1 /* 2 * Local APIC handling, local APIC timers 3 * 4 * (c) 1999, 2000, 2009 Ingo Molnar <mingo@redhat.com> 5 * 6 * Fixes 7 * Maciej W. Rozycki : Bits for genuine 82489DX APICs; 8 * thanks to Eric Gilmore 9 * and Rolf G. Tews 10 * for testing these extensively. 11 * Maciej W. Rozycki : Various updates and fixes. 12 * Mikael Pettersson : Power Management for UP-APIC. 13 * Pavel Machek and 14 * Mikael Pettersson : PM converted to driver model. 15 */ 16 17 #include <linux/perf_event.h> 18 #include <linux/kernel_stat.h> 19 #include <linux/mc146818rtc.h> 20 #include <linux/acpi_pmtmr.h> 21 #include <linux/clockchips.h> 22 #include <linux/interrupt.h> 23 #include <linux/bootmem.h> 24 #include <linux/ftrace.h> 25 #include <linux/ioport.h> 26 #include <linux/module.h> 27 #include <linux/syscore_ops.h> 28 #include <linux/delay.h> 29 #include <linux/timex.h> 30 #include <linux/i8253.h> 31 #include <linux/dmar.h> 32 #include <linux/init.h> 33 #include <linux/cpu.h> 34 #include <linux/dmi.h> 35 #include <linux/smp.h> 36 #include <linux/mm.h> 37 38 #include <asm/trace/irq_vectors.h> 39 #include <asm/irq_remapping.h> 40 #include <asm/perf_event.h> 41 #include <asm/x86_init.h> 42 #include <asm/pgalloc.h> 43 #include <linux/atomic.h> 44 #include <asm/mpspec.h> 45 #include <asm/i8259.h> 46 #include <asm/proto.h> 47 #include <asm/apic.h> 48 #include <asm/io_apic.h> 49 #include <asm/desc.h> 50 #include <asm/hpet.h> 51 #include <asm/idle.h> 52 #include <asm/mtrr.h> 53 #include <asm/time.h> 54 #include <asm/smp.h> 55 #include <asm/mce.h> 56 #include <asm/tsc.h> 57 #include <asm/hypervisor.h> 58 59 unsigned int num_processors; 60 61 unsigned disabled_cpus; 62 63 /* Processor that is doing the boot up */ 64 unsigned int boot_cpu_physical_apicid = -1U; 65 EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid); 66 67 /* 68 * The highest APIC ID seen during enumeration. 69 */ 70 static unsigned int max_physical_apicid; 71 72 /* 73 * Bitmask of physically existing CPUs: 74 */ 75 physid_mask_t phys_cpu_present_map; 76 77 /* 78 * Processor to be disabled specified by kernel parameter 79 * disable_cpu_apicid=<int>, mostly used for the kdump 2nd kernel to 80 * avoid undefined behaviour caused by sending INIT from AP to BSP. 81 */ 82 static unsigned int disabled_cpu_apicid __read_mostly = BAD_APICID; 83 84 /* 85 * Map cpu index to physical APIC ID 86 */ 87 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_cpu_to_apicid, BAD_APICID); 88 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_bios_cpu_apicid, BAD_APICID); 89 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid); 90 EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid); 91 92 #ifdef CONFIG_X86_32 93 94 /* 95 * On x86_32, the mapping between cpu and logical apicid may vary 96 * depending on apic in use. The following early percpu variable is 97 * used for the mapping. This is where the behaviors of x86_64 and 32 98 * actually diverge. Let's keep it ugly for now. 99 */ 100 DEFINE_EARLY_PER_CPU_READ_MOSTLY(int, x86_cpu_to_logical_apicid, BAD_APICID); 101 102 /* Local APIC was disabled by the BIOS and enabled by the kernel */ 103 static int enabled_via_apicbase; 104 105 /* 106 * Handle interrupt mode configuration register (IMCR). 107 * This register controls whether the interrupt signals 108 * that reach the BSP come from the master PIC or from the 109 * local APIC. Before entering Symmetric I/O Mode, either 110 * the BIOS or the operating system must switch out of 111 * PIC Mode by changing the IMCR. 112 */ 113 static inline void imcr_pic_to_apic(void) 114 { 115 /* select IMCR register */ 116 outb(0x70, 0x22); 117 /* NMI and 8259 INTR go through APIC */ 118 outb(0x01, 0x23); 119 } 120 121 static inline void imcr_apic_to_pic(void) 122 { 123 /* select IMCR register */ 124 outb(0x70, 0x22); 125 /* NMI and 8259 INTR go directly to BSP */ 126 outb(0x00, 0x23); 127 } 128 #endif 129 130 /* 131 * Knob to control our willingness to enable the local APIC. 132 * 133 * +1=force-enable 134 */ 135 static int force_enable_local_apic __initdata; 136 137 /* 138 * APIC command line parameters 139 */ 140 static int __init parse_lapic(char *arg) 141 { 142 if (config_enabled(CONFIG_X86_32) && !arg) 143 force_enable_local_apic = 1; 144 else if (arg && !strncmp(arg, "notscdeadline", 13)) 145 setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER); 146 return 0; 147 } 148 early_param("lapic", parse_lapic); 149 150 #ifdef CONFIG_X86_64 151 static int apic_calibrate_pmtmr __initdata; 152 static __init int setup_apicpmtimer(char *s) 153 { 154 apic_calibrate_pmtmr = 1; 155 notsc_setup(NULL); 156 return 0; 157 } 158 __setup("apicpmtimer", setup_apicpmtimer); 159 #endif 160 161 unsigned long mp_lapic_addr; 162 int disable_apic; 163 /* Disable local APIC timer from the kernel commandline or via dmi quirk */ 164 static int disable_apic_timer __initdata; 165 /* Local APIC timer works in C2 */ 166 int local_apic_timer_c2_ok; 167 EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok); 168 169 int first_system_vector = FIRST_SYSTEM_VECTOR; 170 171 /* 172 * Debug level, exported for io_apic.c 173 */ 174 unsigned int apic_verbosity; 175 176 int pic_mode; 177 178 /* Have we found an MP table */ 179 int smp_found_config; 180 181 static struct resource lapic_resource = { 182 .name = "Local APIC", 183 .flags = IORESOURCE_MEM | IORESOURCE_BUSY, 184 }; 185 186 unsigned int lapic_timer_frequency = 0; 187 188 static void apic_pm_activate(void); 189 190 static unsigned long apic_phys; 191 192 /* 193 * Get the LAPIC version 194 */ 195 static inline int lapic_get_version(void) 196 { 197 return GET_APIC_VERSION(apic_read(APIC_LVR)); 198 } 199 200 /* 201 * Check, if the APIC is integrated or a separate chip 202 */ 203 static inline int lapic_is_integrated(void) 204 { 205 #ifdef CONFIG_X86_64 206 return 1; 207 #else 208 return APIC_INTEGRATED(lapic_get_version()); 209 #endif 210 } 211 212 /* 213 * Check, whether this is a modern or a first generation APIC 214 */ 215 static int modern_apic(void) 216 { 217 /* AMD systems use old APIC versions, so check the CPU */ 218 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD && 219 boot_cpu_data.x86 >= 0xf) 220 return 1; 221 return lapic_get_version() >= 0x14; 222 } 223 224 /* 225 * right after this call apic become NOOP driven 226 * so apic->write/read doesn't do anything 227 */ 228 static void __init apic_disable(void) 229 { 230 pr_info("APIC: switched to apic NOOP\n"); 231 apic = &apic_noop; 232 } 233 234 void native_apic_wait_icr_idle(void) 235 { 236 while (apic_read(APIC_ICR) & APIC_ICR_BUSY) 237 cpu_relax(); 238 } 239 240 u32 native_safe_apic_wait_icr_idle(void) 241 { 242 u32 send_status; 243 int timeout; 244 245 timeout = 0; 246 do { 247 send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY; 248 if (!send_status) 249 break; 250 inc_irq_stat(icr_read_retry_count); 251 udelay(100); 252 } while (timeout++ < 1000); 253 254 return send_status; 255 } 256 257 void native_apic_icr_write(u32 low, u32 id) 258 { 259 unsigned long flags; 260 261 local_irq_save(flags); 262 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id)); 263 apic_write(APIC_ICR, low); 264 local_irq_restore(flags); 265 } 266 267 u64 native_apic_icr_read(void) 268 { 269 u32 icr1, icr2; 270 271 icr2 = apic_read(APIC_ICR2); 272 icr1 = apic_read(APIC_ICR); 273 274 return icr1 | ((u64)icr2 << 32); 275 } 276 277 #ifdef CONFIG_X86_32 278 /** 279 * get_physical_broadcast - Get number of physical broadcast IDs 280 */ 281 int get_physical_broadcast(void) 282 { 283 return modern_apic() ? 0xff : 0xf; 284 } 285 #endif 286 287 /** 288 * lapic_get_maxlvt - get the maximum number of local vector table entries 289 */ 290 int lapic_get_maxlvt(void) 291 { 292 unsigned int v; 293 294 v = apic_read(APIC_LVR); 295 /* 296 * - we always have APIC integrated on 64bit mode 297 * - 82489DXs do not report # of LVT entries 298 */ 299 return APIC_INTEGRATED(GET_APIC_VERSION(v)) ? GET_APIC_MAXLVT(v) : 2; 300 } 301 302 /* 303 * Local APIC timer 304 */ 305 306 /* Clock divisor */ 307 #define APIC_DIVISOR 16 308 #define TSC_DIVISOR 32 309 310 /* 311 * This function sets up the local APIC timer, with a timeout of 312 * 'clocks' APIC bus clock. During calibration we actually call 313 * this function twice on the boot CPU, once with a bogus timeout 314 * value, second time for real. The other (noncalibrating) CPUs 315 * call this function only once, with the real, calibrated value. 316 * 317 * We do reads before writes even if unnecessary, to get around the 318 * P5 APIC double write bug. 319 */ 320 static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen) 321 { 322 unsigned int lvtt_value, tmp_value; 323 324 lvtt_value = LOCAL_TIMER_VECTOR; 325 if (!oneshot) 326 lvtt_value |= APIC_LVT_TIMER_PERIODIC; 327 else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) 328 lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE; 329 330 if (!lapic_is_integrated()) 331 lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV); 332 333 if (!irqen) 334 lvtt_value |= APIC_LVT_MASKED; 335 336 apic_write(APIC_LVTT, lvtt_value); 337 338 if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) { 339 /* 340 * See Intel SDM: TSC-Deadline Mode chapter. In xAPIC mode, 341 * writing to the APIC LVTT and TSC_DEADLINE MSR isn't serialized. 342 * According to Intel, MFENCE can do the serialization here. 343 */ 344 asm volatile("mfence" : : : "memory"); 345 346 printk_once(KERN_DEBUG "TSC deadline timer enabled\n"); 347 return; 348 } 349 350 /* 351 * Divide PICLK by 16 352 */ 353 tmp_value = apic_read(APIC_TDCR); 354 apic_write(APIC_TDCR, 355 (tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) | 356 APIC_TDR_DIV_16); 357 358 if (!oneshot) 359 apic_write(APIC_TMICT, clocks / APIC_DIVISOR); 360 } 361 362 /* 363 * Setup extended LVT, AMD specific 364 * 365 * Software should use the LVT offsets the BIOS provides. The offsets 366 * are determined by the subsystems using it like those for MCE 367 * threshold or IBS. On K8 only offset 0 (APIC500) and MCE interrupts 368 * are supported. Beginning with family 10h at least 4 offsets are 369 * available. 370 * 371 * Since the offsets must be consistent for all cores, we keep track 372 * of the LVT offsets in software and reserve the offset for the same 373 * vector also to be used on other cores. An offset is freed by 374 * setting the entry to APIC_EILVT_MASKED. 375 * 376 * If the BIOS is right, there should be no conflicts. Otherwise a 377 * "[Firmware Bug]: ..." error message is generated. However, if 378 * software does not properly determines the offsets, it is not 379 * necessarily a BIOS bug. 380 */ 381 382 static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX]; 383 384 static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new) 385 { 386 return (old & APIC_EILVT_MASKED) 387 || (new == APIC_EILVT_MASKED) 388 || ((new & ~APIC_EILVT_MASKED) == old); 389 } 390 391 static unsigned int reserve_eilvt_offset(int offset, unsigned int new) 392 { 393 unsigned int rsvd, vector; 394 395 if (offset >= APIC_EILVT_NR_MAX) 396 return ~0; 397 398 rsvd = atomic_read(&eilvt_offsets[offset]); 399 do { 400 vector = rsvd & ~APIC_EILVT_MASKED; /* 0: unassigned */ 401 if (vector && !eilvt_entry_is_changeable(vector, new)) 402 /* may not change if vectors are different */ 403 return rsvd; 404 rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new); 405 } while (rsvd != new); 406 407 rsvd &= ~APIC_EILVT_MASKED; 408 if (rsvd && rsvd != vector) 409 pr_info("LVT offset %d assigned for vector 0x%02x\n", 410 offset, rsvd); 411 412 return new; 413 } 414 415 /* 416 * If mask=1, the LVT entry does not generate interrupts while mask=0 417 * enables the vector. See also the BKDGs. Must be called with 418 * preemption disabled. 419 */ 420 421 int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask) 422 { 423 unsigned long reg = APIC_EILVTn(offset); 424 unsigned int new, old, reserved; 425 426 new = (mask << 16) | (msg_type << 8) | vector; 427 old = apic_read(reg); 428 reserved = reserve_eilvt_offset(offset, new); 429 430 if (reserved != new) { 431 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for " 432 "vector 0x%x, but the register is already in use for " 433 "vector 0x%x on another cpu\n", 434 smp_processor_id(), reg, offset, new, reserved); 435 return -EINVAL; 436 } 437 438 if (!eilvt_entry_is_changeable(old, new)) { 439 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for " 440 "vector 0x%x, but the register is already in use for " 441 "vector 0x%x on this cpu\n", 442 smp_processor_id(), reg, offset, new, old); 443 return -EBUSY; 444 } 445 446 apic_write(reg, new); 447 448 return 0; 449 } 450 EXPORT_SYMBOL_GPL(setup_APIC_eilvt); 451 452 /* 453 * Program the next event, relative to now 454 */ 455 static int lapic_next_event(unsigned long delta, 456 struct clock_event_device *evt) 457 { 458 apic_write(APIC_TMICT, delta); 459 return 0; 460 } 461 462 static int lapic_next_deadline(unsigned long delta, 463 struct clock_event_device *evt) 464 { 465 u64 tsc; 466 467 tsc = rdtsc(); 468 wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR)); 469 return 0; 470 } 471 472 static int lapic_timer_shutdown(struct clock_event_device *evt) 473 { 474 unsigned int v; 475 476 /* Lapic used as dummy for broadcast ? */ 477 if (evt->features & CLOCK_EVT_FEAT_DUMMY) 478 return 0; 479 480 v = apic_read(APIC_LVTT); 481 v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR); 482 apic_write(APIC_LVTT, v); 483 apic_write(APIC_TMICT, 0); 484 return 0; 485 } 486 487 static inline int 488 lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot) 489 { 490 /* Lapic used as dummy for broadcast ? */ 491 if (evt->features & CLOCK_EVT_FEAT_DUMMY) 492 return 0; 493 494 __setup_APIC_LVTT(lapic_timer_frequency, oneshot, 1); 495 return 0; 496 } 497 498 static int lapic_timer_set_periodic(struct clock_event_device *evt) 499 { 500 return lapic_timer_set_periodic_oneshot(evt, false); 501 } 502 503 static int lapic_timer_set_oneshot(struct clock_event_device *evt) 504 { 505 return lapic_timer_set_periodic_oneshot(evt, true); 506 } 507 508 /* 509 * Local APIC timer broadcast function 510 */ 511 static void lapic_timer_broadcast(const struct cpumask *mask) 512 { 513 #ifdef CONFIG_SMP 514 apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR); 515 #endif 516 } 517 518 519 /* 520 * The local apic timer can be used for any function which is CPU local. 521 */ 522 static struct clock_event_device lapic_clockevent = { 523 .name = "lapic", 524 .features = CLOCK_EVT_FEAT_PERIODIC | 525 CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP 526 | CLOCK_EVT_FEAT_DUMMY, 527 .shift = 32, 528 .set_state_shutdown = lapic_timer_shutdown, 529 .set_state_periodic = lapic_timer_set_periodic, 530 .set_state_oneshot = lapic_timer_set_oneshot, 531 .set_next_event = lapic_next_event, 532 .broadcast = lapic_timer_broadcast, 533 .rating = 100, 534 .irq = -1, 535 }; 536 static DEFINE_PER_CPU(struct clock_event_device, lapic_events); 537 538 /* 539 * Setup the local APIC timer for this CPU. Copy the initialized values 540 * of the boot CPU and register the clock event in the framework. 541 */ 542 static void setup_APIC_timer(void) 543 { 544 struct clock_event_device *levt = this_cpu_ptr(&lapic_events); 545 546 if (this_cpu_has(X86_FEATURE_ARAT)) { 547 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP; 548 /* Make LAPIC timer preferrable over percpu HPET */ 549 lapic_clockevent.rating = 150; 550 } 551 552 memcpy(levt, &lapic_clockevent, sizeof(*levt)); 553 levt->cpumask = cpumask_of(smp_processor_id()); 554 555 if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) { 556 levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC | 557 CLOCK_EVT_FEAT_DUMMY); 558 levt->set_next_event = lapic_next_deadline; 559 clockevents_config_and_register(levt, 560 (tsc_khz / TSC_DIVISOR) * 1000, 561 0xF, ~0UL); 562 } else 563 clockevents_register_device(levt); 564 } 565 566 /* 567 * In this functions we calibrate APIC bus clocks to the external timer. 568 * 569 * We want to do the calibration only once since we want to have local timer 570 * irqs syncron. CPUs connected by the same APIC bus have the very same bus 571 * frequency. 572 * 573 * This was previously done by reading the PIT/HPET and waiting for a wrap 574 * around to find out, that a tick has elapsed. I have a box, where the PIT 575 * readout is broken, so it never gets out of the wait loop again. This was 576 * also reported by others. 577 * 578 * Monitoring the jiffies value is inaccurate and the clockevents 579 * infrastructure allows us to do a simple substitution of the interrupt 580 * handler. 581 * 582 * The calibration routine also uses the pm_timer when possible, as the PIT 583 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes 584 * back to normal later in the boot process). 585 */ 586 587 #define LAPIC_CAL_LOOPS (HZ/10) 588 589 static __initdata int lapic_cal_loops = -1; 590 static __initdata long lapic_cal_t1, lapic_cal_t2; 591 static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2; 592 static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2; 593 static __initdata unsigned long lapic_cal_j1, lapic_cal_j2; 594 595 /* 596 * Temporary interrupt handler. 597 */ 598 static void __init lapic_cal_handler(struct clock_event_device *dev) 599 { 600 unsigned long long tsc = 0; 601 long tapic = apic_read(APIC_TMCCT); 602 unsigned long pm = acpi_pm_read_early(); 603 604 if (cpu_has_tsc) 605 tsc = rdtsc(); 606 607 switch (lapic_cal_loops++) { 608 case 0: 609 lapic_cal_t1 = tapic; 610 lapic_cal_tsc1 = tsc; 611 lapic_cal_pm1 = pm; 612 lapic_cal_j1 = jiffies; 613 break; 614 615 case LAPIC_CAL_LOOPS: 616 lapic_cal_t2 = tapic; 617 lapic_cal_tsc2 = tsc; 618 if (pm < lapic_cal_pm1) 619 pm += ACPI_PM_OVRRUN; 620 lapic_cal_pm2 = pm; 621 lapic_cal_j2 = jiffies; 622 break; 623 } 624 } 625 626 static int __init 627 calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc) 628 { 629 const long pm_100ms = PMTMR_TICKS_PER_SEC / 10; 630 const long pm_thresh = pm_100ms / 100; 631 unsigned long mult; 632 u64 res; 633 634 #ifndef CONFIG_X86_PM_TIMER 635 return -1; 636 #endif 637 638 apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm); 639 640 /* Check, if the PM timer is available */ 641 if (!deltapm) 642 return -1; 643 644 mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22); 645 646 if (deltapm > (pm_100ms - pm_thresh) && 647 deltapm < (pm_100ms + pm_thresh)) { 648 apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n"); 649 return 0; 650 } 651 652 res = (((u64)deltapm) * mult) >> 22; 653 do_div(res, 1000000); 654 pr_warning("APIC calibration not consistent " 655 "with PM-Timer: %ldms instead of 100ms\n",(long)res); 656 657 /* Correct the lapic counter value */ 658 res = (((u64)(*delta)) * pm_100ms); 659 do_div(res, deltapm); 660 pr_info("APIC delta adjusted to PM-Timer: " 661 "%lu (%ld)\n", (unsigned long)res, *delta); 662 *delta = (long)res; 663 664 /* Correct the tsc counter value */ 665 if (cpu_has_tsc) { 666 res = (((u64)(*deltatsc)) * pm_100ms); 667 do_div(res, deltapm); 668 apic_printk(APIC_VERBOSE, "TSC delta adjusted to " 669 "PM-Timer: %lu (%ld)\n", 670 (unsigned long)res, *deltatsc); 671 *deltatsc = (long)res; 672 } 673 674 return 0; 675 } 676 677 static int __init calibrate_APIC_clock(void) 678 { 679 struct clock_event_device *levt = this_cpu_ptr(&lapic_events); 680 void (*real_handler)(struct clock_event_device *dev); 681 unsigned long deltaj; 682 long delta, deltatsc; 683 int pm_referenced = 0; 684 685 /** 686 * check if lapic timer has already been calibrated by platform 687 * specific routine, such as tsc calibration code. if so, we just fill 688 * in the clockevent structure and return. 689 */ 690 691 if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) { 692 return 0; 693 } else if (lapic_timer_frequency) { 694 apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n", 695 lapic_timer_frequency); 696 lapic_clockevent.mult = div_sc(lapic_timer_frequency/APIC_DIVISOR, 697 TICK_NSEC, lapic_clockevent.shift); 698 lapic_clockevent.max_delta_ns = 699 clockevent_delta2ns(0x7FFFFF, &lapic_clockevent); 700 lapic_clockevent.min_delta_ns = 701 clockevent_delta2ns(0xF, &lapic_clockevent); 702 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY; 703 return 0; 704 } 705 706 apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n" 707 "calibrating APIC timer ...\n"); 708 709 local_irq_disable(); 710 711 /* Replace the global interrupt handler */ 712 real_handler = global_clock_event->event_handler; 713 global_clock_event->event_handler = lapic_cal_handler; 714 715 /* 716 * Setup the APIC counter to maximum. There is no way the lapic 717 * can underflow in the 100ms detection time frame 718 */ 719 __setup_APIC_LVTT(0xffffffff, 0, 0); 720 721 /* Let the interrupts run */ 722 local_irq_enable(); 723 724 while (lapic_cal_loops <= LAPIC_CAL_LOOPS) 725 cpu_relax(); 726 727 local_irq_disable(); 728 729 /* Restore the real event handler */ 730 global_clock_event->event_handler = real_handler; 731 732 /* Build delta t1-t2 as apic timer counts down */ 733 delta = lapic_cal_t1 - lapic_cal_t2; 734 apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta); 735 736 deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1); 737 738 /* we trust the PM based calibration if possible */ 739 pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1, 740 &delta, &deltatsc); 741 742 /* Calculate the scaled math multiplication factor */ 743 lapic_clockevent.mult = div_sc(delta, TICK_NSEC * LAPIC_CAL_LOOPS, 744 lapic_clockevent.shift); 745 lapic_clockevent.max_delta_ns = 746 clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent); 747 lapic_clockevent.min_delta_ns = 748 clockevent_delta2ns(0xF, &lapic_clockevent); 749 750 lapic_timer_frequency = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS; 751 752 apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta); 753 apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult); 754 apic_printk(APIC_VERBOSE, "..... calibration result: %u\n", 755 lapic_timer_frequency); 756 757 if (cpu_has_tsc) { 758 apic_printk(APIC_VERBOSE, "..... CPU clock speed is " 759 "%ld.%04ld MHz.\n", 760 (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ), 761 (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ)); 762 } 763 764 apic_printk(APIC_VERBOSE, "..... host bus clock speed is " 765 "%u.%04u MHz.\n", 766 lapic_timer_frequency / (1000000 / HZ), 767 lapic_timer_frequency % (1000000 / HZ)); 768 769 /* 770 * Do a sanity check on the APIC calibration result 771 */ 772 if (lapic_timer_frequency < (1000000 / HZ)) { 773 local_irq_enable(); 774 pr_warning("APIC frequency too slow, disabling apic timer\n"); 775 return -1; 776 } 777 778 levt->features &= ~CLOCK_EVT_FEAT_DUMMY; 779 780 /* 781 * PM timer calibration failed or not turned on 782 * so lets try APIC timer based calibration 783 */ 784 if (!pm_referenced) { 785 apic_printk(APIC_VERBOSE, "... verify APIC timer\n"); 786 787 /* 788 * Setup the apic timer manually 789 */ 790 levt->event_handler = lapic_cal_handler; 791 lapic_timer_set_periodic(levt); 792 lapic_cal_loops = -1; 793 794 /* Let the interrupts run */ 795 local_irq_enable(); 796 797 while (lapic_cal_loops <= LAPIC_CAL_LOOPS) 798 cpu_relax(); 799 800 /* Stop the lapic timer */ 801 local_irq_disable(); 802 lapic_timer_shutdown(levt); 803 804 /* Jiffies delta */ 805 deltaj = lapic_cal_j2 - lapic_cal_j1; 806 apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj); 807 808 /* Check, if the jiffies result is consistent */ 809 if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2) 810 apic_printk(APIC_VERBOSE, "... jiffies result ok\n"); 811 else 812 levt->features |= CLOCK_EVT_FEAT_DUMMY; 813 } 814 local_irq_enable(); 815 816 if (levt->features & CLOCK_EVT_FEAT_DUMMY) { 817 pr_warning("APIC timer disabled due to verification failure\n"); 818 return -1; 819 } 820 821 return 0; 822 } 823 824 /* 825 * Setup the boot APIC 826 * 827 * Calibrate and verify the result. 828 */ 829 void __init setup_boot_APIC_clock(void) 830 { 831 /* 832 * The local apic timer can be disabled via the kernel 833 * commandline or from the CPU detection code. Register the lapic 834 * timer as a dummy clock event source on SMP systems, so the 835 * broadcast mechanism is used. On UP systems simply ignore it. 836 */ 837 if (disable_apic_timer) { 838 pr_info("Disabling APIC timer\n"); 839 /* No broadcast on UP ! */ 840 if (num_possible_cpus() > 1) { 841 lapic_clockevent.mult = 1; 842 setup_APIC_timer(); 843 } 844 return; 845 } 846 847 if (calibrate_APIC_clock()) { 848 /* No broadcast on UP ! */ 849 if (num_possible_cpus() > 1) 850 setup_APIC_timer(); 851 return; 852 } 853 854 /* 855 * If nmi_watchdog is set to IO_APIC, we need the 856 * PIT/HPET going. Otherwise register lapic as a dummy 857 * device. 858 */ 859 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY; 860 861 /* Setup the lapic or request the broadcast */ 862 setup_APIC_timer(); 863 } 864 865 void setup_secondary_APIC_clock(void) 866 { 867 setup_APIC_timer(); 868 } 869 870 /* 871 * The guts of the apic timer interrupt 872 */ 873 static void local_apic_timer_interrupt(void) 874 { 875 int cpu = smp_processor_id(); 876 struct clock_event_device *evt = &per_cpu(lapic_events, cpu); 877 878 /* 879 * Normally we should not be here till LAPIC has been initialized but 880 * in some cases like kdump, its possible that there is a pending LAPIC 881 * timer interrupt from previous kernel's context and is delivered in 882 * new kernel the moment interrupts are enabled. 883 * 884 * Interrupts are enabled early and LAPIC is setup much later, hence 885 * its possible that when we get here evt->event_handler is NULL. 886 * Check for event_handler being NULL and discard the interrupt as 887 * spurious. 888 */ 889 if (!evt->event_handler) { 890 pr_warning("Spurious LAPIC timer interrupt on cpu %d\n", cpu); 891 /* Switch it off */ 892 lapic_timer_shutdown(evt); 893 return; 894 } 895 896 /* 897 * the NMI deadlock-detector uses this. 898 */ 899 inc_irq_stat(apic_timer_irqs); 900 901 evt->event_handler(evt); 902 } 903 904 /* 905 * Local APIC timer interrupt. This is the most natural way for doing 906 * local interrupts, but local timer interrupts can be emulated by 907 * broadcast interrupts too. [in case the hw doesn't support APIC timers] 908 * 909 * [ if a single-CPU system runs an SMP kernel then we call the local 910 * interrupt as well. Thus we cannot inline the local irq ... ] 911 */ 912 __visible void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs) 913 { 914 struct pt_regs *old_regs = set_irq_regs(regs); 915 916 /* 917 * NOTE! We'd better ACK the irq immediately, 918 * because timer handling can be slow. 919 * 920 * update_process_times() expects us to have done irq_enter(). 921 * Besides, if we don't timer interrupts ignore the global 922 * interrupt lock, which is the WrongThing (tm) to do. 923 */ 924 entering_ack_irq(); 925 local_apic_timer_interrupt(); 926 exiting_irq(); 927 928 set_irq_regs(old_regs); 929 } 930 931 __visible void __irq_entry smp_trace_apic_timer_interrupt(struct pt_regs *regs) 932 { 933 struct pt_regs *old_regs = set_irq_regs(regs); 934 935 /* 936 * NOTE! We'd better ACK the irq immediately, 937 * because timer handling can be slow. 938 * 939 * update_process_times() expects us to have done irq_enter(). 940 * Besides, if we don't timer interrupts ignore the global 941 * interrupt lock, which is the WrongThing (tm) to do. 942 */ 943 entering_ack_irq(); 944 trace_local_timer_entry(LOCAL_TIMER_VECTOR); 945 local_apic_timer_interrupt(); 946 trace_local_timer_exit(LOCAL_TIMER_VECTOR); 947 exiting_irq(); 948 949 set_irq_regs(old_regs); 950 } 951 952 int setup_profiling_timer(unsigned int multiplier) 953 { 954 return -EINVAL; 955 } 956 957 /* 958 * Local APIC start and shutdown 959 */ 960 961 /** 962 * clear_local_APIC - shutdown the local APIC 963 * 964 * This is called, when a CPU is disabled and before rebooting, so the state of 965 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS 966 * leftovers during boot. 967 */ 968 void clear_local_APIC(void) 969 { 970 int maxlvt; 971 u32 v; 972 973 /* APIC hasn't been mapped yet */ 974 if (!x2apic_mode && !apic_phys) 975 return; 976 977 maxlvt = lapic_get_maxlvt(); 978 /* 979 * Masking an LVT entry can trigger a local APIC error 980 * if the vector is zero. Mask LVTERR first to prevent this. 981 */ 982 if (maxlvt >= 3) { 983 v = ERROR_APIC_VECTOR; /* any non-zero vector will do */ 984 apic_write(APIC_LVTERR, v | APIC_LVT_MASKED); 985 } 986 /* 987 * Careful: we have to set masks only first to deassert 988 * any level-triggered sources. 989 */ 990 v = apic_read(APIC_LVTT); 991 apic_write(APIC_LVTT, v | APIC_LVT_MASKED); 992 v = apic_read(APIC_LVT0); 993 apic_write(APIC_LVT0, v | APIC_LVT_MASKED); 994 v = apic_read(APIC_LVT1); 995 apic_write(APIC_LVT1, v | APIC_LVT_MASKED); 996 if (maxlvt >= 4) { 997 v = apic_read(APIC_LVTPC); 998 apic_write(APIC_LVTPC, v | APIC_LVT_MASKED); 999 } 1000 1001 /* lets not touch this if we didn't frob it */ 1002 #ifdef CONFIG_X86_THERMAL_VECTOR 1003 if (maxlvt >= 5) { 1004 v = apic_read(APIC_LVTTHMR); 1005 apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED); 1006 } 1007 #endif 1008 #ifdef CONFIG_X86_MCE_INTEL 1009 if (maxlvt >= 6) { 1010 v = apic_read(APIC_LVTCMCI); 1011 if (!(v & APIC_LVT_MASKED)) 1012 apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED); 1013 } 1014 #endif 1015 1016 /* 1017 * Clean APIC state for other OSs: 1018 */ 1019 apic_write(APIC_LVTT, APIC_LVT_MASKED); 1020 apic_write(APIC_LVT0, APIC_LVT_MASKED); 1021 apic_write(APIC_LVT1, APIC_LVT_MASKED); 1022 if (maxlvt >= 3) 1023 apic_write(APIC_LVTERR, APIC_LVT_MASKED); 1024 if (maxlvt >= 4) 1025 apic_write(APIC_LVTPC, APIC_LVT_MASKED); 1026 1027 /* Integrated APIC (!82489DX) ? */ 1028 if (lapic_is_integrated()) { 1029 if (maxlvt > 3) 1030 /* Clear ESR due to Pentium errata 3AP and 11AP */ 1031 apic_write(APIC_ESR, 0); 1032 apic_read(APIC_ESR); 1033 } 1034 } 1035 1036 /** 1037 * disable_local_APIC - clear and disable the local APIC 1038 */ 1039 void disable_local_APIC(void) 1040 { 1041 unsigned int value; 1042 1043 /* APIC hasn't been mapped yet */ 1044 if (!x2apic_mode && !apic_phys) 1045 return; 1046 1047 clear_local_APIC(); 1048 1049 /* 1050 * Disable APIC (implies clearing of registers 1051 * for 82489DX!). 1052 */ 1053 value = apic_read(APIC_SPIV); 1054 value &= ~APIC_SPIV_APIC_ENABLED; 1055 apic_write(APIC_SPIV, value); 1056 1057 #ifdef CONFIG_X86_32 1058 /* 1059 * When LAPIC was disabled by the BIOS and enabled by the kernel, 1060 * restore the disabled state. 1061 */ 1062 if (enabled_via_apicbase) { 1063 unsigned int l, h; 1064 1065 rdmsr(MSR_IA32_APICBASE, l, h); 1066 l &= ~MSR_IA32_APICBASE_ENABLE; 1067 wrmsr(MSR_IA32_APICBASE, l, h); 1068 } 1069 #endif 1070 } 1071 1072 /* 1073 * If Linux enabled the LAPIC against the BIOS default disable it down before 1074 * re-entering the BIOS on shutdown. Otherwise the BIOS may get confused and 1075 * not power-off. Additionally clear all LVT entries before disable_local_APIC 1076 * for the case where Linux didn't enable the LAPIC. 1077 */ 1078 void lapic_shutdown(void) 1079 { 1080 unsigned long flags; 1081 1082 if (!cpu_has_apic && !apic_from_smp_config()) 1083 return; 1084 1085 local_irq_save(flags); 1086 1087 #ifdef CONFIG_X86_32 1088 if (!enabled_via_apicbase) 1089 clear_local_APIC(); 1090 else 1091 #endif 1092 disable_local_APIC(); 1093 1094 1095 local_irq_restore(flags); 1096 } 1097 1098 /** 1099 * sync_Arb_IDs - synchronize APIC bus arbitration IDs 1100 */ 1101 void __init sync_Arb_IDs(void) 1102 { 1103 /* 1104 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not 1105 * needed on AMD. 1106 */ 1107 if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD) 1108 return; 1109 1110 /* 1111 * Wait for idle. 1112 */ 1113 apic_wait_icr_idle(); 1114 1115 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n"); 1116 apic_write(APIC_ICR, APIC_DEST_ALLINC | 1117 APIC_INT_LEVELTRIG | APIC_DM_INIT); 1118 } 1119 1120 /* 1121 * An initial setup of the virtual wire mode. 1122 */ 1123 void __init init_bsp_APIC(void) 1124 { 1125 unsigned int value; 1126 1127 /* 1128 * Don't do the setup now if we have a SMP BIOS as the 1129 * through-I/O-APIC virtual wire mode might be active. 1130 */ 1131 if (smp_found_config || !cpu_has_apic) 1132 return; 1133 1134 /* 1135 * Do not trust the local APIC being empty at bootup. 1136 */ 1137 clear_local_APIC(); 1138 1139 /* 1140 * Enable APIC. 1141 */ 1142 value = apic_read(APIC_SPIV); 1143 value &= ~APIC_VECTOR_MASK; 1144 value |= APIC_SPIV_APIC_ENABLED; 1145 1146 #ifdef CONFIG_X86_32 1147 /* This bit is reserved on P4/Xeon and should be cleared */ 1148 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) && 1149 (boot_cpu_data.x86 == 15)) 1150 value &= ~APIC_SPIV_FOCUS_DISABLED; 1151 else 1152 #endif 1153 value |= APIC_SPIV_FOCUS_DISABLED; 1154 value |= SPURIOUS_APIC_VECTOR; 1155 apic_write(APIC_SPIV, value); 1156 1157 /* 1158 * Set up the virtual wire mode. 1159 */ 1160 apic_write(APIC_LVT0, APIC_DM_EXTINT); 1161 value = APIC_DM_NMI; 1162 if (!lapic_is_integrated()) /* 82489DX */ 1163 value |= APIC_LVT_LEVEL_TRIGGER; 1164 apic_write(APIC_LVT1, value); 1165 } 1166 1167 static void lapic_setup_esr(void) 1168 { 1169 unsigned int oldvalue, value, maxlvt; 1170 1171 if (!lapic_is_integrated()) { 1172 pr_info("No ESR for 82489DX.\n"); 1173 return; 1174 } 1175 1176 if (apic->disable_esr) { 1177 /* 1178 * Something untraceable is creating bad interrupts on 1179 * secondary quads ... for the moment, just leave the 1180 * ESR disabled - we can't do anything useful with the 1181 * errors anyway - mbligh 1182 */ 1183 pr_info("Leaving ESR disabled.\n"); 1184 return; 1185 } 1186 1187 maxlvt = lapic_get_maxlvt(); 1188 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */ 1189 apic_write(APIC_ESR, 0); 1190 oldvalue = apic_read(APIC_ESR); 1191 1192 /* enables sending errors */ 1193 value = ERROR_APIC_VECTOR; 1194 apic_write(APIC_LVTERR, value); 1195 1196 /* 1197 * spec says clear errors after enabling vector. 1198 */ 1199 if (maxlvt > 3) 1200 apic_write(APIC_ESR, 0); 1201 value = apic_read(APIC_ESR); 1202 if (value != oldvalue) 1203 apic_printk(APIC_VERBOSE, "ESR value before enabling " 1204 "vector: 0x%08x after: 0x%08x\n", 1205 oldvalue, value); 1206 } 1207 1208 /** 1209 * setup_local_APIC - setup the local APIC 1210 * 1211 * Used to setup local APIC while initializing BSP or bringin up APs. 1212 * Always called with preemption disabled. 1213 */ 1214 void setup_local_APIC(void) 1215 { 1216 int cpu = smp_processor_id(); 1217 unsigned int value, queued; 1218 int i, j, acked = 0; 1219 unsigned long long tsc = 0, ntsc; 1220 long long max_loops = cpu_khz ? cpu_khz : 1000000; 1221 1222 if (cpu_has_tsc) 1223 tsc = rdtsc(); 1224 1225 if (disable_apic) { 1226 disable_ioapic_support(); 1227 return; 1228 } 1229 1230 #ifdef CONFIG_X86_32 1231 /* Pound the ESR really hard over the head with a big hammer - mbligh */ 1232 if (lapic_is_integrated() && apic->disable_esr) { 1233 apic_write(APIC_ESR, 0); 1234 apic_write(APIC_ESR, 0); 1235 apic_write(APIC_ESR, 0); 1236 apic_write(APIC_ESR, 0); 1237 } 1238 #endif 1239 perf_events_lapic_init(); 1240 1241 /* 1242 * Double-check whether this APIC is really registered. 1243 * This is meaningless in clustered apic mode, so we skip it. 1244 */ 1245 BUG_ON(!apic->apic_id_registered()); 1246 1247 /* 1248 * Intel recommends to set DFR, LDR and TPR before enabling 1249 * an APIC. See e.g. "AP-388 82489DX User's Manual" (Intel 1250 * document number 292116). So here it goes... 1251 */ 1252 apic->init_apic_ldr(); 1253 1254 #ifdef CONFIG_X86_32 1255 /* 1256 * APIC LDR is initialized. If logical_apicid mapping was 1257 * initialized during get_smp_config(), make sure it matches the 1258 * actual value. 1259 */ 1260 i = early_per_cpu(x86_cpu_to_logical_apicid, cpu); 1261 WARN_ON(i != BAD_APICID && i != logical_smp_processor_id()); 1262 /* always use the value from LDR */ 1263 early_per_cpu(x86_cpu_to_logical_apicid, cpu) = 1264 logical_smp_processor_id(); 1265 #endif 1266 1267 /* 1268 * Set Task Priority to 'accept all'. We never change this 1269 * later on. 1270 */ 1271 value = apic_read(APIC_TASKPRI); 1272 value &= ~APIC_TPRI_MASK; 1273 apic_write(APIC_TASKPRI, value); 1274 1275 /* 1276 * After a crash, we no longer service the interrupts and a pending 1277 * interrupt from previous kernel might still have ISR bit set. 1278 * 1279 * Most probably by now CPU has serviced that pending interrupt and 1280 * it might not have done the ack_APIC_irq() because it thought, 1281 * interrupt came from i8259 as ExtInt. LAPIC did not get EOI so it 1282 * does not clear the ISR bit and cpu thinks it has already serivced 1283 * the interrupt. Hence a vector might get locked. It was noticed 1284 * for timer irq (vector 0x31). Issue an extra EOI to clear ISR. 1285 */ 1286 do { 1287 queued = 0; 1288 for (i = APIC_ISR_NR - 1; i >= 0; i--) 1289 queued |= apic_read(APIC_IRR + i*0x10); 1290 1291 for (i = APIC_ISR_NR - 1; i >= 0; i--) { 1292 value = apic_read(APIC_ISR + i*0x10); 1293 for (j = 31; j >= 0; j--) { 1294 if (value & (1<<j)) { 1295 ack_APIC_irq(); 1296 acked++; 1297 } 1298 } 1299 } 1300 if (acked > 256) { 1301 printk(KERN_ERR "LAPIC pending interrupts after %d EOI\n", 1302 acked); 1303 break; 1304 } 1305 if (queued) { 1306 if (cpu_has_tsc && cpu_khz) { 1307 ntsc = rdtsc(); 1308 max_loops = (cpu_khz << 10) - (ntsc - tsc); 1309 } else 1310 max_loops--; 1311 } 1312 } while (queued && max_loops > 0); 1313 WARN_ON(max_loops <= 0); 1314 1315 /* 1316 * Now that we are all set up, enable the APIC 1317 */ 1318 value = apic_read(APIC_SPIV); 1319 value &= ~APIC_VECTOR_MASK; 1320 /* 1321 * Enable APIC 1322 */ 1323 value |= APIC_SPIV_APIC_ENABLED; 1324 1325 #ifdef CONFIG_X86_32 1326 /* 1327 * Some unknown Intel IO/APIC (or APIC) errata is biting us with 1328 * certain networking cards. If high frequency interrupts are 1329 * happening on a particular IOAPIC pin, plus the IOAPIC routing 1330 * entry is masked/unmasked at a high rate as well then sooner or 1331 * later IOAPIC line gets 'stuck', no more interrupts are received 1332 * from the device. If focus CPU is disabled then the hang goes 1333 * away, oh well :-( 1334 * 1335 * [ This bug can be reproduced easily with a level-triggered 1336 * PCI Ne2000 networking cards and PII/PIII processors, dual 1337 * BX chipset. ] 1338 */ 1339 /* 1340 * Actually disabling the focus CPU check just makes the hang less 1341 * frequent as it makes the interrupt distributon model be more 1342 * like LRU than MRU (the short-term load is more even across CPUs). 1343 * See also the comment in end_level_ioapic_irq(). --macro 1344 */ 1345 1346 /* 1347 * - enable focus processor (bit==0) 1348 * - 64bit mode always use processor focus 1349 * so no need to set it 1350 */ 1351 value &= ~APIC_SPIV_FOCUS_DISABLED; 1352 #endif 1353 1354 /* 1355 * Set spurious IRQ vector 1356 */ 1357 value |= SPURIOUS_APIC_VECTOR; 1358 apic_write(APIC_SPIV, value); 1359 1360 /* 1361 * Set up LVT0, LVT1: 1362 * 1363 * set up through-local-APIC on the BP's LINT0. This is not 1364 * strictly necessary in pure symmetric-IO mode, but sometimes 1365 * we delegate interrupts to the 8259A. 1366 */ 1367 /* 1368 * TODO: set up through-local-APIC from through-I/O-APIC? --macro 1369 */ 1370 value = apic_read(APIC_LVT0) & APIC_LVT_MASKED; 1371 if (!cpu && (pic_mode || !value)) { 1372 value = APIC_DM_EXTINT; 1373 apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu); 1374 } else { 1375 value = APIC_DM_EXTINT | APIC_LVT_MASKED; 1376 apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu); 1377 } 1378 apic_write(APIC_LVT0, value); 1379 1380 /* 1381 * only the BP should see the LINT1 NMI signal, obviously. 1382 */ 1383 if (!cpu) 1384 value = APIC_DM_NMI; 1385 else 1386 value = APIC_DM_NMI | APIC_LVT_MASKED; 1387 if (!lapic_is_integrated()) /* 82489DX */ 1388 value |= APIC_LVT_LEVEL_TRIGGER; 1389 apic_write(APIC_LVT1, value); 1390 1391 #ifdef CONFIG_X86_MCE_INTEL 1392 /* Recheck CMCI information after local APIC is up on CPU #0 */ 1393 if (!cpu) 1394 cmci_recheck(); 1395 #endif 1396 } 1397 1398 static void end_local_APIC_setup(void) 1399 { 1400 lapic_setup_esr(); 1401 1402 #ifdef CONFIG_X86_32 1403 { 1404 unsigned int value; 1405 /* Disable the local apic timer */ 1406 value = apic_read(APIC_LVTT); 1407 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR); 1408 apic_write(APIC_LVTT, value); 1409 } 1410 #endif 1411 1412 apic_pm_activate(); 1413 } 1414 1415 /* 1416 * APIC setup function for application processors. Called from smpboot.c 1417 */ 1418 void apic_ap_setup(void) 1419 { 1420 setup_local_APIC(); 1421 end_local_APIC_setup(); 1422 } 1423 1424 #ifdef CONFIG_X86_X2APIC 1425 int x2apic_mode; 1426 1427 enum { 1428 X2APIC_OFF, 1429 X2APIC_ON, 1430 X2APIC_DISABLED, 1431 }; 1432 static int x2apic_state; 1433 1434 static void __x2apic_disable(void) 1435 { 1436 u64 msr; 1437 1438 if (!cpu_has_apic) 1439 return; 1440 1441 rdmsrl(MSR_IA32_APICBASE, msr); 1442 if (!(msr & X2APIC_ENABLE)) 1443 return; 1444 /* Disable xapic and x2apic first and then reenable xapic mode */ 1445 wrmsrl(MSR_IA32_APICBASE, msr & ~(X2APIC_ENABLE | XAPIC_ENABLE)); 1446 wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE); 1447 printk_once(KERN_INFO "x2apic disabled\n"); 1448 } 1449 1450 static void __x2apic_enable(void) 1451 { 1452 u64 msr; 1453 1454 rdmsrl(MSR_IA32_APICBASE, msr); 1455 if (msr & X2APIC_ENABLE) 1456 return; 1457 wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE); 1458 printk_once(KERN_INFO "x2apic enabled\n"); 1459 } 1460 1461 static int __init setup_nox2apic(char *str) 1462 { 1463 if (x2apic_enabled()) { 1464 int apicid = native_apic_msr_read(APIC_ID); 1465 1466 if (apicid >= 255) { 1467 pr_warning("Apicid: %08x, cannot enforce nox2apic\n", 1468 apicid); 1469 return 0; 1470 } 1471 pr_warning("x2apic already enabled.\n"); 1472 __x2apic_disable(); 1473 } 1474 setup_clear_cpu_cap(X86_FEATURE_X2APIC); 1475 x2apic_state = X2APIC_DISABLED; 1476 x2apic_mode = 0; 1477 return 0; 1478 } 1479 early_param("nox2apic", setup_nox2apic); 1480 1481 /* Called from cpu_init() to enable x2apic on (secondary) cpus */ 1482 void x2apic_setup(void) 1483 { 1484 /* 1485 * If x2apic is not in ON state, disable it if already enabled 1486 * from BIOS. 1487 */ 1488 if (x2apic_state != X2APIC_ON) { 1489 __x2apic_disable(); 1490 return; 1491 } 1492 __x2apic_enable(); 1493 } 1494 1495 static __init void x2apic_disable(void) 1496 { 1497 u32 x2apic_id, state = x2apic_state; 1498 1499 x2apic_mode = 0; 1500 x2apic_state = X2APIC_DISABLED; 1501 1502 if (state != X2APIC_ON) 1503 return; 1504 1505 x2apic_id = read_apic_id(); 1506 if (x2apic_id >= 255) 1507 panic("Cannot disable x2apic, id: %08x\n", x2apic_id); 1508 1509 __x2apic_disable(); 1510 register_lapic_address(mp_lapic_addr); 1511 } 1512 1513 static __init void x2apic_enable(void) 1514 { 1515 if (x2apic_state != X2APIC_OFF) 1516 return; 1517 1518 x2apic_mode = 1; 1519 x2apic_state = X2APIC_ON; 1520 __x2apic_enable(); 1521 } 1522 1523 static __init void try_to_enable_x2apic(int remap_mode) 1524 { 1525 if (x2apic_state == X2APIC_DISABLED) 1526 return; 1527 1528 if (remap_mode != IRQ_REMAP_X2APIC_MODE) { 1529 /* IR is required if there is APIC ID > 255 even when running 1530 * under KVM 1531 */ 1532 if (max_physical_apicid > 255 || 1533 !hypervisor_x2apic_available()) { 1534 pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n"); 1535 x2apic_disable(); 1536 return; 1537 } 1538 1539 /* 1540 * without IR all CPUs can be addressed by IOAPIC/MSI 1541 * only in physical mode 1542 */ 1543 x2apic_phys = 1; 1544 } 1545 x2apic_enable(); 1546 } 1547 1548 void __init check_x2apic(void) 1549 { 1550 if (x2apic_enabled()) { 1551 pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n"); 1552 x2apic_mode = 1; 1553 x2apic_state = X2APIC_ON; 1554 } else if (!cpu_has_x2apic) { 1555 x2apic_state = X2APIC_DISABLED; 1556 } 1557 } 1558 #else /* CONFIG_X86_X2APIC */ 1559 static int __init validate_x2apic(void) 1560 { 1561 if (!apic_is_x2apic_enabled()) 1562 return 0; 1563 /* 1564 * Checkme: Can we simply turn off x2apic here instead of panic? 1565 */ 1566 panic("BIOS has enabled x2apic but kernel doesn't support x2apic, please disable x2apic in BIOS.\n"); 1567 } 1568 early_initcall(validate_x2apic); 1569 1570 static inline void try_to_enable_x2apic(int remap_mode) { } 1571 static inline void __x2apic_enable(void) { } 1572 #endif /* !CONFIG_X86_X2APIC */ 1573 1574 static int __init try_to_enable_IR(void) 1575 { 1576 #ifdef CONFIG_X86_IO_APIC 1577 if (!x2apic_enabled() && skip_ioapic_setup) { 1578 pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n"); 1579 return -1; 1580 } 1581 #endif 1582 return irq_remapping_enable(); 1583 } 1584 1585 void __init enable_IR_x2apic(void) 1586 { 1587 unsigned long flags; 1588 int ret, ir_stat; 1589 1590 ir_stat = irq_remapping_prepare(); 1591 if (ir_stat < 0 && !x2apic_supported()) 1592 return; 1593 1594 ret = save_ioapic_entries(); 1595 if (ret) { 1596 pr_info("Saving IO-APIC state failed: %d\n", ret); 1597 return; 1598 } 1599 1600 local_irq_save(flags); 1601 legacy_pic->mask_all(); 1602 mask_ioapic_entries(); 1603 1604 /* If irq_remapping_prepare() succeded, try to enable it */ 1605 if (ir_stat >= 0) 1606 ir_stat = try_to_enable_IR(); 1607 /* ir_stat contains the remap mode or an error code */ 1608 try_to_enable_x2apic(ir_stat); 1609 1610 if (ir_stat < 0) 1611 restore_ioapic_entries(); 1612 legacy_pic->restore_mask(); 1613 local_irq_restore(flags); 1614 } 1615 1616 #ifdef CONFIG_X86_64 1617 /* 1618 * Detect and enable local APICs on non-SMP boards. 1619 * Original code written by Keir Fraser. 1620 * On AMD64 we trust the BIOS - if it says no APIC it is likely 1621 * not correctly set up (usually the APIC timer won't work etc.) 1622 */ 1623 static int __init detect_init_APIC(void) 1624 { 1625 if (!cpu_has_apic) { 1626 pr_info("No local APIC present\n"); 1627 return -1; 1628 } 1629 1630 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE; 1631 return 0; 1632 } 1633 #else 1634 1635 static int __init apic_verify(void) 1636 { 1637 u32 features, h, l; 1638 1639 /* 1640 * The APIC feature bit should now be enabled 1641 * in `cpuid' 1642 */ 1643 features = cpuid_edx(1); 1644 if (!(features & (1 << X86_FEATURE_APIC))) { 1645 pr_warning("Could not enable APIC!\n"); 1646 return -1; 1647 } 1648 set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC); 1649 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE; 1650 1651 /* The BIOS may have set up the APIC at some other address */ 1652 if (boot_cpu_data.x86 >= 6) { 1653 rdmsr(MSR_IA32_APICBASE, l, h); 1654 if (l & MSR_IA32_APICBASE_ENABLE) 1655 mp_lapic_addr = l & MSR_IA32_APICBASE_BASE; 1656 } 1657 1658 pr_info("Found and enabled local APIC!\n"); 1659 return 0; 1660 } 1661 1662 int __init apic_force_enable(unsigned long addr) 1663 { 1664 u32 h, l; 1665 1666 if (disable_apic) 1667 return -1; 1668 1669 /* 1670 * Some BIOSes disable the local APIC in the APIC_BASE 1671 * MSR. This can only be done in software for Intel P6 or later 1672 * and AMD K7 (Model > 1) or later. 1673 */ 1674 if (boot_cpu_data.x86 >= 6) { 1675 rdmsr(MSR_IA32_APICBASE, l, h); 1676 if (!(l & MSR_IA32_APICBASE_ENABLE)) { 1677 pr_info("Local APIC disabled by BIOS -- reenabling.\n"); 1678 l &= ~MSR_IA32_APICBASE_BASE; 1679 l |= MSR_IA32_APICBASE_ENABLE | addr; 1680 wrmsr(MSR_IA32_APICBASE, l, h); 1681 enabled_via_apicbase = 1; 1682 } 1683 } 1684 return apic_verify(); 1685 } 1686 1687 /* 1688 * Detect and initialize APIC 1689 */ 1690 static int __init detect_init_APIC(void) 1691 { 1692 /* Disabled by kernel option? */ 1693 if (disable_apic) 1694 return -1; 1695 1696 switch (boot_cpu_data.x86_vendor) { 1697 case X86_VENDOR_AMD: 1698 if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) || 1699 (boot_cpu_data.x86 >= 15)) 1700 break; 1701 goto no_apic; 1702 case X86_VENDOR_INTEL: 1703 if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 || 1704 (boot_cpu_data.x86 == 5 && cpu_has_apic)) 1705 break; 1706 goto no_apic; 1707 default: 1708 goto no_apic; 1709 } 1710 1711 if (!cpu_has_apic) { 1712 /* 1713 * Over-ride BIOS and try to enable the local APIC only if 1714 * "lapic" specified. 1715 */ 1716 if (!force_enable_local_apic) { 1717 pr_info("Local APIC disabled by BIOS -- " 1718 "you can enable it with \"lapic\"\n"); 1719 return -1; 1720 } 1721 if (apic_force_enable(APIC_DEFAULT_PHYS_BASE)) 1722 return -1; 1723 } else { 1724 if (apic_verify()) 1725 return -1; 1726 } 1727 1728 apic_pm_activate(); 1729 1730 return 0; 1731 1732 no_apic: 1733 pr_info("No local APIC present or hardware disabled\n"); 1734 return -1; 1735 } 1736 #endif 1737 1738 /** 1739 * init_apic_mappings - initialize APIC mappings 1740 */ 1741 void __init init_apic_mappings(void) 1742 { 1743 unsigned int new_apicid; 1744 1745 if (x2apic_mode) { 1746 boot_cpu_physical_apicid = read_apic_id(); 1747 return; 1748 } 1749 1750 /* If no local APIC can be found return early */ 1751 if (!smp_found_config && detect_init_APIC()) { 1752 /* lets NOP'ify apic operations */ 1753 pr_info("APIC: disable apic facility\n"); 1754 apic_disable(); 1755 } else { 1756 apic_phys = mp_lapic_addr; 1757 1758 /* 1759 * acpi lapic path already maps that address in 1760 * acpi_register_lapic_address() 1761 */ 1762 if (!acpi_lapic && !smp_found_config) 1763 register_lapic_address(apic_phys); 1764 } 1765 1766 /* 1767 * Fetch the APIC ID of the BSP in case we have a 1768 * default configuration (or the MP table is broken). 1769 */ 1770 new_apicid = read_apic_id(); 1771 if (boot_cpu_physical_apicid != new_apicid) { 1772 boot_cpu_physical_apicid = new_apicid; 1773 /* 1774 * yeah -- we lie about apic_version 1775 * in case if apic was disabled via boot option 1776 * but it's not a problem for SMP compiled kernel 1777 * since smp_sanity_check is prepared for such a case 1778 * and disable smp mode 1779 */ 1780 apic_version[new_apicid] = 1781 GET_APIC_VERSION(apic_read(APIC_LVR)); 1782 } 1783 } 1784 1785 void __init register_lapic_address(unsigned long address) 1786 { 1787 mp_lapic_addr = address; 1788 1789 if (!x2apic_mode) { 1790 set_fixmap_nocache(FIX_APIC_BASE, address); 1791 apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n", 1792 APIC_BASE, mp_lapic_addr); 1793 } 1794 if (boot_cpu_physical_apicid == -1U) { 1795 boot_cpu_physical_apicid = read_apic_id(); 1796 apic_version[boot_cpu_physical_apicid] = 1797 GET_APIC_VERSION(apic_read(APIC_LVR)); 1798 } 1799 } 1800 1801 int apic_version[MAX_LOCAL_APIC]; 1802 1803 /* 1804 * Local APIC interrupts 1805 */ 1806 1807 /* 1808 * This interrupt should _never_ happen with our APIC/SMP architecture 1809 */ 1810 static void __smp_spurious_interrupt(u8 vector) 1811 { 1812 u32 v; 1813 1814 /* 1815 * Check if this really is a spurious interrupt and ACK it 1816 * if it is a vectored one. Just in case... 1817 * Spurious interrupts should not be ACKed. 1818 */ 1819 v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1)); 1820 if (v & (1 << (vector & 0x1f))) 1821 ack_APIC_irq(); 1822 1823 inc_irq_stat(irq_spurious_count); 1824 1825 /* see sw-dev-man vol 3, chapter 7.4.13.5 */ 1826 pr_info("spurious APIC interrupt through vector %02x on CPU#%d, " 1827 "should never happen.\n", vector, smp_processor_id()); 1828 } 1829 1830 __visible void smp_spurious_interrupt(struct pt_regs *regs) 1831 { 1832 entering_irq(); 1833 __smp_spurious_interrupt(~regs->orig_ax); 1834 exiting_irq(); 1835 } 1836 1837 __visible void smp_trace_spurious_interrupt(struct pt_regs *regs) 1838 { 1839 u8 vector = ~regs->orig_ax; 1840 1841 entering_irq(); 1842 trace_spurious_apic_entry(vector); 1843 __smp_spurious_interrupt(vector); 1844 trace_spurious_apic_exit(vector); 1845 exiting_irq(); 1846 } 1847 1848 /* 1849 * This interrupt should never happen with our APIC/SMP architecture 1850 */ 1851 static void __smp_error_interrupt(struct pt_regs *regs) 1852 { 1853 u32 v; 1854 u32 i = 0; 1855 static const char * const error_interrupt_reason[] = { 1856 "Send CS error", /* APIC Error Bit 0 */ 1857 "Receive CS error", /* APIC Error Bit 1 */ 1858 "Send accept error", /* APIC Error Bit 2 */ 1859 "Receive accept error", /* APIC Error Bit 3 */ 1860 "Redirectable IPI", /* APIC Error Bit 4 */ 1861 "Send illegal vector", /* APIC Error Bit 5 */ 1862 "Received illegal vector", /* APIC Error Bit 6 */ 1863 "Illegal register address", /* APIC Error Bit 7 */ 1864 }; 1865 1866 /* First tickle the hardware, only then report what went on. -- REW */ 1867 if (lapic_get_maxlvt() > 3) /* Due to the Pentium erratum 3AP. */ 1868 apic_write(APIC_ESR, 0); 1869 v = apic_read(APIC_ESR); 1870 ack_APIC_irq(); 1871 atomic_inc(&irq_err_count); 1872 1873 apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x", 1874 smp_processor_id(), v); 1875 1876 v &= 0xff; 1877 while (v) { 1878 if (v & 0x1) 1879 apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]); 1880 i++; 1881 v >>= 1; 1882 } 1883 1884 apic_printk(APIC_DEBUG, KERN_CONT "\n"); 1885 1886 } 1887 1888 __visible void smp_error_interrupt(struct pt_regs *regs) 1889 { 1890 entering_irq(); 1891 __smp_error_interrupt(regs); 1892 exiting_irq(); 1893 } 1894 1895 __visible void smp_trace_error_interrupt(struct pt_regs *regs) 1896 { 1897 entering_irq(); 1898 trace_error_apic_entry(ERROR_APIC_VECTOR); 1899 __smp_error_interrupt(regs); 1900 trace_error_apic_exit(ERROR_APIC_VECTOR); 1901 exiting_irq(); 1902 } 1903 1904 /** 1905 * connect_bsp_APIC - attach the APIC to the interrupt system 1906 */ 1907 static void __init connect_bsp_APIC(void) 1908 { 1909 #ifdef CONFIG_X86_32 1910 if (pic_mode) { 1911 /* 1912 * Do not trust the local APIC being empty at bootup. 1913 */ 1914 clear_local_APIC(); 1915 /* 1916 * PIC mode, enable APIC mode in the IMCR, i.e. connect BSP's 1917 * local APIC to INT and NMI lines. 1918 */ 1919 apic_printk(APIC_VERBOSE, "leaving PIC mode, " 1920 "enabling APIC mode.\n"); 1921 imcr_pic_to_apic(); 1922 } 1923 #endif 1924 } 1925 1926 /** 1927 * disconnect_bsp_APIC - detach the APIC from the interrupt system 1928 * @virt_wire_setup: indicates, whether virtual wire mode is selected 1929 * 1930 * Virtual wire mode is necessary to deliver legacy interrupts even when the 1931 * APIC is disabled. 1932 */ 1933 void disconnect_bsp_APIC(int virt_wire_setup) 1934 { 1935 unsigned int value; 1936 1937 #ifdef CONFIG_X86_32 1938 if (pic_mode) { 1939 /* 1940 * Put the board back into PIC mode (has an effect only on 1941 * certain older boards). Note that APIC interrupts, including 1942 * IPIs, won't work beyond this point! The only exception are 1943 * INIT IPIs. 1944 */ 1945 apic_printk(APIC_VERBOSE, "disabling APIC mode, " 1946 "entering PIC mode.\n"); 1947 imcr_apic_to_pic(); 1948 return; 1949 } 1950 #endif 1951 1952 /* Go back to Virtual Wire compatibility mode */ 1953 1954 /* For the spurious interrupt use vector F, and enable it */ 1955 value = apic_read(APIC_SPIV); 1956 value &= ~APIC_VECTOR_MASK; 1957 value |= APIC_SPIV_APIC_ENABLED; 1958 value |= 0xf; 1959 apic_write(APIC_SPIV, value); 1960 1961 if (!virt_wire_setup) { 1962 /* 1963 * For LVT0 make it edge triggered, active high, 1964 * external and enabled 1965 */ 1966 value = apic_read(APIC_LVT0); 1967 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING | 1968 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR | 1969 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED); 1970 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING; 1971 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT); 1972 apic_write(APIC_LVT0, value); 1973 } else { 1974 /* Disable LVT0 */ 1975 apic_write(APIC_LVT0, APIC_LVT_MASKED); 1976 } 1977 1978 /* 1979 * For LVT1 make it edge triggered, active high, 1980 * nmi and enabled 1981 */ 1982 value = apic_read(APIC_LVT1); 1983 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING | 1984 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR | 1985 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED); 1986 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING; 1987 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI); 1988 apic_write(APIC_LVT1, value); 1989 } 1990 1991 int generic_processor_info(int apicid, int version) 1992 { 1993 int cpu, max = nr_cpu_ids; 1994 bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid, 1995 phys_cpu_present_map); 1996 1997 /* 1998 * boot_cpu_physical_apicid is designed to have the apicid 1999 * returned by read_apic_id(), i.e, the apicid of the 2000 * currently booting-up processor. However, on some platforms, 2001 * it is temporarily modified by the apicid reported as BSP 2002 * through MP table. Concretely: 2003 * 2004 * - arch/x86/kernel/mpparse.c: MP_processor_info() 2005 * - arch/x86/mm/amdtopology.c: amd_numa_init() 2006 * 2007 * This function is executed with the modified 2008 * boot_cpu_physical_apicid. So, disabled_cpu_apicid kernel 2009 * parameter doesn't work to disable APs on kdump 2nd kernel. 2010 * 2011 * Since fixing handling of boot_cpu_physical_apicid requires 2012 * another discussion and tests on each platform, we leave it 2013 * for now and here we use read_apic_id() directly in this 2014 * function, generic_processor_info(). 2015 */ 2016 if (disabled_cpu_apicid != BAD_APICID && 2017 disabled_cpu_apicid != read_apic_id() && 2018 disabled_cpu_apicid == apicid) { 2019 int thiscpu = num_processors + disabled_cpus; 2020 2021 pr_warning("APIC: Disabling requested cpu." 2022 " Processor %d/0x%x ignored.\n", 2023 thiscpu, apicid); 2024 2025 disabled_cpus++; 2026 return -ENODEV; 2027 } 2028 2029 /* 2030 * If boot cpu has not been detected yet, then only allow upto 2031 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu 2032 */ 2033 if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 && 2034 apicid != boot_cpu_physical_apicid) { 2035 int thiscpu = max + disabled_cpus - 1; 2036 2037 pr_warning( 2038 "ACPI: NR_CPUS/possible_cpus limit of %i almost" 2039 " reached. Keeping one slot for boot cpu." 2040 " Processor %d/0x%x ignored.\n", max, thiscpu, apicid); 2041 2042 disabled_cpus++; 2043 return -ENODEV; 2044 } 2045 2046 if (num_processors >= nr_cpu_ids) { 2047 int thiscpu = max + disabled_cpus; 2048 2049 pr_warning( 2050 "ACPI: NR_CPUS/possible_cpus limit of %i reached." 2051 " Processor %d/0x%x ignored.\n", max, thiscpu, apicid); 2052 2053 disabled_cpus++; 2054 return -EINVAL; 2055 } 2056 2057 num_processors++; 2058 if (apicid == boot_cpu_physical_apicid) { 2059 /* 2060 * x86_bios_cpu_apicid is required to have processors listed 2061 * in same order as logical cpu numbers. Hence the first 2062 * entry is BSP, and so on. 2063 * boot_cpu_init() already hold bit 0 in cpu_present_mask 2064 * for BSP. 2065 */ 2066 cpu = 0; 2067 } else 2068 cpu = cpumask_next_zero(-1, cpu_present_mask); 2069 2070 /* 2071 * Validate version 2072 */ 2073 if (version == 0x0) { 2074 pr_warning("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n", 2075 cpu, apicid); 2076 version = 0x10; 2077 } 2078 apic_version[apicid] = version; 2079 2080 if (version != apic_version[boot_cpu_physical_apicid]) { 2081 pr_warning("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n", 2082 apic_version[boot_cpu_physical_apicid], cpu, version); 2083 } 2084 2085 physid_set(apicid, phys_cpu_present_map); 2086 if (apicid > max_physical_apicid) 2087 max_physical_apicid = apicid; 2088 2089 #if defined(CONFIG_SMP) || defined(CONFIG_X86_64) 2090 early_per_cpu(x86_cpu_to_apicid, cpu) = apicid; 2091 early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid; 2092 #endif 2093 #ifdef CONFIG_X86_32 2094 early_per_cpu(x86_cpu_to_logical_apicid, cpu) = 2095 apic->x86_32_early_logical_apicid(cpu); 2096 #endif 2097 set_cpu_possible(cpu, true); 2098 set_cpu_present(cpu, true); 2099 2100 return cpu; 2101 } 2102 2103 int hard_smp_processor_id(void) 2104 { 2105 return read_apic_id(); 2106 } 2107 2108 void default_init_apic_ldr(void) 2109 { 2110 unsigned long val; 2111 2112 apic_write(APIC_DFR, APIC_DFR_VALUE); 2113 val = apic_read(APIC_LDR) & ~APIC_LDR_MASK; 2114 val |= SET_APIC_LOGICAL_ID(1UL << smp_processor_id()); 2115 apic_write(APIC_LDR, val); 2116 } 2117 2118 int default_cpu_mask_to_apicid_and(const struct cpumask *cpumask, 2119 const struct cpumask *andmask, 2120 unsigned int *apicid) 2121 { 2122 unsigned int cpu; 2123 2124 for_each_cpu_and(cpu, cpumask, andmask) { 2125 if (cpumask_test_cpu(cpu, cpu_online_mask)) 2126 break; 2127 } 2128 2129 if (likely(cpu < nr_cpu_ids)) { 2130 *apicid = per_cpu(x86_cpu_to_apicid, cpu); 2131 return 0; 2132 } 2133 2134 return -EINVAL; 2135 } 2136 2137 /* 2138 * Override the generic EOI implementation with an optimized version. 2139 * Only called during early boot when only one CPU is active and with 2140 * interrupts disabled, so we know this does not race with actual APIC driver 2141 * use. 2142 */ 2143 void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v)) 2144 { 2145 struct apic **drv; 2146 2147 for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) { 2148 /* Should happen once for each apic */ 2149 WARN_ON((*drv)->eoi_write == eoi_write); 2150 (*drv)->eoi_write = eoi_write; 2151 } 2152 } 2153 2154 static void __init apic_bsp_up_setup(void) 2155 { 2156 #ifdef CONFIG_X86_64 2157 apic_write(APIC_ID, SET_APIC_ID(boot_cpu_physical_apicid)); 2158 #else 2159 /* 2160 * Hack: In case of kdump, after a crash, kernel might be booting 2161 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid 2162 * might be zero if read from MP tables. Get it from LAPIC. 2163 */ 2164 # ifdef CONFIG_CRASH_DUMP 2165 boot_cpu_physical_apicid = read_apic_id(); 2166 # endif 2167 #endif 2168 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map); 2169 } 2170 2171 /** 2172 * apic_bsp_setup - Setup function for local apic and io-apic 2173 * @upmode: Force UP mode (for APIC_init_uniprocessor) 2174 * 2175 * Returns: 2176 * apic_id of BSP APIC 2177 */ 2178 int __init apic_bsp_setup(bool upmode) 2179 { 2180 int id; 2181 2182 connect_bsp_APIC(); 2183 if (upmode) 2184 apic_bsp_up_setup(); 2185 setup_local_APIC(); 2186 2187 if (x2apic_mode) 2188 id = apic_read(APIC_LDR); 2189 else 2190 id = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR)); 2191 2192 enable_IO_APIC(); 2193 end_local_APIC_setup(); 2194 irq_remap_enable_fault_handling(); 2195 setup_IO_APIC(); 2196 /* Setup local timer */ 2197 x86_init.timers.setup_percpu_clockev(); 2198 return id; 2199 } 2200 2201 /* 2202 * This initializes the IO-APIC and APIC hardware if this is 2203 * a UP kernel. 2204 */ 2205 int __init APIC_init_uniprocessor(void) 2206 { 2207 if (disable_apic) { 2208 pr_info("Apic disabled\n"); 2209 return -1; 2210 } 2211 #ifdef CONFIG_X86_64 2212 if (!cpu_has_apic) { 2213 disable_apic = 1; 2214 pr_info("Apic disabled by BIOS\n"); 2215 return -1; 2216 } 2217 #else 2218 if (!smp_found_config && !cpu_has_apic) 2219 return -1; 2220 2221 /* 2222 * Complain if the BIOS pretends there is one. 2223 */ 2224 if (!cpu_has_apic && 2225 APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) { 2226 pr_err("BIOS bug, local APIC 0x%x not detected!...\n", 2227 boot_cpu_physical_apicid); 2228 return -1; 2229 } 2230 #endif 2231 2232 if (!smp_found_config) 2233 disable_ioapic_support(); 2234 2235 default_setup_apic_routing(); 2236 apic_bsp_setup(true); 2237 return 0; 2238 } 2239 2240 #ifdef CONFIG_UP_LATE_INIT 2241 void __init up_late_init(void) 2242 { 2243 APIC_init_uniprocessor(); 2244 } 2245 #endif 2246 2247 /* 2248 * Power management 2249 */ 2250 #ifdef CONFIG_PM 2251 2252 static struct { 2253 /* 2254 * 'active' is true if the local APIC was enabled by us and 2255 * not the BIOS; this signifies that we are also responsible 2256 * for disabling it before entering apm/acpi suspend 2257 */ 2258 int active; 2259 /* r/w apic fields */ 2260 unsigned int apic_id; 2261 unsigned int apic_taskpri; 2262 unsigned int apic_ldr; 2263 unsigned int apic_dfr; 2264 unsigned int apic_spiv; 2265 unsigned int apic_lvtt; 2266 unsigned int apic_lvtpc; 2267 unsigned int apic_lvt0; 2268 unsigned int apic_lvt1; 2269 unsigned int apic_lvterr; 2270 unsigned int apic_tmict; 2271 unsigned int apic_tdcr; 2272 unsigned int apic_thmr; 2273 } apic_pm_state; 2274 2275 static int lapic_suspend(void) 2276 { 2277 unsigned long flags; 2278 int maxlvt; 2279 2280 if (!apic_pm_state.active) 2281 return 0; 2282 2283 maxlvt = lapic_get_maxlvt(); 2284 2285 apic_pm_state.apic_id = apic_read(APIC_ID); 2286 apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI); 2287 apic_pm_state.apic_ldr = apic_read(APIC_LDR); 2288 apic_pm_state.apic_dfr = apic_read(APIC_DFR); 2289 apic_pm_state.apic_spiv = apic_read(APIC_SPIV); 2290 apic_pm_state.apic_lvtt = apic_read(APIC_LVTT); 2291 if (maxlvt >= 4) 2292 apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC); 2293 apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0); 2294 apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1); 2295 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR); 2296 apic_pm_state.apic_tmict = apic_read(APIC_TMICT); 2297 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR); 2298 #ifdef CONFIG_X86_THERMAL_VECTOR 2299 if (maxlvt >= 5) 2300 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR); 2301 #endif 2302 2303 local_irq_save(flags); 2304 disable_local_APIC(); 2305 2306 irq_remapping_disable(); 2307 2308 local_irq_restore(flags); 2309 return 0; 2310 } 2311 2312 static void lapic_resume(void) 2313 { 2314 unsigned int l, h; 2315 unsigned long flags; 2316 int maxlvt; 2317 2318 if (!apic_pm_state.active) 2319 return; 2320 2321 local_irq_save(flags); 2322 2323 /* 2324 * IO-APIC and PIC have their own resume routines. 2325 * We just mask them here to make sure the interrupt 2326 * subsystem is completely quiet while we enable x2apic 2327 * and interrupt-remapping. 2328 */ 2329 mask_ioapic_entries(); 2330 legacy_pic->mask_all(); 2331 2332 if (x2apic_mode) { 2333 __x2apic_enable(); 2334 } else { 2335 /* 2336 * Make sure the APICBASE points to the right address 2337 * 2338 * FIXME! This will be wrong if we ever support suspend on 2339 * SMP! We'll need to do this as part of the CPU restore! 2340 */ 2341 if (boot_cpu_data.x86 >= 6) { 2342 rdmsr(MSR_IA32_APICBASE, l, h); 2343 l &= ~MSR_IA32_APICBASE_BASE; 2344 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr; 2345 wrmsr(MSR_IA32_APICBASE, l, h); 2346 } 2347 } 2348 2349 maxlvt = lapic_get_maxlvt(); 2350 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED); 2351 apic_write(APIC_ID, apic_pm_state.apic_id); 2352 apic_write(APIC_DFR, apic_pm_state.apic_dfr); 2353 apic_write(APIC_LDR, apic_pm_state.apic_ldr); 2354 apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri); 2355 apic_write(APIC_SPIV, apic_pm_state.apic_spiv); 2356 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0); 2357 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1); 2358 #if defined(CONFIG_X86_MCE_INTEL) 2359 if (maxlvt >= 5) 2360 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr); 2361 #endif 2362 if (maxlvt >= 4) 2363 apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc); 2364 apic_write(APIC_LVTT, apic_pm_state.apic_lvtt); 2365 apic_write(APIC_TDCR, apic_pm_state.apic_tdcr); 2366 apic_write(APIC_TMICT, apic_pm_state.apic_tmict); 2367 apic_write(APIC_ESR, 0); 2368 apic_read(APIC_ESR); 2369 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr); 2370 apic_write(APIC_ESR, 0); 2371 apic_read(APIC_ESR); 2372 2373 irq_remapping_reenable(x2apic_mode); 2374 2375 local_irq_restore(flags); 2376 } 2377 2378 /* 2379 * This device has no shutdown method - fully functioning local APICs 2380 * are needed on every CPU up until machine_halt/restart/poweroff. 2381 */ 2382 2383 static struct syscore_ops lapic_syscore_ops = { 2384 .resume = lapic_resume, 2385 .suspend = lapic_suspend, 2386 }; 2387 2388 static void apic_pm_activate(void) 2389 { 2390 apic_pm_state.active = 1; 2391 } 2392 2393 static int __init init_lapic_sysfs(void) 2394 { 2395 /* XXX: remove suspend/resume procs if !apic_pm_state.active? */ 2396 if (cpu_has_apic) 2397 register_syscore_ops(&lapic_syscore_ops); 2398 2399 return 0; 2400 } 2401 2402 /* local apic needs to resume before other devices access its registers. */ 2403 core_initcall(init_lapic_sysfs); 2404 2405 #else /* CONFIG_PM */ 2406 2407 static void apic_pm_activate(void) { } 2408 2409 #endif /* CONFIG_PM */ 2410 2411 #ifdef CONFIG_X86_64 2412 2413 static int multi_checked; 2414 static int multi; 2415 2416 static int set_multi(const struct dmi_system_id *d) 2417 { 2418 if (multi) 2419 return 0; 2420 pr_info("APIC: %s detected, Multi Chassis\n", d->ident); 2421 multi = 1; 2422 return 0; 2423 } 2424 2425 static const struct dmi_system_id multi_dmi_table[] = { 2426 { 2427 .callback = set_multi, 2428 .ident = "IBM System Summit2", 2429 .matches = { 2430 DMI_MATCH(DMI_SYS_VENDOR, "IBM"), 2431 DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"), 2432 }, 2433 }, 2434 {} 2435 }; 2436 2437 static void dmi_check_multi(void) 2438 { 2439 if (multi_checked) 2440 return; 2441 2442 dmi_check_system(multi_dmi_table); 2443 multi_checked = 1; 2444 } 2445 2446 /* 2447 * apic_is_clustered_box() -- Check if we can expect good TSC 2448 * 2449 * Thus far, the major user of this is IBM's Summit2 series: 2450 * Clustered boxes may have unsynced TSC problems if they are 2451 * multi-chassis. 2452 * Use DMI to check them 2453 */ 2454 int apic_is_clustered_box(void) 2455 { 2456 dmi_check_multi(); 2457 return multi; 2458 } 2459 #endif 2460 2461 /* 2462 * APIC command line parameters 2463 */ 2464 static int __init setup_disableapic(char *arg) 2465 { 2466 disable_apic = 1; 2467 setup_clear_cpu_cap(X86_FEATURE_APIC); 2468 return 0; 2469 } 2470 early_param("disableapic", setup_disableapic); 2471 2472 /* same as disableapic, for compatibility */ 2473 static int __init setup_nolapic(char *arg) 2474 { 2475 return setup_disableapic(arg); 2476 } 2477 early_param("nolapic", setup_nolapic); 2478 2479 static int __init parse_lapic_timer_c2_ok(char *arg) 2480 { 2481 local_apic_timer_c2_ok = 1; 2482 return 0; 2483 } 2484 early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok); 2485 2486 static int __init parse_disable_apic_timer(char *arg) 2487 { 2488 disable_apic_timer = 1; 2489 return 0; 2490 } 2491 early_param("noapictimer", parse_disable_apic_timer); 2492 2493 static int __init parse_nolapic_timer(char *arg) 2494 { 2495 disable_apic_timer = 1; 2496 return 0; 2497 } 2498 early_param("nolapic_timer", parse_nolapic_timer); 2499 2500 static int __init apic_set_verbosity(char *arg) 2501 { 2502 if (!arg) { 2503 #ifdef CONFIG_X86_64 2504 skip_ioapic_setup = 0; 2505 return 0; 2506 #endif 2507 return -EINVAL; 2508 } 2509 2510 if (strcmp("debug", arg) == 0) 2511 apic_verbosity = APIC_DEBUG; 2512 else if (strcmp("verbose", arg) == 0) 2513 apic_verbosity = APIC_VERBOSE; 2514 else { 2515 pr_warning("APIC Verbosity level %s not recognised" 2516 " use apic=verbose or apic=debug\n", arg); 2517 return -EINVAL; 2518 } 2519 2520 return 0; 2521 } 2522 early_param("apic", apic_set_verbosity); 2523 2524 static int __init lapic_insert_resource(void) 2525 { 2526 if (!apic_phys) 2527 return -1; 2528 2529 /* Put local APIC into the resource map. */ 2530 lapic_resource.start = apic_phys; 2531 lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1; 2532 insert_resource(&iomem_resource, &lapic_resource); 2533 2534 return 0; 2535 } 2536 2537 /* 2538 * need call insert after e820_reserve_resources() 2539 * that is using request_resource 2540 */ 2541 late_initcall(lapic_insert_resource); 2542 2543 static int __init apic_set_disabled_cpu_apicid(char *arg) 2544 { 2545 if (!arg || !get_option(&arg, &disabled_cpu_apicid)) 2546 return -EINVAL; 2547 2548 return 0; 2549 } 2550 early_param("disable_cpu_apicid", apic_set_disabled_cpu_apicid); 2551