Linux Audio

Check our new training course

Loading...
v5.4
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *	Local APIC handling, local APIC timers
   4 *
   5 *	(c) 1999, 2000, 2009 Ingo Molnar <mingo@redhat.com>
   6 *
   7 *	Fixes
   8 *	Maciej W. Rozycki	:	Bits for genuine 82489DX APICs;
   9 *					thanks to Eric Gilmore
  10 *					and Rolf G. Tews
  11 *					for testing these extensively.
  12 *	Maciej W. Rozycki	:	Various updates and fixes.
  13 *	Mikael Pettersson	:	Power Management for UP-APIC.
  14 *	Pavel Machek and
  15 *	Mikael Pettersson	:	PM converted to driver model.
  16 */
  17
  18#include <linux/perf_event.h>
  19#include <linux/kernel_stat.h>
  20#include <linux/mc146818rtc.h>
  21#include <linux/acpi_pmtmr.h>
  22#include <linux/clockchips.h>
  23#include <linux/interrupt.h>
  24#include <linux/memblock.h>
  25#include <linux/ftrace.h>
  26#include <linux/ioport.h>
  27#include <linux/export.h>
  28#include <linux/syscore_ops.h>
  29#include <linux/delay.h>
  30#include <linux/timex.h>
  31#include <linux/i8253.h>
  32#include <linux/dmar.h>
  33#include <linux/init.h>
  34#include <linux/cpu.h>
  35#include <linux/dmi.h>
  36#include <linux/smp.h>
  37#include <linux/mm.h>
  38
  39#include <asm/trace/irq_vectors.h>
  40#include <asm/irq_remapping.h>
  41#include <asm/perf_event.h>
  42#include <asm/x86_init.h>
  43#include <asm/pgalloc.h>
  44#include <linux/atomic.h>
  45#include <asm/mpspec.h>
  46#include <asm/i8259.h>
  47#include <asm/proto.h>
  48#include <asm/traps.h>
  49#include <asm/apic.h>
  50#include <asm/io_apic.h>
  51#include <asm/desc.h>
  52#include <asm/hpet.h>
 
  53#include <asm/mtrr.h>
  54#include <asm/time.h>
  55#include <asm/smp.h>
  56#include <asm/mce.h>
  57#include <asm/tsc.h>
  58#include <asm/hypervisor.h>
  59#include <asm/cpu_device_id.h>
  60#include <asm/intel-family.h>
  61#include <asm/irq_regs.h>
  62
  63unsigned int num_processors;
  64
  65unsigned disabled_cpus;
  66
  67/* Processor that is doing the boot up */
  68unsigned int boot_cpu_physical_apicid __ro_after_init = -1U;
  69EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid);
  70
  71u8 boot_cpu_apic_version __ro_after_init;
  72
  73/*
  74 * The highest APIC ID seen during enumeration.
  75 */
  76static unsigned int max_physical_apicid;
  77
  78/*
  79 * Bitmask of physically existing CPUs:
  80 */
  81physid_mask_t phys_cpu_present_map;
  82
  83/*
  84 * Processor to be disabled specified by kernel parameter
  85 * disable_cpu_apicid=<int>, mostly used for the kdump 2nd kernel to
  86 * avoid undefined behaviour caused by sending INIT from AP to BSP.
  87 */
  88static unsigned int disabled_cpu_apicid __ro_after_init = BAD_APICID;
  89
  90/*
  91 * This variable controls which CPUs receive external NMIs.  By default,
  92 * external NMIs are delivered only to the BSP.
  93 */
  94static int apic_extnmi __ro_after_init = APIC_EXTNMI_BSP;
  95
  96/*
  97 * Map cpu index to physical APIC ID
  98 */
  99DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_cpu_to_apicid, BAD_APICID);
 100DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_bios_cpu_apicid, BAD_APICID);
 101DEFINE_EARLY_PER_CPU_READ_MOSTLY(u32, x86_cpu_to_acpiid, U32_MAX);
 102EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
 103EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
 104EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_acpiid);
 105
 106#ifdef CONFIG_X86_32
 107
 108/*
 109 * On x86_32, the mapping between cpu and logical apicid may vary
 110 * depending on apic in use.  The following early percpu variable is
 111 * used for the mapping.  This is where the behaviors of x86_64 and 32
 112 * actually diverge.  Let's keep it ugly for now.
 113 */
 114DEFINE_EARLY_PER_CPU_READ_MOSTLY(int, x86_cpu_to_logical_apicid, BAD_APICID);
 115
 116/* Local APIC was disabled by the BIOS and enabled by the kernel */
 117static int enabled_via_apicbase __ro_after_init;
 118
 119/*
 120 * Handle interrupt mode configuration register (IMCR).
 121 * This register controls whether the interrupt signals
 122 * that reach the BSP come from the master PIC or from the
 123 * local APIC. Before entering Symmetric I/O Mode, either
 124 * the BIOS or the operating system must switch out of
 125 * PIC Mode by changing the IMCR.
 126 */
 127static inline void imcr_pic_to_apic(void)
 128{
 129	/* select IMCR register */
 130	outb(0x70, 0x22);
 131	/* NMI and 8259 INTR go through APIC */
 132	outb(0x01, 0x23);
 133}
 134
 135static inline void imcr_apic_to_pic(void)
 136{
 137	/* select IMCR register */
 138	outb(0x70, 0x22);
 139	/* NMI and 8259 INTR go directly to BSP */
 140	outb(0x00, 0x23);
 141}
 142#endif
 143
 144/*
 145 * Knob to control our willingness to enable the local APIC.
 146 *
 147 * +1=force-enable
 148 */
 149static int force_enable_local_apic __initdata;
 150
 151/*
 152 * APIC command line parameters
 153 */
 154static int __init parse_lapic(char *arg)
 155{
 156	if (IS_ENABLED(CONFIG_X86_32) && !arg)
 157		force_enable_local_apic = 1;
 158	else if (arg && !strncmp(arg, "notscdeadline", 13))
 159		setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
 160	return 0;
 161}
 162early_param("lapic", parse_lapic);
 163
 164#ifdef CONFIG_X86_64
 165static int apic_calibrate_pmtmr __initdata;
 166static __init int setup_apicpmtimer(char *s)
 167{
 168	apic_calibrate_pmtmr = 1;
 169	notsc_setup(NULL);
 170	return 0;
 171}
 172__setup("apicpmtimer", setup_apicpmtimer);
 173#endif
 174
 175unsigned long mp_lapic_addr __ro_after_init;
 176int disable_apic __ro_after_init;
 177/* Disable local APIC timer from the kernel commandline or via dmi quirk */
 178static int disable_apic_timer __initdata;
 179/* Local APIC timer works in C2 */
 180int local_apic_timer_c2_ok __ro_after_init;
 181EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
 182
 
 
 183/*
 184 * Debug level, exported for io_apic.c
 185 */
 186int apic_verbosity __ro_after_init;
 187
 188int pic_mode __ro_after_init;
 189
 190/* Have we found an MP table */
 191int smp_found_config __ro_after_init;
 192
 193static struct resource lapic_resource = {
 194	.name = "Local APIC",
 195	.flags = IORESOURCE_MEM | IORESOURCE_BUSY,
 196};
 197
 198unsigned int lapic_timer_period = 0;
 199
 200static void apic_pm_activate(void);
 201
 202static unsigned long apic_phys __ro_after_init;
 203
 204/*
 205 * Get the LAPIC version
 206 */
 207static inline int lapic_get_version(void)
 208{
 209	return GET_APIC_VERSION(apic_read(APIC_LVR));
 210}
 211
 212/*
 213 * Check, if the APIC is integrated or a separate chip
 214 */
 215static inline int lapic_is_integrated(void)
 216{
 
 
 
 217	return APIC_INTEGRATED(lapic_get_version());
 
 218}
 219
 220/*
 221 * Check, whether this is a modern or a first generation APIC
 222 */
 223static int modern_apic(void)
 224{
 225	/* AMD systems use old APIC versions, so check the CPU */
 226	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
 227	    boot_cpu_data.x86 >= 0xf)
 228		return 1;
 229
 230	/* Hygon systems use modern APIC */
 231	if (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)
 232		return 1;
 233
 234	return lapic_get_version() >= 0x14;
 235}
 236
 237/*
 238 * right after this call apic become NOOP driven
 239 * so apic->write/read doesn't do anything
 240 */
 241static void __init apic_disable(void)
 242{
 243	pr_info("APIC: switched to apic NOOP\n");
 244	apic = &apic_noop;
 245}
 246
 247void native_apic_wait_icr_idle(void)
 248{
 249	while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
 250		cpu_relax();
 251}
 252
 253u32 native_safe_apic_wait_icr_idle(void)
 254{
 255	u32 send_status;
 256	int timeout;
 257
 258	timeout = 0;
 259	do {
 260		send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
 261		if (!send_status)
 262			break;
 263		inc_irq_stat(icr_read_retry_count);
 264		udelay(100);
 265	} while (timeout++ < 1000);
 266
 267	return send_status;
 268}
 269
 270void native_apic_icr_write(u32 low, u32 id)
 271{
 272	unsigned long flags;
 273
 274	local_irq_save(flags);
 275	apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
 276	apic_write(APIC_ICR, low);
 277	local_irq_restore(flags);
 278}
 279
 280u64 native_apic_icr_read(void)
 281{
 282	u32 icr1, icr2;
 283
 284	icr2 = apic_read(APIC_ICR2);
 285	icr1 = apic_read(APIC_ICR);
 286
 287	return icr1 | ((u64)icr2 << 32);
 288}
 289
 290#ifdef CONFIG_X86_32
 291/**
 292 * get_physical_broadcast - Get number of physical broadcast IDs
 293 */
 294int get_physical_broadcast(void)
 295{
 296	return modern_apic() ? 0xff : 0xf;
 297}
 298#endif
 299
 300/**
 301 * lapic_get_maxlvt - get the maximum number of local vector table entries
 302 */
 303int lapic_get_maxlvt(void)
 304{
 
 
 
 305	/*
 306	 * - we always have APIC integrated on 64bit mode
 307	 * - 82489DXs do not report # of LVT entries
 308	 */
 309	return lapic_is_integrated() ? GET_APIC_MAXLVT(apic_read(APIC_LVR)) : 2;
 310}
 311
 312/*
 313 * Local APIC timer
 314 */
 315
 316/* Clock divisor */
 317#define APIC_DIVISOR 16
 318#define TSC_DIVISOR  8
 319
 320/*
 321 * This function sets up the local APIC timer, with a timeout of
 322 * 'clocks' APIC bus clock. During calibration we actually call
 323 * this function twice on the boot CPU, once with a bogus timeout
 324 * value, second time for real. The other (noncalibrating) CPUs
 325 * call this function only once, with the real, calibrated value.
 326 *
 327 * We do reads before writes even if unnecessary, to get around the
 328 * P5 APIC double write bug.
 329 */
 330static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
 331{
 332	unsigned int lvtt_value, tmp_value;
 333
 334	lvtt_value = LOCAL_TIMER_VECTOR;
 335	if (!oneshot)
 336		lvtt_value |= APIC_LVT_TIMER_PERIODIC;
 337	else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
 338		lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
 339
 340	if (!lapic_is_integrated())
 341		lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
 342
 343	if (!irqen)
 344		lvtt_value |= APIC_LVT_MASKED;
 345
 346	apic_write(APIC_LVTT, lvtt_value);
 347
 348	if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
 349		/*
 350		 * See Intel SDM: TSC-Deadline Mode chapter. In xAPIC mode,
 351		 * writing to the APIC LVTT and TSC_DEADLINE MSR isn't serialized.
 352		 * According to Intel, MFENCE can do the serialization here.
 353		 */
 354		asm volatile("mfence" : : : "memory");
 355
 356		printk_once(KERN_DEBUG "TSC deadline timer enabled\n");
 357		return;
 358	}
 359
 360	/*
 361	 * Divide PICLK by 16
 362	 */
 363	tmp_value = apic_read(APIC_TDCR);
 364	apic_write(APIC_TDCR,
 365		(tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
 366		APIC_TDR_DIV_16);
 367
 368	if (!oneshot)
 369		apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
 370}
 371
 372/*
 373 * Setup extended LVT, AMD specific
 374 *
 375 * Software should use the LVT offsets the BIOS provides.  The offsets
 376 * are determined by the subsystems using it like those for MCE
 377 * threshold or IBS.  On K8 only offset 0 (APIC500) and MCE interrupts
 378 * are supported. Beginning with family 10h at least 4 offsets are
 379 * available.
 380 *
 381 * Since the offsets must be consistent for all cores, we keep track
 382 * of the LVT offsets in software and reserve the offset for the same
 383 * vector also to be used on other cores. An offset is freed by
 384 * setting the entry to APIC_EILVT_MASKED.
 385 *
 386 * If the BIOS is right, there should be no conflicts. Otherwise a
 387 * "[Firmware Bug]: ..." error message is generated. However, if
 388 * software does not properly determines the offsets, it is not
 389 * necessarily a BIOS bug.
 390 */
 391
 392static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
 393
 394static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
 395{
 396	return (old & APIC_EILVT_MASKED)
 397		|| (new == APIC_EILVT_MASKED)
 398		|| ((new & ~APIC_EILVT_MASKED) == old);
 399}
 400
 401static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
 402{
 403	unsigned int rsvd, vector;
 404
 405	if (offset >= APIC_EILVT_NR_MAX)
 406		return ~0;
 407
 408	rsvd = atomic_read(&eilvt_offsets[offset]);
 409	do {
 410		vector = rsvd & ~APIC_EILVT_MASKED;	/* 0: unassigned */
 411		if (vector && !eilvt_entry_is_changeable(vector, new))
 412			/* may not change if vectors are different */
 413			return rsvd;
 414		rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
 415	} while (rsvd != new);
 416
 417	rsvd &= ~APIC_EILVT_MASKED;
 418	if (rsvd && rsvd != vector)
 419		pr_info("LVT offset %d assigned for vector 0x%02x\n",
 420			offset, rsvd);
 421
 422	return new;
 423}
 424
 425/*
 426 * If mask=1, the LVT entry does not generate interrupts while mask=0
 427 * enables the vector. See also the BKDGs. Must be called with
 428 * preemption disabled.
 429 */
 430
 431int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
 432{
 433	unsigned long reg = APIC_EILVTn(offset);
 434	unsigned int new, old, reserved;
 435
 436	new = (mask << 16) | (msg_type << 8) | vector;
 437	old = apic_read(reg);
 438	reserved = reserve_eilvt_offset(offset, new);
 439
 440	if (reserved != new) {
 441		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
 442		       "vector 0x%x, but the register is already in use for "
 443		       "vector 0x%x on another cpu\n",
 444		       smp_processor_id(), reg, offset, new, reserved);
 445		return -EINVAL;
 446	}
 447
 448	if (!eilvt_entry_is_changeable(old, new)) {
 449		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
 450		       "vector 0x%x, but the register is already in use for "
 451		       "vector 0x%x on this cpu\n",
 452		       smp_processor_id(), reg, offset, new, old);
 453		return -EBUSY;
 454	}
 455
 456	apic_write(reg, new);
 457
 458	return 0;
 459}
 460EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
 461
 462/*
 463 * Program the next event, relative to now
 464 */
 465static int lapic_next_event(unsigned long delta,
 466			    struct clock_event_device *evt)
 467{
 468	apic_write(APIC_TMICT, delta);
 469	return 0;
 470}
 471
 472static int lapic_next_deadline(unsigned long delta,
 473			       struct clock_event_device *evt)
 474{
 475	u64 tsc;
 476
 477	tsc = rdtsc();
 478	wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
 479	return 0;
 480}
 481
 482static int lapic_timer_shutdown(struct clock_event_device *evt)
 483{
 484	unsigned int v;
 485
 486	/* Lapic used as dummy for broadcast ? */
 487	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
 488		return 0;
 489
 490	v = apic_read(APIC_LVTT);
 491	v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
 492	apic_write(APIC_LVTT, v);
 493	apic_write(APIC_TMICT, 0);
 494	return 0;
 495}
 496
 497static inline int
 498lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot)
 499{
 500	/* Lapic used as dummy for broadcast ? */
 501	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
 502		return 0;
 503
 504	__setup_APIC_LVTT(lapic_timer_period, oneshot, 1);
 505	return 0;
 506}
 507
 508static int lapic_timer_set_periodic(struct clock_event_device *evt)
 509{
 510	return lapic_timer_set_periodic_oneshot(evt, false);
 511}
 512
 513static int lapic_timer_set_oneshot(struct clock_event_device *evt)
 514{
 515	return lapic_timer_set_periodic_oneshot(evt, true);
 516}
 517
 518/*
 519 * Local APIC timer broadcast function
 520 */
 521static void lapic_timer_broadcast(const struct cpumask *mask)
 522{
 523#ifdef CONFIG_SMP
 524	apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
 525#endif
 526}
 527
 528
 529/*
 530 * The local apic timer can be used for any function which is CPU local.
 531 */
 532static struct clock_event_device lapic_clockevent = {
 533	.name				= "lapic",
 534	.features			= CLOCK_EVT_FEAT_PERIODIC |
 535					  CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP
 536					  | CLOCK_EVT_FEAT_DUMMY,
 537	.shift				= 32,
 538	.set_state_shutdown		= lapic_timer_shutdown,
 539	.set_state_periodic		= lapic_timer_set_periodic,
 540	.set_state_oneshot		= lapic_timer_set_oneshot,
 541	.set_state_oneshot_stopped	= lapic_timer_shutdown,
 542	.set_next_event			= lapic_next_event,
 543	.broadcast			= lapic_timer_broadcast,
 544	.rating				= 100,
 545	.irq				= -1,
 546};
 547static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
 548
 549#define DEADLINE_MODEL_MATCH_FUNC(model, func)	\
 550	{ X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)&func }
 551
 552#define DEADLINE_MODEL_MATCH_REV(model, rev)	\
 553	{ X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)rev }
 554
 555static u32 hsx_deadline_rev(void)
 556{
 557	switch (boot_cpu_data.x86_stepping) {
 558	case 0x02: return 0x3a; /* EP */
 559	case 0x04: return 0x0f; /* EX */
 560	}
 561
 562	return ~0U;
 563}
 564
 565static u32 bdx_deadline_rev(void)
 566{
 567	switch (boot_cpu_data.x86_stepping) {
 568	case 0x02: return 0x00000011;
 569	case 0x03: return 0x0700000e;
 570	case 0x04: return 0x0f00000c;
 571	case 0x05: return 0x0e000003;
 572	}
 573
 574	return ~0U;
 575}
 576
 577static u32 skx_deadline_rev(void)
 578{
 579	switch (boot_cpu_data.x86_stepping) {
 580	case 0x03: return 0x01000136;
 581	case 0x04: return 0x02000014;
 582	}
 583
 584	if (boot_cpu_data.x86_stepping > 4)
 585		return 0;
 586
 587	return ~0U;
 588}
 589
 590static const struct x86_cpu_id deadline_match[] = {
 591	DEADLINE_MODEL_MATCH_FUNC( INTEL_FAM6_HASWELL_X,	hsx_deadline_rev),
 592	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_X,	0x0b000020),
 593	DEADLINE_MODEL_MATCH_FUNC( INTEL_FAM6_BROADWELL_D,	bdx_deadline_rev),
 594	DEADLINE_MODEL_MATCH_FUNC( INTEL_FAM6_SKYLAKE_X,	skx_deadline_rev),
 595
 596	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL,		0x22),
 597	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_L,	0x20),
 598	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_G,	0x17),
 599
 600	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL,	0x25),
 601	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_G,	0x17),
 602
 603	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_SKYLAKE_L,	0xb2),
 604	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_SKYLAKE,		0xb2),
 605
 606	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_KABYLAKE_L,	0x52),
 607	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_KABYLAKE,		0x52),
 608
 609	{},
 610};
 611
 612static void apic_check_deadline_errata(void)
 613{
 614	const struct x86_cpu_id *m;
 615	u32 rev;
 616
 617	if (!boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER) ||
 618	    boot_cpu_has(X86_FEATURE_HYPERVISOR))
 619		return;
 620
 621	m = x86_match_cpu(deadline_match);
 622	if (!m)
 623		return;
 624
 625	/*
 626	 * Function pointers will have the MSB set due to address layout,
 627	 * immediate revisions will not.
 628	 */
 629	if ((long)m->driver_data < 0)
 630		rev = ((u32 (*)(void))(m->driver_data))();
 631	else
 632		rev = (u32)m->driver_data;
 633
 634	if (boot_cpu_data.microcode >= rev)
 635		return;
 636
 637	setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
 638	pr_err(FW_BUG "TSC_DEADLINE disabled due to Errata; "
 639	       "please update microcode to version: 0x%x (or later)\n", rev);
 640}
 641
 642/*
 643 * Setup the local APIC timer for this CPU. Copy the initialized values
 644 * of the boot CPU and register the clock event in the framework.
 645 */
 646static void setup_APIC_timer(void)
 647{
 648	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
 649
 650	if (this_cpu_has(X86_FEATURE_ARAT)) {
 651		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
 652		/* Make LAPIC timer preferrable over percpu HPET */
 653		lapic_clockevent.rating = 150;
 654	}
 655
 656	memcpy(levt, &lapic_clockevent, sizeof(*levt));
 657	levt->cpumask = cpumask_of(smp_processor_id());
 658
 659	if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
 660		levt->name = "lapic-deadline";
 661		levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
 662				    CLOCK_EVT_FEAT_DUMMY);
 663		levt->set_next_event = lapic_next_deadline;
 664		clockevents_config_and_register(levt,
 665						tsc_khz * (1000 / TSC_DIVISOR),
 666						0xF, ~0UL);
 667	} else
 668		clockevents_register_device(levt);
 669}
 670
 671/*
 672 * Install the updated TSC frequency from recalibration at the TSC
 673 * deadline clockevent devices.
 674 */
 675static void __lapic_update_tsc_freq(void *info)
 676{
 677	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
 678
 679	if (!this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
 680		return;
 681
 682	clockevents_update_freq(levt, tsc_khz * (1000 / TSC_DIVISOR));
 683}
 684
 685void lapic_update_tsc_freq(void)
 686{
 687	/*
 688	 * The clockevent device's ->mult and ->shift can both be
 689	 * changed. In order to avoid races, schedule the frequency
 690	 * update code on each CPU.
 691	 */
 692	on_each_cpu(__lapic_update_tsc_freq, NULL, 0);
 693}
 694
 695/*
 696 * In this functions we calibrate APIC bus clocks to the external timer.
 697 *
 698 * We want to do the calibration only once since we want to have local timer
 699 * irqs syncron. CPUs connected by the same APIC bus have the very same bus
 700 * frequency.
 701 *
 702 * This was previously done by reading the PIT/HPET and waiting for a wrap
 703 * around to find out, that a tick has elapsed. I have a box, where the PIT
 704 * readout is broken, so it never gets out of the wait loop again. This was
 705 * also reported by others.
 706 *
 707 * Monitoring the jiffies value is inaccurate and the clockevents
 708 * infrastructure allows us to do a simple substitution of the interrupt
 709 * handler.
 710 *
 711 * The calibration routine also uses the pm_timer when possible, as the PIT
 712 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
 713 * back to normal later in the boot process).
 714 */
 715
 716#define LAPIC_CAL_LOOPS		(HZ/10)
 717
 718static __initdata int lapic_cal_loops = -1;
 719static __initdata long lapic_cal_t1, lapic_cal_t2;
 720static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
 721static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
 722static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
 723
 724/*
 725 * Temporary interrupt handler and polled calibration function.
 726 */
 727static void __init lapic_cal_handler(struct clock_event_device *dev)
 728{
 729	unsigned long long tsc = 0;
 730	long tapic = apic_read(APIC_TMCCT);
 731	unsigned long pm = acpi_pm_read_early();
 732
 733	if (boot_cpu_has(X86_FEATURE_TSC))
 734		tsc = rdtsc();
 735
 736	switch (lapic_cal_loops++) {
 737	case 0:
 738		lapic_cal_t1 = tapic;
 739		lapic_cal_tsc1 = tsc;
 740		lapic_cal_pm1 = pm;
 741		lapic_cal_j1 = jiffies;
 742		break;
 743
 744	case LAPIC_CAL_LOOPS:
 745		lapic_cal_t2 = tapic;
 746		lapic_cal_tsc2 = tsc;
 747		if (pm < lapic_cal_pm1)
 748			pm += ACPI_PM_OVRRUN;
 749		lapic_cal_pm2 = pm;
 750		lapic_cal_j2 = jiffies;
 751		break;
 752	}
 753}
 754
 755static int __init
 756calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
 757{
 758	const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
 759	const long pm_thresh = pm_100ms / 100;
 760	unsigned long mult;
 761	u64 res;
 762
 763#ifndef CONFIG_X86_PM_TIMER
 764	return -1;
 765#endif
 766
 767	apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
 768
 769	/* Check, if the PM timer is available */
 770	if (!deltapm)
 771		return -1;
 772
 773	mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
 774
 775	if (deltapm > (pm_100ms - pm_thresh) &&
 776	    deltapm < (pm_100ms + pm_thresh)) {
 777		apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
 778		return 0;
 779	}
 780
 781	res = (((u64)deltapm) *  mult) >> 22;
 782	do_div(res, 1000000);
 783	pr_warning("APIC calibration not consistent "
 784		   "with PM-Timer: %ldms instead of 100ms\n",(long)res);
 785
 786	/* Correct the lapic counter value */
 787	res = (((u64)(*delta)) * pm_100ms);
 788	do_div(res, deltapm);
 789	pr_info("APIC delta adjusted to PM-Timer: "
 790		"%lu (%ld)\n", (unsigned long)res, *delta);
 791	*delta = (long)res;
 792
 793	/* Correct the tsc counter value */
 794	if (boot_cpu_has(X86_FEATURE_TSC)) {
 795		res = (((u64)(*deltatsc)) * pm_100ms);
 796		do_div(res, deltapm);
 797		apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
 798					  "PM-Timer: %lu (%ld)\n",
 799					(unsigned long)res, *deltatsc);
 800		*deltatsc = (long)res;
 801	}
 802
 803	return 0;
 804}
 805
 806static int __init lapic_init_clockevent(void)
 807{
 808	if (!lapic_timer_period)
 809		return -1;
 810
 811	/* Calculate the scaled math multiplication factor */
 812	lapic_clockevent.mult = div_sc(lapic_timer_period/APIC_DIVISOR,
 813					TICK_NSEC, lapic_clockevent.shift);
 814	lapic_clockevent.max_delta_ns =
 815		clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
 816	lapic_clockevent.max_delta_ticks = 0x7FFFFFFF;
 817	lapic_clockevent.min_delta_ns =
 818		clockevent_delta2ns(0xF, &lapic_clockevent);
 819	lapic_clockevent.min_delta_ticks = 0xF;
 820
 821	return 0;
 822}
 823
 824bool __init apic_needs_pit(void)
 825{
 826	/*
 827	 * If the frequencies are not known, PIT is required for both TSC
 828	 * and apic timer calibration.
 829	 */
 830	if (!tsc_khz || !cpu_khz)
 831		return true;
 832
 833	/* Is there an APIC at all? */
 834	if (!boot_cpu_has(X86_FEATURE_APIC))
 835		return true;
 836
 837	/* Virt guests may lack ARAT, but still have DEADLINE */
 838	if (!boot_cpu_has(X86_FEATURE_ARAT))
 839		return true;
 840
 841	/* Deadline timer is based on TSC so no further PIT action required */
 842	if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
 843		return false;
 844
 845	/* APIC timer disabled? */
 846	if (disable_apic_timer)
 847		return true;
 848	/*
 849	 * The APIC timer frequency is known already, no PIT calibration
 850	 * required. If unknown, let the PIT be initialized.
 851	 */
 852	return lapic_timer_period == 0;
 853}
 854
 855static int __init calibrate_APIC_clock(void)
 856{
 857	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
 858	u64 tsc_perj = 0, tsc_start = 0;
 859	unsigned long jif_start;
 860	unsigned long deltaj;
 861	long delta, deltatsc;
 862	int pm_referenced = 0;
 863
 864	if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
 865		return 0;
 866
 867	/*
 868	 * Check if lapic timer has already been calibrated by platform
 869	 * specific routine, such as tsc calibration code. If so just fill
 870	 * in the clockevent structure and return.
 871	 */
 872	if (!lapic_init_clockevent()) {
 
 
 
 873		apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
 874			    lapic_timer_period);
 875		/*
 876		 * Direct calibration methods must have an always running
 877		 * local APIC timer, no need for broadcast timer.
 878		 */
 
 
 879		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
 880		return 0;
 881	}
 882
 883	apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
 884		    "calibrating APIC timer ...\n");
 885
 886	/*
 887	 * There are platforms w/o global clockevent devices. Instead of
 888	 * making the calibration conditional on that, use a polling based
 889	 * approach everywhere.
 890	 */
 891	local_irq_disable();
 892
 
 
 
 
 893	/*
 894	 * Setup the APIC counter to maximum. There is no way the lapic
 895	 * can underflow in the 100ms detection time frame
 896	 */
 897	__setup_APIC_LVTT(0xffffffff, 0, 0);
 898
 899	/*
 900	 * Methods to terminate the calibration loop:
 901	 *  1) Global clockevent if available (jiffies)
 902	 *  2) TSC if available and frequency is known
 903	 */
 904	jif_start = READ_ONCE(jiffies);
 905
 906	if (tsc_khz) {
 907		tsc_start = rdtsc();
 908		tsc_perj = div_u64((u64)tsc_khz * 1000, HZ);
 909	}
 910
 911	/*
 912	 * Enable interrupts so the tick can fire, if a global
 913	 * clockevent device is available
 914	 */
 915	local_irq_enable();
 916
 917	while (lapic_cal_loops <= LAPIC_CAL_LOOPS) {
 918		/* Wait for a tick to elapse */
 919		while (1) {
 920			if (tsc_khz) {
 921				u64 tsc_now = rdtsc();
 922				if ((tsc_now - tsc_start) >= tsc_perj) {
 923					tsc_start += tsc_perj;
 924					break;
 925				}
 926			} else {
 927				unsigned long jif_now = READ_ONCE(jiffies);
 928
 929				if (time_after(jif_now, jif_start)) {
 930					jif_start = jif_now;
 931					break;
 932				}
 933			}
 934			cpu_relax();
 935		}
 936
 937		/* Invoke the calibration routine */
 938		local_irq_disable();
 939		lapic_cal_handler(NULL);
 940		local_irq_enable();
 941	}
 942
 943	local_irq_disable();
 944
 
 
 
 945	/* Build delta t1-t2 as apic timer counts down */
 946	delta = lapic_cal_t1 - lapic_cal_t2;
 947	apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
 948
 949	deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
 950
 951	/* we trust the PM based calibration if possible */
 952	pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
 953					&delta, &deltatsc);
 954
 955	lapic_timer_period = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
 956	lapic_init_clockevent();
 
 
 
 
 
 
 
 957
 958	apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
 959	apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
 960	apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
 961		    lapic_timer_period);
 962
 963	if (boot_cpu_has(X86_FEATURE_TSC)) {
 964		apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
 965			    "%ld.%04ld MHz.\n",
 966			    (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
 967			    (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
 968	}
 969
 970	apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
 971		    "%u.%04u MHz.\n",
 972		    lapic_timer_period / (1000000 / HZ),
 973		    lapic_timer_period % (1000000 / HZ));
 974
 975	/*
 976	 * Do a sanity check on the APIC calibration result
 977	 */
 978	if (lapic_timer_period < (1000000 / HZ)) {
 979		local_irq_enable();
 980		pr_warning("APIC frequency too slow, disabling apic timer\n");
 981		return -1;
 982	}
 983
 984	levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
 985
 986	/*
 987	 * PM timer calibration failed or not turned on so lets try APIC
 988	 * timer based calibration, if a global clockevent device is
 989	 * available.
 990	 */
 991	if (!pm_referenced && global_clock_event) {
 992		apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
 993
 994		/*
 995		 * Setup the apic timer manually
 996		 */
 997		levt->event_handler = lapic_cal_handler;
 998		lapic_timer_set_periodic(levt);
 999		lapic_cal_loops = -1;
1000
1001		/* Let the interrupts run */
1002		local_irq_enable();
1003
1004		while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
1005			cpu_relax();
1006
1007		/* Stop the lapic timer */
1008		local_irq_disable();
1009		lapic_timer_shutdown(levt);
1010
1011		/* Jiffies delta */
1012		deltaj = lapic_cal_j2 - lapic_cal_j1;
1013		apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
1014
1015		/* Check, if the jiffies result is consistent */
1016		if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
1017			apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
1018		else
1019			levt->features |= CLOCK_EVT_FEAT_DUMMY;
1020	}
1021	local_irq_enable();
1022
1023	if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
1024		pr_warning("APIC timer disabled due to verification failure\n");
1025		return -1;
1026	}
1027
1028	return 0;
1029}
1030
1031/*
1032 * Setup the boot APIC
1033 *
1034 * Calibrate and verify the result.
1035 */
1036void __init setup_boot_APIC_clock(void)
1037{
1038	/*
1039	 * The local apic timer can be disabled via the kernel
1040	 * commandline or from the CPU detection code. Register the lapic
1041	 * timer as a dummy clock event source on SMP systems, so the
1042	 * broadcast mechanism is used. On UP systems simply ignore it.
1043	 */
1044	if (disable_apic_timer) {
1045		pr_info("Disabling APIC timer\n");
1046		/* No broadcast on UP ! */
1047		if (num_possible_cpus() > 1) {
1048			lapic_clockevent.mult = 1;
1049			setup_APIC_timer();
1050		}
1051		return;
1052	}
1053
1054	if (calibrate_APIC_clock()) {
1055		/* No broadcast on UP ! */
1056		if (num_possible_cpus() > 1)
1057			setup_APIC_timer();
1058		return;
1059	}
1060
1061	/*
1062	 * If nmi_watchdog is set to IO_APIC, we need the
1063	 * PIT/HPET going.  Otherwise register lapic as a dummy
1064	 * device.
1065	 */
1066	lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
1067
1068	/* Setup the lapic or request the broadcast */
1069	setup_APIC_timer();
1070	amd_e400_c1e_apic_setup();
1071}
1072
1073void setup_secondary_APIC_clock(void)
1074{
1075	setup_APIC_timer();
1076	amd_e400_c1e_apic_setup();
1077}
1078
1079/*
1080 * The guts of the apic timer interrupt
1081 */
1082static void local_apic_timer_interrupt(void)
1083{
1084	struct clock_event_device *evt = this_cpu_ptr(&lapic_events);
 
1085
1086	/*
1087	 * Normally we should not be here till LAPIC has been initialized but
1088	 * in some cases like kdump, its possible that there is a pending LAPIC
1089	 * timer interrupt from previous kernel's context and is delivered in
1090	 * new kernel the moment interrupts are enabled.
1091	 *
1092	 * Interrupts are enabled early and LAPIC is setup much later, hence
1093	 * its possible that when we get here evt->event_handler is NULL.
1094	 * Check for event_handler being NULL and discard the interrupt as
1095	 * spurious.
1096	 */
1097	if (!evt->event_handler) {
1098		pr_warning("Spurious LAPIC timer interrupt on cpu %d\n",
1099			   smp_processor_id());
1100		/* Switch it off */
1101		lapic_timer_shutdown(evt);
1102		return;
1103	}
1104
1105	/*
1106	 * the NMI deadlock-detector uses this.
1107	 */
1108	inc_irq_stat(apic_timer_irqs);
1109
1110	evt->event_handler(evt);
1111}
1112
1113/*
1114 * Local APIC timer interrupt. This is the most natural way for doing
1115 * local interrupts, but local timer interrupts can be emulated by
1116 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
1117 *
1118 * [ if a single-CPU system runs an SMP kernel then we call the local
1119 *   interrupt as well. Thus we cannot inline the local irq ... ]
1120 */
1121__visible void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
1122{
1123	struct pt_regs *old_regs = set_irq_regs(regs);
1124
1125	/*
1126	 * NOTE! We'd better ACK the irq immediately,
1127	 * because timer handling can be slow.
1128	 *
1129	 * update_process_times() expects us to have done irq_enter().
1130	 * Besides, if we don't timer interrupts ignore the global
1131	 * interrupt lock, which is the WrongThing (tm) to do.
1132	 */
1133	entering_ack_irq();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1134	trace_local_timer_entry(LOCAL_TIMER_VECTOR);
1135	local_apic_timer_interrupt();
1136	trace_local_timer_exit(LOCAL_TIMER_VECTOR);
1137	exiting_irq();
1138
1139	set_irq_regs(old_regs);
1140}
1141
1142int setup_profiling_timer(unsigned int multiplier)
1143{
1144	return -EINVAL;
1145}
1146
1147/*
1148 * Local APIC start and shutdown
1149 */
1150
1151/**
1152 * clear_local_APIC - shutdown the local APIC
1153 *
1154 * This is called, when a CPU is disabled and before rebooting, so the state of
1155 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
1156 * leftovers during boot.
1157 */
1158void clear_local_APIC(void)
1159{
1160	int maxlvt;
1161	u32 v;
1162
1163	/* APIC hasn't been mapped yet */
1164	if (!x2apic_mode && !apic_phys)
1165		return;
1166
1167	maxlvt = lapic_get_maxlvt();
1168	/*
1169	 * Masking an LVT entry can trigger a local APIC error
1170	 * if the vector is zero. Mask LVTERR first to prevent this.
1171	 */
1172	if (maxlvt >= 3) {
1173		v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
1174		apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
1175	}
1176	/*
1177	 * Careful: we have to set masks only first to deassert
1178	 * any level-triggered sources.
1179	 */
1180	v = apic_read(APIC_LVTT);
1181	apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
1182	v = apic_read(APIC_LVT0);
1183	apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1184	v = apic_read(APIC_LVT1);
1185	apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1186	if (maxlvt >= 4) {
1187		v = apic_read(APIC_LVTPC);
1188		apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1189	}
1190
1191	/* lets not touch this if we didn't frob it */
1192#ifdef CONFIG_X86_THERMAL_VECTOR
1193	if (maxlvt >= 5) {
1194		v = apic_read(APIC_LVTTHMR);
1195		apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1196	}
1197#endif
1198#ifdef CONFIG_X86_MCE_INTEL
1199	if (maxlvt >= 6) {
1200		v = apic_read(APIC_LVTCMCI);
1201		if (!(v & APIC_LVT_MASKED))
1202			apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1203	}
1204#endif
1205
1206	/*
1207	 * Clean APIC state for other OSs:
1208	 */
1209	apic_write(APIC_LVTT, APIC_LVT_MASKED);
1210	apic_write(APIC_LVT0, APIC_LVT_MASKED);
1211	apic_write(APIC_LVT1, APIC_LVT_MASKED);
1212	if (maxlvt >= 3)
1213		apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1214	if (maxlvt >= 4)
1215		apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1216
1217	/* Integrated APIC (!82489DX) ? */
1218	if (lapic_is_integrated()) {
1219		if (maxlvt > 3)
1220			/* Clear ESR due to Pentium errata 3AP and 11AP */
1221			apic_write(APIC_ESR, 0);
1222		apic_read(APIC_ESR);
1223	}
1224}
1225
1226/**
1227 * apic_soft_disable - Clears and software disables the local APIC on hotplug
1228 *
1229 * Contrary to disable_local_APIC() this does not touch the enable bit in
1230 * MSR_IA32_APICBASE. Clearing that bit on systems based on the 3 wire APIC
1231 * bus would require a hardware reset as the APIC would lose track of bus
1232 * arbitration. On systems with FSB delivery APICBASE could be disabled,
1233 * but it has to be guaranteed that no interrupt is sent to the APIC while
1234 * in that state and it's not clear from the SDM whether it still responds
1235 * to INIT/SIPI messages. Stay on the safe side and use software disable.
1236 */
1237void apic_soft_disable(void)
1238{
1239	u32 value;
1240
1241	clear_local_APIC();
1242
1243	/* Soft disable APIC (implies clearing of registers for 82489DX!). */
1244	value = apic_read(APIC_SPIV);
1245	value &= ~APIC_SPIV_APIC_ENABLED;
1246	apic_write(APIC_SPIV, value);
1247}
1248
1249/**
1250 * disable_local_APIC - clear and disable the local APIC
1251 */
1252void disable_local_APIC(void)
1253{
 
 
1254	/* APIC hasn't been mapped yet */
1255	if (!x2apic_mode && !apic_phys)
1256		return;
1257
1258	apic_soft_disable();
 
 
 
 
 
 
 
 
1259
1260#ifdef CONFIG_X86_32
1261	/*
1262	 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1263	 * restore the disabled state.
1264	 */
1265	if (enabled_via_apicbase) {
1266		unsigned int l, h;
1267
1268		rdmsr(MSR_IA32_APICBASE, l, h);
1269		l &= ~MSR_IA32_APICBASE_ENABLE;
1270		wrmsr(MSR_IA32_APICBASE, l, h);
1271	}
1272#endif
1273}
1274
1275/*
1276 * If Linux enabled the LAPIC against the BIOS default disable it down before
1277 * re-entering the BIOS on shutdown.  Otherwise the BIOS may get confused and
1278 * not power-off.  Additionally clear all LVT entries before disable_local_APIC
1279 * for the case where Linux didn't enable the LAPIC.
1280 */
1281void lapic_shutdown(void)
1282{
1283	unsigned long flags;
1284
1285	if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
1286		return;
1287
1288	local_irq_save(flags);
1289
1290#ifdef CONFIG_X86_32
1291	if (!enabled_via_apicbase)
1292		clear_local_APIC();
1293	else
1294#endif
1295		disable_local_APIC();
1296
1297
1298	local_irq_restore(flags);
1299}
1300
1301/**
1302 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1303 */
1304void __init sync_Arb_IDs(void)
1305{
1306	/*
1307	 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1308	 * needed on AMD.
1309	 */
1310	if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1311		return;
1312
1313	/*
1314	 * Wait for idle.
1315	 */
1316	apic_wait_icr_idle();
1317
1318	apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1319	apic_write(APIC_ICR, APIC_DEST_ALLINC |
1320			APIC_INT_LEVELTRIG | APIC_DM_INIT);
1321}
1322
1323enum apic_intr_mode_id apic_intr_mode __ro_after_init;
1324
1325static int __init apic_intr_mode_select(void)
1326{
1327	/* Check kernel option */
1328	if (disable_apic) {
1329		pr_info("APIC disabled via kernel command line\n");
1330		return APIC_PIC;
1331	}
1332
1333	/* Check BIOS */
1334#ifdef CONFIG_X86_64
1335	/* On 64-bit, the APIC must be integrated, Check local APIC only */
1336	if (!boot_cpu_has(X86_FEATURE_APIC)) {
1337		disable_apic = 1;
1338		pr_info("APIC disabled by BIOS\n");
1339		return APIC_PIC;
1340	}
1341#else
1342	/* On 32-bit, the APIC may be integrated APIC or 82489DX */
1343
1344	/* Neither 82489DX nor integrated APIC ? */
1345	if (!boot_cpu_has(X86_FEATURE_APIC) && !smp_found_config) {
1346		disable_apic = 1;
1347		return APIC_PIC;
1348	}
1349
1350	/* If the BIOS pretends there is an integrated APIC ? */
1351	if (!boot_cpu_has(X86_FEATURE_APIC) &&
1352		APIC_INTEGRATED(boot_cpu_apic_version)) {
1353		disable_apic = 1;
1354		pr_err(FW_BUG "Local APIC %d not detected, force emulation\n",
1355				       boot_cpu_physical_apicid);
1356		return APIC_PIC;
1357	}
1358#endif
1359
1360	/* Check MP table or ACPI MADT configuration */
1361	if (!smp_found_config) {
1362		disable_ioapic_support();
1363		if (!acpi_lapic) {
1364			pr_info("APIC: ACPI MADT or MP tables are not detected\n");
1365			return APIC_VIRTUAL_WIRE_NO_CONFIG;
1366		}
1367		return APIC_VIRTUAL_WIRE;
1368	}
1369
1370#ifdef CONFIG_SMP
1371	/* If SMP should be disabled, then really disable it! */
1372	if (!setup_max_cpus) {
1373		pr_info("APIC: SMP mode deactivated\n");
1374		return APIC_SYMMETRIC_IO_NO_ROUTING;
1375	}
1376
1377	if (read_apic_id() != boot_cpu_physical_apicid) {
1378		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1379		     read_apic_id(), boot_cpu_physical_apicid);
1380		/* Or can we switch back to PIC here? */
1381	}
1382#endif
1383
1384	return APIC_SYMMETRIC_IO;
1385}
1386
1387/*
1388 * An initial setup of the virtual wire mode.
1389 */
1390void __init init_bsp_APIC(void)
1391{
1392	unsigned int value;
1393
1394	/*
1395	 * Don't do the setup now if we have a SMP BIOS as the
1396	 * through-I/O-APIC virtual wire mode might be active.
1397	 */
1398	if (smp_found_config || !boot_cpu_has(X86_FEATURE_APIC))
1399		return;
1400
1401	/*
1402	 * Do not trust the local APIC being empty at bootup.
1403	 */
1404	clear_local_APIC();
1405
1406	/*
1407	 * Enable APIC.
1408	 */
1409	value = apic_read(APIC_SPIV);
1410	value &= ~APIC_VECTOR_MASK;
1411	value |= APIC_SPIV_APIC_ENABLED;
1412
1413#ifdef CONFIG_X86_32
1414	/* This bit is reserved on P4/Xeon and should be cleared */
1415	if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1416	    (boot_cpu_data.x86 == 15))
1417		value &= ~APIC_SPIV_FOCUS_DISABLED;
1418	else
1419#endif
1420		value |= APIC_SPIV_FOCUS_DISABLED;
1421	value |= SPURIOUS_APIC_VECTOR;
1422	apic_write(APIC_SPIV, value);
1423
1424	/*
1425	 * Set up the virtual wire mode.
1426	 */
1427	apic_write(APIC_LVT0, APIC_DM_EXTINT);
1428	value = APIC_DM_NMI;
1429	if (!lapic_is_integrated())		/* 82489DX */
1430		value |= APIC_LVT_LEVEL_TRIGGER;
1431	if (apic_extnmi == APIC_EXTNMI_NONE)
1432		value |= APIC_LVT_MASKED;
1433	apic_write(APIC_LVT1, value);
1434}
1435
1436static void __init apic_bsp_setup(bool upmode);
1437
1438/* Init the interrupt delivery mode for the BSP */
1439void __init apic_intr_mode_init(void)
1440{
1441	bool upmode = IS_ENABLED(CONFIG_UP_LATE_INIT);
1442
1443	apic_intr_mode = apic_intr_mode_select();
1444
1445	switch (apic_intr_mode) {
1446	case APIC_PIC:
1447		pr_info("APIC: Keep in PIC mode(8259)\n");
1448		return;
1449	case APIC_VIRTUAL_WIRE:
1450		pr_info("APIC: Switch to virtual wire mode setup\n");
1451		default_setup_apic_routing();
1452		break;
1453	case APIC_VIRTUAL_WIRE_NO_CONFIG:
1454		pr_info("APIC: Switch to virtual wire mode setup with no configuration\n");
1455		upmode = true;
1456		default_setup_apic_routing();
1457		break;
1458	case APIC_SYMMETRIC_IO:
1459		pr_info("APIC: Switch to symmetric I/O mode setup\n");
1460		default_setup_apic_routing();
1461		break;
1462	case APIC_SYMMETRIC_IO_NO_ROUTING:
1463		pr_info("APIC: Switch to symmetric I/O mode setup in no SMP routine\n");
1464		break;
1465	}
1466
1467	apic_bsp_setup(upmode);
1468}
1469
1470static void lapic_setup_esr(void)
1471{
1472	unsigned int oldvalue, value, maxlvt;
1473
1474	if (!lapic_is_integrated()) {
1475		pr_info("No ESR for 82489DX.\n");
1476		return;
1477	}
1478
1479	if (apic->disable_esr) {
1480		/*
1481		 * Something untraceable is creating bad interrupts on
1482		 * secondary quads ... for the moment, just leave the
1483		 * ESR disabled - we can't do anything useful with the
1484		 * errors anyway - mbligh
1485		 */
1486		pr_info("Leaving ESR disabled.\n");
1487		return;
1488	}
1489
1490	maxlvt = lapic_get_maxlvt();
1491	if (maxlvt > 3)		/* Due to the Pentium erratum 3AP. */
1492		apic_write(APIC_ESR, 0);
1493	oldvalue = apic_read(APIC_ESR);
1494
1495	/* enables sending errors */
1496	value = ERROR_APIC_VECTOR;
1497	apic_write(APIC_LVTERR, value);
1498
1499	/*
1500	 * spec says clear errors after enabling vector.
1501	 */
1502	if (maxlvt > 3)
1503		apic_write(APIC_ESR, 0);
1504	value = apic_read(APIC_ESR);
1505	if (value != oldvalue)
1506		apic_printk(APIC_VERBOSE, "ESR value before enabling "
1507			"vector: 0x%08x  after: 0x%08x\n",
1508			oldvalue, value);
1509}
1510
1511#define APIC_IR_REGS		APIC_ISR_NR
1512#define APIC_IR_BITS		(APIC_IR_REGS * 32)
1513#define APIC_IR_MAPSIZE		(APIC_IR_BITS / BITS_PER_LONG)
1514
1515union apic_ir {
1516	unsigned long	map[APIC_IR_MAPSIZE];
1517	u32		regs[APIC_IR_REGS];
1518};
1519
1520static bool apic_check_and_ack(union apic_ir *irr, union apic_ir *isr)
1521{
1522	int i, bit;
1523
1524	/* Read the IRRs */
1525	for (i = 0; i < APIC_IR_REGS; i++)
1526		irr->regs[i] = apic_read(APIC_IRR + i * 0x10);
1527
1528	/* Read the ISRs */
1529	for (i = 0; i < APIC_IR_REGS; i++)
1530		isr->regs[i] = apic_read(APIC_ISR + i * 0x10);
1531
1532	/*
1533	 * If the ISR map is not empty. ACK the APIC and run another round
1534	 * to verify whether a pending IRR has been unblocked and turned
1535	 * into a ISR.
1536	 */
1537	if (!bitmap_empty(isr->map, APIC_IR_BITS)) {
1538		/*
1539		 * There can be multiple ISR bits set when a high priority
1540		 * interrupt preempted a lower priority one. Issue an ACK
1541		 * per set bit.
1542		 */
1543		for_each_set_bit(bit, isr->map, APIC_IR_BITS)
1544			ack_APIC_irq();
1545		return true;
1546	}
1547
1548	return !bitmap_empty(irr->map, APIC_IR_BITS);
1549}
1550
1551/*
1552 * After a crash, we no longer service the interrupts and a pending
1553 * interrupt from previous kernel might still have ISR bit set.
1554 *
1555 * Most probably by now the CPU has serviced that pending interrupt and it
1556 * might not have done the ack_APIC_irq() because it thought, interrupt
1557 * came from i8259 as ExtInt. LAPIC did not get EOI so it does not clear
1558 * the ISR bit and cpu thinks it has already serivced the interrupt. Hence
1559 * a vector might get locked. It was noticed for timer irq (vector
1560 * 0x31). Issue an extra EOI to clear ISR.
1561 *
1562 * If there are pending IRR bits they turn into ISR bits after a higher
1563 * priority ISR bit has been acked.
1564 */
1565static void apic_pending_intr_clear(void)
1566{
1567	union apic_ir irr, isr;
1568	unsigned int i;
1569
1570	/* 512 loops are way oversized and give the APIC a chance to obey. */
1571	for (i = 0; i < 512; i++) {
1572		if (!apic_check_and_ack(&irr, &isr))
1573			return;
1574	}
1575	/* Dump the IRR/ISR content if that failed */
1576	pr_warn("APIC: Stale IRR: %256pb ISR: %256pb\n", irr.map, isr.map);
1577}
1578
1579/**
1580 * setup_local_APIC - setup the local APIC
1581 *
1582 * Used to setup local APIC while initializing BSP or bringing up APs.
1583 * Always called with preemption disabled.
1584 */
1585static void setup_local_APIC(void)
1586{
1587	int cpu = smp_processor_id();
1588	unsigned int value;
 
 
 
 
 
 
1589
1590	if (disable_apic) {
1591		disable_ioapic_support();
1592		return;
1593	}
1594
1595	/*
1596	 * If this comes from kexec/kcrash the APIC might be enabled in
1597	 * SPIV. Soft disable it before doing further initialization.
1598	 */
1599	value = apic_read(APIC_SPIV);
1600	value &= ~APIC_SPIV_APIC_ENABLED;
1601	apic_write(APIC_SPIV, value);
1602
1603#ifdef CONFIG_X86_32
1604	/* Pound the ESR really hard over the head with a big hammer - mbligh */
1605	if (lapic_is_integrated() && apic->disable_esr) {
1606		apic_write(APIC_ESR, 0);
1607		apic_write(APIC_ESR, 0);
1608		apic_write(APIC_ESR, 0);
1609		apic_write(APIC_ESR, 0);
1610	}
1611#endif
 
 
1612	/*
1613	 * Double-check whether this APIC is really registered.
1614	 * This is meaningless in clustered apic mode, so we skip it.
1615	 */
1616	BUG_ON(!apic->apic_id_registered());
1617
1618	/*
1619	 * Intel recommends to set DFR, LDR and TPR before enabling
1620	 * an APIC.  See e.g. "AP-388 82489DX User's Manual" (Intel
1621	 * document number 292116).  So here it goes...
1622	 */
1623	apic->init_apic_ldr();
1624
1625#ifdef CONFIG_X86_32
1626	if (apic->dest_logical) {
1627		int logical_apicid, ldr_apicid;
1628
1629		/*
1630		 * APIC LDR is initialized.  If logical_apicid mapping was
1631		 * initialized during get_smp_config(), make sure it matches
1632		 * the actual value.
1633		 */
1634		logical_apicid = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1635		ldr_apicid = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
1636		if (logical_apicid != BAD_APICID)
1637			WARN_ON(logical_apicid != ldr_apicid);
1638		/* Always use the value from LDR. */
1639		early_per_cpu(x86_cpu_to_logical_apicid, cpu) = ldr_apicid;
1640	}
1641#endif
1642
1643	/*
1644	 * Set Task Priority to 'accept all except vectors 0-31'.  An APIC
1645	 * vector in the 16-31 range could be delivered if TPR == 0, but we
1646	 * would think it's an exception and terrible things will happen.  We
1647	 * never change this later on.
1648	 */
1649	value = apic_read(APIC_TASKPRI);
1650	value &= ~APIC_TPRI_MASK;
1651	value |= 0x10;
1652	apic_write(APIC_TASKPRI, value);
1653
1654	/* Clear eventually stale ISR/IRR bits */
1655	apic_pending_intr_clear();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1656
1657	/*
1658	 * Now that we are all set up, enable the APIC
1659	 */
1660	value = apic_read(APIC_SPIV);
1661	value &= ~APIC_VECTOR_MASK;
1662	/*
1663	 * Enable APIC
1664	 */
1665	value |= APIC_SPIV_APIC_ENABLED;
1666
1667#ifdef CONFIG_X86_32
1668	/*
1669	 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1670	 * certain networking cards. If high frequency interrupts are
1671	 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1672	 * entry is masked/unmasked at a high rate as well then sooner or
1673	 * later IOAPIC line gets 'stuck', no more interrupts are received
1674	 * from the device. If focus CPU is disabled then the hang goes
1675	 * away, oh well :-(
1676	 *
1677	 * [ This bug can be reproduced easily with a level-triggered
1678	 *   PCI Ne2000 networking cards and PII/PIII processors, dual
1679	 *   BX chipset. ]
1680	 */
1681	/*
1682	 * Actually disabling the focus CPU check just makes the hang less
1683	 * frequent as it makes the interrupt distributon model be more
1684	 * like LRU than MRU (the short-term load is more even across CPUs).
 
1685	 */
1686
1687	/*
1688	 * - enable focus processor (bit==0)
1689	 * - 64bit mode always use processor focus
1690	 *   so no need to set it
1691	 */
1692	value &= ~APIC_SPIV_FOCUS_DISABLED;
1693#endif
1694
1695	/*
1696	 * Set spurious IRQ vector
1697	 */
1698	value |= SPURIOUS_APIC_VECTOR;
1699	apic_write(APIC_SPIV, value);
1700
1701	perf_events_lapic_init();
1702
1703	/*
1704	 * Set up LVT0, LVT1:
1705	 *
1706	 * set up through-local-APIC on the boot CPU's LINT0. This is not
1707	 * strictly necessary in pure symmetric-IO mode, but sometimes
1708	 * we delegate interrupts to the 8259A.
1709	 */
1710	/*
1711	 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1712	 */
1713	value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1714	if (!cpu && (pic_mode || !value || skip_ioapic_setup)) {
1715		value = APIC_DM_EXTINT;
1716		apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1717	} else {
1718		value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1719		apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1720	}
1721	apic_write(APIC_LVT0, value);
1722
1723	/*
1724	 * Only the BSP sees the LINT1 NMI signal by default. This can be
1725	 * modified by apic_extnmi= boot option.
1726	 */
1727	if ((!cpu && apic_extnmi != APIC_EXTNMI_NONE) ||
1728	    apic_extnmi == APIC_EXTNMI_ALL)
1729		value = APIC_DM_NMI;
1730	else
1731		value = APIC_DM_NMI | APIC_LVT_MASKED;
1732
1733	/* Is 82489DX ? */
1734	if (!lapic_is_integrated())
1735		value |= APIC_LVT_LEVEL_TRIGGER;
1736	apic_write(APIC_LVT1, value);
1737
1738#ifdef CONFIG_X86_MCE_INTEL
1739	/* Recheck CMCI information after local APIC is up on CPU #0 */
1740	if (!cpu)
1741		cmci_recheck();
1742#endif
1743}
1744
1745static void end_local_APIC_setup(void)
1746{
1747	lapic_setup_esr();
1748
1749#ifdef CONFIG_X86_32
1750	{
1751		unsigned int value;
1752		/* Disable the local apic timer */
1753		value = apic_read(APIC_LVTT);
1754		value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1755		apic_write(APIC_LVTT, value);
1756	}
1757#endif
1758
1759	apic_pm_activate();
1760}
1761
1762/*
1763 * APIC setup function for application processors. Called from smpboot.c
1764 */
1765void apic_ap_setup(void)
1766{
1767	setup_local_APIC();
1768	end_local_APIC_setup();
1769}
1770
1771#ifdef CONFIG_X86_X2APIC
1772int x2apic_mode;
1773
1774enum {
1775	X2APIC_OFF,
1776	X2APIC_ON,
1777	X2APIC_DISABLED,
1778};
1779static int x2apic_state;
1780
1781static void __x2apic_disable(void)
1782{
1783	u64 msr;
1784
1785	if (!boot_cpu_has(X86_FEATURE_APIC))
1786		return;
1787
1788	rdmsrl(MSR_IA32_APICBASE, msr);
1789	if (!(msr & X2APIC_ENABLE))
1790		return;
1791	/* Disable xapic and x2apic first and then reenable xapic mode */
1792	wrmsrl(MSR_IA32_APICBASE, msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1793	wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1794	printk_once(KERN_INFO "x2apic disabled\n");
1795}
1796
1797static void __x2apic_enable(void)
1798{
1799	u64 msr;
1800
1801	rdmsrl(MSR_IA32_APICBASE, msr);
1802	if (msr & X2APIC_ENABLE)
1803		return;
1804	wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1805	printk_once(KERN_INFO "x2apic enabled\n");
1806}
1807
1808static int __init setup_nox2apic(char *str)
1809{
1810	if (x2apic_enabled()) {
1811		int apicid = native_apic_msr_read(APIC_ID);
1812
1813		if (apicid >= 255) {
1814			pr_warning("Apicid: %08x, cannot enforce nox2apic\n",
1815				   apicid);
1816			return 0;
1817		}
1818		pr_warning("x2apic already enabled.\n");
1819		__x2apic_disable();
1820	}
1821	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1822	x2apic_state = X2APIC_DISABLED;
1823	x2apic_mode = 0;
1824	return 0;
1825}
1826early_param("nox2apic", setup_nox2apic);
1827
1828/* Called from cpu_init() to enable x2apic on (secondary) cpus */
1829void x2apic_setup(void)
1830{
1831	/*
1832	 * If x2apic is not in ON state, disable it if already enabled
1833	 * from BIOS.
1834	 */
1835	if (x2apic_state != X2APIC_ON) {
1836		__x2apic_disable();
1837		return;
1838	}
1839	__x2apic_enable();
1840}
1841
1842static __init void x2apic_disable(void)
1843{
1844	u32 x2apic_id, state = x2apic_state;
1845
1846	x2apic_mode = 0;
1847	x2apic_state = X2APIC_DISABLED;
1848
1849	if (state != X2APIC_ON)
1850		return;
1851
1852	x2apic_id = read_apic_id();
1853	if (x2apic_id >= 255)
1854		panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1855
1856	__x2apic_disable();
1857	register_lapic_address(mp_lapic_addr);
1858}
1859
1860static __init void x2apic_enable(void)
1861{
1862	if (x2apic_state != X2APIC_OFF)
1863		return;
1864
1865	x2apic_mode = 1;
1866	x2apic_state = X2APIC_ON;
1867	__x2apic_enable();
1868}
1869
1870static __init void try_to_enable_x2apic(int remap_mode)
1871{
1872	if (x2apic_state == X2APIC_DISABLED)
1873		return;
1874
1875	if (remap_mode != IRQ_REMAP_X2APIC_MODE) {
1876		/* IR is required if there is APIC ID > 255 even when running
1877		 * under KVM
1878		 */
1879		if (max_physical_apicid > 255 ||
1880		    !x86_init.hyper.x2apic_available()) {
1881			pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n");
1882			x2apic_disable();
1883			return;
1884		}
1885
1886		/*
1887		 * without IR all CPUs can be addressed by IOAPIC/MSI
1888		 * only in physical mode
1889		 */
1890		x2apic_phys = 1;
1891	}
1892	x2apic_enable();
1893}
1894
1895void __init check_x2apic(void)
1896{
1897	if (x2apic_enabled()) {
1898		pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n");
1899		x2apic_mode = 1;
1900		x2apic_state = X2APIC_ON;
1901	} else if (!boot_cpu_has(X86_FEATURE_X2APIC)) {
1902		x2apic_state = X2APIC_DISABLED;
1903	}
1904}
1905#else /* CONFIG_X86_X2APIC */
1906static int __init validate_x2apic(void)
1907{
1908	if (!apic_is_x2apic_enabled())
1909		return 0;
1910	/*
1911	 * Checkme: Can we simply turn off x2apic here instead of panic?
1912	 */
1913	panic("BIOS has enabled x2apic but kernel doesn't support x2apic, please disable x2apic in BIOS.\n");
1914}
1915early_initcall(validate_x2apic);
1916
1917static inline void try_to_enable_x2apic(int remap_mode) { }
1918static inline void __x2apic_enable(void) { }
1919#endif /* !CONFIG_X86_X2APIC */
1920
 
 
 
 
 
 
 
 
 
 
 
1921void __init enable_IR_x2apic(void)
1922{
1923	unsigned long flags;
1924	int ret, ir_stat;
1925
1926	if (skip_ioapic_setup) {
1927		pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n");
1928		return;
1929	}
1930
1931	ir_stat = irq_remapping_prepare();
1932	if (ir_stat < 0 && !x2apic_supported())
1933		return;
1934
1935	ret = save_ioapic_entries();
1936	if (ret) {
1937		pr_info("Saving IO-APIC state failed: %d\n", ret);
1938		return;
1939	}
1940
1941	local_irq_save(flags);
1942	legacy_pic->mask_all();
1943	mask_ioapic_entries();
1944
1945	/* If irq_remapping_prepare() succeeded, try to enable it */
1946	if (ir_stat >= 0)
1947		ir_stat = irq_remapping_enable();
1948	/* ir_stat contains the remap mode or an error code */
1949	try_to_enable_x2apic(ir_stat);
1950
1951	if (ir_stat < 0)
1952		restore_ioapic_entries();
1953	legacy_pic->restore_mask();
1954	local_irq_restore(flags);
1955}
1956
1957#ifdef CONFIG_X86_64
1958/*
1959 * Detect and enable local APICs on non-SMP boards.
1960 * Original code written by Keir Fraser.
1961 * On AMD64 we trust the BIOS - if it says no APIC it is likely
1962 * not correctly set up (usually the APIC timer won't work etc.)
1963 */
1964static int __init detect_init_APIC(void)
1965{
1966	if (!boot_cpu_has(X86_FEATURE_APIC)) {
1967		pr_info("No local APIC present\n");
1968		return -1;
1969	}
1970
1971	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1972	return 0;
1973}
1974#else
1975
1976static int __init apic_verify(void)
1977{
1978	u32 features, h, l;
1979
1980	/*
1981	 * The APIC feature bit should now be enabled
1982	 * in `cpuid'
1983	 */
1984	features = cpuid_edx(1);
1985	if (!(features & (1 << X86_FEATURE_APIC))) {
1986		pr_warning("Could not enable APIC!\n");
1987		return -1;
1988	}
1989	set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1990	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1991
1992	/* The BIOS may have set up the APIC at some other address */
1993	if (boot_cpu_data.x86 >= 6) {
1994		rdmsr(MSR_IA32_APICBASE, l, h);
1995		if (l & MSR_IA32_APICBASE_ENABLE)
1996			mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1997	}
1998
1999	pr_info("Found and enabled local APIC!\n");
2000	return 0;
2001}
2002
2003int __init apic_force_enable(unsigned long addr)
2004{
2005	u32 h, l;
2006
2007	if (disable_apic)
2008		return -1;
2009
2010	/*
2011	 * Some BIOSes disable the local APIC in the APIC_BASE
2012	 * MSR. This can only be done in software for Intel P6 or later
2013	 * and AMD K7 (Model > 1) or later.
2014	 */
2015	if (boot_cpu_data.x86 >= 6) {
2016		rdmsr(MSR_IA32_APICBASE, l, h);
2017		if (!(l & MSR_IA32_APICBASE_ENABLE)) {
2018			pr_info("Local APIC disabled by BIOS -- reenabling.\n");
2019			l &= ~MSR_IA32_APICBASE_BASE;
2020			l |= MSR_IA32_APICBASE_ENABLE | addr;
2021			wrmsr(MSR_IA32_APICBASE, l, h);
2022			enabled_via_apicbase = 1;
2023		}
2024	}
2025	return apic_verify();
2026}
2027
2028/*
2029 * Detect and initialize APIC
2030 */
2031static int __init detect_init_APIC(void)
2032{
2033	/* Disabled by kernel option? */
2034	if (disable_apic)
2035		return -1;
2036
2037	switch (boot_cpu_data.x86_vendor) {
2038	case X86_VENDOR_AMD:
2039		if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
2040		    (boot_cpu_data.x86 >= 15))
2041			break;
2042		goto no_apic;
2043	case X86_VENDOR_HYGON:
2044		break;
2045	case X86_VENDOR_INTEL:
2046		if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
2047		    (boot_cpu_data.x86 == 5 && boot_cpu_has(X86_FEATURE_APIC)))
2048			break;
2049		goto no_apic;
2050	default:
2051		goto no_apic;
2052	}
2053
2054	if (!boot_cpu_has(X86_FEATURE_APIC)) {
2055		/*
2056		 * Over-ride BIOS and try to enable the local APIC only if
2057		 * "lapic" specified.
2058		 */
2059		if (!force_enable_local_apic) {
2060			pr_info("Local APIC disabled by BIOS -- "
2061				"you can enable it with \"lapic\"\n");
2062			return -1;
2063		}
2064		if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
2065			return -1;
2066	} else {
2067		if (apic_verify())
2068			return -1;
2069	}
2070
2071	apic_pm_activate();
2072
2073	return 0;
2074
2075no_apic:
2076	pr_info("No local APIC present or hardware disabled\n");
2077	return -1;
2078}
2079#endif
2080
2081/**
2082 * init_apic_mappings - initialize APIC mappings
2083 */
2084void __init init_apic_mappings(void)
2085{
2086	unsigned int new_apicid;
2087
2088	apic_check_deadline_errata();
2089
2090	if (x2apic_mode) {
2091		boot_cpu_physical_apicid = read_apic_id();
2092		return;
2093	}
2094
2095	/* If no local APIC can be found return early */
2096	if (!smp_found_config && detect_init_APIC()) {
2097		/* lets NOP'ify apic operations */
2098		pr_info("APIC: disable apic facility\n");
2099		apic_disable();
2100	} else {
2101		apic_phys = mp_lapic_addr;
2102
2103		/*
2104		 * If the system has ACPI MADT tables or MP info, the LAPIC
2105		 * address is already registered.
2106		 */
2107		if (!acpi_lapic && !smp_found_config)
2108			register_lapic_address(apic_phys);
2109	}
2110
2111	/*
2112	 * Fetch the APIC ID of the BSP in case we have a
2113	 * default configuration (or the MP table is broken).
2114	 */
2115	new_apicid = read_apic_id();
2116	if (boot_cpu_physical_apicid != new_apicid) {
2117		boot_cpu_physical_apicid = new_apicid;
2118		/*
2119		 * yeah -- we lie about apic_version
2120		 * in case if apic was disabled via boot option
2121		 * but it's not a problem for SMP compiled kernel
2122		 * since apic_intr_mode_select is prepared for such
2123		 * a case and disable smp mode
2124		 */
2125		boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
 
2126	}
2127}
2128
2129void __init register_lapic_address(unsigned long address)
2130{
2131	mp_lapic_addr = address;
2132
2133	if (!x2apic_mode) {
2134		set_fixmap_nocache(FIX_APIC_BASE, address);
2135		apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
2136			    APIC_BASE, address);
2137	}
2138	if (boot_cpu_physical_apicid == -1U) {
2139		boot_cpu_physical_apicid  = read_apic_id();
2140		boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
 
2141	}
2142}
2143
 
 
2144/*
2145 * Local APIC interrupts
2146 */
2147
2148/*
2149 * This interrupt should _never_ happen with our APIC/SMP architecture
2150 */
2151__visible void __irq_entry smp_spurious_interrupt(struct pt_regs *regs)
2152{
2153	u8 vector = ~regs->orig_ax;
2154	u32 v;
2155
2156	entering_irq();
2157	trace_spurious_apic_entry(vector);
2158
2159	inc_irq_stat(irq_spurious_count);
2160
2161	/*
2162	 * If this is a spurious interrupt then do not acknowledge
2163	 */
2164	if (vector == SPURIOUS_APIC_VECTOR) {
2165		/* See SDM vol 3 */
2166		pr_info("Spurious APIC interrupt (vector 0xFF) on CPU#%d, should never happen.\n",
2167			smp_processor_id());
2168		goto out;
2169	}
2170
2171	/*
2172	 * If it is a vectored one, verify it's set in the ISR. If set,
2173	 * acknowledge it.
 
2174	 */
2175	v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1));
2176	if (v & (1 << (vector & 0x1f))) {
2177		pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Acked\n",
2178			vector, smp_processor_id());
2179		ack_APIC_irq();
2180	} else {
2181		pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Not pending!\n",
2182			vector, smp_processor_id());
2183	}
2184out:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2185	trace_spurious_apic_exit(vector);
2186	exiting_irq();
2187}
2188
2189/*
2190 * This interrupt should never happen with our APIC/SMP architecture
2191 */
2192__visible void __irq_entry smp_error_interrupt(struct pt_regs *regs)
2193{
 
 
2194	static const char * const error_interrupt_reason[] = {
2195		"Send CS error",		/* APIC Error Bit 0 */
2196		"Receive CS error",		/* APIC Error Bit 1 */
2197		"Send accept error",		/* APIC Error Bit 2 */
2198		"Receive accept error",		/* APIC Error Bit 3 */
2199		"Redirectable IPI",		/* APIC Error Bit 4 */
2200		"Send illegal vector",		/* APIC Error Bit 5 */
2201		"Received illegal vector",	/* APIC Error Bit 6 */
2202		"Illegal register address",	/* APIC Error Bit 7 */
2203	};
2204	u32 v, i = 0;
2205
2206	entering_irq();
2207	trace_error_apic_entry(ERROR_APIC_VECTOR);
2208
2209	/* First tickle the hardware, only then report what went on. -- REW */
2210	if (lapic_get_maxlvt() > 3)	/* Due to the Pentium erratum 3AP. */
2211		apic_write(APIC_ESR, 0);
2212	v = apic_read(APIC_ESR);
2213	ack_APIC_irq();
2214	atomic_inc(&irq_err_count);
2215
2216	apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x",
2217		    smp_processor_id(), v);
2218
2219	v &= 0xff;
2220	while (v) {
2221		if (v & 0x1)
2222			apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
2223		i++;
2224		v >>= 1;
2225	}
2226
2227	apic_printk(APIC_DEBUG, KERN_CONT "\n");
2228
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2229	trace_error_apic_exit(ERROR_APIC_VECTOR);
2230	exiting_irq();
2231}
2232
2233/**
2234 * connect_bsp_APIC - attach the APIC to the interrupt system
2235 */
2236static void __init connect_bsp_APIC(void)
2237{
2238#ifdef CONFIG_X86_32
2239	if (pic_mode) {
2240		/*
2241		 * Do not trust the local APIC being empty at bootup.
2242		 */
2243		clear_local_APIC();
2244		/*
2245		 * PIC mode, enable APIC mode in the IMCR, i.e.  connect BSP's
2246		 * local APIC to INT and NMI lines.
2247		 */
2248		apic_printk(APIC_VERBOSE, "leaving PIC mode, "
2249				"enabling APIC mode.\n");
2250		imcr_pic_to_apic();
2251	}
2252#endif
2253}
2254
2255/**
2256 * disconnect_bsp_APIC - detach the APIC from the interrupt system
2257 * @virt_wire_setup:	indicates, whether virtual wire mode is selected
2258 *
2259 * Virtual wire mode is necessary to deliver legacy interrupts even when the
2260 * APIC is disabled.
2261 */
2262void disconnect_bsp_APIC(int virt_wire_setup)
2263{
2264	unsigned int value;
2265
2266#ifdef CONFIG_X86_32
2267	if (pic_mode) {
2268		/*
2269		 * Put the board back into PIC mode (has an effect only on
2270		 * certain older boards).  Note that APIC interrupts, including
2271		 * IPIs, won't work beyond this point!  The only exception are
2272		 * INIT IPIs.
2273		 */
2274		apic_printk(APIC_VERBOSE, "disabling APIC mode, "
2275				"entering PIC mode.\n");
2276		imcr_apic_to_pic();
2277		return;
2278	}
2279#endif
2280
2281	/* Go back to Virtual Wire compatibility mode */
2282
2283	/* For the spurious interrupt use vector F, and enable it */
2284	value = apic_read(APIC_SPIV);
2285	value &= ~APIC_VECTOR_MASK;
2286	value |= APIC_SPIV_APIC_ENABLED;
2287	value |= 0xf;
2288	apic_write(APIC_SPIV, value);
2289
2290	if (!virt_wire_setup) {
2291		/*
2292		 * For LVT0 make it edge triggered, active high,
2293		 * external and enabled
2294		 */
2295		value = apic_read(APIC_LVT0);
2296		value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2297			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2298			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2299		value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2300		value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2301		apic_write(APIC_LVT0, value);
2302	} else {
2303		/* Disable LVT0 */
2304		apic_write(APIC_LVT0, APIC_LVT_MASKED);
2305	}
2306
2307	/*
2308	 * For LVT1 make it edge triggered, active high,
2309	 * nmi and enabled
2310	 */
2311	value = apic_read(APIC_LVT1);
2312	value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2313			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2314			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2315	value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2316	value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2317	apic_write(APIC_LVT1, value);
2318}
2319
2320/*
2321 * The number of allocated logical CPU IDs. Since logical CPU IDs are allocated
2322 * contiguously, it equals to current allocated max logical CPU ID plus 1.
2323 * All allocated CPU IDs should be in the [0, nr_logical_cpuids) range,
2324 * so the maximum of nr_logical_cpuids is nr_cpu_ids.
2325 *
2326 * NOTE: Reserve 0 for BSP.
2327 */
2328static int nr_logical_cpuids = 1;
2329
2330/*
2331 * Used to store mapping between logical CPU IDs and APIC IDs.
2332 */
2333static int cpuid_to_apicid[] = {
2334	[0 ... NR_CPUS - 1] = -1,
2335};
2336
2337#ifdef CONFIG_SMP
2338/**
2339 * apic_id_is_primary_thread - Check whether APIC ID belongs to a primary thread
2340 * @id:	APIC ID to check
2341 */
2342bool apic_id_is_primary_thread(unsigned int apicid)
2343{
2344	u32 mask;
2345
2346	if (smp_num_siblings == 1)
2347		return true;
2348	/* Isolate the SMT bit(s) in the APICID and check for 0 */
2349	mask = (1U << (fls(smp_num_siblings) - 1)) - 1;
2350	return !(apicid & mask);
2351}
2352#endif
2353
2354/*
2355 * Should use this API to allocate logical CPU IDs to keep nr_logical_cpuids
2356 * and cpuid_to_apicid[] synchronized.
2357 */
2358static int allocate_logical_cpuid(int apicid)
2359{
2360	int i;
2361
2362	/*
2363	 * cpuid <-> apicid mapping is persistent, so when a cpu is up,
2364	 * check if the kernel has allocated a cpuid for it.
2365	 */
2366	for (i = 0; i < nr_logical_cpuids; i++) {
2367		if (cpuid_to_apicid[i] == apicid)
2368			return i;
2369	}
2370
2371	/* Allocate a new cpuid. */
2372	if (nr_logical_cpuids >= nr_cpu_ids) {
2373		WARN_ONCE(1, "APIC: NR_CPUS/possible_cpus limit of %u reached. "
2374			     "Processor %d/0x%x and the rest are ignored.\n",
2375			     nr_cpu_ids, nr_logical_cpuids, apicid);
2376		return -EINVAL;
2377	}
2378
2379	cpuid_to_apicid[nr_logical_cpuids] = apicid;
2380	return nr_logical_cpuids++;
2381}
2382
2383int generic_processor_info(int apicid, int version)
2384{
2385	int cpu, max = nr_cpu_ids;
2386	bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
2387				phys_cpu_present_map);
2388
2389	/*
2390	 * boot_cpu_physical_apicid is designed to have the apicid
2391	 * returned by read_apic_id(), i.e, the apicid of the
2392	 * currently booting-up processor. However, on some platforms,
2393	 * it is temporarily modified by the apicid reported as BSP
2394	 * through MP table. Concretely:
2395	 *
2396	 * - arch/x86/kernel/mpparse.c: MP_processor_info()
2397	 * - arch/x86/mm/amdtopology.c: amd_numa_init()
2398	 *
2399	 * This function is executed with the modified
2400	 * boot_cpu_physical_apicid. So, disabled_cpu_apicid kernel
2401	 * parameter doesn't work to disable APs on kdump 2nd kernel.
2402	 *
2403	 * Since fixing handling of boot_cpu_physical_apicid requires
2404	 * another discussion and tests on each platform, we leave it
2405	 * for now and here we use read_apic_id() directly in this
2406	 * function, generic_processor_info().
2407	 */
2408	if (disabled_cpu_apicid != BAD_APICID &&
2409	    disabled_cpu_apicid != read_apic_id() &&
2410	    disabled_cpu_apicid == apicid) {
2411		int thiscpu = num_processors + disabled_cpus;
2412
2413		pr_warning("APIC: Disabling requested cpu."
2414			   " Processor %d/0x%x ignored.\n",
2415			   thiscpu, apicid);
2416
2417		disabled_cpus++;
2418		return -ENODEV;
2419	}
2420
2421	/*
2422	 * If boot cpu has not been detected yet, then only allow upto
2423	 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2424	 */
2425	if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2426	    apicid != boot_cpu_physical_apicid) {
2427		int thiscpu = max + disabled_cpus - 1;
2428
2429		pr_warning(
2430			"APIC: NR_CPUS/possible_cpus limit of %i almost"
2431			" reached. Keeping one slot for boot cpu."
2432			"  Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2433
2434		disabled_cpus++;
2435		return -ENODEV;
2436	}
2437
2438	if (num_processors >= nr_cpu_ids) {
2439		int thiscpu = max + disabled_cpus;
2440
2441		pr_warning("APIC: NR_CPUS/possible_cpus limit of %i "
2442			   "reached. Processor %d/0x%x ignored.\n",
2443			   max, thiscpu, apicid);
2444
2445		disabled_cpus++;
2446		return -EINVAL;
2447	}
2448
 
2449	if (apicid == boot_cpu_physical_apicid) {
2450		/*
2451		 * x86_bios_cpu_apicid is required to have processors listed
2452		 * in same order as logical cpu numbers. Hence the first
2453		 * entry is BSP, and so on.
2454		 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2455		 * for BSP.
2456		 */
2457		cpu = 0;
 
 
2458
2459		/* Logical cpuid 0 is reserved for BSP. */
2460		cpuid_to_apicid[0] = apicid;
2461	} else {
2462		cpu = allocate_logical_cpuid(apicid);
2463		if (cpu < 0) {
2464			disabled_cpus++;
2465			return -EINVAL;
2466		}
 
 
 
 
2467	}
2468
2469	/*
2470	 * Validate version
2471	 */
2472	if (version == 0x0) {
2473		pr_warning("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2474			   cpu, apicid);
2475		version = 0x10;
2476	}
 
2477
2478	if (version != boot_cpu_apic_version) {
2479		pr_warning("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2480			boot_cpu_apic_version, cpu, version);
2481	}
2482
 
2483	if (apicid > max_physical_apicid)
2484		max_physical_apicid = apicid;
2485
2486#if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2487	early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2488	early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2489#endif
2490#ifdef CONFIG_X86_32
2491	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2492		apic->x86_32_early_logical_apicid(cpu);
2493#endif
2494	set_cpu_possible(cpu, true);
2495	physid_set(apicid, phys_cpu_present_map);
2496	set_cpu_present(cpu, true);
2497	num_processors++;
2498
2499	return cpu;
2500}
2501
2502int hard_smp_processor_id(void)
2503{
2504	return read_apic_id();
2505}
2506
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2507/*
2508 * Override the generic EOI implementation with an optimized version.
2509 * Only called during early boot when only one CPU is active and with
2510 * interrupts disabled, so we know this does not race with actual APIC driver
2511 * use.
2512 */
2513void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2514{
2515	struct apic **drv;
2516
2517	for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2518		/* Should happen once for each apic */
2519		WARN_ON((*drv)->eoi_write == eoi_write);
2520		(*drv)->native_eoi_write = (*drv)->eoi_write;
2521		(*drv)->eoi_write = eoi_write;
2522	}
2523}
2524
2525static void __init apic_bsp_up_setup(void)
2526{
2527#ifdef CONFIG_X86_64
2528	apic_write(APIC_ID, apic->set_apic_id(boot_cpu_physical_apicid));
2529#else
2530	/*
2531	 * Hack: In case of kdump, after a crash, kernel might be booting
2532	 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
2533	 * might be zero if read from MP tables. Get it from LAPIC.
2534	 */
2535# ifdef CONFIG_CRASH_DUMP
2536	boot_cpu_physical_apicid = read_apic_id();
2537# endif
2538#endif
2539	physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
2540}
2541
2542/**
2543 * apic_bsp_setup - Setup function for local apic and io-apic
2544 * @upmode:		Force UP mode (for APIC_init_uniprocessor)
 
 
 
2545 */
2546static void __init apic_bsp_setup(bool upmode)
2547{
 
 
2548	connect_bsp_APIC();
2549	if (upmode)
2550		apic_bsp_up_setup();
2551	setup_local_APIC();
2552
 
 
 
 
 
2553	enable_IO_APIC();
2554	end_local_APIC_setup();
2555	irq_remap_enable_fault_handling();
2556	setup_IO_APIC();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2557}
2558
2559#ifdef CONFIG_UP_LATE_INIT
2560void __init up_late_init(void)
2561{
2562	if (apic_intr_mode == APIC_PIC)
2563		return;
2564
2565	/* Setup local timer */
2566	x86_init.timers.setup_percpu_clockev();
2567}
2568#endif
2569
2570/*
2571 * Power management
2572 */
2573#ifdef CONFIG_PM
2574
2575static struct {
2576	/*
2577	 * 'active' is true if the local APIC was enabled by us and
2578	 * not the BIOS; this signifies that we are also responsible
2579	 * for disabling it before entering apm/acpi suspend
2580	 */
2581	int active;
2582	/* r/w apic fields */
2583	unsigned int apic_id;
2584	unsigned int apic_taskpri;
2585	unsigned int apic_ldr;
2586	unsigned int apic_dfr;
2587	unsigned int apic_spiv;
2588	unsigned int apic_lvtt;
2589	unsigned int apic_lvtpc;
2590	unsigned int apic_lvt0;
2591	unsigned int apic_lvt1;
2592	unsigned int apic_lvterr;
2593	unsigned int apic_tmict;
2594	unsigned int apic_tdcr;
2595	unsigned int apic_thmr;
2596	unsigned int apic_cmci;
2597} apic_pm_state;
2598
2599static int lapic_suspend(void)
2600{
2601	unsigned long flags;
2602	int maxlvt;
2603
2604	if (!apic_pm_state.active)
2605		return 0;
2606
2607	maxlvt = lapic_get_maxlvt();
2608
2609	apic_pm_state.apic_id = apic_read(APIC_ID);
2610	apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2611	apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2612	apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2613	apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2614	apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2615	if (maxlvt >= 4)
2616		apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2617	apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2618	apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2619	apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2620	apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2621	apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2622#ifdef CONFIG_X86_THERMAL_VECTOR
2623	if (maxlvt >= 5)
2624		apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2625#endif
2626#ifdef CONFIG_X86_MCE_INTEL
2627	if (maxlvt >= 6)
2628		apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI);
2629#endif
2630
2631	local_irq_save(flags);
2632	disable_local_APIC();
2633
2634	irq_remapping_disable();
2635
2636	local_irq_restore(flags);
2637	return 0;
2638}
2639
2640static void lapic_resume(void)
2641{
2642	unsigned int l, h;
2643	unsigned long flags;
2644	int maxlvt;
2645
2646	if (!apic_pm_state.active)
2647		return;
2648
2649	local_irq_save(flags);
2650
2651	/*
2652	 * IO-APIC and PIC have their own resume routines.
2653	 * We just mask them here to make sure the interrupt
2654	 * subsystem is completely quiet while we enable x2apic
2655	 * and interrupt-remapping.
2656	 */
2657	mask_ioapic_entries();
2658	legacy_pic->mask_all();
2659
2660	if (x2apic_mode) {
2661		__x2apic_enable();
2662	} else {
2663		/*
2664		 * Make sure the APICBASE points to the right address
2665		 *
2666		 * FIXME! This will be wrong if we ever support suspend on
2667		 * SMP! We'll need to do this as part of the CPU restore!
2668		 */
2669		if (boot_cpu_data.x86 >= 6) {
2670			rdmsr(MSR_IA32_APICBASE, l, h);
2671			l &= ~MSR_IA32_APICBASE_BASE;
2672			l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2673			wrmsr(MSR_IA32_APICBASE, l, h);
2674		}
2675	}
2676
2677	maxlvt = lapic_get_maxlvt();
2678	apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2679	apic_write(APIC_ID, apic_pm_state.apic_id);
2680	apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2681	apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2682	apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2683	apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2684	apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2685	apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2686#ifdef CONFIG_X86_THERMAL_VECTOR
2687	if (maxlvt >= 5)
2688		apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2689#endif
2690#ifdef CONFIG_X86_MCE_INTEL
2691	if (maxlvt >= 6)
2692		apic_write(APIC_LVTCMCI, apic_pm_state.apic_cmci);
2693#endif
2694	if (maxlvt >= 4)
2695		apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2696	apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2697	apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2698	apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2699	apic_write(APIC_ESR, 0);
2700	apic_read(APIC_ESR);
2701	apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2702	apic_write(APIC_ESR, 0);
2703	apic_read(APIC_ESR);
2704
2705	irq_remapping_reenable(x2apic_mode);
2706
2707	local_irq_restore(flags);
2708}
2709
2710/*
2711 * This device has no shutdown method - fully functioning local APICs
2712 * are needed on every CPU up until machine_halt/restart/poweroff.
2713 */
2714
2715static struct syscore_ops lapic_syscore_ops = {
2716	.resume		= lapic_resume,
2717	.suspend	= lapic_suspend,
2718};
2719
2720static void apic_pm_activate(void)
2721{
2722	apic_pm_state.active = 1;
2723}
2724
2725static int __init init_lapic_sysfs(void)
2726{
2727	/* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2728	if (boot_cpu_has(X86_FEATURE_APIC))
2729		register_syscore_ops(&lapic_syscore_ops);
2730
2731	return 0;
2732}
2733
2734/* local apic needs to resume before other devices access its registers. */
2735core_initcall(init_lapic_sysfs);
2736
2737#else	/* CONFIG_PM */
2738
2739static void apic_pm_activate(void) { }
2740
2741#endif	/* CONFIG_PM */
2742
2743#ifdef CONFIG_X86_64
2744
2745static int multi_checked;
2746static int multi;
2747
2748static int set_multi(const struct dmi_system_id *d)
2749{
2750	if (multi)
2751		return 0;
2752	pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2753	multi = 1;
2754	return 0;
2755}
2756
2757static const struct dmi_system_id multi_dmi_table[] = {
2758	{
2759		.callback = set_multi,
2760		.ident = "IBM System Summit2",
2761		.matches = {
2762			DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2763			DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2764		},
2765	},
2766	{}
2767};
2768
2769static void dmi_check_multi(void)
2770{
2771	if (multi_checked)
2772		return;
2773
2774	dmi_check_system(multi_dmi_table);
2775	multi_checked = 1;
2776}
2777
2778/*
2779 * apic_is_clustered_box() -- Check if we can expect good TSC
2780 *
2781 * Thus far, the major user of this is IBM's Summit2 series:
2782 * Clustered boxes may have unsynced TSC problems if they are
2783 * multi-chassis.
2784 * Use DMI to check them
2785 */
2786int apic_is_clustered_box(void)
2787{
2788	dmi_check_multi();
2789	return multi;
2790}
2791#endif
2792
2793/*
2794 * APIC command line parameters
2795 */
2796static int __init setup_disableapic(char *arg)
2797{
2798	disable_apic = 1;
2799	setup_clear_cpu_cap(X86_FEATURE_APIC);
2800	return 0;
2801}
2802early_param("disableapic", setup_disableapic);
2803
2804/* same as disableapic, for compatibility */
2805static int __init setup_nolapic(char *arg)
2806{
2807	return setup_disableapic(arg);
2808}
2809early_param("nolapic", setup_nolapic);
2810
2811static int __init parse_lapic_timer_c2_ok(char *arg)
2812{
2813	local_apic_timer_c2_ok = 1;
2814	return 0;
2815}
2816early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2817
2818static int __init parse_disable_apic_timer(char *arg)
2819{
2820	disable_apic_timer = 1;
2821	return 0;
2822}
2823early_param("noapictimer", parse_disable_apic_timer);
2824
2825static int __init parse_nolapic_timer(char *arg)
2826{
2827	disable_apic_timer = 1;
2828	return 0;
2829}
2830early_param("nolapic_timer", parse_nolapic_timer);
2831
2832static int __init apic_set_verbosity(char *arg)
2833{
2834	if (!arg)  {
2835#ifdef CONFIG_X86_64
2836		skip_ioapic_setup = 0;
2837		return 0;
2838#endif
2839		return -EINVAL;
2840	}
2841
2842	if (strcmp("debug", arg) == 0)
2843		apic_verbosity = APIC_DEBUG;
2844	else if (strcmp("verbose", arg) == 0)
2845		apic_verbosity = APIC_VERBOSE;
2846#ifdef CONFIG_X86_64
2847	else {
2848		pr_warning("APIC Verbosity level %s not recognised"
2849			" use apic=verbose or apic=debug\n", arg);
2850		return -EINVAL;
2851	}
2852#endif
2853
2854	return 0;
2855}
2856early_param("apic", apic_set_verbosity);
2857
2858static int __init lapic_insert_resource(void)
2859{
2860	if (!apic_phys)
2861		return -1;
2862
2863	/* Put local APIC into the resource map. */
2864	lapic_resource.start = apic_phys;
2865	lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2866	insert_resource(&iomem_resource, &lapic_resource);
2867
2868	return 0;
2869}
2870
2871/*
2872 * need call insert after e820__reserve_resources()
2873 * that is using request_resource
2874 */
2875late_initcall(lapic_insert_resource);
2876
2877static int __init apic_set_disabled_cpu_apicid(char *arg)
2878{
2879	if (!arg || !get_option(&arg, &disabled_cpu_apicid))
2880		return -EINVAL;
2881
2882	return 0;
2883}
2884early_param("disable_cpu_apicid", apic_set_disabled_cpu_apicid);
2885
2886static int __init apic_set_extnmi(char *arg)
2887{
2888	if (!arg)
2889		return -EINVAL;
2890
2891	if (!strncmp("all", arg, 3))
2892		apic_extnmi = APIC_EXTNMI_ALL;
2893	else if (!strncmp("none", arg, 4))
2894		apic_extnmi = APIC_EXTNMI_NONE;
2895	else if (!strncmp("bsp", arg, 3))
2896		apic_extnmi = APIC_EXTNMI_BSP;
2897	else {
2898		pr_warn("Unknown external NMI delivery mode `%s' ignored\n", arg);
2899		return -EINVAL;
2900	}
2901
2902	return 0;
2903}
2904early_param("apic_extnmi", apic_set_extnmi);
v4.6
 
   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
  59unsigned int num_processors;
  60
  61unsigned disabled_cpus;
  62
  63/* Processor that is doing the boot up */
  64unsigned int boot_cpu_physical_apicid = -1U;
  65EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid);
  66
 
 
  67/*
  68 * The highest APIC ID seen during enumeration.
  69 */
  70static unsigned int max_physical_apicid;
  71
  72/*
  73 * Bitmask of physically existing CPUs:
  74 */
  75physid_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 */
  82static unsigned int disabled_cpu_apicid __read_mostly = BAD_APICID;
  83
  84/*
  85 * This variable controls which CPUs receive external NMIs.  By default,
  86 * external NMIs are delivered only to the BSP.
  87 */
  88static int apic_extnmi = APIC_EXTNMI_BSP;
  89
  90/*
  91 * Map cpu index to physical APIC ID
  92 */
  93DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_cpu_to_apicid, BAD_APICID);
  94DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_bios_cpu_apicid, BAD_APICID);
 
  95EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
  96EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
 
  97
  98#ifdef CONFIG_X86_32
  99
 100/*
 101 * On x86_32, the mapping between cpu and logical apicid may vary
 102 * depending on apic in use.  The following early percpu variable is
 103 * used for the mapping.  This is where the behaviors of x86_64 and 32
 104 * actually diverge.  Let's keep it ugly for now.
 105 */
 106DEFINE_EARLY_PER_CPU_READ_MOSTLY(int, x86_cpu_to_logical_apicid, BAD_APICID);
 107
 108/* Local APIC was disabled by the BIOS and enabled by the kernel */
 109static int enabled_via_apicbase;
 110
 111/*
 112 * Handle interrupt mode configuration register (IMCR).
 113 * This register controls whether the interrupt signals
 114 * that reach the BSP come from the master PIC or from the
 115 * local APIC. Before entering Symmetric I/O Mode, either
 116 * the BIOS or the operating system must switch out of
 117 * PIC Mode by changing the IMCR.
 118 */
 119static inline void imcr_pic_to_apic(void)
 120{
 121	/* select IMCR register */
 122	outb(0x70, 0x22);
 123	/* NMI and 8259 INTR go through APIC */
 124	outb(0x01, 0x23);
 125}
 126
 127static inline void imcr_apic_to_pic(void)
 128{
 129	/* select IMCR register */
 130	outb(0x70, 0x22);
 131	/* NMI and 8259 INTR go directly to BSP */
 132	outb(0x00, 0x23);
 133}
 134#endif
 135
 136/*
 137 * Knob to control our willingness to enable the local APIC.
 138 *
 139 * +1=force-enable
 140 */
 141static int force_enable_local_apic __initdata;
 142
 143/*
 144 * APIC command line parameters
 145 */
 146static int __init parse_lapic(char *arg)
 147{
 148	if (config_enabled(CONFIG_X86_32) && !arg)
 149		force_enable_local_apic = 1;
 150	else if (arg && !strncmp(arg, "notscdeadline", 13))
 151		setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
 152	return 0;
 153}
 154early_param("lapic", parse_lapic);
 155
 156#ifdef CONFIG_X86_64
 157static int apic_calibrate_pmtmr __initdata;
 158static __init int setup_apicpmtimer(char *s)
 159{
 160	apic_calibrate_pmtmr = 1;
 161	notsc_setup(NULL);
 162	return 0;
 163}
 164__setup("apicpmtimer", setup_apicpmtimer);
 165#endif
 166
 167unsigned long mp_lapic_addr;
 168int disable_apic;
 169/* Disable local APIC timer from the kernel commandline or via dmi quirk */
 170static int disable_apic_timer __initdata;
 171/* Local APIC timer works in C2 */
 172int local_apic_timer_c2_ok;
 173EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
 174
 175int first_system_vector = FIRST_SYSTEM_VECTOR;
 176
 177/*
 178 * Debug level, exported for io_apic.c
 179 */
 180unsigned int apic_verbosity;
 181
 182int pic_mode;
 183
 184/* Have we found an MP table */
 185int smp_found_config;
 186
 187static struct resource lapic_resource = {
 188	.name = "Local APIC",
 189	.flags = IORESOURCE_MEM | IORESOURCE_BUSY,
 190};
 191
 192unsigned int lapic_timer_frequency = 0;
 193
 194static void apic_pm_activate(void);
 195
 196static unsigned long apic_phys;
 197
 198/*
 199 * Get the LAPIC version
 200 */
 201static inline int lapic_get_version(void)
 202{
 203	return GET_APIC_VERSION(apic_read(APIC_LVR));
 204}
 205
 206/*
 207 * Check, if the APIC is integrated or a separate chip
 208 */
 209static inline int lapic_is_integrated(void)
 210{
 211#ifdef CONFIG_X86_64
 212	return 1;
 213#else
 214	return APIC_INTEGRATED(lapic_get_version());
 215#endif
 216}
 217
 218/*
 219 * Check, whether this is a modern or a first generation APIC
 220 */
 221static int modern_apic(void)
 222{
 223	/* AMD systems use old APIC versions, so check the CPU */
 224	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
 225	    boot_cpu_data.x86 >= 0xf)
 226		return 1;
 
 
 
 
 
 227	return lapic_get_version() >= 0x14;
 228}
 229
 230/*
 231 * right after this call apic become NOOP driven
 232 * so apic->write/read doesn't do anything
 233 */
 234static void __init apic_disable(void)
 235{
 236	pr_info("APIC: switched to apic NOOP\n");
 237	apic = &apic_noop;
 238}
 239
 240void native_apic_wait_icr_idle(void)
 241{
 242	while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
 243		cpu_relax();
 244}
 245
 246u32 native_safe_apic_wait_icr_idle(void)
 247{
 248	u32 send_status;
 249	int timeout;
 250
 251	timeout = 0;
 252	do {
 253		send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
 254		if (!send_status)
 255			break;
 256		inc_irq_stat(icr_read_retry_count);
 257		udelay(100);
 258	} while (timeout++ < 1000);
 259
 260	return send_status;
 261}
 262
 263void native_apic_icr_write(u32 low, u32 id)
 264{
 265	unsigned long flags;
 266
 267	local_irq_save(flags);
 268	apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
 269	apic_write(APIC_ICR, low);
 270	local_irq_restore(flags);
 271}
 272
 273u64 native_apic_icr_read(void)
 274{
 275	u32 icr1, icr2;
 276
 277	icr2 = apic_read(APIC_ICR2);
 278	icr1 = apic_read(APIC_ICR);
 279
 280	return icr1 | ((u64)icr2 << 32);
 281}
 282
 283#ifdef CONFIG_X86_32
 284/**
 285 * get_physical_broadcast - Get number of physical broadcast IDs
 286 */
 287int get_physical_broadcast(void)
 288{
 289	return modern_apic() ? 0xff : 0xf;
 290}
 291#endif
 292
 293/**
 294 * lapic_get_maxlvt - get the maximum number of local vector table entries
 295 */
 296int lapic_get_maxlvt(void)
 297{
 298	unsigned int v;
 299
 300	v = apic_read(APIC_LVR);
 301	/*
 302	 * - we always have APIC integrated on 64bit mode
 303	 * - 82489DXs do not report # of LVT entries
 304	 */
 305	return APIC_INTEGRATED(GET_APIC_VERSION(v)) ? GET_APIC_MAXLVT(v) : 2;
 306}
 307
 308/*
 309 * Local APIC timer
 310 */
 311
 312/* Clock divisor */
 313#define APIC_DIVISOR 16
 314#define TSC_DIVISOR  32
 315
 316/*
 317 * This function sets up the local APIC timer, with a timeout of
 318 * 'clocks' APIC bus clock. During calibration we actually call
 319 * this function twice on the boot CPU, once with a bogus timeout
 320 * value, second time for real. The other (noncalibrating) CPUs
 321 * call this function only once, with the real, calibrated value.
 322 *
 323 * We do reads before writes even if unnecessary, to get around the
 324 * P5 APIC double write bug.
 325 */
 326static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
 327{
 328	unsigned int lvtt_value, tmp_value;
 329
 330	lvtt_value = LOCAL_TIMER_VECTOR;
 331	if (!oneshot)
 332		lvtt_value |= APIC_LVT_TIMER_PERIODIC;
 333	else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
 334		lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
 335
 336	if (!lapic_is_integrated())
 337		lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
 338
 339	if (!irqen)
 340		lvtt_value |= APIC_LVT_MASKED;
 341
 342	apic_write(APIC_LVTT, lvtt_value);
 343
 344	if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
 345		/*
 346		 * See Intel SDM: TSC-Deadline Mode chapter. In xAPIC mode,
 347		 * writing to the APIC LVTT and TSC_DEADLINE MSR isn't serialized.
 348		 * According to Intel, MFENCE can do the serialization here.
 349		 */
 350		asm volatile("mfence" : : : "memory");
 351
 352		printk_once(KERN_DEBUG "TSC deadline timer enabled\n");
 353		return;
 354	}
 355
 356	/*
 357	 * Divide PICLK by 16
 358	 */
 359	tmp_value = apic_read(APIC_TDCR);
 360	apic_write(APIC_TDCR,
 361		(tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
 362		APIC_TDR_DIV_16);
 363
 364	if (!oneshot)
 365		apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
 366}
 367
 368/*
 369 * Setup extended LVT, AMD specific
 370 *
 371 * Software should use the LVT offsets the BIOS provides.  The offsets
 372 * are determined by the subsystems using it like those for MCE
 373 * threshold or IBS.  On K8 only offset 0 (APIC500) and MCE interrupts
 374 * are supported. Beginning with family 10h at least 4 offsets are
 375 * available.
 376 *
 377 * Since the offsets must be consistent for all cores, we keep track
 378 * of the LVT offsets in software and reserve the offset for the same
 379 * vector also to be used on other cores. An offset is freed by
 380 * setting the entry to APIC_EILVT_MASKED.
 381 *
 382 * If the BIOS is right, there should be no conflicts. Otherwise a
 383 * "[Firmware Bug]: ..." error message is generated. However, if
 384 * software does not properly determines the offsets, it is not
 385 * necessarily a BIOS bug.
 386 */
 387
 388static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
 389
 390static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
 391{
 392	return (old & APIC_EILVT_MASKED)
 393		|| (new == APIC_EILVT_MASKED)
 394		|| ((new & ~APIC_EILVT_MASKED) == old);
 395}
 396
 397static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
 398{
 399	unsigned int rsvd, vector;
 400
 401	if (offset >= APIC_EILVT_NR_MAX)
 402		return ~0;
 403
 404	rsvd = atomic_read(&eilvt_offsets[offset]);
 405	do {
 406		vector = rsvd & ~APIC_EILVT_MASKED;	/* 0: unassigned */
 407		if (vector && !eilvt_entry_is_changeable(vector, new))
 408			/* may not change if vectors are different */
 409			return rsvd;
 410		rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
 411	} while (rsvd != new);
 412
 413	rsvd &= ~APIC_EILVT_MASKED;
 414	if (rsvd && rsvd != vector)
 415		pr_info("LVT offset %d assigned for vector 0x%02x\n",
 416			offset, rsvd);
 417
 418	return new;
 419}
 420
 421/*
 422 * If mask=1, the LVT entry does not generate interrupts while mask=0
 423 * enables the vector. See also the BKDGs. Must be called with
 424 * preemption disabled.
 425 */
 426
 427int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
 428{
 429	unsigned long reg = APIC_EILVTn(offset);
 430	unsigned int new, old, reserved;
 431
 432	new = (mask << 16) | (msg_type << 8) | vector;
 433	old = apic_read(reg);
 434	reserved = reserve_eilvt_offset(offset, new);
 435
 436	if (reserved != new) {
 437		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
 438		       "vector 0x%x, but the register is already in use for "
 439		       "vector 0x%x on another cpu\n",
 440		       smp_processor_id(), reg, offset, new, reserved);
 441		return -EINVAL;
 442	}
 443
 444	if (!eilvt_entry_is_changeable(old, new)) {
 445		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
 446		       "vector 0x%x, but the register is already in use for "
 447		       "vector 0x%x on this cpu\n",
 448		       smp_processor_id(), reg, offset, new, old);
 449		return -EBUSY;
 450	}
 451
 452	apic_write(reg, new);
 453
 454	return 0;
 455}
 456EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
 457
 458/*
 459 * Program the next event, relative to now
 460 */
 461static int lapic_next_event(unsigned long delta,
 462			    struct clock_event_device *evt)
 463{
 464	apic_write(APIC_TMICT, delta);
 465	return 0;
 466}
 467
 468static int lapic_next_deadline(unsigned long delta,
 469			       struct clock_event_device *evt)
 470{
 471	u64 tsc;
 472
 473	tsc = rdtsc();
 474	wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
 475	return 0;
 476}
 477
 478static int lapic_timer_shutdown(struct clock_event_device *evt)
 479{
 480	unsigned int v;
 481
 482	/* Lapic used as dummy for broadcast ? */
 483	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
 484		return 0;
 485
 486	v = apic_read(APIC_LVTT);
 487	v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
 488	apic_write(APIC_LVTT, v);
 489	apic_write(APIC_TMICT, 0);
 490	return 0;
 491}
 492
 493static inline int
 494lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot)
 495{
 496	/* Lapic used as dummy for broadcast ? */
 497	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
 498		return 0;
 499
 500	__setup_APIC_LVTT(lapic_timer_frequency, oneshot, 1);
 501	return 0;
 502}
 503
 504static int lapic_timer_set_periodic(struct clock_event_device *evt)
 505{
 506	return lapic_timer_set_periodic_oneshot(evt, false);
 507}
 508
 509static int lapic_timer_set_oneshot(struct clock_event_device *evt)
 510{
 511	return lapic_timer_set_periodic_oneshot(evt, true);
 512}
 513
 514/*
 515 * Local APIC timer broadcast function
 516 */
 517static void lapic_timer_broadcast(const struct cpumask *mask)
 518{
 519#ifdef CONFIG_SMP
 520	apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
 521#endif
 522}
 523
 524
 525/*
 526 * The local apic timer can be used for any function which is CPU local.
 527 */
 528static struct clock_event_device lapic_clockevent = {
 529	.name			= "lapic",
 530	.features		= CLOCK_EVT_FEAT_PERIODIC |
 531				  CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP
 532				  | CLOCK_EVT_FEAT_DUMMY,
 533	.shift			= 32,
 534	.set_state_shutdown	= lapic_timer_shutdown,
 535	.set_state_periodic	= lapic_timer_set_periodic,
 536	.set_state_oneshot	= lapic_timer_set_oneshot,
 537	.set_next_event		= lapic_next_event,
 538	.broadcast		= lapic_timer_broadcast,
 539	.rating			= 100,
 540	.irq			= -1,
 
 541};
 542static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
 543
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 544/*
 545 * Setup the local APIC timer for this CPU. Copy the initialized values
 546 * of the boot CPU and register the clock event in the framework.
 547 */
 548static void setup_APIC_timer(void)
 549{
 550	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
 551
 552	if (this_cpu_has(X86_FEATURE_ARAT)) {
 553		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
 554		/* Make LAPIC timer preferrable over percpu HPET */
 555		lapic_clockevent.rating = 150;
 556	}
 557
 558	memcpy(levt, &lapic_clockevent, sizeof(*levt));
 559	levt->cpumask = cpumask_of(smp_processor_id());
 560
 561	if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
 
 562		levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
 563				    CLOCK_EVT_FEAT_DUMMY);
 564		levt->set_next_event = lapic_next_deadline;
 565		clockevents_config_and_register(levt,
 566						(tsc_khz / TSC_DIVISOR) * 1000,
 567						0xF, ~0UL);
 568	} else
 569		clockevents_register_device(levt);
 570}
 571
 572/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 573 * In this functions we calibrate APIC bus clocks to the external timer.
 574 *
 575 * We want to do the calibration only once since we want to have local timer
 576 * irqs syncron. CPUs connected by the same APIC bus have the very same bus
 577 * frequency.
 578 *
 579 * This was previously done by reading the PIT/HPET and waiting for a wrap
 580 * around to find out, that a tick has elapsed. I have a box, where the PIT
 581 * readout is broken, so it never gets out of the wait loop again. This was
 582 * also reported by others.
 583 *
 584 * Monitoring the jiffies value is inaccurate and the clockevents
 585 * infrastructure allows us to do a simple substitution of the interrupt
 586 * handler.
 587 *
 588 * The calibration routine also uses the pm_timer when possible, as the PIT
 589 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
 590 * back to normal later in the boot process).
 591 */
 592
 593#define LAPIC_CAL_LOOPS		(HZ/10)
 594
 595static __initdata int lapic_cal_loops = -1;
 596static __initdata long lapic_cal_t1, lapic_cal_t2;
 597static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
 598static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
 599static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
 600
 601/*
 602 * Temporary interrupt handler.
 603 */
 604static void __init lapic_cal_handler(struct clock_event_device *dev)
 605{
 606	unsigned long long tsc = 0;
 607	long tapic = apic_read(APIC_TMCCT);
 608	unsigned long pm = acpi_pm_read_early();
 609
 610	if (cpu_has_tsc)
 611		tsc = rdtsc();
 612
 613	switch (lapic_cal_loops++) {
 614	case 0:
 615		lapic_cal_t1 = tapic;
 616		lapic_cal_tsc1 = tsc;
 617		lapic_cal_pm1 = pm;
 618		lapic_cal_j1 = jiffies;
 619		break;
 620
 621	case LAPIC_CAL_LOOPS:
 622		lapic_cal_t2 = tapic;
 623		lapic_cal_tsc2 = tsc;
 624		if (pm < lapic_cal_pm1)
 625			pm += ACPI_PM_OVRRUN;
 626		lapic_cal_pm2 = pm;
 627		lapic_cal_j2 = jiffies;
 628		break;
 629	}
 630}
 631
 632static int __init
 633calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
 634{
 635	const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
 636	const long pm_thresh = pm_100ms / 100;
 637	unsigned long mult;
 638	u64 res;
 639
 640#ifndef CONFIG_X86_PM_TIMER
 641	return -1;
 642#endif
 643
 644	apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
 645
 646	/* Check, if the PM timer is available */
 647	if (!deltapm)
 648		return -1;
 649
 650	mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
 651
 652	if (deltapm > (pm_100ms - pm_thresh) &&
 653	    deltapm < (pm_100ms + pm_thresh)) {
 654		apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
 655		return 0;
 656	}
 657
 658	res = (((u64)deltapm) *  mult) >> 22;
 659	do_div(res, 1000000);
 660	pr_warning("APIC calibration not consistent "
 661		   "with PM-Timer: %ldms instead of 100ms\n",(long)res);
 662
 663	/* Correct the lapic counter value */
 664	res = (((u64)(*delta)) * pm_100ms);
 665	do_div(res, deltapm);
 666	pr_info("APIC delta adjusted to PM-Timer: "
 667		"%lu (%ld)\n", (unsigned long)res, *delta);
 668	*delta = (long)res;
 669
 670	/* Correct the tsc counter value */
 671	if (cpu_has_tsc) {
 672		res = (((u64)(*deltatsc)) * pm_100ms);
 673		do_div(res, deltapm);
 674		apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
 675					  "PM-Timer: %lu (%ld)\n",
 676					(unsigned long)res, *deltatsc);
 677		*deltatsc = (long)res;
 678	}
 679
 680	return 0;
 681}
 682
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 683static int __init calibrate_APIC_clock(void)
 684{
 685	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
 686	void (*real_handler)(struct clock_event_device *dev);
 
 687	unsigned long deltaj;
 688	long delta, deltatsc;
 689	int pm_referenced = 0;
 690
 691	/**
 692	 * check if lapic timer has already been calibrated by platform
 693	 * specific routine, such as tsc calibration code. if so, we just fill
 
 
 
 694	 * in the clockevent structure and return.
 695	 */
 696
 697	if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
 698		return 0;
 699	} else if (lapic_timer_frequency) {
 700		apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
 701				lapic_timer_frequency);
 702		lapic_clockevent.mult = div_sc(lapic_timer_frequency/APIC_DIVISOR,
 703					TICK_NSEC, lapic_clockevent.shift);
 704		lapic_clockevent.max_delta_ns =
 705			clockevent_delta2ns(0x7FFFFF, &lapic_clockevent);
 706		lapic_clockevent.min_delta_ns =
 707			clockevent_delta2ns(0xF, &lapic_clockevent);
 708		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
 709		return 0;
 710	}
 711
 712	apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
 713		    "calibrating APIC timer ...\n");
 714
 
 
 
 
 
 715	local_irq_disable();
 716
 717	/* Replace the global interrupt handler */
 718	real_handler = global_clock_event->event_handler;
 719	global_clock_event->event_handler = lapic_cal_handler;
 720
 721	/*
 722	 * Setup the APIC counter to maximum. There is no way the lapic
 723	 * can underflow in the 100ms detection time frame
 724	 */
 725	__setup_APIC_LVTT(0xffffffff, 0, 0);
 726
 727	/* Let the interrupts run */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 728	local_irq_enable();
 729
 730	while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
 731		cpu_relax();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 732
 733	local_irq_disable();
 734
 735	/* Restore the real event handler */
 736	global_clock_event->event_handler = real_handler;
 737
 738	/* Build delta t1-t2 as apic timer counts down */
 739	delta = lapic_cal_t1 - lapic_cal_t2;
 740	apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
 741
 742	deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
 743
 744	/* we trust the PM based calibration if possible */
 745	pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
 746					&delta, &deltatsc);
 747
 748	/* Calculate the scaled math multiplication factor */
 749	lapic_clockevent.mult = div_sc(delta, TICK_NSEC * LAPIC_CAL_LOOPS,
 750				       lapic_clockevent.shift);
 751	lapic_clockevent.max_delta_ns =
 752		clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
 753	lapic_clockevent.min_delta_ns =
 754		clockevent_delta2ns(0xF, &lapic_clockevent);
 755
 756	lapic_timer_frequency = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
 757
 758	apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
 759	apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
 760	apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
 761		    lapic_timer_frequency);
 762
 763	if (cpu_has_tsc) {
 764		apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
 765			    "%ld.%04ld MHz.\n",
 766			    (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
 767			    (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
 768	}
 769
 770	apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
 771		    "%u.%04u MHz.\n",
 772		    lapic_timer_frequency / (1000000 / HZ),
 773		    lapic_timer_frequency % (1000000 / HZ));
 774
 775	/*
 776	 * Do a sanity check on the APIC calibration result
 777	 */
 778	if (lapic_timer_frequency < (1000000 / HZ)) {
 779		local_irq_enable();
 780		pr_warning("APIC frequency too slow, disabling apic timer\n");
 781		return -1;
 782	}
 783
 784	levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
 785
 786	/*
 787	 * PM timer calibration failed or not turned on
 788	 * so lets try APIC timer based calibration
 
 789	 */
 790	if (!pm_referenced) {
 791		apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
 792
 793		/*
 794		 * Setup the apic timer manually
 795		 */
 796		levt->event_handler = lapic_cal_handler;
 797		lapic_timer_set_periodic(levt);
 798		lapic_cal_loops = -1;
 799
 800		/* Let the interrupts run */
 801		local_irq_enable();
 802
 803		while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
 804			cpu_relax();
 805
 806		/* Stop the lapic timer */
 807		local_irq_disable();
 808		lapic_timer_shutdown(levt);
 809
 810		/* Jiffies delta */
 811		deltaj = lapic_cal_j2 - lapic_cal_j1;
 812		apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
 813
 814		/* Check, if the jiffies result is consistent */
 815		if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
 816			apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
 817		else
 818			levt->features |= CLOCK_EVT_FEAT_DUMMY;
 819	}
 820	local_irq_enable();
 821
 822	if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
 823		pr_warning("APIC timer disabled due to verification failure\n");
 824			return -1;
 825	}
 826
 827	return 0;
 828}
 829
 830/*
 831 * Setup the boot APIC
 832 *
 833 * Calibrate and verify the result.
 834 */
 835void __init setup_boot_APIC_clock(void)
 836{
 837	/*
 838	 * The local apic timer can be disabled via the kernel
 839	 * commandline or from the CPU detection code. Register the lapic
 840	 * timer as a dummy clock event source on SMP systems, so the
 841	 * broadcast mechanism is used. On UP systems simply ignore it.
 842	 */
 843	if (disable_apic_timer) {
 844		pr_info("Disabling APIC timer\n");
 845		/* No broadcast on UP ! */
 846		if (num_possible_cpus() > 1) {
 847			lapic_clockevent.mult = 1;
 848			setup_APIC_timer();
 849		}
 850		return;
 851	}
 852
 853	if (calibrate_APIC_clock()) {
 854		/* No broadcast on UP ! */
 855		if (num_possible_cpus() > 1)
 856			setup_APIC_timer();
 857		return;
 858	}
 859
 860	/*
 861	 * If nmi_watchdog is set to IO_APIC, we need the
 862	 * PIT/HPET going.  Otherwise register lapic as a dummy
 863	 * device.
 864	 */
 865	lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
 866
 867	/* Setup the lapic or request the broadcast */
 868	setup_APIC_timer();
 
 869}
 870
 871void setup_secondary_APIC_clock(void)
 872{
 873	setup_APIC_timer();
 
 874}
 875
 876/*
 877 * The guts of the apic timer interrupt
 878 */
 879static void local_apic_timer_interrupt(void)
 880{
 881	int cpu = smp_processor_id();
 882	struct clock_event_device *evt = &per_cpu(lapic_events, cpu);
 883
 884	/*
 885	 * Normally we should not be here till LAPIC has been initialized but
 886	 * in some cases like kdump, its possible that there is a pending LAPIC
 887	 * timer interrupt from previous kernel's context and is delivered in
 888	 * new kernel the moment interrupts are enabled.
 889	 *
 890	 * Interrupts are enabled early and LAPIC is setup much later, hence
 891	 * its possible that when we get here evt->event_handler is NULL.
 892	 * Check for event_handler being NULL and discard the interrupt as
 893	 * spurious.
 894	 */
 895	if (!evt->event_handler) {
 896		pr_warning("Spurious LAPIC timer interrupt on cpu %d\n", cpu);
 
 897		/* Switch it off */
 898		lapic_timer_shutdown(evt);
 899		return;
 900	}
 901
 902	/*
 903	 * the NMI deadlock-detector uses this.
 904	 */
 905	inc_irq_stat(apic_timer_irqs);
 906
 907	evt->event_handler(evt);
 908}
 909
 910/*
 911 * Local APIC timer interrupt. This is the most natural way for doing
 912 * local interrupts, but local timer interrupts can be emulated by
 913 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
 914 *
 915 * [ if a single-CPU system runs an SMP kernel then we call the local
 916 *   interrupt as well. Thus we cannot inline the local irq ... ]
 917 */
 918__visible void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
 919{
 920	struct pt_regs *old_regs = set_irq_regs(regs);
 921
 922	/*
 923	 * NOTE! We'd better ACK the irq immediately,
 924	 * because timer handling can be slow.
 925	 *
 926	 * update_process_times() expects us to have done irq_enter().
 927	 * Besides, if we don't timer interrupts ignore the global
 928	 * interrupt lock, which is the WrongThing (tm) to do.
 929	 */
 930	entering_ack_irq();
 931	local_apic_timer_interrupt();
 932	exiting_irq();
 933
 934	set_irq_regs(old_regs);
 935}
 936
 937__visible void __irq_entry smp_trace_apic_timer_interrupt(struct pt_regs *regs)
 938{
 939	struct pt_regs *old_regs = set_irq_regs(regs);
 940
 941	/*
 942	 * NOTE! We'd better ACK the irq immediately,
 943	 * because timer handling can be slow.
 944	 *
 945	 * update_process_times() expects us to have done irq_enter().
 946	 * Besides, if we don't timer interrupts ignore the global
 947	 * interrupt lock, which is the WrongThing (tm) to do.
 948	 */
 949	entering_ack_irq();
 950	trace_local_timer_entry(LOCAL_TIMER_VECTOR);
 951	local_apic_timer_interrupt();
 952	trace_local_timer_exit(LOCAL_TIMER_VECTOR);
 953	exiting_irq();
 954
 955	set_irq_regs(old_regs);
 956}
 957
 958int setup_profiling_timer(unsigned int multiplier)
 959{
 960	return -EINVAL;
 961}
 962
 963/*
 964 * Local APIC start and shutdown
 965 */
 966
 967/**
 968 * clear_local_APIC - shutdown the local APIC
 969 *
 970 * This is called, when a CPU is disabled and before rebooting, so the state of
 971 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
 972 * leftovers during boot.
 973 */
 974void clear_local_APIC(void)
 975{
 976	int maxlvt;
 977	u32 v;
 978
 979	/* APIC hasn't been mapped yet */
 980	if (!x2apic_mode && !apic_phys)
 981		return;
 982
 983	maxlvt = lapic_get_maxlvt();
 984	/*
 985	 * Masking an LVT entry can trigger a local APIC error
 986	 * if the vector is zero. Mask LVTERR first to prevent this.
 987	 */
 988	if (maxlvt >= 3) {
 989		v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
 990		apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
 991	}
 992	/*
 993	 * Careful: we have to set masks only first to deassert
 994	 * any level-triggered sources.
 995	 */
 996	v = apic_read(APIC_LVTT);
 997	apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
 998	v = apic_read(APIC_LVT0);
 999	apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1000	v = apic_read(APIC_LVT1);
1001	apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1002	if (maxlvt >= 4) {
1003		v = apic_read(APIC_LVTPC);
1004		apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1005	}
1006
1007	/* lets not touch this if we didn't frob it */
1008#ifdef CONFIG_X86_THERMAL_VECTOR
1009	if (maxlvt >= 5) {
1010		v = apic_read(APIC_LVTTHMR);
1011		apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1012	}
1013#endif
1014#ifdef CONFIG_X86_MCE_INTEL
1015	if (maxlvt >= 6) {
1016		v = apic_read(APIC_LVTCMCI);
1017		if (!(v & APIC_LVT_MASKED))
1018			apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1019	}
1020#endif
1021
1022	/*
1023	 * Clean APIC state for other OSs:
1024	 */
1025	apic_write(APIC_LVTT, APIC_LVT_MASKED);
1026	apic_write(APIC_LVT0, APIC_LVT_MASKED);
1027	apic_write(APIC_LVT1, APIC_LVT_MASKED);
1028	if (maxlvt >= 3)
1029		apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1030	if (maxlvt >= 4)
1031		apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1032
1033	/* Integrated APIC (!82489DX) ? */
1034	if (lapic_is_integrated()) {
1035		if (maxlvt > 3)
1036			/* Clear ESR due to Pentium errata 3AP and 11AP */
1037			apic_write(APIC_ESR, 0);
1038		apic_read(APIC_ESR);
1039	}
1040}
1041
1042/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1043 * disable_local_APIC - clear and disable the local APIC
1044 */
1045void disable_local_APIC(void)
1046{
1047	unsigned int value;
1048
1049	/* APIC hasn't been mapped yet */
1050	if (!x2apic_mode && !apic_phys)
1051		return;
1052
1053	clear_local_APIC();
1054
1055	/*
1056	 * Disable APIC (implies clearing of registers
1057	 * for 82489DX!).
1058	 */
1059	value = apic_read(APIC_SPIV);
1060	value &= ~APIC_SPIV_APIC_ENABLED;
1061	apic_write(APIC_SPIV, value);
1062
1063#ifdef CONFIG_X86_32
1064	/*
1065	 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1066	 * restore the disabled state.
1067	 */
1068	if (enabled_via_apicbase) {
1069		unsigned int l, h;
1070
1071		rdmsr(MSR_IA32_APICBASE, l, h);
1072		l &= ~MSR_IA32_APICBASE_ENABLE;
1073		wrmsr(MSR_IA32_APICBASE, l, h);
1074	}
1075#endif
1076}
1077
1078/*
1079 * If Linux enabled the LAPIC against the BIOS default disable it down before
1080 * re-entering the BIOS on shutdown.  Otherwise the BIOS may get confused and
1081 * not power-off.  Additionally clear all LVT entries before disable_local_APIC
1082 * for the case where Linux didn't enable the LAPIC.
1083 */
1084void lapic_shutdown(void)
1085{
1086	unsigned long flags;
1087
1088	if (!cpu_has_apic && !apic_from_smp_config())
1089		return;
1090
1091	local_irq_save(flags);
1092
1093#ifdef CONFIG_X86_32
1094	if (!enabled_via_apicbase)
1095		clear_local_APIC();
1096	else
1097#endif
1098		disable_local_APIC();
1099
1100
1101	local_irq_restore(flags);
1102}
1103
1104/**
1105 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1106 */
1107void __init sync_Arb_IDs(void)
1108{
1109	/*
1110	 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1111	 * needed on AMD.
1112	 */
1113	if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1114		return;
1115
1116	/*
1117	 * Wait for idle.
1118	 */
1119	apic_wait_icr_idle();
1120
1121	apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1122	apic_write(APIC_ICR, APIC_DEST_ALLINC |
1123			APIC_INT_LEVELTRIG | APIC_DM_INIT);
1124}
1125
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1126/*
1127 * An initial setup of the virtual wire mode.
1128 */
1129void __init init_bsp_APIC(void)
1130{
1131	unsigned int value;
1132
1133	/*
1134	 * Don't do the setup now if we have a SMP BIOS as the
1135	 * through-I/O-APIC virtual wire mode might be active.
1136	 */
1137	if (smp_found_config || !cpu_has_apic)
1138		return;
1139
1140	/*
1141	 * Do not trust the local APIC being empty at bootup.
1142	 */
1143	clear_local_APIC();
1144
1145	/*
1146	 * Enable APIC.
1147	 */
1148	value = apic_read(APIC_SPIV);
1149	value &= ~APIC_VECTOR_MASK;
1150	value |= APIC_SPIV_APIC_ENABLED;
1151
1152#ifdef CONFIG_X86_32
1153	/* This bit is reserved on P4/Xeon and should be cleared */
1154	if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1155	    (boot_cpu_data.x86 == 15))
1156		value &= ~APIC_SPIV_FOCUS_DISABLED;
1157	else
1158#endif
1159		value |= APIC_SPIV_FOCUS_DISABLED;
1160	value |= SPURIOUS_APIC_VECTOR;
1161	apic_write(APIC_SPIV, value);
1162
1163	/*
1164	 * Set up the virtual wire mode.
1165	 */
1166	apic_write(APIC_LVT0, APIC_DM_EXTINT);
1167	value = APIC_DM_NMI;
1168	if (!lapic_is_integrated())		/* 82489DX */
1169		value |= APIC_LVT_LEVEL_TRIGGER;
1170	if (apic_extnmi == APIC_EXTNMI_NONE)
1171		value |= APIC_LVT_MASKED;
1172	apic_write(APIC_LVT1, value);
1173}
1174
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1175static void lapic_setup_esr(void)
1176{
1177	unsigned int oldvalue, value, maxlvt;
1178
1179	if (!lapic_is_integrated()) {
1180		pr_info("No ESR for 82489DX.\n");
1181		return;
1182	}
1183
1184	if (apic->disable_esr) {
1185		/*
1186		 * Something untraceable is creating bad interrupts on
1187		 * secondary quads ... for the moment, just leave the
1188		 * ESR disabled - we can't do anything useful with the
1189		 * errors anyway - mbligh
1190		 */
1191		pr_info("Leaving ESR disabled.\n");
1192		return;
1193	}
1194
1195	maxlvt = lapic_get_maxlvt();
1196	if (maxlvt > 3)		/* Due to the Pentium erratum 3AP. */
1197		apic_write(APIC_ESR, 0);
1198	oldvalue = apic_read(APIC_ESR);
1199
1200	/* enables sending errors */
1201	value = ERROR_APIC_VECTOR;
1202	apic_write(APIC_LVTERR, value);
1203
1204	/*
1205	 * spec says clear errors after enabling vector.
1206	 */
1207	if (maxlvt > 3)
1208		apic_write(APIC_ESR, 0);
1209	value = apic_read(APIC_ESR);
1210	if (value != oldvalue)
1211		apic_printk(APIC_VERBOSE, "ESR value before enabling "
1212			"vector: 0x%08x  after: 0x%08x\n",
1213			oldvalue, value);
1214}
1215
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1216/**
1217 * setup_local_APIC - setup the local APIC
1218 *
1219 * Used to setup local APIC while initializing BSP or bringin up APs.
1220 * Always called with preemption disabled.
1221 */
1222void setup_local_APIC(void)
1223{
1224	int cpu = smp_processor_id();
1225	unsigned int value, queued;
1226	int i, j, acked = 0;
1227	unsigned long long tsc = 0, ntsc;
1228	long long max_loops = cpu_khz ? cpu_khz : 1000000;
1229
1230	if (cpu_has_tsc)
1231		tsc = rdtsc();
1232
1233	if (disable_apic) {
1234		disable_ioapic_support();
1235		return;
1236	}
1237
 
 
 
 
 
 
 
 
1238#ifdef CONFIG_X86_32
1239	/* Pound the ESR really hard over the head with a big hammer - mbligh */
1240	if (lapic_is_integrated() && apic->disable_esr) {
1241		apic_write(APIC_ESR, 0);
1242		apic_write(APIC_ESR, 0);
1243		apic_write(APIC_ESR, 0);
1244		apic_write(APIC_ESR, 0);
1245	}
1246#endif
1247	perf_events_lapic_init();
1248
1249	/*
1250	 * Double-check whether this APIC is really registered.
1251	 * This is meaningless in clustered apic mode, so we skip it.
1252	 */
1253	BUG_ON(!apic->apic_id_registered());
1254
1255	/*
1256	 * Intel recommends to set DFR, LDR and TPR before enabling
1257	 * an APIC.  See e.g. "AP-388 82489DX User's Manual" (Intel
1258	 * document number 292116).  So here it goes...
1259	 */
1260	apic->init_apic_ldr();
1261
1262#ifdef CONFIG_X86_32
1263	/*
1264	 * APIC LDR is initialized.  If logical_apicid mapping was
1265	 * initialized during get_smp_config(), make sure it matches the
1266	 * actual value.
1267	 */
1268	i = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1269	WARN_ON(i != BAD_APICID && i != logical_smp_processor_id());
1270	/* always use the value from LDR */
1271	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
1272		logical_smp_processor_id();
 
 
 
 
 
1273#endif
1274
1275	/*
1276	 * Set Task Priority to 'accept all'. We never change this
1277	 * later on.
 
 
1278	 */
1279	value = apic_read(APIC_TASKPRI);
1280	value &= ~APIC_TPRI_MASK;
 
1281	apic_write(APIC_TASKPRI, value);
1282
1283	/*
1284	 * After a crash, we no longer service the interrupts and a pending
1285	 * interrupt from previous kernel might still have ISR bit set.
1286	 *
1287	 * Most probably by now CPU has serviced that pending interrupt and
1288	 * it might not have done the ack_APIC_irq() because it thought,
1289	 * interrupt came from i8259 as ExtInt. LAPIC did not get EOI so it
1290	 * does not clear the ISR bit and cpu thinks it has already serivced
1291	 * the interrupt. Hence a vector might get locked. It was noticed
1292	 * for timer irq (vector 0x31). Issue an extra EOI to clear ISR.
1293	 */
1294	do {
1295		queued = 0;
1296		for (i = APIC_ISR_NR - 1; i >= 0; i--)
1297			queued |= apic_read(APIC_IRR + i*0x10);
1298
1299		for (i = APIC_ISR_NR - 1; i >= 0; i--) {
1300			value = apic_read(APIC_ISR + i*0x10);
1301			for (j = 31; j >= 0; j--) {
1302				if (value & (1<<j)) {
1303					ack_APIC_irq();
1304					acked++;
1305				}
1306			}
1307		}
1308		if (acked > 256) {
1309			printk(KERN_ERR "LAPIC pending interrupts after %d EOI\n",
1310			       acked);
1311			break;
1312		}
1313		if (queued) {
1314			if (cpu_has_tsc && cpu_khz) {
1315				ntsc = rdtsc();
1316				max_loops = (cpu_khz << 10) - (ntsc - tsc);
1317			} else
1318				max_loops--;
1319		}
1320	} while (queued && max_loops > 0);
1321	WARN_ON(max_loops <= 0);
1322
1323	/*
1324	 * Now that we are all set up, enable the APIC
1325	 */
1326	value = apic_read(APIC_SPIV);
1327	value &= ~APIC_VECTOR_MASK;
1328	/*
1329	 * Enable APIC
1330	 */
1331	value |= APIC_SPIV_APIC_ENABLED;
1332
1333#ifdef CONFIG_X86_32
1334	/*
1335	 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1336	 * certain networking cards. If high frequency interrupts are
1337	 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1338	 * entry is masked/unmasked at a high rate as well then sooner or
1339	 * later IOAPIC line gets 'stuck', no more interrupts are received
1340	 * from the device. If focus CPU is disabled then the hang goes
1341	 * away, oh well :-(
1342	 *
1343	 * [ This bug can be reproduced easily with a level-triggered
1344	 *   PCI Ne2000 networking cards and PII/PIII processors, dual
1345	 *   BX chipset. ]
1346	 */
1347	/*
1348	 * Actually disabling the focus CPU check just makes the hang less
1349	 * frequent as it makes the interrupt distributon model be more
1350	 * like LRU than MRU (the short-term load is more even across CPUs).
1351	 * See also the comment in end_level_ioapic_irq().  --macro
1352	 */
1353
1354	/*
1355	 * - enable focus processor (bit==0)
1356	 * - 64bit mode always use processor focus
1357	 *   so no need to set it
1358	 */
1359	value &= ~APIC_SPIV_FOCUS_DISABLED;
1360#endif
1361
1362	/*
1363	 * Set spurious IRQ vector
1364	 */
1365	value |= SPURIOUS_APIC_VECTOR;
1366	apic_write(APIC_SPIV, value);
1367
 
 
1368	/*
1369	 * Set up LVT0, LVT1:
1370	 *
1371	 * set up through-local-APIC on the BP's LINT0. This is not
1372	 * strictly necessary in pure symmetric-IO mode, but sometimes
1373	 * we delegate interrupts to the 8259A.
1374	 */
1375	/*
1376	 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1377	 */
1378	value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1379	if (!cpu && (pic_mode || !value)) {
1380		value = APIC_DM_EXTINT;
1381		apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1382	} else {
1383		value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1384		apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1385	}
1386	apic_write(APIC_LVT0, value);
1387
1388	/*
1389	 * Only the BSP sees the LINT1 NMI signal by default. This can be
1390	 * modified by apic_extnmi= boot option.
1391	 */
1392	if ((!cpu && apic_extnmi != APIC_EXTNMI_NONE) ||
1393	    apic_extnmi == APIC_EXTNMI_ALL)
1394		value = APIC_DM_NMI;
1395	else
1396		value = APIC_DM_NMI | APIC_LVT_MASKED;
1397	if (!lapic_is_integrated())		/* 82489DX */
 
 
1398		value |= APIC_LVT_LEVEL_TRIGGER;
1399	apic_write(APIC_LVT1, value);
1400
1401#ifdef CONFIG_X86_MCE_INTEL
1402	/* Recheck CMCI information after local APIC is up on CPU #0 */
1403	if (!cpu)
1404		cmci_recheck();
1405#endif
1406}
1407
1408static void end_local_APIC_setup(void)
1409{
1410	lapic_setup_esr();
1411
1412#ifdef CONFIG_X86_32
1413	{
1414		unsigned int value;
1415		/* Disable the local apic timer */
1416		value = apic_read(APIC_LVTT);
1417		value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1418		apic_write(APIC_LVTT, value);
1419	}
1420#endif
1421
1422	apic_pm_activate();
1423}
1424
1425/*
1426 * APIC setup function for application processors. Called from smpboot.c
1427 */
1428void apic_ap_setup(void)
1429{
1430	setup_local_APIC();
1431	end_local_APIC_setup();
1432}
1433
1434#ifdef CONFIG_X86_X2APIC
1435int x2apic_mode;
1436
1437enum {
1438	X2APIC_OFF,
1439	X2APIC_ON,
1440	X2APIC_DISABLED,
1441};
1442static int x2apic_state;
1443
1444static void __x2apic_disable(void)
1445{
1446	u64 msr;
1447
1448	if (!cpu_has_apic)
1449		return;
1450
1451	rdmsrl(MSR_IA32_APICBASE, msr);
1452	if (!(msr & X2APIC_ENABLE))
1453		return;
1454	/* Disable xapic and x2apic first and then reenable xapic mode */
1455	wrmsrl(MSR_IA32_APICBASE, msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1456	wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1457	printk_once(KERN_INFO "x2apic disabled\n");
1458}
1459
1460static void __x2apic_enable(void)
1461{
1462	u64 msr;
1463
1464	rdmsrl(MSR_IA32_APICBASE, msr);
1465	if (msr & X2APIC_ENABLE)
1466		return;
1467	wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1468	printk_once(KERN_INFO "x2apic enabled\n");
1469}
1470
1471static int __init setup_nox2apic(char *str)
1472{
1473	if (x2apic_enabled()) {
1474		int apicid = native_apic_msr_read(APIC_ID);
1475
1476		if (apicid >= 255) {
1477			pr_warning("Apicid: %08x, cannot enforce nox2apic\n",
1478				   apicid);
1479			return 0;
1480		}
1481		pr_warning("x2apic already enabled.\n");
1482		__x2apic_disable();
1483	}
1484	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1485	x2apic_state = X2APIC_DISABLED;
1486	x2apic_mode = 0;
1487	return 0;
1488}
1489early_param("nox2apic", setup_nox2apic);
1490
1491/* Called from cpu_init() to enable x2apic on (secondary) cpus */
1492void x2apic_setup(void)
1493{
1494	/*
1495	 * If x2apic is not in ON state, disable it if already enabled
1496	 * from BIOS.
1497	 */
1498	if (x2apic_state != X2APIC_ON) {
1499		__x2apic_disable();
1500		return;
1501	}
1502	__x2apic_enable();
1503}
1504
1505static __init void x2apic_disable(void)
1506{
1507	u32 x2apic_id, state = x2apic_state;
1508
1509	x2apic_mode = 0;
1510	x2apic_state = X2APIC_DISABLED;
1511
1512	if (state != X2APIC_ON)
1513		return;
1514
1515	x2apic_id = read_apic_id();
1516	if (x2apic_id >= 255)
1517		panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1518
1519	__x2apic_disable();
1520	register_lapic_address(mp_lapic_addr);
1521}
1522
1523static __init void x2apic_enable(void)
1524{
1525	if (x2apic_state != X2APIC_OFF)
1526		return;
1527
1528	x2apic_mode = 1;
1529	x2apic_state = X2APIC_ON;
1530	__x2apic_enable();
1531}
1532
1533static __init void try_to_enable_x2apic(int remap_mode)
1534{
1535	if (x2apic_state == X2APIC_DISABLED)
1536		return;
1537
1538	if (remap_mode != IRQ_REMAP_X2APIC_MODE) {
1539		/* IR is required if there is APIC ID > 255 even when running
1540		 * under KVM
1541		 */
1542		if (max_physical_apicid > 255 ||
1543		    !hypervisor_x2apic_available()) {
1544			pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n");
1545			x2apic_disable();
1546			return;
1547		}
1548
1549		/*
1550		 * without IR all CPUs can be addressed by IOAPIC/MSI
1551		 * only in physical mode
1552		 */
1553		x2apic_phys = 1;
1554	}
1555	x2apic_enable();
1556}
1557
1558void __init check_x2apic(void)
1559{
1560	if (x2apic_enabled()) {
1561		pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n");
1562		x2apic_mode = 1;
1563		x2apic_state = X2APIC_ON;
1564	} else if (!cpu_has_x2apic) {
1565		x2apic_state = X2APIC_DISABLED;
1566	}
1567}
1568#else /* CONFIG_X86_X2APIC */
1569static int __init validate_x2apic(void)
1570{
1571	if (!apic_is_x2apic_enabled())
1572		return 0;
1573	/*
1574	 * Checkme: Can we simply turn off x2apic here instead of panic?
1575	 */
1576	panic("BIOS has enabled x2apic but kernel doesn't support x2apic, please disable x2apic in BIOS.\n");
1577}
1578early_initcall(validate_x2apic);
1579
1580static inline void try_to_enable_x2apic(int remap_mode) { }
1581static inline void __x2apic_enable(void) { }
1582#endif /* !CONFIG_X86_X2APIC */
1583
1584static int __init try_to_enable_IR(void)
1585{
1586#ifdef CONFIG_X86_IO_APIC
1587	if (!x2apic_enabled() && skip_ioapic_setup) {
1588		pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n");
1589		return -1;
1590	}
1591#endif
1592	return irq_remapping_enable();
1593}
1594
1595void __init enable_IR_x2apic(void)
1596{
1597	unsigned long flags;
1598	int ret, ir_stat;
1599
 
 
 
 
 
1600	ir_stat = irq_remapping_prepare();
1601	if (ir_stat < 0 && !x2apic_supported())
1602		return;
1603
1604	ret = save_ioapic_entries();
1605	if (ret) {
1606		pr_info("Saving IO-APIC state failed: %d\n", ret);
1607		return;
1608	}
1609
1610	local_irq_save(flags);
1611	legacy_pic->mask_all();
1612	mask_ioapic_entries();
1613
1614	/* If irq_remapping_prepare() succeeded, try to enable it */
1615	if (ir_stat >= 0)
1616		ir_stat = try_to_enable_IR();
1617	/* ir_stat contains the remap mode or an error code */
1618	try_to_enable_x2apic(ir_stat);
1619
1620	if (ir_stat < 0)
1621		restore_ioapic_entries();
1622	legacy_pic->restore_mask();
1623	local_irq_restore(flags);
1624}
1625
1626#ifdef CONFIG_X86_64
1627/*
1628 * Detect and enable local APICs on non-SMP boards.
1629 * Original code written by Keir Fraser.
1630 * On AMD64 we trust the BIOS - if it says no APIC it is likely
1631 * not correctly set up (usually the APIC timer won't work etc.)
1632 */
1633static int __init detect_init_APIC(void)
1634{
1635	if (!cpu_has_apic) {
1636		pr_info("No local APIC present\n");
1637		return -1;
1638	}
1639
1640	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1641	return 0;
1642}
1643#else
1644
1645static int __init apic_verify(void)
1646{
1647	u32 features, h, l;
1648
1649	/*
1650	 * The APIC feature bit should now be enabled
1651	 * in `cpuid'
1652	 */
1653	features = cpuid_edx(1);
1654	if (!(features & (1 << X86_FEATURE_APIC))) {
1655		pr_warning("Could not enable APIC!\n");
1656		return -1;
1657	}
1658	set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1659	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1660
1661	/* The BIOS may have set up the APIC at some other address */
1662	if (boot_cpu_data.x86 >= 6) {
1663		rdmsr(MSR_IA32_APICBASE, l, h);
1664		if (l & MSR_IA32_APICBASE_ENABLE)
1665			mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1666	}
1667
1668	pr_info("Found and enabled local APIC!\n");
1669	return 0;
1670}
1671
1672int __init apic_force_enable(unsigned long addr)
1673{
1674	u32 h, l;
1675
1676	if (disable_apic)
1677		return -1;
1678
1679	/*
1680	 * Some BIOSes disable the local APIC in the APIC_BASE
1681	 * MSR. This can only be done in software for Intel P6 or later
1682	 * and AMD K7 (Model > 1) or later.
1683	 */
1684	if (boot_cpu_data.x86 >= 6) {
1685		rdmsr(MSR_IA32_APICBASE, l, h);
1686		if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1687			pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1688			l &= ~MSR_IA32_APICBASE_BASE;
1689			l |= MSR_IA32_APICBASE_ENABLE | addr;
1690			wrmsr(MSR_IA32_APICBASE, l, h);
1691			enabled_via_apicbase = 1;
1692		}
1693	}
1694	return apic_verify();
1695}
1696
1697/*
1698 * Detect and initialize APIC
1699 */
1700static int __init detect_init_APIC(void)
1701{
1702	/* Disabled by kernel option? */
1703	if (disable_apic)
1704		return -1;
1705
1706	switch (boot_cpu_data.x86_vendor) {
1707	case X86_VENDOR_AMD:
1708		if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
1709		    (boot_cpu_data.x86 >= 15))
1710			break;
1711		goto no_apic;
 
 
1712	case X86_VENDOR_INTEL:
1713		if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
1714		    (boot_cpu_data.x86 == 5 && cpu_has_apic))
1715			break;
1716		goto no_apic;
1717	default:
1718		goto no_apic;
1719	}
1720
1721	if (!cpu_has_apic) {
1722		/*
1723		 * Over-ride BIOS and try to enable the local APIC only if
1724		 * "lapic" specified.
1725		 */
1726		if (!force_enable_local_apic) {
1727			pr_info("Local APIC disabled by BIOS -- "
1728				"you can enable it with \"lapic\"\n");
1729			return -1;
1730		}
1731		if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
1732			return -1;
1733	} else {
1734		if (apic_verify())
1735			return -1;
1736	}
1737
1738	apic_pm_activate();
1739
1740	return 0;
1741
1742no_apic:
1743	pr_info("No local APIC present or hardware disabled\n");
1744	return -1;
1745}
1746#endif
1747
1748/**
1749 * init_apic_mappings - initialize APIC mappings
1750 */
1751void __init init_apic_mappings(void)
1752{
1753	unsigned int new_apicid;
1754
 
 
1755	if (x2apic_mode) {
1756		boot_cpu_physical_apicid = read_apic_id();
1757		return;
1758	}
1759
1760	/* If no local APIC can be found return early */
1761	if (!smp_found_config && detect_init_APIC()) {
1762		/* lets NOP'ify apic operations */
1763		pr_info("APIC: disable apic facility\n");
1764		apic_disable();
1765	} else {
1766		apic_phys = mp_lapic_addr;
1767
1768		/*
1769		 * acpi lapic path already maps that address in
1770		 * acpi_register_lapic_address()
1771		 */
1772		if (!acpi_lapic && !smp_found_config)
1773			register_lapic_address(apic_phys);
1774	}
1775
1776	/*
1777	 * Fetch the APIC ID of the BSP in case we have a
1778	 * default configuration (or the MP table is broken).
1779	 */
1780	new_apicid = read_apic_id();
1781	if (boot_cpu_physical_apicid != new_apicid) {
1782		boot_cpu_physical_apicid = new_apicid;
1783		/*
1784		 * yeah -- we lie about apic_version
1785		 * in case if apic was disabled via boot option
1786		 * but it's not a problem for SMP compiled kernel
1787		 * since smp_sanity_check is prepared for such a case
1788		 * and disable smp mode
1789		 */
1790		apic_version[new_apicid] =
1791			 GET_APIC_VERSION(apic_read(APIC_LVR));
1792	}
1793}
1794
1795void __init register_lapic_address(unsigned long address)
1796{
1797	mp_lapic_addr = address;
1798
1799	if (!x2apic_mode) {
1800		set_fixmap_nocache(FIX_APIC_BASE, address);
1801		apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
1802			    APIC_BASE, mp_lapic_addr);
1803	}
1804	if (boot_cpu_physical_apicid == -1U) {
1805		boot_cpu_physical_apicid  = read_apic_id();
1806		apic_version[boot_cpu_physical_apicid] =
1807			 GET_APIC_VERSION(apic_read(APIC_LVR));
1808	}
1809}
1810
1811int apic_version[MAX_LOCAL_APIC];
1812
1813/*
1814 * Local APIC interrupts
1815 */
1816
1817/*
1818 * This interrupt should _never_ happen with our APIC/SMP architecture
1819 */
1820static void __smp_spurious_interrupt(u8 vector)
1821{
 
1822	u32 v;
1823
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1824	/*
1825	 * Check if this really is a spurious interrupt and ACK it
1826	 * if it is a vectored one.  Just in case...
1827	 * Spurious interrupts should not be ACKed.
1828	 */
1829	v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1));
1830	if (v & (1 << (vector & 0x1f)))
 
 
1831		ack_APIC_irq();
1832
1833	inc_irq_stat(irq_spurious_count);
1834
1835	/* see sw-dev-man vol 3, chapter 7.4.13.5 */
1836	pr_info("spurious APIC interrupt through vector %02x on CPU#%d, "
1837		"should never happen.\n", vector, smp_processor_id());
1838}
1839
1840__visible void smp_spurious_interrupt(struct pt_regs *regs)
1841{
1842	entering_irq();
1843	__smp_spurious_interrupt(~regs->orig_ax);
1844	exiting_irq();
1845}
1846
1847__visible void smp_trace_spurious_interrupt(struct pt_regs *regs)
1848{
1849	u8 vector = ~regs->orig_ax;
1850
1851	entering_irq();
1852	trace_spurious_apic_entry(vector);
1853	__smp_spurious_interrupt(vector);
1854	trace_spurious_apic_exit(vector);
1855	exiting_irq();
1856}
1857
1858/*
1859 * This interrupt should never happen with our APIC/SMP architecture
1860 */
1861static void __smp_error_interrupt(struct pt_regs *regs)
1862{
1863	u32 v;
1864	u32 i = 0;
1865	static const char * const error_interrupt_reason[] = {
1866		"Send CS error",		/* APIC Error Bit 0 */
1867		"Receive CS error",		/* APIC Error Bit 1 */
1868		"Send accept error",		/* APIC Error Bit 2 */
1869		"Receive accept error",		/* APIC Error Bit 3 */
1870		"Redirectable IPI",		/* APIC Error Bit 4 */
1871		"Send illegal vector",		/* APIC Error Bit 5 */
1872		"Received illegal vector",	/* APIC Error Bit 6 */
1873		"Illegal register address",	/* APIC Error Bit 7 */
1874	};
 
 
 
 
1875
1876	/* First tickle the hardware, only then report what went on. -- REW */
1877	if (lapic_get_maxlvt() > 3)	/* Due to the Pentium erratum 3AP. */
1878		apic_write(APIC_ESR, 0);
1879	v = apic_read(APIC_ESR);
1880	ack_APIC_irq();
1881	atomic_inc(&irq_err_count);
1882
1883	apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x",
1884		    smp_processor_id(), v);
1885
1886	v &= 0xff;
1887	while (v) {
1888		if (v & 0x1)
1889			apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
1890		i++;
1891		v >>= 1;
1892	}
1893
1894	apic_printk(APIC_DEBUG, KERN_CONT "\n");
1895
1896}
1897
1898__visible void smp_error_interrupt(struct pt_regs *regs)
1899{
1900	entering_irq();
1901	__smp_error_interrupt(regs);
1902	exiting_irq();
1903}
1904
1905__visible void smp_trace_error_interrupt(struct pt_regs *regs)
1906{
1907	entering_irq();
1908	trace_error_apic_entry(ERROR_APIC_VECTOR);
1909	__smp_error_interrupt(regs);
1910	trace_error_apic_exit(ERROR_APIC_VECTOR);
1911	exiting_irq();
1912}
1913
1914/**
1915 * connect_bsp_APIC - attach the APIC to the interrupt system
1916 */
1917static void __init connect_bsp_APIC(void)
1918{
1919#ifdef CONFIG_X86_32
1920	if (pic_mode) {
1921		/*
1922		 * Do not trust the local APIC being empty at bootup.
1923		 */
1924		clear_local_APIC();
1925		/*
1926		 * PIC mode, enable APIC mode in the IMCR, i.e.  connect BSP's
1927		 * local APIC to INT and NMI lines.
1928		 */
1929		apic_printk(APIC_VERBOSE, "leaving PIC mode, "
1930				"enabling APIC mode.\n");
1931		imcr_pic_to_apic();
1932	}
1933#endif
1934}
1935
1936/**
1937 * disconnect_bsp_APIC - detach the APIC from the interrupt system
1938 * @virt_wire_setup:	indicates, whether virtual wire mode is selected
1939 *
1940 * Virtual wire mode is necessary to deliver legacy interrupts even when the
1941 * APIC is disabled.
1942 */
1943void disconnect_bsp_APIC(int virt_wire_setup)
1944{
1945	unsigned int value;
1946
1947#ifdef CONFIG_X86_32
1948	if (pic_mode) {
1949		/*
1950		 * Put the board back into PIC mode (has an effect only on
1951		 * certain older boards).  Note that APIC interrupts, including
1952		 * IPIs, won't work beyond this point!  The only exception are
1953		 * INIT IPIs.
1954		 */
1955		apic_printk(APIC_VERBOSE, "disabling APIC mode, "
1956				"entering PIC mode.\n");
1957		imcr_apic_to_pic();
1958		return;
1959	}
1960#endif
1961
1962	/* Go back to Virtual Wire compatibility mode */
1963
1964	/* For the spurious interrupt use vector F, and enable it */
1965	value = apic_read(APIC_SPIV);
1966	value &= ~APIC_VECTOR_MASK;
1967	value |= APIC_SPIV_APIC_ENABLED;
1968	value |= 0xf;
1969	apic_write(APIC_SPIV, value);
1970
1971	if (!virt_wire_setup) {
1972		/*
1973		 * For LVT0 make it edge triggered, active high,
1974		 * external and enabled
1975		 */
1976		value = apic_read(APIC_LVT0);
1977		value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
1978			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
1979			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
1980		value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
1981		value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
1982		apic_write(APIC_LVT0, value);
1983	} else {
1984		/* Disable LVT0 */
1985		apic_write(APIC_LVT0, APIC_LVT_MASKED);
1986	}
1987
1988	/*
1989	 * For LVT1 make it edge triggered, active high,
1990	 * nmi and enabled
1991	 */
1992	value = apic_read(APIC_LVT1);
1993	value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
1994			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
1995			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
1996	value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
1997	value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
1998	apic_write(APIC_LVT1, value);
1999}
2000
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2001int generic_processor_info(int apicid, int version)
2002{
2003	int cpu, max = nr_cpu_ids;
2004	bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
2005				phys_cpu_present_map);
2006
2007	/*
2008	 * boot_cpu_physical_apicid is designed to have the apicid
2009	 * returned by read_apic_id(), i.e, the apicid of the
2010	 * currently booting-up processor. However, on some platforms,
2011	 * it is temporarily modified by the apicid reported as BSP
2012	 * through MP table. Concretely:
2013	 *
2014	 * - arch/x86/kernel/mpparse.c: MP_processor_info()
2015	 * - arch/x86/mm/amdtopology.c: amd_numa_init()
2016	 *
2017	 * This function is executed with the modified
2018	 * boot_cpu_physical_apicid. So, disabled_cpu_apicid kernel
2019	 * parameter doesn't work to disable APs on kdump 2nd kernel.
2020	 *
2021	 * Since fixing handling of boot_cpu_physical_apicid requires
2022	 * another discussion and tests on each platform, we leave it
2023	 * for now and here we use read_apic_id() directly in this
2024	 * function, generic_processor_info().
2025	 */
2026	if (disabled_cpu_apicid != BAD_APICID &&
2027	    disabled_cpu_apicid != read_apic_id() &&
2028	    disabled_cpu_apicid == apicid) {
2029		int thiscpu = num_processors + disabled_cpus;
2030
2031		pr_warning("APIC: Disabling requested cpu."
2032			   " Processor %d/0x%x ignored.\n",
2033			   thiscpu, apicid);
2034
2035		disabled_cpus++;
2036		return -ENODEV;
2037	}
2038
2039	/*
2040	 * If boot cpu has not been detected yet, then only allow upto
2041	 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2042	 */
2043	if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2044	    apicid != boot_cpu_physical_apicid) {
2045		int thiscpu = max + disabled_cpus - 1;
2046
2047		pr_warning(
2048			"ACPI: NR_CPUS/possible_cpus limit of %i almost"
2049			" reached. Keeping one slot for boot cpu."
2050			"  Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2051
2052		disabled_cpus++;
2053		return -ENODEV;
2054	}
2055
2056	if (num_processors >= nr_cpu_ids) {
2057		int thiscpu = max + disabled_cpus;
2058
2059		pr_warning(
2060			"ACPI: NR_CPUS/possible_cpus limit of %i reached."
2061			"  Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2062
2063		disabled_cpus++;
2064		return -EINVAL;
2065	}
2066
2067	num_processors++;
2068	if (apicid == boot_cpu_physical_apicid) {
2069		/*
2070		 * x86_bios_cpu_apicid is required to have processors listed
2071		 * in same order as logical cpu numbers. Hence the first
2072		 * entry is BSP, and so on.
2073		 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2074		 * for BSP.
2075		 */
2076		cpu = 0;
2077	} else
2078		cpu = cpumask_next_zero(-1, cpu_present_mask);
2079
2080	/*
2081	 * This can happen on physical hotplug. The sanity check at boot time
2082	 * is done from native_smp_prepare_cpus() after num_possible_cpus() is
2083	 * established.
2084	 */
2085	if (topology_update_package_map(apicid, cpu) < 0) {
2086		int thiscpu = max + disabled_cpus;
2087
2088		pr_warning("ACPI: Package limit reached. Processor %d/0x%x ignored.\n",
2089			   thiscpu, apicid);
2090		disabled_cpus++;
2091		return -ENOSPC;
2092	}
2093
2094	/*
2095	 * Validate version
2096	 */
2097	if (version == 0x0) {
2098		pr_warning("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2099			   cpu, apicid);
2100		version = 0x10;
2101	}
2102	apic_version[apicid] = version;
2103
2104	if (version != apic_version[boot_cpu_physical_apicid]) {
2105		pr_warning("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2106			apic_version[boot_cpu_physical_apicid], cpu, version);
2107	}
2108
2109	physid_set(apicid, phys_cpu_present_map);
2110	if (apicid > max_physical_apicid)
2111		max_physical_apicid = apicid;
2112
2113#if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2114	early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2115	early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2116#endif
2117#ifdef CONFIG_X86_32
2118	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2119		apic->x86_32_early_logical_apicid(cpu);
2120#endif
2121	set_cpu_possible(cpu, true);
 
2122	set_cpu_present(cpu, true);
 
2123
2124	return cpu;
2125}
2126
2127int hard_smp_processor_id(void)
2128{
2129	return read_apic_id();
2130}
2131
2132void default_init_apic_ldr(void)
2133{
2134	unsigned long val;
2135
2136	apic_write(APIC_DFR, APIC_DFR_VALUE);
2137	val = apic_read(APIC_LDR) & ~APIC_LDR_MASK;
2138	val |= SET_APIC_LOGICAL_ID(1UL << smp_processor_id());
2139	apic_write(APIC_LDR, val);
2140}
2141
2142int default_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
2143				   const struct cpumask *andmask,
2144				   unsigned int *apicid)
2145{
2146	unsigned int cpu;
2147
2148	for_each_cpu_and(cpu, cpumask, andmask) {
2149		if (cpumask_test_cpu(cpu, cpu_online_mask))
2150			break;
2151	}
2152
2153	if (likely(cpu < nr_cpu_ids)) {
2154		*apicid = per_cpu(x86_cpu_to_apicid, cpu);
2155		return 0;
2156	}
2157
2158	return -EINVAL;
2159}
2160
2161/*
2162 * Override the generic EOI implementation with an optimized version.
2163 * Only called during early boot when only one CPU is active and with
2164 * interrupts disabled, so we know this does not race with actual APIC driver
2165 * use.
2166 */
2167void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2168{
2169	struct apic **drv;
2170
2171	for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2172		/* Should happen once for each apic */
2173		WARN_ON((*drv)->eoi_write == eoi_write);
 
2174		(*drv)->eoi_write = eoi_write;
2175	}
2176}
2177
2178static void __init apic_bsp_up_setup(void)
2179{
2180#ifdef CONFIG_X86_64
2181	apic_write(APIC_ID, SET_APIC_ID(boot_cpu_physical_apicid));
2182#else
2183	/*
2184	 * Hack: In case of kdump, after a crash, kernel might be booting
2185	 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
2186	 * might be zero if read from MP tables. Get it from LAPIC.
2187	 */
2188# ifdef CONFIG_CRASH_DUMP
2189	boot_cpu_physical_apicid = read_apic_id();
2190# endif
2191#endif
2192	physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
2193}
2194
2195/**
2196 * apic_bsp_setup - Setup function for local apic and io-apic
2197 * @upmode:		Force UP mode (for APIC_init_uniprocessor)
2198 *
2199 * Returns:
2200 * apic_id of BSP APIC
2201 */
2202int __init apic_bsp_setup(bool upmode)
2203{
2204	int id;
2205
2206	connect_bsp_APIC();
2207	if (upmode)
2208		apic_bsp_up_setup();
2209	setup_local_APIC();
2210
2211	if (x2apic_mode)
2212		id = apic_read(APIC_LDR);
2213	else
2214		id = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
2215
2216	enable_IO_APIC();
2217	end_local_APIC_setup();
2218	irq_remap_enable_fault_handling();
2219	setup_IO_APIC();
2220	/* Setup local timer */
2221	x86_init.timers.setup_percpu_clockev();
2222	return id;
2223}
2224
2225/*
2226 * This initializes the IO-APIC and APIC hardware if this is
2227 * a UP kernel.
2228 */
2229int __init APIC_init_uniprocessor(void)
2230{
2231	if (disable_apic) {
2232		pr_info("Apic disabled\n");
2233		return -1;
2234	}
2235#ifdef CONFIG_X86_64
2236	if (!cpu_has_apic) {
2237		disable_apic = 1;
2238		pr_info("Apic disabled by BIOS\n");
2239		return -1;
2240	}
2241#else
2242	if (!smp_found_config && !cpu_has_apic)
2243		return -1;
2244
2245	/*
2246	 * Complain if the BIOS pretends there is one.
2247	 */
2248	if (!cpu_has_apic &&
2249	    APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
2250		pr_err("BIOS bug, local APIC 0x%x not detected!...\n",
2251			boot_cpu_physical_apicid);
2252		return -1;
2253	}
2254#endif
2255
2256	if (!smp_found_config)
2257		disable_ioapic_support();
2258
2259	default_setup_apic_routing();
2260	apic_bsp_setup(true);
2261	return 0;
2262}
2263
2264#ifdef CONFIG_UP_LATE_INIT
2265void __init up_late_init(void)
2266{
2267	APIC_init_uniprocessor();
 
 
 
 
2268}
2269#endif
2270
2271/*
2272 * Power management
2273 */
2274#ifdef CONFIG_PM
2275
2276static struct {
2277	/*
2278	 * 'active' is true if the local APIC was enabled by us and
2279	 * not the BIOS; this signifies that we are also responsible
2280	 * for disabling it before entering apm/acpi suspend
2281	 */
2282	int active;
2283	/* r/w apic fields */
2284	unsigned int apic_id;
2285	unsigned int apic_taskpri;
2286	unsigned int apic_ldr;
2287	unsigned int apic_dfr;
2288	unsigned int apic_spiv;
2289	unsigned int apic_lvtt;
2290	unsigned int apic_lvtpc;
2291	unsigned int apic_lvt0;
2292	unsigned int apic_lvt1;
2293	unsigned int apic_lvterr;
2294	unsigned int apic_tmict;
2295	unsigned int apic_tdcr;
2296	unsigned int apic_thmr;
2297	unsigned int apic_cmci;
2298} apic_pm_state;
2299
2300static int lapic_suspend(void)
2301{
2302	unsigned long flags;
2303	int maxlvt;
2304
2305	if (!apic_pm_state.active)
2306		return 0;
2307
2308	maxlvt = lapic_get_maxlvt();
2309
2310	apic_pm_state.apic_id = apic_read(APIC_ID);
2311	apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2312	apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2313	apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2314	apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2315	apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2316	if (maxlvt >= 4)
2317		apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2318	apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2319	apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2320	apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2321	apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2322	apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2323#ifdef CONFIG_X86_THERMAL_VECTOR
2324	if (maxlvt >= 5)
2325		apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2326#endif
2327#ifdef CONFIG_X86_MCE_INTEL
2328	if (maxlvt >= 6)
2329		apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI);
2330#endif
2331
2332	local_irq_save(flags);
2333	disable_local_APIC();
2334
2335	irq_remapping_disable();
2336
2337	local_irq_restore(flags);
2338	return 0;
2339}
2340
2341static void lapic_resume(void)
2342{
2343	unsigned int l, h;
2344	unsigned long flags;
2345	int maxlvt;
2346
2347	if (!apic_pm_state.active)
2348		return;
2349
2350	local_irq_save(flags);
2351
2352	/*
2353	 * IO-APIC and PIC have their own resume routines.
2354	 * We just mask them here to make sure the interrupt
2355	 * subsystem is completely quiet while we enable x2apic
2356	 * and interrupt-remapping.
2357	 */
2358	mask_ioapic_entries();
2359	legacy_pic->mask_all();
2360
2361	if (x2apic_mode) {
2362		__x2apic_enable();
2363	} else {
2364		/*
2365		 * Make sure the APICBASE points to the right address
2366		 *
2367		 * FIXME! This will be wrong if we ever support suspend on
2368		 * SMP! We'll need to do this as part of the CPU restore!
2369		 */
2370		if (boot_cpu_data.x86 >= 6) {
2371			rdmsr(MSR_IA32_APICBASE, l, h);
2372			l &= ~MSR_IA32_APICBASE_BASE;
2373			l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2374			wrmsr(MSR_IA32_APICBASE, l, h);
2375		}
2376	}
2377
2378	maxlvt = lapic_get_maxlvt();
2379	apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2380	apic_write(APIC_ID, apic_pm_state.apic_id);
2381	apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2382	apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2383	apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2384	apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2385	apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2386	apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2387#ifdef CONFIG_X86_THERMAL_VECTOR
2388	if (maxlvt >= 5)
2389		apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2390#endif
2391#ifdef CONFIG_X86_MCE_INTEL
2392	if (maxlvt >= 6)
2393		apic_write(APIC_LVTCMCI, apic_pm_state.apic_cmci);
2394#endif
2395	if (maxlvt >= 4)
2396		apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2397	apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2398	apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2399	apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2400	apic_write(APIC_ESR, 0);
2401	apic_read(APIC_ESR);
2402	apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2403	apic_write(APIC_ESR, 0);
2404	apic_read(APIC_ESR);
2405
2406	irq_remapping_reenable(x2apic_mode);
2407
2408	local_irq_restore(flags);
2409}
2410
2411/*
2412 * This device has no shutdown method - fully functioning local APICs
2413 * are needed on every CPU up until machine_halt/restart/poweroff.
2414 */
2415
2416static struct syscore_ops lapic_syscore_ops = {
2417	.resume		= lapic_resume,
2418	.suspend	= lapic_suspend,
2419};
2420
2421static void apic_pm_activate(void)
2422{
2423	apic_pm_state.active = 1;
2424}
2425
2426static int __init init_lapic_sysfs(void)
2427{
2428	/* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2429	if (cpu_has_apic)
2430		register_syscore_ops(&lapic_syscore_ops);
2431
2432	return 0;
2433}
2434
2435/* local apic needs to resume before other devices access its registers. */
2436core_initcall(init_lapic_sysfs);
2437
2438#else	/* CONFIG_PM */
2439
2440static void apic_pm_activate(void) { }
2441
2442#endif	/* CONFIG_PM */
2443
2444#ifdef CONFIG_X86_64
2445
2446static int multi_checked;
2447static int multi;
2448
2449static int set_multi(const struct dmi_system_id *d)
2450{
2451	if (multi)
2452		return 0;
2453	pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2454	multi = 1;
2455	return 0;
2456}
2457
2458static const struct dmi_system_id multi_dmi_table[] = {
2459	{
2460		.callback = set_multi,
2461		.ident = "IBM System Summit2",
2462		.matches = {
2463			DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2464			DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2465		},
2466	},
2467	{}
2468};
2469
2470static void dmi_check_multi(void)
2471{
2472	if (multi_checked)
2473		return;
2474
2475	dmi_check_system(multi_dmi_table);
2476	multi_checked = 1;
2477}
2478
2479/*
2480 * apic_is_clustered_box() -- Check if we can expect good TSC
2481 *
2482 * Thus far, the major user of this is IBM's Summit2 series:
2483 * Clustered boxes may have unsynced TSC problems if they are
2484 * multi-chassis.
2485 * Use DMI to check them
2486 */
2487int apic_is_clustered_box(void)
2488{
2489	dmi_check_multi();
2490	return multi;
2491}
2492#endif
2493
2494/*
2495 * APIC command line parameters
2496 */
2497static int __init setup_disableapic(char *arg)
2498{
2499	disable_apic = 1;
2500	setup_clear_cpu_cap(X86_FEATURE_APIC);
2501	return 0;
2502}
2503early_param("disableapic", setup_disableapic);
2504
2505/* same as disableapic, for compatibility */
2506static int __init setup_nolapic(char *arg)
2507{
2508	return setup_disableapic(arg);
2509}
2510early_param("nolapic", setup_nolapic);
2511
2512static int __init parse_lapic_timer_c2_ok(char *arg)
2513{
2514	local_apic_timer_c2_ok = 1;
2515	return 0;
2516}
2517early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2518
2519static int __init parse_disable_apic_timer(char *arg)
2520{
2521	disable_apic_timer = 1;
2522	return 0;
2523}
2524early_param("noapictimer", parse_disable_apic_timer);
2525
2526static int __init parse_nolapic_timer(char *arg)
2527{
2528	disable_apic_timer = 1;
2529	return 0;
2530}
2531early_param("nolapic_timer", parse_nolapic_timer);
2532
2533static int __init apic_set_verbosity(char *arg)
2534{
2535	if (!arg)  {
2536#ifdef CONFIG_X86_64
2537		skip_ioapic_setup = 0;
2538		return 0;
2539#endif
2540		return -EINVAL;
2541	}
2542
2543	if (strcmp("debug", arg) == 0)
2544		apic_verbosity = APIC_DEBUG;
2545	else if (strcmp("verbose", arg) == 0)
2546		apic_verbosity = APIC_VERBOSE;
 
2547	else {
2548		pr_warning("APIC Verbosity level %s not recognised"
2549			" use apic=verbose or apic=debug\n", arg);
2550		return -EINVAL;
2551	}
 
2552
2553	return 0;
2554}
2555early_param("apic", apic_set_verbosity);
2556
2557static int __init lapic_insert_resource(void)
2558{
2559	if (!apic_phys)
2560		return -1;
2561
2562	/* Put local APIC into the resource map. */
2563	lapic_resource.start = apic_phys;
2564	lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2565	insert_resource(&iomem_resource, &lapic_resource);
2566
2567	return 0;
2568}
2569
2570/*
2571 * need call insert after e820_reserve_resources()
2572 * that is using request_resource
2573 */
2574late_initcall(lapic_insert_resource);
2575
2576static int __init apic_set_disabled_cpu_apicid(char *arg)
2577{
2578	if (!arg || !get_option(&arg, &disabled_cpu_apicid))
2579		return -EINVAL;
2580
2581	return 0;
2582}
2583early_param("disable_cpu_apicid", apic_set_disabled_cpu_apicid);
2584
2585static int __init apic_set_extnmi(char *arg)
2586{
2587	if (!arg)
2588		return -EINVAL;
2589
2590	if (!strncmp("all", arg, 3))
2591		apic_extnmi = APIC_EXTNMI_ALL;
2592	else if (!strncmp("none", arg, 4))
2593		apic_extnmi = APIC_EXTNMI_NONE;
2594	else if (!strncmp("bsp", arg, 3))
2595		apic_extnmi = APIC_EXTNMI_BSP;
2596	else {
2597		pr_warn("Unknown external NMI delivery mode `%s' ignored\n", arg);
2598		return -EINVAL;
2599	}
2600
2601	return 0;
2602}
2603early_param("apic_extnmi", apic_set_extnmi);