mp_x86.c revision 34019
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 * $Id: mp_machdep.c,v 1.66 1998/03/01 04:18:50 dyson Exp $ 26 */ 27 28#include "opt_smp.h" 29#include "opt_vm86.h" 30 31#ifdef SMP 32#include <machine/smptests.h> 33#else 34#error 35#endif 36 37#include <sys/param.h> 38#include <sys/systm.h> 39#include <sys/kernel.h> 40#include <sys/proc.h> 41#include <sys/sysctl.h> 42#ifdef BETTER_CLOCK 43#include <sys/dkstat.h> 44#endif 45 46#include <vm/vm.h> 47#include <vm/vm_param.h> 48#include <vm/pmap.h> 49#include <vm/vm_kern.h> 50#include <vm/vm_extern.h> 51#ifdef BETTER_CLOCK 52#include <sys/lock.h> 53#include <vm/vm_map.h> 54#include <sys/user.h> 55#ifdef GPROF 56#include <sys/gmon.h> 57#endif 58#endif 59 60#include <machine/smp.h> 61#include <machine/apic.h> 62#include <machine/mpapic.h> 63#include <machine/segments.h> 64#include <machine/smptests.h> /** TEST_DEFAULT_CONFIG, TEST_TEST1 */ 65#include <machine/tss.h> 66#include <machine/specialreg.h> 67#include <machine/cputypes.h> 68 69#include <i386/i386/cons.h> /* cngetc() */ 70 71#if defined(APIC_IO) 72#include <machine/md_var.h> /* setidt() */ 73#include <i386/isa/icu.h> /* IPIs */ 74#include <i386/isa/intr_machdep.h> /* IPIs */ 75#endif /* APIC_IO */ 76 77#if defined(TEST_DEFAULT_CONFIG) 78#define MPFPS_MPFB1 TEST_DEFAULT_CONFIG 79#else 80#define MPFPS_MPFB1 mpfps->mpfb1 81#endif /* TEST_DEFAULT_CONFIG */ 82 83#define WARMBOOT_TARGET 0 84#define WARMBOOT_OFF (KERNBASE + 0x0467) 85#define WARMBOOT_SEG (KERNBASE + 0x0469) 86 87#define BIOS_BASE (0xf0000) 88#define BIOS_SIZE (0x10000) 89#define BIOS_COUNT (BIOS_SIZE/4) 90 91#define CMOS_REG (0x70) 92#define CMOS_DATA (0x71) 93#define BIOS_RESET (0x0f) 94#define BIOS_WARM (0x0a) 95 96#define PROCENTRY_FLAG_EN 0x01 97#define PROCENTRY_FLAG_BP 0x02 98#define IOAPICENTRY_FLAG_EN 0x01 99 100 101/* MP Floating Pointer Structure */ 102typedef struct MPFPS { 103 char signature[4]; 104 void *pap; 105 u_char length; 106 u_char spec_rev; 107 u_char checksum; 108 u_char mpfb1; 109 u_char mpfb2; 110 u_char mpfb3; 111 u_char mpfb4; 112 u_char mpfb5; 113} *mpfps_t; 114 115/* MP Configuration Table Header */ 116typedef struct MPCTH { 117 char signature[4]; 118 u_short base_table_length; 119 u_char spec_rev; 120 u_char checksum; 121 u_char oem_id[8]; 122 u_char product_id[12]; 123 void *oem_table_pointer; 124 u_short oem_table_size; 125 u_short entry_count; 126 void *apic_address; 127 u_short extended_table_length; 128 u_char extended_table_checksum; 129 u_char reserved; 130} *mpcth_t; 131 132 133typedef struct PROCENTRY { 134 u_char type; 135 u_char apic_id; 136 u_char apic_version; 137 u_char cpu_flags; 138 u_long cpu_signature; 139 u_long feature_flags; 140 u_long reserved1; 141 u_long reserved2; 142} *proc_entry_ptr; 143 144typedef struct BUSENTRY { 145 u_char type; 146 u_char bus_id; 147 char bus_type[6]; 148} *bus_entry_ptr; 149 150typedef struct IOAPICENTRY { 151 u_char type; 152 u_char apic_id; 153 u_char apic_version; 154 u_char apic_flags; 155 void *apic_address; 156} *io_apic_entry_ptr; 157 158typedef struct INTENTRY { 159 u_char type; 160 u_char int_type; 161 u_short int_flags; 162 u_char src_bus_id; 163 u_char src_bus_irq; 164 u_char dst_apic_id; 165 u_char dst_apic_int; 166} *int_entry_ptr; 167 168/* descriptions of MP basetable entries */ 169typedef struct BASETABLE_ENTRY { 170 u_char type; 171 u_char length; 172 char name[16]; 173} basetable_entry; 174 175/* 176 * this code MUST be enabled here and in mpboot.s. 177 * it follows the very early stages of AP boot by placing values in CMOS ram. 178 * it NORMALLY will never be needed and thus the primitive method for enabling. 179 * 180#define CHECK_POINTS 181 */ 182 183#if defined(CHECK_POINTS) 184#define CHECK_READ(A) (outb(CMOS_REG, (A)), inb(CMOS_DATA)) 185#define CHECK_WRITE(A,D) (outb(CMOS_REG, (A)), outb(CMOS_DATA, (D))) 186 187#define CHECK_INIT(D); \ 188 CHECK_WRITE(0x34, (D)); \ 189 CHECK_WRITE(0x35, (D)); \ 190 CHECK_WRITE(0x36, (D)); \ 191 CHECK_WRITE(0x37, (D)); \ 192 CHECK_WRITE(0x38, (D)); \ 193 CHECK_WRITE(0x39, (D)); 194 195#define CHECK_PRINT(S); \ 196 printf("%s: %d, %d, %d, %d, %d, %d\n", \ 197 (S), \ 198 CHECK_READ(0x34), \ 199 CHECK_READ(0x35), \ 200 CHECK_READ(0x36), \ 201 CHECK_READ(0x37), \ 202 CHECK_READ(0x38), \ 203 CHECK_READ(0x39)); 204 205#else /* CHECK_POINTS */ 206 207#define CHECK_INIT(D) 208#define CHECK_PRINT(S) 209 210#endif /* CHECK_POINTS */ 211 212/* 213 * Values to send to the POST hardware. 214 */ 215#define MP_BOOTADDRESS_POST 0x10 216#define MP_PROBE_POST 0x11 217#define MPTABLE_PASS1_POST 0x12 218 219#define MP_START_POST 0x13 220#define MP_ENABLE_POST 0x14 221#define MPTABLE_PASS2_POST 0x15 222 223#define START_ALL_APS_POST 0x16 224#define INSTALL_AP_TRAMP_POST 0x17 225#define START_AP_POST 0x18 226 227#define MP_ANNOUNCE_POST 0x19 228 229 230/** XXX FIXME: where does this really belong, isa.h/isa.c perhaps? */ 231int current_postcode; 232 233/** XXX FIXME: what system files declare these??? */ 234extern struct region_descriptor r_gdt, r_idt; 235 236int bsp_apic_ready = 0; /* flags useability of BSP apic */ 237int mp_ncpus; /* # of CPUs, including BSP */ 238int mp_naps; /* # of Applications processors */ 239int mp_nbusses; /* # of busses */ 240int mp_napics; /* # of IO APICs */ 241int boot_cpu_id; /* designated BSP */ 242vm_offset_t cpu_apic_address; 243vm_offset_t io_apic_address[NAPICID]; /* NAPICID is more than enough */ 244extern int nkpt; 245 246u_int32_t cpu_apic_versions[NCPU]; 247u_int32_t io_apic_versions[NAPIC]; 248 249/* 250 * APIC ID logical/physical mapping structures. 251 * We oversize these to simplify boot-time config. 252 */ 253int cpu_num_to_apic_id[NAPICID]; 254int io_num_to_apic_id[NAPICID]; 255int apic_id_to_logical[NAPICID]; 256 257 258#define NPPROVMTRR 8 259#define PPRO_VMTRRphysBase0 0x200 260#define PPRO_VMTRRphysMask0 0x201 261static struct { 262 u_int64_t base, mask; 263} PPro_vmtrr[NPPROVMTRR]; 264 265/* Bitmap of all available CPUs */ 266u_int all_cpus; 267 268/* AP uses this PTD during bootstrap. Do not staticize. */ 269pd_entry_t *bootPTD; 270 271/* Hotwire a 0->4MB V==P mapping */ 272extern pt_entry_t *KPTphys; 273 274/* Virtual address of per-cpu common_tss */ 275extern struct i386tss common_tss; 276#ifdef VM86 277extern struct segment_descriptor common_tssd; 278extern u_int private_tss; /* flag indicating private tss */ 279extern u_int my_tr; 280#endif /* VM86 */ 281 282/* IdlePTD per cpu */ 283pd_entry_t *IdlePTDS[NCPU]; 284 285/* "my" private page table page, for BSP init */ 286extern pt_entry_t SMP_prvpt[]; 287 288/* Private page pointer to curcpu's PTD, used during BSP init */ 289extern pd_entry_t *my_idlePTD; 290 291static int smp_started; /* has the system started? */ 292 293/* 294 * Local data and functions. 295 */ 296 297static int mp_capable; 298static u_int boot_address; 299static u_int base_memory; 300 301static int picmode; /* 0: virtual wire mode, 1: PIC mode */ 302static mpfps_t mpfps; 303static int search_for_sig(u_int32_t target, int count); 304static void mp_enable(u_int boot_addr); 305 306static int mptable_pass1(void); 307static int mptable_pass2(void); 308static void default_mp_table(int type); 309static void fix_mp_table(void); 310static void init_locks(void); 311static int start_all_aps(u_int boot_addr); 312static void install_ap_tramp(u_int boot_addr); 313static int start_ap(int logicalCpu, u_int boot_addr); 314static void getmtrr(void); 315static void putmtrr(void); 316static void putfmtrr(void); 317 318 319/* 320 * Calculate usable address in base memory for AP trampoline code. 321 */ 322u_int 323mp_bootaddress(u_int basemem) 324{ 325 POSTCODE(MP_BOOTADDRESS_POST); 326 327 base_memory = basemem * 1024; /* convert to bytes */ 328 329 boot_address = base_memory & ~0xfff; /* round down to 4k boundary */ 330 if ((base_memory - boot_address) < bootMP_size) 331 boot_address -= 4096; /* not enough, lower by 4k */ 332 333 return boot_address; 334} 335 336 337/* 338 * Look for an Intel MP spec table (ie, SMP capable hardware). 339 */ 340int 341mp_probe(void) 342{ 343 int x; 344 u_long segment; 345 u_int32_t target; 346 347 POSTCODE(MP_PROBE_POST); 348 349 /* see if EBDA exists */ 350 if (segment = (u_long) * (u_short *) (KERNBASE + 0x40e)) { 351 /* search first 1K of EBDA */ 352 target = (u_int32_t) (segment << 4); 353 if ((x = search_for_sig(target, 1024 / 4)) >= 0) 354 goto found; 355 } else { 356 /* last 1K of base memory, effective 'top of base' passed in */ 357 target = (u_int32_t) (base_memory - 0x400); 358 if ((x = search_for_sig(target, 1024 / 4)) >= 0) 359 goto found; 360 } 361 362 /* search the BIOS */ 363 target = (u_int32_t) BIOS_BASE; 364 if ((x = search_for_sig(target, BIOS_COUNT)) >= 0) 365 goto found; 366 367 /* nothing found */ 368 mpfps = (mpfps_t)0; 369 mp_capable = 0; 370 return 0; 371 372found: 373 /* calculate needed resources */ 374 mpfps = (mpfps_t)x; 375 if (mptable_pass1()) 376 panic("you must reconfigure your kernel"); 377 378 /* flag fact that we are running multiple processors */ 379 mp_capable = 1; 380 return 1; 381} 382 383 384/* 385 * Startup the SMP processors. 386 */ 387void 388mp_start(void) 389{ 390 POSTCODE(MP_START_POST); 391 392 /* look for MP capable motherboard */ 393 if (mp_capable) 394 mp_enable(boot_address); 395 else 396 panic("MP hardware not found!"); 397} 398 399 400/* 401 * Print various information about the SMP system hardware and setup. 402 */ 403void 404mp_announce(void) 405{ 406 int x; 407 408 POSTCODE(MP_ANNOUNCE_POST); 409 410 printf("FreeBSD/SMP: Multiprocessor motherboard\n"); 411 printf(" cpu0 (BSP): apic id: %2d", CPU_TO_ID(0)); 412 printf(", version: 0x%08x", cpu_apic_versions[0]); 413 printf(", at 0x%08x\n", cpu_apic_address); 414 for (x = 1; x <= mp_naps; ++x) { 415 printf(" cpu%d (AP): apic id: %2d", x, CPU_TO_ID(x)); 416 printf(", version: 0x%08x", cpu_apic_versions[x]); 417 printf(", at 0x%08x\n", cpu_apic_address); 418 } 419 420#if defined(APIC_IO) 421 for (x = 0; x < mp_napics; ++x) { 422 printf(" io%d (APIC): apic id: %2d", x, IO_TO_ID(x)); 423 printf(", version: 0x%08x", io_apic_versions[x]); 424 printf(", at 0x%08x\n", io_apic_address[x]); 425 } 426#else 427 printf(" Warning: APIC I/O disabled\n"); 428#endif /* APIC_IO */ 429} 430 431/* 432 * AP cpu's call this to sync up protected mode. 433 */ 434void 435init_secondary(void) 436{ 437 int gsel_tss; 438#ifndef VM86 439 u_int my_tr; 440#endif 441 442 r_gdt.rd_limit = sizeof(gdt[0]) * (NGDT + NCPU) - 1; 443 r_gdt.rd_base = (int) gdt; 444 lgdt(&r_gdt); /* does magic intra-segment return */ 445 lidt(&r_idt); 446 lldt(_default_ldt); 447 448 my_tr = NGDT + cpuid; 449 gsel_tss = GSEL(my_tr, SEL_KPL); 450 gdt[my_tr].sd.sd_type = SDT_SYS386TSS; 451 common_tss.tss_esp0 = 0; /* not used until after switch */ 452 common_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL); 453 common_tss.tss_ioopt = (sizeof common_tss) << 16; 454#ifdef VM86 455 common_tssd = gdt[my_tr].sd; 456 private_tss = 0; 457#endif /* VM86 */ 458 ltr(gsel_tss); 459 460 load_cr0(0x8005003b); /* XXX! */ 461 462 PTD[0] = 0; 463 pmap_set_opt((unsigned *)PTD); 464 465 putmtrr(); 466 putfmtrr(); 467 468 invltlb(); 469} 470 471 472#if defined(APIC_IO) 473/* 474 * Final configuration of the BSP's local APIC: 475 * - disable 'pic mode'. 476 * - disable 'virtual wire mode'. 477 * - enable NMI. 478 */ 479void 480bsp_apic_configure(void) 481{ 482 u_char byte; 483 u_int32_t temp; 484 485 /* leave 'pic mode' if necessary */ 486 if (picmode) { 487 outb(0x22, 0x70); /* select IMCR */ 488 byte = inb(0x23); /* current contents */ 489 byte |= 0x01; /* mask external INTR */ 490 outb(0x23, byte); /* disconnect 8259s/NMI */ 491 } 492 493 /* mask lint0 (the 8259 'virtual wire' connection) */ 494 temp = lapic.lvt_lint0; 495 temp |= APIC_LVT_M; /* set the mask */ 496 lapic.lvt_lint0 = temp; 497 498 /* setup lint1 to handle NMI */ 499 temp = lapic.lvt_lint1; 500 temp &= ~APIC_LVT_M; /* clear the mask */ 501 lapic.lvt_lint1 = temp; 502 503 if (bootverbose) 504 apic_dump("bsp_apic_configure()"); 505} 506#endif /* APIC_IO */ 507 508 509/******************************************************************* 510 * local functions and data 511 */ 512 513/* 514 * start the SMP system 515 */ 516static void 517mp_enable(u_int boot_addr) 518{ 519 int x; 520#if defined(APIC_IO) 521 int apic; 522 u_int ux; 523#endif /* APIC_IO */ 524 525 getmtrr(); 526 putfmtrr(); 527 528 POSTCODE(MP_ENABLE_POST); 529 530 /* turn on 4MB of V == P addressing so we can get to MP table */ 531 *(int *)PTD = PG_V | PG_RW | ((u_long)KPTphys & PG_FRAME); 532 invltlb(); 533 534 /* examine the MP table for needed info, uses physical addresses */ 535 x = mptable_pass2(); 536 537 *(int *)PTD = 0; 538 invltlb(); 539 540 /* can't process default configs till the CPU APIC is pmapped */ 541 if (x) 542 default_mp_table(x); 543 544 /* post scan cleanup */ 545 fix_mp_table(); 546 547#if defined(APIC_IO) 548 549 /* fill the LOGICAL io_apic_versions table */ 550 for (apic = 0; apic < mp_napics; ++apic) { 551 ux = io_apic_read(apic, IOAPIC_VER); 552 io_apic_versions[apic] = ux; 553 } 554 555 /* program each IO APIC in the system */ 556 for (apic = 0; apic < mp_napics; ++apic) 557 if (io_apic_setup(apic) < 0) 558 panic("IO APIC setup failure"); 559 560 /* install a 'Spurious INTerrupt' vector */ 561 setidt(XSPURIOUSINT_OFFSET, Xspuriousint, 562 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 563 564 /* install an inter-CPU IPI for TLB invalidation */ 565 setidt(XINVLTLB_OFFSET, Xinvltlb, 566 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 567 568#ifdef BETTER_CLOCK 569 /* install an inter-CPU IPI for reading processor state */ 570 setidt(XCPUCHECKSTATE_OFFSET, Xcpucheckstate, 571 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 572 573 /* install an inter-CPU IPI for forcing an additional software trap */ 574 setidt(XCPUAST_OFFSET, Xcpuast, 575 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 576#endif 577 578 /* install an inter-CPU IPI for CPU stop/restart */ 579 setidt(XCPUSTOP_OFFSET, Xcpustop, 580 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 581 582#if defined(TEST_TEST1) 583 /* install a "fake hardware INTerrupt" vector */ 584 setidt(XTEST1_OFFSET, Xtest1, 585 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 586#endif /** TEST_TEST1 */ 587 588#endif /* APIC_IO */ 589 590 /* initialize all SMP locks */ 591 init_locks(); 592 593 /* start each Application Processor */ 594 start_all_aps(boot_addr); 595 596 /* 597 * The init process might be started on a different CPU now, 598 * and the boot CPU might not call prepare_usermode to get 599 * cr0 correctly configured. Thus we initialize cr0 here. 600 */ 601 load_cr0(rcr0() | CR0_WP | CR0_AM); 602} 603 604 605/* 606 * look for the MP spec signature 607 */ 608 609/* string defined by the Intel MP Spec as identifying the MP table */ 610#define MP_SIG 0x5f504d5f /* _MP_ */ 611#define NEXT(X) ((X) += 4) 612static int 613search_for_sig(u_int32_t target, int count) 614{ 615 int x; 616 u_int32_t *addr = (u_int32_t *) (KERNBASE + target); 617 618 for (x = 0; x < count; NEXT(x)) 619 if (addr[x] == MP_SIG) 620 /* make array index a byte index */ 621 return (target + (x * sizeof(u_int32_t))); 622 623 return -1; 624} 625 626 627static basetable_entry basetable_entry_types[] = 628{ 629 {0, 20, "Processor"}, 630 {1, 8, "Bus"}, 631 {2, 8, "I/O APIC"}, 632 {3, 8, "I/O INT"}, 633 {4, 8, "Local INT"} 634}; 635 636typedef struct BUSDATA { 637 u_char bus_id; 638 enum busTypes bus_type; 639} bus_datum; 640 641typedef struct INTDATA { 642 u_char int_type; 643 u_short int_flags; 644 u_char src_bus_id; 645 u_char src_bus_irq; 646 u_char dst_apic_id; 647 u_char dst_apic_int; 648} io_int, local_int; 649 650typedef struct BUSTYPENAME { 651 u_char type; 652 char name[7]; 653} bus_type_name; 654 655static bus_type_name bus_type_table[] = 656{ 657 {CBUS, "CBUS"}, 658 {CBUSII, "CBUSII"}, 659 {EISA, "EISA"}, 660 {UNKNOWN_BUSTYPE, "---"}, 661 {UNKNOWN_BUSTYPE, "---"}, 662 {ISA, "ISA"}, 663 {UNKNOWN_BUSTYPE, "---"}, 664 {UNKNOWN_BUSTYPE, "---"}, 665 {UNKNOWN_BUSTYPE, "---"}, 666 {UNKNOWN_BUSTYPE, "---"}, 667 {UNKNOWN_BUSTYPE, "---"}, 668 {UNKNOWN_BUSTYPE, "---"}, 669 {PCI, "PCI"}, 670 {UNKNOWN_BUSTYPE, "---"}, 671 {UNKNOWN_BUSTYPE, "---"}, 672 {UNKNOWN_BUSTYPE, "---"}, 673 {UNKNOWN_BUSTYPE, "---"}, 674 {XPRESS, "XPRESS"}, 675 {UNKNOWN_BUSTYPE, "---"} 676}; 677/* from MP spec v1.4, table 5-1 */ 678static int default_data[7][5] = 679{ 680/* nbus, id0, type0, id1, type1 */ 681 {1, 0, ISA, 255, 255}, 682 {1, 0, EISA, 255, 255}, 683 {1, 0, EISA, 255, 255}, 684 {0, 255, 255, 255, 255},/* MCA not supported */ 685 {2, 0, ISA, 1, PCI}, 686 {2, 0, EISA, 1, PCI}, 687 {0, 255, 255, 255, 255} /* MCA not supported */ 688}; 689 690 691/* the bus data */ 692static bus_datum bus_data[NBUS]; 693 694/* the IO INT data, one entry per possible APIC INTerrupt */ 695static io_int io_apic_ints[NINTR]; 696 697static int nintrs; 698 699static int processor_entry __P((proc_entry_ptr entry, int cpu)); 700static int bus_entry __P((bus_entry_ptr entry, int bus)); 701static int io_apic_entry __P((io_apic_entry_ptr entry, int apic)); 702static int int_entry __P((int_entry_ptr entry, int intr)); 703static int lookup_bus_type __P((char *name)); 704 705 706/* 707 * 1st pass on motherboard's Intel MP specification table. 708 * 709 * initializes: 710 * mp_ncpus = 1 711 * 712 * determines: 713 * cpu_apic_address (common to all CPUs) 714 * io_apic_address[N] 715 * mp_naps 716 * mp_nbusses 717 * mp_napics 718 * nintrs 719 */ 720static int 721mptable_pass1(void) 722{ 723 int x; 724 mpcth_t cth; 725 int totalSize; 726 void* position; 727 int count; 728 int type; 729 int mustpanic; 730 731 POSTCODE(MPTABLE_PASS1_POST); 732 733 mustpanic = 0; 734 735 /* clear various tables */ 736 for (x = 0; x < NAPICID; ++x) { 737 io_apic_address[x] = ~0; /* IO APIC address table */ 738 } 739 740 /* init everything to empty */ 741 mp_naps = 0; 742 mp_nbusses = 0; 743 mp_napics = 0; 744 nintrs = 0; 745 746 /* check for use of 'default' configuration */ 747 if (MPFPS_MPFB1 != 0) { 748 /* use default addresses */ 749 cpu_apic_address = DEFAULT_APIC_BASE; 750 io_apic_address[0] = DEFAULT_IO_APIC_BASE; 751 752 /* fill in with defaults */ 753 mp_naps = 2; /* includes BSP */ 754 mp_nbusses = default_data[MPFPS_MPFB1 - 1][0]; 755#if defined(APIC_IO) 756 mp_napics = 1; 757 nintrs = 16; 758#endif /* APIC_IO */ 759 } 760 else { 761 if ((cth = mpfps->pap) == 0) 762 panic("MP Configuration Table Header MISSING!"); 763 764 cpu_apic_address = (vm_offset_t) cth->apic_address; 765 766 /* walk the table, recording info of interest */ 767 totalSize = cth->base_table_length - sizeof(struct MPCTH); 768 position = (u_char *) cth + sizeof(struct MPCTH); 769 count = cth->entry_count; 770 771 while (count--) { 772 switch (type = *(u_char *) position) { 773 case 0: /* processor_entry */ 774 if (((proc_entry_ptr)position)->cpu_flags 775 & PROCENTRY_FLAG_EN) 776 ++mp_naps; 777 break; 778 case 1: /* bus_entry */ 779 ++mp_nbusses; 780 break; 781 case 2: /* io_apic_entry */ 782 if (((io_apic_entry_ptr)position)->apic_flags 783 & IOAPICENTRY_FLAG_EN) 784 io_apic_address[mp_napics++] = 785 (vm_offset_t)((io_apic_entry_ptr) 786 position)->apic_address; 787 break; 788 case 3: /* int_entry */ 789 ++nintrs; 790 break; 791 case 4: /* int_entry */ 792 break; 793 default: 794 panic("mpfps Base Table HOSED!"); 795 /* NOTREACHED */ 796 } 797 798 totalSize -= basetable_entry_types[type].length; 799 (u_char*)position += basetable_entry_types[type].length; 800 } 801 } 802 803 /* qualify the numbers */ 804 if (mp_naps > NCPU) 805#if 0 /* XXX FIXME: kern/4255 */ 806 printf("Warning: only using %d of %d available CPUs!\n", 807 NCPU, mp_naps); 808#else 809 { 810 printf("NCPU cannot be different than actual CPU count.\n"); 811 printf(" add 'options NCPU=%d' to your kernel config file,\n", 812 mp_naps); 813 printf(" then rerun config & rebuild your SMP kernel\n"); 814 mustpanic = 1; 815 } 816#endif /* XXX FIXME: kern/4255 */ 817 if (mp_nbusses > NBUS) { 818 printf("found %d busses, increase NBUS\n", mp_nbusses); 819 mustpanic = 1; 820 } 821 if (mp_napics > NAPIC) { 822 printf("found %d apics, increase NAPIC\n", mp_napics); 823 mustpanic = 1; 824 } 825 if (nintrs > NINTR) { 826 printf("found %d intrs, increase NINTR\n", nintrs); 827 mustpanic = 1; 828 } 829 830 /* 831 * Count the BSP. 832 * This is also used as a counter while starting the APs. 833 */ 834 mp_ncpus = 1; 835 836 --mp_naps; /* subtract the BSP */ 837 838 return mustpanic; 839} 840 841 842/* 843 * 2nd pass on motherboard's Intel MP specification table. 844 * 845 * sets: 846 * boot_cpu_id 847 * ID_TO_IO(N), phy APIC ID to log CPU/IO table 848 * CPU_TO_ID(N), logical CPU to APIC ID table 849 * IO_TO_ID(N), logical IO to APIC ID table 850 * bus_data[N] 851 * io_apic_ints[N] 852 */ 853static int 854mptable_pass2(void) 855{ 856 int x; 857 mpcth_t cth; 858 int totalSize; 859 void* position; 860 int count; 861 int type; 862 int apic, bus, cpu, intr; 863 864 POSTCODE(MPTABLE_PASS2_POST); 865 866 /* clear various tables */ 867 for (x = 0; x < NAPICID; ++x) { 868 ID_TO_IO(x) = -1; /* phy APIC ID to log CPU/IO table */ 869 CPU_TO_ID(x) = -1; /* logical CPU to APIC ID table */ 870 IO_TO_ID(x) = -1; /* logical IO to APIC ID table */ 871 } 872 873 /* clear bus data table */ 874 for (x = 0; x < NBUS; ++x) 875 bus_data[x].bus_id = 0xff; 876 877 /* clear IO APIC INT table */ 878 for (x = 0; x < NINTR; ++x) 879 io_apic_ints[x].int_type = 0xff; 880 881 /* setup the cpu/apic mapping arrays */ 882 boot_cpu_id = -1; 883 884 /* record whether PIC or virtual-wire mode */ 885 picmode = (mpfps->mpfb2 & 0x80) ? 1 : 0; 886 887 /* check for use of 'default' configuration */ 888 if (MPFPS_MPFB1 != 0) 889 return MPFPS_MPFB1; /* return default configuration type */ 890 891 if ((cth = mpfps->pap) == 0) 892 panic("MP Configuration Table Header MISSING!"); 893 894 /* walk the table, recording info of interest */ 895 totalSize = cth->base_table_length - sizeof(struct MPCTH); 896 position = (u_char *) cth + sizeof(struct MPCTH); 897 count = cth->entry_count; 898 apic = bus = intr = 0; 899 cpu = 1; /* pre-count the BSP */ 900 901 while (count--) { 902 switch (type = *(u_char *) position) { 903 case 0: 904 if (processor_entry(position, cpu)) 905 ++cpu; 906 break; 907 case 1: 908 if (bus_entry(position, bus)) 909 ++bus; 910 break; 911 case 2: 912 if (io_apic_entry(position, apic)) 913 ++apic; 914 break; 915 case 3: 916 if (int_entry(position, intr)) 917 ++intr; 918 break; 919 case 4: 920 /* int_entry(position); */ 921 break; 922 default: 923 panic("mpfps Base Table HOSED!"); 924 /* NOTREACHED */ 925 } 926 927 totalSize -= basetable_entry_types[type].length; 928 (u_char *) position += basetable_entry_types[type].length; 929 } 930 931 if (boot_cpu_id == -1) 932 panic("NO BSP found!"); 933 934 /* report fact that its NOT a default configuration */ 935 return 0; 936} 937 938 939/* 940 * parse an Intel MP specification table 941 */ 942static void 943fix_mp_table(void) 944{ 945 int x; 946 int id; 947 int bus_0; 948 int bus_pci; 949 int num_pci_bus; 950 951 /* 952 * Fix mis-numbering of the PCI bus and its INT entries if the BIOS 953 * did it wrong. The MP spec says that when more than 1 PCI bus 954 * exists the BIOS must begin with bus entries for the PCI bus and use 955 * actual PCI bus numbering. This implies that when only 1 PCI bus 956 * exists the BIOS can choose to ignore this ordering, and indeed many 957 * MP motherboards do ignore it. This causes a problem when the PCI 958 * sub-system makes requests of the MP sub-system based on PCI bus 959 * numbers. So here we look for the situation and renumber the 960 * busses and associated INTs in an effort to "make it right". 961 */ 962 963 /* find bus 0, PCI bus, count the number of PCI busses */ 964 for (num_pci_bus = 0, x = 0; x < mp_nbusses; ++x) { 965 if (bus_data[x].bus_id == 0) { 966 bus_0 = x; 967 } 968 if (bus_data[x].bus_type == PCI) { 969 ++num_pci_bus; 970 bus_pci = x; 971 } 972 } 973 /* 974 * bus_0 == slot of bus with ID of 0 975 * bus_pci == slot of last PCI bus encountered 976 */ 977 978 /* check the 1 PCI bus case for sanity */ 979 if (num_pci_bus == 1) { 980 981 /* if it is number 0 all is well */ 982 if (bus_data[bus_pci].bus_id == 0) 983 return; 984 985 /* mis-numbered, swap with whichever bus uses slot 0 */ 986 987 /* swap the bus entry types */ 988 bus_data[bus_pci].bus_type = bus_data[bus_0].bus_type; 989 bus_data[bus_0].bus_type = PCI; 990 991 /* swap each relavant INTerrupt entry */ 992 id = bus_data[bus_pci].bus_id; 993 for (x = 0; x < nintrs; ++x) { 994 if (io_apic_ints[x].src_bus_id == id) { 995 io_apic_ints[x].src_bus_id = 0; 996 } 997 else if (io_apic_ints[x].src_bus_id == 0) { 998 io_apic_ints[x].src_bus_id = id; 999 } 1000 } 1001 } 1002 /* sanity check if more than 1 PCI bus */ 1003 else if (num_pci_bus > 1) { 1004 for (x = 0; x < mp_nbusses; ++x) { 1005 if (bus_data[x].bus_type != PCI) 1006 continue; 1007 if (bus_data[x].bus_id >= num_pci_bus) 1008 panic("bad PCI bus numbering"); 1009 } 1010 } 1011} 1012 1013 1014static int 1015processor_entry(proc_entry_ptr entry, int cpu) 1016{ 1017 /* check for usability */ 1018 if ((cpu >= NCPU) || !(entry->cpu_flags & PROCENTRY_FLAG_EN)) 1019 return 0; 1020 1021 /* check for BSP flag */ 1022 if (entry->cpu_flags & PROCENTRY_FLAG_BP) { 1023 boot_cpu_id = entry->apic_id; 1024 CPU_TO_ID(0) = entry->apic_id; 1025 ID_TO_CPU(entry->apic_id) = 0; 1026 return 0; /* its already been counted */ 1027 } 1028 1029 /* add another AP to list, if less than max number of CPUs */ 1030 else { 1031 CPU_TO_ID(cpu) = entry->apic_id; 1032 ID_TO_CPU(entry->apic_id) = cpu; 1033 return 1; 1034 } 1035} 1036 1037 1038static int 1039bus_entry(bus_entry_ptr entry, int bus) 1040{ 1041 int x; 1042 char c, name[8]; 1043 1044 /* encode the name into an index */ 1045 for (x = 0; x < 6; ++x) { 1046 if ((c = entry->bus_type[x]) == ' ') 1047 break; 1048 name[x] = c; 1049 } 1050 name[x] = '\0'; 1051 1052 if ((x = lookup_bus_type(name)) == UNKNOWN_BUSTYPE) 1053 panic("unknown bus type: '%s'", name); 1054 1055 bus_data[bus].bus_id = entry->bus_id; 1056 bus_data[bus].bus_type = x; 1057 1058 return 1; 1059} 1060 1061 1062static int 1063io_apic_entry(io_apic_entry_ptr entry, int apic) 1064{ 1065 if (!(entry->apic_flags & IOAPICENTRY_FLAG_EN)) 1066 return 0; 1067 1068 IO_TO_ID(apic) = entry->apic_id; 1069 ID_TO_IO(entry->apic_id) = apic; 1070 1071 return 1; 1072} 1073 1074 1075static int 1076lookup_bus_type(char *name) 1077{ 1078 int x; 1079 1080 for (x = 0; x < MAX_BUSTYPE; ++x) 1081 if (strcmp(bus_type_table[x].name, name) == 0) 1082 return bus_type_table[x].type; 1083 1084 return UNKNOWN_BUSTYPE; 1085} 1086 1087 1088static int 1089int_entry(int_entry_ptr entry, int intr) 1090{ 1091 io_apic_ints[intr].int_type = entry->int_type; 1092 io_apic_ints[intr].int_flags = entry->int_flags; 1093 io_apic_ints[intr].src_bus_id = entry->src_bus_id; 1094 io_apic_ints[intr].src_bus_irq = entry->src_bus_irq; 1095 io_apic_ints[intr].dst_apic_id = entry->dst_apic_id; 1096 io_apic_ints[intr].dst_apic_int = entry->dst_apic_int; 1097 1098 return 1; 1099} 1100 1101 1102static int 1103apic_int_is_bus_type(int intr, int bus_type) 1104{ 1105 int bus; 1106 1107 for (bus = 0; bus < mp_nbusses; ++bus) 1108 if ((bus_data[bus].bus_id == io_apic_ints[intr].src_bus_id) 1109 && ((int) bus_data[bus].bus_type == bus_type)) 1110 return 1; 1111 1112 return 0; 1113} 1114 1115 1116/* 1117 * Given a traditional ISA INT mask, return an APIC mask. 1118 */ 1119u_int 1120isa_apic_mask(u_int isa_mask) 1121{ 1122 int isa_irq; 1123 int apic_pin; 1124 1125#if defined(SKIP_IRQ15_REDIRECT) 1126 if (isa_mask == (1 << 15)) { 1127 printf("skipping ISA IRQ15 redirect\n"); 1128 return isa_mask; 1129 } 1130#endif /* SKIP_IRQ15_REDIRECT */ 1131 1132 isa_irq = ffs(isa_mask); /* find its bit position */ 1133 if (isa_irq == 0) /* doesn't exist */ 1134 return 0; 1135 --isa_irq; /* make it zero based */ 1136 1137 apic_pin = isa_apic_pin(isa_irq); /* look for APIC connection */ 1138 if (apic_pin == -1) 1139 return 0; 1140 1141 return (1 << apic_pin); /* convert pin# to a mask */ 1142} 1143 1144 1145/* 1146 * Determine which APIC pin an ISA/EISA INT is attached to. 1147 */ 1148#define INTTYPE(I) (io_apic_ints[(I)].int_type) 1149#define INTPIN(I) (io_apic_ints[(I)].dst_apic_int) 1150 1151#define SRCBUSIRQ(I) (io_apic_ints[(I)].src_bus_irq) 1152int 1153isa_apic_pin(int isa_irq) 1154{ 1155 int intr; 1156 1157 for (intr = 0; intr < nintrs; ++intr) { /* check each record */ 1158 if (INTTYPE(intr) == 0) { /* standard INT */ 1159 if (SRCBUSIRQ(intr) == isa_irq) { 1160 if (apic_int_is_bus_type(intr, ISA) || 1161 apic_int_is_bus_type(intr, EISA)) 1162 return INTPIN(intr); /* found */ 1163 } 1164 } 1165 } 1166 return -1; /* NOT found */ 1167} 1168#undef SRCBUSIRQ 1169 1170 1171/* 1172 * Determine which APIC pin a PCI INT is attached to. 1173 */ 1174#define SRCBUSID(I) (io_apic_ints[(I)].src_bus_id) 1175#define SRCBUSDEVICE(I) ((io_apic_ints[(I)].src_bus_irq >> 2) & 0x1f) 1176#define SRCBUSLINE(I) (io_apic_ints[(I)].src_bus_irq & 0x03) 1177int 1178pci_apic_pin(int pciBus, int pciDevice, int pciInt) 1179{ 1180 int intr; 1181 1182 --pciInt; /* zero based */ 1183 1184 for (intr = 0; intr < nintrs; ++intr) /* check each record */ 1185 if ((INTTYPE(intr) == 0) /* standard INT */ 1186 && (SRCBUSID(intr) == pciBus) 1187 && (SRCBUSDEVICE(intr) == pciDevice) 1188 && (SRCBUSLINE(intr) == pciInt)) /* a candidate IRQ */ 1189 if (apic_int_is_bus_type(intr, PCI)) 1190 return INTPIN(intr); /* exact match */ 1191 1192 return -1; /* NOT found */ 1193} 1194#undef SRCBUSLINE 1195#undef SRCBUSDEVICE 1196#undef SRCBUSID 1197 1198#undef INTPIN 1199#undef INTTYPE 1200 1201 1202/* 1203 * Reprogram the MB chipset to NOT redirect an ISA INTerrupt. 1204 * 1205 * XXX FIXME: 1206 * Exactly what this means is unclear at this point. It is a solution 1207 * for motherboards that redirect the MBIRQ0 pin. Generically a motherboard 1208 * could route any of the ISA INTs to upper (>15) IRQ values. But most would 1209 * NOT be redirected via MBIRQ0, thus "undirect()ing" them would NOT be an 1210 * option. 1211 */ 1212int 1213undirect_isa_irq(int rirq) 1214{ 1215#if defined(READY) 1216 printf("Freeing redirected ISA irq %d.\n", rirq); 1217 /** FIXME: tickle the MB redirector chip */ 1218 return ???; 1219#else 1220 printf("Freeing (NOT implemented) redirected ISA irq %d.\n", rirq); 1221 return 0; 1222#endif /* READY */ 1223} 1224 1225 1226/* 1227 * Reprogram the MB chipset to NOT redirect a PCI INTerrupt 1228 */ 1229int 1230undirect_pci_irq(int rirq) 1231{ 1232#if defined(READY) 1233 if (bootverbose) 1234 printf("Freeing redirected PCI irq %d.\n", rirq); 1235 1236 /** FIXME: tickle the MB redirector chip */ 1237 return ???; 1238#else 1239 if (bootverbose) 1240 printf("Freeing (NOT implemented) redirected PCI irq %d.\n", 1241 rirq); 1242 return 0; 1243#endif /* READY */ 1244} 1245 1246 1247/* 1248 * given a bus ID, return: 1249 * the bus type if found 1250 * -1 if NOT found 1251 */ 1252int 1253apic_bus_type(int id) 1254{ 1255 int x; 1256 1257 for (x = 0; x < mp_nbusses; ++x) 1258 if (bus_data[x].bus_id == id) 1259 return bus_data[x].bus_type; 1260 1261 return -1; 1262} 1263 1264 1265/* 1266 * given a LOGICAL APIC# and pin#, return: 1267 * the associated src bus ID if found 1268 * -1 if NOT found 1269 */ 1270int 1271apic_src_bus_id(int apic, int pin) 1272{ 1273 int x; 1274 1275 /* search each of the possible INTerrupt sources */ 1276 for (x = 0; x < nintrs; ++x) 1277 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) && 1278 (pin == io_apic_ints[x].dst_apic_int)) 1279 return (io_apic_ints[x].src_bus_id); 1280 1281 return -1; /* NOT found */ 1282} 1283 1284 1285/* 1286 * given a LOGICAL APIC# and pin#, return: 1287 * the associated src bus IRQ if found 1288 * -1 if NOT found 1289 */ 1290int 1291apic_src_bus_irq(int apic, int pin) 1292{ 1293 int x; 1294 1295 for (x = 0; x < nintrs; x++) 1296 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) && 1297 (pin == io_apic_ints[x].dst_apic_int)) 1298 return (io_apic_ints[x].src_bus_irq); 1299 1300 return -1; /* NOT found */ 1301} 1302 1303 1304/* 1305 * given a LOGICAL APIC# and pin#, return: 1306 * the associated INTerrupt type if found 1307 * -1 if NOT found 1308 */ 1309int 1310apic_int_type(int apic, int pin) 1311{ 1312 int x; 1313 1314 /* search each of the possible INTerrupt sources */ 1315 for (x = 0; x < nintrs; ++x) 1316 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) && 1317 (pin == io_apic_ints[x].dst_apic_int)) 1318 return (io_apic_ints[x].int_type); 1319 1320 return -1; /* NOT found */ 1321} 1322 1323 1324/* 1325 * given a LOGICAL APIC# and pin#, return: 1326 * the associated trigger mode if found 1327 * -1 if NOT found 1328 */ 1329int 1330apic_trigger(int apic, int pin) 1331{ 1332 int x; 1333 1334 /* search each of the possible INTerrupt sources */ 1335 for (x = 0; x < nintrs; ++x) 1336 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) && 1337 (pin == io_apic_ints[x].dst_apic_int)) 1338 return ((io_apic_ints[x].int_flags >> 2) & 0x03); 1339 1340 return -1; /* NOT found */ 1341} 1342 1343 1344/* 1345 * given a LOGICAL APIC# and pin#, return: 1346 * the associated 'active' level if found 1347 * -1 if NOT found 1348 */ 1349int 1350apic_polarity(int apic, int pin) 1351{ 1352 int x; 1353 1354 /* search each of the possible INTerrupt sources */ 1355 for (x = 0; x < nintrs; ++x) 1356 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) && 1357 (pin == io_apic_ints[x].dst_apic_int)) 1358 return (io_apic_ints[x].int_flags & 0x03); 1359 1360 return -1; /* NOT found */ 1361} 1362 1363 1364/* 1365 * set data according to MP defaults 1366 * FIXME: probably not complete yet... 1367 */ 1368static void 1369default_mp_table(int type) 1370{ 1371 int ap_cpu_id; 1372#if defined(APIC_IO) 1373 u_int32_t ux; 1374 int io_apic_id; 1375 int pin; 1376#endif /* APIC_IO */ 1377 1378#if 0 1379 printf(" MP default config type: %d\n", type); 1380 switch (type) { 1381 case 1: 1382 printf(" bus: ISA, APIC: 82489DX\n"); 1383 break; 1384 case 2: 1385 printf(" bus: EISA, APIC: 82489DX\n"); 1386 break; 1387 case 3: 1388 printf(" bus: EISA, APIC: 82489DX\n"); 1389 break; 1390 case 4: 1391 printf(" bus: MCA, APIC: 82489DX\n"); 1392 break; 1393 case 5: 1394 printf(" bus: ISA+PCI, APIC: Integrated\n"); 1395 break; 1396 case 6: 1397 printf(" bus: EISA+PCI, APIC: Integrated\n"); 1398 break; 1399 case 7: 1400 printf(" bus: MCA+PCI, APIC: Integrated\n"); 1401 break; 1402 default: 1403 printf(" future type\n"); 1404 break; 1405 /* NOTREACHED */ 1406 } 1407#endif /* 0 */ 1408 1409 boot_cpu_id = (lapic.id & APIC_ID_MASK) >> 24; 1410 ap_cpu_id = (boot_cpu_id == 0) ? 1 : 0; 1411 1412 /* BSP */ 1413 CPU_TO_ID(0) = boot_cpu_id; 1414 ID_TO_CPU(boot_cpu_id) = 0; 1415 1416 /* one and only AP */ 1417 CPU_TO_ID(1) = ap_cpu_id; 1418 ID_TO_CPU(ap_cpu_id) = 1; 1419 1420#if defined(APIC_IO) 1421 /* one and only IO APIC */ 1422 io_apic_id = (io_apic_read(0, IOAPIC_ID) & APIC_ID_MASK) >> 24; 1423 1424 /* 1425 * sanity check, refer to MP spec section 3.6.6, last paragraph 1426 * necessary as some hardware isn't properly setting up the IO APIC 1427 */ 1428#if defined(REALLY_ANAL_IOAPICID_VALUE) 1429 if (io_apic_id != 2) { 1430#else 1431 if ((io_apic_id == 0) || (io_apic_id == 1) || (io_apic_id == 15)) { 1432#endif /* REALLY_ANAL_IOAPICID_VALUE */ 1433 ux = io_apic_read(0, IOAPIC_ID); /* get current contents */ 1434 ux &= ~APIC_ID_MASK; /* clear the ID field */ 1435 ux |= 0x02000000; /* set it to '2' */ 1436 io_apic_write(0, IOAPIC_ID, ux); /* write new value */ 1437 ux = io_apic_read(0, IOAPIC_ID); /* re-read && test */ 1438 if ((ux & APIC_ID_MASK) != 0x02000000) 1439 panic("can't control IO APIC ID, reg: 0x%08x", ux); 1440 io_apic_id = 2; 1441 } 1442 IO_TO_ID(0) = io_apic_id; 1443 ID_TO_IO(io_apic_id) = 0; 1444#endif /* APIC_IO */ 1445 1446 /* fill out bus entries */ 1447 switch (type) { 1448 case 1: 1449 case 2: 1450 case 3: 1451 case 5: 1452 case 6: 1453 bus_data[0].bus_id = default_data[type - 1][1]; 1454 bus_data[0].bus_type = default_data[type - 1][2]; 1455 bus_data[1].bus_id = default_data[type - 1][3]; 1456 bus_data[1].bus_type = default_data[type - 1][4]; 1457 break; 1458 1459 /* case 4: case 7: MCA NOT supported */ 1460 default: /* illegal/reserved */ 1461 panic("BAD default MP config: %d", type); 1462 /* NOTREACHED */ 1463 } 1464 1465#if defined(APIC_IO) 1466 /* general cases from MP v1.4, table 5-2 */ 1467 for (pin = 0; pin < 16; ++pin) { 1468 io_apic_ints[pin].int_type = 0; 1469 io_apic_ints[pin].int_flags = 0x05; /* edge/active-hi */ 1470 io_apic_ints[pin].src_bus_id = 0; 1471 io_apic_ints[pin].src_bus_irq = pin; /* IRQ2 caught below */ 1472 io_apic_ints[pin].dst_apic_id = io_apic_id; 1473 io_apic_ints[pin].dst_apic_int = pin; /* 1-to-1 */ 1474 } 1475 1476 /* special cases from MP v1.4, table 5-2 */ 1477 if (type == 2) { 1478 io_apic_ints[2].int_type = 0xff; /* N/C */ 1479 io_apic_ints[13].int_type = 0xff; /* N/C */ 1480#if !defined(APIC_MIXED_MODE) 1481 /** FIXME: ??? */ 1482 panic("sorry, can't support type 2 default yet"); 1483#endif /* APIC_MIXED_MODE */ 1484 } 1485 else 1486 io_apic_ints[2].src_bus_irq = 0; /* ISA IRQ0 is on APIC INT 2 */ 1487 1488 if (type == 7) 1489 io_apic_ints[0].int_type = 0xff; /* N/C */ 1490 else 1491 io_apic_ints[0].int_type = 3; /* vectored 8259 */ 1492#endif /* APIC_IO */ 1493} 1494 1495 1496/* 1497 * initialize all the SMP locks 1498 */ 1499 1500/* critical region around IO APIC, apic_imen */ 1501struct simplelock imen_lock; 1502 1503/* critical region around splxx(), cpl, cml, cil, ipending */ 1504struct simplelock cpl_lock; 1505 1506/* Make FAST_INTR() routines sequential */ 1507struct simplelock fast_intr_lock; 1508 1509/* critical region around INTR() routines */ 1510struct simplelock intr_lock; 1511 1512/* lock regions protected in UP kernel via cli/sti */ 1513struct simplelock mpintr_lock; 1514 1515/* lock region used by kernel profiling */ 1516struct simplelock mcount_lock; 1517 1518#ifdef USE_COMLOCK 1519/* locks com (tty) data/hardware accesses: a FASTINTR() */ 1520struct simplelock com_lock; 1521#endif /* USE_COMLOCK */ 1522 1523#ifdef USE_CLOCKLOCK 1524/* lock regions around the clock hardware */ 1525struct simplelock clock_lock; 1526#endif /* USE_CLOCKLOCK */ 1527 1528static void 1529init_locks(void) 1530{ 1531 /* 1532 * Get the initial mp_lock with a count of 1 for the BSP. 1533 * This uses a LOGICAL cpu ID, ie BSP == 0. 1534 */ 1535 mp_lock = 0x00000001; 1536 1537 /* ISR uses its own "giant lock" */ 1538 isr_lock = FREE_LOCK; 1539 1540 s_lock_init((struct simplelock*)&mpintr_lock); 1541 1542 s_lock_init((struct simplelock*)&mcount_lock); 1543 1544 s_lock_init((struct simplelock*)&fast_intr_lock); 1545 s_lock_init((struct simplelock*)&intr_lock); 1546 s_lock_init((struct simplelock*)&imen_lock); 1547 s_lock_init((struct simplelock*)&cpl_lock); 1548 1549#ifdef USE_COMLOCK 1550 s_lock_init((struct simplelock*)&com_lock); 1551#endif /* USE_COMLOCK */ 1552#ifdef USE_CLOCKLOCK 1553 s_lock_init((struct simplelock*)&clock_lock); 1554#endif /* USE_CLOCKLOCK */ 1555} 1556 1557 1558/* 1559 * start each AP in our list 1560 */ 1561static int 1562start_all_aps(u_int boot_addr) 1563{ 1564 int x, i; 1565 u_char mpbiosreason; 1566 u_long mpbioswarmvec; 1567 pd_entry_t *newptd; 1568 pt_entry_t *newpt; 1569 int *newpp; 1570 char *stack; 1571 pd_entry_t *myPTD; 1572 1573 POSTCODE(START_ALL_APS_POST); 1574 1575 /* initialize BSP's local APIC */ 1576 apic_initialize(); 1577 bsp_apic_ready = 1; 1578 1579 /* install the AP 1st level boot code */ 1580 install_ap_tramp(boot_addr); 1581 1582 1583 /* save the current value of the warm-start vector */ 1584 mpbioswarmvec = *((u_long *) WARMBOOT_OFF); 1585 outb(CMOS_REG, BIOS_RESET); 1586 mpbiosreason = inb(CMOS_DATA); 1587 1588 /* record BSP in CPU map */ 1589 all_cpus = 1; 1590 1591 /* start each AP */ 1592 for (x = 1; x <= mp_naps; ++x) { 1593 1594 /* This is a bit verbose, it will go away soon. */ 1595 1596 /* alloc new page table directory */ 1597 newptd = (pd_entry_t *)(kmem_alloc(kernel_map, PAGE_SIZE)); 1598 1599 /* Store the virtual PTD address for this CPU */ 1600 IdlePTDS[x] = newptd; 1601 1602 /* clone currently active one (ie: IdlePTD) */ 1603 bcopy(PTD, newptd, PAGE_SIZE); /* inc prv page pde */ 1604 1605 /* set up 0 -> 4MB P==V mapping for AP boot */ 1606 newptd[0] = (pd_entry_t) (PG_V | PG_RW | 1607 ((u_long)KPTphys & PG_FRAME)); 1608 1609 /* store PTD for this AP's boot sequence */ 1610 myPTD = (pd_entry_t *)vtophys(newptd); 1611 1612 /* alloc new page table page */ 1613 newpt = (pt_entry_t *)(kmem_alloc(kernel_map, PAGE_SIZE)); 1614 1615 /* set the new PTD's private page to point there */ 1616 newptd[MPPTDI] = (pt_entry_t)(PG_V | PG_RW | vtophys(newpt)); 1617 1618 /* install self referential entry */ 1619 newptd[PTDPTDI] = (pd_entry_t)(PG_V | PG_RW | vtophys(newptd)); 1620 1621 /* allocate a new private data page */ 1622 newpp = (int *)kmem_alloc(kernel_map, PAGE_SIZE); 1623 1624 /* wire it into the private page table page */ 1625 newpt[0] = (pt_entry_t)(PG_V | PG_RW | vtophys(newpp)); 1626 1627 /* wire the ptp into itself for access */ 1628 newpt[1] = (pt_entry_t)(PG_V | PG_RW | vtophys(newpt)); 1629 1630 /* copy in the pointer to the local apic */ 1631 newpt[2] = SMP_prvpt[2]; 1632 1633 /* and the IO apic mapping[s] */ 1634 for (i = 16; i < 32; i++) 1635 newpt[i] = SMP_prvpt[i]; 1636 1637 /* allocate and set up an idle stack data page */ 1638 stack = (char *)kmem_alloc(kernel_map, UPAGES*PAGE_SIZE); 1639 for (i = 0; i < UPAGES; i++) 1640 newpt[i + 3] = (pt_entry_t)(PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack)); 1641 1642 newpt[3 + UPAGES] = 0; /* *prv_CMAP1 */ 1643 newpt[4 + UPAGES] = 0; /* *prv_CMAP2 */ 1644 newpt[5 + UPAGES] = 0; /* *prv_CMAP3 */ 1645 1646 /* prime data page for it to use */ 1647 newpp[0] = x; /* cpuid */ 1648 newpp[1] = 0; /* curproc */ 1649 newpp[2] = 0; /* curpcb */ 1650 newpp[3] = 0; /* npxproc */ 1651 newpp[4] = 0; /* runtime.tv_sec */ 1652 newpp[5] = 0; /* runtime.tv_usec */ 1653 newpp[6] = x << 24; /* cpu_lockid */ 1654 newpp[7] = 0; /* other_cpus */ 1655 newpp[8] = (int)myPTD; /* my_idlePTD */ 1656 newpp[9] = 0; /* ss_tpr */ 1657 newpp[10] = (int)&newpt[3 + UPAGES]; /* prv_CMAP1 */ 1658 newpp[11] = (int)&newpt[4 + UPAGES]; /* prv_CMAP2 */ 1659 newpp[12] = (int)&newpt[5 + UPAGES]; /* prv_CMAP3 */ 1660 1661 /* setup a vector to our boot code */ 1662 *((volatile u_short *) WARMBOOT_OFF) = WARMBOOT_TARGET; 1663 *((volatile u_short *) WARMBOOT_SEG) = (boot_addr >> 4); 1664 outb(CMOS_REG, BIOS_RESET); 1665 outb(CMOS_DATA, BIOS_WARM); /* 'warm-start' */ 1666 1667 bootPTD = myPTD; 1668 /* attempt to start the Application Processor */ 1669 CHECK_INIT(99); /* setup checkpoints */ 1670 if (!start_ap(x, boot_addr)) { 1671 printf("AP #%d (PHY# %d) failed!\n", x, CPU_TO_ID(x)); 1672 CHECK_PRINT("trace"); /* show checkpoints */ 1673 /* better panic as the AP may be running loose */ 1674 printf("panic y/n? [y] "); 1675 if (cngetc() != 'n') 1676 panic("bye-bye"); 1677 } 1678 CHECK_PRINT("trace"); /* show checkpoints */ 1679 1680 /* record its version info */ 1681 cpu_apic_versions[x] = cpu_apic_versions[0]; 1682 1683 all_cpus |= (1 << x); /* record AP in CPU map */ 1684 } 1685 1686 /* build our map of 'other' CPUs */ 1687 other_cpus = all_cpus & ~(1 << cpuid); 1688 1689 /* fill in our (BSP) APIC version */ 1690 cpu_apic_versions[0] = lapic.version; 1691 1692 /* restore the warmstart vector */ 1693 *(u_long *) WARMBOOT_OFF = mpbioswarmvec; 1694 outb(CMOS_REG, BIOS_RESET); 1695 outb(CMOS_DATA, mpbiosreason); 1696 1697 /* 1698 * Set up the idle context for the BSP. Similar to above except 1699 * that some was done by locore, some by pmap.c and some is implicit 1700 * because the BSP is cpu#0 and the page is initially zero, and also 1701 * because we can refer to variables by name on the BSP.. 1702 */ 1703 newptd = (pd_entry_t *)(kmem_alloc(kernel_map, PAGE_SIZE)); 1704 1705 bcopy(PTD, newptd, PAGE_SIZE); /* inc prv page pde */ 1706 IdlePTDS[0] = newptd; 1707 1708 /* Point PTD[] to this page instead of IdlePTD's physical page */ 1709 newptd[PTDPTDI] = (pd_entry_t)(PG_V | PG_RW | vtophys(newptd)); 1710 1711 my_idlePTD = (pd_entry_t *)vtophys(newptd); 1712 1713 /* Allocate and setup BSP idle stack */ 1714 stack = (char *)kmem_alloc(kernel_map, UPAGES * PAGE_SIZE); 1715 for (i = 0; i < UPAGES; i++) 1716 SMP_prvpt[i + 3] = (pt_entry_t)(PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack)); 1717 1718 pmap_set_opt_bsp(); 1719 1720 for (i = 0; i < mp_ncpus; i++) { 1721 bcopy( (int *) PTD + KPTDI, (int *) IdlePTDS[i] + KPTDI, NKPDE * sizeof (int)); 1722 } 1723 1724 /* number of APs actually started */ 1725 return mp_ncpus - 1; 1726} 1727 1728 1729/* 1730 * load the 1st level AP boot code into base memory. 1731 */ 1732 1733/* targets for relocation */ 1734extern void bigJump(void); 1735extern void bootCodeSeg(void); 1736extern void bootDataSeg(void); 1737extern void MPentry(void); 1738extern u_int MP_GDT; 1739extern u_int mp_gdtbase; 1740 1741static void 1742install_ap_tramp(u_int boot_addr) 1743{ 1744 int x; 1745 int size = *(int *) ((u_long) & bootMP_size); 1746 u_char *src = (u_char *) ((u_long) bootMP); 1747 u_char *dst = (u_char *) boot_addr + KERNBASE; 1748 u_int boot_base = (u_int) bootMP; 1749 u_int8_t *dst8; 1750 u_int16_t *dst16; 1751 u_int32_t *dst32; 1752 1753 POSTCODE(INSTALL_AP_TRAMP_POST); 1754 1755 for (x = 0; x < size; ++x) 1756 *dst++ = *src++; 1757 1758 /* 1759 * modify addresses in code we just moved to basemem. unfortunately we 1760 * need fairly detailed info about mpboot.s for this to work. changes 1761 * to mpboot.s might require changes here. 1762 */ 1763 1764 /* boot code is located in KERNEL space */ 1765 dst = (u_char *) boot_addr + KERNBASE; 1766 1767 /* modify the lgdt arg */ 1768 dst32 = (u_int32_t *) (dst + ((u_int) & mp_gdtbase - boot_base)); 1769 *dst32 = boot_addr + ((u_int) & MP_GDT - boot_base); 1770 1771 /* modify the ljmp target for MPentry() */ 1772 dst32 = (u_int32_t *) (dst + ((u_int) bigJump - boot_base) + 1); 1773 *dst32 = ((u_int) MPentry - KERNBASE); 1774 1775 /* modify the target for boot code segment */ 1776 dst16 = (u_int16_t *) (dst + ((u_int) bootCodeSeg - boot_base)); 1777 dst8 = (u_int8_t *) (dst16 + 1); 1778 *dst16 = (u_int) boot_addr & 0xffff; 1779 *dst8 = ((u_int) boot_addr >> 16) & 0xff; 1780 1781 /* modify the target for boot data segment */ 1782 dst16 = (u_int16_t *) (dst + ((u_int) bootDataSeg - boot_base)); 1783 dst8 = (u_int8_t *) (dst16 + 1); 1784 *dst16 = (u_int) boot_addr & 0xffff; 1785 *dst8 = ((u_int) boot_addr >> 16) & 0xff; 1786} 1787 1788 1789/* 1790 * this function starts the AP (application processor) identified 1791 * by the APIC ID 'physicalCpu'. It does quite a "song and dance" 1792 * to accomplish this. This is necessary because of the nuances 1793 * of the different hardware we might encounter. It ain't pretty, 1794 * but it seems to work. 1795 */ 1796static int 1797start_ap(int logical_cpu, u_int boot_addr) 1798{ 1799 int physical_cpu; 1800 int vector; 1801 int cpus; 1802 u_long icr_lo, icr_hi; 1803 1804 POSTCODE(START_AP_POST); 1805 1806 /* get the PHYSICAL APIC ID# */ 1807 physical_cpu = CPU_TO_ID(logical_cpu); 1808 1809 /* calculate the vector */ 1810 vector = (boot_addr >> 12) & 0xff; 1811 1812 /* used as a watchpoint to signal AP startup */ 1813 cpus = mp_ncpus; 1814 1815 /* 1816 * first we do an INIT/RESET IPI this INIT IPI might be run, reseting 1817 * and running the target CPU. OR this INIT IPI might be latched (P5 1818 * bug), CPU waiting for STARTUP IPI. OR this INIT IPI might be 1819 * ignored. 1820 */ 1821 1822 /* setup the address for the target AP */ 1823 icr_hi = lapic.icr_hi & ~APIC_ID_MASK; 1824 icr_hi |= (physical_cpu << 24); 1825 lapic.icr_hi = icr_hi; 1826 1827 /* do an INIT IPI: assert RESET */ 1828 icr_lo = lapic.icr_lo & 0xfff00000; 1829 lapic.icr_lo = icr_lo | 0x0000c500; 1830 1831 /* wait for pending status end */ 1832 while (lapic.icr_lo & APIC_DELSTAT_MASK) 1833 /* spin */ ; 1834 1835 /* do an INIT IPI: deassert RESET */ 1836 lapic.icr_lo = icr_lo | 0x00008500; 1837 1838 /* wait for pending status end */ 1839 u_sleep(10000); /* wait ~10mS */ 1840 while (lapic.icr_lo & APIC_DELSTAT_MASK) 1841 /* spin */ ; 1842 1843 /* 1844 * next we do a STARTUP IPI: the previous INIT IPI might still be 1845 * latched, (P5 bug) this 1st STARTUP would then terminate 1846 * immediately, and the previously started INIT IPI would continue. OR 1847 * the previous INIT IPI has already run. and this STARTUP IPI will 1848 * run. OR the previous INIT IPI was ignored. and this STARTUP IPI 1849 * will run. 1850 */ 1851 1852 /* do a STARTUP IPI */ 1853 lapic.icr_lo = icr_lo | 0x00000600 | vector; 1854 while (lapic.icr_lo & APIC_DELSTAT_MASK) 1855 /* spin */ ; 1856 u_sleep(200); /* wait ~200uS */ 1857 1858 /* 1859 * finally we do a 2nd STARTUP IPI: this 2nd STARTUP IPI should run IF 1860 * the previous STARTUP IPI was cancelled by a latched INIT IPI. OR 1861 * this STARTUP IPI will be ignored, as only ONE STARTUP IPI is 1862 * recognized after hardware RESET or INIT IPI. 1863 */ 1864 1865 lapic.icr_lo = icr_lo | 0x00000600 | vector; 1866 while (lapic.icr_lo & APIC_DELSTAT_MASK) 1867 /* spin */ ; 1868 u_sleep(200); /* wait ~200uS */ 1869 1870 /* wait for it to start */ 1871 set_apic_timer(5000000);/* == 5 seconds */ 1872 while (read_apic_timer()) 1873 if (mp_ncpus > cpus) 1874 return 1; /* return SUCCESS */ 1875 1876 return 0; /* return FAILURE */ 1877} 1878 1879 1880/* 1881 * Flush the TLB on all other CPU's 1882 * 1883 * XXX: Needs to handshake and wait for completion before proceding. 1884 */ 1885void 1886smp_invltlb(void) 1887{ 1888#if defined(APIC_IO) 1889 if (smp_started && invltlb_ok) 1890 all_but_self_ipi(XINVLTLB_OFFSET); 1891#endif /* APIC_IO */ 1892} 1893 1894void 1895invlpg(u_int addr) 1896{ 1897 __asm __volatile("invlpg (%0)"::"r"(addr):"memory"); 1898 1899 /* send a message to the other CPUs */ 1900 smp_invltlb(); 1901} 1902 1903void 1904invltlb(void) 1905{ 1906 u_long temp; 1907 1908 /* 1909 * This should be implemented as load_cr3(rcr3()) when load_cr3() is 1910 * inlined. 1911 */ 1912 __asm __volatile("movl %%cr3, %0; movl %0, %%cr3":"=r"(temp) :: "memory"); 1913 1914 /* send a message to the other CPUs */ 1915 smp_invltlb(); 1916} 1917 1918 1919/* 1920 * When called the executing CPU will send an IPI to all other CPUs 1921 * requesting that they halt execution. 1922 * 1923 * Usually (but not necessarily) called with 'other_cpus' as its arg. 1924 * 1925 * - Signals all CPUs in map to stop. 1926 * - Waits for each to stop. 1927 * 1928 * Returns: 1929 * -1: error 1930 * 0: NA 1931 * 1: ok 1932 * 1933 * XXX FIXME: this is not MP-safe, needs a lock to prevent multiple CPUs 1934 * from executing at same time. 1935 */ 1936int 1937stop_cpus(u_int map) 1938{ 1939 if (!smp_started) 1940 return 0; 1941 1942 /* send IPI to all CPUs in map */ 1943 stopped_cpus = 0; 1944 1945 /* send the Xcpustop IPI to all CPUs in map */ 1946 selected_apic_ipi(map, XCPUSTOP_OFFSET, APIC_DELMODE_FIXED); 1947 1948 while (stopped_cpus != map) 1949 /* spin */ ; 1950 1951 return 1; 1952} 1953 1954 1955/* 1956 * Called by a CPU to restart stopped CPUs. 1957 * 1958 * Usually (but not necessarily) called with 'stopped_cpus' as its arg. 1959 * 1960 * - Signals all CPUs in map to restart. 1961 * - Waits for each to restart. 1962 * 1963 * Returns: 1964 * -1: error 1965 * 0: NA 1966 * 1: ok 1967 */ 1968int 1969restart_cpus(u_int map) 1970{ 1971 if (!smp_started) 1972 return 0; 1973 1974 started_cpus = map; /* signal other cpus to restart */ 1975 1976 while (started_cpus) /* wait for each to clear its bit */ 1977 /* spin */ ; 1978 stopped_cpus = 0; 1979 1980 return 1; 1981} 1982 1983int smp_active = 0; /* are the APs allowed to run? */ 1984SYSCTL_INT(_machdep, OID_AUTO, smp_active, CTLFLAG_RW, &smp_active, 0, ""); 1985 1986/* XXX maybe should be hw.ncpu */ 1987static int smp_cpus = 1; /* how many cpu's running */ 1988SYSCTL_INT(_machdep, OID_AUTO, smp_cpus, CTLFLAG_RD, &smp_cpus, 0, ""); 1989 1990int invltlb_ok = 0; /* throttle smp_invltlb() till safe */ 1991SYSCTL_INT(_machdep, OID_AUTO, invltlb_ok, CTLFLAG_RW, &invltlb_ok, 0, ""); 1992 1993/* Warning: Do not staticize. Used from swtch.s */ 1994int do_page_zero_idle = 1; /* bzero pages for fun and profit in idleloop */ 1995SYSCTL_INT(_machdep, OID_AUTO, do_page_zero_idle, CTLFLAG_RW, 1996 &do_page_zero_idle, 0, ""); 1997 1998 1999/* 2000 * This is called once the rest of the system is up and running and we're 2001 * ready to let the AP's out of the pen. 2002 */ 2003void ap_init(void); 2004 2005void 2006ap_init() 2007{ 2008 u_int temp; 2009 u_int apic_id; 2010 2011 smp_cpus++; 2012 2013 /* Build our map of 'other' CPUs. */ 2014 other_cpus = all_cpus & ~(1 << cpuid); 2015 2016 printf("SMP: AP CPU #%d Launched!\n", cpuid); 2017 2018 /* XXX FIXME: i386 specific, and redundant: Setup the FPU. */ 2019 load_cr0((rcr0() & ~CR0_EM) | CR0_MP | CR0_NE | CR0_TS); 2020 2021 /* A quick check from sanity claus */ 2022 apic_id = (apic_id_to_logical[(lapic.id & 0x0f000000) >> 24]); 2023 if (cpuid != apic_id) { 2024 printf("SMP: cpuid = %d\n", cpuid); 2025 printf("SMP: apic_id = %d\n", apic_id); 2026 printf("PTD[MPPTDI] = %08x\n", PTD[MPPTDI]); 2027 panic("cpuid mismatch! boom!!"); 2028 } 2029 2030 /* Init local apic for irq's */ 2031 apic_initialize(); 2032 2033 /* 2034 * Activate smp_invltlb, although strictly speaking, this isn't 2035 * quite correct yet. We should have a bitfield for cpus willing 2036 * to accept TLB flush IPI's or something and sync them. 2037 */ 2038 invltlb_ok = 1; 2039 smp_started = 1; /* enable IPI's, tlb shootdown, freezes etc */ 2040 smp_active = 1; /* historic */ 2041 2042 curproc = NULL; /* make sure */ 2043} 2044 2045void 2046getmtrr() 2047{ 2048 int i; 2049 2050 if (cpu_class == CPUCLASS_686) { 2051 for(i = 0; i < NPPROVMTRR; i++) { 2052 PPro_vmtrr[i].base = rdmsr(PPRO_VMTRRphysBase0 + i * 2); 2053 PPro_vmtrr[i].mask = rdmsr(PPRO_VMTRRphysMask0 + i * 2); 2054 } 2055 } 2056} 2057 2058void 2059putmtrr() 2060{ 2061 int i; 2062 2063 if (cpu_class == CPUCLASS_686) { 2064 wbinvd(); 2065 for(i = 0; i < NPPROVMTRR; i++) { 2066 wrmsr(PPRO_VMTRRphysBase0 + i * 2, PPro_vmtrr[i].base); 2067 wrmsr(PPRO_VMTRRphysMask0 + i * 2, PPro_vmtrr[i].mask); 2068 } 2069 } 2070} 2071 2072void 2073putfmtrr() 2074{ 2075 if (cpu_class == CPUCLASS_686) { 2076 wbinvd(); 2077 /* 2078 * Set memory between 0-640K to be WB 2079 */ 2080 wrmsr(0x250, 0x0606060606060606LL); 2081 wrmsr(0x258, 0x0606060606060606LL); 2082 /* 2083 * Set normal, PC video memory to be WC 2084 */ 2085 wrmsr(0x259, 0x0101010101010101LL); 2086 } 2087} 2088 2089 2090#ifdef BETTER_CLOCK 2091 2092#define CHECKSTATE_USER 0 2093#define CHECKSTATE_SYS 1 2094#define CHECKSTATE_INTR 2 2095 2096/* Do not staticize. Used from apic_vector.s */ 2097struct proc* checkstate_curproc[NCPU]; 2098int checkstate_cpustate[NCPU]; 2099u_long checkstate_pc[NCPU]; 2100 2101extern long cp_time[CPUSTATES]; 2102 2103#define PC_TO_INDEX(pc, prof) \ 2104 ((int)(((u_quad_t)((pc) - (prof)->pr_off) * \ 2105 (u_quad_t)((prof)->pr_scale)) >> 16) & ~1) 2106 2107static void 2108addupc_intr_forwarded(struct proc *p, int id, int *astmap) 2109{ 2110 int i; 2111 struct uprof *prof; 2112 u_long pc; 2113 2114 pc = checkstate_pc[id]; 2115 prof = &p->p_stats->p_prof; 2116 if (pc >= prof->pr_off && 2117 (i = PC_TO_INDEX(pc, prof)) < prof->pr_size) { 2118 if ((p->p_flag & P_OWEUPC) == 0) { 2119 prof->pr_addr = pc; 2120 prof->pr_ticks = 1; 2121 p->p_flag |= P_OWEUPC; 2122 } 2123 *astmap |= (1 << id); 2124 } 2125} 2126 2127static void 2128forwarded_statclock(int id, int pscnt, int *astmap) 2129{ 2130 struct pstats *pstats; 2131 long rss; 2132 struct rusage *ru; 2133 struct vmspace *vm; 2134 int cpustate; 2135 struct proc *p; 2136#ifdef GPROF 2137 register struct gmonparam *g; 2138 int i; 2139#endif 2140 2141 p = checkstate_curproc[id]; 2142 cpustate = checkstate_cpustate[id]; 2143 2144 switch (cpustate) { 2145 case CHECKSTATE_USER: 2146 if (p->p_flag & P_PROFIL) 2147 addupc_intr_forwarded(p, id, astmap); 2148 if (pscnt > 1) 2149 return; 2150 p->p_uticks++; 2151 if (p->p_nice > NZERO) 2152 cp_time[CP_NICE]++; 2153 else 2154 cp_time[CP_USER]++; 2155 break; 2156 case CHECKSTATE_SYS: 2157#ifdef GPROF 2158 /* 2159 * Kernel statistics are just like addupc_intr, only easier. 2160 */ 2161 g = &_gmonparam; 2162 if (g->state == GMON_PROF_ON) { 2163 i = checkstate_pc[id] - g->lowpc; 2164 if (i < g->textsize) { 2165 i /= HISTFRACTION * sizeof(*g->kcount); 2166 g->kcount[i]++; 2167 } 2168 } 2169#endif 2170 if (pscnt > 1) 2171 return; 2172 2173 if (!p) 2174 cp_time[CP_IDLE]++; 2175 else { 2176 p->p_sticks++; 2177 cp_time[CP_SYS]++; 2178 } 2179 break; 2180 case CHECKSTATE_INTR: 2181 default: 2182#ifdef GPROF 2183 /* 2184 * Kernel statistics are just like addupc_intr, only easier. 2185 */ 2186 g = &_gmonparam; 2187 if (g->state == GMON_PROF_ON) { 2188 i = checkstate_pc[id] - g->lowpc; 2189 if (i < g->textsize) { 2190 i /= HISTFRACTION * sizeof(*g->kcount); 2191 g->kcount[i]++; 2192 } 2193 } 2194#endif 2195 if (pscnt > 1) 2196 return; 2197 if (p) 2198 p->p_iticks++; 2199 cp_time[CP_INTR]++; 2200 } 2201 if (p != NULL) { 2202 p->p_cpticks++; 2203 if (++p->p_estcpu == 0) 2204 p->p_estcpu--; 2205 if ((p->p_estcpu & 3) == 0) { 2206 resetpriority(p); 2207 if (p->p_priority >= PUSER) 2208 p->p_priority = p->p_usrpri; 2209 } 2210 2211 /* Update resource usage integrals and maximums. */ 2212 if ((pstats = p->p_stats) != NULL && 2213 (ru = &pstats->p_ru) != NULL && 2214 (vm = p->p_vmspace) != NULL) { 2215 ru->ru_ixrss += vm->vm_tsize * PAGE_SIZE / 1024; 2216 ru->ru_idrss += vm->vm_dsize * PAGE_SIZE / 1024; 2217 ru->ru_isrss += vm->vm_ssize * PAGE_SIZE / 1024; 2218 rss = vm->vm_pmap.pm_stats.resident_count * 2219 PAGE_SIZE / 1024; 2220 if (ru->ru_maxrss < rss) 2221 ru->ru_maxrss = rss; 2222 } 2223 } 2224} 2225 2226void 2227forward_statclock(int pscnt) 2228{ 2229 int map; 2230 int id; 2231 int i; 2232 2233 /* Kludge. We don't yet have separate locks for the interrupts 2234 * and the kernel. This means that we cannot let the other processors 2235 * handle complex interrupts while inhibiting them from entering 2236 * the kernel in a non-interrupt context. 2237 * 2238 * What we can do, without changing the locking mechanisms yet, 2239 * is letting the other processors handle a very simple interrupt 2240 * (wich determines the processor states), and do the main 2241 * work ourself. 2242 */ 2243 2244 if (!smp_started || !invltlb_ok || cold || panicstr) 2245 return; 2246 2247 /* Step 1: Probe state (user, cpu, interrupt, spinlock, idle ) */ 2248 2249 map = other_cpus & ~stopped_cpus ; 2250 checkstate_probed_cpus = 0; 2251 if (map != 0) 2252 selected_apic_ipi(map, 2253 XCPUCHECKSTATE_OFFSET, APIC_DELMODE_FIXED); 2254 2255 i = 0; 2256 while (checkstate_probed_cpus != map) { 2257 /* spin */ 2258 i++; 2259 if (i == 1000000) { 2260 printf("forward_statclock: checkstate %x\n", 2261 checkstate_probed_cpus); 2262 break; 2263 } 2264 } 2265 2266 /* 2267 * Step 2: walk through other processors processes, update ticks and 2268 * profiling info. 2269 */ 2270 2271 map = 0; 2272 for (id = 0; id < mp_ncpus; id++) { 2273 if (id == cpuid) 2274 continue; 2275 if (((1 << id) & checkstate_probed_cpus) == 0) 2276 continue; 2277 forwarded_statclock(id, pscnt, &map); 2278 } 2279 if (map != 0) { 2280 checkstate_need_ast |= map; 2281 selected_apic_ipi(map, XCPUAST_OFFSET, APIC_DELMODE_FIXED); 2282 i = 0; 2283 while ((checkstate_need_ast & map) != 0) { 2284 /* spin */ 2285 i++; 2286 if (i > 100000) { 2287#ifdef BETTER_CLOCK_DIAGNOSTIC 2288 printf("forward_statclock: dropped ast 0x%x\n", 2289 checkstate_need_ast & map); 2290#endif 2291 break; 2292 } 2293 } 2294 } 2295} 2296 2297void 2298forward_hardclock(int pscnt) 2299{ 2300 int map; 2301 int id; 2302 struct proc *p; 2303 struct pstats *pstats; 2304 int i; 2305 2306 /* Kludge. We don't yet have separate locks for the interrupts 2307 * and the kernel. This means that we cannot let the other processors 2308 * handle complex interrupts while inhibiting them from entering 2309 * the kernel in a non-interrupt context. 2310 * 2311 * What we can do, without changing the locking mechanisms yet, 2312 * is letting the other processors handle a very simple interrupt 2313 * (wich determines the processor states), and do the main 2314 * work ourself. 2315 */ 2316 2317 if (!smp_started || !invltlb_ok || cold || panicstr) 2318 return; 2319 2320 /* Step 1: Probe state (user, cpu, interrupt, spinlock, idle) */ 2321 2322 map = other_cpus & ~stopped_cpus ; 2323 checkstate_probed_cpus = 0; 2324 if (map != 0) 2325 selected_apic_ipi(map, 2326 XCPUCHECKSTATE_OFFSET, APIC_DELMODE_FIXED); 2327 2328 i = 0; 2329 while (checkstate_probed_cpus != map) { 2330 /* spin */ 2331 i++; 2332 if (i == 1000000) { 2333 printf("forward_hardclock: checkstate %x\n", 2334 checkstate_probed_cpus); 2335 break; 2336 } 2337 } 2338 2339 /* 2340 * Step 2: walk through other processors processes, update virtual 2341 * timer and profiling timer. If stathz == 0, also update ticks and 2342 * profiling info. 2343 */ 2344 2345 map = 0; 2346 for (id = 0; id < mp_ncpus; id++) { 2347 if (id == cpuid) 2348 continue; 2349 if (((1 << id) & checkstate_probed_cpus) == 0) 2350 continue; 2351 p = checkstate_curproc[id]; 2352 if (p) { 2353 pstats = p->p_stats; 2354 if (checkstate_cpustate[id] == CHECKSTATE_USER && 2355 timerisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && 2356 itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) { 2357 psignal(p, SIGVTALRM); 2358 map |= (1 << id); 2359 } 2360 if (timerisset(&pstats->p_timer[ITIMER_PROF].it_value) && 2361 itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) { 2362 psignal(p, SIGPROF); 2363 map |= (1 << id); 2364 } 2365 } 2366 if (stathz == 0) { 2367 forwarded_statclock( id, pscnt, &map); 2368 } 2369 } 2370 if (map != 0) { 2371 checkstate_need_ast |= map; 2372 selected_apic_ipi(map, XCPUAST_OFFSET, APIC_DELMODE_FIXED); 2373 i = 0; 2374 while ((checkstate_need_ast & map) != 0) { 2375 /* spin */ 2376 i++; 2377 if (i > 100000) { 2378#ifdef BETTER_CLOCK_DIAGNOSTIC 2379 printf("forward_hardclock: dropped ast 0x%x\n", 2380 checkstate_need_ast & map); 2381#endif 2382 break; 2383 } 2384 } 2385 } 2386} 2387 2388#endif /* BETTER_CLOCK */ 2389