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 * $FreeBSD$ 26 */ 27 28 #include "opt_cpu.h" 29 #include "opt_user_ldt.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/bus.h> 39 #include <sys/systm.h> 40 #include <sys/kernel.h> 41 #include <sys/proc.h> 42 #include <sys/sysctl.h> 43 #include <sys/malloc.h> 44 #include <sys/memrange.h> 45 #include <sys/mutex.h> 46 #ifdef BETTER_CLOCK 47 #include <sys/dkstat.h> 48 #endif 49 #include <sys/cons.h> /* cngetc() */ 50 51 #include <vm/vm.h> 52 #include <vm/vm_param.h> 53 #include <vm/pmap.h> 54 #include <vm/vm_kern.h> 55 #include <vm/vm_extern.h> 56 #ifdef BETTER_CLOCK 57 #include <sys/lock.h> 58 #include <vm/vm_map.h> 59 #include <sys/user.h> 60 #ifdef GPROF 61 #include <sys/gmon.h> 62 #endif 63 #endif 64 65 #include <machine/smp.h> 66 #include <machine/apic.h> 67 #include <machine/atomic.h> 68 #include <machine/cpufunc.h> 69 #include <machine/mpapic.h> 70 #include <machine/psl.h> 71 #include <machine/segments.h> 72 #include <machine/smptests.h> /** TEST_DEFAULT_CONFIG, TEST_TEST1 */ 73 #include <machine/tss.h> 74 #include <machine/specialreg.h> 75 #include <machine/globaldata.h> 76 77 #if defined(APIC_IO) 78 #include <machine/md_var.h> /* setidt() */ 79 #include <i386/isa/icu.h> /* IPIs */ 80 #include <i386/isa/intr_machdep.h> /* IPIs */ 81 #endif /* APIC_IO */ 82 83 #if defined(TEST_DEFAULT_CONFIG) 84 #define MPFPS_MPFB1 TEST_DEFAULT_CONFIG 85 #else 86 #define MPFPS_MPFB1 mpfps->mpfb1 87 #endif /* TEST_DEFAULT_CONFIG */ 88 89 #define WARMBOOT_TARGET 0 90 #define WARMBOOT_OFF (KERNBASE + 0x0467) 91 #define WARMBOOT_SEG (KERNBASE + 0x0469) 92 93 #ifdef PC98 94 #define BIOS_BASE (0xe8000) 95 #define BIOS_SIZE (0x18000) 96 #else 97 #define BIOS_BASE (0xf0000) 98 #define BIOS_SIZE (0x10000) 99 #endif 100 #define BIOS_COUNT (BIOS_SIZE/4) 101 102 #define CMOS_REG (0x70) 103 #define CMOS_DATA (0x71) 104 #define BIOS_RESET (0x0f) 105 #define BIOS_WARM (0x0a) 106 107 #define PROCENTRY_FLAG_EN 0x01 108 #define PROCENTRY_FLAG_BP 0x02 109 #define IOAPICENTRY_FLAG_EN 0x01 110 111 112 /* MP Floating Pointer Structure */ 113 typedef struct MPFPS { 114 char signature[4]; 115 void *pap; 116 u_char length; 117 u_char spec_rev; 118 u_char checksum; 119 u_char mpfb1; 120 u_char mpfb2; 121 u_char mpfb3; 122 u_char mpfb4; 123 u_char mpfb5; 124 } *mpfps_t; 125 126 /* MP Configuration Table Header */ 127 typedef struct MPCTH { 128 char signature[4]; 129 u_short base_table_length; 130 u_char spec_rev; 131 u_char checksum; 132 u_char oem_id[8]; 133 u_char product_id[12]; 134 void *oem_table_pointer; 135 u_short oem_table_size; 136 u_short entry_count; 137 void *apic_address; 138 u_short extended_table_length; 139 u_char extended_table_checksum; 140 u_char reserved; 141 } *mpcth_t; 142 143 144 typedef struct PROCENTRY { 145 u_char type; 146 u_char apic_id; 147 u_char apic_version; 148 u_char cpu_flags; 149 u_long cpu_signature; 150 u_long feature_flags; 151 u_long reserved1; 152 u_long reserved2; 153 } *proc_entry_ptr; 154 155 typedef struct BUSENTRY { 156 u_char type; 157 u_char bus_id; 158 char bus_type[6]; 159 } *bus_entry_ptr; 160 161 typedef struct IOAPICENTRY { 162 u_char type; 163 u_char apic_id; 164 u_char apic_version; 165 u_char apic_flags; 166 void *apic_address; 167 } *io_apic_entry_ptr; 168 169 typedef struct INTENTRY { 170 u_char type; 171 u_char int_type; 172 u_short int_flags; 173 u_char src_bus_id; 174 u_char src_bus_irq; 175 u_char dst_apic_id; 176 u_char dst_apic_int; 177 } *int_entry_ptr; 178 179 /* descriptions of MP basetable entries */ 180 typedef struct BASETABLE_ENTRY { 181 u_char type; 182 u_char length; 183 char name[16]; 184 } basetable_entry; 185 186 /* 187 * this code MUST be enabled here and in mpboot.s. 188 * it follows the very early stages of AP boot by placing values in CMOS ram. 189 * it NORMALLY will never be needed and thus the primitive method for enabling. 190 * 191 #define CHECK_POINTS 192 */ 193 194 #if defined(CHECK_POINTS) && !defined(PC98) 195 #define CHECK_READ(A) (outb(CMOS_REG, (A)), inb(CMOS_DATA)) 196 #define CHECK_WRITE(A,D) (outb(CMOS_REG, (A)), outb(CMOS_DATA, (D))) 197 198 #define CHECK_INIT(D); \ 199 CHECK_WRITE(0x34, (D)); \ 200 CHECK_WRITE(0x35, (D)); \ 201 CHECK_WRITE(0x36, (D)); \ 202 CHECK_WRITE(0x37, (D)); \ 203 CHECK_WRITE(0x38, (D)); \ 204 CHECK_WRITE(0x39, (D)); 205 206 #define CHECK_PRINT(S); \ 207 printf("%s: %d, %d, %d, %d, %d, %d\n", \ 208 (S), \ 209 CHECK_READ(0x34), \ 210 CHECK_READ(0x35), \ 211 CHECK_READ(0x36), \ 212 CHECK_READ(0x37), \ 213 CHECK_READ(0x38), \ 214 CHECK_READ(0x39)); 215 216 #else /* CHECK_POINTS */ 217 218 #define CHECK_INIT(D) 219 #define CHECK_PRINT(S) 220 221 #endif /* CHECK_POINTS */ 222 223 /* 224 * Values to send to the POST hardware. 225 */ 226 #define MP_BOOTADDRESS_POST 0x10 227 #define MP_PROBE_POST 0x11 228 #define MPTABLE_PASS1_POST 0x12 229 230 #define MP_START_POST 0x13 231 #define MP_ENABLE_POST 0x14 232 #define MPTABLE_PASS2_POST 0x15 233 234 #define START_ALL_APS_POST 0x16 235 #define INSTALL_AP_TRAMP_POST 0x17 236 #define START_AP_POST 0x18 237 238 #define MP_ANNOUNCE_POST 0x19 239 240 /* used to hold the AP's until we are ready to release them */ 241 struct simplelock ap_boot_lock; 242 243 /** XXX FIXME: where does this really belong, isa.h/isa.c perhaps? */ 244 int current_postcode; 245 246 /** XXX FIXME: what system files declare these??? */ 247 extern struct region_descriptor r_gdt, r_idt; 248 249 int bsp_apic_ready = 0; /* flags useability of BSP apic */ 250 int mp_ncpus; /* # of CPUs, including BSP */ 251 int mp_naps; /* # of Applications processors */ 252 int mp_nbusses; /* # of busses */ 253 int mp_napics; /* # of IO APICs */ 254 int boot_cpu_id; /* designated BSP */ 255 vm_offset_t cpu_apic_address; 256 vm_offset_t io_apic_address[NAPICID]; /* NAPICID is more than enough */ 257 extern int nkpt; 258 259 u_int32_t cpu_apic_versions[MAXCPU]; 260 u_int32_t *io_apic_versions; 261 262 #ifdef APIC_INTR_DIAGNOSTIC 263 int apic_itrace_enter[32]; 264 int apic_itrace_tryisrlock[32]; 265 int apic_itrace_gotisrlock[32]; 266 int apic_itrace_active[32]; 267 int apic_itrace_masked[32]; 268 int apic_itrace_noisrlock[32]; 269 int apic_itrace_masked2[32]; 270 int apic_itrace_unmask[32]; 271 int apic_itrace_noforward[32]; 272 int apic_itrace_leave[32]; 273 int apic_itrace_enter2[32]; 274 int apic_itrace_doreti[32]; 275 int apic_itrace_splz[32]; 276 int apic_itrace_eoi[32]; 277 #ifdef APIC_INTR_DIAGNOSTIC_IRQ 278 unsigned short apic_itrace_debugbuffer[32768]; 279 int apic_itrace_debugbuffer_idx; 280 struct simplelock apic_itrace_debuglock; 281 #endif 282 #endif 283 284 #ifdef APIC_INTR_REORDER 285 struct { 286 volatile int *location; 287 int bit; 288 } apic_isrbit_location[32]; 289 #endif 290 291 struct apic_intmapinfo int_to_apicintpin[APIC_INTMAPSIZE]; 292 293 /* 294 * APIC ID logical/physical mapping structures. 295 * We oversize these to simplify boot-time config. 296 */ 297 int cpu_num_to_apic_id[NAPICID]; 298 int io_num_to_apic_id[NAPICID]; 299 int apic_id_to_logical[NAPICID]; 300 301 302 /* Bitmap of all available CPUs */ 303 u_int all_cpus; 304 305 /* AP uses this during bootstrap. Do not staticize. */ 306 char *bootSTK; 307 static int bootAP; 308 309 /* Hotwire a 0->4MB V==P mapping */ 310 extern pt_entry_t *KPTphys; 311 312 /* SMP page table page */ 313 extern pt_entry_t *SMPpt; 314 315 struct pcb stoppcbs[MAXCPU]; 316 317 int smp_started; /* has the system started? */ 318 319 /* 320 * Local data and functions. 321 */ 322 323 static int mp_capable; 324 static u_int boot_address; 325 static u_int base_memory; 326 327 static int picmode; /* 0: virtual wire mode, 1: PIC mode */ 328 static mpfps_t mpfps; 329 static int search_for_sig(u_int32_t target, int count); 330 static void mp_enable(u_int boot_addr); 331 332 static void mptable_pass1(void); 333 static int mptable_pass2(void); 334 static void default_mp_table(int type); 335 static void fix_mp_table(void); 336 static void setup_apic_irq_mapping(void); 337 static void init_locks(void); 338 static int start_all_aps(u_int boot_addr); 339 static void install_ap_tramp(u_int boot_addr); 340 static int start_ap(int logicalCpu, u_int boot_addr); 341 static int apic_int_is_bus_type(int intr, int bus_type); 342 static void release_aps(void *dummy); 343 344 /* 345 * Calculate usable address in base memory for AP trampoline code. 346 */ 347 u_int 348 mp_bootaddress(u_int basemem) 349 { 350 POSTCODE(MP_BOOTADDRESS_POST); 351 352 base_memory = basemem * 1024; /* convert to bytes */ 353 354 boot_address = base_memory & ~0xfff; /* round down to 4k boundary */ 355 if ((base_memory - boot_address) < bootMP_size) 356 boot_address -= 4096; /* not enough, lower by 4k */ 357 358 return boot_address; 359 } 360 361 362 /* 363 * Look for an Intel MP spec table (ie, SMP capable hardware). 364 */ 365 int 366 mp_probe(void) 367 { 368 int x; 369 u_long segment; 370 u_int32_t target; 371 372 POSTCODE(MP_PROBE_POST); 373 374 /* see if EBDA exists */ 375 if ((segment = (u_long) * (u_short *) (KERNBASE + 0x40e)) != 0) { 376 /* search first 1K of EBDA */ 377 target = (u_int32_t) (segment << 4); 378 if ((x = search_for_sig(target, 1024 / 4)) >= 0) 379 goto found; 380 } else { 381 /* last 1K of base memory, effective 'top of base' passed in */ 382 target = (u_int32_t) (base_memory - 0x400); 383 if ((x = search_for_sig(target, 1024 / 4)) >= 0) 384 goto found; 385 } 386 387 /* search the BIOS */ 388 target = (u_int32_t) BIOS_BASE; 389 if ((x = search_for_sig(target, BIOS_COUNT)) >= 0) 390 goto found; 391 392 /* nothing found */ 393 mpfps = (mpfps_t)0; 394 mp_capable = 0; 395 return 0; 396 397 found: 398 /* calculate needed resources */ 399 mpfps = (mpfps_t)x; 400 mptable_pass1(); 401 402 /* flag fact that we are running multiple processors */ 403 mp_capable = 1; 404 return 1; 405 } 406 407 408 /* 409 * Initialize the SMP hardware and the APIC and start up the AP's. 410 */ 411 void 412 mp_start(void) 413 { 414 POSTCODE(MP_START_POST); 415 416 /* look for MP capable motherboard */ 417 if (mp_capable) 418 mp_enable(boot_address); 419 else 420 panic("MP hardware not found!"); 421 } 422 423 424 /* 425 * Print various information about the SMP system hardware and setup. 426 */ 427 void 428 mp_announce(void) 429 { 430 int x; 431 432 POSTCODE(MP_ANNOUNCE_POST); 433 434 printf("FreeBSD/SMP: Multiprocessor motherboard\n"); 435 printf(" cpu0 (BSP): apic id: %2d", CPU_TO_ID(0)); 436 printf(", version: 0x%08x", cpu_apic_versions[0]); 437 printf(", at 0x%08x\n", cpu_apic_address); 438 for (x = 1; x <= mp_naps; ++x) { 439 printf(" cpu%d (AP): apic id: %2d", x, CPU_TO_ID(x)); 440 printf(", version: 0x%08x", cpu_apic_versions[x]); 441 printf(", at 0x%08x\n", cpu_apic_address); 442 } 443 444 #if defined(APIC_IO) 445 for (x = 0; x < mp_napics; ++x) { 446 printf(" io%d (APIC): apic id: %2d", x, IO_TO_ID(x)); 447 printf(", version: 0x%08x", io_apic_versions[x]); 448 printf(", at 0x%08x\n", io_apic_address[x]); 449 } 450 #else 451 printf(" Warning: APIC I/O disabled\n"); 452 #endif /* APIC_IO */ 453 } 454 455 /* 456 * AP cpu's call this to sync up protected mode. 457 */ 458 void 459 init_secondary(void) 460 { 461 int gsel_tss; 462 int x, myid = bootAP; 463 464 gdt_segs[GPRIV_SEL].ssd_base = (int) &SMP_prvspace[myid]; 465 gdt_segs[GPROC0_SEL].ssd_base = 466 (int) &SMP_prvspace[myid].globaldata.gd_common_tss; 467 SMP_prvspace[myid].globaldata.gd_prvspace = 468 &SMP_prvspace[myid].globaldata; 469 470 for (x = 0; x < NGDT; x++) { 471 ssdtosd(&gdt_segs[x], &gdt[myid * NGDT + x].sd); 472 } 473 474 r_gdt.rd_limit = NGDT * sizeof(gdt[0]) - 1; 475 r_gdt.rd_base = (int) &gdt[myid * NGDT]; 476 lgdt(&r_gdt); /* does magic intra-segment return */ 477 478 lidt(&r_idt); 479 480 lldt(_default_ldt); 481 #ifdef USER_LDT 482 PCPU_SET(currentldt, _default_ldt); 483 #endif 484 485 gsel_tss = GSEL(GPROC0_SEL, SEL_KPL); 486 gdt[myid * NGDT + GPROC0_SEL].sd.sd_type = SDT_SYS386TSS; 487 PCPU_SET(common_tss.tss_esp0, 0); /* not used until after switch */ 488 PCPU_SET(common_tss.tss_ss0, GSEL(GDATA_SEL, SEL_KPL)); 489 PCPU_SET(common_tss.tss_ioopt, (sizeof (struct i386tss)) << 16); 490 PCPU_SET(tss_gdt, &gdt[myid * NGDT + GPROC0_SEL].sd); 491 PCPU_SET(common_tssd, *PCPU_GET(tss_gdt)); 492 ltr(gsel_tss); 493 494 pmap_set_opt(); 495 } 496 497 498 #if defined(APIC_IO) 499 /* 500 * Final configuration of the BSP's local APIC: 501 * - disable 'pic mode'. 502 * - disable 'virtual wire mode'. 503 * - enable NMI. 504 */ 505 void 506 bsp_apic_configure(void) 507 { 508 u_char byte; 509 u_int32_t temp; 510 511 /* leave 'pic mode' if necessary */ 512 if (picmode) { 513 outb(0x22, 0x70); /* select IMCR */ 514 byte = inb(0x23); /* current contents */ 515 byte |= 0x01; /* mask external INTR */ 516 outb(0x23, byte); /* disconnect 8259s/NMI */ 517 } 518 519 /* mask lint0 (the 8259 'virtual wire' connection) */ 520 temp = lapic.lvt_lint0; 521 temp |= APIC_LVT_M; /* set the mask */ 522 lapic.lvt_lint0 = temp; 523 524 /* setup lint1 to handle NMI */ 525 temp = lapic.lvt_lint1; 526 temp &= ~APIC_LVT_M; /* clear the mask */ 527 lapic.lvt_lint1 = temp; 528 529 if (bootverbose) 530 apic_dump("bsp_apic_configure()"); 531 } 532 #endif /* APIC_IO */ 533 534 535 /******************************************************************* 536 * local functions and data 537 */ 538 539 /* 540 * start the SMP system 541 */ 542 static void 543 mp_enable(u_int boot_addr) 544 { 545 int x; 546 #if defined(APIC_IO) 547 int apic; 548 u_int ux; 549 #endif /* APIC_IO */ 550 551 POSTCODE(MP_ENABLE_POST); 552 553 /* turn on 4MB of V == P addressing so we can get to MP table */ 554 *(int *)PTD = PG_V | PG_RW | ((uintptr_t)(void *)KPTphys & PG_FRAME); 555 invltlb(); 556 557 /* examine the MP table for needed info, uses physical addresses */ 558 x = mptable_pass2(); 559 560 *(int *)PTD = 0; 561 invltlb(); 562 563 /* can't process default configs till the CPU APIC is pmapped */ 564 if (x) 565 default_mp_table(x); 566 567 /* post scan cleanup */ 568 fix_mp_table(); 569 setup_apic_irq_mapping(); 570 571 #if defined(APIC_IO) 572 573 /* fill the LOGICAL io_apic_versions table */ 574 for (apic = 0; apic < mp_napics; ++apic) { 575 ux = io_apic_read(apic, IOAPIC_VER); 576 io_apic_versions[apic] = ux; 577 io_apic_set_id(apic, IO_TO_ID(apic)); 578 } 579 580 /* program each IO APIC in the system */ 581 for (apic = 0; apic < mp_napics; ++apic) 582 if (io_apic_setup(apic) < 0) 583 panic("IO APIC setup failure"); 584 585 /* install a 'Spurious INTerrupt' vector */ 586 setidt(XSPURIOUSINT_OFFSET, Xspuriousint, 587 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 588 589 /* install an inter-CPU IPI for TLB invalidation */ 590 setidt(XINVLTLB_OFFSET, Xinvltlb, 591 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 592 593 #ifdef BETTER_CLOCK 594 /* install an inter-CPU IPI for reading processor state */ 595 setidt(XCPUCHECKSTATE_OFFSET, Xcpucheckstate, 596 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 597 #endif 598 599 /* install an inter-CPU IPI for all-CPU rendezvous */ 600 setidt(XRENDEZVOUS_OFFSET, Xrendezvous, 601 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 602 603 /* install an inter-CPU IPI for forcing an additional software trap */ 604 setidt(XCPUAST_OFFSET, Xcpuast, 605 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 606 607 /* install an inter-CPU IPI for interrupt forwarding */ 608 setidt(XFORWARD_IRQ_OFFSET, Xforward_irq, 609 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 610 611 /* install an inter-CPU IPI for CPU stop/restart */ 612 setidt(XCPUSTOP_OFFSET, Xcpustop, 613 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 614 615 #if defined(TEST_TEST1) 616 /* install a "fake hardware INTerrupt" vector */ 617 setidt(XTEST1_OFFSET, Xtest1, 618 SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 619 #endif /** TEST_TEST1 */ 620 621 #endif /* APIC_IO */ 622 623 /* initialize all SMP locks */ 624 init_locks(); 625 626 /* obtain the ap_boot_lock */ 627 s_lock(&ap_boot_lock); 628 629 /* start each Application Processor */ 630 start_all_aps(boot_addr); 631 } 632 633 634 /* 635 * look for the MP spec signature 636 */ 637 638 /* string defined by the Intel MP Spec as identifying the MP table */ 639 #define MP_SIG 0x5f504d5f /* _MP_ */ 640 #define NEXT(X) ((X) += 4) 641 static int 642 search_for_sig(u_int32_t target, int count) 643 { 644 int x; 645 u_int32_t *addr = (u_int32_t *) (KERNBASE + target); 646 647 for (x = 0; x < count; NEXT(x)) 648 if (addr[x] == MP_SIG) 649 /* make array index a byte index */ 650 return (target + (x * sizeof(u_int32_t))); 651 652 return -1; 653 } 654 655 656 static basetable_entry basetable_entry_types[] = 657 { 658 {0, 20, "Processor"}, 659 {1, 8, "Bus"}, 660 {2, 8, "I/O APIC"}, 661 {3, 8, "I/O INT"}, 662 {4, 8, "Local INT"} 663 }; 664 665 typedef struct BUSDATA { 666 u_char bus_id; 667 enum busTypes bus_type; 668 } bus_datum; 669 670 typedef struct INTDATA { 671 u_char int_type; 672 u_short int_flags; 673 u_char src_bus_id; 674 u_char src_bus_irq; 675 u_char dst_apic_id; 676 u_char dst_apic_int; 677 u_char int_vector; 678 } io_int, local_int; 679 680 typedef struct BUSTYPENAME { 681 u_char type; 682 char name[7]; 683 } bus_type_name; 684 685 static bus_type_name bus_type_table[] = 686 { 687 {CBUS, "CBUS"}, 688 {CBUSII, "CBUSII"}, 689 {EISA, "EISA"}, 690 {MCA, "MCA"}, 691 {UNKNOWN_BUSTYPE, "---"}, 692 {ISA, "ISA"}, 693 {MCA, "MCA"}, 694 {UNKNOWN_BUSTYPE, "---"}, 695 {UNKNOWN_BUSTYPE, "---"}, 696 {UNKNOWN_BUSTYPE, "---"}, 697 {UNKNOWN_BUSTYPE, "---"}, 698 {UNKNOWN_BUSTYPE, "---"}, 699 {PCI, "PCI"}, 700 {UNKNOWN_BUSTYPE, "---"}, 701 {UNKNOWN_BUSTYPE, "---"}, 702 {UNKNOWN_BUSTYPE, "---"}, 703 {UNKNOWN_BUSTYPE, "---"}, 704 {XPRESS, "XPRESS"}, 705 {UNKNOWN_BUSTYPE, "---"} 706 }; 707 /* from MP spec v1.4, table 5-1 */ 708 static int default_data[7][5] = 709 { 710 /* nbus, id0, type0, id1, type1 */ 711 {1, 0, ISA, 255, 255}, 712 {1, 0, EISA, 255, 255}, 713 {1, 0, EISA, 255, 255}, 714 {1, 0, MCA, 255, 255}, 715 {2, 0, ISA, 1, PCI}, 716 {2, 0, EISA, 1, PCI}, 717 {2, 0, MCA, 1, PCI} 718 }; 719 720 721 /* the bus data */ 722 static bus_datum *bus_data; 723 724 /* the IO INT data, one entry per possible APIC INTerrupt */ 725 static io_int *io_apic_ints; 726 727 static int nintrs; 728 729 static int processor_entry __P((proc_entry_ptr entry, int cpu)); 730 static int bus_entry __P((bus_entry_ptr entry, int bus)); 731 static int io_apic_entry __P((io_apic_entry_ptr entry, int apic)); 732 static int int_entry __P((int_entry_ptr entry, int intr)); 733 static int lookup_bus_type __P((char *name)); 734 735 736 /* 737 * 1st pass on motherboard's Intel MP specification table. 738 * 739 * initializes: 740 * mp_ncpus = 1 741 * 742 * determines: 743 * cpu_apic_address (common to all CPUs) 744 * io_apic_address[N] 745 * mp_naps 746 * mp_nbusses 747 * mp_napics 748 * nintrs 749 */ 750 static void 751 mptable_pass1(void) 752 { 753 int x; 754 mpcth_t cth; 755 int totalSize; 756 void* position; 757 int count; 758 int type; 759 760 POSTCODE(MPTABLE_PASS1_POST); 761 762 /* clear various tables */ 763 for (x = 0; x < NAPICID; ++x) { 764 io_apic_address[x] = ~0; /* IO APIC address table */ 765 } 766 767 /* init everything to empty */ 768 mp_naps = 0; 769 mp_nbusses = 0; 770 mp_napics = 0; 771 nintrs = 0; 772 773 /* check for use of 'default' configuration */ 774 if (MPFPS_MPFB1 != 0) { 775 /* use default addresses */ 776 cpu_apic_address = DEFAULT_APIC_BASE; 777 io_apic_address[0] = DEFAULT_IO_APIC_BASE; 778 779 /* fill in with defaults */ 780 mp_naps = 2; /* includes BSP */ 781 mp_nbusses = default_data[MPFPS_MPFB1 - 1][0]; 782 #if defined(APIC_IO) 783 mp_napics = 1; 784 nintrs = 16; 785 #endif /* APIC_IO */ 786 } 787 else { 788 if ((cth = mpfps->pap) == 0) 789 panic("MP Configuration Table Header MISSING!"); 790 791 cpu_apic_address = (vm_offset_t) cth->apic_address; 792 793 /* walk the table, recording info of interest */ 794 totalSize = cth->base_table_length - sizeof(struct MPCTH); 795 position = (u_char *) cth + sizeof(struct MPCTH); 796 count = cth->entry_count; 797 798 while (count--) { 799 switch (type = *(u_char *) position) { 800 case 0: /* processor_entry */ 801 if (((proc_entry_ptr)position)->cpu_flags 802 & PROCENTRY_FLAG_EN) 803 ++mp_naps; 804 break; 805 case 1: /* bus_entry */ 806 ++mp_nbusses; 807 break; 808 case 2: /* io_apic_entry */ 809 if (((io_apic_entry_ptr)position)->apic_flags 810 & IOAPICENTRY_FLAG_EN) 811 io_apic_address[mp_napics++] = 812 (vm_offset_t)((io_apic_entry_ptr) 813 position)->apic_address; 814 break; 815 case 3: /* int_entry */ 816 ++nintrs; 817 break; 818 case 4: /* int_entry */ 819 break; 820 default: 821 panic("mpfps Base Table HOSED!"); 822 /* NOTREACHED */ 823 } 824 825 totalSize -= basetable_entry_types[type].length; 826 (u_char*)position += basetable_entry_types[type].length; 827 } 828 } 829 830 /* qualify the numbers */ 831 if (mp_naps > MAXCPU) { 832 printf("Warning: only using %d of %d available CPUs!\n", 833 MAXCPU, mp_naps); 834 mp_naps = MAXCPU; 835 } 836 837 /* 838 * Count the BSP. 839 * This is also used as a counter while starting the APs. 840 */ 841 mp_ncpus = 1; 842 843 --mp_naps; /* subtract the BSP */ 844 } 845 846 847 /* 848 * 2nd pass on motherboard's Intel MP specification table. 849 * 850 * sets: 851 * boot_cpu_id 852 * ID_TO_IO(N), phy APIC ID to log CPU/IO table 853 * CPU_TO_ID(N), logical CPU to APIC ID table 854 * IO_TO_ID(N), logical IO to APIC ID table 855 * bus_data[N] 856 * io_apic_ints[N] 857 */ 858 static int 859 mptable_pass2(void) 860 { 861 int x; 862 mpcth_t cth; 863 int totalSize; 864 void* position; 865 int count; 866 int type; 867 int apic, bus, cpu, intr; 868 int i, j; 869 int pgeflag; 870 871 POSTCODE(MPTABLE_PASS2_POST); 872 873 pgeflag = 0; /* XXX - Not used under SMP yet. */ 874 875 MALLOC(io_apic_versions, u_int32_t *, sizeof(u_int32_t) * mp_napics, 876 M_DEVBUF, M_WAITOK); 877 MALLOC(ioapic, volatile ioapic_t **, sizeof(ioapic_t *) * mp_napics, 878 M_DEVBUF, M_WAITOK); 879 MALLOC(io_apic_ints, io_int *, sizeof(io_int) * (nintrs + 1), 880 M_DEVBUF, M_WAITOK); 881 MALLOC(bus_data, bus_datum *, sizeof(bus_datum) * mp_nbusses, 882 M_DEVBUF, M_WAITOK); 883 884 bzero(ioapic, sizeof(ioapic_t *) * mp_napics); 885 886 for (i = 0; i < mp_napics; i++) { 887 for (j = 0; j < mp_napics; j++) { 888 /* same page frame as a previous IO apic? */ 889 if (((vm_offset_t)SMPpt[NPTEPG-2-j] & PG_FRAME) == 890 (io_apic_address[i] & PG_FRAME)) { 891 ioapic[i] = (ioapic_t *)((u_int)SMP_prvspace 892 + (NPTEPG-2-j) * PAGE_SIZE 893 + (io_apic_address[i] & PAGE_MASK)); 894 break; 895 } 896 /* use this slot if available */ 897 if (((vm_offset_t)SMPpt[NPTEPG-2-j] & PG_FRAME) == 0) { 898 SMPpt[NPTEPG-2-j] = (pt_entry_t)(PG_V | PG_RW | 899 pgeflag | (io_apic_address[i] & PG_FRAME)); 900 ioapic[i] = (ioapic_t *)((u_int)SMP_prvspace 901 + (NPTEPG-2-j) * PAGE_SIZE 902 + (io_apic_address[i] & PAGE_MASK)); 903 break; 904 } 905 } 906 } 907 908 /* clear various tables */ 909 for (x = 0; x < NAPICID; ++x) { 910 ID_TO_IO(x) = -1; /* phy APIC ID to log CPU/IO table */ 911 CPU_TO_ID(x) = -1; /* logical CPU to APIC ID table */ 912 IO_TO_ID(x) = -1; /* logical IO to APIC ID table */ 913 } 914 915 /* clear bus data table */ 916 for (x = 0; x < mp_nbusses; ++x) 917 bus_data[x].bus_id = 0xff; 918 919 /* clear IO APIC INT table */ 920 for (x = 0; x < (nintrs + 1); ++x) { 921 io_apic_ints[x].int_type = 0xff; 922 io_apic_ints[x].int_vector = 0xff; 923 } 924 925 /* setup the cpu/apic mapping arrays */ 926 boot_cpu_id = -1; 927 928 /* record whether PIC or virtual-wire mode */ 929 picmode = (mpfps->mpfb2 & 0x80) ? 1 : 0; 930 931 /* check for use of 'default' configuration */ 932 if (MPFPS_MPFB1 != 0) 933 return MPFPS_MPFB1; /* return default configuration type */ 934 935 if ((cth = mpfps->pap) == 0) 936 panic("MP Configuration Table Header MISSING!"); 937 938 /* walk the table, recording info of interest */ 939 totalSize = cth->base_table_length - sizeof(struct MPCTH); 940 position = (u_char *) cth + sizeof(struct MPCTH); 941 count = cth->entry_count; 942 apic = bus = intr = 0; 943 cpu = 1; /* pre-count the BSP */ 944 945 while (count--) { 946 switch (type = *(u_char *) position) { 947 case 0: 948 if (processor_entry(position, cpu)) 949 ++cpu; 950 break; 951 case 1: 952 if (bus_entry(position, bus)) 953 ++bus; 954 break; 955 case 2: 956 if (io_apic_entry(position, apic)) 957 ++apic; 958 break; 959 case 3: 960 if (int_entry(position, intr)) 961 ++intr; 962 break; 963 case 4: 964 /* int_entry(position); */ 965 break; 966 default: 967 panic("mpfps Base Table HOSED!"); 968 /* NOTREACHED */ 969 } 970 971 totalSize -= basetable_entry_types[type].length; 972 (u_char *) position += basetable_entry_types[type].length; 973 } 974 975 if (boot_cpu_id == -1) 976 panic("NO BSP found!"); 977 978 /* report fact that its NOT a default configuration */ 979 return 0; 980 } 981 982 983 void 984 assign_apic_irq(int apic, int intpin, int irq) 985 { 986 int x; 987 988 if (int_to_apicintpin[irq].ioapic != -1) 989 panic("assign_apic_irq: inconsistent table"); 990 991 int_to_apicintpin[irq].ioapic = apic; 992 int_to_apicintpin[irq].int_pin = intpin; 993 int_to_apicintpin[irq].apic_address = ioapic[apic]; 994 int_to_apicintpin[irq].redirindex = IOAPIC_REDTBL + 2 * intpin; 995 996 for (x = 0; x < nintrs; x++) { 997 if ((io_apic_ints[x].int_type == 0 || 998 io_apic_ints[x].int_type == 3) && 999 io_apic_ints[x].int_vector == 0xff && 1000 io_apic_ints[x].dst_apic_id == IO_TO_ID(apic) && 1001 io_apic_ints[x].dst_apic_int == intpin) 1002 io_apic_ints[x].int_vector = irq; 1003 } 1004 } 1005 1006 void 1007 revoke_apic_irq(int irq) 1008 { 1009 int x; 1010 int oldapic; 1011 int oldintpin; 1012 1013 if (int_to_apicintpin[irq].ioapic == -1) 1014 panic("assign_apic_irq: inconsistent table"); 1015 1016 oldapic = int_to_apicintpin[irq].ioapic; 1017 oldintpin = int_to_apicintpin[irq].int_pin; 1018 1019 int_to_apicintpin[irq].ioapic = -1; 1020 int_to_apicintpin[irq].int_pin = 0; 1021 int_to_apicintpin[irq].apic_address = NULL; 1022 int_to_apicintpin[irq].redirindex = 0; 1023 1024 for (x = 0; x < nintrs; x++) { 1025 if ((io_apic_ints[x].int_type == 0 || 1026 io_apic_ints[x].int_type == 3) && 1027 io_apic_ints[x].int_vector == 0xff && 1028 io_apic_ints[x].dst_apic_id == IO_TO_ID(oldapic) && 1029 io_apic_ints[x].dst_apic_int == oldintpin) 1030 io_apic_ints[x].int_vector = 0xff; 1031 } 1032 } 1033 1034 1035 1036 static void 1037 swap_apic_id(int apic, int oldid, int newid) 1038 { 1039 int x; 1040 int oapic; 1041 1042 1043 if (oldid == newid) 1044 return; /* Nothing to do */ 1045 1046 printf("Changing APIC ID for IO APIC #%d from %d to %d in MP table\n", 1047 apic, oldid, newid); 1048 1049 /* Swap physical APIC IDs in interrupt entries */ 1050 for (x = 0; x < nintrs; x++) { 1051 if (io_apic_ints[x].dst_apic_id == oldid) 1052 io_apic_ints[x].dst_apic_id = newid; 1053 else if (io_apic_ints[x].dst_apic_id == newid) 1054 io_apic_ints[x].dst_apic_id = oldid; 1055 } 1056 1057 /* Swap physical APIC IDs in IO_TO_ID mappings */ 1058 for (oapic = 0; oapic < mp_napics; oapic++) 1059 if (IO_TO_ID(oapic) == newid) 1060 break; 1061 1062 if (oapic < mp_napics) { 1063 printf("Changing APIC ID for IO APIC #%d from " 1064 "%d to %d in MP table\n", 1065 oapic, newid, oldid); 1066 IO_TO_ID(oapic) = oldid; 1067 } 1068 IO_TO_ID(apic) = newid; 1069 } 1070 1071 1072 static void 1073 fix_id_to_io_mapping(void) 1074 { 1075 int x; 1076 1077 for (x = 0; x < NAPICID; x++) 1078 ID_TO_IO(x) = -1; 1079 1080 for (x = 0; x <= mp_naps; x++) 1081 if (CPU_TO_ID(x) < NAPICID) 1082 ID_TO_IO(CPU_TO_ID(x)) = x; 1083 1084 for (x = 0; x < mp_napics; x++) 1085 if (IO_TO_ID(x) < NAPICID) 1086 ID_TO_IO(IO_TO_ID(x)) = x; 1087 } 1088 1089 1090 static int 1091 first_free_apic_id(void) 1092 { 1093 int freeid, x; 1094 1095 for (freeid = 0; freeid < NAPICID; freeid++) { 1096 for (x = 0; x <= mp_naps; x++) 1097 if (CPU_TO_ID(x) == freeid) 1098 break; 1099 if (x <= mp_naps) 1100 continue; 1101 for (x = 0; x < mp_napics; x++) 1102 if (IO_TO_ID(x) == freeid) 1103 break; 1104 if (x < mp_napics) 1105 continue; 1106 return freeid; 1107 } 1108 return freeid; 1109 } 1110 1111 1112 static int 1113 io_apic_id_acceptable(int apic, int id) 1114 { 1115 int cpu; /* Logical CPU number */ 1116 int oapic; /* Logical IO APIC number for other IO APIC */ 1117 1118 if (id >= NAPICID) 1119 return 0; /* Out of range */ 1120 1121 for (cpu = 0; cpu <= mp_naps; cpu++) 1122 if (CPU_TO_ID(cpu) == id) 1123 return 0; /* Conflict with CPU */ 1124 1125 for (oapic = 0; oapic < mp_napics && oapic < apic; oapic++) 1126 if (IO_TO_ID(oapic) == id) 1127 return 0; /* Conflict with other APIC */ 1128 1129 return 1; /* ID is acceptable for IO APIC */ 1130 } 1131 1132 1133 /* 1134 * parse an Intel MP specification table 1135 */ 1136 static void 1137 fix_mp_table(void) 1138 { 1139 int x; 1140 int id; 1141 int bus_0 = 0; /* Stop GCC warning */ 1142 int bus_pci = 0; /* Stop GCC warning */ 1143 int num_pci_bus; 1144 int apic; /* IO APIC unit number */ 1145 int freeid; /* Free physical APIC ID */ 1146 int physid; /* Current physical IO APIC ID */ 1147 1148 /* 1149 * Fix mis-numbering of the PCI bus and its INT entries if the BIOS 1150 * did it wrong. The MP spec says that when more than 1 PCI bus 1151 * exists the BIOS must begin with bus entries for the PCI bus and use 1152 * actual PCI bus numbering. This implies that when only 1 PCI bus 1153 * exists the BIOS can choose to ignore this ordering, and indeed many 1154 * MP motherboards do ignore it. This causes a problem when the PCI 1155 * sub-system makes requests of the MP sub-system based on PCI bus 1156 * numbers. So here we look for the situation and renumber the 1157 * busses and associated INTs in an effort to "make it right". 1158 */ 1159 1160 /* find bus 0, PCI bus, count the number of PCI busses */ 1161 for (num_pci_bus = 0, x = 0; x < mp_nbusses; ++x) { 1162 if (bus_data[x].bus_id == 0) { 1163 bus_0 = x; 1164 } 1165 if (bus_data[x].bus_type == PCI) { 1166 ++num_pci_bus; 1167 bus_pci = x; 1168 } 1169 } 1170 /* 1171 * bus_0 == slot of bus with ID of 0 1172 * bus_pci == slot of last PCI bus encountered 1173 */ 1174 1175 /* check the 1 PCI bus case for sanity */ 1176 /* if it is number 0 all is well */ 1177 if (num_pci_bus == 1 && 1178 bus_data[bus_pci].bus_id != 0) { 1179 1180 /* mis-numbered, swap with whichever bus uses slot 0 */ 1181 1182 /* swap the bus entry types */ 1183 bus_data[bus_pci].bus_type = bus_data[bus_0].bus_type; 1184 bus_data[bus_0].bus_type = PCI; 1185 1186 /* swap each relavant INTerrupt entry */ 1187 id = bus_data[bus_pci].bus_id; 1188 for (x = 0; x < nintrs; ++x) { 1189 if (io_apic_ints[x].src_bus_id == id) { 1190 io_apic_ints[x].src_bus_id = 0; 1191 } 1192 else if (io_apic_ints[x].src_bus_id == 0) { 1193 io_apic_ints[x].src_bus_id = id; 1194 } 1195 } 1196 } 1197 1198 /* Assign IO APIC IDs. 1199 * 1200 * First try the existing ID. If a conflict is detected, try 1201 * the ID in the MP table. If a conflict is still detected, find 1202 * a free id. 1203 * 1204 * We cannot use the ID_TO_IO table before all conflicts has been 1205 * resolved and the table has been corrected. 1206 */ 1207 for (apic = 0; apic < mp_napics; ++apic) { /* For all IO APICs */ 1208 1209 /* First try to use the value set by the BIOS */ 1210 physid = io_apic_get_id(apic); 1211 if (io_apic_id_acceptable(apic, physid)) { 1212 if (IO_TO_ID(apic) != physid) 1213 swap_apic_id(apic, IO_TO_ID(apic), physid); 1214 continue; 1215 } 1216 1217 /* Then check if the value in the MP table is acceptable */ 1218 if (io_apic_id_acceptable(apic, IO_TO_ID(apic))) 1219 continue; 1220 1221 /* Last resort, find a free APIC ID and use it */ 1222 freeid = first_free_apic_id(); 1223 if (freeid >= NAPICID) 1224 panic("No free physical APIC IDs found"); 1225 1226 if (io_apic_id_acceptable(apic, freeid)) { 1227 swap_apic_id(apic, IO_TO_ID(apic), freeid); 1228 continue; 1229 } 1230 panic("Free physical APIC ID not usable"); 1231 } 1232 fix_id_to_io_mapping(); 1233 1234 /* detect and fix broken Compaq MP table */ 1235 if (apic_int_type(0, 0) == -1) { 1236 printf("APIC_IO: MP table broken: 8259->APIC entry missing!\n"); 1237 io_apic_ints[nintrs].int_type = 3; /* ExtInt */ 1238 io_apic_ints[nintrs].int_vector = 0xff; /* Unassigned */ 1239 /* XXX fixme, set src bus id etc, but it doesn't seem to hurt */ 1240 io_apic_ints[nintrs].dst_apic_id = IO_TO_ID(0); 1241 io_apic_ints[nintrs].dst_apic_int = 0; /* Pin 0 */ 1242 nintrs++; 1243 } 1244 } 1245 1246 1247 /* Assign low level interrupt handlers */ 1248 static void 1249 setup_apic_irq_mapping(void) 1250 { 1251 int x; 1252 int int_vector; 1253 1254 /* Clear array */ 1255 for (x = 0; x < APIC_INTMAPSIZE; x++) { 1256 int_to_apicintpin[x].ioapic = -1; 1257 int_to_apicintpin[x].int_pin = 0; 1258 int_to_apicintpin[x].apic_address = NULL; 1259 int_to_apicintpin[x].redirindex = 0; 1260 } 1261 1262 /* First assign ISA/EISA interrupts */ 1263 for (x = 0; x < nintrs; x++) { 1264 int_vector = io_apic_ints[x].src_bus_irq; 1265 if (int_vector < APIC_INTMAPSIZE && 1266 io_apic_ints[x].int_vector == 0xff && 1267 int_to_apicintpin[int_vector].ioapic == -1 && 1268 (apic_int_is_bus_type(x, ISA) || 1269 apic_int_is_bus_type(x, EISA)) && 1270 io_apic_ints[x].int_type == 0) { 1271 assign_apic_irq(ID_TO_IO(io_apic_ints[x].dst_apic_id), 1272 io_apic_ints[x].dst_apic_int, 1273 int_vector); 1274 } 1275 } 1276 1277 /* Assign first set of interrupts to intpins on IOAPIC #0 */ 1278 for (x = 0; x < nintrs; x++) { 1279 int_vector = io_apic_ints[x].dst_apic_int; 1280 if (int_vector < APIC_INTMAPSIZE && 1281 io_apic_ints[x].dst_apic_id == IO_TO_ID(0) && 1282 io_apic_ints[x].int_vector == 0xff && 1283 int_to_apicintpin[int_vector].ioapic == -1 && 1284 (io_apic_ints[x].int_type == 0 || 1285 io_apic_ints[x].int_type == 3)) { 1286 assign_apic_irq(0, 1287 io_apic_ints[x].dst_apic_int, 1288 int_vector); 1289 } 1290 } 1291 /* 1292 * Assign interrupts for remaining intpins. 1293 * Skip IOAPIC #0 intpin 0 if the type is ExtInt, since this indicates 1294 * that an entry for ISA/EISA irq 0 exist, and a fallback to mixed mode 1295 * due to 8254 interrupts not being delivered can reuse that low level 1296 * interrupt handler. 1297 */ 1298 int_vector = 0; 1299 while (int_vector < APIC_INTMAPSIZE && 1300 int_to_apicintpin[int_vector].ioapic != -1) 1301 int_vector++; 1302 for (x = 0; x < nintrs && int_vector < APIC_INTMAPSIZE; x++) { 1303 if ((io_apic_ints[x].int_type == 0 || 1304 (io_apic_ints[x].int_type == 3 && 1305 (io_apic_ints[x].dst_apic_id != IO_TO_ID(0) || 1306 io_apic_ints[x].dst_apic_int != 0))) && 1307 io_apic_ints[x].int_vector == 0xff) { 1308 assign_apic_irq(ID_TO_IO(io_apic_ints[x].dst_apic_id), 1309 io_apic_ints[x].dst_apic_int, 1310 int_vector); 1311 int_vector++; 1312 while (int_vector < APIC_INTMAPSIZE && 1313 int_to_apicintpin[int_vector].ioapic != -1) 1314 int_vector++; 1315 } 1316 } 1317 } 1318 1319 1320 static int 1321 processor_entry(proc_entry_ptr entry, int cpu) 1322 { 1323 /* check for usability */ 1324 if (!(entry->cpu_flags & PROCENTRY_FLAG_EN)) 1325 return 0; 1326 1327 if(entry->apic_id >= NAPICID) 1328 panic("CPU APIC ID out of range (0..%d)", NAPICID - 1); 1329 /* check for BSP flag */ 1330 if (entry->cpu_flags & PROCENTRY_FLAG_BP) { 1331 boot_cpu_id = entry->apic_id; 1332 CPU_TO_ID(0) = entry->apic_id; 1333 ID_TO_CPU(entry->apic_id) = 0; 1334 return 0; /* its already been counted */ 1335 } 1336 1337 /* add another AP to list, if less than max number of CPUs */ 1338 else if (cpu < MAXCPU) { 1339 CPU_TO_ID(cpu) = entry->apic_id; 1340 ID_TO_CPU(entry->apic_id) = cpu; 1341 return 1; 1342 } 1343 1344 return 0; 1345 } 1346 1347 1348 static int 1349 bus_entry(bus_entry_ptr entry, int bus) 1350 { 1351 int x; 1352 char c, name[8]; 1353 1354 /* encode the name into an index */ 1355 for (x = 0; x < 6; ++x) { 1356 if ((c = entry->bus_type[x]) == ' ') 1357 break; 1358 name[x] = c; 1359 } 1360 name[x] = '\0'; 1361 1362 if ((x = lookup_bus_type(name)) == UNKNOWN_BUSTYPE) 1363 panic("unknown bus type: '%s'", name); 1364 1365 bus_data[bus].bus_id = entry->bus_id; 1366 bus_data[bus].bus_type = x; 1367 1368 return 1; 1369 } 1370 1371 1372 static int 1373 io_apic_entry(io_apic_entry_ptr entry, int apic) 1374 { 1375 if (!(entry->apic_flags & IOAPICENTRY_FLAG_EN)) 1376 return 0; 1377 1378 IO_TO_ID(apic) = entry->apic_id; 1379 if (entry->apic_id < NAPICID) 1380 ID_TO_IO(entry->apic_id) = apic; 1381 1382 return 1; 1383 } 1384 1385 1386 static int 1387 lookup_bus_type(char *name) 1388 { 1389 int x; 1390 1391 for (x = 0; x < MAX_BUSTYPE; ++x) 1392 if (strcmp(bus_type_table[x].name, name) == 0) 1393 return bus_type_table[x].type; 1394 1395 return UNKNOWN_BUSTYPE; 1396 } 1397 1398 1399 static int 1400 int_entry(int_entry_ptr entry, int intr) 1401 { 1402 int apic; 1403 1404 io_apic_ints[intr].int_type = entry->int_type; 1405 io_apic_ints[intr].int_flags = entry->int_flags; 1406 io_apic_ints[intr].src_bus_id = entry->src_bus_id; 1407 io_apic_ints[intr].src_bus_irq = entry->src_bus_irq; 1408 if (entry->dst_apic_id == 255) { 1409 /* This signal goes to all IO APICS. Select an IO APIC 1410 with sufficient number of interrupt pins */ 1411 for (apic = 0; apic < mp_napics; apic++) 1412 if (((io_apic_read(apic, IOAPIC_VER) & 1413 IOART_VER_MAXREDIR) >> MAXREDIRSHIFT) >= 1414 entry->dst_apic_int) 1415 break; 1416 if (apic < mp_napics) 1417 io_apic_ints[intr].dst_apic_id = IO_TO_ID(apic); 1418 else 1419 io_apic_ints[intr].dst_apic_id = entry->dst_apic_id; 1420 } else 1421 io_apic_ints[intr].dst_apic_id = entry->dst_apic_id; 1422 io_apic_ints[intr].dst_apic_int = entry->dst_apic_int; 1423 1424 return 1; 1425 } 1426 1427 1428 static int 1429 apic_int_is_bus_type(int intr, int bus_type) 1430 { 1431 int bus; 1432 1433 for (bus = 0; bus < mp_nbusses; ++bus) 1434 if ((bus_data[bus].bus_id == io_apic_ints[intr].src_bus_id) 1435 && ((int) bus_data[bus].bus_type == bus_type)) 1436 return 1; 1437 1438 return 0; 1439 } 1440 1441 1442 /* 1443 * Given a traditional ISA INT mask, return an APIC mask. 1444 */ 1445 u_int 1446 isa_apic_mask(u_int isa_mask) 1447 { 1448 int isa_irq; 1449 int apic_pin; 1450 1451 #if defined(SKIP_IRQ15_REDIRECT) 1452 if (isa_mask == (1 << 15)) { 1453 printf("skipping ISA IRQ15 redirect\n"); 1454 return isa_mask; 1455 } 1456 #endif /* SKIP_IRQ15_REDIRECT */ 1457 1458 isa_irq = ffs(isa_mask); /* find its bit position */ 1459 if (isa_irq == 0) /* doesn't exist */ 1460 return 0; 1461 --isa_irq; /* make it zero based */ 1462 1463 apic_pin = isa_apic_irq(isa_irq); /* look for APIC connection */ 1464 if (apic_pin == -1) 1465 return 0; 1466 1467 return (1 << apic_pin); /* convert pin# to a mask */ 1468 } 1469 1470 1471 /* 1472 * Determine which APIC pin an ISA/EISA INT is attached to. 1473 */ 1474 #define INTTYPE(I) (io_apic_ints[(I)].int_type) 1475 #define INTPIN(I) (io_apic_ints[(I)].dst_apic_int) 1476 #define INTIRQ(I) (io_apic_ints[(I)].int_vector) 1477 #define INTAPIC(I) (ID_TO_IO(io_apic_ints[(I)].dst_apic_id)) 1478 1479 #define SRCBUSIRQ(I) (io_apic_ints[(I)].src_bus_irq) 1480 int 1481 isa_apic_irq(int isa_irq) 1482 { 1483 int intr; 1484 1485 for (intr = 0; intr < nintrs; ++intr) { /* check each record */ 1486 if (INTTYPE(intr) == 0) { /* standard INT */ 1487 if (SRCBUSIRQ(intr) == isa_irq) { 1488 if (apic_int_is_bus_type(intr, ISA) || 1489 apic_int_is_bus_type(intr, EISA)) 1490 return INTIRQ(intr); /* found */ 1491 } 1492 } 1493 } 1494 return -1; /* NOT found */ 1495 } 1496 1497 1498 /* 1499 * Determine which APIC pin a PCI INT is attached to. 1500 */ 1501 #define SRCBUSID(I) (io_apic_ints[(I)].src_bus_id) 1502 #define SRCBUSDEVICE(I) ((io_apic_ints[(I)].src_bus_irq >> 2) & 0x1f) 1503 #define SRCBUSLINE(I) (io_apic_ints[(I)].src_bus_irq & 0x03) 1504 int 1505 pci_apic_irq(int pciBus, int pciDevice, int pciInt) 1506 { 1507 int intr; 1508 1509 --pciInt; /* zero based */ 1510 1511 for (intr = 0; intr < nintrs; ++intr) /* check each record */ 1512 if ((INTTYPE(intr) == 0) /* standard INT */ 1513 && (SRCBUSID(intr) == pciBus) 1514 && (SRCBUSDEVICE(intr) == pciDevice) 1515 && (SRCBUSLINE(intr) == pciInt)) /* a candidate IRQ */ 1516 if (apic_int_is_bus_type(intr, PCI)) 1517 return INTIRQ(intr); /* exact match */ 1518 1519 return -1; /* NOT found */ 1520 } 1521 1522 int 1523 next_apic_irq(int irq) 1524 { 1525 int intr, ointr; 1526 int bus, bustype; 1527 1528 bus = 0; 1529 bustype = 0; 1530 for (intr = 0; intr < nintrs; intr++) { 1531 if (INTIRQ(intr) != irq || INTTYPE(intr) != 0) 1532 continue; 1533 bus = SRCBUSID(intr); 1534 bustype = apic_bus_type(bus); 1535 if (bustype != ISA && 1536 bustype != EISA && 1537 bustype != PCI) 1538 continue; 1539 break; 1540 } 1541 if (intr >= nintrs) { 1542 return -1; 1543 } 1544 for (ointr = intr + 1; ointr < nintrs; ointr++) { 1545 if (INTTYPE(ointr) != 0) 1546 continue; 1547 if (bus != SRCBUSID(ointr)) 1548 continue; 1549 if (bustype == PCI) { 1550 if (SRCBUSDEVICE(intr) != SRCBUSDEVICE(ointr)) 1551 continue; 1552 if (SRCBUSLINE(intr) != SRCBUSLINE(ointr)) 1553 continue; 1554 } 1555 if (bustype == ISA || bustype == EISA) { 1556 if (SRCBUSIRQ(intr) != SRCBUSIRQ(ointr)) 1557 continue; 1558 } 1559 if (INTPIN(intr) == INTPIN(ointr)) 1560 continue; 1561 break; 1562 } 1563 if (ointr >= nintrs) { 1564 return -1; 1565 } 1566 return INTIRQ(ointr); 1567 } 1568 #undef SRCBUSLINE 1569 #undef SRCBUSDEVICE 1570 #undef SRCBUSID 1571 #undef SRCBUSIRQ 1572 1573 #undef INTPIN 1574 #undef INTIRQ 1575 #undef INTAPIC 1576 #undef INTTYPE 1577 1578 1579 /* 1580 * Reprogram the MB chipset to NOT redirect an ISA INTerrupt. 1581 * 1582 * XXX FIXME: 1583 * Exactly what this means is unclear at this point. It is a solution 1584 * for motherboards that redirect the MBIRQ0 pin. Generically a motherboard 1585 * could route any of the ISA INTs to upper (>15) IRQ values. But most would 1586 * NOT be redirected via MBIRQ0, thus "undirect()ing" them would NOT be an 1587 * option. 1588 */ 1589 int 1590 undirect_isa_irq(int rirq) 1591 { 1592 #if defined(READY) 1593 if (bootverbose) 1594 printf("Freeing redirected ISA irq %d.\n", rirq); 1595 /** FIXME: tickle the MB redirector chip */ 1596 return ???; 1597 #else 1598 if (bootverbose) 1599 printf("Freeing (NOT implemented) redirected ISA irq %d.\n", rirq); 1600 return 0; 1601 #endif /* READY */ 1602 } 1603 1604 1605 /* 1606 * Reprogram the MB chipset to NOT redirect a PCI INTerrupt 1607 */ 1608 int 1609 undirect_pci_irq(int rirq) 1610 { 1611 #if defined(READY) 1612 if (bootverbose) 1613 printf("Freeing redirected PCI irq %d.\n", rirq); 1614 1615 /** FIXME: tickle the MB redirector chip */ 1616 return ???; 1617 #else 1618 if (bootverbose) 1619 printf("Freeing (NOT implemented) redirected PCI irq %d.\n", 1620 rirq); 1621 return 0; 1622 #endif /* READY */ 1623 } 1624 1625 1626 /* 1627 * given a bus ID, return: 1628 * the bus type if found 1629 * -1 if NOT found 1630 */ 1631 int 1632 apic_bus_type(int id) 1633 { 1634 int x; 1635 1636 for (x = 0; x < mp_nbusses; ++x) 1637 if (bus_data[x].bus_id == id) 1638 return bus_data[x].bus_type; 1639 1640 return -1; 1641 } 1642 1643 1644 /* 1645 * given a LOGICAL APIC# and pin#, return: 1646 * the associated src bus ID if found 1647 * -1 if NOT found 1648 */ 1649 int 1650 apic_src_bus_id(int apic, int pin) 1651 { 1652 int x; 1653 1654 /* search each of the possible INTerrupt sources */ 1655 for (x = 0; x < nintrs; ++x) 1656 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) && 1657 (pin == io_apic_ints[x].dst_apic_int)) 1658 return (io_apic_ints[x].src_bus_id); 1659 1660 return -1; /* NOT found */ 1661 } 1662 1663 1664 /* 1665 * given a LOGICAL APIC# and pin#, return: 1666 * the associated src bus IRQ if found 1667 * -1 if NOT found 1668 */ 1669 int 1670 apic_src_bus_irq(int apic, int pin) 1671 { 1672 int x; 1673 1674 for (x = 0; x < nintrs; x++) 1675 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) && 1676 (pin == io_apic_ints[x].dst_apic_int)) 1677 return (io_apic_ints[x].src_bus_irq); 1678 1679 return -1; /* NOT found */ 1680 } 1681 1682 1683 /* 1684 * given a LOGICAL APIC# and pin#, return: 1685 * the associated INTerrupt type if found 1686 * -1 if NOT found 1687 */ 1688 int 1689 apic_int_type(int apic, int pin) 1690 { 1691 int x; 1692 1693 /* search each of the possible INTerrupt sources */ 1694 for (x = 0; x < nintrs; ++x) 1695 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) && 1696 (pin == io_apic_ints[x].dst_apic_int)) 1697 return (io_apic_ints[x].int_type); 1698 1699 return -1; /* NOT found */ 1700 } 1701 1702 int 1703 apic_irq(int apic, int pin) 1704 { 1705 int x; 1706 int res; 1707 1708 for (x = 0; x < nintrs; ++x) 1709 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) && 1710 (pin == io_apic_ints[x].dst_apic_int)) { 1711 res = io_apic_ints[x].int_vector; 1712 if (res == 0xff) 1713 return -1; 1714 if (apic != int_to_apicintpin[res].ioapic) 1715 panic("apic_irq: inconsistent table"); 1716 if (pin != int_to_apicintpin[res].int_pin) 1717 panic("apic_irq inconsistent table (2)"); 1718 return res; 1719 } 1720 return -1; 1721 } 1722 1723 1724 /* 1725 * given a LOGICAL APIC# and pin#, return: 1726 * the associated trigger mode if found 1727 * -1 if NOT found 1728 */ 1729 int 1730 apic_trigger(int apic, int pin) 1731 { 1732 int x; 1733 1734 /* search each of the possible INTerrupt sources */ 1735 for (x = 0; x < nintrs; ++x) 1736 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) && 1737 (pin == io_apic_ints[x].dst_apic_int)) 1738 return ((io_apic_ints[x].int_flags >> 2) & 0x03); 1739 1740 return -1; /* NOT found */ 1741 } 1742 1743 1744 /* 1745 * given a LOGICAL APIC# and pin#, return: 1746 * the associated 'active' level if found 1747 * -1 if NOT found 1748 */ 1749 int 1750 apic_polarity(int apic, int pin) 1751 { 1752 int x; 1753 1754 /* search each of the possible INTerrupt sources */ 1755 for (x = 0; x < nintrs; ++x) 1756 if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) && 1757 (pin == io_apic_ints[x].dst_apic_int)) 1758 return (io_apic_ints[x].int_flags & 0x03); 1759 1760 return -1; /* NOT found */ 1761 } 1762 1763 1764 /* 1765 * set data according to MP defaults 1766 * FIXME: probably not complete yet... 1767 */ 1768 static void 1769 default_mp_table(int type) 1770 { 1771 int ap_cpu_id; 1772 #if defined(APIC_IO) 1773 int io_apic_id; 1774 int pin; 1775 #endif /* APIC_IO */ 1776 1777 #if 0 1778 printf(" MP default config type: %d\n", type); 1779 switch (type) { 1780 case 1: 1781 printf(" bus: ISA, APIC: 82489DX\n"); 1782 break; 1783 case 2: 1784 printf(" bus: EISA, APIC: 82489DX\n"); 1785 break; 1786 case 3: 1787 printf(" bus: EISA, APIC: 82489DX\n"); 1788 break; 1789 case 4: 1790 printf(" bus: MCA, APIC: 82489DX\n"); 1791 break; 1792 case 5: 1793 printf(" bus: ISA+PCI, APIC: Integrated\n"); 1794 break; 1795 case 6: 1796 printf(" bus: EISA+PCI, APIC: Integrated\n"); 1797 break; 1798 case 7: 1799 printf(" bus: MCA+PCI, APIC: Integrated\n"); 1800 break; 1801 default: 1802 printf(" future type\n"); 1803 break; 1804 /* NOTREACHED */ 1805 } 1806 #endif /* 0 */ 1807 1808 boot_cpu_id = (lapic.id & APIC_ID_MASK) >> 24; 1809 ap_cpu_id = (boot_cpu_id == 0) ? 1 : 0; 1810 1811 /* BSP */ 1812 CPU_TO_ID(0) = boot_cpu_id; 1813 ID_TO_CPU(boot_cpu_id) = 0; 1814 1815 /* one and only AP */ 1816 CPU_TO_ID(1) = ap_cpu_id; 1817 ID_TO_CPU(ap_cpu_id) = 1; 1818 1819 #if defined(APIC_IO) 1820 /* one and only IO APIC */ 1821 io_apic_id = (io_apic_read(0, IOAPIC_ID) & APIC_ID_MASK) >> 24; 1822 1823 /* 1824 * sanity check, refer to MP spec section 3.6.6, last paragraph 1825 * necessary as some hardware isn't properly setting up the IO APIC 1826 */ 1827 #if defined(REALLY_ANAL_IOAPICID_VALUE) 1828 if (io_apic_id != 2) { 1829 #else 1830 if ((io_apic_id == 0) || (io_apic_id == 1) || (io_apic_id == 15)) { 1831 #endif /* REALLY_ANAL_IOAPICID_VALUE */ 1832 io_apic_set_id(0, 2); 1833 io_apic_id = 2; 1834 } 1835 IO_TO_ID(0) = io_apic_id; 1836 ID_TO_IO(io_apic_id) = 0; 1837 #endif /* APIC_IO */ 1838 1839 /* fill out bus entries */ 1840 switch (type) { 1841 case 1: 1842 case 2: 1843 case 3: 1844 case 4: 1845 case 5: 1846 case 6: 1847 case 7: 1848 bus_data[0].bus_id = default_data[type - 1][1]; 1849 bus_data[0].bus_type = default_data[type - 1][2]; 1850 bus_data[1].bus_id = default_data[type - 1][3]; 1851 bus_data[1].bus_type = default_data[type - 1][4]; 1852 break; 1853 1854 /* case 4: case 7: MCA NOT supported */ 1855 default: /* illegal/reserved */ 1856 panic("BAD default MP config: %d", type); 1857 /* NOTREACHED */ 1858 } 1859 1860 #if defined(APIC_IO) 1861 /* general cases from MP v1.4, table 5-2 */ 1862 for (pin = 0; pin < 16; ++pin) { 1863 io_apic_ints[pin].int_type = 0; 1864 io_apic_ints[pin].int_flags = 0x05; /* edge/active-hi */ 1865 io_apic_ints[pin].src_bus_id = 0; 1866 io_apic_ints[pin].src_bus_irq = pin; /* IRQ2 caught below */ 1867 io_apic_ints[pin].dst_apic_id = io_apic_id; 1868 io_apic_ints[pin].dst_apic_int = pin; /* 1-to-1 */ 1869 } 1870 1871 /* special cases from MP v1.4, table 5-2 */ 1872 if (type == 2) { 1873 io_apic_ints[2].int_type = 0xff; /* N/C */ 1874 io_apic_ints[13].int_type = 0xff; /* N/C */ 1875 #if !defined(APIC_MIXED_MODE) 1876 /** FIXME: ??? */ 1877 panic("sorry, can't support type 2 default yet"); 1878 #endif /* APIC_MIXED_MODE */ 1879 } 1880 else 1881 io_apic_ints[2].src_bus_irq = 0; /* ISA IRQ0 is on APIC INT 2 */ 1882 1883 if (type == 7) 1884 io_apic_ints[0].int_type = 0xff; /* N/C */ 1885 else 1886 io_apic_ints[0].int_type = 3; /* vectored 8259 */ 1887 #endif /* APIC_IO */ 1888 } 1889 1890 1891 /* 1892 * initialize all the SMP locks 1893 */ 1894 1895 /* critical region around IO APIC, apic_imen */ 1896 struct simplelock imen_lock; 1897 1898 /* critical region around splxx(), cpl, cml, cil, ipending */ 1899 struct simplelock cpl_lock; 1900 1901 /* Make FAST_INTR() routines sequential */ 1902 struct simplelock fast_intr_lock; 1903 1904 /* critical region around INTR() routines */ 1905 struct simplelock intr_lock; 1906 1907 /* lock region used by kernel profiling */ 1908 struct simplelock mcount_lock; 1909 1910 #ifdef USE_COMLOCK 1911 /* locks com (tty) data/hardware accesses: a FASTINTR() */ 1912 struct simplelock com_lock; 1913 #endif /* USE_COMLOCK */ 1914 1915 /* lock around the MP rendezvous */ 1916 static struct simplelock smp_rv_lock; 1917 1918 /* only 1 CPU can panic at a time :) */ 1919 struct simplelock panic_lock; 1920 1921 static void 1922 init_locks(void) 1923 { 1924 #if defined(APIC_INTR_DIAGNOSTIC) && defined(APIC_INTR_DIAGNOSTIC_IRQ) 1925 s_lock_init((struct simplelock*)&apic_itrace_debuglock); 1926 #endif 1927 1928 s_lock_init((struct simplelock*)&mcount_lock); 1929 1930 s_lock_init((struct simplelock*)&fast_intr_lock); 1931 s_lock_init((struct simplelock*)&intr_lock); 1932 s_lock_init((struct simplelock*)&imen_lock); 1933 s_lock_init((struct simplelock*)&cpl_lock); 1934 s_lock_init(&smp_rv_lock); 1935 s_lock_init(&panic_lock); 1936 1937 #ifdef USE_COMLOCK 1938 s_lock_init((struct simplelock*)&com_lock); 1939 #endif /* USE_COMLOCK */ 1940 1941 s_lock_init(&ap_boot_lock); 1942 } 1943 1944 /* 1945 * start each AP in our list 1946 */ 1947 static int 1948 start_all_aps(u_int boot_addr) 1949 { 1950 int x, i, pg; 1951 u_char mpbiosreason; 1952 u_long mpbioswarmvec; 1953 struct globaldata *gd; 1954 char *stack; 1955 1956 POSTCODE(START_ALL_APS_POST); 1957 1958 /* initialize BSP's local APIC */ 1959 apic_initialize(); 1960 bsp_apic_ready = 1; 1961 1962 /* install the AP 1st level boot code */ 1963 install_ap_tramp(boot_addr); 1964 1965 1966 /* save the current value of the warm-start vector */ 1967 mpbioswarmvec = *((u_long *) WARMBOOT_OFF); 1968 #ifndef PC98 1969 outb(CMOS_REG, BIOS_RESET); 1970 mpbiosreason = inb(CMOS_DATA); 1971 #endif 1972 1973 /* record BSP in CPU map */ 1974 all_cpus = 1; 1975 1976 /* set up 0 -> 4MB P==V mapping for AP boot */ 1977 *(int *)PTD = PG_V | PG_RW | ((uintptr_t)(void *)KPTphys & PG_FRAME); 1978 invltlb(); 1979 1980 /* start each AP */ 1981 for (x = 1; x <= mp_naps; ++x) { 1982 1983 /* This is a bit verbose, it will go away soon. */ 1984 1985 /* first page of AP's private space */ 1986 pg = x * i386_btop(sizeof(struct privatespace)); 1987 1988 /* allocate a new private data page */ 1989 gd = (struct globaldata *)kmem_alloc(kernel_map, PAGE_SIZE); 1990 1991 /* wire it into the private page table page */ 1992 SMPpt[pg] = (pt_entry_t)(PG_V | PG_RW | vtophys(gd)); 1993 1994 /* allocate and set up an idle stack data page */ 1995 stack = (char *)kmem_alloc(kernel_map, UPAGES*PAGE_SIZE); 1996 for (i = 0; i < UPAGES; i++) 1997 SMPpt[pg + 5 + i] = (pt_entry_t) 1998 (PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack)); 1999 2000 SMPpt[pg + 1] = 0; /* *prv_CMAP1 */ 2001 SMPpt[pg + 2] = 0; /* *prv_CMAP2 */ 2002 SMPpt[pg + 3] = 0; /* *prv_CMAP3 */ 2003 SMPpt[pg + 4] = 0; /* *prv_PMAP1 */ 2004 2005 /* prime data page for it to use */ 2006 SLIST_INSERT_HEAD(&cpuhead, gd, gd_allcpu); 2007 gd->gd_cpuid = x; 2008 gd->gd_cpu_lockid = x << 24; 2009 gd->gd_prv_CMAP1 = &SMPpt[pg + 1]; 2010 gd->gd_prv_CMAP2 = &SMPpt[pg + 2]; 2011 gd->gd_prv_CMAP3 = &SMPpt[pg + 3]; 2012 gd->gd_prv_PMAP1 = &SMPpt[pg + 4]; 2013 gd->gd_prv_CADDR1 = SMP_prvspace[x].CPAGE1; 2014 gd->gd_prv_CADDR2 = SMP_prvspace[x].CPAGE2; 2015 gd->gd_prv_CADDR3 = SMP_prvspace[x].CPAGE3; 2016 gd->gd_prv_PADDR1 = (unsigned *)SMP_prvspace[x].PPAGE1; 2017 2018 /* setup a vector to our boot code */ 2019 *((volatile u_short *) WARMBOOT_OFF) = WARMBOOT_TARGET; 2020 *((volatile u_short *) WARMBOOT_SEG) = (boot_addr >> 4); 2021 #ifndef PC98 2022 outb(CMOS_REG, BIOS_RESET); 2023 outb(CMOS_DATA, BIOS_WARM); /* 'warm-start' */ 2024 #endif 2025 2026 bootSTK = &SMP_prvspace[x].idlestack[UPAGES*PAGE_SIZE]; 2027 bootAP = x; 2028 2029 /* attempt to start the Application Processor */ 2030 CHECK_INIT(99); /* setup checkpoints */ 2031 if (!start_ap(x, boot_addr)) { 2032 printf("AP #%d (PHY# %d) failed!\n", x, CPU_TO_ID(x)); 2033 CHECK_PRINT("trace"); /* show checkpoints */ 2034 /* better panic as the AP may be running loose */ 2035 printf("panic y/n? [y] "); 2036 if (cngetc() != 'n') 2037 panic("bye-bye"); 2038 } 2039 CHECK_PRINT("trace"); /* show checkpoints */ 2040 2041 /* record its version info */ 2042 cpu_apic_versions[x] = cpu_apic_versions[0]; 2043 2044 all_cpus |= (1 << x); /* record AP in CPU map */ 2045 } 2046 2047 /* build our map of 'other' CPUs */ 2048 PCPU_SET(other_cpus, all_cpus & ~(1 << PCPU_GET(cpuid))); 2049 2050 /* fill in our (BSP) APIC version */ 2051 cpu_apic_versions[0] = lapic.version; 2052 2053 /* restore the warmstart vector */ 2054 *(u_long *) WARMBOOT_OFF = mpbioswarmvec; 2055 #ifndef PC98 2056 outb(CMOS_REG, BIOS_RESET); 2057 outb(CMOS_DATA, mpbiosreason); 2058 #endif 2059 2060 /* 2061 * Set up the idle context for the BSP. Similar to above except 2062 * that some was done by locore, some by pmap.c and some is implicit 2063 * because the BSP is cpu#0 and the page is initially zero, and also 2064 * because we can refer to variables by name on the BSP.. 2065 */ 2066 2067 /* Allocate and setup BSP idle stack */ 2068 stack = (char *)kmem_alloc(kernel_map, UPAGES * PAGE_SIZE); 2069 for (i = 0; i < UPAGES; i++) 2070 SMPpt[5 + i] = (pt_entry_t) 2071 (PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack)); 2072 2073 *(int *)PTD = 0; 2074 pmap_set_opt(); 2075 2076 /* number of APs actually started */ 2077 return mp_ncpus - 1; 2078 } 2079 2080 2081 /* 2082 * load the 1st level AP boot code into base memory. 2083 */ 2084 2085 /* targets for relocation */ 2086 extern void bigJump(void); 2087 extern void bootCodeSeg(void); 2088 extern void bootDataSeg(void); 2089 extern void MPentry(void); 2090 extern u_int MP_GDT; 2091 extern u_int mp_gdtbase; 2092 2093 static void 2094 install_ap_tramp(u_int boot_addr) 2095 { 2096 int x; 2097 int size = *(int *) ((u_long) & bootMP_size); 2098 u_char *src = (u_char *) ((u_long) bootMP); 2099 u_char *dst = (u_char *) boot_addr + KERNBASE; 2100 u_int boot_base = (u_int) bootMP; 2101 u_int8_t *dst8; 2102 u_int16_t *dst16; 2103 u_int32_t *dst32; 2104 2105 POSTCODE(INSTALL_AP_TRAMP_POST); 2106 2107 for (x = 0; x < size; ++x) 2108 *dst++ = *src++; 2109 2110 /* 2111 * modify addresses in code we just moved to basemem. unfortunately we 2112 * need fairly detailed info about mpboot.s for this to work. changes 2113 * to mpboot.s might require changes here. 2114 */ 2115 2116 /* boot code is located in KERNEL space */ 2117 dst = (u_char *) boot_addr + KERNBASE; 2118 2119 /* modify the lgdt arg */ 2120 dst32 = (u_int32_t *) (dst + ((u_int) & mp_gdtbase - boot_base)); 2121 *dst32 = boot_addr + ((u_int) & MP_GDT - boot_base); 2122 2123 /* modify the ljmp target for MPentry() */ 2124 dst32 = (u_int32_t *) (dst + ((u_int) bigJump - boot_base) + 1); 2125 *dst32 = ((u_int) MPentry - KERNBASE); 2126 2127 /* modify the target for boot code segment */ 2128 dst16 = (u_int16_t *) (dst + ((u_int) bootCodeSeg - boot_base)); 2129 dst8 = (u_int8_t *) (dst16 + 1); 2130 *dst16 = (u_int) boot_addr & 0xffff; 2131 *dst8 = ((u_int) boot_addr >> 16) & 0xff; 2132 2133 /* modify the target for boot data segment */ 2134 dst16 = (u_int16_t *) (dst + ((u_int) bootDataSeg - boot_base)); 2135 dst8 = (u_int8_t *) (dst16 + 1); 2136 *dst16 = (u_int) boot_addr & 0xffff; 2137 *dst8 = ((u_int) boot_addr >> 16) & 0xff; 2138 } 2139 2140 2141 /* 2142 * this function starts the AP (application processor) identified 2143 * by the APIC ID 'physicalCpu'. It does quite a "song and dance" 2144 * to accomplish this. This is necessary because of the nuances 2145 * of the different hardware we might encounter. It ain't pretty, 2146 * but it seems to work. 2147 */ 2148 static int 2149 start_ap(int logical_cpu, u_int boot_addr) 2150 { 2151 int physical_cpu; 2152 int vector; 2153 int cpus; 2154 u_long icr_lo, icr_hi; 2155 2156 POSTCODE(START_AP_POST); 2157 2158 /* get the PHYSICAL APIC ID# */ 2159 physical_cpu = CPU_TO_ID(logical_cpu); 2160 2161 /* calculate the vector */ 2162 vector = (boot_addr >> 12) & 0xff; 2163 2164 /* used as a watchpoint to signal AP startup */ 2165 cpus = mp_ncpus; 2166 2167 /* 2168 * first we do an INIT/RESET IPI this INIT IPI might be run, reseting 2169 * and running the target CPU. OR this INIT IPI might be latched (P5 2170 * bug), CPU waiting for STARTUP IPI. OR this INIT IPI might be 2171 * ignored. 2172 */ 2173 2174 /* setup the address for the target AP */ 2175 icr_hi = lapic.icr_hi & ~APIC_ID_MASK; 2176 icr_hi |= (physical_cpu << 24); 2177 lapic.icr_hi = icr_hi; 2178 2179 /* do an INIT IPI: assert RESET */ 2180 icr_lo = lapic.icr_lo & 0xfff00000; 2181 lapic.icr_lo = icr_lo | 0x0000c500; 2182 2183 /* wait for pending status end */ 2184 while (lapic.icr_lo & APIC_DELSTAT_MASK) 2185 /* spin */ ; 2186 2187 /* do an INIT IPI: deassert RESET */ 2188 lapic.icr_lo = icr_lo | 0x00008500; 2189 2190 /* wait for pending status end */ 2191 u_sleep(10000); /* wait ~10mS */ 2192 while (lapic.icr_lo & APIC_DELSTAT_MASK) 2193 /* spin */ ; 2194 2195 /* 2196 * next we do a STARTUP IPI: the previous INIT IPI might still be 2197 * latched, (P5 bug) this 1st STARTUP would then terminate 2198 * immediately, and the previously started INIT IPI would continue. OR 2199 * the previous INIT IPI has already run. and this STARTUP IPI will 2200 * run. OR the previous INIT IPI was ignored. and this STARTUP IPI 2201 * will run. 2202 */ 2203 2204 /* do a STARTUP IPI */ 2205 lapic.icr_lo = icr_lo | 0x00000600 | vector; 2206 while (lapic.icr_lo & APIC_DELSTAT_MASK) 2207 /* spin */ ; 2208 u_sleep(200); /* wait ~200uS */ 2209 2210 /* 2211 * finally we do a 2nd STARTUP IPI: this 2nd STARTUP IPI should run IF 2212 * the previous STARTUP IPI was cancelled by a latched INIT IPI. OR 2213 * this STARTUP IPI will be ignored, as only ONE STARTUP IPI is 2214 * recognized after hardware RESET or INIT IPI. 2215 */ 2216 2217 lapic.icr_lo = icr_lo | 0x00000600 | vector; 2218 while (lapic.icr_lo & APIC_DELSTAT_MASK) 2219 /* spin */ ; 2220 u_sleep(200); /* wait ~200uS */ 2221 2222 /* wait for it to start */ 2223 set_apic_timer(5000000);/* == 5 seconds */ 2224 while (read_apic_timer()) 2225 if (mp_ncpus > cpus) 2226 return 1; /* return SUCCESS */ 2227 2228 return 0; /* return FAILURE */ 2229 } 2230 2231 /* 2232 * Flush the TLB on all other CPU's 2233 * 2234 * XXX: Needs to handshake and wait for completion before proceding. 2235 */ 2236 void 2237 smp_invltlb(void) 2238 { 2239 #if defined(APIC_IO) 2240 if (smp_started && invltlb_ok) 2241 all_but_self_ipi(XINVLTLB_OFFSET); 2242 #endif /* APIC_IO */ 2243 } 2244 2245 void 2246 invlpg(u_int addr) 2247 { 2248 __asm __volatile("invlpg (%0)"::"r"(addr):"memory"); 2249 2250 /* send a message to the other CPUs */ 2251 smp_invltlb(); 2252 } 2253 2254 void 2255 invltlb(void) 2256 { 2257 u_long temp; 2258 2259 /* 2260 * This should be implemented as load_cr3(rcr3()) when load_cr3() is 2261 * inlined. 2262 */ 2263 __asm __volatile("movl %%cr3, %0; movl %0, %%cr3":"=r"(temp) :: "memory"); 2264 2265 /* send a message to the other CPUs */ 2266 smp_invltlb(); 2267 } 2268 2269 2270 /* 2271 * When called the executing CPU will send an IPI to all other CPUs 2272 * requesting that they halt execution. 2273 * 2274 * Usually (but not necessarily) called with 'other_cpus' as its arg. 2275 * 2276 * - Signals all CPUs in map to stop. 2277 * - Waits for each to stop. 2278 * 2279 * Returns: 2280 * -1: error 2281 * 0: NA 2282 * 1: ok 2283 * 2284 * XXX FIXME: this is not MP-safe, needs a lock to prevent multiple CPUs 2285 * from executing at same time. 2286 */ 2287 int 2288 stop_cpus(u_int map) 2289 { 2290 int count = 0; 2291 2292 if (!smp_started) 2293 return 0; 2294 2295 /* send the Xcpustop IPI to all CPUs in map */ 2296 selected_apic_ipi(map, XCPUSTOP_OFFSET, APIC_DELMODE_FIXED); 2297 2298 while (count++ < 100000 && (stopped_cpus & map) != map) 2299 /* spin */ ; 2300 2301 #ifdef DIAGNOSTIC 2302 if ((stopped_cpus & map) != map) 2303 printf("Warning: CPUs 0x%x did not stop!\n", 2304 (~(stopped_cpus & map)) & map); 2305 #endif 2306 2307 return 1; 2308 } 2309 2310 2311 /* 2312 * Called by a CPU to restart stopped CPUs. 2313 * 2314 * Usually (but not necessarily) called with 'stopped_cpus' as its arg. 2315 * 2316 * - Signals all CPUs in map to restart. 2317 * - Waits for each to restart. 2318 * 2319 * Returns: 2320 * -1: error 2321 * 0: NA 2322 * 1: ok 2323 */ 2324 int 2325 restart_cpus(u_int map) 2326 { 2327 int count = 0; 2328 2329 if (!smp_started) 2330 return 0; 2331 2332 started_cpus = map; /* signal other cpus to restart */ 2333 2334 /* wait for each to clear its bit */ 2335 while (count++ < 100000 && (stopped_cpus & map) != 0) 2336 /* spin */ ; 2337 2338 #ifdef DIAGNOSTIC 2339 if ((stopped_cpus & map) != 0) 2340 printf("Warning: CPUs 0x%x did not restart!\n", 2341 (~(stopped_cpus & map)) & map); 2342 #endif 2343 2344 return 1; 2345 } 2346 2347 int smp_active = 0; /* are the APs allowed to run? */ 2348 SYSCTL_INT(_machdep, OID_AUTO, smp_active, CTLFLAG_RW, &smp_active, 0, ""); 2349 2350 /* XXX maybe should be hw.ncpu */ 2351 static int smp_cpus = 1; /* how many cpu's running */ 2352 SYSCTL_INT(_machdep, OID_AUTO, smp_cpus, CTLFLAG_RD, &smp_cpus, 0, ""); 2353 2354 int invltlb_ok = 0; /* throttle smp_invltlb() till safe */ 2355 SYSCTL_INT(_machdep, OID_AUTO, invltlb_ok, CTLFLAG_RW, &invltlb_ok, 0, ""); 2356 2357 /* Warning: Do not staticize. Used from swtch.s */ 2358 int do_page_zero_idle = 1; /* bzero pages for fun and profit in idleloop */ 2359 SYSCTL_INT(_machdep, OID_AUTO, do_page_zero_idle, CTLFLAG_RW, 2360 &do_page_zero_idle, 0, ""); 2361 2362 /* Is forwarding of a interrupt to the CPU holding the ISR lock enabled ? */ 2363 int forward_irq_enabled = 1; 2364 SYSCTL_INT(_machdep, OID_AUTO, forward_irq_enabled, CTLFLAG_RW, 2365 &forward_irq_enabled, 0, ""); 2366 2367 /* Enable forwarding of a signal to a process running on a different CPU */ 2368 static int forward_signal_enabled = 1; 2369 SYSCTL_INT(_machdep, OID_AUTO, forward_signal_enabled, CTLFLAG_RW, 2370 &forward_signal_enabled, 0, ""); 2371 2372 /* Enable forwarding of roundrobin to all other cpus */ 2373 static int forward_roundrobin_enabled = 1; 2374 SYSCTL_INT(_machdep, OID_AUTO, forward_roundrobin_enabled, CTLFLAG_RW, 2375 &forward_roundrobin_enabled, 0, ""); 2376 2377 /* 2378 * This is called once the rest of the system is up and running and we're 2379 * ready to let the AP's out of the pen. 2380 */ 2381 void ap_init(void); 2382 2383 void 2384 ap_init(void) 2385 { 2386 u_int apic_id; 2387 2388 /* lock against other AP's that are waking up */ 2389 s_lock(&ap_boot_lock); 2390 2391 /* BSP may have changed PTD while we're waiting for the lock */ 2392 cpu_invltlb(); 2393 2394 smp_cpus++; 2395 2396 #if defined(I586_CPU) && !defined(NO_F00F_HACK) 2397 lidt(&r_idt); 2398 #endif 2399 2400 /* Build our map of 'other' CPUs. */ 2401 PCPU_SET(other_cpus, all_cpus & ~(1 << PCPU_GET(cpuid))); 2402 2403 printf("SMP: AP CPU #%d Launched!\n", PCPU_GET(cpuid)); 2404 2405 /* set up CPU registers and state */ 2406 cpu_setregs(); 2407 2408 /* set up FPU state on the AP */ 2409 npxinit(__INITIAL_NPXCW__); 2410 2411 /* A quick check from sanity claus */ 2412 apic_id = (apic_id_to_logical[(lapic.id & 0x0f000000) >> 24]); 2413 if (PCPU_GET(cpuid) != apic_id) { 2414 printf("SMP: cpuid = %d\n", PCPU_GET(cpuid)); 2415 printf("SMP: apic_id = %d\n", apic_id); 2416 printf("PTD[MPPTDI] = %p\n", (void *)PTD[MPPTDI]); 2417 panic("cpuid mismatch! boom!!"); 2418 } 2419 2420 /* Init local apic for irq's */ 2421 apic_initialize(); 2422 2423 /* Set memory range attributes for this CPU to match the BSP */ 2424 mem_range_AP_init(); 2425 2426 /* 2427 * Activate smp_invltlb, although strictly speaking, this isn't 2428 * quite correct yet. We should have a bitfield for cpus willing 2429 * to accept TLB flush IPI's or something and sync them. 2430 */ 2431 if (smp_cpus == mp_ncpus) { 2432 invltlb_ok = 1; 2433 smp_started = 1; /* enable IPI's, tlb shootdown, freezes etc */ 2434 smp_active = 1; /* historic */ 2435 } 2436 2437 /* let other AP's wake up now */ 2438 s_unlock(&ap_boot_lock); 2439 2440 /* wait until all the AP's are up */ 2441 while (smp_started == 0) 2442 ; /* nothing */ 2443 2444 /* 2445 * Set curproc to our per-cpu idleproc so that mutexes have 2446 * something unique to lock with. 2447 */ 2448 PCPU_SET(curproc, PCPU_GET(idleproc)); 2449 2450 microuptime(PCPU_PTR(switchtime)); 2451 PCPU_SET(switchticks, ticks); 2452 2453 /* ok, now grab sched_lock and enter the scheduler */ 2454 enable_intr(); 2455 mtx_enter(&sched_lock, MTX_SPIN); 2456 cpu_throw(); /* doesn't return */ 2457 2458 panic("scheduler returned us to ap_init"); 2459 } 2460 2461 #ifdef BETTER_CLOCK 2462 2463 #define CHECKSTATE_USER 0 2464 #define CHECKSTATE_SYS 1 2465 #define CHECKSTATE_INTR 2 2466 2467 /* Do not staticize. Used from apic_vector.s */ 2468 struct proc* checkstate_curproc[MAXCPU]; 2469 int checkstate_cpustate[MAXCPU]; 2470 u_long checkstate_pc[MAXCPU]; 2471 2472 #define PC_TO_INDEX(pc, prof) \ 2473 ((int)(((u_quad_t)((pc) - (prof)->pr_off) * \ 2474 (u_quad_t)((prof)->pr_scale)) >> 16) & ~1) 2475 2476 static void 2477 addupc_intr_forwarded(struct proc *p, int id, int *astmap) 2478 { 2479 int i; 2480 struct uprof *prof; 2481 u_long pc; 2482 2483 pc = checkstate_pc[id]; 2484 prof = &p->p_stats->p_prof; 2485 if (pc >= prof->pr_off && 2486 (i = PC_TO_INDEX(pc, prof)) < prof->pr_size) { 2487 if ((p->p_flag & P_OWEUPC) == 0) { 2488 prof->pr_addr = pc; 2489 prof->pr_ticks = 1; 2490 p->p_flag |= P_OWEUPC; 2491 } 2492 *astmap |= (1 << id); 2493 } 2494 } 2495 2496 static void 2497 forwarded_statclock(int id, int pscnt, int *astmap) 2498 { 2499 struct pstats *pstats; 2500 long rss; 2501 struct rusage *ru; 2502 struct vmspace *vm; 2503 int cpustate; 2504 struct proc *p; 2505 #ifdef GPROF 2506 register struct gmonparam *g; 2507 int i; 2508 #endif 2509 2510 p = checkstate_curproc[id]; 2511 cpustate = checkstate_cpustate[id]; 2512 2513 /* XXX */ 2514 if (p->p_ithd) 2515 cpustate = CHECKSTATE_INTR; 2516 else if (p == SMP_prvspace[id].globaldata.gd_idleproc) 2517 cpustate = CHECKSTATE_SYS; 2518 2519 switch (cpustate) { 2520 case CHECKSTATE_USER: 2521 if (p->p_flag & P_PROFIL) 2522 addupc_intr_forwarded(p, id, astmap); 2523 if (pscnt > 1) 2524 return; 2525 p->p_uticks++; 2526 if (p->p_nice > NZERO) 2527 cp_time[CP_NICE]++; 2528 else 2529 cp_time[CP_USER]++; 2530 break; 2531 case CHECKSTATE_SYS: 2532 #ifdef GPROF 2533 /* 2534 * Kernel statistics are just like addupc_intr, only easier. 2535 */ 2536 g = &_gmonparam; 2537 if (g->state == GMON_PROF_ON) { 2538 i = checkstate_pc[id] - g->lowpc; 2539 if (i < g->textsize) { 2540 i /= HISTFRACTION * sizeof(*g->kcount); 2541 g->kcount[i]++; 2542 } 2543 } 2544 #endif 2545 if (pscnt > 1) 2546 return; 2547 2548 p->p_sticks++; 2549 if (p == SMP_prvspace[id].globaldata.gd_idleproc) 2550 cp_time[CP_IDLE]++; 2551 else 2552 cp_time[CP_SYS]++; 2553 break; 2554 case CHECKSTATE_INTR: 2555 default: 2556 #ifdef GPROF 2557 /* 2558 * Kernel statistics are just like addupc_intr, only easier. 2559 */ 2560 g = &_gmonparam; 2561 if (g->state == GMON_PROF_ON) { 2562 i = checkstate_pc[id] - g->lowpc; 2563 if (i < g->textsize) { 2564 i /= HISTFRACTION * sizeof(*g->kcount); 2565 g->kcount[i]++; 2566 } 2567 } 2568 #endif 2569 if (pscnt > 1) 2570 return; 2571 if (p) 2572 p->p_iticks++; 2573 cp_time[CP_INTR]++; 2574 } 2575 schedclock(p); 2576 2577 /* Update resource usage integrals and maximums. */ 2578 if ((pstats = p->p_stats) != NULL && 2579 (ru = &pstats->p_ru) != NULL && 2580 (vm = p->p_vmspace) != NULL) { 2581 ru->ru_ixrss += pgtok(vm->vm_tsize); 2582 ru->ru_idrss += pgtok(vm->vm_dsize); 2583 ru->ru_isrss += pgtok(vm->vm_ssize); 2584 rss = pgtok(vmspace_resident_count(vm)); 2585 if (ru->ru_maxrss < rss) 2586 ru->ru_maxrss = rss; 2587 } 2588 } 2589 2590 void 2591 forward_statclock(int pscnt) 2592 { 2593 int map; 2594 int id; 2595 int i; 2596 2597 /* Kludge. We don't yet have separate locks for the interrupts 2598 * and the kernel. This means that we cannot let the other processors 2599 * handle complex interrupts while inhibiting them from entering 2600 * the kernel in a non-interrupt context. 2601 * 2602 * What we can do, without changing the locking mechanisms yet, 2603 * is letting the other processors handle a very simple interrupt 2604 * (wich determines the processor states), and do the main 2605 * work ourself. 2606 */ 2607 2608 if (!smp_started || !invltlb_ok || cold || panicstr) 2609 return; 2610 2611 /* Step 1: Probe state (user, cpu, interrupt, spinlock, idle ) */ 2612 2613 map = PCPU_GET(other_cpus) & ~stopped_cpus ; 2614 checkstate_probed_cpus = 0; 2615 if (map != 0) 2616 selected_apic_ipi(map, 2617 XCPUCHECKSTATE_OFFSET, APIC_DELMODE_FIXED); 2618 2619 i = 0; 2620 while (checkstate_probed_cpus != map) { 2621 /* spin */ 2622 i++; 2623 if (i == 100000) { 2624 #ifdef BETTER_CLOCK_DIAGNOSTIC 2625 printf("forward_statclock: checkstate %x\n", 2626 checkstate_probed_cpus); 2627 #endif 2628 break; 2629 } 2630 } 2631 2632 /* 2633 * Step 2: walk through other processors processes, update ticks and 2634 * profiling info. 2635 */ 2636 2637 map = 0; 2638 for (id = 0; id < mp_ncpus; id++) { 2639 if (id == PCPU_GET(cpuid)) 2640 continue; 2641 if (((1 << id) & checkstate_probed_cpus) == 0) 2642 continue; 2643 forwarded_statclock(id, pscnt, &map); 2644 } 2645 if (map != 0) { 2646 checkstate_need_ast |= map; 2647 selected_apic_ipi(map, XCPUAST_OFFSET, APIC_DELMODE_FIXED); 2648 i = 0; 2649 while ((checkstate_need_ast & map) != 0) { 2650 /* spin */ 2651 i++; 2652 if (i > 100000) { 2653 #ifdef BETTER_CLOCK_DIAGNOSTIC 2654 printf("forward_statclock: dropped ast 0x%x\n", 2655 checkstate_need_ast & map); 2656 #endif 2657 break; 2658 } 2659 } 2660 } 2661 } 2662 2663 void 2664 forward_hardclock(int pscnt) 2665 { 2666 int map; 2667 int id; 2668 struct proc *p; 2669 struct pstats *pstats; 2670 int i; 2671 2672 /* Kludge. We don't yet have separate locks for the interrupts 2673 * and the kernel. This means that we cannot let the other processors 2674 * handle complex interrupts while inhibiting them from entering 2675 * the kernel in a non-interrupt context. 2676 * 2677 * What we can do, without changing the locking mechanisms yet, 2678 * is letting the other processors handle a very simple interrupt 2679 * (wich determines the processor states), and do the main 2680 * work ourself. 2681 */ 2682 2683 if (!smp_started || !invltlb_ok || cold || panicstr) 2684 return; 2685 2686 /* Step 1: Probe state (user, cpu, interrupt, spinlock, idle) */ 2687 2688 map = PCPU_GET(other_cpus) & ~stopped_cpus ; 2689 checkstate_probed_cpus = 0; 2690 if (map != 0) 2691 selected_apic_ipi(map, 2692 XCPUCHECKSTATE_OFFSET, APIC_DELMODE_FIXED); 2693 2694 i = 0; 2695 while (checkstate_probed_cpus != map) { 2696 /* spin */ 2697 i++; 2698 if (i == 100000) { 2699 #ifdef BETTER_CLOCK_DIAGNOSTIC 2700 printf("forward_hardclock: checkstate %x\n", 2701 checkstate_probed_cpus); 2702 #endif 2703 break; 2704 } 2705 } 2706 2707 /* 2708 * Step 2: walk through other processors processes, update virtual 2709 * timer and profiling timer. If stathz == 0, also update ticks and 2710 * profiling info. 2711 */ 2712 2713 map = 0; 2714 for (id = 0; id < mp_ncpus; id++) { 2715 if (id == PCPU_GET(cpuid)) 2716 continue; 2717 if (((1 << id) & checkstate_probed_cpus) == 0) 2718 continue; 2719 p = checkstate_curproc[id]; 2720 if (p) { 2721 pstats = p->p_stats; 2722 if (checkstate_cpustate[id] == CHECKSTATE_USER && 2723 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && 2724 itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) { 2725 psignal(p, SIGVTALRM); 2726 map |= (1 << id); 2727 } 2728 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) && 2729 itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) { 2730 psignal(p, SIGPROF); 2731 map |= (1 << id); 2732 } 2733 } 2734 if (stathz == 0) { 2735 forwarded_statclock( id, pscnt, &map); 2736 } 2737 } 2738 if (map != 0) { 2739 checkstate_need_ast |= map; 2740 selected_apic_ipi(map, XCPUAST_OFFSET, APIC_DELMODE_FIXED); 2741 i = 0; 2742 while ((checkstate_need_ast & map) != 0) { 2743 /* spin */ 2744 i++; 2745 if (i > 100000) { 2746 #ifdef BETTER_CLOCK_DIAGNOSTIC 2747 printf("forward_hardclock: dropped ast 0x%x\n", 2748 checkstate_need_ast & map); 2749 #endif 2750 break; 2751 } 2752 } 2753 } 2754 } 2755 2756 #endif /* BETTER_CLOCK */ 2757 2758 void 2759 forward_signal(struct proc *p) 2760 { 2761 int map; 2762 int id; 2763 int i; 2764 2765 /* Kludge. We don't yet have separate locks for the interrupts 2766 * and the kernel. This means that we cannot let the other processors 2767 * handle complex interrupts while inhibiting them from entering 2768 * the kernel in a non-interrupt context. 2769 * 2770 * What we can do, without changing the locking mechanisms yet, 2771 * is letting the other processors handle a very simple interrupt 2772 * (wich determines the processor states), and do the main 2773 * work ourself. 2774 */ 2775 2776 if (!smp_started || !invltlb_ok || cold || panicstr) 2777 return; 2778 if (!forward_signal_enabled) 2779 return; 2780 while (1) { 2781 if (p->p_stat != SRUN) 2782 return; 2783 id = p->p_oncpu; 2784 if (id == 0xff) 2785 return; 2786 map = (1<<id); 2787 checkstate_need_ast |= map; 2788 selected_apic_ipi(map, XCPUAST_OFFSET, APIC_DELMODE_FIXED); 2789 i = 0; 2790 while ((checkstate_need_ast & map) != 0) { 2791 /* spin */ 2792 i++; 2793 if (i > 100000) { 2794 #if 0 2795 printf("forward_signal: dropped ast 0x%x\n", 2796 checkstate_need_ast & map); 2797 #endif 2798 break; 2799 } 2800 } 2801 if (id == p->p_oncpu) 2802 return; 2803 } 2804 } 2805 2806 void 2807 forward_roundrobin(void) 2808 { 2809 u_int map; 2810 int i; 2811 2812 if (!smp_started || !invltlb_ok || cold || panicstr) 2813 return; 2814 if (!forward_roundrobin_enabled) 2815 return; 2816 resched_cpus |= PCPU_GET(other_cpus); 2817 map = PCPU_GET(other_cpus) & ~stopped_cpus ; 2818 #if 1 2819 selected_apic_ipi(map, XCPUAST_OFFSET, APIC_DELMODE_FIXED); 2820 #else 2821 (void) all_but_self_ipi(XCPUAST_OFFSET); 2822 #endif 2823 i = 0; 2824 while ((checkstate_need_ast & map) != 0) { 2825 /* spin */ 2826 i++; 2827 if (i > 100000) { 2828 #if 0 2829 printf("forward_roundrobin: dropped ast 0x%x\n", 2830 checkstate_need_ast & map); 2831 #endif 2832 break; 2833 } 2834 } 2835 } 2836 2837 2838 #ifdef APIC_INTR_REORDER 2839 /* 2840 * Maintain mapping from softintr vector to isr bit in local apic. 2841 */ 2842 void 2843 set_lapic_isrloc(int intr, int vector) 2844 { 2845 if (intr < 0 || intr > 32) 2846 panic("set_apic_isrloc: bad intr argument: %d",intr); 2847 if (vector < ICU_OFFSET || vector > 255) 2848 panic("set_apic_isrloc: bad vector argument: %d",vector); 2849 apic_isrbit_location[intr].location = &lapic.isr0 + ((vector>>5)<<2); 2850 apic_isrbit_location[intr].bit = (1<<(vector & 31)); 2851 } 2852 #endif 2853 2854 /* 2855 * All-CPU rendezvous. CPUs are signalled, all execute the setup function 2856 * (if specified), rendezvous, execute the action function (if specified), 2857 * rendezvous again, execute the teardown function (if specified), and then 2858 * resume. 2859 * 2860 * Note that the supplied external functions _must_ be reentrant and aware 2861 * that they are running in parallel and in an unknown lock context. 2862 */ 2863 static void (*smp_rv_setup_func)(void *arg); 2864 static void (*smp_rv_action_func)(void *arg); 2865 static void (*smp_rv_teardown_func)(void *arg); 2866 static void *smp_rv_func_arg; 2867 static volatile int smp_rv_waiters[2]; 2868 2869 void 2870 smp_rendezvous_action(void) 2871 { 2872 /* setup function */ 2873 if (smp_rv_setup_func != NULL) 2874 smp_rv_setup_func(smp_rv_func_arg); 2875 /* spin on entry rendezvous */ 2876 atomic_add_int(&smp_rv_waiters[0], 1); 2877 while (smp_rv_waiters[0] < mp_ncpus) 2878 ; 2879 /* action function */ 2880 if (smp_rv_action_func != NULL) 2881 smp_rv_action_func(smp_rv_func_arg); 2882 /* spin on exit rendezvous */ 2883 atomic_add_int(&smp_rv_waiters[1], 1); 2884 while (smp_rv_waiters[1] < mp_ncpus) 2885 ; 2886 /* teardown function */ 2887 if (smp_rv_teardown_func != NULL) 2888 smp_rv_teardown_func(smp_rv_func_arg); 2889 } 2890 2891 void 2892 smp_rendezvous(void (* setup_func)(void *), 2893 void (* action_func)(void *), 2894 void (* teardown_func)(void *), 2895 void *arg) 2896 { 2897 u_int efl; 2898 2899 /* obtain rendezvous lock */ 2900 s_lock(&smp_rv_lock); /* XXX sleep here? NOWAIT flag? */ 2901 2902 /* set static function pointers */ 2903 smp_rv_setup_func = setup_func; 2904 smp_rv_action_func = action_func; 2905 smp_rv_teardown_func = teardown_func; 2906 smp_rv_func_arg = arg; 2907 smp_rv_waiters[0] = 0; 2908 smp_rv_waiters[1] = 0; 2909 2910 /* disable interrupts on this CPU, save interrupt status */ 2911 efl = read_eflags(); 2912 write_eflags(efl & ~PSL_I); 2913 2914 /* signal other processors, which will enter the IPI with interrupts off */ 2915 all_but_self_ipi(XRENDEZVOUS_OFFSET); 2916 2917 /* call executor function */ 2918 smp_rendezvous_action(); 2919 2920 /* restore interrupt flag */ 2921 write_eflags(efl); 2922 2923 /* release lock */ 2924 s_unlock(&smp_rv_lock); 2925 } 2926 2927 void 2928 release_aps(void *dummy __unused) 2929 { 2930 s_unlock(&ap_boot_lock); 2931 } 2932 2933 SYSINIT(start_aps, SI_SUB_SMP, SI_ORDER_FIRST, release_aps, NULL); 2934