(XINVLPG_OFFSET, Xinvlpg, 634 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 635 setidt(XINVLRNG_OFFSET, Xinvlrng, 636 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 637 638 /* install an inter-CPU IPI for forwarding hardclock() */ 639 setidt(XHARDCLOCK_OFFSET, Xhardclock, 640 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 641 642 /* install an inter-CPU IPI for forwarding statclock() */ 643 setidt(XSTATCLOCK_OFFSET, Xstatclock, 644 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 645 646 /* install an inter-CPU IPI for lazy pmap release */ 647 setidt(XLAZYPMAP_OFFSET, Xlazypmap, 648 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 649 650 /* install an inter-CPU IPI for all-CPU rendezvous */ 651 setidt(XRENDEZVOUS_OFFSET, Xrendezvous, 652 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 653 654 /* install an inter-CPU IPI for forcing an additional software trap */ 655 setidt(XCPUAST_OFFSET, Xcpuast, 656 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 657 658 /* install an inter-CPU IPI for CPU stop/restart */ 659 setidt(XCPUSTOP_OFFSET, Xcpustop, 660 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 661 662#if defined(TEST_TEST1) 663 /* install a "fake hardware INTerrupt" vector */ 664 setidt(XTEST1_OFFSET, Xtest1, 665 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 666#endif /** TEST_TEST1 */ 667 668#endif /* APIC_IO */ 669 670 /* initialize all SMP locks */ 671 init_locks(); 672 673 /* start each Application Processor */ 674 start_all_aps(boot_addr); 675} 676 677 678/* 679 * look for the MP spec signature 680 */ 681 682/* string defined by the Intel MP Spec as identifying the MP table */ 683#define MP_SIG 0x5f504d5f /* _MP_ */ 684#define NEXT(X) ((X) += 4) 685static int 686search_for_sig(u_int32_t target, int count) 687{ 688 int x; 689 u_int32_t *addr = (u_int32_t *) (KERNBASE + target); 690 691 for (x = 0; x < count; NEXT(x)) 692 if (addr[x] == MP_SIG) 693 /* make array index a byte index */ 694 return (target + (x * sizeof(u_int32_t))); 695 696 return -1; 697} 698 699 700static basetable_entry basetable_entry_types[] = 701{ 702 {0, 20, "Processor"}, 703 {1, 8, "Bus"}, 704 {2, 8, "I/O APIC"}, 705 {3, 8, "I/O INT"}, 706 {4, 8, "Local INT"} 707}; 708 709typedef struct BUSDATA { 710 u_char bus_id; 711 enum busTypes bus_type; 712} bus_datum; 713 714typedef struct INTDATA { 715 u_char int_type; 716 u_short int_flags; 717 u_char src_bus_id; 718 u_char src_bus_irq; 719 u_char dst_apic_id; 720 u_char dst_apic_int; 721 u_char int_vector; 722} io_int, local_int; 723 724typedef struct BUSTYPENAME { 725 u_char type; 726 char name[7]; 727} bus_type_name; 728 729static bus_type_name bus_type_table[] = 730{ 731 {CBUS, "CBUS"}, 732 {CBUSII, "CBUSII"}, 733 {EISA, "EISA"}, 734 {MCA, "MCA"}, 735 {UNKNOWN_BUSTYPE, "---"}, 736 {ISA, "ISA"}, 737 {MCA, "MCA"}, 738 {UNKNOWN_BUSTYPE, "---"}, 739 {UNKNOWN_BUSTYPE, "---"}, 740 {UNKNOWN_BUSTYPE, "---"}, 741 {UNKNOWN_BUSTYPE, "---"}, 742 {UNKNOWN_BUSTYPE, "---"}, 743 {PCI, "PCI"}, 744 {UNKNOWN_BUSTYPE, "---"}, 745 {UNKNOWN_BUSTYPE, "---"}, 746 {UNKNOWN_BUSTYPE, "---"}, 747 {UNKNOWN_BUSTYPE, "---"}, 748 {XPRESS, "XPRESS"}, 749 {UNKNOWN_BUSTYPE, "---"} 750}; 751/* from MP spec v1.4, table 5-1 */ 752static int default_data[7][5] = 753{ 754/* nbus, id0, type0, id1, type1 */ 755 {1, 0, ISA, 255, 255}, 756 {1, 0, EISA, 255, 255}, 757 {1, 0, EISA, 255, 255}, 758 {1, 0, MCA, 255, 255}, 759 {2, 0, ISA, 1, PCI}, 760 {2, 0, EISA, 1, PCI}, 761 {2, 0, MCA, 1, PCI} 762}; 763 764 765/* the bus data */ 766static bus_datum *bus_data; 767 768/* the IO INT data, one entry per possible APIC INTerrupt */ 769static io_int *io_apic_ints; 770 771static int nintrs; 772 773static int processor_entry(proc_entry_ptr entry, int cpu); 774static int bus_entry(bus_entry_ptr entry, int bus); 775static int io_apic_entry(io_apic_entry_ptr entry, int apic); 776static int int_entry(int_entry_ptr entry, int intr); 777static int lookup_bus_type(char *name); 778 779 780/* 781 * 1st pass on motherboard's Intel MP specification table. 782 * 783 * initializes: 784 * mp_ncpus = 1 785 * 786 * determines: 787 * cpu_apic_address (common to all CPUs) 788 * io_apic_address[N] 789 * mp_naps 790 * mp_nbusses 791 * mp_napics 792 * nintrs 793 */ 794static void 795mptable_pass1(void) 796{ 797 int x; 798 mpcth_t cth; 799 int totalSize; 800 void* position; 801 int count; 802 int type; 803 u_int id_mask; 804 805 POSTCODE(MPTABLE_PASS1_POST); 806 807 /* clear various tables */ 808 for (x = 0; x < NAPICID; ++x) { 809 io_apic_address[x] = ~0; /* IO APIC address table */ 810 } 811 812 /* init everything to empty */ 813 mp_naps = 0; 814 mp_nbusses = 0; 815 mp_napics = 0; 816 nintrs = 0; 817 id_mask = 0; 818 819 /* check for use of 'default' configuration */ 820 if (MPFPS_MPFB1 != 0) { 821 /* use default addresses */ 822 cpu_apic_address = DEFAULT_APIC_BASE; 823 io_apic_address[0] = DEFAULT_IO_APIC_BASE; 824 825 /* fill in with defaults */ 826 mp_naps = 2; /* includes BSP */ 827 mp_maxid = 1; 828 mp_nbusses = default_data[MPFPS_MPFB1 - 1][0]; 829#if defined(APIC_IO) 830 mp_napics = 1; 831 nintrs = 16; 832#endif /* APIC_IO */ 833 } 834 else { 835 if ((cth = mpfps->pap) == 0) 836 panic("MP Configuration Table Header MISSING!"); 837 838 cpu_apic_address = (vm_offset_t) cth->apic_address; 839 840 /* walk the table, recording info of interest */ 841 totalSize = cth->base_table_length - sizeof(struct MPCTH); 842 position = (u_char *) cth + sizeof(struct MPCTH); 843 count = cth->entry_count; 844 845 while (count--) { 846 switch (type = *(u_char *) position) { 847 case 0: /* processor_entry */ 848 if (((proc_entry_ptr)position)->cpu_flags 849 & PROCENTRY_FLAG_EN) { 850 ++mp_naps; 851 mp_maxid++; 852 id_mask |= 1 << 853 ((proc_entry_ptr)position)->apic_id; 854 } 855 break; 856 case 1: /* bus_entry */ 857 ++mp_nbusses; 858 break; 859 case 2: /* io_apic_entry */ 860 if (((io_apic_entry_ptr)position)->apic_flags 861 & IOAPICENTRY_FLAG_EN) 862 io_apic_address[mp_napics++] = 863 (vm_offset_t)((io_apic_entry_ptr) 864 position)->apic_address; 865 break; 866 case 3: /* int_entry */ 867 ++nintrs; 868 break; 869 case 4: /* int_entry */ 870 break; 871 default: 872 panic("mpfps Base Table HOSED!"); 873 /* NOTREACHED */ 874 } 875 876 totalSize -= basetable_entry_types[type].length; 877 (u_char*)position += basetable_entry_types[type].length; 878 } 879 } 880 881 /* qualify the numbers */ 882 if (mp_naps > MAXCPU) { 883 printf("Warning: only using %d of %d available CPUs!\n", 884 MAXCPU, mp_naps); 885 mp_naps = MAXCPU; 886 } 887 888 /* See if we need to fixup HT logical CPUs. */ 889 mptable_hyperthread_fixup(id_mask); 890 891 /* 892 * Count the BSP. 893 * This is also used as a counter while starting the APs. 894 */ 895 mp_ncpus = 1; 896 897 --mp_naps; /* subtract the BSP */ 898} 899 900 901/* 902 * 2nd pass on motherboard's Intel MP specification table. 903 * 904 * sets: 905 * boot_cpu_id 906 * ID_TO_IO(N), phy APIC ID to log CPU/IO table 907 * CPU_TO_ID(N), logical CPU to APIC ID table 908 * IO_TO_ID(N), logical IO to APIC ID table 909 * bus_data[N] 910 * io_apic_ints[N] 911 */ 912static int 913mptable_pass2(void) 914{ 915 struct PROCENTRY proc; 916 int x; 917 mpcth_t cth; 918 int totalSize; 919 void* position; 920 int count; 921 int type; 922 int apic, bus, cpu, intr; 923 int i, j; 924 925 POSTCODE(MPTABLE_PASS2_POST); 926 927 /* Initialize fake proc entry for use with HT fixup. */ 928 bzero(&proc, sizeof(proc)); 929 proc.type = 0; 930 proc.cpu_flags = PROCENTRY_FLAG_EN; 931 932 MALLOC(io_apic_versions, u_int32_t *, sizeof(u_int32_t) * mp_napics, 933 M_DEVBUF, M_WAITOK); 934 MALLOC(ioapic, volatile ioapic_t **, sizeof(ioapic_t *) * mp_napics, 935 M_DEVBUF, M_WAITOK); 936 MALLOC(io_apic_ints, io_int *, sizeof(io_int) * (nintrs + 1), 937 M_DEVBUF, M_WAITOK); 938 MALLOC(bus_data, bus_datum *, sizeof(bus_datum) * mp_nbusses, 939 M_DEVBUF, M_WAITOK); 940 941 bzero(ioapic, sizeof(ioapic_t *) * mp_napics); 942 943 for (i = 0; i < mp_napics; i++) { 944 for (j = 0; j < mp_napics; j++) { 945 /* same page frame as a previous IO apic? */ 946 if (((vm_offset_t)SMPpt[NPTEPG-2-j] & PG_FRAME) == 947 (io_apic_address[i] & PG_FRAME)) { 948 ioapic[i] = (ioapic_t *)((u_int)SMP_prvspace 949 + (NPTEPG-2-j) * PAGE_SIZE 950 + (io_apic_address[i] & PAGE_MASK)); 951 break; 952 } 953 /* use this slot if available */ 954 if (((vm_offset_t)SMPpt[NPTEPG-2-j] & PG_FRAME) == 0) { 955 SMPpt[NPTEPG-2-j] = (pt_entry_t)(PG_V | PG_RW | 956 (io_apic_address[i] & PG_FRAME)); 957 ioapic[i] = (ioapic_t *)((u_int)SMP_prvspace 958 + (NPTEPG-2-j) * PAGE_SIZE 959 + (io_apic_address[i] & PAGE_MASK)); 960 break; 961 } 962 } 963 } 964 965 /* clear various tables */ 966 for (x = 0; x < NAPICID; ++x) { 967 ID_TO_IO(x) = -1; /* phy APIC ID to log CPU/IO table */ 968 CPU_TO_ID(x) = -1; /* logical CPU to APIC ID table */ 969 IO_TO_ID(x) = -1; /* logical IO to APIC ID table */ 970 } 971 972 /* clear bus data table */ 973 for (x = 0; x < mp_nbusses; ++x) 974 bus_data[x].bus_id = 0xff; 975 976 /* clear IO APIC INT table */ 977 for (x = 0; x < (nintrs + 1); ++x) { 978 io_apic_ints[x].int_type = 0xff; 979 io_apic_ints[x].int_vector = 0xff; 980 } 981 982 /* setup the cpu/apic mapping arrays */ 983 boot_cpu_id = -1; 984 985 /* record whether PIC or virtual-wire mode */ 986 picmode = (mpfps->mpfb2 & 0x80) ? 1 : 0; 987 988 /* check for use of 'default' configuration */ 989 if (MPFPS_MPFB1 != 0) 990 return MPFPS_MPFB1; /* return default configuration type */ 991 992 if ((cth = mpfps->pap) == 0) 993 panic("MP Configuration Table Header MISSING!"); 994 995 /* walk the table, recording info of interest */ 996 totalSize = cth->base_table_length - sizeof(struct MPCTH); 997 position = (u_char *) cth + sizeof(struct MPCTH); 998 count = cth->entry_count; 999 apic = bus = intr = 0; 1000 cpu = 1; /* pre-count the BSP */ 1001 1002 while (count--) { 1003 switch (type = *(u_char *) position) { 1004 case 0: 1005 if (processor_entry(position, cpu)) { 1006 if (logical_cpus != 0 && 1007 cpu % logical_cpus != 0) 1008 logical_cpus_mask |= (1 << cpu); 1009 ++cpu; 1010 } 1011 if (need_hyperthreading_fixup) { 1012 /* 1013 * Create fake mptable processor entries 1014 * and feed them to processor_entry() to 1015 * enumerate the logical CPUs. 1016 */ 1017 proc.apic_id = ((proc_entry_ptr)position)->apic_id; 1018 for (i = 1; i < logical_cpus; i++) { 1019 proc.apic_id++; 1020 (void)processor_entry(&proc, cpu); 1021 logical_cpus_mask |= (1 << cpu); 1022 cpu++; 1023 } 1024 } 1025 break; 1026 case 1: 1027 if (bus_entry(position, bus)) 1028 ++bus; 1029 break; 1030 case 2: 1031 if (io_apic_entry(position, apic)) 1032 ++apic; 1033 break; 1034 case 3: 1035 if (int_entry(position, intr)) 1036 ++intr; 1037 break; 1038 case 4: 1039 /* int_entry(position); */ 1040 break; 1041 default: 1042 panic("mpfps Base Table HOSED!"); 1043 /* NOTREACHED */ 1044 } 1045 1046 totalSize -= basetable_entry_types[type].length; 1047 (u_char *) position += basetable_entry_types[type].length; 1048 } 1049 1050 if (boot_cpu_id == -1) 1051 panic("NO BSP found!"); 1052 1053 /* report fact that its NOT a default configuration */ 1054 return 0; 1055} 1056 1057/* 1058 * Check if we should perform a hyperthreading "fix-up" to 1059 * enumerate any logical CPU's that aren't already listed 1060 * in the table. 1061 * 1062 * XXX: We assume that all of the physical CPUs in the 1063 * system have the same number of logical CPUs. 1064 * 1065 * XXX: We assume that APIC ID's are allocated such that 1066 * the APIC ID's for a physical processor are aligned 1067 * with the number of logical CPU's in the processor. 1068 */ 1069static void 1070mptable_hyperthread_fixup(u_int id_mask) 1071{ 1072 u_int i, id; 1073 int logical; 1074 1075 /* Nothing to do if there is no HTT support. */ 1076 if ((cpu_feature & CPUID_HTT) == 0) 1077 return; 1078 logical_cpus = (cpu_procinfo & CPUID_HTT_CORES) >> 16; 1079 if (logical_cpus <= 1) 1080 return; 1081 1082 /* 1083 * For each APIC ID of a CPU that is set in the mask, 1084 * scan the other candidate APIC ID's for this 1085 * physical processor. If any of those ID's are 1086 * already in the table, then kill the fixup. 1087 */ 1088 for (id = 0; id <= MAXCPU; id++) { 1089 if ((id_mask & 1 << id) == 0) 1090 continue; 1091 /* First, make sure we are on a logical_cpus boundary. */ 1092 if (id % logical_cpus != 0) 1093 return; 1094 for (i = id + 1; i < id + logical_cpus; i++) 1095 if ((id_mask & 1 << i) != 0) 1096 return; 1097 } 1098 1099 /* 1100 * Ok, the ID's checked out, so enable the fixup. We have to fixup 1101 * mp_naps and mp_maxid right now. 1102 */ 1103 need_hyperthreading_fixup = 1; 1104 mp_maxid *= logical_cpus; 1105 mp_naps *= logical_cpus; 1106 1107 /* 1108 * Now setup the cpu topology map. 1109 */ 1110 mp_top.ct_count = mp_naps / logical_cpus; 1111 mp_top.ct_group = mp_groups; 1112 1113 /* 1114 * The first logical id is directly after the last valid physical id. 1115 */ 1116 logical = mp_top.ct_count + 1; 1117 1118 for (i = 0; i < mp_top.ct_count; i++) { 1119 int j; 1120 1121 mp_groups[i].cg_mask = (1 << i); 1122 for (j = 1; j < logical_cpus; j++) 1123 mp_groups[i].cg_mask |= (1 << logical++); 1124 mp_groups[i].cg_count = logical_cpus; 1125 mp_groups[i].cg_children = 0; 1126 } 1127 1128 smp_topology = &mp_top; 1129} 1130 1131void 1132assign_apic_irq(int apic, int intpin, int irq) 1133{ 1134 int x; 1135 1136 if (int_to_apicintpin[irq].ioapic != -1) 1137 panic("assign_apic_irq: inconsistent table"); 1138 1139 int_to_apicintpin[irq].ioapic = apic; 1140 int_to_apicintpin[irq].int_pin = intpin; 1141 int_to_apicintpin[irq].apic_address = ioapic[apic]; 1142 int_to_apicintpin[irq].redirindex = IOAPIC_REDTBL + 2 * intpin; 1143 1144 for (x = 0; x < nintrs; x++) { 1145 if ((io_apic_ints[x].int_type == 0 || 1146 io_apic_ints[x].int_type == 3) && 1147 io_apic_ints[x].int_vector == 0xff && 1148 io_apic_ints[x].dst_apic_id == IO_TO_ID(apic) && 1149 io_apic_ints[x].dst_apic_int == intpin) 1150 io_apic_ints[x].int_vector = irq; 1151 } 1152} 1153 1154void 1155revoke_apic_irq(int irq) 1156{ 1157 int x; 1158 int oldapic; 1159 int oldintpin; 1160 1161 if (int_to_apicintpin[irq].ioapic == -1) 1162 panic("revoke_apic_irq: inconsistent table"); 1163 1164 oldapic = int_to_apicintpin[irq].ioapic; 1165 oldintpin = int_to_apicintpin[irq].int_pin; 1166 1167 int_to_apicintpin[irq].ioapic = -1; 1168 int_to_apicintpin[irq].int_pin = 0; 1169 int_to_apicintpin[irq].apic_address = NULL; 1170 int_to_apicintpin[irq].redirindex = 0; 1171 1172 for (x = 0; x < nintrs; x++) { 1173 if ((io_apic_ints[x].int_type == 0 || 1174 io_apic_ints[x].int_type == 3) && 1175 io_apic_ints[x].int_vector != 0xff && 1176 io_apic_ints[x].dst_apic_id == IO_TO_ID(oldapic) && 1177 io_apic_ints[x].dst_apic_int == oldintpin) 1178 io_apic_ints[x].int_vector = 0xff; 1179 } 1180} 1181 1182 1183static void 1184allocate_apic_irq(int intr) 1185{ 1186 int apic; 1187 int intpin; 1188 int irq; 1189 1190 if (io_apic_ints[intr].int_vector != 0xff) 1191 return; /* Interrupt handler already assigned */ 1192 1193 if (io_apic_ints[intr].int_type != 0 && 1194 (io_apic_ints[intr].int_type != 3 || 1195 (io_apic_ints[intr].dst_apic_id == IO_TO_ID(0) && 1196 io_apic_ints[intr].dst_apic_int == 0))) 1197 return; /* Not INT or ExtInt on != (0, 0) */ 1198 1199 irq = 0; 1200 while (irq < APIC_INTMAPSIZE && 1201 int_to_apicintpin[irq].ioapic != -1) 1202 irq++; 1203 1204 if (irq >= APIC_INTMAPSIZE) 1205 return; /* No free interrupt handlers */ 1206 1207 apic = ID_TO_IO(io_apic_ints[intr].dst_apic_id); 1208 intpin = io_apic_ints[intr].dst_apic_int; 1209 1210 assign_apic_irq(apic, intpin, irq); 1211 io_apic_setup_intpin(apic, intpin); 1212} 1213 1214 1215static void 1216swap_apic_id(int apic, int oldid, int newid) 1217{ 1218 int x; 1219 int oapic; 1220 1221 1222 if (oldid == newid) 1223 return; /* Nothing to do */ 1224 1225 printf("Changing APIC ID for IO APIC #%d from %d to %d in MP table\n", 1226 apic, oldid, newid); 1227 1228 /* Swap physical APIC IDs in interrupt entries */ 1229 for (x = 0; x < nintrs; x++) { 1230 if (io_apic_ints[x].dst_apic_id == oldid) 1231 io_apic_ints[x].dst_apic_id = newid; 1232 else if (io_apic_ints[x].dst_apic_id == newid) 1233 io_apic_ints[x].dst_apic_id = oldid; 1234 } 1235 1236 /* Swap physical APIC IDs in IO_TO_ID mappings */ 1237 for (oapic = 0; oapic < mp_napics; oapic++) 1238 if (IO_TO_ID(oapic) == newid) 1239 break; 1240 1241 if (oapic < mp_napics) { 1242 printf("Changing APIC ID for IO APIC #%d from " 1243 "%d to %d in MP table\n", 1244 oapic, newid, oldid); 1245 IO_TO_ID(oapic) = oldid; 1246 } 1247 IO_TO_ID(apic) = newid; 1248} 1249 1250 1251static void 1252fix_id_to_io_mapping(void) 1253{ 1254 int x; 1255 1256 for (x = 0; x < NAPICID; x++) 1257 ID_TO_IO(x) = -1; 1258 1259 for (x = 0; x <= mp_naps; x++) 1260 if (CPU_TO_ID(x) < NAPICID) 1261 ID_TO_IO(CPU_TO_ID(x)) = x; 1262 1263 for (x = 0; x < mp_napics; x++) 1264 if (IO_TO_ID(x) < NAPICID) 1265 ID_TO_IO(IO_TO_ID(x)) = x; 1266} 1267 1268 1269static int 1270first_free_apic_id(void) 1271{ 1272 int freeid, x; 1273 1274 for (freeid = 0; freeid < NAPICID; freeid++) { 1275 for (x = 0; x <= mp_naps; x++) 1276 if (CPU_TO_ID(x) == freeid) 1277 break; 1278 if (x <= mp_naps) 1279 continue; 1280 for (x = 0; x < mp_napics; x++) 1281 if (IO_TO_ID(x) == freeid) 1282 break; 1283 if (x < mp_napics) 1284 continue; 1285 return freeid; 1286 } 1287 return freeid; 1288} 1289 1290 1291static int 1292io_apic_id_acceptable(int apic, int id) 1293{ 1294 int cpu; /* Logical CPU number */ 1295 int oapic; /* Logical IO APIC number for other IO APIC */ 1296 1297 if (id >= NAPICID) 1298 return 0; /* Out of range */ 1299 1300 for (cpu = 0; cpu <= mp_naps; cpu++) 1301 if (CPU_TO_ID(cpu) == id) 1302 return 0; /* Conflict with CPU */ 1303 1304 for (oapic = 0; oapic < mp_napics && oapic < apic; oapic++) 1305 if (IO_TO_ID(oapic) == id) 1306 return 0; /* Conflict with other APIC */ 1307 1308 return 1; /* ID is acceptable for IO APIC */ 1309} 1310 1311 1312/* 1313 * parse an Intel MP specification table 1314 */ 1315static void 1316fix_mp_table(void) 1317{ 1318 int x; 1319 int id; 1320 int bus_0 = 0; /* Stop GCC warning */ 1321 int bus_pci = 0; /* Stop GCC warning */ 1322 int num_pci_bus; 1323 int apic; /* IO APIC unit number */ 1324 int freeid; /* Free physical APIC ID */ 1325 int physid; /* Current physical IO APIC ID */ 1326 1327 /* 1328 * Fix mis-numbering of the PCI bus and its INT entries if the BIOS 1329 * did it wrong. The MP spec says that when more than 1 PCI bus 1330 * exists the BIOS must begin with bus entries for the PCI bus and use 1331 * actual PCI bus numbering. This implies that when only 1 PCI bus 1332 * exists the BIOS can choose to ignore this ordering, and indeed many 1333 * MP motherboards do ignore it. This causes a problem when the PCI 1334 * sub-system makes requests of the MP sub-system based on PCI bus 1335 * numbers. So here we look for the situation and renumber the 1336 * busses and associated INTs in an effort to "make it right". 1337 */ 1338 1339 /* find bus 0, PCI bus, count the number of PCI busses */ 1340 for (num_pci_bus = 0, x = 0; x < mp_nbusses; ++x) { 1341 if (bus_data[x].bus_id == 0) { 1342 bus_0 = x; 1343 } 1344 if (bus_data[x].bus_type == PCI) { 1345 ++num_pci_bus; 1346 bus_pci = x; 1347 } 1348 } 1349 /* 1350 * bus_0 == slot of bus with ID of 0 1351 * bus_pci == slot of last PCI bus encountered 1352 */ 1353 1354 /* check the 1 PCI bus case for sanity */ 1355 /* if it is number 0 all is well */ 1356 if (num_pci_bus == 1 && 1357 bus_data[bus_pci].bus_id != 0) { 1358 1359 /* mis-numbered, swap with whichever bus uses slot 0 */ 1360 1361 /* swap the bus entry types */ 1362 bus_data[bus_pci].bus_type = bus_data[bus_0].bus_type; 1363 bus_data[bus_0].bus_type = PCI; 1364 1365 /* swap each relavant INTerrupt entry */ 1366 id = bus_data[bus_pci].bus_id; 1367 for (x = 0; x < nintrs; ++x) { 1368 if (io_apic_ints[x].src_bus_id == id) { 1369 io_apic_ints[x].src_bus_id = 0; 1370 } 1371 else if (io_apic_ints[x].src_bus_id == 0) { 1372 io_apic_ints[x].src_bus_id = id; 1373 } 1374 } 1375 } 1376 1377 /* Assign IO APIC IDs. 1378 * 1379 * First try the existing ID. If a conflict is detected, try 1380 * the ID in the MP table. If a conflict is still detected, find 1381 * a free id. 1382 * 1383 * We cannot use the ID_TO_IO table before all conflicts has been 1384 * resolved and the table has been corrected. 1385 */ 1386 for (apic = 0; apic < mp_napics; ++apic) { /* For all IO APICs */ 1387 1388 /* First try to use the value set by the BIOS */ 1389 physid = io_apic_get_id(apic); 1390 if (io_apic_id_acceptable(apic, physid)) { 1391 if (IO_TO_ID(apic) != physid) 1392 swap_apic_id(apic, IO_TO_ID(apic), physid); 1393 continue; 1394 } 1395 1396 /* Then check if the value in the MP table is acceptable */ 1397 if (io_apic_id_acceptable(apic, IO_TO_ID(apic))) 1398 continue; 1399 1400 /* Last resort, find a free APIC ID and use it */ 1401 freeid = first_free_apic_id(); 1402 if (freeid >= NAPICID) 1403 panic("No free physical APIC IDs found"); 1404 1405 if (io_apic_id_acceptable(apic, freeid)) { 1406 swap_apic_id(apic, IO_TO_ID(apic), freeid); 1407 continue; 1408 } 1409 panic("Free physical APIC ID not usable"); 1410 } 1411 fix_id_to_io_mapping(); 1412 1413 /* detect and fix broken Compaq MP table */ 1414 if (apic_int_type(0, 0) == -1) { 1415 printf("APIC_IO: MP table broken: 8259->APIC entry missing!\n"); 1416 io_apic_ints[nintrs].int_type = 3; /* ExtInt */ 1417 io_apic_ints[nintrs].int_vector = 0xff; /* Unassigned */ 1418 /* XXX fixme, set src bus id etc, but it doesn't seem to hurt */ 1419 io_apic_ints[nintrs].dst_apic_id = IO_TO_ID(0); 1420 io_apic_ints[nintrs].dst_apic_int = 0; /* Pin 0 */ 1421 nintrs++; 1422 } 1423} 1424 1425 1426/* Assign low level interrupt handlers */ 1427static void 1428setup_apic_irq_mapping(void) 1429{ 1430 int x; 1431 int int_vector; 1432 1433 /* Clear array */ 1434 for (x = 0; x < APIC_INTMAPSIZE; x++) { 1435 int_to_apicintpin[x].ioapic = -1; 1436 int_to_apicintpin[x].int_pin = 0; 1437 int_to_apicintpin[x].apic_address = NULL; 1438 int_to_apicintpin[x].redirindex = 0; 1439 } 1440 1441 /* First assign ISA/EISA interrupts */ 1442 for (x = 0; x < nintrs; x++) { 1443 int_vector = io_apic_ints[x].src_bus_irq; 1444 if (int_vector < APIC_INTMAPSIZE && 1445 io_apic_ints[x].int_vector == 0xff && 1446 int_to_apicintpin[int_vector].ioapic == -1 && 1447 (apic_int_is_bus_type(x, ISA) || 1448 apic_int_is_bus_type(x, EISA)) && 1449 io_apic_ints[x].int_type == 0) { 1450 assign_apic_irq(ID_TO_IO(io_apic_ints[x].dst_apic_id), 1451 io_apic_ints[x].dst_apic_int, 1452 int_vector); 1453 } 1454 } 1455 1456 /* Assign ExtInt entry if no ISA/EISA interrupt 0 entry */ 1457 for (x = 0; x < nintrs; x++) { 1458 if (io_apic_ints[x].dst_apic_int == 0 && 1459 io_apic_ints[x].dst_apic_id == IO_TO_ID(0) && 1460 io_apic_ints[x].int_vector == 0xff && 1461 int_to_apicintpin[0].ioapic == -1 && 1462 io_apic_ints[x].int_type == 3) { 1463 assign_apic_irq(0, 0, 0); 1464 break; 1465 } 1466 } 1467 /* PCI interrupt assignment is deferred */ 1468} 1469 1470 1471static int 1472processor_entry(proc_entry_ptr entry, int cpu) 1473{ 1474 /* check for usability */ 1475 if (!(entry->cpu_flags & PROCENTRY_FLAG_EN)) 1476 return 0; 1477 1478 if(entry->apic_id >= NAPICID) 1479 panic("CPU APIC ID out of range (0..%d)", NAPICID - 1); 1480 /* check for BSP flag */ 1481 if (entry->cpu_flags & PROCENTRY_FLAG_BP) { 1482 boot_cpu_id = entry->apic_id; 1483 CPU_TO_ID(0) = entry->apic_id; 1484 ID_TO_CPU(entry->apic_id) = 0; 1485 return 0; /* its already been counted */ 1486 } 1487 1488 /* add another AP to list, if less than max number of CPUs */ 1489 else if (cpu < MAXCPU) { 1490 CPU_TO_ID(cpu) = entry->apic_id; 1491 ID_TO_CPU(entry->apic_id) = cpu; 1492 return 1; 1493 } 1494 1495 return 0; 1496} 1497 1498 1499static int 1500bus_entry(bus_entry_ptr entry, int bus) 1501{ 1502 int x; 1503 char c, name[8]; 1504 1505 /* encode the name into an index */ 1506 for (x = 0; x < 6; ++x) { 1507 if ((c = entry->bus_type[x]) == ' ') 1508 break; 1509 name[x] = c; 1510 } 1511 name[x] = '\0'; 1512 1513 if ((x = lookup_bus_type(name)) == UNKNOWN_BUSTYPE) 1514 panic("unknown bus type: '%s'", name); 1515 1516 bus_data[bus].bus_id = entry->bus_id; 1517 bus_data[bus].bus_type = x; 1518 1519 return 1; 1520} 1521 1522 1523static int 1524io_apic_entry(io_apic_entry_ptr entry, int apic) 1525{ 1526 if (!(entry->apic_flags & IOAPICENTRY_FLAG_EN)) 1527 return 0; 1528 1529 IO_TO_ID(apic) = entry->apic_id; 1530 if (entry->apic_id < NAPICID) 1531 ID_TO_IO(entry->apic_id) = apic; 1532 1533 return 1; 1534} 1535 1536 1537static int 1538lookup_bus_type(char *name) 1539{ 1540 int x; 1541 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; 1545 1546 return UNKNOWN_BUSTYPE; 1547} 1548 1549 1550static int 1551int_entry(int_entry_ptr entry, int intr) 1552{ 1553 int apic; 1554 1555 io_apic_ints[intr].int_type = entry->int_type; 1556 io_apic_ints[intr].int_flags = entry->int_flags; 1557 io_apic_ints[intr].src_bus_id = entry->src_bus_id; 1558 io_apic_ints[intr].src_bus_irq = entry->src_bus_irq; 1559 if (entry->dst_apic_id == 255) { 1560 /* This signal goes to all IO APICS. Select an IO APIC 1561 with sufficient number of interrupt pins */ 1562 for (apic = 0; apic < mp_napics; apic++) 1563 if (((io_apic_read(apic, IOAPIC_VER) & 1564 IOART_VER_MAXREDIR) >> MAXREDIRSHIFT) >= 1565 entry->dst_apic_int) 1566 break; 1567 if (apic < mp_napics) 1568 io_apic_ints[intr].dst_apic_id = IO_TO_ID(apic); 1569 else 1570 io_apic_ints[intr].dst_apic_id = entry->dst_apic_id; 1571 } else 1572 io_apic_ints[intr].dst_apic_id = entry->dst_apic_id; 1573 io_apic_ints[intr].dst_apic_int = entry->dst_apic_int; 1574 1575 return 1; 1576} 1577 1578 1579static int 1580apic_int_is_bus_type(int intr, int bus_type) 1581{ 1582 int bus; 1583 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)) 1587 return 1; 1588 1589 return 0; 1590} 1591 1592 1593/* 1594 * Given a traditional ISA INT mask, return an APIC mask. 1595 */ 1596u_int 1597isa_apic_mask(u_int isa_mask) 1598{ 1599 int isa_irq; 1600 int apic_pin; 1601 1602#if defined(SKIP_IRQ15_REDIRECT) 1603 if (isa_mask == (1 << 15)) { 1604 printf("skipping ISA IRQ15 redirect\n"); 1605 return isa_mask; 1606 } 1607#endif /* SKIP_IRQ15_REDIRECT */ 1608 1609 isa_irq = ffs(isa_mask); /* find its bit position */ 1610 if (isa_irq == 0) /* doesn't exist */ 1611 return 0; 1612 --isa_irq; /* make it zero based */ 1613 1614 apic_pin = isa_apic_irq(isa_irq); /* look for APIC connection */ 1615 if (apic_pin == -1) 1616 return 0; 1617 1618 return (1 << apic_pin); /* convert pin# to a mask */ 1619} 1620 1621 1622/* 1623 * Determine which APIC pin an ISA/EISA INT is attached to. 1624 */ 1625#define INTTYPE(I) (io_apic_ints[(I)].int_type) 1626#define INTPIN(I) (io_apic_ints[(I)].dst_apic_int) 1627#define INTIRQ(I) (io_apic_ints[(I)].int_vector) 1628#define INTAPIC(I) (ID_TO_IO(io_apic_ints[(I)].dst_apic_id)) 1629 1630#define SRCBUSIRQ(I) (io_apic_ints[(I)].src_bus_irq) 1631int 1632isa_apic_irq(int isa_irq) 1633{ 1634 int intr; 1635 1636 for (intr = 0; intr < nintrs; ++intr) { /* check each record */ 1637 if (INTTYPE(intr) == 0) { /* standard INT */ 1638 if (SRCBUSIRQ(intr) == isa_irq) { 1639 if (apic_int_is_bus_type(intr, ISA) || 1640 apic_int_is_bus_type(intr, EISA)) { 1641 if (INTIRQ(intr) == 0xff) 1642 return -1; /* unassigned */ 1643 return INTIRQ(intr); /* found */ 1644 } 1645 } 1646 } 1647 } 1648 return -1; /* NOT found */ 1649} 1650 1651 1652/* 1653 * Determine which APIC pin a PCI INT is attached to. 1654 */ 1655#define SRCBUSID(I) (io_apic_ints[(I)].src_bus_id) 1656#define SRCBUSDEVICE(I) ((io_apic_ints[(I)].src_bus_irq >> 2) & 0x1f) 1657#define SRCBUSLINE(I) (io_apic_ints[(I)].src_bus_irq & 0x03) 1658int 1659pci_apic_irq(int pciBus, int pciDevice, int pciInt) 1660{ 1661 int intr; 1662 1663 --pciInt; /* zero based */ 1664 1665 for (intr = 0; intr < nintrs; ++intr) /* check each record */ 1666 if ((INTTYPE(intr) == 0) /* standard INT */ 1667 && (SRCBUSID(intr) == pciBus) 1668 && (SRCBUSDEVICE(intr) == pciDevice) 1669 && (SRCBUSLINE(intr) == pciInt)) /* a candidate IRQ */ 1670 if (apic_int_is_bus_type(intr, PCI)) { 1671 if (INTIRQ(intr) == 0xff) 1672 allocate_apic_irq(intr); 1673 if (INTIRQ(intr) == 0xff) 1674 return -1; /* unassigned */ 1675 return INTIRQ(intr); /* exact match */ 1676 } 1677 1678 return -1; /* NOT found */ 1679} 1680 1681int 1682next_apic_irq(int irq) 1683{ 1684 int intr, ointr; 1685 int bus, bustype; 1686 1687 bus = 0; 1688 bustype = 0; 1689 for (intr = 0; intr < nintrs; intr++) { 1690 if (INTIRQ(intr) != irq || INTTYPE(intr) != 0) 1691 continue; 1692 bus = SRCBUSID(intr); 1693 bustype = apic_bus_type(bus); 1694 if (bustype != ISA && 1695 bustype != EISA && 1696 bustype != PCI) 1697 continue; 1698 break; 1699 } 1700 if (intr >= nintrs) { 1701 return -1; 1702 } 1703 for (ointr = intr + 1; ointr < nintrs; ointr++) { 1704 if (INTTYPE(ointr) != 0) 1705 continue; 1706 if (bus != SRCBUSID(ointr)) 1707 continue; 1708 if (bustype == PCI) { 1709 if (SRCBUSDEVICE(intr) != SRCBUSDEVICE(ointr)) 1710 continue; 1711 if (SRCBUSLINE(intr) != SRCBUSLINE(ointr)) 1712 continue; 1713 } 1714 if (bustype == ISA || bustype == EISA) { 1715 if (SRCBUSIRQ(intr) != SRCBUSIRQ(ointr)) 1716 continue; 1717 } 1718 if (INTPIN(intr) == INTPIN(ointr)) 1719 continue; 1720 break; 1721 } 1722 if (ointr >= nintrs) { 1723 return -1; 1724 } 1725 return INTIRQ(ointr); 1726} 1727#undef SRCBUSLINE 1728#undef SRCBUSDEVICE 1729#undef SRCBUSID 1730#undef SRCBUSIRQ 1731 1732#undef INTPIN 1733#undef INTIRQ 1734#undef INTAPIC 1735#undef INTTYPE 1736 1737 1738/* 1739 * Reprogram the MB chipset to NOT redirect an ISA INTerrupt. 1740 * 1741 * XXX FIXME: 1742 * Exactly what this means is unclear at this point. It is a solution 1743 * for motherboards that redirect the MBIRQ0 pin. Generically a motherboard 1744 * could route any of the ISA INTs to upper (>15) IRQ values. But most would 1745 * NOT be redirected via MBIRQ0, thus "undirect()ing" them would NOT be an 1746 * option. 1747 */ 1748int 1749undirect_isa_irq(int rirq) 1750{ 1751#if defined(READY) 1752 if (bootverbose) 1753 printf("Freeing redirected ISA irq %d.\n", rirq); 1754 /** FIXME: tickle the MB redirector chip */ 1755 return -1; 1756#else 1757 if (bootverbose) 1758 printf("Freeing (NOT implemented) redirected ISA irq %d.\n", rirq); 1759 return 0; 1760#endif /* READY */ 1761} 1762 1763 1764/* 1765 * Reprogram the MB chipset to NOT redirect a PCI INTerrupt 1766 */ 1767int 1768undirect_pci_irq(int rirq) 1769{ 1770#if defined(READY) 1771 if (bootverbose) 1772 printf("Freeing redirected PCI irq %d.\n", rirq); 1773 1774 /** FIXME: tickle the MB redirector chip */ 1775 return -1; 1776#else 1777 if (bootverbose) 1778 printf("Freeing (NOT implemented) redirected PCI irq %d.\n", 1779 rirq); 1780 return 0; 1781#endif /* READY */ 1782} 1783 1784 1785/* 1786 * given a bus ID, return: 1787 * the bus type if found 1788 * -1 if NOT found 1789 */ 1790int 1791apic_bus_type(int id) 1792{ 1793 int x; 1794 1795 for (x = 0; x < mp_nbusses; ++x) 1796 if (bus_data[x].bus_id == id) 1797 return bus_data[x].bus_type; 1798 1799 return -1; 1800} 1801 1802 1803/* 1804 * given a LOGICAL APIC# and pin#, return: 1805 * the associated src bus ID if found 1806 * -1 if NOT found 1807 */ 1808int 1809apic_src_bus_id(int apic, int pin) 1810{ 1811 int x; 1812 1813 /* search each of the possible INTerrupt sources */ 1814 for (x = 0; x < nintrs; ++x) 1815 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) && 1816 (pin == io_apic_ints[x].dst_apic_int)) 1817 return (io_apic_ints[x].src_bus_id); 1818 1819 return -1; /* NOT found */ 1820} 1821 1822 1823/* 1824 * given a LOGICAL APIC# and pin#, return: 1825 * the associated src bus IRQ if found 1826 * -1 if NOT found 1827 */ 1828int 1829apic_src_bus_irq(int apic, int pin) 1830{ 1831 int x; 1832 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].src_bus_irq); 1837 1838 return -1; /* NOT found */ 1839} 1840 1841 1842/* 1843 * given a LOGICAL APIC# and pin#, return: 1844 * the associated INTerrupt type if found 1845 * -1 if NOT found 1846 */ 1847int 1848apic_int_type(int apic, int pin) 1849{ 1850 int x; 1851 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_type); 1857 1858 return -1; /* NOT found */ 1859} 1860 1861int 1862apic_irq(int apic, int pin) 1863{ 1864 int x; 1865 int res; 1866 1867 for (x = 0; x < nintrs; ++x) 1868 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) && 1869 (pin == io_apic_ints[x].dst_apic_int)) { 1870 res = io_apic_ints[x].int_vector; 1871 if (res == 0xff) 1872 return -1; 1873 if (apic != int_to_apicintpin[res].ioapic) 1874 panic("apic_irq: inconsistent table"); 1875 if (pin != int_to_apicintpin[res].int_pin) 1876 panic("apic_irq inconsistent table (2)"); 1877 return res; 1878 } 1879 return -1; 1880} 1881 1882 1883/* 1884 * given a LOGICAL APIC# and pin#, return: 1885 * the associated trigger mode if found 1886 * -1 if NOT found 1887 */ 1888int 1889apic_trigger(int apic, int pin) 1890{ 1891 int x; 1892 1893 /* search each of the possible INTerrupt sources */ 1894 for (x = 0; x < nintrs; ++x) 1895 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) && 1896 (pin == io_apic_ints[x].dst_apic_int)) 1897 return ((io_apic_ints[x].int_flags >> 2) & 0x03); 1898 1899 return -1; /* NOT found */ 1900} 1901 1902 1903/* 1904 * given a LOGICAL APIC# and pin#, return: 1905 * the associated 'active' level if found 1906 * -1 if NOT found 1907 */ 1908int 1909apic_polarity(int apic, int pin) 1910{ 1911 int x; 1912 1913 /* search each of the possible INTerrupt sources */ 1914 for (x = 0; x < nintrs; ++x) 1915 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) && 1916 (pin == io_apic_ints[x].dst_apic_int)) 1917 return (io_apic_ints[x].int_flags & 0x03); 1918 1919 return -1; /* NOT found */ 1920} 1921 1922 1923/* 1924 * set data according to MP defaults 1925 * FIXME: probably not complete yet... 1926 */ 1927static void 1928default_mp_table(int type) 1929{ 1930 int ap_cpu_id; 1931#if defined(APIC_IO) 1932 int io_apic_id; 1933 int pin; 1934#endif /* APIC_IO */ 1935 1936#if 0 1937 printf(" MP default config type: %d\n", type); 1938 switch (type) { 1939 case 1: 1940 printf(" bus: ISA, APIC: 82489DX\n"); 1941 break; 1942 case 2: 1943 printf(" bus: EISA, APIC: 82489DX\n"); 1944 break; 1945 case 3: 1946 printf(" bus: EISA, APIC: 82489DX\n"); 1947 break; 1948 case 4: 1949 printf(" bus: MCA, APIC: 82489DX\n"); 1950 break; 1951 case 5: 1952 printf(" bus: ISA+PCI, APIC: Integrated\n"); 1953 break; 1954 case 6: 1955 printf(" bus: EISA+PCI, APIC: Integrated\n"); 1956 break; 1957 case 7: 1958 printf(" bus: MCA+PCI, APIC: Integrated\n"); 1959 break; 1960 default: 1961 printf(" future type\n"); 1962 break; 1963 /* NOTREACHED */ 1964 } 1965#endif /* 0 */ 1966 1967 boot_cpu_id = (lapic.id & APIC_ID_MASK) >> 24; 1968 ap_cpu_id = (boot_cpu_id == 0) ? 1 : 0; 1969 1970 /* BSP */ 1971 CPU_TO_ID(0) = boot_cpu_id; 1972 ID_TO_CPU(boot_cpu_id) = 0; 1973 1974 /* one and only AP */ 1975 CPU_TO_ID(1) = ap_cpu_id; 1976 ID_TO_CPU(ap_cpu_id) = 1; 1977 1978#if defined(APIC_IO) 1979 /* one and only IO APIC */ 1980 io_apic_id = (io_apic_read(0, IOAPIC_ID) & APIC_ID_MASK) >> 24; 1981 1982 /* 1983 * sanity check, refer to MP spec section 3.6.6, last paragraph 1984 * necessary as some hardware isn't properly setting up the IO APIC 1985 */ 1986#if defined(REALLY_ANAL_IOAPICID_VALUE) 1987 if (io_apic_id != 2) { 1988#else 1989 if ((io_apic_id == 0) || (io_apic_id == 1) || (io_apic_id == 15)) { 1990#endif /* REALLY_ANAL_IOAPICID_VALUE */ 1991 io_apic_set_id(0, 2); 1992 io_apic_id = 2; 1993 } 1994 IO_TO_ID(0) = io_apic_id; 1995 ID_TO_IO(io_apic_id) = 0; 1996#endif /* APIC_IO */ 1997 1998 /* fill out bus entries */ 1999 switch (type) { 2000 case 1: 2001 case 2: 2002 case 3: 2003 case 4: 2004 case 5: 2005 case 6: 2006 case 7: 2007 bus_data[0].bus_id = default_data[type - 1][1]; 2008 bus_data[0].bus_type = default_data[type - 1][2]; 2009 bus_data[1].bus_id = default_data[type - 1][3]; 2010 bus_data[1].bus_type = default_data[type - 1][4]; 2011 break; 2012 2013 /* case 4: case 7: MCA NOT supported */ 2014 default: /* illegal/reserved */ 2015 panic("BAD default MP config: %d", type); 2016 /* NOTREACHED */ 2017 } 2018 2019#if defined(APIC_IO) 2020 /* general cases from MP v1.4, table 5-2 */ 2021 for (pin = 0; pin < 16; ++pin) { 2022 io_apic_ints[pin].int_type = 0; 2023 io_apic_ints[pin].int_flags = 0x05; /* edge/active-hi */ 2024 io_apic_ints[pin].src_bus_id = 0; 2025 io_apic_ints[pin].src_bus_irq = pin; /* IRQ2 caught below */ 2026 io_apic_ints[pin].dst_apic_id = io_apic_id; 2027 io_apic_ints[pin].dst_apic_int = pin; /* 1-to-1 */ 2028 } 2029 2030 /* special cases from MP v1.4, table 5-2 */ 2031 if (type == 2) { 2032 io_apic_ints[2].int_type = 0xff; /* N/C */ 2033 io_apic_ints[13].int_type = 0xff; /* N/C */ 2034#if !defined(APIC_MIXED_MODE) 2035 /** FIXME: ??? */ 2036 panic("sorry, can't support type 2 default yet"); 2037#endif /* APIC_MIXED_MODE */ 2038 } 2039 else 2040 io_apic_ints[2].src_bus_irq = 0; /* ISA IRQ0 is on APIC INT 2 */ 2041 2042 if (type == 7) 2043 io_apic_ints[0].int_type = 0xff; /* N/C */ 2044 else 2045 io_apic_ints[0].int_type = 3; /* vectored 8259 */ 2046#endif /* APIC_IO */ 2047} 2048 2049 2050/* 2051 * start each AP in our list 2052 */ 2053static int 2054start_all_aps(u_int boot_addr) 2055{ 2056 int x, i, pg; 2057#ifndef PC98 2058 u_char mpbiosreason; 2059#endif 2060 u_long mpbioswarmvec; 2061 struct pcpu *pc; 2062 char *stack; 2063 uintptr_t kptbase; 2064 2065 POSTCODE(START_ALL_APS_POST); 2066 2067 mtx_init(&ap_boot_mtx, "ap boot", NULL, MTX_SPIN); 2068 2069 /* initialize BSP's local APIC */ 2070 apic_initialize(); 2071 bsp_apic_ready = 1; 2072 2073 /* install the AP 1st level boot code */ 2074 install_ap_tramp(boot_addr); 2075 2076 2077 /* save the current value of the warm-start vector */ 2078 mpbioswarmvec = *((u_long *) WARMBOOT_OFF); 2079#ifndef PC98 2080 outb(CMOS_REG, BIOS_RESET); 2081 mpbiosreason = inb(CMOS_DATA); 2082#endif 2083 2084 /* set up temporary P==V mapping for AP boot */ 2085 /* XXX this is a hack, we should boot the AP on its own stack/PTD */ 2086 kptbase = (uintptr_t)(void *)KPTphys; 2087 for (x = 0; x < NKPT; x++) 2088 PTD[x] = (pd_entry_t)(PG_V | PG_RW | 2089 ((kptbase + x * PAGE_SIZE) & PG_FRAME)); 2090 invltlb(); 2091 2092 /* start each AP */ 2093 for (x = 1; x <= mp_naps; ++x) { 2094 2095 /* This is a bit verbose, it will go away soon. */ 2096 2097 /* first page of AP's private space */ 2098 pg = x * i386_btop(sizeof(struct privatespace)); 2099 2100 /* allocate a new private data page */ 2101 pc = (struct pcpu *)kmem_alloc(kernel_map, PAGE_SIZE); 2102 2103 /* wire it into the private page table page */ 2104 SMPpt[pg] = (pt_entry_t)(PG_V | PG_RW | vtophys(pc)); 2105 2106 /* allocate and set up an idle stack data page */ 2107 stack = (char *)kmem_alloc(kernel_map, KSTACK_PAGES * PAGE_SIZE); /* XXXKSE */ 2108 for (i = 0; i < KSTACK_PAGES; i++) 2109 SMPpt[pg + 1 + i] = (pt_entry_t) 2110 (PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack)); 2111 2112 /* prime data page for it to use */ 2113 pcpu_init(pc, x, sizeof(struct pcpu)); 2114 2115 /* setup a vector to our boot code */ 2116 *((volatile u_short *) WARMBOOT_OFF) = WARMBOOT_TARGET; 2117 *((volatile u_short *) WARMBOOT_SEG) = (boot_addr >> 4); 2118#ifndef PC98 2119 outb(CMOS_REG, BIOS_RESET); 2120 outb(CMOS_DATA, BIOS_WARM); /* 'warm-start' */ 2121#endif 2122 2123 bootSTK = &SMP_prvspace[x].idlekstack[KSTACK_PAGES * PAGE_SIZE]; 2124 bootAP = x; 2125 2126 /* attempt to start the Application Processor */ 2127 CHECK_INIT(99); /* setup checkpoints */ 2128 if (!start_ap(x, boot_addr)) { 2129 printf("AP #%d (PHY# %d) failed!\n", x, CPU_TO_ID(x)); 2130 CHECK_PRINT("trace"); /* show checkpoints */ 2131 /* better panic as the AP may be running loose */ 2132 printf("panic y/n? [y] "); 2133 if (cngetc() != 'n') 2134 panic("bye-bye"); 2135 } 2136 CHECK_PRINT("trace"); /* show checkpoints */ 2137 2138 /* record its version info */ 2139 cpu_apic_versions[x] = cpu_apic_versions[0]; 2140 2141 all_cpus |= (1 << x); /* record AP in CPU map */ 2142 } 2143 2144 /* build our map of 'other' CPUs */ 2145 PCPU_SET(other_cpus, all_cpus & ~PCPU_GET(cpumask)); 2146 2147 /* fill in our (BSP) APIC version */ 2148 cpu_apic_versions[0] = lapic.version; 2149 2150 /* restore the warmstart vector */ 2151 *(u_long *) WARMBOOT_OFF = mpbioswarmvec; 2152#ifndef PC98 2153 outb(CMOS_REG, BIOS_RESET); 2154 outb(CMOS_DATA, mpbiosreason); 2155#endif 2156 2157 /* 2158 * Set up the idle context for the BSP. Similar to above except 2159 * that some was done by locore, some by pmap.c and some is implicit 2160 * because the BSP is cpu#0 and the page is initially zero and also 2161 * because we can refer to variables by name on the BSP.. 2162 */ 2163 2164 /* Allocate and setup BSP idle stack */ 2165 stack = (char *)kmem_alloc(kernel_map, KSTACK_PAGES * PAGE_SIZE); 2166 for (i = 0; i < KSTACK_PAGES; i++) 2167 SMPpt[1 + i] = (pt_entry_t) 2168 (PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack)); 2169 2170 for (x = 0; x < NKPT; x++) 2171 PTD[x] = 0; 2172 2173 /* number of APs actually started */ 2174 return mp_ncpus - 1; 2175} 2176 2177/* 2178 * load the 1st level AP boot code into base memory. 2179 */ 2180 2181/* targets for relocation */ 2182extern void bigJump(void); 2183extern void bootCodeSeg(void); 2184extern void bootDataSeg(void); 2185extern void MPentry(void); 2186extern u_int MP_GDT; 2187extern u_int mp_gdtbase; 2188 2189static void 2190install_ap_tramp(u_int boot_addr) 2191{ 2192 int x; 2193 int size = *(int *) ((u_long) & bootMP_size); 2194 u_char *src = (u_char *) ((u_long) bootMP); 2195 u_char *dst = (u_char *) boot_addr + KERNBASE; 2196 u_int boot_base = (u_int) bootMP; 2197 u_int8_t *dst8; 2198 u_int16_t *dst16; 2199 u_int32_t *dst32; 2200 2201 POSTCODE(INSTALL_AP_TRAMP_POST); 2202 2203 for (x = 0; x < size; ++x) 2204 *dst++ = *src++; 2205 2206 /* 2207 * modify addresses in code we just moved to basemem. unfortunately we 2208 * need fairly detailed info about mpboot.s for this to work. changes 2209 * to mpboot.s might require changes here. 2210 */ 2211 2212 /* boot code is located in KERNEL space */ 2213 dst = (u_char *) boot_addr + KERNBASE; 2214 2215 /* modify the lgdt arg */ 2216 dst32 = (u_int32_t *) (dst + ((u_int) & mp_gdtbase - boot_base)); 2217 *dst32 = boot_addr + ((u_int) & MP_GDT - boot_base); 2218 2219 /* modify the ljmp target for MPentry() */ 2220 dst32 = (u_int32_t *) (dst + ((u_int) bigJump - boot_base) + 1); 2221 *dst32 = ((u_int) MPentry - KERNBASE); 2222 2223 /* modify the target for boot code segment */ 2224 dst16 = (u_int16_t *) (dst + ((u_int) bootCodeSeg - boot_base)); 2225 dst8 = (u_int8_t *) (dst16 + 1); 2226 *dst16 = (u_int) boot_addr & 0xffff; 2227 *dst8 = ((u_int) boot_addr >> 16) & 0xff; 2228 2229 /* modify the target for boot data segment */ 2230 dst16 = (u_int16_t *) (dst + ((u_int) bootDataSeg - boot_base)); 2231 dst8 = (u_int8_t *) (dst16 + 1); 2232 *dst16 = (u_int) boot_addr & 0xffff; 2233 *dst8 = ((u_int) boot_addr >> 16) & 0xff; 2234} 2235 2236/* 2237 * This function starts the AP (application processor) identified 2238 * by the APIC ID 'physicalCpu'. It does quite a "song and dance" 2239 * to accomplish this. This is necessary because of the nuances 2240 * of the different hardware we might encounter. It isn't pretty, 2241 * but it seems to work. 2242 */ 2243static int 2244start_ap(int logical_cpu, u_int boot_addr) 2245{ 2246 int physical_cpu; 2247 int vector; 2248 int cpus; 2249 u_long icr_lo, icr_hi; 2250 2251 POSTCODE(START_AP_POST); 2252 2253 /* get the PHYSICAL APIC ID# */ 2254 physical_cpu = CPU_TO_ID(logical_cpu); 2255 2256 /* calculate the vector */ 2257 vector = (boot_addr >> 12) & 0xff; 2258 2259 /* used as a watchpoint to signal AP startup */ 2260 cpus = mp_ncpus; 2261 2262 /* 2263 * first we do an INIT/RESET IPI this INIT IPI might be run, reseting 2264 * and running the target CPU. OR this INIT IPI might be latched (P5 2265 * bug), CPU waiting for STARTUP IPI. OR this INIT IPI might be 2266 * ignored. 2267 */ 2268 2269 /* setup the address for the target AP */ 2270 icr_hi = lapic.icr_hi & ~APIC_ID_MASK; 2271 icr_hi |= (physical_cpu << 24); 2272 lapic.icr_hi = icr_hi; 2273 2274 /* setup common fields for subsequent IPIs */ 2275 icr_lo = lapic.icr_lo & APIC_ICRLO_RESV_MASK; 2276 icr_lo |= APIC_DESTMODE_PHY; 2277 2278 /* do an INIT IPI: assert RESET */ 2279 lapic.icr_lo = icr_lo | APIC_DEST_DESTFLD | APIC_TRIGMOD_EDGE | 2280 APIC_LEVEL_ASSERT | APIC_DELMODE_INIT; 2281 2282 /* wait for pending status end */ 2283 while (lapic.icr_lo & APIC_DELSTAT_MASK) 2284 /* spin */ ; 2285 2286 /* do an INIT IPI: deassert RESET */ 2287 lapic.icr_lo = icr_lo | APIC_DEST_ALLESELF | APIC_TRIGMOD_LEVEL | 2288 APIC_LEVEL_DEASSERT | APIC_DELMODE_INIT; 2289 2290 /* wait for pending status end */ 2291 u_sleep(10000); /* wait ~10mS */ 2292 while (lapic.icr_lo & APIC_DELSTAT_MASK) 2293 /* spin */ ; 2294 2295 /* 2296 * next we do a STARTUP IPI: the previous INIT IPI might still be 2297 * latched, (P5 bug) this 1st STARTUP would then terminate 2298 * immediately, and the previously started INIT IPI would continue. OR 2299 * the previous INIT IPI has already run. and this STARTUP IPI will 2300 * run. OR the previous INIT IPI was ignored. and this STARTUP IPI 2301 * will run. 2302 */ 2303 2304 /* do a STARTUP IPI */ 2305 lapic.icr_lo = icr_lo | APIC_DEST_DESTFLD | APIC_TRIGMOD_EDGE | 2306 APIC_LEVEL_DEASSERT | APIC_DELMODE_STARTUP | vector; 2307 while (lapic.icr_lo & APIC_DELSTAT_MASK) 2308 /* spin */ ; 2309 u_sleep(200); /* wait ~200uS */ 2310 2311 /* 2312 * finally we do a 2nd STARTUP IPI: this 2nd STARTUP IPI should run IF 2313 * the previous STARTUP IPI was cancelled by a latched INIT IPI. OR 2314 * this STARTUP IPI will be ignored, as only ONE STARTUP IPI is 2315 * recognized after hardware RESET or INIT IPI. 2316 */ 2317 2318 lapic.icr_lo = icr_lo | APIC_DEST_DESTFLD | APIC_TRIGMOD_EDGE | 2319 APIC_LEVEL_DEASSERT | APIC_DELMODE_STARTUP | vector; 2320 while (lapic.icr_lo & APIC_DELSTAT_MASK) 2321 /* spin */ ; 2322 u_sleep(200); /* wait ~200uS */ 2323 2324 /* wait for it to start */ 2325 set_apic_timer(5000000);/* == 5 seconds */ 2326 while (read_apic_timer()) 2327 if (mp_ncpus > cpus) 2328 return 1; /* return SUCCESS */ 2329 2330 return 0; /* return FAILURE */ 2331} 2332 2333#if defined(APIC_IO) 2334 2335#ifdef COUNT_XINVLTLB_HITS 2336u_int xhits_gbl[MAXCPU]; 2337u_int xhits_pg[MAXCPU]; 2338u_int xhits_rng[MAXCPU]; 2339SYSCTL_NODE(_debug, OID_AUTO, xhits, CTLFLAG_RW, 0, ""); 2340SYSCTL_OPAQUE(_debug_xhits, OID_AUTO, global, CTLFLAG_RW, &xhits_gbl, 2341 sizeof(xhits_gbl), "IU", ""); 2342SYSCTL_OPAQUE(_debug_xhits, OID_AUTO, page, CTLFLAG_RW, &xhits_pg, 2343 sizeof(xhits_pg), "IU", ""); 2344SYSCTL_OPAQUE(_debug_xhits, OID_AUTO, range, CTLFLAG_RW, &xhits_rng, 2345 sizeof(xhits_rng), "IU", ""); 2346 2347u_int ipi_global; 2348u_int ipi_page; 2349u_int ipi_range; 2350u_int ipi_range_size; 2351SYSCTL_INT(_debug_xhits, OID_AUTO, ipi_global, CTLFLAG_RW, &ipi_global, 0, ""); 2352SYSCTL_INT(_debug_xhits, OID_AUTO, ipi_page, CTLFLAG_RW, &ipi_page, 0, ""); 2353SYSCTL_INT(_debug_xhits, OID_AUTO, ipi_range, CTLFLAG_RW, &ipi_range, 0, ""); 2354SYSCTL_INT(_debug_xhits, OID_AUTO, ipi_range_size, CTLFLAG_RW, &ipi_range_size, 2355 0, ""); 2356 2357u_int ipi_masked_global; 2358u_int ipi_masked_page; 2359u_int ipi_masked_range; 2360u_int ipi_masked_range_size; 2361SYSCTL_INT(_debug_xhits, OID_AUTO, ipi_masked_global, CTLFLAG_RW, 2362 &ipi_masked_global, 0, ""); 2363SYSCTL_INT(_debug_xhits, OID_AUTO, ipi_masked_page, CTLFLAG_RW, 2364 &ipi_masked_page, 0, ""); 2365SYSCTL_INT(_debug_xhits, OID_AUTO, ipi_masked_range, CTLFLAG_RW, 2366 &ipi_masked_range, 0, ""); 2367SYSCTL_INT(_debug_xhits, OID_AUTO, ipi_masked_range_size, CTLFLAG_RW, 2368 &ipi_masked_range_size, 0, ""); 2369#endif 2370 2371/* 2372 * Flush the TLB on all other CPU's 2373 */ 2374static void 2375smp_tlb_shootdown(u_int vector, vm_offset_t addr1, vm_offset_t addr2) 2376{ 2377 u_int ncpu; 2378 register_t eflags; 2379 2380 ncpu = mp_ncpus - 1; /* does not shootdown self */ 2381 if (ncpu < 1) 2382 return; /* no other cpus */ 2383 eflags = read_eflags(); 2384 if ((eflags & PSL_I) == 0) 2385 panic("absolutely cannot call smp_ipi_shootdown with interrupts already disabled"); 2386 mtx_lock_spin(&smp_tlb_mtx); 2387 smp_tlb_addr1 = addr1; 2388 smp_tlb_addr2 = addr2; 2389 atomic_store_rel_int(&smp_tlb_wait, 0); 2390 ipi_all_but_self(vector); 2391 while (smp_tlb_wait < ncpu) 2392 ia32_pause(); 2393 mtx_unlock_spin(&smp_tlb_mtx); 2394} 2395 2396/* 2397 * This is about as magic as it gets. fortune(1) has got similar code 2398 * for reversing bits in a word. Who thinks up this stuff?? 2399 * 2400 * Yes, it does appear to be consistently faster than: 2401 * while (i = ffs(m)) { 2402 * m >>= i; 2403 * bits++; 2404 * } 2405 * and 2406 * while (lsb = (m & -m)) { // This is magic too 2407 * m &= ~lsb; // or: m ^= lsb 2408 * bits++; 2409 * } 2410 * Both of these latter forms do some very strange things on gcc-3.1 with 2411 * -mcpu=pentiumpro and/or -march=pentiumpro and/or -O or -O2. 2412 * There is probably an SSE or MMX popcnt instruction. 2413 * 2414 * I wonder if this should be in libkern? 2415 * 2416 * XXX Stop the presses! Another one: 2417 * static __inline u_int32_t 2418 * popcnt1(u_int32_t v) 2419 * { 2420 * v -= ((v >> 1) & 0x55555555); 2421 * v = (v & 0x33333333) + ((v >> 2) & 0x33333333); 2422 * v = (v + (v >> 4)) & 0x0F0F0F0F; 2423 * return (v * 0x01010101) >> 24; 2424 * } 2425 * The downside is that it has a multiply. With a pentium3 with 2426 * -mcpu=pentiumpro and -march=pentiumpro then gcc-3.1 will use 2427 * an imull, and in that case it is faster. In most other cases 2428 * it appears slightly slower. 2429 */ 2430static __inline u_int32_t 2431popcnt(u_int32_t m) 2432{ 2433 2434 m = (m & 0x55555555) + ((m & 0xaaaaaaaa) >> 1); 2435 m = (m & 0x33333333) + ((m & 0xcccccccc) >> 2); 2436 m = (m & 0x0f0f0f0f) + ((m & 0xf0f0f0f0) >> 4); 2437 m = (m & 0x00ff00ff) + ((m & 0xff00ff00) >> 8); 2438 m = (m & 0x0000ffff) + ((m & 0xffff0000) >> 16); 2439 return m; 2440} 2441 2442static void 2443smp_targeted_tlb_shootdown(u_int mask, u_int vector, vm_offset_t addr1, vm_offset_t addr2) 2444{ 2445 int ncpu, othercpus; 2446 register_t eflags; 2447 2448 othercpus = mp_ncpus - 1; 2449 if (mask == (u_int)-1) { 2450 ncpu = othercpus; 2451 if (ncpu < 1) 2452 return; 2453 } else { 2454 mask &= ~PCPU_GET(cpumask); 2455 if (mask == 0) 2456 return; 2457 ncpu = popcnt(mask); 2458 if (ncpu > othercpus) { 2459 /* XXX this should be a panic offence */ 2460 printf("SMP: tlb shootdown to %d other cpus (only have %d)\n", 2461 ncpu, othercpus); 2462 ncpu = othercpus; 2463 } 2464 /* XXX should be a panic, implied by mask == 0 above */ 2465 if (ncpu < 1) 2466 return; 2467 } 2468 eflags = read_eflags(); 2469 if ((eflags & PSL_I) == 0) 2470 panic("absolutely cannot call smp_targeted_ipi_shootdown with interrupts already disabled"); 2471 mtx_lock_spin(&smp_tlb_mtx); 2472 smp_tlb_addr1 = addr1; 2473 smp_tlb_addr2 = addr2; 2474 atomic_store_rel_int(&smp_tlb_wait, 0); 2475 if (mask == (u_int)-1) 2476 ipi_all_but_self(vector); 2477 else 2478 ipi_selected(mask, vector); 2479 while (smp_tlb_wait < ncpu) 2480 ia32_pause(); 2481 mtx_unlock_spin(&smp_tlb_mtx); 2482} 2483#endif 2484 2485void 2486smp_invltlb(void) 2487{ 2488#if defined(APIC_IO) 2489 if (smp_started) { 2490 smp_tlb_shootdown(IPI_INVLTLB, 0, 0); 2491#ifdef COUNT_XINVLTLB_HITS 2492 ipi_global++; 2493#endif 2494 } 2495#endif /* APIC_IO */ 2496} 2497 2498void 2499smp_invlpg(vm_offset_t addr) 2500{ 2501#if defined(APIC_IO) 2502 if (smp_started) { 2503 smp_tlb_shootdown(IPI_INVLPG, addr, 0); 2504#ifdef COUNT_XINVLTLB_HITS 2505 ipi_page++; 2506#endif 2507 } 2508#endif /* APIC_IO */ 2509} 2510 2511void 2512smp_invlpg_range(vm_offset_t addr1, vm_offset_t addr2) 2513{ 2514#if defined(APIC_IO) 2515 if (smp_started) { 2516 smp_tlb_shootdown(IPI_INVLRNG, addr1, addr2); 2517#ifdef COUNT_XINVLTLB_HITS 2518 ipi_range++; 2519 ipi_range_size += (addr2 - addr1) / PAGE_SIZE; 2520#endif 2521 } 2522#endif /* APIC_IO */ 2523} 2524 2525void 2526smp_masked_invltlb(u_int mask) 2527{ 2528#if defined(APIC_IO) 2529 if (smp_started) { 2530 smp_targeted_tlb_shootdown(mask, IPI_INVLTLB, 0, 0); 2531#ifdef COUNT_XINVLTLB_HITS 2532 ipi_masked_global++; 2533#endif 2534 } 2535#endif /* APIC_IO */ 2536} 2537 2538void 2539smp_masked_invlpg(u_int mask, vm_offset_t addr) 2540{ 2541#if defined(APIC_IO) 2542 if (smp_started) { 2543 smp_targeted_tlb_shootdown(mask, IPI_INVLPG, addr, 0); 2544#ifdef COUNT_XINVLTLB_HITS 2545 ipi_masked_page++; 2546#endif 2547 } 2548#endif /* APIC_IO */ 2549} 2550 2551void 2552smp_masked_invlpg_range(u_int mask, vm_offset_t addr1, vm_offset_t addr2) 2553{ 2554#if defined(APIC_IO) 2555 if (smp_started) { 2556 smp_targeted_tlb_shootdown(mask, IPI_INVLRNG, addr1, addr2); 2557#ifdef COUNT_XINVLTLB_HITS 2558 ipi_masked_range++; 2559 ipi_masked_range_size += (addr2 - addr1) / PAGE_SIZE; 2560#endif 2561 } 2562#endif /* APIC_IO */ 2563} 2564 2565 2566/* 2567 * This is called once the rest of the system is up and running and we're 2568 * ready to let the AP's out of the pen. 2569 */ 2570void 2571ap_init(void) 2572{ 2573 u_int apic_id; 2574 2575 /* spin until all the AP's are ready */ 2576 while (!aps_ready) 2577 ia32_pause(); 2578 2579 /* BSP may have changed PTD while we were waiting */ 2580 invltlb(); 2581 2582#if defined(I586_CPU) && !defined(NO_F00F_HACK) 2583 lidt(&r_idt); 2584#endif 2585 2586 /* set up CPU registers and state */ 2587 cpu_setregs(); 2588 2589 /* set up FPU state on the AP */ 2590 npxinit(__INITIAL_NPXCW__); 2591 2592 /* set up SSE registers */ 2593 enable_sse(); 2594 2595 /* A quick check from sanity claus */ 2596 apic_id = (apic_id_to_logical[(lapic.id & 0x0f000000) >> 24]); 2597 if (PCPU_GET(cpuid) != apic_id) { 2598 printf("SMP: cpuid = %d\n", PCPU_GET(cpuid)); 2599 printf("SMP: apic_id = %d\n", apic_id); 2600 printf("PTD[MPPTDI] = %#jx\n", (uintmax_t)PTD[MPPTDI]); 2601 panic("cpuid mismatch! boom!!"); 2602 } 2603 2604 /* Init local apic for irq's */ 2605 apic_initialize(); 2606 2607 /* Set memory range attributes for this CPU to match the BSP */ 2608 mem_range_AP_init(); 2609 2610 mtx_lock_spin(&ap_boot_mtx); 2611 2612 smp_cpus++; 2613 2614 CTR1(KTR_SMP, "SMP: AP CPU #%d Launched", PCPU_GET(cpuid)); 2615 printf("SMP: AP CPU #%d Launched!\n", PCPU_GET(cpuid)); 2616 2617 /* Build our map of 'other' CPUs. */ 2618 PCPU_SET(other_cpus, all_cpus & ~PCPU_GET(cpumask)); 2619 2620 if (bootverbose) 2621 apic_dump("ap_init()"); 2622 2623 if (smp_cpus == mp_ncpus) { 2624 /* enable IPI's, tlb shootdown, freezes etc */ 2625 atomic_store_rel_int(&smp_started, 1); 2626 smp_active = 1; /* historic */ 2627 } 2628 2629 mtx_unlock_spin(&ap_boot_mtx); 2630 2631 /* wait until all the AP's are up */ 2632 while (smp_started == 0) 2633 ia32_pause(); 2634 2635 /* ok, now grab sched_lock and enter the scheduler */ 2636 mtx_lock_spin(&sched_lock); 2637 2638 binuptime(PCPU_PTR(switchtime)); 2639 PCPU_SET(switchticks, ticks); 2640 2641 cpu_throw(NULL, choosethread()); /* doesn't return */ 2642 2643 panic("scheduler returned us to %s", __func__); 2644} 2645 2646/* 2647 * For statclock, we send an IPI to all CPU's to have them call this 2648 * function. 2649 * 2650 * WARNING! unpend() will call statclock() directly and skip this 2651 * routine. 2652 */ 2653void 2654forwarded_statclock(struct clockframe frame) 2655{ 2656 2657 if (profprocs != 0) 2658 profclock(&frame); 2659 if (pscnt == psdiv) 2660 statclock(&frame); 2661} 2662 2663void 2664forward_statclock(void) 2665{ 2666 int map; 2667 2668 CTR0(KTR_SMP, "forward_statclock"); 2669 2670 if (!smp_started || cold || panicstr) 2671 return; 2672 2673 map = PCPU_GET(other_cpus) & ~(stopped_cpus|hlt_cpus_mask); 2674 if (map != 0) 2675 ipi_selected(map, IPI_STATCLOCK); 2676} 2677 2678/* 2679 * For each hardclock(), we send an IPI to all other CPU's to have them 2680 * execute this function. It would be nice to reduce contention on 2681 * sched_lock if we could simply peek at the CPU to determine the user/kernel 2682 * state and call hardclock_process() on the CPU receiving the clock interrupt 2683 * and then just use a simple IPI to handle any ast's if needed. 2684 * 2685 * WARNING! unpend() will call hardclock_process() directly and skip this 2686 * routine. 2687 */ 2688void 2689forwarded_hardclock(struct clockframe frame) 2690{ 2691 2692 hardclock_process(&frame); 2693} 2694 2695void 2696forward_hardclock(void) 2697{ 2698 u_int map; 2699 2700 CTR0(KTR_SMP, "forward_hardclock"); 2701 2702 if (!smp_started || cold || panicstr) 2703 return; 2704 2705 map = PCPU_GET(other_cpus) & ~(stopped_cpus|hlt_cpus_mask); 2706 if (map != 0) 2707 ipi_selected(map, IPI_HARDCLOCK); 2708} 2709 2710#ifdef APIC_INTR_REORDER 2711/* 2712 * Maintain mapping from softintr vector to isr bit in local apic. 2713 */ 2714void 2715set_lapic_isrloc(int intr, int vector) 2716{ 2717 if (intr < 0 || intr > 32) 2718 panic("set_apic_isrloc: bad intr argument: %d",intr); 2719 if (vector < ICU_OFFSET || vector > 255) 2720 panic("set_apic_isrloc: bad vector argument: %d",vector); 2721 apic_isrbit_location[intr].location = &lapic.isr0 + ((vector>>5)<<2); 2722 apic_isrbit_location[intr].bit = (1<<(vector & 31)); 2723} 2724#endif 2725 2726/* 2727 * send an IPI to a set of cpus. 2728 */ 2729void 2730ipi_selected(u_int32_t cpus, u_int ipi) 2731{ 2732 2733 CTR3(KTR_SMP, "%s: cpus: %x ipi: %x", __func__, cpus, ipi); 2734 selected_apic_ipi(cpus, ipi, APIC_DELMODE_FIXED); 2735} 2736 2737/* 2738 * send an IPI INTerrupt containing 'vector' to all CPUs, including myself 2739 */ 2740void 2741ipi_all(u_int ipi) 2742{ 2743 2744 CTR2(KTR_SMP, "%s: ipi: %x", __func__, ipi); 2745 apic_ipi(APIC_DEST_ALLISELF, ipi, APIC_DELMODE_FIXED); 2746} 2747 2748/* 2749 * send an IPI to all CPUs EXCEPT myself 2750 */ 2751void 2752ipi_all_but_self(u_int ipi) 2753{ 2754 2755 CTR2(KTR_SMP, "%s: ipi: %x", __func__, ipi); 2756 apic_ipi(APIC_DEST_ALLESELF, ipi, APIC_DELMODE_FIXED); 2757} 2758 2759/* 2760 * send an IPI to myself 2761 */ 2762void 2763ipi_self(u_int ipi) 2764{ 2765 2766 CTR2(KTR_SMP, "%s: ipi: %x", __func__, ipi); 2767 apic_ipi(APIC_DEST_SELF, ipi, APIC_DELMODE_FIXED); 2768} 2769 2770static void 2771release_aps(void *dummy __unused) 2772{ 2773 2774 if (mp_ncpus == 1) 2775 return; 2776 mtx_lock_spin(&sched_lock); 2777 atomic_store_rel_int(&aps_ready, 1); 2778 while (smp_started == 0) 2779 ia32_pause(); 2780 mtx_unlock_spin(&sched_lock); 2781} 2782SYSINIT(start_aps, SI_SUB_SMP, SI_ORDER_FIRST, release_aps, NULL); 2783 2784static int 2785sysctl_hlt_cpus(SYSCTL_HANDLER_ARGS) 2786{ 2787 u_int mask; 2788 int error; 2789 2790 mask = hlt_cpus_mask; 2791 error = sysctl_handle_int(oidp, &mask, 0, req); 2792 if (error || !req->newptr) 2793 return (error); 2794 2795 if (logical_cpus_mask != 0 && 2796 (mask & logical_cpus_mask) == logical_cpus_mask) 2797 hlt_logical_cpus = 1; 2798 else 2799 hlt_logical_cpus = 0; 2800 2801 if ((mask & all_cpus) == all_cpus) 2802 mask &= ~(1<<0); 2803 hlt_cpus_mask = mask; 2804 return (error); 2805} 2806SYSCTL_PROC(_machdep, OID_AUTO, hlt_cpus, CTLTYPE_INT|CTLFLAG_RW, 2807 0, 0, sysctl_hlt_cpus, "IU", ""); 2808 2809static int 2810sysctl_hlt_logical_cpus(SYSCTL_HANDLER_ARGS) 2811{ 2812 int disable, error; 2813 2814 disable = hlt_logical_cpus; 2815 error = sysctl_handle_int(oidp, &disable, 0, req); 2816 if (error || !req->newptr) 2817 return (error); 2818 2819 if (disable) 2820 hlt_cpus_mask |= logical_cpus_mask; 2821 else 2822 hlt_cpus_mask &= ~logical_cpus_mask; 2823 2824 if ((hlt_cpus_mask & all_cpus) == all_cpus) 2825 hlt_cpus_mask &= ~(1<<0); 2826 2827 hlt_logical_cpus = disable; 2828 return (error); 2829} 2830 2831static void 2832cpu_hlt_setup(void *dummy __unused) 2833{ 2834 2835 if (logical_cpus_mask != 0) { 2836 TUNABLE_INT_FETCH("machdep.hlt_logical_cpus", 2837 &hlt_logical_cpus); 2838 sysctl_ctx_init(&logical_cpu_clist); 2839 SYSCTL_ADD_PROC(&logical_cpu_clist, 2840 SYSCTL_STATIC_CHILDREN(_machdep), OID_AUTO, 2841 "hlt_logical_cpus", CTLTYPE_INT|CTLFLAG_RW, 0, 0, 2842 sysctl_hlt_logical_cpus, "IU", ""); 2843 SYSCTL_ADD_UINT(&logical_cpu_clist, 2844 SYSCTL_STATIC_CHILDREN(_machdep), OID_AUTO, 2845 "logical_cpus_mask", CTLTYPE_INT|CTLFLAG_RD, 2846 &logical_cpus_mask, 0, ""); 2847 2848 if (hlt_logical_cpus) 2849 hlt_cpus_mask |= logical_cpus_mask; 2850 } 2851} 2852SYSINIT(cpu_hlt, SI_SUB_SMP, SI_ORDER_ANY, cpu_hlt_setup, NULL); 2853 2854int 2855mp_grab_cpu_hlt(void) 2856{ 2857 u_int mask = PCPU_GET(cpumask); 2858 int retval; 2859 2860 retval = mask & hlt_cpus_mask; 2861 while (mask & hlt_cpus_mask) 2862 __asm __volatile("sti; hlt" : : : "memory"); 2863 return (retval); 2864} 2865