cpu.c revision 1.42.2.3
1 /* $NetBSD: cpu.c,v 1.42.2.3 2010/07/03 01:19:30 rmind Exp $ */ 2 /* NetBSD: cpu.c,v 1.18 2004/02/20 17:35:01 yamt Exp */ 3 4 /*- 5 * Copyright (c) 2000 The NetBSD Foundation, Inc. 6 * Copyright (c) 2002, 2006, 2007 YAMAMOTO Takashi, 7 * All rights reserved. 8 * 9 * This code is derived from software contributed to The NetBSD Foundation 10 * by RedBack Networks Inc. 11 * 12 * Author: Bill Sommerfeld 13 * 14 * Redistribution and use in source and binary forms, with or without 15 * modification, are permitted provided that the following conditions 16 * are met: 17 * 1. Redistributions of source code must retain the above copyright 18 * notice, this list of conditions and the following disclaimer. 19 * 2. Redistributions in binary form must reproduce the above copyright 20 * notice, this list of conditions and the following disclaimer in the 21 * documentation and/or other materials provided with the distribution. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 24 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 25 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 26 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 27 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 28 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 29 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 30 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 31 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 32 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 33 * POSSIBILITY OF SUCH DAMAGE. 34 */ 35 36 /* 37 * Copyright (c) 1999 Stefan Grefen 38 * 39 * Redistribution and use in source and binary forms, with or without 40 * modification, are permitted provided that the following conditions 41 * are met: 42 * 1. Redistributions of source code must retain the above copyright 43 * notice, this list of conditions and the following disclaimer. 44 * 2. Redistributions in binary form must reproduce the above copyright 45 * notice, this list of conditions and the following disclaimer in the 46 * documentation and/or other materials provided with the distribution. 47 * 3. All advertising materials mentioning features or use of this software 48 * must display the following acknowledgement: 49 * This product includes software developed by the NetBSD 50 * Foundation, Inc. and its contributors. 51 * 4. Neither the name of The NetBSD Foundation nor the names of its 52 * contributors may be used to endorse or promote products derived 53 * from this software without specific prior written permission. 54 * 55 * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND ANY 56 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 57 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 58 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR AND CONTRIBUTORS BE LIABLE 59 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 60 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 61 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 62 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 63 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 64 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 65 * SUCH DAMAGE. 66 */ 67 68 #include <sys/cdefs.h> 69 __KERNEL_RCSID(0, "$NetBSD: cpu.c,v 1.42.2.3 2010/07/03 01:19:30 rmind Exp $"); 70 71 #include "opt_ddb.h" 72 #include "opt_multiprocessor.h" 73 #include "opt_mpbios.h" /* for MPDEBUG */ 74 #include "opt_mtrr.h" 75 #include "opt_xen.h" 76 77 #include "lapic.h" 78 #include "ioapic.h" 79 80 #include <sys/param.h> 81 #include <sys/proc.h> 82 #include <sys/systm.h> 83 #include <sys/device.h> 84 #include <sys/kmem.h> 85 #include <sys/cpu.h> 86 #include <sys/atomic.h> 87 #include <sys/reboot.h> 88 89 #include <uvm/uvm_extern.h> 90 91 #include <machine/cpufunc.h> 92 #include <machine/cpuvar.h> 93 #include <machine/pmap.h> 94 #include <machine/vmparam.h> 95 #include <machine/mpbiosvar.h> 96 #include <machine/pcb.h> 97 #include <machine/specialreg.h> 98 #include <machine/segments.h> 99 #include <machine/gdt.h> 100 #include <machine/mtrr.h> 101 #include <machine/pio.h> 102 103 #include <xen/vcpuvar.h> 104 105 #if NLAPIC > 0 106 #include <machine/apicvar.h> 107 #include <machine/i82489reg.h> 108 #include <machine/i82489var.h> 109 #endif 110 111 #include <dev/ic/mc146818reg.h> 112 #include <dev/isa/isareg.h> 113 114 #if MAXCPUS > 32 115 #error cpu_info contains 32bit bitmasks 116 #endif 117 118 int cpu_match(device_t, cfdata_t, void *); 119 void cpu_attach(device_t, device_t, void *); 120 int vcpu_match(device_t, cfdata_t, void *); 121 void vcpu_attach(device_t, device_t, void *); 122 void cpu_attach_common(device_t, device_t, void *); 123 void cpu_offline_md(void); 124 125 struct cpu_softc { 126 device_t sc_dev; /* device tree glue */ 127 struct cpu_info *sc_info; /* pointer to CPU info */ 128 bool sc_wasonline; 129 }; 130 131 int mp_cpu_start(struct cpu_info *, paddr_t); 132 void mp_cpu_start_cleanup(struct cpu_info *); 133 const struct cpu_functions mp_cpu_funcs = { mp_cpu_start, NULL, 134 mp_cpu_start_cleanup }; 135 136 CFATTACH_DECL_NEW(cpu, sizeof(struct cpu_softc), 137 cpu_match, cpu_attach, NULL, NULL); 138 CFATTACH_DECL_NEW(vcpu, sizeof(struct cpu_softc), 139 vcpu_match, vcpu_attach, NULL, NULL); 140 141 /* 142 * Statically-allocated CPU info for the primary CPU (or the only 143 * CPU, on uniprocessors). The CPU info list is initialized to 144 * point at it. 145 */ 146 #ifdef TRAPLOG 147 #include <machine/tlog.h> 148 struct tlog tlog_primary; 149 #endif 150 struct cpu_info cpu_info_primary __aligned(CACHE_LINE_SIZE) = { 151 .ci_dev = 0, 152 .ci_self = &cpu_info_primary, 153 .ci_idepth = -1, 154 .ci_curlwp = &lwp0, 155 .ci_curldt = -1, 156 #ifdef TRAPLOG 157 .ci_tlog = &tlog_primary, 158 #endif 159 160 }; 161 struct cpu_info phycpu_info_primary __aligned(CACHE_LINE_SIZE) = { 162 .ci_dev = 0, 163 .ci_self = &phycpu_info_primary, 164 }; 165 166 struct cpu_info *cpu_info_list = &cpu_info_primary; 167 struct cpu_info *phycpu_info_list = &phycpu_info_primary; 168 169 static void cpu_set_tss_gates(struct cpu_info *ci); 170 171 uint32_t cpus_attached = 0; 172 uint32_t cpus_running = 0; 173 174 uint32_t phycpus_attached = 0; 175 uint32_t phycpus_running = 0; 176 177 uint32_t cpu_feature[5]; /* X86 CPUID feature bits 178 * [0] basic features %edx 179 * [1] basic features %ecx 180 * [2] extended features %edx 181 * [3] extended features %ecx 182 * [4] VIA padlock features 183 */ 184 185 bool x86_mp_online; 186 paddr_t mp_trampoline_paddr = MP_TRAMPOLINE; 187 188 #if defined(MULTIPROCESSOR) 189 void cpu_hatch(void *); 190 static void cpu_boot_secondary(struct cpu_info *ci); 191 static void cpu_start_secondary(struct cpu_info *ci); 192 static void cpu_copy_trampoline(void); 193 194 /* 195 * Runs once per boot once multiprocessor goo has been detected and 196 * the local APIC on the boot processor has been mapped. 197 * 198 * Called from lapic_boot_init() (from mpbios_scan()). 199 */ 200 void 201 cpu_init_first(void) 202 { 203 204 cpu_info_primary.ci_cpuid = lapic_cpu_number(); 205 cpu_copy_trampoline(); 206 } 207 #endif /* MULTIPROCESSOR */ 208 209 int 210 cpu_match(device_t parent, cfdata_t match, void *aux) 211 { 212 213 return 1; 214 } 215 216 void 217 cpu_attach(device_t parent, device_t self, void *aux) 218 { 219 struct cpu_softc *sc = device_private(self); 220 struct cpu_attach_args *caa = aux; 221 struct cpu_info *ci; 222 uintptr_t ptr; 223 static bool again = false; 224 225 sc->sc_dev = self; 226 227 if (phycpus_attached == ~0) { 228 aprint_error(": increase MAXCPUS\n"); 229 return; 230 } 231 232 /* 233 * If we're an Application Processor, allocate a cpu_info 234 * structure, otherwise use the primary's. 235 */ 236 if (caa->cpu_role == CPU_ROLE_AP) { 237 if ((boothowto & RB_MD1) != 0) { 238 aprint_error(": multiprocessor boot disabled\n"); 239 if (!pmf_device_register(self, NULL, NULL)) 240 aprint_error_dev(self, 241 "couldn't establish power handler\n"); 242 return; 243 } 244 aprint_naive(": Application Processor\n"); 245 ptr = (uintptr_t)kmem_zalloc(sizeof(*ci) + CACHE_LINE_SIZE - 1, 246 KM_SLEEP); 247 ci = (struct cpu_info *)roundup2(ptr, CACHE_LINE_SIZE); 248 ci->ci_curldt = -1; 249 } else { 250 aprint_naive(": %s Processor\n", 251 caa->cpu_role == CPU_ROLE_SP ? "Single" : "Boot"); 252 ci = &phycpu_info_primary; 253 } 254 255 ci->ci_self = ci; 256 sc->sc_info = ci; 257 258 ci->ci_dev = self; 259 ci->ci_cpuid = caa->cpu_number; 260 ci->ci_vcpu = NULL; 261 262 /* 263 * Boot processor may not be attached first, but the below 264 * must be done to allow booting other processors. 265 */ 266 if (!again) { 267 atomic_or_32(&ci->ci_flags, CPUF_PRESENT | CPUF_PRIMARY); 268 /* Basic init */ 269 again = true; 270 } 271 272 printf(": "); 273 switch (caa->cpu_role) { 274 case CPU_ROLE_SP: 275 printf("(uniprocessor)\n"); 276 atomic_or_32(&ci->ci_flags, CPUF_SP); 277 break; 278 279 case CPU_ROLE_BP: 280 printf("(boot processor)\n"); 281 atomic_or_32(&ci->ci_flags, CPUF_BSP); 282 break; 283 284 case CPU_ROLE_AP: 285 /* 286 * report on an AP 287 */ 288 printf("(application processor)\n"); 289 if (ci->ci_flags & CPUF_PRESENT) { 290 struct cpu_info *tmp; 291 292 tmp = phycpu_info_list; 293 while (tmp->ci_next) 294 tmp = tmp->ci_next; 295 296 tmp->ci_next = ci; 297 } 298 break; 299 300 default: 301 panic("unknown processor type??\n"); 302 } 303 304 atomic_or_32(&phycpus_attached, ci->ci_cpumask); 305 306 return; 307 } 308 309 int 310 vcpu_match(device_t parent, cfdata_t match, void *aux) 311 { 312 struct vcpu_attach_args *vcaa = aux; 313 314 if (strcmp(vcaa->vcaa_name, match->cf_name) == 0) 315 return 1; 316 return 0; 317 } 318 319 void 320 vcpu_attach(device_t parent, device_t self, void *aux) 321 { 322 struct vcpu_attach_args *vcaa = aux; 323 324 cpu_attach_common(parent, self, &vcaa->vcaa_caa); 325 } 326 327 static void 328 cpu_vm_init(struct cpu_info *ci) 329 { 330 int ncolors = 2, i; 331 332 for (i = CAI_ICACHE; i <= CAI_L2CACHE; i++) { 333 struct x86_cache_info *cai; 334 int tcolors; 335 336 cai = &ci->ci_cinfo[i]; 337 338 tcolors = atop(cai->cai_totalsize); 339 switch(cai->cai_associativity) { 340 case 0xff: 341 tcolors = 1; /* fully associative */ 342 break; 343 case 0: 344 case 1: 345 break; 346 default: 347 tcolors /= cai->cai_associativity; 348 } 349 ncolors = max(ncolors, tcolors); 350 } 351 352 /* 353 * Knowing the size of the largest cache on this CPU, re-color 354 * our pages. 355 */ 356 if (ncolors <= uvmexp.ncolors) 357 return; 358 aprint_debug_dev(ci->ci_dev, "%d page colors\n", ncolors); 359 uvm_page_recolor(ncolors); 360 } 361 362 void 363 cpu_attach_common(device_t parent, device_t self, void *aux) 364 { 365 struct cpu_softc *sc = device_private(self); 366 struct cpu_attach_args *caa = aux; 367 struct cpu_info *ci; 368 uintptr_t ptr; 369 int cpunum = caa->cpu_number; 370 static bool again = false; 371 372 sc->sc_dev = self; 373 374 /* 375 * If we're an Application Processor, allocate a cpu_info 376 * structure, otherwise use the primary's. 377 */ 378 if (caa->cpu_role == CPU_ROLE_AP) { 379 aprint_naive(": Application Processor\n"); 380 ptr = (uintptr_t)kmem_alloc(sizeof(*ci) + CACHE_LINE_SIZE - 1, 381 KM_SLEEP); 382 ci = (struct cpu_info *)roundup2(ptr, CACHE_LINE_SIZE); 383 memset(ci, 0, sizeof(*ci)); 384 #ifdef TRAPLOG 385 ci->ci_tlog_base = kmem_zalloc(sizeof(struct tlog), KM_SLEEP); 386 #endif 387 } else { 388 aprint_naive(": %s Processor\n", 389 caa->cpu_role == CPU_ROLE_SP ? "Single" : "Boot"); 390 ci = &cpu_info_primary; 391 #if NLAPIC > 0 392 if (cpunum != lapic_cpu_number()) { 393 /* XXX should be done earlier */ 394 uint32_t reg; 395 aprint_verbose("\n"); 396 aprint_verbose_dev(self, "running CPU at apic %d" 397 " instead of at expected %d", lapic_cpu_number(), 398 cpunum); 399 reg = i82489_readreg(LAPIC_ID); 400 i82489_writereg(LAPIC_ID, (reg & ~LAPIC_ID_MASK) | 401 (cpunum << LAPIC_ID_SHIFT)); 402 } 403 if (cpunum != lapic_cpu_number()) { 404 aprint_error_dev(self, "unable to reset apic id\n"); 405 } 406 #endif 407 } 408 409 ci->ci_self = ci; 410 sc->sc_info = ci; 411 ci->ci_dev = self; 412 ci->ci_cpuid = cpunum; 413 414 KASSERT(HYPERVISOR_shared_info != NULL); 415 ci->ci_vcpu = &HYPERVISOR_shared_info->vcpu_info[cpunum]; 416 417 ci->ci_func = caa->cpu_func; 418 419 /* Must be called before mi_cpu_attach(). */ 420 cpu_vm_init(ci); 421 422 if (caa->cpu_role == CPU_ROLE_AP) { 423 int error; 424 425 error = mi_cpu_attach(ci); 426 if (error != 0) { 427 aprint_normal("\n"); 428 aprint_error_dev(self, 429 "mi_cpu_attach failed with %d\n", error); 430 return; 431 } 432 } else { 433 KASSERT(ci->ci_data.cpu_idlelwp != NULL); 434 } 435 436 ci->ci_cpumask = (1 << cpu_index(ci)); 437 pmap_reference(pmap_kernel()); 438 ci->ci_pmap = pmap_kernel(); 439 ci->ci_tlbstate = TLBSTATE_STALE; 440 441 /* 442 * Boot processor may not be attached first, but the below 443 * must be done to allow booting other processors. 444 */ 445 if (!again) { 446 atomic_or_32(&ci->ci_flags, CPUF_PRESENT | CPUF_PRIMARY); 447 /* Basic init. */ 448 cpu_intr_init(ci); 449 cpu_get_tsc_freq(ci); 450 cpu_init(ci); 451 cpu_set_tss_gates(ci); 452 #if NLAPIC > 0 453 if (caa->cpu_role != CPU_ROLE_SP) { 454 /* Enable lapic. */ 455 lapic_enable(); 456 lapic_set_lvt(); 457 lapic_calibrate_timer(); 458 } 459 #endif 460 /* Make sure DELAY() is initialized. */ 461 DELAY(1); 462 again = true; 463 } 464 465 /* further PCB init done later. */ 466 467 switch (caa->cpu_role) { 468 case CPU_ROLE_SP: 469 atomic_or_32(&ci->ci_flags, CPUF_SP); 470 cpu_identify(ci); 471 #if 0 472 x86_errata(); 473 #endif 474 x86_cpu_idle_init(); 475 break; 476 477 case CPU_ROLE_BP: 478 atomic_or_32(&ci->ci_flags, CPUF_BSP); 479 cpu_identify(ci); 480 cpu_init(ci); 481 #if 0 482 x86_errata(); 483 #endif 484 x86_cpu_idle_init(); 485 break; 486 487 case CPU_ROLE_AP: 488 /* 489 * report on an AP 490 */ 491 492 #if defined(MULTIPROCESSOR) 493 cpu_intr_init(ci); 494 gdt_alloc_cpu(ci); 495 cpu_set_tss_gates(ci); 496 cpu_start_secondary(ci); 497 if (ci->ci_flags & CPUF_PRESENT) { 498 struct cpu_info *tmp; 499 500 identifycpu(ci); 501 tmp = cpu_info_list; 502 while (tmp->ci_next) 503 tmp = tmp->ci_next; 504 505 tmp->ci_next = ci; 506 } 507 #else 508 aprint_error_dev(self, "not started\n"); 509 #endif 510 break; 511 512 default: 513 aprint_normal("\n"); 514 panic("unknown processor type??\n"); 515 } 516 517 atomic_or_32(&cpus_attached, ci->ci_cpumask); 518 519 #if 0 520 if (!pmf_device_register(self, cpu_suspend, cpu_resume)) 521 aprint_error_dev(self, "couldn't establish power handler\n"); 522 #endif 523 524 #if defined(MULTIPROCESSOR) 525 if (mp_verbose) { 526 struct lwp *l = ci->ci_data.cpu_idlelwp; 527 struct pcb *pcb = lwp_getpcb(l); 528 529 aprint_verbose_dev(self, 530 "idle lwp at %p, idle sp at 0x%p\n", 531 l, 532 #ifdef i386 533 (void *)pcb->pcb_esp 534 #else 535 (void *)pcb->pcb_rsp 536 #endif 537 ); 538 539 } 540 #endif 541 } 542 543 /* 544 * Initialize the processor appropriately. 545 */ 546 547 void 548 cpu_init(struct cpu_info *ci) 549 { 550 551 /* 552 * On a P6 or above, enable global TLB caching if the 553 * hardware supports it. 554 */ 555 if (cpu_feature[0] & CPUID_PGE) 556 lcr4(rcr4() | CR4_PGE); /* enable global TLB caching */ 557 558 #ifdef XXXMTRR 559 /* 560 * On a P6 or above, initialize MTRR's if the hardware supports them. 561 */ 562 if (cpu_feature[0] & CPUID_MTRR) { 563 if ((ci->ci_flags & CPUF_AP) == 0) 564 i686_mtrr_init_first(); 565 mtrr_init_cpu(ci); 566 } 567 #endif 568 /* 569 * If we have FXSAVE/FXRESTOR, use them. 570 */ 571 if (cpu_feature[0] & CPUID_FXSR) { 572 lcr4(rcr4() | CR4_OSFXSR); 573 574 /* 575 * If we have SSE/SSE2, enable XMM exceptions. 576 */ 577 if (cpu_feature[0] & (CPUID_SSE|CPUID_SSE2)) 578 lcr4(rcr4() | CR4_OSXMMEXCPT); 579 } 580 581 atomic_or_32(&cpus_running, ci->ci_cpumask); 582 atomic_or_32(&ci->ci_flags, CPUF_RUNNING); 583 } 584 585 586 #ifdef MULTIPROCESSOR 587 void 588 cpu_boot_secondary_processors(void) 589 { 590 struct cpu_info *ci; 591 u_long i; 592 593 for (i = 0; i < maxcpus; i++) { 594 ci = cpu_lookup(i); 595 if (ci == NULL) 596 continue; 597 if (ci->ci_data.cpu_idlelwp == NULL) 598 continue; 599 if ((ci->ci_flags & CPUF_PRESENT) == 0) 600 continue; 601 if (ci->ci_flags & (CPUF_BSP|CPUF_SP|CPUF_PRIMARY)) 602 continue; 603 cpu_boot_secondary(ci); 604 } 605 606 x86_mp_online = true; 607 } 608 609 static void 610 cpu_init_idle_lwp(struct cpu_info *ci) 611 { 612 struct lwp *l = ci->ci_data.cpu_idlelwp; 613 struct pcb *pcb = lwp_getpcb(l); 614 615 pcb->pcb_cr0 = rcr0(); 616 } 617 618 void 619 cpu_init_idle_lwps(void) 620 { 621 struct cpu_info *ci; 622 u_long i; 623 624 for (i = 0; i < maxcpus; i++) { 625 ci = cpu_lookup(i); 626 if (ci == NULL) 627 continue; 628 if (ci->ci_data.cpu_idlelwp == NULL) 629 continue; 630 if ((ci->ci_flags & CPUF_PRESENT) == 0) 631 continue; 632 cpu_init_idle_lwp(ci); 633 } 634 } 635 636 void 637 cpu_start_secondary(struct cpu_info *ci) 638 { 639 int i; 640 struct pmap *kpm = pmap_kernel(); 641 extern uint32_t mp_pdirpa; 642 643 mp_pdirpa = kpm->pm_pdirpa; /* XXX move elsewhere, not per CPU. */ 644 645 atomic_or_32(&ci->ci_flags, CPUF_AP); 646 647 aprint_debug_dev(ci->ci_dev, "starting\n"); 648 649 ci->ci_curlwp = ci->ci_data.cpu_idlelwp; 650 if (CPU_STARTUP(ci, mp_trampoline_paddr) != 0) 651 return; 652 653 /* 654 * wait for it to become ready 655 */ 656 for (i = 100000; (!(ci->ci_flags & CPUF_PRESENT)) && i > 0; i--) { 657 #ifdef MPDEBUG 658 extern int cpu_trace[3]; 659 static int otrace[3]; 660 if (memcmp(otrace, cpu_trace, sizeof(otrace)) != 0) { 661 aprint_debug_dev(ci->ci_dev, "trace %02x %02x %02x\n", 662 cpu_trace[0], cpu_trace[1], cpu_trace[2]); 663 memcpy(otrace, cpu_trace, sizeof(otrace)); 664 } 665 #endif 666 delay(10); 667 } 668 if ((ci->ci_flags & CPUF_PRESENT) == 0) { 669 aprint_error_dev(ci->ci_dev, "failed to become ready\n"); 670 #if defined(MPDEBUG) && defined(DDB) 671 printf("dropping into debugger; continue from here to resume boot\n"); 672 Debugger(); 673 #endif 674 } 675 676 CPU_START_CLEANUP(ci); 677 } 678 679 void 680 cpu_boot_secondary(struct cpu_info *ci) 681 { 682 int i; 683 684 atomic_or_32(&ci->ci_flags, CPUF_GO); 685 for (i = 100000; (!(ci->ci_flags & CPUF_RUNNING)) && i > 0; i--) { 686 delay(10); 687 } 688 if ((ci->ci_flags & CPUF_RUNNING) == 0) { 689 aprint_error_dev(ci->ci_dev, "CPU failed to start\n"); 690 #if defined(MPDEBUG) && defined(DDB) 691 printf("dropping into debugger; continue from here to resume boot\n"); 692 Debugger(); 693 #endif 694 } 695 } 696 697 /* 698 * The CPU ends up here when its ready to run 699 * This is called from code in mptramp.s; at this point, we are running 700 * in the idle pcb/idle stack of the new CPU. When this function returns, 701 * this processor will enter the idle loop and start looking for work. 702 * 703 * XXX should share some of this with init386 in machdep.c 704 */ 705 void 706 cpu_hatch(void *v) 707 { 708 struct cpu_info *ci = (struct cpu_info *)v; 709 struct pcb *pcb; 710 int s, i; 711 712 cpu_probe(ci); 713 714 cpu_feature[0] &= ~CPUID_FEAT_BLACKLIST; 715 cpu_feature[2] &= ~CPUID_FEAT_EXT_BLACKLIST; 716 717 cpu_init_msrs(ci, true); 718 719 KDASSERT((ci->ci_flags & CPUF_PRESENT) == 0); 720 atomic_or_32(&ci->ci_flags, CPUF_PRESENT); 721 while ((ci->ci_flags & CPUF_GO) == 0) { 722 /* Don't use delay, boot CPU may be patching the text. */ 723 for (i = 10000; i != 0; i--) 724 x86_pause(); 725 } 726 727 /* Because the text may have been patched in x86_patch(). */ 728 wbinvd(); 729 x86_flush(); 730 731 KASSERT((ci->ci_flags & CPUF_RUNNING) == 0); 732 733 pcb = lwp_getpcb(curlwp); 734 lcr3(pmap_kernel()->pm_pdirpa); 735 pcb->pcb_cr3 = pmap_kernel()->pm_pdirpa; 736 pcb = lwp_getpcb(ci->ci_data.cpu_idlelwp); 737 lcr0(pcb->pcb_cr0); 738 739 cpu_init_idt(); 740 gdt_init_cpu(ci); 741 lapic_enable(); 742 lapic_set_lvt(); 743 lapic_initclocks(); 744 745 #ifdef i386 746 npxinit(ci); 747 #else 748 fpuinit(ci); 749 #endif 750 751 lldt(GSEL(GLDT_SEL, SEL_KPL)); 752 ltr(ci->ci_tss_sel); 753 754 cpu_init(ci); 755 cpu_get_tsc_freq(ci); 756 757 s = splhigh(); 758 #ifdef i386 759 lapic_tpr = 0; 760 #else 761 lcr8(0); 762 #endif 763 x86_enable_intr(); 764 splx(s); 765 #if 0 766 x86_errata(); 767 #endif 768 769 aprint_debug_dev(ci->ci_dev, "CPU %ld running\n", 770 (long)ci->ci_cpuid); 771 } 772 773 #if defined(DDB) 774 775 #include <ddb/db_output.h> 776 #include <machine/db_machdep.h> 777 778 /* 779 * Dump CPU information from ddb. 780 */ 781 void 782 cpu_debug_dump(void) 783 { 784 struct cpu_info *ci; 785 CPU_INFO_ITERATOR cii; 786 787 db_printf("addr dev id flags ipis curlwp fpcurlwp\n"); 788 for (CPU_INFO_FOREACH(cii, ci)) { 789 db_printf("%p %s %ld %x %x %10p %10p\n", 790 ci, 791 ci->ci_dev == NULL ? "BOOT" : device_xname(ci->ci_dev), 792 (long)ci->ci_cpuid, 793 ci->ci_flags, ci->ci_ipis, 794 ci->ci_curlwp, 795 ci->ci_fpcurlwp); 796 } 797 } 798 #endif /* DDB */ 799 800 static void 801 cpu_copy_trampoline(void) 802 { 803 /* 804 * Copy boot code. 805 */ 806 extern u_char cpu_spinup_trampoline[]; 807 extern u_char cpu_spinup_trampoline_end[]; 808 809 vaddr_t mp_trampoline_vaddr; 810 811 mp_trampoline_vaddr = uvm_km_alloc(kernel_map, PAGE_SIZE, 0, 812 UVM_KMF_VAONLY); 813 814 pmap_kenter_pa(mp_trampoline_vaddr, mp_trampoline_paddr, 815 VM_PROT_READ | VM_PROT_WRITE, 0); 816 pmap_update(pmap_kernel()); 817 memcpy((void *)mp_trampoline_vaddr, 818 cpu_spinup_trampoline, 819 cpu_spinup_trampoline_end - cpu_spinup_trampoline); 820 821 pmap_kremove(mp_trampoline_vaddr, PAGE_SIZE); 822 pmap_update(pmap_kernel()); 823 uvm_km_free(kernel_map, mp_trampoline_vaddr, PAGE_SIZE, UVM_KMF_VAONLY); 824 } 825 826 #endif /* MULTIPROCESSOR */ 827 828 #ifdef i386 829 #if 0 830 static void 831 tss_init(struct i386tss *tss, void *stack, void *func) 832 { 833 memset(tss, 0, sizeof *tss); 834 tss->tss_esp0 = tss->tss_esp = (int)((char *)stack + USPACE - 16); 835 tss->tss_ss0 = GSEL(GDATA_SEL, SEL_KPL); 836 tss->__tss_cs = GSEL(GCODE_SEL, SEL_KPL); 837 tss->tss_fs = GSEL(GCPU_SEL, SEL_KPL); 838 tss->tss_gs = tss->__tss_es = tss->__tss_ds = 839 tss->__tss_ss = GSEL(GDATA_SEL, SEL_KPL); 840 tss->tss_cr3 = pmap_kernel()->pm_pdirpa; 841 tss->tss_esp = (int)((char *)stack + USPACE - 16); 842 tss->tss_ldt = GSEL(GLDT_SEL, SEL_KPL); 843 tss->__tss_eflags = PSL_MBO | PSL_NT; /* XXX not needed? */ 844 tss->__tss_eip = (int)func; 845 } 846 #endif 847 848 /* XXX */ 849 #define IDTVEC(name) __CONCAT(X, name) 850 typedef void (vector)(void); 851 extern vector IDTVEC(tss_trap08); 852 #ifdef DDB 853 extern vector Xintrddbipi; 854 extern int ddb_vec; 855 #endif 856 857 static void 858 cpu_set_tss_gates(struct cpu_info *ci) 859 { 860 #if 0 861 struct segment_descriptor sd; 862 863 ci->ci_doubleflt_stack = (char *)uvm_km_alloc(kernel_map, USPACE, 0, 864 UVM_KMF_WIRED); 865 tss_init(&ci->ci_doubleflt_tss, ci->ci_doubleflt_stack, 866 IDTVEC(tss_trap08)); 867 setsegment(&sd, &ci->ci_doubleflt_tss, sizeof(struct i386tss) - 1, 868 SDT_SYS386TSS, SEL_KPL, 0, 0); 869 ci->ci_gdt[GTRAPTSS_SEL].sd = sd; 870 setgate(&idt[8], NULL, 0, SDT_SYSTASKGT, SEL_KPL, 871 GSEL(GTRAPTSS_SEL, SEL_KPL)); 872 #endif 873 874 #if defined(DDB) && defined(MULTIPROCESSOR) 875 /* 876 * Set up separate handler for the DDB IPI, so that it doesn't 877 * stomp on a possibly corrupted stack. 878 * 879 * XXX overwriting the gate set in db_machine_init. 880 * Should rearrange the code so that it's set only once. 881 */ 882 ci->ci_ddbipi_stack = (char *)uvm_km_alloc(kernel_map, USPACE, 0, 883 UVM_KMF_WIRED); 884 tss_init(&ci->ci_ddbipi_tss, ci->ci_ddbipi_stack, 885 Xintrddbipi); 886 887 setsegment(&sd, &ci->ci_ddbipi_tss, sizeof(struct i386tss) - 1, 888 SDT_SYS386TSS, SEL_KPL, 0, 0); 889 ci->ci_gdt[GIPITSS_SEL].sd = sd; 890 891 setgate(&idt[ddb_vec], NULL, 0, SDT_SYSTASKGT, SEL_KPL, 892 GSEL(GIPITSS_SEL, SEL_KPL)); 893 #endif 894 } 895 #else 896 static void 897 cpu_set_tss_gates(struct cpu_info *ci) 898 { 899 900 } 901 #endif /* i386 */ 902 903 int 904 mp_cpu_start(struct cpu_info *ci, paddr_t target) 905 { 906 #if 0 907 #if NLAPIC > 0 908 int error; 909 #endif 910 unsigned short dwordptr[2]; 911 912 /* 913 * Bootstrap code must be addressable in real mode 914 * and it must be page aligned. 915 */ 916 KASSERT(target < 0x10000 && target % PAGE_SIZE == 0); 917 918 /* 919 * "The BSP must initialize CMOS shutdown code to 0Ah ..." 920 */ 921 922 outb(IO_RTC, NVRAM_RESET); 923 outb(IO_RTC+1, NVRAM_RESET_JUMP); 924 925 /* 926 * "and the warm reset vector (DWORD based at 40:67) to point 927 * to the AP startup code ..." 928 */ 929 930 dwordptr[0] = 0; 931 dwordptr[1] = target >> 4; 932 933 pmap_kenter_pa (0, 0, VM_PROT_READ|VM_PROT_WRITE, 0); 934 pmap_update(pmap_kernel()); 935 936 memcpy ((uint8_t *) 0x467, dwordptr, 4); 937 938 pmap_kremove (0, PAGE_SIZE); 939 pmap_update(pmap_kernel()); 940 941 #if NLAPIC > 0 942 /* 943 * ... prior to executing the following sequence:" 944 */ 945 946 if (ci->ci_flags & CPUF_AP) { 947 if ((error = x86_ipi_init(ci->ci_cpuid)) != 0) 948 return error; 949 950 delay(10000); 951 952 if (cpu_feature & CPUID_APIC) { 953 error = x86_ipi_init(ci->ci_cpuid); 954 if (error != 0) { 955 aprint_error_dev(ci->ci_dev, "%s: IPI not taken (1)\n", 956 __func__); 957 return error; 958 } 959 960 delay(10000); 961 962 error = x86_ipi(target / PAGE_SIZE, ci->ci_cpuid, 963 LAPIC_DLMODE_STARTUP); 964 if (error != 0) { 965 aprint_error_dev(ci->ci_dev, "%s: IPI not taken (2)\n", 966 __func__); 967 return error; 968 } 969 delay(200); 970 971 error = x86_ipi(target / PAGE_SIZE, ci->ci_cpuid, 972 LAPIC_DLMODE_STARTUP); 973 if (error != 0) { 974 aprint_error_dev(ci->ci_dev, "%s: IPI not taken ((3)\n", 975 __func__); 976 return error; 977 } 978 delay(200); 979 } 980 } 981 #endif 982 #endif /* 0 */ 983 return 0; 984 } 985 986 void 987 mp_cpu_start_cleanup(struct cpu_info *ci) 988 { 989 #if 0 990 /* 991 * Ensure the NVRAM reset byte contains something vaguely sane. 992 */ 993 994 outb(IO_RTC, NVRAM_RESET); 995 outb(IO_RTC+1, NVRAM_RESET_RST); 996 #endif 997 } 998 999 void 1000 cpu_init_msrs(struct cpu_info *ci, bool full) 1001 { 1002 #ifdef __x86_64__ 1003 if (full) { 1004 HYPERVISOR_set_segment_base (SEGBASE_FS, 0); 1005 HYPERVISOR_set_segment_base (SEGBASE_GS_KERNEL, (uint64_t) ci); 1006 HYPERVISOR_set_segment_base (SEGBASE_GS_USER, 0); 1007 } 1008 #endif /* __x86_64__ */ 1009 1010 if (cpu_feature[2] & CPUID_NOX) 1011 wrmsr(MSR_EFER, rdmsr(MSR_EFER) | EFER_NXE); 1012 } 1013 1014 void 1015 cpu_offline_md(void) 1016 { 1017 int s; 1018 1019 s = splhigh(); 1020 #ifdef __i386__ 1021 npxsave_cpu(true); 1022 #else 1023 fpusave_cpu(true); 1024 #endif 1025 splx(s); 1026 } 1027 1028 #if 0 1029 /* XXX joerg restructure and restart CPUs individually */ 1030 static bool 1031 cpu_suspend(device_t dv, const pmf_qual_t *qual) 1032 { 1033 struct cpu_softc *sc = device_private(dv); 1034 struct cpu_info *ci = sc->sc_info; 1035 int err; 1036 1037 if (ci->ci_flags & CPUF_PRIMARY) 1038 return true; 1039 if (ci->ci_data.cpu_idlelwp == NULL) 1040 return true; 1041 if ((ci->ci_flags & CPUF_PRESENT) == 0) 1042 return true; 1043 1044 sc->sc_wasonline = !(ci->ci_schedstate.spc_flags & SPCF_OFFLINE); 1045 1046 if (sc->sc_wasonline) { 1047 mutex_enter(&cpu_lock); 1048 err = cpu_setstate(ci, false); 1049 mutex_exit(&cpu_lock); 1050 1051 if (err) 1052 return false; 1053 } 1054 1055 return true; 1056 } 1057 1058 static bool 1059 cpu_resume(device_t dv, const pmf_qual_t *qual) 1060 { 1061 struct cpu_softc *sc = device_private(dv); 1062 struct cpu_info *ci = sc->sc_info; 1063 int err = 0; 1064 1065 if (ci->ci_flags & CPUF_PRIMARY) 1066 return true; 1067 if (ci->ci_data.cpu_idlelwp == NULL) 1068 return true; 1069 if ((ci->ci_flags & CPUF_PRESENT) == 0) 1070 return true; 1071 1072 if (sc->sc_wasonline) { 1073 mutex_enter(&cpu_lock); 1074 err = cpu_setstate(ci, true); 1075 mutex_exit(&cpu_lock); 1076 } 1077 1078 return err == 0; 1079 } 1080 #endif 1081 1082 void 1083 cpu_get_tsc_freq(struct cpu_info *ci) 1084 { 1085 const volatile vcpu_time_info_t *tinfo = &ci->ci_vcpu->time; 1086 delay(1000000); 1087 uint64_t freq = 1000000000ULL << 32; 1088 freq = freq / (uint64_t)tinfo->tsc_to_system_mul; 1089 if ( tinfo->tsc_shift < 0 ) 1090 freq = freq << -tinfo->tsc_shift; 1091 else 1092 freq = freq >> tinfo->tsc_shift; 1093 ci->ci_data.cpu_cc_freq = freq; 1094 } 1095 1096 void 1097 x86_cpu_idle_xen(void) 1098 { 1099 struct cpu_info *ci = curcpu(); 1100 1101 KASSERT(ci->ci_ilevel == IPL_NONE); 1102 1103 x86_disable_intr(); 1104 if (!__predict_false(ci->ci_want_resched)) { 1105 idle_block(); 1106 } else { 1107 x86_enable_intr(); 1108 } 1109 } 1110
Indexes created Fri Oct 03 02:09:57 GMT 2025