Linux Audio

Check our new training course

Loading...
v6.13.7
   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/bitmap.h>
  23#include <linux/clockchips.h>
  24#include <linux/interrupt.h>
  25#include <linux/memblock.h>
  26#include <linux/ftrace.h>
  27#include <linux/ioport.h>
  28#include <linux/export.h>
  29#include <linux/syscore_ops.h>
  30#include <linux/delay.h>
  31#include <linux/timex.h>
  32#include <linux/i8253.h>
  33#include <linux/dmar.h>
  34#include <linux/init.h>
  35#include <linux/cpu.h>
  36#include <linux/dmi.h>
  37#include <linux/smp.h>
  38#include <linux/mm.h>
  39
  40#include <xen/xen.h>
  41
  42#include <asm/trace/irq_vectors.h>
  43#include <asm/irq_remapping.h>
  44#include <asm/pc-conf-reg.h>
  45#include <asm/perf_event.h>
  46#include <asm/x86_init.h>
 
  47#include <linux/atomic.h>
  48#include <asm/barrier.h>
  49#include <asm/mpspec.h>
  50#include <asm/i8259.h>
  51#include <asm/proto.h>
  52#include <asm/traps.h>
  53#include <asm/apic.h>
  54#include <asm/acpi.h>
  55#include <asm/io_apic.h>
  56#include <asm/desc.h>
  57#include <asm/hpet.h>
  58#include <asm/mtrr.h>
  59#include <asm/time.h>
  60#include <asm/smp.h>
  61#include <asm/mce.h>
  62#include <asm/tsc.h>
  63#include <asm/hypervisor.h>
  64#include <asm/cpu_device_id.h>
  65#include <asm/intel-family.h>
  66#include <asm/irq_regs.h>
  67#include <asm/cpu.h>
  68
  69#include "local.h"
 
 
  70
  71/* Processor that is doing the boot up */
  72u32 boot_cpu_physical_apicid __ro_after_init = BAD_APICID;
  73EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid);
  74
  75u8 boot_cpu_apic_version __ro_after_init;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  76
  77/*
  78 * This variable controls which CPUs receive external NMIs.  By default,
  79 * external NMIs are delivered only to the BSP.
  80 */
  81static int apic_extnmi __ro_after_init = APIC_EXTNMI_BSP;
  82
  83/*
  84 * Hypervisor supports 15 bits of APIC ID in MSI Extended Destination ID
  85 */
  86static bool virt_ext_dest_id __ro_after_init;
 
 
 
 
 
  87
  88/* For parallel bootup. */
  89unsigned long apic_mmio_base __ro_after_init;
  90
  91static inline bool apic_accessible(void)
  92{
  93	return x2apic_mode || apic_mmio_base;
  94}
 
 
 
  95
  96#ifdef CONFIG_X86_32
  97/* Local APIC was disabled by the BIOS and enabled by the kernel */
  98static int enabled_via_apicbase __ro_after_init;
  99
 100/*
 101 * Handle interrupt mode configuration register (IMCR).
 102 * This register controls whether the interrupt signals
 103 * that reach the BSP come from the master PIC or from the
 104 * local APIC. Before entering Symmetric I/O Mode, either
 105 * the BIOS or the operating system must switch out of
 106 * PIC Mode by changing the IMCR.
 107 */
 108static inline void imcr_pic_to_apic(void)
 109{
 
 
 110	/* NMI and 8259 INTR go through APIC */
 111	pc_conf_set(PC_CONF_MPS_IMCR, 0x01);
 112}
 113
 114static inline void imcr_apic_to_pic(void)
 115{
 
 
 116	/* NMI and 8259 INTR go directly to BSP */
 117	pc_conf_set(PC_CONF_MPS_IMCR, 0x00);
 118}
 119#endif
 120
 121/*
 122 * Knob to control our willingness to enable the local APIC.
 123 *
 124 * +1=force-enable
 125 */
 126static int force_enable_local_apic __initdata;
 127
 128/*
 129 * APIC command line parameters
 130 */
 131static int __init parse_lapic(char *arg)
 132{
 133	if (IS_ENABLED(CONFIG_X86_32) && !arg)
 134		force_enable_local_apic = 1;
 135	else if (arg && !strncmp(arg, "notscdeadline", 13))
 136		setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
 137	return 0;
 138}
 139early_param("lapic", parse_lapic);
 140
 141#ifdef CONFIG_X86_64
 142static int apic_calibrate_pmtmr __initdata;
 143static __init int setup_apicpmtimer(char *s)
 144{
 145	apic_calibrate_pmtmr = 1;
 146	notsc_setup(NULL);
 147	return 1;
 148}
 149__setup("apicpmtimer", setup_apicpmtimer);
 150#endif
 151
 152static unsigned long mp_lapic_addr __ro_after_init;
 153bool apic_is_disabled __ro_after_init;
 154/* Disable local APIC timer from the kernel commandline or via dmi quirk */
 155static int disable_apic_timer __initdata;
 156/* Local APIC timer works in C2 */
 157int local_apic_timer_c2_ok __ro_after_init;
 158EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
 159
 160/*
 161 * Debug level, exported for io_apic.c
 162 */
 163int apic_verbosity __ro_after_init;
 164
 165int pic_mode __ro_after_init;
 166
 167/* Have we found an MP table */
 168int smp_found_config __ro_after_init;
 169
 170static struct resource lapic_resource = {
 171	.name = "Local APIC",
 172	.flags = IORESOURCE_MEM | IORESOURCE_BUSY,
 173};
 174
 175unsigned int lapic_timer_period = 0;
 176
 177static void apic_pm_activate(void);
 178
 
 
 179/*
 180 * Get the LAPIC version
 181 */
 182static inline int lapic_get_version(void)
 183{
 184	return GET_APIC_VERSION(apic_read(APIC_LVR));
 185}
 186
 187/*
 188 * Check, if the APIC is integrated or a separate chip
 189 */
 190static inline int lapic_is_integrated(void)
 191{
 192	return APIC_INTEGRATED(lapic_get_version());
 193}
 194
 195/*
 196 * Check, whether this is a modern or a first generation APIC
 197 */
 198static int modern_apic(void)
 199{
 200	/* AMD systems use old APIC versions, so check the CPU */
 201	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
 202	    boot_cpu_data.x86 >= 0xf)
 203		return 1;
 204
 205	/* Hygon systems use modern APIC */
 206	if (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)
 207		return 1;
 208
 209	return lapic_get_version() >= 0x14;
 210}
 211
 212/*
 213 * right after this call apic become NOOP driven
 214 * so apic->write/read doesn't do anything
 215 */
 216static void __init apic_disable(void)
 217{
 218	apic_install_driver(&apic_noop);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 219}
 220
 221void native_apic_icr_write(u32 low, u32 id)
 222{
 223	unsigned long flags;
 224
 225	local_irq_save(flags);
 226	apic_write(APIC_ICR2, SET_XAPIC_DEST_FIELD(id));
 227	apic_write(APIC_ICR, low);
 228	local_irq_restore(flags);
 229}
 230
 231u64 native_apic_icr_read(void)
 232{
 233	u32 icr1, icr2;
 234
 235	icr2 = apic_read(APIC_ICR2);
 236	icr1 = apic_read(APIC_ICR);
 237
 238	return icr1 | ((u64)icr2 << 32);
 239}
 240
 
 
 
 
 
 
 
 
 
 
 241/**
 242 * lapic_get_maxlvt - get the maximum number of local vector table entries
 243 */
 244int lapic_get_maxlvt(void)
 245{
 246	/*
 247	 * - we always have APIC integrated on 64bit mode
 248	 * - 82489DXs do not report # of LVT entries
 249	 */
 250	return lapic_is_integrated() ? GET_APIC_MAXLVT(apic_read(APIC_LVR)) : 2;
 251}
 252
 253/*
 254 * Local APIC timer
 255 */
 256
 257/* Clock divisor */
 258#define APIC_DIVISOR 16
 259#define TSC_DIVISOR  8
 260
 261/* i82489DX specific */
 262#define		I82489DX_BASE_DIVIDER		(((0x2) << 18))
 263
 264/*
 265 * This function sets up the local APIC timer, with a timeout of
 266 * 'clocks' APIC bus clock. During calibration we actually call
 267 * this function twice on the boot CPU, once with a bogus timeout
 268 * value, second time for real. The other (noncalibrating) CPUs
 269 * call this function only once, with the real, calibrated value.
 270 *
 271 * We do reads before writes even if unnecessary, to get around the
 272 * P5 APIC double write bug.
 273 */
 274static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
 275{
 276	unsigned int lvtt_value, tmp_value;
 277
 278	lvtt_value = LOCAL_TIMER_VECTOR;
 279	if (!oneshot)
 280		lvtt_value |= APIC_LVT_TIMER_PERIODIC;
 281	else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
 282		lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
 283
 284	/*
 285	 * The i82489DX APIC uses bit 18 and 19 for the base divider.  This
 286	 * overlaps with bit 18 on integrated APICs, but is not documented
 287	 * in the SDM. No problem though. i82489DX equipped systems do not
 288	 * have TSC deadline timer.
 289	 */
 290	if (!lapic_is_integrated())
 291		lvtt_value |= I82489DX_BASE_DIVIDER;
 292
 293	if (!irqen)
 294		lvtt_value |= APIC_LVT_MASKED;
 295
 296	apic_write(APIC_LVTT, lvtt_value);
 297
 298	if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
 299		/*
 300		 * See Intel SDM: TSC-Deadline Mode chapter. In xAPIC mode,
 301		 * writing to the APIC LVTT and TSC_DEADLINE MSR isn't serialized.
 302		 * According to Intel, MFENCE can do the serialization here.
 303		 */
 304		asm volatile("mfence" : : : "memory");
 
 
 305		return;
 306	}
 307
 308	/*
 309	 * Divide PICLK by 16
 310	 */
 311	tmp_value = apic_read(APIC_TDCR);
 312	apic_write(APIC_TDCR,
 313		(tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
 314		APIC_TDR_DIV_16);
 315
 316	if (!oneshot)
 317		apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
 318}
 319
 320/*
 321 * Setup extended LVT, AMD specific
 322 *
 323 * Software should use the LVT offsets the BIOS provides.  The offsets
 324 * are determined by the subsystems using it like those for MCE
 325 * threshold or IBS.  On K8 only offset 0 (APIC500) and MCE interrupts
 326 * are supported. Beginning with family 10h at least 4 offsets are
 327 * available.
 328 *
 329 * Since the offsets must be consistent for all cores, we keep track
 330 * of the LVT offsets in software and reserve the offset for the same
 331 * vector also to be used on other cores. An offset is freed by
 332 * setting the entry to APIC_EILVT_MASKED.
 333 *
 334 * If the BIOS is right, there should be no conflicts. Otherwise a
 335 * "[Firmware Bug]: ..." error message is generated. However, if
 336 * software does not properly determines the offsets, it is not
 337 * necessarily a BIOS bug.
 338 */
 339
 340static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
 341
 342static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
 343{
 344	return (old & APIC_EILVT_MASKED)
 345		|| (new == APIC_EILVT_MASKED)
 346		|| ((new & ~APIC_EILVT_MASKED) == old);
 347}
 348
 349static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
 350{
 351	unsigned int rsvd, vector;
 352
 353	if (offset >= APIC_EILVT_NR_MAX)
 354		return ~0;
 355
 356	rsvd = atomic_read(&eilvt_offsets[offset]);
 357	do {
 358		vector = rsvd & ~APIC_EILVT_MASKED;	/* 0: unassigned */
 359		if (vector && !eilvt_entry_is_changeable(vector, new))
 360			/* may not change if vectors are different */
 361			return rsvd;
 362	} while (!atomic_try_cmpxchg(&eilvt_offsets[offset], &rsvd, new));
 
 363
 364	rsvd = new & ~APIC_EILVT_MASKED;
 365	if (rsvd && rsvd != vector)
 366		pr_info("LVT offset %d assigned for vector 0x%02x\n",
 367			offset, rsvd);
 368
 369	return new;
 370}
 371
 372/*
 373 * If mask=1, the LVT entry does not generate interrupts while mask=0
 374 * enables the vector. See also the BKDGs. Must be called with
 375 * preemption disabled.
 376 */
 377
 378int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
 379{
 380	unsigned long reg = APIC_EILVTn(offset);
 381	unsigned int new, old, reserved;
 382
 383	new = (mask << 16) | (msg_type << 8) | vector;
 384	old = apic_read(reg);
 385	reserved = reserve_eilvt_offset(offset, new);
 386
 387	if (reserved != new) {
 388		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
 389		       "vector 0x%x, but the register is already in use for "
 390		       "vector 0x%x on another cpu\n",
 391		       smp_processor_id(), reg, offset, new, reserved);
 392		return -EINVAL;
 393	}
 394
 395	if (!eilvt_entry_is_changeable(old, new)) {
 396		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
 397		       "vector 0x%x, but the register is already in use for "
 398		       "vector 0x%x on this cpu\n",
 399		       smp_processor_id(), reg, offset, new, old);
 400		return -EBUSY;
 401	}
 402
 403	apic_write(reg, new);
 404
 405	return 0;
 406}
 407EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
 408
 409/*
 410 * Program the next event, relative to now
 411 */
 412static int lapic_next_event(unsigned long delta,
 413			    struct clock_event_device *evt)
 414{
 415	apic_write(APIC_TMICT, delta);
 416	return 0;
 417}
 418
 419static int lapic_next_deadline(unsigned long delta,
 420			       struct clock_event_device *evt)
 421{
 422	u64 tsc;
 423
 424	/* This MSR is special and need a special fence: */
 425	weak_wrmsr_fence();
 426
 427	tsc = rdtsc();
 428	wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
 429	return 0;
 430}
 431
 432static int lapic_timer_shutdown(struct clock_event_device *evt)
 433{
 434	unsigned int v;
 435
 436	/* Lapic used as dummy for broadcast ? */
 437	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
 438		return 0;
 439
 440	v = apic_read(APIC_LVTT);
 441	v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
 442	apic_write(APIC_LVTT, v);
 443
 444	/*
 445	 * Setting APIC_LVT_MASKED (above) should be enough to tell
 446	 * the hardware that this timer will never fire. But AMD
 447	 * erratum 411 and some Intel CPU behavior circa 2024 say
 448	 * otherwise.  Time for belt and suspenders programming: mask
 449	 * the timer _and_ zero the counter registers:
 450	 */
 451	if (v & APIC_LVT_TIMER_TSCDEADLINE)
 452		wrmsrl(MSR_IA32_TSC_DEADLINE, 0);
 453	else
 454		apic_write(APIC_TMICT, 0);
 455
 456	return 0;
 457}
 458
 459static inline int
 460lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot)
 461{
 462	/* Lapic used as dummy for broadcast ? */
 463	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
 464		return 0;
 465
 466	__setup_APIC_LVTT(lapic_timer_period, oneshot, 1);
 467	return 0;
 468}
 469
 470static int lapic_timer_set_periodic(struct clock_event_device *evt)
 471{
 472	return lapic_timer_set_periodic_oneshot(evt, false);
 473}
 474
 475static int lapic_timer_set_oneshot(struct clock_event_device *evt)
 476{
 477	return lapic_timer_set_periodic_oneshot(evt, true);
 478}
 479
 480/*
 481 * Local APIC timer broadcast function
 482 */
 483static void lapic_timer_broadcast(const struct cpumask *mask)
 484{
 485#ifdef CONFIG_SMP
 486	__apic_send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
 487#endif
 488}
 489
 490
 491/*
 492 * The local apic timer can be used for any function which is CPU local.
 493 */
 494static struct clock_event_device lapic_clockevent = {
 495	.name				= "lapic",
 496	.features			= CLOCK_EVT_FEAT_PERIODIC |
 497					  CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP
 498					  | CLOCK_EVT_FEAT_DUMMY,
 499	.shift				= 32,
 500	.set_state_shutdown		= lapic_timer_shutdown,
 501	.set_state_periodic		= lapic_timer_set_periodic,
 502	.set_state_oneshot		= lapic_timer_set_oneshot,
 503	.set_state_oneshot_stopped	= lapic_timer_shutdown,
 504	.set_next_event			= lapic_next_event,
 505	.broadcast			= lapic_timer_broadcast,
 506	.rating				= 100,
 507	.irq				= -1,
 508};
 509static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
 510
 511static const struct x86_cpu_id deadline_match[] __initconst = {
 512	X86_MATCH_VFM_STEPPINGS(INTEL_HASWELL_X, X86_STEPPINGS(0x2, 0x2), 0x3a), /* EP */
 513	X86_MATCH_VFM_STEPPINGS(INTEL_HASWELL_X, X86_STEPPINGS(0x4, 0x4), 0x0f), /* EX */
 
 
 
 
 
 
 
 
 
 514
 515	X86_MATCH_VFM(INTEL_BROADWELL_X,	0x0b000020),
 
 516
 517	X86_MATCH_VFM_STEPPINGS(INTEL_BROADWELL_D, X86_STEPPINGS(0x2, 0x2), 0x00000011),
 518	X86_MATCH_VFM_STEPPINGS(INTEL_BROADWELL_D, X86_STEPPINGS(0x3, 0x3), 0x0700000e),
 519	X86_MATCH_VFM_STEPPINGS(INTEL_BROADWELL_D, X86_STEPPINGS(0x4, 0x4), 0x0f00000c),
 520	X86_MATCH_VFM_STEPPINGS(INTEL_BROADWELL_D, X86_STEPPINGS(0x5, 0x5), 0x0e000003),
 
 
 
 
 521
 522	X86_MATCH_VFM_STEPPINGS(INTEL_SKYLAKE_X, X86_STEPPINGS(0x3, 0x3), 0x01000136),
 523	X86_MATCH_VFM_STEPPINGS(INTEL_SKYLAKE_X, X86_STEPPINGS(0x4, 0x4), 0x02000014),
 524	X86_MATCH_VFM_STEPPINGS(INTEL_SKYLAKE_X, X86_STEPPINGS(0x5, 0xf), 0),
 525
 526	X86_MATCH_VFM(INTEL_HASWELL,		0x22),
 527	X86_MATCH_VFM(INTEL_HASWELL_L,		0x20),
 528	X86_MATCH_VFM(INTEL_HASWELL_G,		0x17),
 
 
 
 
 
 
 529
 530	X86_MATCH_VFM(INTEL_BROADWELL,		0x25),
 531	X86_MATCH_VFM(INTEL_BROADWELL_G,	0x17),
 
 
 
 532
 533	X86_MATCH_VFM(INTEL_SKYLAKE_L,		0xb2),
 534	X86_MATCH_VFM(INTEL_SKYLAKE,		0xb2),
 
 535
 536	X86_MATCH_VFM(INTEL_KABYLAKE_L,		0x52),
 537	X86_MATCH_VFM(INTEL_KABYLAKE,		0x52),
 
 
 
 
 
 
 538
 539	{},
 540};
 541
 542static __init bool apic_validate_deadline_timer(void)
 543{
 544	const struct x86_cpu_id *m;
 545	u32 rev;
 546
 547	if (!boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
 548		return false;
 549	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
 550		return true;
 551
 552	m = x86_match_cpu(deadline_match);
 553	if (!m)
 554		return true;
 555
 556	rev = (u32)m->driver_data;
 
 
 
 
 
 
 
 557
 558	if (boot_cpu_data.microcode >= rev)
 559		return true;
 560
 561	setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
 562	pr_err(FW_BUG "TSC_DEADLINE disabled due to Errata; "
 563	       "please update microcode to version: 0x%x (or later)\n", rev);
 564	return false;
 565}
 566
 567/*
 568 * Setup the local APIC timer for this CPU. Copy the initialized values
 569 * of the boot CPU and register the clock event in the framework.
 570 */
 571static void setup_APIC_timer(void)
 572{
 573	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
 574
 575	if (this_cpu_has(X86_FEATURE_ARAT)) {
 576		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
 577		/* Make LAPIC timer preferable over percpu HPET */
 578		lapic_clockevent.rating = 150;
 579	}
 580
 581	memcpy(levt, &lapic_clockevent, sizeof(*levt));
 582	levt->cpumask = cpumask_of(smp_processor_id());
 583
 584	if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
 585		levt->name = "lapic-deadline";
 586		levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
 587				    CLOCK_EVT_FEAT_DUMMY);
 588		levt->set_next_event = lapic_next_deadline;
 589		clockevents_config_and_register(levt,
 590						tsc_khz * (1000 / TSC_DIVISOR),
 591						0xF, ~0UL);
 592	} else
 593		clockevents_register_device(levt);
 594}
 595
 596/*
 597 * Install the updated TSC frequency from recalibration at the TSC
 598 * deadline clockevent devices.
 599 */
 600static void __lapic_update_tsc_freq(void *info)
 601{
 602	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
 603
 604	if (!this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
 605		return;
 606
 607	clockevents_update_freq(levt, tsc_khz * (1000 / TSC_DIVISOR));
 608}
 609
 610void lapic_update_tsc_freq(void)
 611{
 612	/*
 613	 * The clockevent device's ->mult and ->shift can both be
 614	 * changed. In order to avoid races, schedule the frequency
 615	 * update code on each CPU.
 616	 */
 617	on_each_cpu(__lapic_update_tsc_freq, NULL, 0);
 618}
 619
 620/*
 621 * In this functions we calibrate APIC bus clocks to the external timer.
 622 *
 623 * We want to do the calibration only once since we want to have local timer
 624 * irqs synchronous. CPUs connected by the same APIC bus have the very same bus
 625 * frequency.
 626 *
 627 * This was previously done by reading the PIT/HPET and waiting for a wrap
 628 * around to find out, that a tick has elapsed. I have a box, where the PIT
 629 * readout is broken, so it never gets out of the wait loop again. This was
 630 * also reported by others.
 631 *
 632 * Monitoring the jiffies value is inaccurate and the clockevents
 633 * infrastructure allows us to do a simple substitution of the interrupt
 634 * handler.
 635 *
 636 * The calibration routine also uses the pm_timer when possible, as the PIT
 637 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
 638 * back to normal later in the boot process).
 639 */
 640
 641#define LAPIC_CAL_LOOPS		(HZ/10)
 642
 643static __initdata int lapic_cal_loops = -1;
 644static __initdata long lapic_cal_t1, lapic_cal_t2;
 645static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
 646static __initdata u32 lapic_cal_pm1, lapic_cal_pm2;
 647static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
 648
 649/*
 650 * Temporary interrupt handler and polled calibration function.
 651 */
 652static void __init lapic_cal_handler(struct clock_event_device *dev)
 653{
 654	unsigned long long tsc = 0;
 655	long tapic = apic_read(APIC_TMCCT);
 656	u32 pm = acpi_pm_read_early();
 657
 658	if (boot_cpu_has(X86_FEATURE_TSC))
 659		tsc = rdtsc();
 660
 661	switch (lapic_cal_loops++) {
 662	case 0:
 663		lapic_cal_t1 = tapic;
 664		lapic_cal_tsc1 = tsc;
 665		lapic_cal_pm1 = pm;
 666		lapic_cal_j1 = jiffies;
 667		break;
 668
 669	case LAPIC_CAL_LOOPS:
 670		lapic_cal_t2 = tapic;
 671		lapic_cal_tsc2 = tsc;
 672		if (pm < lapic_cal_pm1)
 673			pm += ACPI_PM_OVRRUN;
 674		lapic_cal_pm2 = pm;
 675		lapic_cal_j2 = jiffies;
 676		break;
 677	}
 678}
 679
 680static int __init
 681calibrate_by_pmtimer(u32 deltapm, long *delta, long *deltatsc)
 682{
 683	const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
 684	const long pm_thresh = pm_100ms / 100;
 685	unsigned long mult;
 686	u64 res;
 687
 688#ifndef CONFIG_X86_PM_TIMER
 689	return -1;
 690#endif
 691
 692	apic_pr_verbose("... PM-Timer delta = %u\n", deltapm);
 693
 694	/* Check, if the PM timer is available */
 695	if (!deltapm)
 696		return -1;
 697
 698	mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
 699
 700	if (deltapm > (pm_100ms - pm_thresh) &&
 701	    deltapm < (pm_100ms + pm_thresh)) {
 702		apic_pr_verbose("... PM-Timer result ok\n");
 703		return 0;
 704	}
 705
 706	res = (((u64)deltapm) *  mult) >> 22;
 707	do_div(res, 1000000);
 708	pr_warn("APIC calibration not consistent with PM-Timer: %ldms instead of 100ms\n",
 709		(long)res);
 710
 711	/* Correct the lapic counter value */
 712	res = (((u64)(*delta)) * pm_100ms);
 713	do_div(res, deltapm);
 714	pr_info("APIC delta adjusted to PM-Timer: "
 715		"%lu (%ld)\n", (unsigned long)res, *delta);
 716	*delta = (long)res;
 717
 718	/* Correct the tsc counter value */
 719	if (boot_cpu_has(X86_FEATURE_TSC)) {
 720		res = (((u64)(*deltatsc)) * pm_100ms);
 721		do_div(res, deltapm);
 722		apic_pr_verbose("TSC delta adjusted to PM-Timer: %lu (%ld)\n",
 723				(unsigned long)res, *deltatsc);
 
 724		*deltatsc = (long)res;
 725	}
 726
 727	return 0;
 728}
 729
 730static int __init lapic_init_clockevent(void)
 731{
 732	if (!lapic_timer_period)
 733		return -1;
 734
 735	/* Calculate the scaled math multiplication factor */
 736	lapic_clockevent.mult = div_sc(lapic_timer_period/APIC_DIVISOR,
 737					TICK_NSEC, lapic_clockevent.shift);
 738	lapic_clockevent.max_delta_ns =
 739		clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
 740	lapic_clockevent.max_delta_ticks = 0x7FFFFFFF;
 741	lapic_clockevent.min_delta_ns =
 742		clockevent_delta2ns(0xF, &lapic_clockevent);
 743	lapic_clockevent.min_delta_ticks = 0xF;
 744
 745	return 0;
 746}
 747
 748bool __init apic_needs_pit(void)
 749{
 750	/*
 751	 * If the frequencies are not known, PIT is required for both TSC
 752	 * and apic timer calibration.
 753	 */
 754	if (!tsc_khz || !cpu_khz)
 755		return true;
 756
 757	/* Is there an APIC at all or is it disabled? */
 758	if (!boot_cpu_has(X86_FEATURE_APIC) || apic_is_disabled)
 759		return true;
 760
 761	/*
 762	 * If interrupt delivery mode is legacy PIC or virtual wire without
 763	 * configuration, the local APIC timer won't be set up. Make sure
 764	 * that the PIT is initialized.
 765	 */
 766	if (apic_intr_mode == APIC_PIC ||
 767	    apic_intr_mode == APIC_VIRTUAL_WIRE_NO_CONFIG)
 768		return true;
 769
 770	/* Virt guests may lack ARAT, but still have DEADLINE */
 771	if (!boot_cpu_has(X86_FEATURE_ARAT))
 772		return true;
 773
 774	/* Deadline timer is based on TSC so no further PIT action required */
 775	if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
 776		return false;
 777
 778	/* APIC timer disabled? */
 779	if (disable_apic_timer)
 780		return true;
 781	/*
 782	 * The APIC timer frequency is known already, no PIT calibration
 783	 * required. If unknown, let the PIT be initialized.
 784	 */
 785	return lapic_timer_period == 0;
 786}
 787
 788static int __init calibrate_APIC_clock(void)
 789{
 790	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
 791	u64 tsc_perj = 0, tsc_start = 0;
 792	unsigned long jif_start;
 793	unsigned long deltaj;
 794	long delta, deltatsc;
 795	int pm_referenced = 0;
 796
 797	if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
 798		return 0;
 799
 800	/*
 801	 * Check if lapic timer has already been calibrated by platform
 802	 * specific routine, such as tsc calibration code. If so just fill
 803	 * in the clockevent structure and return.
 804	 */
 805	if (!lapic_init_clockevent()) {
 806		apic_pr_verbose("lapic timer already calibrated %d\n", lapic_timer_period);
 807		/*
 808		 * Direct calibration methods must have an always running
 809		 * local APIC timer, no need for broadcast timer.
 810		 */
 
 
 
 
 
 
 
 
 811		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
 812		return 0;
 813	}
 814
 815	apic_pr_verbose("Using local APIC timer interrupts. Calibrating APIC timer ...\n");
 
 816
 817	/*
 818	 * There are platforms w/o global clockevent devices. Instead of
 819	 * making the calibration conditional on that, use a polling based
 820	 * approach everywhere.
 821	 */
 822	local_irq_disable();
 823
 
 
 
 
 824	/*
 825	 * Setup the APIC counter to maximum. There is no way the lapic
 826	 * can underflow in the 100ms detection time frame
 827	 */
 828	__setup_APIC_LVTT(0xffffffff, 0, 0);
 829
 830	/*
 831	 * Methods to terminate the calibration loop:
 832	 *  1) Global clockevent if available (jiffies)
 833	 *  2) TSC if available and frequency is known
 834	 */
 835	jif_start = READ_ONCE(jiffies);
 836
 837	if (tsc_khz) {
 838		tsc_start = rdtsc();
 839		tsc_perj = div_u64((u64)tsc_khz * 1000, HZ);
 840	}
 841
 842	/*
 843	 * Enable interrupts so the tick can fire, if a global
 844	 * clockevent device is available
 845	 */
 846	local_irq_enable();
 847
 848	while (lapic_cal_loops <= LAPIC_CAL_LOOPS) {
 849		/* Wait for a tick to elapse */
 850		while (1) {
 851			if (tsc_khz) {
 852				u64 tsc_now = rdtsc();
 853				if ((tsc_now - tsc_start) >= tsc_perj) {
 854					tsc_start += tsc_perj;
 855					break;
 856				}
 857			} else {
 858				unsigned long jif_now = READ_ONCE(jiffies);
 859
 860				if (time_after(jif_now, jif_start)) {
 861					jif_start = jif_now;
 862					break;
 863				}
 864			}
 865			cpu_relax();
 866		}
 867
 868		/* Invoke the calibration routine */
 869		local_irq_disable();
 870		lapic_cal_handler(NULL);
 871		local_irq_enable();
 872	}
 873
 874	local_irq_disable();
 875
 
 
 
 876	/* Build delta t1-t2 as apic timer counts down */
 877	delta = lapic_cal_t1 - lapic_cal_t2;
 878	apic_pr_verbose("... lapic delta = %ld\n", delta);
 879
 880	deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
 881
 882	/* we trust the PM based calibration if possible */
 883	pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
 884					&delta, &deltatsc);
 885
 886	lapic_timer_period = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
 887	lapic_init_clockevent();
 
 
 
 
 
 
 
 
 
 888
 889	apic_pr_verbose("..... delta %ld\n", delta);
 890	apic_pr_verbose("..... mult: %u\n", lapic_clockevent.mult);
 891	apic_pr_verbose("..... calibration result: %u\n", lapic_timer_period);
 
 892
 893	if (boot_cpu_has(X86_FEATURE_TSC)) {
 894		apic_pr_verbose("..... CPU clock speed is %ld.%04ld MHz.\n",
 895				(deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
 896				(deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
 
 897	}
 898
 899	apic_pr_verbose("..... host bus clock speed is %u.%04u MHz.\n",
 900			lapic_timer_period / (1000000 / HZ),
 901			lapic_timer_period % (1000000 / HZ));
 
 902
 903	/*
 904	 * Do a sanity check on the APIC calibration result
 905	 */
 906	if (lapic_timer_period < (1000000 / HZ)) {
 907		local_irq_enable();
 908		pr_warn("APIC frequency too slow, disabling apic timer\n");
 909		return -1;
 910	}
 911
 912	levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
 913
 914	/*
 915	 * PM timer calibration failed or not turned on so lets try APIC
 916	 * timer based calibration, if a global clockevent device is
 917	 * available.
 918	 */
 919	if (!pm_referenced && global_clock_event) {
 920		apic_pr_verbose("... verify APIC timer\n");
 921
 922		/*
 923		 * Setup the apic timer manually
 924		 */
 925		levt->event_handler = lapic_cal_handler;
 926		lapic_timer_set_periodic(levt);
 927		lapic_cal_loops = -1;
 928
 929		/* Let the interrupts run */
 930		local_irq_enable();
 931
 932		while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
 933			cpu_relax();
 934
 935		/* Stop the lapic timer */
 936		local_irq_disable();
 937		lapic_timer_shutdown(levt);
 938
 939		/* Jiffies delta */
 940		deltaj = lapic_cal_j2 - lapic_cal_j1;
 941		apic_pr_verbose("... jiffies delta = %lu\n", deltaj);
 942
 943		/* Check, if the jiffies result is consistent */
 944		if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
 945			apic_pr_verbose("... jiffies result ok\n");
 946		else
 947			levt->features |= CLOCK_EVT_FEAT_DUMMY;
 948	}
 949	local_irq_enable();
 950
 951	if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
 952		pr_warn("APIC timer disabled due to verification failure\n");
 953		return -1;
 954	}
 955
 956	return 0;
 957}
 958
 959/*
 960 * Setup the boot APIC
 961 *
 962 * Calibrate and verify the result.
 963 */
 964void __init setup_boot_APIC_clock(void)
 965{
 966	/*
 967	 * The local apic timer can be disabled via the kernel
 968	 * commandline or from the CPU detection code. Register the lapic
 969	 * timer as a dummy clock event source on SMP systems, so the
 970	 * broadcast mechanism is used. On UP systems simply ignore it.
 971	 */
 972	if (disable_apic_timer) {
 973		pr_info("Disabling APIC timer\n");
 974		/* No broadcast on UP ! */
 975		if (num_possible_cpus() > 1) {
 976			lapic_clockevent.mult = 1;
 977			setup_APIC_timer();
 978		}
 979		return;
 980	}
 981
 982	if (calibrate_APIC_clock()) {
 983		/* No broadcast on UP ! */
 984		if (num_possible_cpus() > 1)
 985			setup_APIC_timer();
 986		return;
 987	}
 988
 989	/*
 990	 * If nmi_watchdog is set to IO_APIC, we need the
 991	 * PIT/HPET going.  Otherwise register lapic as a dummy
 992	 * device.
 993	 */
 994	lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
 995
 996	/* Setup the lapic or request the broadcast */
 997	setup_APIC_timer();
 998	amd_e400_c1e_apic_setup();
 999}
1000
1001void setup_secondary_APIC_clock(void)
1002{
1003	setup_APIC_timer();
1004	amd_e400_c1e_apic_setup();
1005}
1006
1007/*
1008 * The guts of the apic timer interrupt
1009 */
1010static void local_apic_timer_interrupt(void)
1011{
1012	struct clock_event_device *evt = this_cpu_ptr(&lapic_events);
1013
1014	/*
1015	 * Normally we should not be here till LAPIC has been initialized but
1016	 * in some cases like kdump, its possible that there is a pending LAPIC
1017	 * timer interrupt from previous kernel's context and is delivered in
1018	 * new kernel the moment interrupts are enabled.
1019	 *
1020	 * Interrupts are enabled early and LAPIC is setup much later, hence
1021	 * its possible that when we get here evt->event_handler is NULL.
1022	 * Check for event_handler being NULL and discard the interrupt as
1023	 * spurious.
1024	 */
1025	if (!evt->event_handler) {
1026		pr_warn("Spurious LAPIC timer interrupt on cpu %d\n",
1027			smp_processor_id());
1028		/* Switch it off */
1029		lapic_timer_shutdown(evt);
1030		return;
1031	}
1032
1033	/*
1034	 * the NMI deadlock-detector uses this.
1035	 */
1036	inc_irq_stat(apic_timer_irqs);
1037
1038	evt->event_handler(evt);
1039}
1040
1041/*
1042 * Local APIC timer interrupt. This is the most natural way for doing
1043 * local interrupts, but local timer interrupts can be emulated by
1044 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
1045 *
1046 * [ if a single-CPU system runs an SMP kernel then we call the local
1047 *   interrupt as well. Thus we cannot inline the local irq ... ]
1048 */
1049DEFINE_IDTENTRY_SYSVEC(sysvec_apic_timer_interrupt)
1050{
1051	struct pt_regs *old_regs = set_irq_regs(regs);
1052
1053	apic_eoi();
 
 
 
 
 
 
 
 
1054	trace_local_timer_entry(LOCAL_TIMER_VECTOR);
1055	local_apic_timer_interrupt();
1056	trace_local_timer_exit(LOCAL_TIMER_VECTOR);
 
1057
1058	set_irq_regs(old_regs);
1059}
1060
 
 
 
 
 
1061/*
1062 * Local APIC start and shutdown
1063 */
1064
1065/**
1066 * clear_local_APIC - shutdown the local APIC
1067 *
1068 * This is called, when a CPU is disabled and before rebooting, so the state of
1069 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
1070 * leftovers during boot.
1071 */
1072void clear_local_APIC(void)
1073{
1074	int maxlvt;
1075	u32 v;
1076
1077	if (!apic_accessible())
 
1078		return;
1079
1080	maxlvt = lapic_get_maxlvt();
1081	/*
1082	 * Masking an LVT entry can trigger a local APIC error
1083	 * if the vector is zero. Mask LVTERR first to prevent this.
1084	 */
1085	if (maxlvt >= 3) {
1086		v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
1087		apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
1088	}
1089	/*
1090	 * Careful: we have to set masks only first to deassert
1091	 * any level-triggered sources.
1092	 */
1093	v = apic_read(APIC_LVTT);
1094	apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
1095	v = apic_read(APIC_LVT0);
1096	apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1097	v = apic_read(APIC_LVT1);
1098	apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1099	if (maxlvt >= 4) {
1100		v = apic_read(APIC_LVTPC);
1101		apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1102	}
1103
1104	/* lets not touch this if we didn't frob it */
1105#ifdef CONFIG_X86_THERMAL_VECTOR
1106	if (maxlvt >= 5) {
1107		v = apic_read(APIC_LVTTHMR);
1108		apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1109	}
1110#endif
1111#ifdef CONFIG_X86_MCE_INTEL
1112	if (maxlvt >= 6) {
1113		v = apic_read(APIC_LVTCMCI);
1114		if (!(v & APIC_LVT_MASKED))
1115			apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1116	}
1117#endif
1118
1119	/*
1120	 * Clean APIC state for other OSs:
1121	 */
1122	apic_write(APIC_LVTT, APIC_LVT_MASKED);
1123	apic_write(APIC_LVT0, APIC_LVT_MASKED);
1124	apic_write(APIC_LVT1, APIC_LVT_MASKED);
1125	if (maxlvt >= 3)
1126		apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1127	if (maxlvt >= 4)
1128		apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1129
1130	/* Integrated APIC (!82489DX) ? */
1131	if (lapic_is_integrated()) {
1132		if (maxlvt > 3)
1133			/* Clear ESR due to Pentium errata 3AP and 11AP */
1134			apic_write(APIC_ESR, 0);
1135		apic_read(APIC_ESR);
1136	}
1137}
1138
1139/**
1140 * apic_soft_disable - Clears and software disables the local APIC on hotplug
1141 *
1142 * Contrary to disable_local_APIC() this does not touch the enable bit in
1143 * MSR_IA32_APICBASE. Clearing that bit on systems based on the 3 wire APIC
1144 * bus would require a hardware reset as the APIC would lose track of bus
1145 * arbitration. On systems with FSB delivery APICBASE could be disabled,
1146 * but it has to be guaranteed that no interrupt is sent to the APIC while
1147 * in that state and it's not clear from the SDM whether it still responds
1148 * to INIT/SIPI messages. Stay on the safe side and use software disable.
1149 */
1150void apic_soft_disable(void)
1151{
1152	u32 value;
 
 
 
 
1153
1154	clear_local_APIC();
1155
1156	/* Soft disable APIC (implies clearing of registers for 82489DX!). */
 
 
 
1157	value = apic_read(APIC_SPIV);
1158	value &= ~APIC_SPIV_APIC_ENABLED;
1159	apic_write(APIC_SPIV, value);
1160}
1161
1162/**
1163 * disable_local_APIC - clear and disable the local APIC
1164 */
1165void disable_local_APIC(void)
1166{
1167	if (!apic_accessible())
1168		return;
1169
1170	apic_soft_disable();
1171
1172#ifdef CONFIG_X86_32
1173	/*
1174	 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1175	 * restore the disabled state.
1176	 */
1177	if (enabled_via_apicbase) {
1178		unsigned int l, h;
1179
1180		rdmsr(MSR_IA32_APICBASE, l, h);
1181		l &= ~MSR_IA32_APICBASE_ENABLE;
1182		wrmsr(MSR_IA32_APICBASE, l, h);
1183	}
1184#endif
1185}
1186
1187/*
1188 * If Linux enabled the LAPIC against the BIOS default disable it down before
1189 * re-entering the BIOS on shutdown.  Otherwise the BIOS may get confused and
1190 * not power-off.  Additionally clear all LVT entries before disable_local_APIC
1191 * for the case where Linux didn't enable the LAPIC.
1192 */
1193void lapic_shutdown(void)
1194{
1195	unsigned long flags;
1196
1197	if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
1198		return;
1199
1200	local_irq_save(flags);
1201
1202#ifdef CONFIG_X86_32
1203	if (!enabled_via_apicbase)
1204		clear_local_APIC();
1205	else
1206#endif
1207		disable_local_APIC();
1208
1209
1210	local_irq_restore(flags);
1211}
1212
1213/**
1214 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1215 */
1216void __init sync_Arb_IDs(void)
1217{
1218	/*
1219	 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1220	 * needed on AMD.
1221	 */
1222	if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1223		return;
1224
1225	/*
1226	 * Wait for idle.
1227	 */
1228	apic_wait_icr_idle();
1229
1230	apic_pr_debug("Synchronizing Arb IDs.\n");
1231	apic_write(APIC_ICR, APIC_DEST_ALLINC | APIC_INT_LEVELTRIG | APIC_DM_INIT);
 
1232}
1233
1234enum apic_intr_mode_id apic_intr_mode __ro_after_init;
1235
1236static int __init __apic_intr_mode_select(void)
1237{
1238	/* Check kernel option */
1239	if (apic_is_disabled) {
1240		pr_info("APIC disabled via kernel command line\n");
1241		return APIC_PIC;
1242	}
1243
1244	/* Check BIOS */
1245#ifdef CONFIG_X86_64
1246	/* On 64-bit, the APIC must be integrated, Check local APIC only */
1247	if (!boot_cpu_has(X86_FEATURE_APIC)) {
1248		apic_is_disabled = true;
1249		pr_info("APIC disabled by BIOS\n");
1250		return APIC_PIC;
1251	}
1252#else
1253	/* On 32-bit, the APIC may be integrated APIC or 82489DX */
1254
1255	/* Neither 82489DX nor integrated APIC ? */
1256	if (!boot_cpu_has(X86_FEATURE_APIC) && !smp_found_config) {
1257		apic_is_disabled = true;
1258		return APIC_PIC;
1259	}
1260
1261	/* If the BIOS pretends there is an integrated APIC ? */
1262	if (!boot_cpu_has(X86_FEATURE_APIC) &&
1263		APIC_INTEGRATED(boot_cpu_apic_version)) {
1264		apic_is_disabled = true;
1265		pr_err(FW_BUG "Local APIC not detected, force emulation\n");
 
1266		return APIC_PIC;
1267	}
1268#endif
1269
1270	/* Check MP table or ACPI MADT configuration */
1271	if (!smp_found_config) {
1272		disable_ioapic_support();
1273		if (!acpi_lapic) {
1274			pr_info("APIC: ACPI MADT or MP tables are not detected\n");
1275			return APIC_VIRTUAL_WIRE_NO_CONFIG;
1276		}
1277		return APIC_VIRTUAL_WIRE;
1278	}
1279
1280#ifdef CONFIG_SMP
1281	/* If SMP should be disabled, then really disable it! */
1282	if (!setup_max_cpus) {
1283		pr_info("APIC: SMP mode deactivated\n");
1284		return APIC_SYMMETRIC_IO_NO_ROUTING;
1285	}
 
 
 
 
 
 
1286#endif
1287
1288	return APIC_SYMMETRIC_IO;
1289}
1290
1291/* Select the interrupt delivery mode for the BSP */
1292void __init apic_intr_mode_select(void)
1293{
1294	apic_intr_mode = __apic_intr_mode_select();
1295}
1296
1297/*
1298 * An initial setup of the virtual wire mode.
1299 */
1300void __init init_bsp_APIC(void)
1301{
1302	unsigned int value;
1303
1304	/*
1305	 * Don't do the setup now if we have a SMP BIOS as the
1306	 * through-I/O-APIC virtual wire mode might be active.
1307	 */
1308	if (smp_found_config || !boot_cpu_has(X86_FEATURE_APIC))
1309		return;
1310
1311	/*
1312	 * Do not trust the local APIC being empty at bootup.
1313	 */
1314	clear_local_APIC();
1315
1316	/*
1317	 * Enable APIC.
1318	 */
1319	value = apic_read(APIC_SPIV);
1320	value &= ~APIC_VECTOR_MASK;
1321	value |= APIC_SPIV_APIC_ENABLED;
1322
1323#ifdef CONFIG_X86_32
1324	/* This bit is reserved on P4/Xeon and should be cleared */
1325	if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1326	    (boot_cpu_data.x86 == 15))
1327		value &= ~APIC_SPIV_FOCUS_DISABLED;
1328	else
1329#endif
1330		value |= APIC_SPIV_FOCUS_DISABLED;
1331	value |= SPURIOUS_APIC_VECTOR;
1332	apic_write(APIC_SPIV, value);
1333
1334	/*
1335	 * Set up the virtual wire mode.
1336	 */
1337	apic_write(APIC_LVT0, APIC_DM_EXTINT);
1338	value = APIC_DM_NMI;
1339	if (!lapic_is_integrated())		/* 82489DX */
1340		value |= APIC_LVT_LEVEL_TRIGGER;
1341	if (apic_extnmi == APIC_EXTNMI_NONE)
1342		value |= APIC_LVT_MASKED;
1343	apic_write(APIC_LVT1, value);
1344}
1345
1346static void __init apic_bsp_setup(bool upmode);
1347
1348/* Init the interrupt delivery mode for the BSP */
1349void __init apic_intr_mode_init(void)
1350{
1351	bool upmode = IS_ENABLED(CONFIG_UP_LATE_INIT);
1352
 
 
1353	switch (apic_intr_mode) {
1354	case APIC_PIC:
1355		pr_info("APIC: Keep in PIC mode(8259)\n");
1356		return;
1357	case APIC_VIRTUAL_WIRE:
1358		pr_info("APIC: Switch to virtual wire mode setup\n");
 
1359		break;
1360	case APIC_VIRTUAL_WIRE_NO_CONFIG:
1361		pr_info("APIC: Switch to virtual wire mode setup with no configuration\n");
1362		upmode = true;
 
1363		break;
1364	case APIC_SYMMETRIC_IO:
1365		pr_info("APIC: Switch to symmetric I/O mode setup\n");
 
1366		break;
1367	case APIC_SYMMETRIC_IO_NO_ROUTING:
1368		pr_info("APIC: Switch to symmetric I/O mode setup in no SMP routine\n");
1369		break;
1370	}
1371
1372	x86_64_probe_apic();
1373
1374	x86_32_install_bigsmp();
1375
1376	if (x86_platform.apic_post_init)
1377		x86_platform.apic_post_init();
1378
1379	apic_bsp_setup(upmode);
1380}
1381
1382static void lapic_setup_esr(void)
1383{
1384	unsigned int oldvalue, value, maxlvt;
1385
1386	if (!lapic_is_integrated()) {
1387		pr_info("No ESR for 82489DX.\n");
1388		return;
1389	}
1390
1391	if (apic->disable_esr) {
1392		/*
1393		 * Something untraceable is creating bad interrupts on
1394		 * secondary quads ... for the moment, just leave the
1395		 * ESR disabled - we can't do anything useful with the
1396		 * errors anyway - mbligh
1397		 */
1398		pr_info("Leaving ESR disabled.\n");
1399		return;
1400	}
1401
1402	maxlvt = lapic_get_maxlvt();
1403	if (maxlvt > 3)		/* Due to the Pentium erratum 3AP. */
1404		apic_write(APIC_ESR, 0);
1405	oldvalue = apic_read(APIC_ESR);
1406
1407	/* enables sending errors */
1408	value = ERROR_APIC_VECTOR;
1409	apic_write(APIC_LVTERR, value);
1410
1411	/*
1412	 * spec says clear errors after enabling vector.
1413	 */
1414	if (maxlvt > 3)
1415		apic_write(APIC_ESR, 0);
1416	value = apic_read(APIC_ESR);
1417	if (value != oldvalue) {
1418		apic_pr_verbose("ESR value before enabling vector: 0x%08x  after: 0x%08x\n",
1419				oldvalue, value);
1420	}
1421}
1422
1423#define APIC_IR_REGS		APIC_ISR_NR
1424#define APIC_IR_BITS		(APIC_IR_REGS * 32)
1425#define APIC_IR_MAPSIZE		(APIC_IR_BITS / BITS_PER_LONG)
1426
1427union apic_ir {
1428	unsigned long	map[APIC_IR_MAPSIZE];
1429	u32		regs[APIC_IR_REGS];
1430};
1431
1432static bool apic_check_and_ack(union apic_ir *irr, union apic_ir *isr)
1433{
1434	int i, bit;
1435
1436	/* Read the IRRs */
1437	for (i = 0; i < APIC_IR_REGS; i++)
1438		irr->regs[i] = apic_read(APIC_IRR + i * 0x10);
1439
1440	/* Read the ISRs */
1441	for (i = 0; i < APIC_IR_REGS; i++)
1442		isr->regs[i] = apic_read(APIC_ISR + i * 0x10);
1443
 
 
1444	/*
1445	 * If the ISR map is not empty. ACK the APIC and run another round
1446	 * to verify whether a pending IRR has been unblocked and turned
1447	 * into a ISR.
 
 
 
 
 
 
1448	 */
1449	if (!bitmap_empty(isr->map, APIC_IR_BITS)) {
1450		/*
1451		 * There can be multiple ISR bits set when a high priority
1452		 * interrupt preempted a lower priority one. Issue an ACK
1453		 * per set bit.
1454		 */
1455		for_each_set_bit(bit, isr->map, APIC_IR_BITS)
1456			apic_eoi();
1457		return true;
1458	}
1459
1460	return !bitmap_empty(irr->map, APIC_IR_BITS);
1461}
1462
1463/*
1464 * After a crash, we no longer service the interrupts and a pending
1465 * interrupt from previous kernel might still have ISR bit set.
1466 *
1467 * Most probably by now the CPU has serviced that pending interrupt and it
1468 * might not have done the apic_eoi() because it thought, interrupt
1469 * came from i8259 as ExtInt. LAPIC did not get EOI so it does not clear
1470 * the ISR bit and cpu thinks it has already serviced the interrupt. Hence
1471 * a vector might get locked. It was noticed for timer irq (vector
1472 * 0x31). Issue an extra EOI to clear ISR.
1473 *
1474 * If there are pending IRR bits they turn into ISR bits after a higher
1475 * priority ISR bit has been acked.
1476 */
1477static void apic_pending_intr_clear(void)
1478{
1479	union apic_ir irr, isr;
1480	unsigned int i;
1481
1482	/* 512 loops are way oversized and give the APIC a chance to obey. */
1483	for (i = 0; i < 512; i++) {
1484		if (!apic_check_and_ack(&irr, &isr))
1485			return;
1486	}
1487	/* Dump the IRR/ISR content if that failed */
1488	pr_warn("APIC: Stale IRR: %256pb ISR: %256pb\n", irr.map, isr.map);
1489}
1490
1491/**
1492 * setup_local_APIC - setup the local APIC
1493 *
1494 * Used to setup local APIC while initializing BSP or bringing up APs.
1495 * Always called with preemption disabled.
1496 */
1497static void setup_local_APIC(void)
1498{
1499	int cpu = smp_processor_id();
1500	unsigned int value;
 
 
 
1501
1502	if (apic_is_disabled) {
 
1503		disable_ioapic_support();
1504		return;
1505	}
1506
1507	/*
1508	 * If this comes from kexec/kcrash the APIC might be enabled in
1509	 * SPIV. Soft disable it before doing further initialization.
1510	 */
1511	value = apic_read(APIC_SPIV);
1512	value &= ~APIC_SPIV_APIC_ENABLED;
1513	apic_write(APIC_SPIV, value);
1514
1515#ifdef CONFIG_X86_32
1516	/* Pound the ESR really hard over the head with a big hammer - mbligh */
1517	if (lapic_is_integrated() && apic->disable_esr) {
1518		apic_write(APIC_ESR, 0);
1519		apic_write(APIC_ESR, 0);
1520		apic_write(APIC_ESR, 0);
1521		apic_write(APIC_ESR, 0);
1522	}
1523#endif
 
 
 
 
 
 
 
 
1524	/*
1525	 * Intel recommends to set DFR, LDR and TPR before enabling
1526	 * an APIC.  See e.g. "AP-388 82489DX User's Manual" (Intel
1527	 * document number 292116).
1528	 *
1529	 * Except for APICs which operate in physical destination mode.
1530	 */
1531	if (apic->init_apic_ldr)
1532		apic->init_apic_ldr();
1533
 
1534	/*
1535	 * Set Task Priority to 'accept all except vectors 0-31'.  An APIC
1536	 * vector in the 16-31 range could be delivered if TPR == 0, but we
1537	 * would think it's an exception and terrible things will happen.  We
1538	 * never change this later on.
 
 
 
 
 
 
 
 
 
 
1539	 */
1540	value = apic_read(APIC_TASKPRI);
1541	value &= ~APIC_TPRI_MASK;
1542	value |= 0x10;
1543	apic_write(APIC_TASKPRI, value);
1544
1545	/* Clear eventually stale ISR/IRR bits */
1546	apic_pending_intr_clear();
1547
1548	/*
1549	 * Now that we are all set up, enable the APIC
1550	 */
1551	value = apic_read(APIC_SPIV);
1552	value &= ~APIC_VECTOR_MASK;
1553	/*
1554	 * Enable APIC
1555	 */
1556	value |= APIC_SPIV_APIC_ENABLED;
1557
1558#ifdef CONFIG_X86_32
1559	/*
1560	 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1561	 * certain networking cards. If high frequency interrupts are
1562	 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1563	 * entry is masked/unmasked at a high rate as well then sooner or
1564	 * later IOAPIC line gets 'stuck', no more interrupts are received
1565	 * from the device. If focus CPU is disabled then the hang goes
1566	 * away, oh well :-(
1567	 *
1568	 * [ This bug can be reproduced easily with a level-triggered
1569	 *   PCI Ne2000 networking cards and PII/PIII processors, dual
1570	 *   BX chipset. ]
1571	 */
1572	/*
1573	 * Actually disabling the focus CPU check just makes the hang less
1574	 * frequent as it makes the interrupt distribution model be more
1575	 * like LRU than MRU (the short-term load is more even across CPUs).
1576	 */
1577
1578	/*
1579	 * - enable focus processor (bit==0)
1580	 * - 64bit mode always use processor focus
1581	 *   so no need to set it
1582	 */
1583	value &= ~APIC_SPIV_FOCUS_DISABLED;
1584#endif
1585
1586	/*
1587	 * Set spurious IRQ vector
1588	 */
1589	value |= SPURIOUS_APIC_VECTOR;
1590	apic_write(APIC_SPIV, value);
1591
1592	perf_events_lapic_init();
1593
1594	/*
1595	 * Set up LVT0, LVT1:
1596	 *
1597	 * set up through-local-APIC on the boot CPU's LINT0. This is not
1598	 * strictly necessary in pure symmetric-IO mode, but sometimes
1599	 * we delegate interrupts to the 8259A.
1600	 */
1601	/*
1602	 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1603	 */
1604	value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1605	if (!cpu && (pic_mode || !value || ioapic_is_disabled)) {
1606		value = APIC_DM_EXTINT;
1607		apic_pr_verbose("Enabled ExtINT on CPU#%d\n", cpu);
1608	} else {
1609		value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1610		apic_pr_verbose("Masked ExtINT on CPU#%d\n", cpu);
1611	}
1612	apic_write(APIC_LVT0, value);
1613
1614	/*
1615	 * Only the BSP sees the LINT1 NMI signal by default. This can be
1616	 * modified by apic_extnmi= boot option.
1617	 */
1618	if ((!cpu && apic_extnmi != APIC_EXTNMI_NONE) ||
1619	    apic_extnmi == APIC_EXTNMI_ALL)
1620		value = APIC_DM_NMI;
1621	else
1622		value = APIC_DM_NMI | APIC_LVT_MASKED;
1623
1624	/* Is 82489DX ? */
1625	if (!lapic_is_integrated())
1626		value |= APIC_LVT_LEVEL_TRIGGER;
1627	apic_write(APIC_LVT1, value);
1628
1629#ifdef CONFIG_X86_MCE_INTEL
1630	/* Recheck CMCI information after local APIC is up on CPU #0 */
1631	if (!cpu)
1632		cmci_recheck();
1633#endif
1634}
1635
1636static void end_local_APIC_setup(void)
1637{
1638	lapic_setup_esr();
1639
1640#ifdef CONFIG_X86_32
1641	{
1642		unsigned int value;
1643		/* Disable the local apic timer */
1644		value = apic_read(APIC_LVTT);
1645		value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1646		apic_write(APIC_LVTT, value);
1647	}
1648#endif
1649
1650	apic_pm_activate();
1651}
1652
1653/*
1654 * APIC setup function for application processors. Called from smpboot.c
1655 */
1656void apic_ap_setup(void)
1657{
1658	setup_local_APIC();
1659	end_local_APIC_setup();
1660}
1661
1662static __init void apic_read_boot_cpu_id(bool x2apic)
1663{
1664	/*
1665	 * This can be invoked from check_x2apic() before the APIC has been
1666	 * selected. But that code knows for sure that the BIOS enabled
1667	 * X2APIC.
1668	 */
1669	if (x2apic) {
1670		boot_cpu_physical_apicid = native_apic_msr_read(APIC_ID);
1671		boot_cpu_apic_version = GET_APIC_VERSION(native_apic_msr_read(APIC_LVR));
1672	} else {
1673		boot_cpu_physical_apicid = read_apic_id();
1674		boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
1675	}
1676	topology_register_boot_apic(boot_cpu_physical_apicid);
1677	x86_32_probe_bigsmp_early();
1678}
1679
1680#ifdef CONFIG_X86_X2APIC
1681int x2apic_mode;
1682EXPORT_SYMBOL_GPL(x2apic_mode);
1683
1684enum {
1685	X2APIC_OFF,
1686	X2APIC_DISABLED,
1687	/* All states below here have X2APIC enabled */
1688	X2APIC_ON,
1689	X2APIC_ON_LOCKED
1690};
1691static int x2apic_state;
1692
1693static bool x2apic_hw_locked(void)
1694{
1695	u64 x86_arch_cap_msr;
1696	u64 msr;
1697
1698	x86_arch_cap_msr = x86_read_arch_cap_msr();
1699	if (x86_arch_cap_msr & ARCH_CAP_XAPIC_DISABLE) {
1700		rdmsrl(MSR_IA32_XAPIC_DISABLE_STATUS, msr);
1701		return (msr & LEGACY_XAPIC_DISABLED);
1702	}
1703	return false;
1704}
1705
1706static void __x2apic_disable(void)
1707{
1708	u64 msr;
1709
1710	if (!boot_cpu_has(X86_FEATURE_APIC))
1711		return;
1712
1713	rdmsrl(MSR_IA32_APICBASE, msr);
1714	if (!(msr & X2APIC_ENABLE))
1715		return;
1716	/* Disable xapic and x2apic first and then reenable xapic mode */
1717	wrmsrl(MSR_IA32_APICBASE, msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1718	wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1719	printk_once(KERN_INFO "x2apic disabled\n");
1720}
1721
1722static void __x2apic_enable(void)
1723{
1724	u64 msr;
1725
1726	rdmsrl(MSR_IA32_APICBASE, msr);
1727	if (msr & X2APIC_ENABLE)
1728		return;
1729	wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1730	printk_once(KERN_INFO "x2apic enabled\n");
1731}
1732
1733static int __init setup_nox2apic(char *str)
1734{
1735	if (x2apic_enabled()) {
1736		u32 apicid = native_apic_msr_read(APIC_ID);
1737
1738		if (apicid >= 255) {
1739			pr_warn("Apicid: %08x, cannot enforce nox2apic\n",
1740				apicid);
1741			return 0;
1742		}
1743		if (x2apic_hw_locked()) {
1744			pr_warn("APIC locked in x2apic mode, can't disable\n");
1745			return 0;
1746		}
1747		pr_warn("x2apic already enabled.\n");
1748		__x2apic_disable();
1749	}
1750	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1751	x2apic_state = X2APIC_DISABLED;
1752	x2apic_mode = 0;
1753	return 0;
1754}
1755early_param("nox2apic", setup_nox2apic);
1756
1757/* Called from cpu_init() to enable x2apic on (secondary) cpus */
1758void x2apic_setup(void)
1759{
1760	/*
1761	 * Try to make the AP's APIC state match that of the BSP,  but if the
1762	 * BSP is unlocked and the AP is locked then there is a state mismatch.
1763	 * Warn about the mismatch in case a GP fault occurs due to a locked AP
1764	 * trying to be turned off.
1765	 */
1766	if (x2apic_state != X2APIC_ON_LOCKED && x2apic_hw_locked())
1767		pr_warn("x2apic lock mismatch between BSP and AP.\n");
1768	/*
1769	 * If x2apic is not in ON or LOCKED state, disable it if already enabled
1770	 * from BIOS.
1771	 */
1772	if (x2apic_state < X2APIC_ON) {
1773		__x2apic_disable();
1774		return;
1775	}
1776	__x2apic_enable();
1777}
1778
1779static __init void apic_set_fixmap(bool read_apic);
1780
1781static __init void x2apic_disable(void)
1782{
1783	u32 x2apic_id;
 
 
 
1784
1785	if (x2apic_state < X2APIC_ON)
1786		return;
1787
1788	x2apic_id = read_apic_id();
1789	if (x2apic_id >= 255)
1790		panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1791
1792	if (x2apic_hw_locked()) {
1793		pr_warn("Cannot disable locked x2apic, id: %08x\n", x2apic_id);
1794		return;
1795	}
1796
1797	__x2apic_disable();
1798
1799	x2apic_mode = 0;
1800	x2apic_state = X2APIC_DISABLED;
1801
1802	/*
1803	 * Don't reread the APIC ID as it was already done from
1804	 * check_x2apic() and the APIC driver still is a x2APIC variant,
1805	 * which fails to do the read after x2APIC was disabled.
1806	 */
1807	apic_set_fixmap(false);
1808}
1809
1810static __init void x2apic_enable(void)
1811{
1812	if (x2apic_state != X2APIC_OFF)
1813		return;
1814
1815	x2apic_mode = 1;
1816	x2apic_state = X2APIC_ON;
1817	__x2apic_enable();
1818}
1819
1820static __init void try_to_enable_x2apic(int remap_mode)
1821{
1822	if (x2apic_state == X2APIC_DISABLED)
1823		return;
1824
1825	if (remap_mode != IRQ_REMAP_X2APIC_MODE) {
1826		u32 apic_limit = 255;
1827
1828		/*
1829		 * Using X2APIC without IR is not architecturally supported
1830		 * on bare metal but may be supported in guests.
1831		 */
1832		if (!x86_init.hyper.x2apic_available()) {
 
1833			pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n");
1834			x2apic_disable();
1835			return;
1836		}
1837
1838		/*
1839		 * If the hypervisor supports extended destination ID in
1840		 * MSI, that increases the maximum APIC ID that can be
1841		 * used for non-remapped IRQ domains.
1842		 */
1843		if (x86_init.hyper.msi_ext_dest_id()) {
1844			virt_ext_dest_id = 1;
1845			apic_limit = 32767;
1846		}
1847
1848		/*
1849		 * Without IR, all CPUs can be addressed by IOAPIC/MSI only
1850		 * in physical mode, and CPUs with an APIC ID that cannot
1851		 * be addressed must not be brought online.
1852		 */
1853		x2apic_set_max_apicid(apic_limit);
1854		x2apic_phys = 1;
1855	}
1856	x2apic_enable();
1857}
1858
1859void __init check_x2apic(void)
1860{
1861	if (x2apic_enabled()) {
1862		pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n");
1863		x2apic_mode = 1;
1864		if (x2apic_hw_locked())
1865			x2apic_state = X2APIC_ON_LOCKED;
1866		else
1867			x2apic_state = X2APIC_ON;
1868		apic_read_boot_cpu_id(true);
1869	} else if (!boot_cpu_has(X86_FEATURE_X2APIC)) {
1870		x2apic_state = X2APIC_DISABLED;
1871	}
1872}
1873#else /* CONFIG_X86_X2APIC */
1874void __init check_x2apic(void)
1875{
1876	if (!apic_is_x2apic_enabled())
1877		return;
1878	/*
1879	 * Checkme: Can we simply turn off x2APIC here instead of disabling the APIC?
1880	 */
1881	pr_err("Kernel does not support x2APIC, please recompile with CONFIG_X86_X2APIC.\n");
1882	pr_err("Disabling APIC, expect reduced performance and functionality.\n");
1883
1884	apic_is_disabled = true;
1885	setup_clear_cpu_cap(X86_FEATURE_APIC);
1886}
 
1887
1888static inline void try_to_enable_x2apic(int remap_mode) { }
1889static inline void __x2apic_enable(void) { }
1890#endif /* !CONFIG_X86_X2APIC */
1891
1892void __init enable_IR_x2apic(void)
1893{
1894	unsigned long flags;
1895	int ret, ir_stat;
1896
1897	if (ioapic_is_disabled) {
1898		pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n");
1899		return;
1900	}
1901
1902	ir_stat = irq_remapping_prepare();
1903	if (ir_stat < 0 && !x2apic_supported())
1904		return;
1905
1906	ret = save_ioapic_entries();
1907	if (ret) {
1908		pr_info("Saving IO-APIC state failed: %d\n", ret);
1909		return;
1910	}
1911
1912	local_irq_save(flags);
1913	legacy_pic->mask_all();
1914	mask_ioapic_entries();
1915
1916	/* If irq_remapping_prepare() succeeded, try to enable it */
1917	if (ir_stat >= 0)
1918		ir_stat = irq_remapping_enable();
1919	/* ir_stat contains the remap mode or an error code */
1920	try_to_enable_x2apic(ir_stat);
1921
1922	if (ir_stat < 0)
1923		restore_ioapic_entries();
1924	legacy_pic->restore_mask();
1925	local_irq_restore(flags);
1926}
1927
1928#ifdef CONFIG_X86_64
1929/*
1930 * Detect and enable local APICs on non-SMP boards.
1931 * Original code written by Keir Fraser.
1932 * On AMD64 we trust the BIOS - if it says no APIC it is likely
1933 * not correctly set up (usually the APIC timer won't work etc.)
1934 */
1935static bool __init detect_init_APIC(void)
1936{
1937	if (!boot_cpu_has(X86_FEATURE_APIC)) {
1938		pr_info("No local APIC present\n");
1939		return false;
1940	}
1941
1942	register_lapic_address(APIC_DEFAULT_PHYS_BASE);
1943	return true;
1944}
1945#else
1946
1947static bool __init apic_verify(unsigned long addr)
1948{
1949	u32 features, h, l;
1950
1951	/*
1952	 * The APIC feature bit should now be enabled
1953	 * in `cpuid'
1954	 */
1955	features = cpuid_edx(1);
1956	if (!(features & (1 << X86_FEATURE_APIC))) {
1957		pr_warn("Could not enable APIC!\n");
1958		return false;
1959	}
1960	set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
 
1961
1962	/* The BIOS may have set up the APIC at some other address */
1963	if (boot_cpu_data.x86 >= 6) {
1964		rdmsr(MSR_IA32_APICBASE, l, h);
1965		if (l & MSR_IA32_APICBASE_ENABLE)
1966			addr = l & MSR_IA32_APICBASE_BASE;
1967	}
1968
1969	register_lapic_address(addr);
1970	pr_info("Found and enabled local APIC!\n");
1971	return true;
1972}
1973
1974bool __init apic_force_enable(unsigned long addr)
1975{
1976	u32 h, l;
1977
1978	if (apic_is_disabled)
1979		return false;
1980
1981	/*
1982	 * Some BIOSes disable the local APIC in the APIC_BASE
1983	 * MSR. This can only be done in software for Intel P6 or later
1984	 * and AMD K7 (Model > 1) or later.
1985	 */
1986	if (boot_cpu_data.x86 >= 6) {
1987		rdmsr(MSR_IA32_APICBASE, l, h);
1988		if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1989			pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1990			l &= ~MSR_IA32_APICBASE_BASE;
1991			l |= MSR_IA32_APICBASE_ENABLE | addr;
1992			wrmsr(MSR_IA32_APICBASE, l, h);
1993			enabled_via_apicbase = 1;
1994		}
1995	}
1996	return apic_verify(addr);
1997}
1998
1999/*
2000 * Detect and initialize APIC
2001 */
2002static bool __init detect_init_APIC(void)
2003{
2004	/* Disabled by kernel option? */
2005	if (apic_is_disabled)
2006		return false;
2007
2008	switch (boot_cpu_data.x86_vendor) {
2009	case X86_VENDOR_AMD:
2010		if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
2011		    (boot_cpu_data.x86 >= 15))
2012			break;
2013		goto no_apic;
2014	case X86_VENDOR_HYGON:
2015		break;
2016	case X86_VENDOR_INTEL:
2017		if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
2018		    (boot_cpu_data.x86 == 5 && boot_cpu_has(X86_FEATURE_APIC)))
2019			break;
2020		goto no_apic;
2021	default:
2022		goto no_apic;
2023	}
2024
2025	if (!boot_cpu_has(X86_FEATURE_APIC)) {
2026		/*
2027		 * Over-ride BIOS and try to enable the local APIC only if
2028		 * "lapic" specified.
2029		 */
2030		if (!force_enable_local_apic) {
2031			pr_info("Local APIC disabled by BIOS -- "
2032				"you can enable it with \"lapic\"\n");
2033			return false;
2034		}
2035		if (!apic_force_enable(APIC_DEFAULT_PHYS_BASE))
2036			return false;
2037	} else {
2038		if (!apic_verify(APIC_DEFAULT_PHYS_BASE))
2039			return false;
2040	}
2041
2042	apic_pm_activate();
2043
2044	return true;
2045
2046no_apic:
2047	pr_info("No local APIC present or hardware disabled\n");
2048	return false;
2049}
2050#endif
2051
2052/**
2053 * init_apic_mappings - initialize APIC mappings
2054 */
2055void __init init_apic_mappings(void)
2056{
2057	if (apic_validate_deadline_timer())
2058		pr_info("TSC deadline timer available\n");
 
2059
2060	if (x2apic_mode)
 
2061		return;
 
2062
2063	if (!smp_found_config) {
2064		if (!detect_init_APIC()) {
2065			pr_info("APIC: disable apic facility\n");
2066			apic_disable();
2067		}
 
 
 
 
 
 
 
 
 
2068	}
2069}
2070
2071static __init void apic_set_fixmap(bool read_apic)
2072{
2073	set_fixmap_nocache(FIX_APIC_BASE, mp_lapic_addr);
2074	apic_mmio_base = APIC_BASE;
2075	apic_pr_verbose("Mapped APIC to %16lx (%16lx)\n", apic_mmio_base, mp_lapic_addr);
2076	if (read_apic)
2077		apic_read_boot_cpu_id(false);
 
 
 
 
 
 
 
 
 
2078}
2079
2080void __init register_lapic_address(unsigned long address)
2081{
2082	/* This should only happen once */
2083	WARN_ON_ONCE(mp_lapic_addr);
2084	mp_lapic_addr = address;
2085
2086	if (!x2apic_mode)
2087		apic_set_fixmap(true);
 
 
 
 
 
 
 
2088}
2089
2090/*
2091 * Local APIC interrupts
2092 */
2093
2094/*
2095 * Common handling code for spurious_interrupt and spurious_vector entry
2096 * points below. No point in allowing the compiler to inline it twice.
2097 */
2098static noinline void handle_spurious_interrupt(u8 vector)
2099{
 
2100	u32 v;
2101
 
2102	trace_spurious_apic_entry(vector);
2103
2104	inc_irq_stat(irq_spurious_count);
2105
2106	/*
2107	 * If this is a spurious interrupt then do not acknowledge
2108	 */
2109	if (vector == SPURIOUS_APIC_VECTOR) {
2110		/* See SDM vol 3 */
2111		pr_info("Spurious APIC interrupt (vector 0xFF) on CPU#%d, should never happen.\n",
2112			smp_processor_id());
2113		goto out;
2114	}
2115
2116	/*
2117	 * If it is a vectored one, verify it's set in the ISR. If set,
2118	 * acknowledge it.
 
2119	 */
2120	v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1));
2121	if (v & (1 << (vector & 0x1f))) {
2122		pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Acked\n",
2123			vector, smp_processor_id());
2124		apic_eoi();
2125	} else {
2126		pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Not pending!\n",
2127			vector, smp_processor_id());
2128	}
2129out:
2130	trace_spurious_apic_exit(vector);
2131}
2132
2133/**
2134 * spurious_interrupt - Catch all for interrupts raised on unused vectors
2135 * @regs:	Pointer to pt_regs on stack
2136 * @vector:	The vector number
2137 *
2138 * This is invoked from ASM entry code to catch all interrupts which
2139 * trigger on an entry which is routed to the common_spurious idtentry
2140 * point.
2141 */
2142DEFINE_IDTENTRY_IRQ(spurious_interrupt)
2143{
2144	handle_spurious_interrupt(vector);
2145}
2146
2147DEFINE_IDTENTRY_SYSVEC(sysvec_spurious_apic_interrupt)
2148{
2149	handle_spurious_interrupt(SPURIOUS_APIC_VECTOR);
 
 
 
2150}
2151
2152/*
2153 * This interrupt should never happen with our APIC/SMP architecture
2154 */
2155DEFINE_IDTENTRY_SYSVEC(sysvec_error_interrupt)
2156{
2157	static const char * const error_interrupt_reason[] = {
2158		"Send CS error",		/* APIC Error Bit 0 */
2159		"Receive CS error",		/* APIC Error Bit 1 */
2160		"Send accept error",		/* APIC Error Bit 2 */
2161		"Receive accept error",		/* APIC Error Bit 3 */
2162		"Redirectable IPI",		/* APIC Error Bit 4 */
2163		"Send illegal vector",		/* APIC Error Bit 5 */
2164		"Received illegal vector",	/* APIC Error Bit 6 */
2165		"Illegal register address",	/* APIC Error Bit 7 */
2166	};
2167	u32 v, i = 0;
2168
 
2169	trace_error_apic_entry(ERROR_APIC_VECTOR);
2170
2171	/* First tickle the hardware, only then report what went on. -- REW */
2172	if (lapic_get_maxlvt() > 3)	/* Due to the Pentium erratum 3AP. */
2173		apic_write(APIC_ESR, 0);
2174	v = apic_read(APIC_ESR);
2175	apic_eoi();
2176	atomic_inc(&irq_err_count);
2177
2178	apic_pr_debug("APIC error on CPU%d: %02x", smp_processor_id(), v);
 
2179
2180	v &= 0xff;
2181	while (v) {
2182		if (v & 0x1)
2183			apic_pr_debug_cont(" : %s", error_interrupt_reason[i]);
2184		i++;
2185		v >>= 1;
2186	}
2187
2188	apic_pr_debug_cont("\n");
2189
2190	trace_error_apic_exit(ERROR_APIC_VECTOR);
 
2191}
2192
2193/**
2194 * connect_bsp_APIC - attach the APIC to the interrupt system
2195 */
2196static void __init connect_bsp_APIC(void)
2197{
2198#ifdef CONFIG_X86_32
2199	if (pic_mode) {
2200		/*
2201		 * Do not trust the local APIC being empty at bootup.
2202		 */
2203		clear_local_APIC();
2204		/*
2205		 * PIC mode, enable APIC mode in the IMCR, i.e.  connect BSP's
2206		 * local APIC to INT and NMI lines.
2207		 */
2208		apic_pr_verbose("Leaving PIC mode, enabling APIC mode.\n");
 
2209		imcr_pic_to_apic();
2210	}
2211#endif
2212}
2213
2214/**
2215 * disconnect_bsp_APIC - detach the APIC from the interrupt system
2216 * @virt_wire_setup:	indicates, whether virtual wire mode is selected
2217 *
2218 * Virtual wire mode is necessary to deliver legacy interrupts even when the
2219 * APIC is disabled.
2220 */
2221void disconnect_bsp_APIC(int virt_wire_setup)
2222{
2223	unsigned int value;
2224
2225#ifdef CONFIG_X86_32
2226	if (pic_mode) {
2227		/*
2228		 * Put the board back into PIC mode (has an effect only on
2229		 * certain older boards).  Note that APIC interrupts, including
2230		 * IPIs, won't work beyond this point!  The only exception are
2231		 * INIT IPIs.
2232		 */
2233		apic_pr_verbose("Disabling APIC mode, entering PIC mode.\n");
 
2234		imcr_apic_to_pic();
2235		return;
2236	}
2237#endif
2238
2239	/* Go back to Virtual Wire compatibility mode */
2240
2241	/* For the spurious interrupt use vector F, and enable it */
2242	value = apic_read(APIC_SPIV);
2243	value &= ~APIC_VECTOR_MASK;
2244	value |= APIC_SPIV_APIC_ENABLED;
2245	value |= 0xf;
2246	apic_write(APIC_SPIV, value);
2247
2248	if (!virt_wire_setup) {
2249		/*
2250		 * For LVT0 make it edge triggered, active high,
2251		 * external and enabled
2252		 */
2253		value = apic_read(APIC_LVT0);
2254		value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2255			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2256			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2257		value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2258		value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2259		apic_write(APIC_LVT0, value);
2260	} else {
2261		/* Disable LVT0 */
2262		apic_write(APIC_LVT0, APIC_LVT_MASKED);
2263	}
2264
2265	/*
2266	 * For LVT1 make it edge triggered, active high,
2267	 * nmi and enabled
2268	 */
2269	value = apic_read(APIC_LVT1);
2270	value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2271			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2272			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2273	value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2274	value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2275	apic_write(APIC_LVT1, value);
2276}
2277
2278void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg,
2279			   bool dmar)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2280{
2281	memset(msg, 0, sizeof(*msg));
2282
2283	msg->arch_addr_lo.base_address = X86_MSI_BASE_ADDRESS_LOW;
2284	msg->arch_addr_lo.dest_mode_logical = apic->dest_mode_logical;
2285	msg->arch_addr_lo.destid_0_7 = cfg->dest_apicid & 0xFF;
2286
2287	msg->arch_data.delivery_mode = APIC_DELIVERY_MODE_FIXED;
2288	msg->arch_data.vector = cfg->vector;
2289
2290	msg->address_hi = X86_MSI_BASE_ADDRESS_HIGH;
2291	/*
2292	 * Only the IOMMU itself can use the trick of putting destination
2293	 * APIC ID into the high bits of the address. Anything else would
2294	 * just be writing to memory if it tried that, and needs IR to
2295	 * address APICs which can't be addressed in the normal 32-bit
2296	 * address range at 0xFFExxxxx. That is typically just 8 bits, but
2297	 * some hypervisors allow the extended destination ID field in bits
2298	 * 5-11 to be used, giving support for 15 bits of APIC IDs in total.
2299	 */
2300	if (dmar)
2301		msg->arch_addr_hi.destid_8_31 = cfg->dest_apicid >> 8;
2302	else if (virt_ext_dest_id && cfg->dest_apicid < 0x8000)
2303		msg->arch_addr_lo.virt_destid_8_14 = cfg->dest_apicid >> 8;
2304	else
2305		WARN_ON_ONCE(cfg->dest_apicid > 0xFF);
2306}
2307
2308u32 x86_msi_msg_get_destid(struct msi_msg *msg, bool extid)
2309{
2310	u32 dest = msg->arch_addr_lo.destid_0_7;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2311
2312	if (extid)
2313		dest |= msg->arch_addr_hi.destid_8_31 << 8;
2314	return dest;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2315}
2316EXPORT_SYMBOL_GPL(x86_msi_msg_get_destid);
2317
2318static void __init apic_bsp_up_setup(void)
2319{
2320	reset_phys_cpu_present_map(boot_cpu_physical_apicid);
 
 
 
 
 
 
 
 
 
 
 
 
2321}
2322
2323/**
2324 * apic_bsp_setup - Setup function for local apic and io-apic
2325 * @upmode:		Force UP mode (for APIC_init_uniprocessor)
 
 
 
2326 */
2327static void __init apic_bsp_setup(bool upmode)
2328{
2329	connect_bsp_APIC();
2330	if (upmode)
2331		apic_bsp_up_setup();
2332	setup_local_APIC();
2333
2334	enable_IO_APIC();
2335	end_local_APIC_setup();
2336	irq_remap_enable_fault_handling();
2337	setup_IO_APIC();
2338	lapic_update_legacy_vectors();
2339}
2340
2341#ifdef CONFIG_UP_LATE_INIT
2342void __init up_late_init(void)
2343{
2344	if (apic_intr_mode == APIC_PIC)
2345		return;
2346
2347	/* Setup local timer */
2348	x86_init.timers.setup_percpu_clockev();
2349}
2350#endif
2351
2352/*
2353 * Power management
2354 */
2355#ifdef CONFIG_PM
2356
2357static struct {
2358	/*
2359	 * 'active' is true if the local APIC was enabled by us and
2360	 * not the BIOS; this signifies that we are also responsible
2361	 * for disabling it before entering apm/acpi suspend
2362	 */
2363	int active;
2364	/* r/w apic fields */
2365	u32 apic_id;
2366	unsigned int apic_taskpri;
2367	unsigned int apic_ldr;
2368	unsigned int apic_dfr;
2369	unsigned int apic_spiv;
2370	unsigned int apic_lvtt;
2371	unsigned int apic_lvtpc;
2372	unsigned int apic_lvt0;
2373	unsigned int apic_lvt1;
2374	unsigned int apic_lvterr;
2375	unsigned int apic_tmict;
2376	unsigned int apic_tdcr;
2377	unsigned int apic_thmr;
2378	unsigned int apic_cmci;
2379} apic_pm_state;
2380
2381static int lapic_suspend(void)
2382{
2383	unsigned long flags;
2384	int maxlvt;
2385
2386	if (!apic_pm_state.active)
2387		return 0;
2388
2389	maxlvt = lapic_get_maxlvt();
2390
2391	apic_pm_state.apic_id = apic_read(APIC_ID);
2392	apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2393	apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2394	apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2395	apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2396	apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2397	if (maxlvt >= 4)
2398		apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2399	apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2400	apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2401	apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2402	apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2403	apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2404#ifdef CONFIG_X86_THERMAL_VECTOR
2405	if (maxlvt >= 5)
2406		apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2407#endif
2408#ifdef CONFIG_X86_MCE_INTEL
2409	if (maxlvt >= 6)
2410		apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI);
2411#endif
2412
2413	local_irq_save(flags);
2414
2415	/*
2416	 * Mask IOAPIC before disabling the local APIC to prevent stale IRR
2417	 * entries on some implementations.
2418	 */
2419	mask_ioapic_entries();
2420
2421	disable_local_APIC();
2422
2423	irq_remapping_disable();
2424
2425	local_irq_restore(flags);
2426	return 0;
2427}
2428
2429static void lapic_resume(void)
2430{
2431	unsigned int l, h;
2432	unsigned long flags;
2433	int maxlvt;
2434
2435	if (!apic_pm_state.active)
2436		return;
2437
2438	local_irq_save(flags);
2439
2440	/*
2441	 * IO-APIC and PIC have their own resume routines.
2442	 * We just mask them here to make sure the interrupt
2443	 * subsystem is completely quiet while we enable x2apic
2444	 * and interrupt-remapping.
2445	 */
2446	mask_ioapic_entries();
2447	legacy_pic->mask_all();
2448
2449	if (x2apic_mode) {
2450		__x2apic_enable();
2451	} else {
2452		/*
2453		 * Make sure the APICBASE points to the right address
2454		 *
2455		 * FIXME! This will be wrong if we ever support suspend on
2456		 * SMP! We'll need to do this as part of the CPU restore!
2457		 */
2458		if (boot_cpu_data.x86 >= 6) {
2459			rdmsr(MSR_IA32_APICBASE, l, h);
2460			l &= ~MSR_IA32_APICBASE_BASE;
2461			l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2462			wrmsr(MSR_IA32_APICBASE, l, h);
2463		}
2464	}
2465
2466	maxlvt = lapic_get_maxlvt();
2467	apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2468	apic_write(APIC_ID, apic_pm_state.apic_id);
2469	apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2470	apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2471	apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2472	apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2473	apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2474	apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2475#ifdef CONFIG_X86_THERMAL_VECTOR
2476	if (maxlvt >= 5)
2477		apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2478#endif
2479#ifdef CONFIG_X86_MCE_INTEL
2480	if (maxlvt >= 6)
2481		apic_write(APIC_LVTCMCI, apic_pm_state.apic_cmci);
2482#endif
2483	if (maxlvt >= 4)
2484		apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2485	apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2486	apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2487	apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2488	apic_write(APIC_ESR, 0);
2489	apic_read(APIC_ESR);
2490	apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2491	apic_write(APIC_ESR, 0);
2492	apic_read(APIC_ESR);
2493
2494	irq_remapping_reenable(x2apic_mode);
2495
2496	local_irq_restore(flags);
2497}
2498
2499/*
2500 * This device has no shutdown method - fully functioning local APICs
2501 * are needed on every CPU up until machine_halt/restart/poweroff.
2502 */
2503
2504static struct syscore_ops lapic_syscore_ops = {
2505	.resume		= lapic_resume,
2506	.suspend	= lapic_suspend,
2507};
2508
2509static void apic_pm_activate(void)
2510{
2511	apic_pm_state.active = 1;
2512}
2513
2514static int __init init_lapic_sysfs(void)
2515{
2516	/* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2517	if (boot_cpu_has(X86_FEATURE_APIC))
2518		register_syscore_ops(&lapic_syscore_ops);
2519
2520	return 0;
2521}
2522
2523/* local apic needs to resume before other devices access its registers. */
2524core_initcall(init_lapic_sysfs);
2525
2526#else	/* CONFIG_PM */
2527
2528static void apic_pm_activate(void) { }
2529
2530#endif	/* CONFIG_PM */
2531
2532#ifdef CONFIG_X86_64
2533
2534static int multi_checked;
2535static int multi;
2536
2537static int set_multi(const struct dmi_system_id *d)
2538{
2539	if (multi)
2540		return 0;
2541	pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2542	multi = 1;
2543	return 0;
2544}
2545
2546static const struct dmi_system_id multi_dmi_table[] = {
2547	{
2548		.callback = set_multi,
2549		.ident = "IBM System Summit2",
2550		.matches = {
2551			DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2552			DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2553		},
2554	},
2555	{}
2556};
2557
2558static void dmi_check_multi(void)
2559{
2560	if (multi_checked)
2561		return;
2562
2563	dmi_check_system(multi_dmi_table);
2564	multi_checked = 1;
2565}
2566
2567/*
2568 * apic_is_clustered_box() -- Check if we can expect good TSC
2569 *
2570 * Thus far, the major user of this is IBM's Summit2 series:
2571 * Clustered boxes may have unsynced TSC problems if they are
2572 * multi-chassis.
2573 * Use DMI to check them
2574 */
2575int apic_is_clustered_box(void)
2576{
2577	dmi_check_multi();
2578	return multi;
2579}
2580#endif
2581
2582/*
2583 * APIC command line parameters
2584 */
2585static int __init setup_disableapic(char *arg)
2586{
2587	apic_is_disabled = true;
2588	setup_clear_cpu_cap(X86_FEATURE_APIC);
2589	return 0;
2590}
2591early_param("disableapic", setup_disableapic);
2592
2593/* same as disableapic, for compatibility */
2594static int __init setup_nolapic(char *arg)
2595{
2596	return setup_disableapic(arg);
2597}
2598early_param("nolapic", setup_nolapic);
2599
2600static int __init parse_lapic_timer_c2_ok(char *arg)
2601{
2602	local_apic_timer_c2_ok = 1;
2603	return 0;
2604}
2605early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2606
2607static int __init parse_disable_apic_timer(char *arg)
2608{
2609	disable_apic_timer = 1;
2610	return 0;
2611}
2612early_param("noapictimer", parse_disable_apic_timer);
2613
2614static int __init parse_nolapic_timer(char *arg)
2615{
2616	disable_apic_timer = 1;
2617	return 0;
2618}
2619early_param("nolapic_timer", parse_nolapic_timer);
2620
2621static int __init apic_set_verbosity(char *arg)
2622{
2623	if (!arg)  {
2624		if (IS_ENABLED(CONFIG_X86_32))
2625			return -EINVAL;
2626
2627		ioapic_is_disabled = false;
2628		return 0;
 
 
2629	}
2630
2631	if (strcmp("debug", arg) == 0)
2632		apic_verbosity = APIC_DEBUG;
2633	else if (strcmp("verbose", arg) == 0)
2634		apic_verbosity = APIC_VERBOSE;
2635#ifdef CONFIG_X86_64
2636	else {
2637		pr_warn("APIC Verbosity level %s not recognised"
2638			" use apic=verbose or apic=debug\n", arg);
2639		return -EINVAL;
2640	}
2641#endif
2642
2643	return 0;
2644}
2645early_param("apic", apic_set_verbosity);
2646
2647static int __init lapic_insert_resource(void)
2648{
2649	if (!apic_mmio_base)
2650		return -1;
2651
2652	/* Put local APIC into the resource map. */
2653	lapic_resource.start = apic_mmio_base;
2654	lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2655	insert_resource(&iomem_resource, &lapic_resource);
2656
2657	return 0;
2658}
2659
2660/*
2661 * need call insert after e820__reserve_resources()
2662 * that is using request_resource
2663 */
2664late_initcall(lapic_insert_resource);
 
 
 
 
 
 
 
 
 
2665
2666static int __init apic_set_extnmi(char *arg)
2667{
2668	if (!arg)
2669		return -EINVAL;
2670
2671	if (!strncmp("all", arg, 3))
2672		apic_extnmi = APIC_EXTNMI_ALL;
2673	else if (!strncmp("none", arg, 4))
2674		apic_extnmi = APIC_EXTNMI_NONE;
2675	else if (!strncmp("bsp", arg, 3))
2676		apic_extnmi = APIC_EXTNMI_BSP;
2677	else {
2678		pr_warn("Unknown external NMI delivery mode `%s' ignored\n", arg);
2679		return -EINVAL;
2680	}
2681
2682	return 0;
2683}
2684early_param("apic_extnmi", apic_set_extnmi);
v4.17
 
   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/export.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/mtrr.h>
  52#include <asm/time.h>
  53#include <asm/smp.h>
  54#include <asm/mce.h>
  55#include <asm/tsc.h>
  56#include <asm/hypervisor.h>
  57#include <asm/cpu_device_id.h>
  58#include <asm/intel-family.h>
 
 
  59
  60unsigned int num_processors;
  61
  62unsigned disabled_cpus;
  63
  64/* Processor that is doing the boot up */
  65unsigned int boot_cpu_physical_apicid = -1U;
  66EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid);
  67
  68u8 boot_cpu_apic_version;
  69
  70/*
  71 * The highest APIC ID seen during enumeration.
  72 */
  73static unsigned int max_physical_apicid;
  74
  75/*
  76 * Bitmask of physically existing CPUs:
  77 */
  78physid_mask_t phys_cpu_present_map;
  79
  80/*
  81 * Processor to be disabled specified by kernel parameter
  82 * disable_cpu_apicid=<int>, mostly used for the kdump 2nd kernel to
  83 * avoid undefined behaviour caused by sending INIT from AP to BSP.
  84 */
  85static unsigned int disabled_cpu_apicid __read_mostly = BAD_APICID;
  86
  87/*
  88 * This variable controls which CPUs receive external NMIs.  By default,
  89 * external NMIs are delivered only to the BSP.
  90 */
  91static int apic_extnmi = APIC_EXTNMI_BSP;
  92
  93/*
  94 * Map cpu index to physical APIC ID
  95 */
  96DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_cpu_to_apicid, BAD_APICID);
  97DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_bios_cpu_apicid, BAD_APICID);
  98DEFINE_EARLY_PER_CPU_READ_MOSTLY(u32, x86_cpu_to_acpiid, U32_MAX);
  99EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
 100EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
 101EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_acpiid);
 102
 103#ifdef CONFIG_X86_32
 
 104
 105/*
 106 * On x86_32, the mapping between cpu and logical apicid may vary
 107 * depending on apic in use.  The following early percpu variable is
 108 * used for the mapping.  This is where the behaviors of x86_64 and 32
 109 * actually diverge.  Let's keep it ugly for now.
 110 */
 111DEFINE_EARLY_PER_CPU_READ_MOSTLY(int, x86_cpu_to_logical_apicid, BAD_APICID);
 112
 
 113/* Local APIC was disabled by the BIOS and enabled by the kernel */
 114static int enabled_via_apicbase;
 115
 116/*
 117 * Handle interrupt mode configuration register (IMCR).
 118 * This register controls whether the interrupt signals
 119 * that reach the BSP come from the master PIC or from the
 120 * local APIC. Before entering Symmetric I/O Mode, either
 121 * the BIOS or the operating system must switch out of
 122 * PIC Mode by changing the IMCR.
 123 */
 124static inline void imcr_pic_to_apic(void)
 125{
 126	/* select IMCR register */
 127	outb(0x70, 0x22);
 128	/* NMI and 8259 INTR go through APIC */
 129	outb(0x01, 0x23);
 130}
 131
 132static inline void imcr_apic_to_pic(void)
 133{
 134	/* select IMCR register */
 135	outb(0x70, 0x22);
 136	/* NMI and 8259 INTR go directly to BSP */
 137	outb(0x00, 0x23);
 138}
 139#endif
 140
 141/*
 142 * Knob to control our willingness to enable the local APIC.
 143 *
 144 * +1=force-enable
 145 */
 146static int force_enable_local_apic __initdata;
 147
 148/*
 149 * APIC command line parameters
 150 */
 151static int __init parse_lapic(char *arg)
 152{
 153	if (IS_ENABLED(CONFIG_X86_32) && !arg)
 154		force_enable_local_apic = 1;
 155	else if (arg && !strncmp(arg, "notscdeadline", 13))
 156		setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
 157	return 0;
 158}
 159early_param("lapic", parse_lapic);
 160
 161#ifdef CONFIG_X86_64
 162static int apic_calibrate_pmtmr __initdata;
 163static __init int setup_apicpmtimer(char *s)
 164{
 165	apic_calibrate_pmtmr = 1;
 166	notsc_setup(NULL);
 167	return 0;
 168}
 169__setup("apicpmtimer", setup_apicpmtimer);
 170#endif
 171
 172unsigned long mp_lapic_addr;
 173int disable_apic;
 174/* Disable local APIC timer from the kernel commandline or via dmi quirk */
 175static int disable_apic_timer __initdata;
 176/* Local APIC timer works in C2 */
 177int local_apic_timer_c2_ok;
 178EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
 179
 180/*
 181 * Debug level, exported for io_apic.c
 182 */
 183unsigned int apic_verbosity;
 184
 185int pic_mode;
 186
 187/* Have we found an MP table */
 188int smp_found_config;
 189
 190static struct resource lapic_resource = {
 191	.name = "Local APIC",
 192	.flags = IORESOURCE_MEM | IORESOURCE_BUSY,
 193};
 194
 195unsigned int lapic_timer_frequency = 0;
 196
 197static void apic_pm_activate(void);
 198
 199static unsigned long apic_phys;
 200
 201/*
 202 * Get the LAPIC version
 203 */
 204static inline int lapic_get_version(void)
 205{
 206	return GET_APIC_VERSION(apic_read(APIC_LVR));
 207}
 208
 209/*
 210 * Check, if the APIC is integrated or a separate chip
 211 */
 212static inline int lapic_is_integrated(void)
 213{
 214	return APIC_INTEGRATED(lapic_get_version());
 215}
 216
 217/*
 218 * Check, whether this is a modern or a first generation APIC
 219 */
 220static int modern_apic(void)
 221{
 222	/* AMD systems use old APIC versions, so check the CPU */
 223	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
 224	    boot_cpu_data.x86 >= 0xf)
 225		return 1;
 
 
 
 
 
 226	return lapic_get_version() >= 0x14;
 227}
 228
 229/*
 230 * right after this call apic become NOOP driven
 231 * so apic->write/read doesn't do anything
 232 */
 233static void __init apic_disable(void)
 234{
 235	pr_info("APIC: switched to apic NOOP\n");
 236	apic = &apic_noop;
 237}
 238
 239void native_apic_wait_icr_idle(void)
 240{
 241	while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
 242		cpu_relax();
 243}
 244
 245u32 native_safe_apic_wait_icr_idle(void)
 246{
 247	u32 send_status;
 248	int timeout;
 249
 250	timeout = 0;
 251	do {
 252		send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
 253		if (!send_status)
 254			break;
 255		inc_irq_stat(icr_read_retry_count);
 256		udelay(100);
 257	} while (timeout++ < 1000);
 258
 259	return send_status;
 260}
 261
 262void native_apic_icr_write(u32 low, u32 id)
 263{
 264	unsigned long flags;
 265
 266	local_irq_save(flags);
 267	apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
 268	apic_write(APIC_ICR, low);
 269	local_irq_restore(flags);
 270}
 271
 272u64 native_apic_icr_read(void)
 273{
 274	u32 icr1, icr2;
 275
 276	icr2 = apic_read(APIC_ICR2);
 277	icr1 = apic_read(APIC_ICR);
 278
 279	return icr1 | ((u64)icr2 << 32);
 280}
 281
 282#ifdef CONFIG_X86_32
 283/**
 284 * get_physical_broadcast - Get number of physical broadcast IDs
 285 */
 286int get_physical_broadcast(void)
 287{
 288	return modern_apic() ? 0xff : 0xf;
 289}
 290#endif
 291
 292/**
 293 * lapic_get_maxlvt - get the maximum number of local vector table entries
 294 */
 295int lapic_get_maxlvt(void)
 296{
 297	/*
 298	 * - we always have APIC integrated on 64bit mode
 299	 * - 82489DXs do not report # of LVT entries
 300	 */
 301	return lapic_is_integrated() ? GET_APIC_MAXLVT(apic_read(APIC_LVR)) : 2;
 302}
 303
 304/*
 305 * Local APIC timer
 306 */
 307
 308/* Clock divisor */
 309#define APIC_DIVISOR 16
 310#define TSC_DIVISOR  8
 311
 
 
 
 312/*
 313 * This function sets up the local APIC timer, with a timeout of
 314 * 'clocks' APIC bus clock. During calibration we actually call
 315 * this function twice on the boot CPU, once with a bogus timeout
 316 * value, second time for real. The other (noncalibrating) CPUs
 317 * call this function only once, with the real, calibrated value.
 318 *
 319 * We do reads before writes even if unnecessary, to get around the
 320 * P5 APIC double write bug.
 321 */
 322static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
 323{
 324	unsigned int lvtt_value, tmp_value;
 325
 326	lvtt_value = LOCAL_TIMER_VECTOR;
 327	if (!oneshot)
 328		lvtt_value |= APIC_LVT_TIMER_PERIODIC;
 329	else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
 330		lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
 331
 
 
 
 
 
 
 332	if (!lapic_is_integrated())
 333		lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
 334
 335	if (!irqen)
 336		lvtt_value |= APIC_LVT_MASKED;
 337
 338	apic_write(APIC_LVTT, lvtt_value);
 339
 340	if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
 341		/*
 342		 * See Intel SDM: TSC-Deadline Mode chapter. In xAPIC mode,
 343		 * writing to the APIC LVTT and TSC_DEADLINE MSR isn't serialized.
 344		 * According to Intel, MFENCE can do the serialization here.
 345		 */
 346		asm volatile("mfence" : : : "memory");
 347
 348		printk_once(KERN_DEBUG "TSC deadline timer enabled\n");
 349		return;
 350	}
 351
 352	/*
 353	 * Divide PICLK by 16
 354	 */
 355	tmp_value = apic_read(APIC_TDCR);
 356	apic_write(APIC_TDCR,
 357		(tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
 358		APIC_TDR_DIV_16);
 359
 360	if (!oneshot)
 361		apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
 362}
 363
 364/*
 365 * Setup extended LVT, AMD specific
 366 *
 367 * Software should use the LVT offsets the BIOS provides.  The offsets
 368 * are determined by the subsystems using it like those for MCE
 369 * threshold or IBS.  On K8 only offset 0 (APIC500) and MCE interrupts
 370 * are supported. Beginning with family 10h at least 4 offsets are
 371 * available.
 372 *
 373 * Since the offsets must be consistent for all cores, we keep track
 374 * of the LVT offsets in software and reserve the offset for the same
 375 * vector also to be used on other cores. An offset is freed by
 376 * setting the entry to APIC_EILVT_MASKED.
 377 *
 378 * If the BIOS is right, there should be no conflicts. Otherwise a
 379 * "[Firmware Bug]: ..." error message is generated. However, if
 380 * software does not properly determines the offsets, it is not
 381 * necessarily a BIOS bug.
 382 */
 383
 384static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
 385
 386static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
 387{
 388	return (old & APIC_EILVT_MASKED)
 389		|| (new == APIC_EILVT_MASKED)
 390		|| ((new & ~APIC_EILVT_MASKED) == old);
 391}
 392
 393static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
 394{
 395	unsigned int rsvd, vector;
 396
 397	if (offset >= APIC_EILVT_NR_MAX)
 398		return ~0;
 399
 400	rsvd = atomic_read(&eilvt_offsets[offset]);
 401	do {
 402		vector = rsvd & ~APIC_EILVT_MASKED;	/* 0: unassigned */
 403		if (vector && !eilvt_entry_is_changeable(vector, new))
 404			/* may not change if vectors are different */
 405			return rsvd;
 406		rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
 407	} while (rsvd != new);
 408
 409	rsvd &= ~APIC_EILVT_MASKED;
 410	if (rsvd && rsvd != vector)
 411		pr_info("LVT offset %d assigned for vector 0x%02x\n",
 412			offset, rsvd);
 413
 414	return new;
 415}
 416
 417/*
 418 * If mask=1, the LVT entry does not generate interrupts while mask=0
 419 * enables the vector. See also the BKDGs. Must be called with
 420 * preemption disabled.
 421 */
 422
 423int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
 424{
 425	unsigned long reg = APIC_EILVTn(offset);
 426	unsigned int new, old, reserved;
 427
 428	new = (mask << 16) | (msg_type << 8) | vector;
 429	old = apic_read(reg);
 430	reserved = reserve_eilvt_offset(offset, new);
 431
 432	if (reserved != new) {
 433		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
 434		       "vector 0x%x, but the register is already in use for "
 435		       "vector 0x%x on another cpu\n",
 436		       smp_processor_id(), reg, offset, new, reserved);
 437		return -EINVAL;
 438	}
 439
 440	if (!eilvt_entry_is_changeable(old, 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 this cpu\n",
 444		       smp_processor_id(), reg, offset, new, old);
 445		return -EBUSY;
 446	}
 447
 448	apic_write(reg, new);
 449
 450	return 0;
 451}
 452EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
 453
 454/*
 455 * Program the next event, relative to now
 456 */
 457static int lapic_next_event(unsigned long delta,
 458			    struct clock_event_device *evt)
 459{
 460	apic_write(APIC_TMICT, delta);
 461	return 0;
 462}
 463
 464static int lapic_next_deadline(unsigned long delta,
 465			       struct clock_event_device *evt)
 466{
 467	u64 tsc;
 468
 
 
 
 469	tsc = rdtsc();
 470	wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
 471	return 0;
 472}
 473
 474static int lapic_timer_shutdown(struct clock_event_device *evt)
 475{
 476	unsigned int v;
 477
 478	/* Lapic used as dummy for broadcast ? */
 479	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
 480		return 0;
 481
 482	v = apic_read(APIC_LVTT);
 483	v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
 484	apic_write(APIC_LVTT, v);
 485	apic_write(APIC_TMICT, 0);
 
 
 
 
 
 
 
 
 
 
 
 
 486	return 0;
 487}
 488
 489static inline int
 490lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot)
 491{
 492	/* Lapic used as dummy for broadcast ? */
 493	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
 494		return 0;
 495
 496	__setup_APIC_LVTT(lapic_timer_frequency, oneshot, 1);
 497	return 0;
 498}
 499
 500static int lapic_timer_set_periodic(struct clock_event_device *evt)
 501{
 502	return lapic_timer_set_periodic_oneshot(evt, false);
 503}
 504
 505static int lapic_timer_set_oneshot(struct clock_event_device *evt)
 506{
 507	return lapic_timer_set_periodic_oneshot(evt, true);
 508}
 509
 510/*
 511 * Local APIC timer broadcast function
 512 */
 513static void lapic_timer_broadcast(const struct cpumask *mask)
 514{
 515#ifdef CONFIG_SMP
 516	apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
 517#endif
 518}
 519
 520
 521/*
 522 * The local apic timer can be used for any function which is CPU local.
 523 */
 524static struct clock_event_device lapic_clockevent = {
 525	.name				= "lapic",
 526	.features			= CLOCK_EVT_FEAT_PERIODIC |
 527					  CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP
 528					  | CLOCK_EVT_FEAT_DUMMY,
 529	.shift				= 32,
 530	.set_state_shutdown		= lapic_timer_shutdown,
 531	.set_state_periodic		= lapic_timer_set_periodic,
 532	.set_state_oneshot		= lapic_timer_set_oneshot,
 533	.set_state_oneshot_stopped	= lapic_timer_shutdown,
 534	.set_next_event			= lapic_next_event,
 535	.broadcast			= lapic_timer_broadcast,
 536	.rating				= 100,
 537	.irq				= -1,
 538};
 539static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
 540
 541#define DEADLINE_MODEL_MATCH_FUNC(model, func)	\
 542	{ X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)&func }
 543
 544#define DEADLINE_MODEL_MATCH_REV(model, rev)	\
 545	{ X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)rev }
 546
 547static u32 hsx_deadline_rev(void)
 548{
 549	switch (boot_cpu_data.x86_stepping) {
 550	case 0x02: return 0x3a; /* EP */
 551	case 0x04: return 0x0f; /* EX */
 552	}
 553
 554	return ~0U;
 555}
 556
 557static u32 bdx_deadline_rev(void)
 558{
 559	switch (boot_cpu_data.x86_stepping) {
 560	case 0x02: return 0x00000011;
 561	case 0x03: return 0x0700000e;
 562	case 0x04: return 0x0f00000c;
 563	case 0x05: return 0x0e000003;
 564	}
 565
 566	return ~0U;
 567}
 
 568
 569static u32 skx_deadline_rev(void)
 570{
 571	switch (boot_cpu_data.x86_stepping) {
 572	case 0x03: return 0x01000136;
 573	case 0x04: return 0x02000014;
 574	}
 575
 576	return ~0U;
 577}
 578
 579static const struct x86_cpu_id deadline_match[] = {
 580	DEADLINE_MODEL_MATCH_FUNC( INTEL_FAM6_HASWELL_X,	hsx_deadline_rev),
 581	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_X,	0x0b000020),
 582	DEADLINE_MODEL_MATCH_FUNC( INTEL_FAM6_BROADWELL_XEON_D,	bdx_deadline_rev),
 583	DEADLINE_MODEL_MATCH_FUNC( INTEL_FAM6_SKYLAKE_X,	skx_deadline_rev),
 584
 585	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_CORE,	0x22),
 586	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_ULT,	0x20),
 587	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_GT3E,	0x17),
 588
 589	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_CORE,	0x25),
 590	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_GT3E,	0x17),
 591
 592	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_SKYLAKE_MOBILE,	0xb2),
 593	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_SKYLAKE_DESKTOP,	0xb2),
 594
 595	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_KABYLAKE_MOBILE,	0x52),
 596	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_KABYLAKE_DESKTOP,	0x52),
 597
 598	{},
 599};
 600
 601static void apic_check_deadline_errata(void)
 602{
 603	const struct x86_cpu_id *m;
 604	u32 rev;
 605
 606	if (!boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER) ||
 607	    boot_cpu_has(X86_FEATURE_HYPERVISOR))
 608		return;
 
 609
 610	m = x86_match_cpu(deadline_match);
 611	if (!m)
 612		return;
 613
 614	/*
 615	 * Function pointers will have the MSB set due to address layout,
 616	 * immediate revisions will not.
 617	 */
 618	if ((long)m->driver_data < 0)
 619		rev = ((u32 (*)(void))(m->driver_data))();
 620	else
 621		rev = (u32)m->driver_data;
 622
 623	if (boot_cpu_data.microcode >= rev)
 624		return;
 625
 626	setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
 627	pr_err(FW_BUG "TSC_DEADLINE disabled due to Errata; "
 628	       "please update microcode to version: 0x%x (or later)\n", rev);
 
 629}
 630
 631/*
 632 * Setup the local APIC timer for this CPU. Copy the initialized values
 633 * of the boot CPU and register the clock event in the framework.
 634 */
 635static void setup_APIC_timer(void)
 636{
 637	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
 638
 639	if (this_cpu_has(X86_FEATURE_ARAT)) {
 640		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
 641		/* Make LAPIC timer preferrable over percpu HPET */
 642		lapic_clockevent.rating = 150;
 643	}
 644
 645	memcpy(levt, &lapic_clockevent, sizeof(*levt));
 646	levt->cpumask = cpumask_of(smp_processor_id());
 647
 648	if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
 649		levt->name = "lapic-deadline";
 650		levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
 651				    CLOCK_EVT_FEAT_DUMMY);
 652		levt->set_next_event = lapic_next_deadline;
 653		clockevents_config_and_register(levt,
 654						tsc_khz * (1000 / TSC_DIVISOR),
 655						0xF, ~0UL);
 656	} else
 657		clockevents_register_device(levt);
 658}
 659
 660/*
 661 * Install the updated TSC frequency from recalibration at the TSC
 662 * deadline clockevent devices.
 663 */
 664static void __lapic_update_tsc_freq(void *info)
 665{
 666	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
 667
 668	if (!this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
 669		return;
 670
 671	clockevents_update_freq(levt, tsc_khz * (1000 / TSC_DIVISOR));
 672}
 673
 674void lapic_update_tsc_freq(void)
 675{
 676	/*
 677	 * The clockevent device's ->mult and ->shift can both be
 678	 * changed. In order to avoid races, schedule the frequency
 679	 * update code on each CPU.
 680	 */
 681	on_each_cpu(__lapic_update_tsc_freq, NULL, 0);
 682}
 683
 684/*
 685 * In this functions we calibrate APIC bus clocks to the external timer.
 686 *
 687 * We want to do the calibration only once since we want to have local timer
 688 * irqs syncron. CPUs connected by the same APIC bus have the very same bus
 689 * frequency.
 690 *
 691 * This was previously done by reading the PIT/HPET and waiting for a wrap
 692 * around to find out, that a tick has elapsed. I have a box, where the PIT
 693 * readout is broken, so it never gets out of the wait loop again. This was
 694 * also reported by others.
 695 *
 696 * Monitoring the jiffies value is inaccurate and the clockevents
 697 * infrastructure allows us to do a simple substitution of the interrupt
 698 * handler.
 699 *
 700 * The calibration routine also uses the pm_timer when possible, as the PIT
 701 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
 702 * back to normal later in the boot process).
 703 */
 704
 705#define LAPIC_CAL_LOOPS		(HZ/10)
 706
 707static __initdata int lapic_cal_loops = -1;
 708static __initdata long lapic_cal_t1, lapic_cal_t2;
 709static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
 710static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
 711static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
 712
 713/*
 714 * Temporary interrupt handler.
 715 */
 716static void __init lapic_cal_handler(struct clock_event_device *dev)
 717{
 718	unsigned long long tsc = 0;
 719	long tapic = apic_read(APIC_TMCCT);
 720	unsigned long pm = acpi_pm_read_early();
 721
 722	if (boot_cpu_has(X86_FEATURE_TSC))
 723		tsc = rdtsc();
 724
 725	switch (lapic_cal_loops++) {
 726	case 0:
 727		lapic_cal_t1 = tapic;
 728		lapic_cal_tsc1 = tsc;
 729		lapic_cal_pm1 = pm;
 730		lapic_cal_j1 = jiffies;
 731		break;
 732
 733	case LAPIC_CAL_LOOPS:
 734		lapic_cal_t2 = tapic;
 735		lapic_cal_tsc2 = tsc;
 736		if (pm < lapic_cal_pm1)
 737			pm += ACPI_PM_OVRRUN;
 738		lapic_cal_pm2 = pm;
 739		lapic_cal_j2 = jiffies;
 740		break;
 741	}
 742}
 743
 744static int __init
 745calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
 746{
 747	const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
 748	const long pm_thresh = pm_100ms / 100;
 749	unsigned long mult;
 750	u64 res;
 751
 752#ifndef CONFIG_X86_PM_TIMER
 753	return -1;
 754#endif
 755
 756	apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
 757
 758	/* Check, if the PM timer is available */
 759	if (!deltapm)
 760		return -1;
 761
 762	mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
 763
 764	if (deltapm > (pm_100ms - pm_thresh) &&
 765	    deltapm < (pm_100ms + pm_thresh)) {
 766		apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
 767		return 0;
 768	}
 769
 770	res = (((u64)deltapm) *  mult) >> 22;
 771	do_div(res, 1000000);
 772	pr_warning("APIC calibration not consistent "
 773		   "with PM-Timer: %ldms instead of 100ms\n",(long)res);
 774
 775	/* Correct the lapic counter value */
 776	res = (((u64)(*delta)) * pm_100ms);
 777	do_div(res, deltapm);
 778	pr_info("APIC delta adjusted to PM-Timer: "
 779		"%lu (%ld)\n", (unsigned long)res, *delta);
 780	*delta = (long)res;
 781
 782	/* Correct the tsc counter value */
 783	if (boot_cpu_has(X86_FEATURE_TSC)) {
 784		res = (((u64)(*deltatsc)) * pm_100ms);
 785		do_div(res, deltapm);
 786		apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
 787					  "PM-Timer: %lu (%ld)\n",
 788					(unsigned long)res, *deltatsc);
 789		*deltatsc = (long)res;
 790	}
 791
 792	return 0;
 793}
 794
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 795static int __init calibrate_APIC_clock(void)
 796{
 797	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
 798	void (*real_handler)(struct clock_event_device *dev);
 
 799	unsigned long deltaj;
 800	long delta, deltatsc;
 801	int pm_referenced = 0;
 802
 803	/**
 804	 * check if lapic timer has already been calibrated by platform
 805	 * specific routine, such as tsc calibration code. if so, we just fill
 
 
 
 806	 * in the clockevent structure and return.
 807	 */
 808
 809	if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
 810		return 0;
 811	} else if (lapic_timer_frequency) {
 812		apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
 813				lapic_timer_frequency);
 814		lapic_clockevent.mult = div_sc(lapic_timer_frequency/APIC_DIVISOR,
 815					TICK_NSEC, lapic_clockevent.shift);
 816		lapic_clockevent.max_delta_ns =
 817			clockevent_delta2ns(0x7FFFFF, &lapic_clockevent);
 818		lapic_clockevent.max_delta_ticks = 0x7FFFFF;
 819		lapic_clockevent.min_delta_ns =
 820			clockevent_delta2ns(0xF, &lapic_clockevent);
 821		lapic_clockevent.min_delta_ticks = 0xF;
 822		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
 823		return 0;
 824	}
 825
 826	apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
 827		    "calibrating APIC timer ...\n");
 828
 
 
 
 
 
 829	local_irq_disable();
 830
 831	/* Replace the global interrupt handler */
 832	real_handler = global_clock_event->event_handler;
 833	global_clock_event->event_handler = lapic_cal_handler;
 834
 835	/*
 836	 * Setup the APIC counter to maximum. There is no way the lapic
 837	 * can underflow in the 100ms detection time frame
 838	 */
 839	__setup_APIC_LVTT(0xffffffff, 0, 0);
 840
 841	/* Let the interrupts run */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 842	local_irq_enable();
 843
 844	while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
 845		cpu_relax();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 846
 847	local_irq_disable();
 848
 849	/* Restore the real event handler */
 850	global_clock_event->event_handler = real_handler;
 851
 852	/* Build delta t1-t2 as apic timer counts down */
 853	delta = lapic_cal_t1 - lapic_cal_t2;
 854	apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
 855
 856	deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
 857
 858	/* we trust the PM based calibration if possible */
 859	pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
 860					&delta, &deltatsc);
 861
 862	/* Calculate the scaled math multiplication factor */
 863	lapic_clockevent.mult = div_sc(delta, TICK_NSEC * LAPIC_CAL_LOOPS,
 864				       lapic_clockevent.shift);
 865	lapic_clockevent.max_delta_ns =
 866		clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
 867	lapic_clockevent.max_delta_ticks = 0x7FFFFFFF;
 868	lapic_clockevent.min_delta_ns =
 869		clockevent_delta2ns(0xF, &lapic_clockevent);
 870	lapic_clockevent.min_delta_ticks = 0xF;
 871
 872	lapic_timer_frequency = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
 873
 874	apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
 875	apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
 876	apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
 877		    lapic_timer_frequency);
 878
 879	if (boot_cpu_has(X86_FEATURE_TSC)) {
 880		apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
 881			    "%ld.%04ld MHz.\n",
 882			    (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
 883			    (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
 884	}
 885
 886	apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
 887		    "%u.%04u MHz.\n",
 888		    lapic_timer_frequency / (1000000 / HZ),
 889		    lapic_timer_frequency % (1000000 / HZ));
 890
 891	/*
 892	 * Do a sanity check on the APIC calibration result
 893	 */
 894	if (lapic_timer_frequency < (1000000 / HZ)) {
 895		local_irq_enable();
 896		pr_warning("APIC frequency too slow, disabling apic timer\n");
 897		return -1;
 898	}
 899
 900	levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
 901
 902	/*
 903	 * PM timer calibration failed or not turned on
 904	 * so lets try APIC timer based calibration
 
 905	 */
 906	if (!pm_referenced) {
 907		apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
 908
 909		/*
 910		 * Setup the apic timer manually
 911		 */
 912		levt->event_handler = lapic_cal_handler;
 913		lapic_timer_set_periodic(levt);
 914		lapic_cal_loops = -1;
 915
 916		/* Let the interrupts run */
 917		local_irq_enable();
 918
 919		while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
 920			cpu_relax();
 921
 922		/* Stop the lapic timer */
 923		local_irq_disable();
 924		lapic_timer_shutdown(levt);
 925
 926		/* Jiffies delta */
 927		deltaj = lapic_cal_j2 - lapic_cal_j1;
 928		apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
 929
 930		/* Check, if the jiffies result is consistent */
 931		if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
 932			apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
 933		else
 934			levt->features |= CLOCK_EVT_FEAT_DUMMY;
 935	}
 936	local_irq_enable();
 937
 938	if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
 939		pr_warning("APIC timer disabled due to verification failure\n");
 940			return -1;
 941	}
 942
 943	return 0;
 944}
 945
 946/*
 947 * Setup the boot APIC
 948 *
 949 * Calibrate and verify the result.
 950 */
 951void __init setup_boot_APIC_clock(void)
 952{
 953	/*
 954	 * The local apic timer can be disabled via the kernel
 955	 * commandline or from the CPU detection code. Register the lapic
 956	 * timer as a dummy clock event source on SMP systems, so the
 957	 * broadcast mechanism is used. On UP systems simply ignore it.
 958	 */
 959	if (disable_apic_timer) {
 960		pr_info("Disabling APIC timer\n");
 961		/* No broadcast on UP ! */
 962		if (num_possible_cpus() > 1) {
 963			lapic_clockevent.mult = 1;
 964			setup_APIC_timer();
 965		}
 966		return;
 967	}
 968
 969	if (calibrate_APIC_clock()) {
 970		/* No broadcast on UP ! */
 971		if (num_possible_cpus() > 1)
 972			setup_APIC_timer();
 973		return;
 974	}
 975
 976	/*
 977	 * If nmi_watchdog is set to IO_APIC, we need the
 978	 * PIT/HPET going.  Otherwise register lapic as a dummy
 979	 * device.
 980	 */
 981	lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
 982
 983	/* Setup the lapic or request the broadcast */
 984	setup_APIC_timer();
 985	amd_e400_c1e_apic_setup();
 986}
 987
 988void setup_secondary_APIC_clock(void)
 989{
 990	setup_APIC_timer();
 991	amd_e400_c1e_apic_setup();
 992}
 993
 994/*
 995 * The guts of the apic timer interrupt
 996 */
 997static void local_apic_timer_interrupt(void)
 998{
 999	struct clock_event_device *evt = this_cpu_ptr(&lapic_events);
1000
1001	/*
1002	 * Normally we should not be here till LAPIC has been initialized but
1003	 * in some cases like kdump, its possible that there is a pending LAPIC
1004	 * timer interrupt from previous kernel's context and is delivered in
1005	 * new kernel the moment interrupts are enabled.
1006	 *
1007	 * Interrupts are enabled early and LAPIC is setup much later, hence
1008	 * its possible that when we get here evt->event_handler is NULL.
1009	 * Check for event_handler being NULL and discard the interrupt as
1010	 * spurious.
1011	 */
1012	if (!evt->event_handler) {
1013		pr_warning("Spurious LAPIC timer interrupt on cpu %d\n",
1014			   smp_processor_id());
1015		/* Switch it off */
1016		lapic_timer_shutdown(evt);
1017		return;
1018	}
1019
1020	/*
1021	 * the NMI deadlock-detector uses this.
1022	 */
1023	inc_irq_stat(apic_timer_irqs);
1024
1025	evt->event_handler(evt);
1026}
1027
1028/*
1029 * Local APIC timer interrupt. This is the most natural way for doing
1030 * local interrupts, but local timer interrupts can be emulated by
1031 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
1032 *
1033 * [ if a single-CPU system runs an SMP kernel then we call the local
1034 *   interrupt as well. Thus we cannot inline the local irq ... ]
1035 */
1036__visible void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
1037{
1038	struct pt_regs *old_regs = set_irq_regs(regs);
1039
1040	/*
1041	 * NOTE! We'd better ACK the irq immediately,
1042	 * because timer handling can be slow.
1043	 *
1044	 * update_process_times() expects us to have done irq_enter().
1045	 * Besides, if we don't timer interrupts ignore the global
1046	 * interrupt lock, which is the WrongThing (tm) to do.
1047	 */
1048	entering_ack_irq();
1049	trace_local_timer_entry(LOCAL_TIMER_VECTOR);
1050	local_apic_timer_interrupt();
1051	trace_local_timer_exit(LOCAL_TIMER_VECTOR);
1052	exiting_irq();
1053
1054	set_irq_regs(old_regs);
1055}
1056
1057int setup_profiling_timer(unsigned int multiplier)
1058{
1059	return -EINVAL;
1060}
1061
1062/*
1063 * Local APIC start and shutdown
1064 */
1065
1066/**
1067 * clear_local_APIC - shutdown the local APIC
1068 *
1069 * This is called, when a CPU is disabled and before rebooting, so the state of
1070 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
1071 * leftovers during boot.
1072 */
1073void clear_local_APIC(void)
1074{
1075	int maxlvt;
1076	u32 v;
1077
1078	/* APIC hasn't been mapped yet */
1079	if (!x2apic_mode && !apic_phys)
1080		return;
1081
1082	maxlvt = lapic_get_maxlvt();
1083	/*
1084	 * Masking an LVT entry can trigger a local APIC error
1085	 * if the vector is zero. Mask LVTERR first to prevent this.
1086	 */
1087	if (maxlvt >= 3) {
1088		v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
1089		apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
1090	}
1091	/*
1092	 * Careful: we have to set masks only first to deassert
1093	 * any level-triggered sources.
1094	 */
1095	v = apic_read(APIC_LVTT);
1096	apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
1097	v = apic_read(APIC_LVT0);
1098	apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1099	v = apic_read(APIC_LVT1);
1100	apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1101	if (maxlvt >= 4) {
1102		v = apic_read(APIC_LVTPC);
1103		apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1104	}
1105
1106	/* lets not touch this if we didn't frob it */
1107#ifdef CONFIG_X86_THERMAL_VECTOR
1108	if (maxlvt >= 5) {
1109		v = apic_read(APIC_LVTTHMR);
1110		apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1111	}
1112#endif
1113#ifdef CONFIG_X86_MCE_INTEL
1114	if (maxlvt >= 6) {
1115		v = apic_read(APIC_LVTCMCI);
1116		if (!(v & APIC_LVT_MASKED))
1117			apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1118	}
1119#endif
1120
1121	/*
1122	 * Clean APIC state for other OSs:
1123	 */
1124	apic_write(APIC_LVTT, APIC_LVT_MASKED);
1125	apic_write(APIC_LVT0, APIC_LVT_MASKED);
1126	apic_write(APIC_LVT1, APIC_LVT_MASKED);
1127	if (maxlvt >= 3)
1128		apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1129	if (maxlvt >= 4)
1130		apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1131
1132	/* Integrated APIC (!82489DX) ? */
1133	if (lapic_is_integrated()) {
1134		if (maxlvt > 3)
1135			/* Clear ESR due to Pentium errata 3AP and 11AP */
1136			apic_write(APIC_ESR, 0);
1137		apic_read(APIC_ESR);
1138	}
1139}
1140
1141/**
1142 * disable_local_APIC - clear and disable the local APIC
 
 
 
 
 
 
 
 
1143 */
1144void disable_local_APIC(void)
1145{
1146	unsigned int value;
1147
1148	/* APIC hasn't been mapped yet */
1149	if (!x2apic_mode && !apic_phys)
1150		return;
1151
1152	clear_local_APIC();
1153
1154	/*
1155	 * Disable APIC (implies clearing of registers
1156	 * for 82489DX!).
1157	 */
1158	value = apic_read(APIC_SPIV);
1159	value &= ~APIC_SPIV_APIC_ENABLED;
1160	apic_write(APIC_SPIV, value);
 
 
 
 
 
 
 
 
 
 
 
1161
1162#ifdef CONFIG_X86_32
1163	/*
1164	 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1165	 * restore the disabled state.
1166	 */
1167	if (enabled_via_apicbase) {
1168		unsigned int l, h;
1169
1170		rdmsr(MSR_IA32_APICBASE, l, h);
1171		l &= ~MSR_IA32_APICBASE_ENABLE;
1172		wrmsr(MSR_IA32_APICBASE, l, h);
1173	}
1174#endif
1175}
1176
1177/*
1178 * If Linux enabled the LAPIC against the BIOS default disable it down before
1179 * re-entering the BIOS on shutdown.  Otherwise the BIOS may get confused and
1180 * not power-off.  Additionally clear all LVT entries before disable_local_APIC
1181 * for the case where Linux didn't enable the LAPIC.
1182 */
1183void lapic_shutdown(void)
1184{
1185	unsigned long flags;
1186
1187	if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
1188		return;
1189
1190	local_irq_save(flags);
1191
1192#ifdef CONFIG_X86_32
1193	if (!enabled_via_apicbase)
1194		clear_local_APIC();
1195	else
1196#endif
1197		disable_local_APIC();
1198
1199
1200	local_irq_restore(flags);
1201}
1202
1203/**
1204 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1205 */
1206void __init sync_Arb_IDs(void)
1207{
1208	/*
1209	 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1210	 * needed on AMD.
1211	 */
1212	if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1213		return;
1214
1215	/*
1216	 * Wait for idle.
1217	 */
1218	apic_wait_icr_idle();
1219
1220	apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1221	apic_write(APIC_ICR, APIC_DEST_ALLINC |
1222			APIC_INT_LEVELTRIG | APIC_DM_INIT);
1223}
1224
1225enum apic_intr_mode_id apic_intr_mode;
1226
1227static int __init apic_intr_mode_select(void)
1228{
1229	/* Check kernel option */
1230	if (disable_apic) {
1231		pr_info("APIC disabled via kernel command line\n");
1232		return APIC_PIC;
1233	}
1234
1235	/* Check BIOS */
1236#ifdef CONFIG_X86_64
1237	/* On 64-bit, the APIC must be integrated, Check local APIC only */
1238	if (!boot_cpu_has(X86_FEATURE_APIC)) {
1239		disable_apic = 1;
1240		pr_info("APIC disabled by BIOS\n");
1241		return APIC_PIC;
1242	}
1243#else
1244	/* On 32-bit, the APIC may be integrated APIC or 82489DX */
1245
1246	/* Neither 82489DX nor integrated APIC ? */
1247	if (!boot_cpu_has(X86_FEATURE_APIC) && !smp_found_config) {
1248		disable_apic = 1;
1249		return APIC_PIC;
1250	}
1251
1252	/* If the BIOS pretends there is an integrated APIC ? */
1253	if (!boot_cpu_has(X86_FEATURE_APIC) &&
1254		APIC_INTEGRATED(boot_cpu_apic_version)) {
1255		disable_apic = 1;
1256		pr_err(FW_BUG "Local APIC %d not detected, force emulation\n",
1257				       boot_cpu_physical_apicid);
1258		return APIC_PIC;
1259	}
1260#endif
1261
1262	/* Check MP table or ACPI MADT configuration */
1263	if (!smp_found_config) {
1264		disable_ioapic_support();
1265		if (!acpi_lapic) {
1266			pr_info("APIC: ACPI MADT or MP tables are not detected\n");
1267			return APIC_VIRTUAL_WIRE_NO_CONFIG;
1268		}
1269		return APIC_VIRTUAL_WIRE;
1270	}
1271
1272#ifdef CONFIG_SMP
1273	/* If SMP should be disabled, then really disable it! */
1274	if (!setup_max_cpus) {
1275		pr_info("APIC: SMP mode deactivated\n");
1276		return APIC_SYMMETRIC_IO_NO_ROUTING;
1277	}
1278
1279	if (read_apic_id() != boot_cpu_physical_apicid) {
1280		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1281		     read_apic_id(), boot_cpu_physical_apicid);
1282		/* Or can we switch back to PIC here? */
1283	}
1284#endif
1285
1286	return APIC_SYMMETRIC_IO;
1287}
1288
 
 
 
 
 
 
1289/*
1290 * An initial setup of the virtual wire mode.
1291 */
1292void __init init_bsp_APIC(void)
1293{
1294	unsigned int value;
1295
1296	/*
1297	 * Don't do the setup now if we have a SMP BIOS as the
1298	 * through-I/O-APIC virtual wire mode might be active.
1299	 */
1300	if (smp_found_config || !boot_cpu_has(X86_FEATURE_APIC))
1301		return;
1302
1303	/*
1304	 * Do not trust the local APIC being empty at bootup.
1305	 */
1306	clear_local_APIC();
1307
1308	/*
1309	 * Enable APIC.
1310	 */
1311	value = apic_read(APIC_SPIV);
1312	value &= ~APIC_VECTOR_MASK;
1313	value |= APIC_SPIV_APIC_ENABLED;
1314
1315#ifdef CONFIG_X86_32
1316	/* This bit is reserved on P4/Xeon and should be cleared */
1317	if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1318	    (boot_cpu_data.x86 == 15))
1319		value &= ~APIC_SPIV_FOCUS_DISABLED;
1320	else
1321#endif
1322		value |= APIC_SPIV_FOCUS_DISABLED;
1323	value |= SPURIOUS_APIC_VECTOR;
1324	apic_write(APIC_SPIV, value);
1325
1326	/*
1327	 * Set up the virtual wire mode.
1328	 */
1329	apic_write(APIC_LVT0, APIC_DM_EXTINT);
1330	value = APIC_DM_NMI;
1331	if (!lapic_is_integrated())		/* 82489DX */
1332		value |= APIC_LVT_LEVEL_TRIGGER;
1333	if (apic_extnmi == APIC_EXTNMI_NONE)
1334		value |= APIC_LVT_MASKED;
1335	apic_write(APIC_LVT1, value);
1336}
1337
 
 
1338/* Init the interrupt delivery mode for the BSP */
1339void __init apic_intr_mode_init(void)
1340{
1341	bool upmode = IS_ENABLED(CONFIG_UP_LATE_INIT);
1342
1343	apic_intr_mode = apic_intr_mode_select();
1344
1345	switch (apic_intr_mode) {
1346	case APIC_PIC:
1347		pr_info("APIC: Keep in PIC mode(8259)\n");
1348		return;
1349	case APIC_VIRTUAL_WIRE:
1350		pr_info("APIC: Switch to virtual wire mode setup\n");
1351		default_setup_apic_routing();
1352		break;
1353	case APIC_VIRTUAL_WIRE_NO_CONFIG:
1354		pr_info("APIC: Switch to virtual wire mode setup with no configuration\n");
1355		upmode = true;
1356		default_setup_apic_routing();
1357		break;
1358	case APIC_SYMMETRIC_IO:
1359		pr_info("APIC: Switch to symmetric I/O mode setup\n");
1360		default_setup_apic_routing();
1361		break;
1362	case APIC_SYMMETRIC_IO_NO_ROUTING:
1363		pr_info("APIC: Switch to symmetric I/O mode setup in no SMP routine\n");
1364		break;
1365	}
1366
 
 
 
 
 
 
 
1367	apic_bsp_setup(upmode);
1368}
1369
1370static void lapic_setup_esr(void)
1371{
1372	unsigned int oldvalue, value, maxlvt;
1373
1374	if (!lapic_is_integrated()) {
1375		pr_info("No ESR for 82489DX.\n");
1376		return;
1377	}
1378
1379	if (apic->disable_esr) {
1380		/*
1381		 * Something untraceable is creating bad interrupts on
1382		 * secondary quads ... for the moment, just leave the
1383		 * ESR disabled - we can't do anything useful with the
1384		 * errors anyway - mbligh
1385		 */
1386		pr_info("Leaving ESR disabled.\n");
1387		return;
1388	}
1389
1390	maxlvt = lapic_get_maxlvt();
1391	if (maxlvt > 3)		/* Due to the Pentium erratum 3AP. */
1392		apic_write(APIC_ESR, 0);
1393	oldvalue = apic_read(APIC_ESR);
1394
1395	/* enables sending errors */
1396	value = ERROR_APIC_VECTOR;
1397	apic_write(APIC_LVTERR, value);
1398
1399	/*
1400	 * spec says clear errors after enabling vector.
1401	 */
1402	if (maxlvt > 3)
1403		apic_write(APIC_ESR, 0);
1404	value = apic_read(APIC_ESR);
1405	if (value != oldvalue)
1406		apic_printk(APIC_VERBOSE, "ESR value before enabling "
1407			"vector: 0x%08x  after: 0x%08x\n",
1408			oldvalue, value);
1409}
1410
1411static void apic_pending_intr_clear(void)
 
 
 
 
 
 
 
 
 
1412{
1413	long long max_loops = cpu_khz ? cpu_khz : 1000000;
1414	unsigned long long tsc = 0, ntsc;
1415	unsigned int queued;
1416	unsigned long value;
1417	int i, j, acked = 0;
 
 
 
 
1418
1419	if (boot_cpu_has(X86_FEATURE_TSC))
1420		tsc = rdtsc();
1421	/*
1422	 * After a crash, we no longer service the interrupts and a pending
1423	 * interrupt from previous kernel might still have ISR bit set.
1424	 *
1425	 * Most probably by now CPU has serviced that pending interrupt and
1426	 * it might not have done the ack_APIC_irq() because it thought,
1427	 * interrupt came from i8259 as ExtInt. LAPIC did not get EOI so it
1428	 * does not clear the ISR bit and cpu thinks it has already serivced
1429	 * the interrupt. Hence a vector might get locked. It was noticed
1430	 * for timer irq (vector 0x31). Issue an extra EOI to clear ISR.
1431	 */
1432	do {
1433		queued = 0;
1434		for (i = APIC_ISR_NR - 1; i >= 0; i--)
1435			queued |= apic_read(APIC_IRR + i*0x10);
1436
1437		for (i = APIC_ISR_NR - 1; i >= 0; i--) {
1438			value = apic_read(APIC_ISR + i*0x10);
1439			for_each_set_bit(j, &value, 32) {
1440				ack_APIC_irq();
1441				acked++;
1442			}
1443		}
1444		if (acked > 256) {
1445			pr_err("LAPIC pending interrupts after %d EOI\n", acked);
1446			break;
1447		}
1448		if (queued) {
1449			if (boot_cpu_has(X86_FEATURE_TSC) && cpu_khz) {
1450				ntsc = rdtsc();
1451				max_loops = (cpu_khz << 10) - (ntsc - tsc);
1452			} else {
1453				max_loops--;
1454			}
1455		}
1456	} while (queued && max_loops > 0);
1457	WARN_ON(max_loops <= 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1458}
1459
1460/**
1461 * setup_local_APIC - setup the local APIC
1462 *
1463 * Used to setup local APIC while initializing BSP or bringing up APs.
1464 * Always called with preemption disabled.
1465 */
1466static void setup_local_APIC(void)
1467{
1468	int cpu = smp_processor_id();
1469	unsigned int value;
1470#ifdef CONFIG_X86_32
1471	int logical_apicid, ldr_apicid;
1472#endif
1473
1474
1475	if (disable_apic) {
1476		disable_ioapic_support();
1477		return;
1478	}
1479
 
 
 
 
 
 
 
 
1480#ifdef CONFIG_X86_32
1481	/* Pound the ESR really hard over the head with a big hammer - mbligh */
1482	if (lapic_is_integrated() && apic->disable_esr) {
1483		apic_write(APIC_ESR, 0);
1484		apic_write(APIC_ESR, 0);
1485		apic_write(APIC_ESR, 0);
1486		apic_write(APIC_ESR, 0);
1487	}
1488#endif
1489	perf_events_lapic_init();
1490
1491	/*
1492	 * Double-check whether this APIC is really registered.
1493	 * This is meaningless in clustered apic mode, so we skip it.
1494	 */
1495	BUG_ON(!apic->apic_id_registered());
1496
1497	/*
1498	 * Intel recommends to set DFR, LDR and TPR before enabling
1499	 * an APIC.  See e.g. "AP-388 82489DX User's Manual" (Intel
1500	 * document number 292116).  So here it goes...
 
 
1501	 */
1502	apic->init_apic_ldr();
 
1503
1504#ifdef CONFIG_X86_32
1505	/*
1506	 * APIC LDR is initialized.  If logical_apicid mapping was
1507	 * initialized during get_smp_config(), make sure it matches the
1508	 * actual value.
1509	 */
1510	logical_apicid = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1511	ldr_apicid = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
1512	WARN_ON(logical_apicid != BAD_APICID && logical_apicid != ldr_apicid);
1513	/* always use the value from LDR */
1514	early_per_cpu(x86_cpu_to_logical_apicid, cpu) = ldr_apicid;
1515#endif
1516
1517	/*
1518	 * Set Task Priority to 'accept all'. We never change this
1519	 * later on.
1520	 */
1521	value = apic_read(APIC_TASKPRI);
1522	value &= ~APIC_TPRI_MASK;
 
1523	apic_write(APIC_TASKPRI, value);
1524
 
1525	apic_pending_intr_clear();
1526
1527	/*
1528	 * Now that we are all set up, enable the APIC
1529	 */
1530	value = apic_read(APIC_SPIV);
1531	value &= ~APIC_VECTOR_MASK;
1532	/*
1533	 * Enable APIC
1534	 */
1535	value |= APIC_SPIV_APIC_ENABLED;
1536
1537#ifdef CONFIG_X86_32
1538	/*
1539	 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1540	 * certain networking cards. If high frequency interrupts are
1541	 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1542	 * entry is masked/unmasked at a high rate as well then sooner or
1543	 * later IOAPIC line gets 'stuck', no more interrupts are received
1544	 * from the device. If focus CPU is disabled then the hang goes
1545	 * away, oh well :-(
1546	 *
1547	 * [ This bug can be reproduced easily with a level-triggered
1548	 *   PCI Ne2000 networking cards and PII/PIII processors, dual
1549	 *   BX chipset. ]
1550	 */
1551	/*
1552	 * Actually disabling the focus CPU check just makes the hang less
1553	 * frequent as it makes the interrupt distributon model be more
1554	 * like LRU than MRU (the short-term load is more even across CPUs).
1555	 */
1556
1557	/*
1558	 * - enable focus processor (bit==0)
1559	 * - 64bit mode always use processor focus
1560	 *   so no need to set it
1561	 */
1562	value &= ~APIC_SPIV_FOCUS_DISABLED;
1563#endif
1564
1565	/*
1566	 * Set spurious IRQ vector
1567	 */
1568	value |= SPURIOUS_APIC_VECTOR;
1569	apic_write(APIC_SPIV, value);
1570
 
 
1571	/*
1572	 * Set up LVT0, LVT1:
1573	 *
1574	 * set up through-local-APIC on the boot CPU's LINT0. This is not
1575	 * strictly necessary in pure symmetric-IO mode, but sometimes
1576	 * we delegate interrupts to the 8259A.
1577	 */
1578	/*
1579	 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1580	 */
1581	value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1582	if (!cpu && (pic_mode || !value || skip_ioapic_setup)) {
1583		value = APIC_DM_EXTINT;
1584		apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1585	} else {
1586		value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1587		apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1588	}
1589	apic_write(APIC_LVT0, value);
1590
1591	/*
1592	 * Only the BSP sees the LINT1 NMI signal by default. This can be
1593	 * modified by apic_extnmi= boot option.
1594	 */
1595	if ((!cpu && apic_extnmi != APIC_EXTNMI_NONE) ||
1596	    apic_extnmi == APIC_EXTNMI_ALL)
1597		value = APIC_DM_NMI;
1598	else
1599		value = APIC_DM_NMI | APIC_LVT_MASKED;
1600
1601	/* Is 82489DX ? */
1602	if (!lapic_is_integrated())
1603		value |= APIC_LVT_LEVEL_TRIGGER;
1604	apic_write(APIC_LVT1, value);
1605
1606#ifdef CONFIG_X86_MCE_INTEL
1607	/* Recheck CMCI information after local APIC is up on CPU #0 */
1608	if (!cpu)
1609		cmci_recheck();
1610#endif
1611}
1612
1613static void end_local_APIC_setup(void)
1614{
1615	lapic_setup_esr();
1616
1617#ifdef CONFIG_X86_32
1618	{
1619		unsigned int value;
1620		/* Disable the local apic timer */
1621		value = apic_read(APIC_LVTT);
1622		value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1623		apic_write(APIC_LVTT, value);
1624	}
1625#endif
1626
1627	apic_pm_activate();
1628}
1629
1630/*
1631 * APIC setup function for application processors. Called from smpboot.c
1632 */
1633void apic_ap_setup(void)
1634{
1635	setup_local_APIC();
1636	end_local_APIC_setup();
1637}
1638
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1639#ifdef CONFIG_X86_X2APIC
1640int x2apic_mode;
 
1641
1642enum {
1643	X2APIC_OFF,
 
 
1644	X2APIC_ON,
1645	X2APIC_DISABLED,
1646};
1647static int x2apic_state;
1648
 
 
 
 
 
 
 
 
 
 
 
 
 
1649static void __x2apic_disable(void)
1650{
1651	u64 msr;
1652
1653	if (!boot_cpu_has(X86_FEATURE_APIC))
1654		return;
1655
1656	rdmsrl(MSR_IA32_APICBASE, msr);
1657	if (!(msr & X2APIC_ENABLE))
1658		return;
1659	/* Disable xapic and x2apic first and then reenable xapic mode */
1660	wrmsrl(MSR_IA32_APICBASE, msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1661	wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1662	printk_once(KERN_INFO "x2apic disabled\n");
1663}
1664
1665static void __x2apic_enable(void)
1666{
1667	u64 msr;
1668
1669	rdmsrl(MSR_IA32_APICBASE, msr);
1670	if (msr & X2APIC_ENABLE)
1671		return;
1672	wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1673	printk_once(KERN_INFO "x2apic enabled\n");
1674}
1675
1676static int __init setup_nox2apic(char *str)
1677{
1678	if (x2apic_enabled()) {
1679		int apicid = native_apic_msr_read(APIC_ID);
1680
1681		if (apicid >= 255) {
1682			pr_warning("Apicid: %08x, cannot enforce nox2apic\n",
1683				   apicid);
 
 
 
 
1684			return 0;
1685		}
1686		pr_warning("x2apic already enabled.\n");
1687		__x2apic_disable();
1688	}
1689	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1690	x2apic_state = X2APIC_DISABLED;
1691	x2apic_mode = 0;
1692	return 0;
1693}
1694early_param("nox2apic", setup_nox2apic);
1695
1696/* Called from cpu_init() to enable x2apic on (secondary) cpus */
1697void x2apic_setup(void)
1698{
1699	/*
1700	 * If x2apic is not in ON state, disable it if already enabled
 
 
 
 
 
 
 
 
1701	 * from BIOS.
1702	 */
1703	if (x2apic_state != X2APIC_ON) {
1704		__x2apic_disable();
1705		return;
1706	}
1707	__x2apic_enable();
1708}
1709
 
 
1710static __init void x2apic_disable(void)
1711{
1712	u32 x2apic_id, state = x2apic_state;
1713
1714	x2apic_mode = 0;
1715	x2apic_state = X2APIC_DISABLED;
1716
1717	if (state != X2APIC_ON)
1718		return;
1719
1720	x2apic_id = read_apic_id();
1721	if (x2apic_id >= 255)
1722		panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1723
 
 
 
 
 
1724	__x2apic_disable();
1725	register_lapic_address(mp_lapic_addr);
 
 
 
 
 
 
 
 
 
1726}
1727
1728static __init void x2apic_enable(void)
1729{
1730	if (x2apic_state != X2APIC_OFF)
1731		return;
1732
1733	x2apic_mode = 1;
1734	x2apic_state = X2APIC_ON;
1735	__x2apic_enable();
1736}
1737
1738static __init void try_to_enable_x2apic(int remap_mode)
1739{
1740	if (x2apic_state == X2APIC_DISABLED)
1741		return;
1742
1743	if (remap_mode != IRQ_REMAP_X2APIC_MODE) {
1744		/* IR is required if there is APIC ID > 255 even when running
1745		 * under KVM
 
 
 
1746		 */
1747		if (max_physical_apicid > 255 ||
1748		    !x86_init.hyper.x2apic_available()) {
1749			pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n");
1750			x2apic_disable();
1751			return;
1752		}
1753
1754		/*
1755		 * without IR all CPUs can be addressed by IOAPIC/MSI
1756		 * only in physical mode
 
1757		 */
 
 
 
 
 
 
 
 
 
 
 
1758		x2apic_phys = 1;
1759	}
1760	x2apic_enable();
1761}
1762
1763void __init check_x2apic(void)
1764{
1765	if (x2apic_enabled()) {
1766		pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n");
1767		x2apic_mode = 1;
1768		x2apic_state = X2APIC_ON;
 
 
 
 
1769	} else if (!boot_cpu_has(X86_FEATURE_X2APIC)) {
1770		x2apic_state = X2APIC_DISABLED;
1771	}
1772}
1773#else /* CONFIG_X86_X2APIC */
1774static int __init validate_x2apic(void)
1775{
1776	if (!apic_is_x2apic_enabled())
1777		return 0;
1778	/*
1779	 * Checkme: Can we simply turn off x2apic here instead of panic?
1780	 */
1781	panic("BIOS has enabled x2apic but kernel doesn't support x2apic, please disable x2apic in BIOS.\n");
 
 
 
 
1782}
1783early_initcall(validate_x2apic);
1784
1785static inline void try_to_enable_x2apic(int remap_mode) { }
1786static inline void __x2apic_enable(void) { }
1787#endif /* !CONFIG_X86_X2APIC */
1788
1789void __init enable_IR_x2apic(void)
1790{
1791	unsigned long flags;
1792	int ret, ir_stat;
1793
1794	if (skip_ioapic_setup) {
1795		pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n");
1796		return;
1797	}
1798
1799	ir_stat = irq_remapping_prepare();
1800	if (ir_stat < 0 && !x2apic_supported())
1801		return;
1802
1803	ret = save_ioapic_entries();
1804	if (ret) {
1805		pr_info("Saving IO-APIC state failed: %d\n", ret);
1806		return;
1807	}
1808
1809	local_irq_save(flags);
1810	legacy_pic->mask_all();
1811	mask_ioapic_entries();
1812
1813	/* If irq_remapping_prepare() succeeded, try to enable it */
1814	if (ir_stat >= 0)
1815		ir_stat = irq_remapping_enable();
1816	/* ir_stat contains the remap mode or an error code */
1817	try_to_enable_x2apic(ir_stat);
1818
1819	if (ir_stat < 0)
1820		restore_ioapic_entries();
1821	legacy_pic->restore_mask();
1822	local_irq_restore(flags);
1823}
1824
1825#ifdef CONFIG_X86_64
1826/*
1827 * Detect and enable local APICs on non-SMP boards.
1828 * Original code written by Keir Fraser.
1829 * On AMD64 we trust the BIOS - if it says no APIC it is likely
1830 * not correctly set up (usually the APIC timer won't work etc.)
1831 */
1832static int __init detect_init_APIC(void)
1833{
1834	if (!boot_cpu_has(X86_FEATURE_APIC)) {
1835		pr_info("No local APIC present\n");
1836		return -1;
1837	}
1838
1839	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1840	return 0;
1841}
1842#else
1843
1844static int __init apic_verify(void)
1845{
1846	u32 features, h, l;
1847
1848	/*
1849	 * The APIC feature bit should now be enabled
1850	 * in `cpuid'
1851	 */
1852	features = cpuid_edx(1);
1853	if (!(features & (1 << X86_FEATURE_APIC))) {
1854		pr_warning("Could not enable APIC!\n");
1855		return -1;
1856	}
1857	set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1858	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1859
1860	/* The BIOS may have set up the APIC at some other address */
1861	if (boot_cpu_data.x86 >= 6) {
1862		rdmsr(MSR_IA32_APICBASE, l, h);
1863		if (l & MSR_IA32_APICBASE_ENABLE)
1864			mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1865	}
1866
 
1867	pr_info("Found and enabled local APIC!\n");
1868	return 0;
1869}
1870
1871int __init apic_force_enable(unsigned long addr)
1872{
1873	u32 h, l;
1874
1875	if (disable_apic)
1876		return -1;
1877
1878	/*
1879	 * Some BIOSes disable the local APIC in the APIC_BASE
1880	 * MSR. This can only be done in software for Intel P6 or later
1881	 * and AMD K7 (Model > 1) or later.
1882	 */
1883	if (boot_cpu_data.x86 >= 6) {
1884		rdmsr(MSR_IA32_APICBASE, l, h);
1885		if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1886			pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1887			l &= ~MSR_IA32_APICBASE_BASE;
1888			l |= MSR_IA32_APICBASE_ENABLE | addr;
1889			wrmsr(MSR_IA32_APICBASE, l, h);
1890			enabled_via_apicbase = 1;
1891		}
1892	}
1893	return apic_verify();
1894}
1895
1896/*
1897 * Detect and initialize APIC
1898 */
1899static int __init detect_init_APIC(void)
1900{
1901	/* Disabled by kernel option? */
1902	if (disable_apic)
1903		return -1;
1904
1905	switch (boot_cpu_data.x86_vendor) {
1906	case X86_VENDOR_AMD:
1907		if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
1908		    (boot_cpu_data.x86 >= 15))
1909			break;
1910		goto no_apic;
 
 
1911	case X86_VENDOR_INTEL:
1912		if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
1913		    (boot_cpu_data.x86 == 5 && boot_cpu_has(X86_FEATURE_APIC)))
1914			break;
1915		goto no_apic;
1916	default:
1917		goto no_apic;
1918	}
1919
1920	if (!boot_cpu_has(X86_FEATURE_APIC)) {
1921		/*
1922		 * Over-ride BIOS and try to enable the local APIC only if
1923		 * "lapic" specified.
1924		 */
1925		if (!force_enable_local_apic) {
1926			pr_info("Local APIC disabled by BIOS -- "
1927				"you can enable it with \"lapic\"\n");
1928			return -1;
1929		}
1930		if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
1931			return -1;
1932	} else {
1933		if (apic_verify())
1934			return -1;
1935	}
1936
1937	apic_pm_activate();
1938
1939	return 0;
1940
1941no_apic:
1942	pr_info("No local APIC present or hardware disabled\n");
1943	return -1;
1944}
1945#endif
1946
1947/**
1948 * init_apic_mappings - initialize APIC mappings
1949 */
1950void __init init_apic_mappings(void)
1951{
1952	unsigned int new_apicid;
1953
1954	apic_check_deadline_errata();
1955
1956	if (x2apic_mode) {
1957		boot_cpu_physical_apicid = read_apic_id();
1958		return;
1959	}
1960
1961	/* If no local APIC can be found return early */
1962	if (!smp_found_config && detect_init_APIC()) {
1963		/* lets NOP'ify apic operations */
1964		pr_info("APIC: disable apic facility\n");
1965		apic_disable();
1966	} else {
1967		apic_phys = mp_lapic_addr;
1968
1969		/*
1970		 * If the system has ACPI MADT tables or MP info, the LAPIC
1971		 * address is already registered.
1972		 */
1973		if (!acpi_lapic && !smp_found_config)
1974			register_lapic_address(apic_phys);
1975	}
 
1976
1977	/*
1978	 * Fetch the APIC ID of the BSP in case we have a
1979	 * default configuration (or the MP table is broken).
1980	 */
1981	new_apicid = read_apic_id();
1982	if (boot_cpu_physical_apicid != new_apicid) {
1983		boot_cpu_physical_apicid = new_apicid;
1984		/*
1985		 * yeah -- we lie about apic_version
1986		 * in case if apic was disabled via boot option
1987		 * but it's not a problem for SMP compiled kernel
1988		 * since apic_intr_mode_select is prepared for such
1989		 * a case and disable smp mode
1990		 */
1991		boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
1992	}
1993}
1994
1995void __init register_lapic_address(unsigned long address)
1996{
 
 
1997	mp_lapic_addr = address;
1998
1999	if (!x2apic_mode) {
2000		set_fixmap_nocache(FIX_APIC_BASE, address);
2001		apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
2002			    APIC_BASE, address);
2003	}
2004	if (boot_cpu_physical_apicid == -1U) {
2005		boot_cpu_physical_apicid  = read_apic_id();
2006		boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
2007	}
2008}
2009
2010/*
2011 * Local APIC interrupts
2012 */
2013
2014/*
2015 * This interrupt should _never_ happen with our APIC/SMP architecture
 
2016 */
2017__visible void __irq_entry smp_spurious_interrupt(struct pt_regs *regs)
2018{
2019	u8 vector = ~regs->orig_ax;
2020	u32 v;
2021
2022	entering_irq();
2023	trace_spurious_apic_entry(vector);
2024
 
 
 
 
 
 
 
 
 
 
 
 
2025	/*
2026	 * Check if this really is a spurious interrupt and ACK it
2027	 * if it is a vectored one.  Just in case...
2028	 * Spurious interrupts should not be ACKed.
2029	 */
2030	v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1));
2031	if (v & (1 << (vector & 0x1f)))
2032		ack_APIC_irq();
 
 
 
 
 
 
 
 
 
2033
2034	inc_irq_stat(irq_spurious_count);
 
 
 
 
 
 
 
 
 
 
 
 
2035
2036	/* see sw-dev-man vol 3, chapter 7.4.13.5 */
2037	pr_info("spurious APIC interrupt through vector %02x on CPU#%d, "
2038		"should never happen.\n", vector, smp_processor_id());
2039
2040	trace_spurious_apic_exit(vector);
2041	exiting_irq();
2042}
2043
2044/*
2045 * This interrupt should never happen with our APIC/SMP architecture
2046 */
2047__visible void __irq_entry smp_error_interrupt(struct pt_regs *regs)
2048{
2049	static const char * const error_interrupt_reason[] = {
2050		"Send CS error",		/* APIC Error Bit 0 */
2051		"Receive CS error",		/* APIC Error Bit 1 */
2052		"Send accept error",		/* APIC Error Bit 2 */
2053		"Receive accept error",		/* APIC Error Bit 3 */
2054		"Redirectable IPI",		/* APIC Error Bit 4 */
2055		"Send illegal vector",		/* APIC Error Bit 5 */
2056		"Received illegal vector",	/* APIC Error Bit 6 */
2057		"Illegal register address",	/* APIC Error Bit 7 */
2058	};
2059	u32 v, i = 0;
2060
2061	entering_irq();
2062	trace_error_apic_entry(ERROR_APIC_VECTOR);
2063
2064	/* First tickle the hardware, only then report what went on. -- REW */
2065	if (lapic_get_maxlvt() > 3)	/* Due to the Pentium erratum 3AP. */
2066		apic_write(APIC_ESR, 0);
2067	v = apic_read(APIC_ESR);
2068	ack_APIC_irq();
2069	atomic_inc(&irq_err_count);
2070
2071	apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x",
2072		    smp_processor_id(), v);
2073
2074	v &= 0xff;
2075	while (v) {
2076		if (v & 0x1)
2077			apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
2078		i++;
2079		v >>= 1;
2080	}
2081
2082	apic_printk(APIC_DEBUG, KERN_CONT "\n");
2083
2084	trace_error_apic_exit(ERROR_APIC_VECTOR);
2085	exiting_irq();
2086}
2087
2088/**
2089 * connect_bsp_APIC - attach the APIC to the interrupt system
2090 */
2091static void __init connect_bsp_APIC(void)
2092{
2093#ifdef CONFIG_X86_32
2094	if (pic_mode) {
2095		/*
2096		 * Do not trust the local APIC being empty at bootup.
2097		 */
2098		clear_local_APIC();
2099		/*
2100		 * PIC mode, enable APIC mode in the IMCR, i.e.  connect BSP's
2101		 * local APIC to INT and NMI lines.
2102		 */
2103		apic_printk(APIC_VERBOSE, "leaving PIC mode, "
2104				"enabling APIC mode.\n");
2105		imcr_pic_to_apic();
2106	}
2107#endif
2108}
2109
2110/**
2111 * disconnect_bsp_APIC - detach the APIC from the interrupt system
2112 * @virt_wire_setup:	indicates, whether virtual wire mode is selected
2113 *
2114 * Virtual wire mode is necessary to deliver legacy interrupts even when the
2115 * APIC is disabled.
2116 */
2117void disconnect_bsp_APIC(int virt_wire_setup)
2118{
2119	unsigned int value;
2120
2121#ifdef CONFIG_X86_32
2122	if (pic_mode) {
2123		/*
2124		 * Put the board back into PIC mode (has an effect only on
2125		 * certain older boards).  Note that APIC interrupts, including
2126		 * IPIs, won't work beyond this point!  The only exception are
2127		 * INIT IPIs.
2128		 */
2129		apic_printk(APIC_VERBOSE, "disabling APIC mode, "
2130				"entering PIC mode.\n");
2131		imcr_apic_to_pic();
2132		return;
2133	}
2134#endif
2135
2136	/* Go back to Virtual Wire compatibility mode */
2137
2138	/* For the spurious interrupt use vector F, and enable it */
2139	value = apic_read(APIC_SPIV);
2140	value &= ~APIC_VECTOR_MASK;
2141	value |= APIC_SPIV_APIC_ENABLED;
2142	value |= 0xf;
2143	apic_write(APIC_SPIV, value);
2144
2145	if (!virt_wire_setup) {
2146		/*
2147		 * For LVT0 make it edge triggered, active high,
2148		 * external and enabled
2149		 */
2150		value = apic_read(APIC_LVT0);
2151		value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2152			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2153			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2154		value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2155		value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2156		apic_write(APIC_LVT0, value);
2157	} else {
2158		/* Disable LVT0 */
2159		apic_write(APIC_LVT0, APIC_LVT_MASKED);
2160	}
2161
2162	/*
2163	 * For LVT1 make it edge triggered, active high,
2164	 * nmi and enabled
2165	 */
2166	value = apic_read(APIC_LVT1);
2167	value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2168			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2169			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2170	value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2171	value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2172	apic_write(APIC_LVT1, value);
2173}
2174
2175/*
2176 * The number of allocated logical CPU IDs. Since logical CPU IDs are allocated
2177 * contiguously, it equals to current allocated max logical CPU ID plus 1.
2178 * All allocated CPU IDs should be in the [0, nr_logical_cpuids) range,
2179 * so the maximum of nr_logical_cpuids is nr_cpu_ids.
2180 *
2181 * NOTE: Reserve 0 for BSP.
2182 */
2183static int nr_logical_cpuids = 1;
2184
2185/*
2186 * Used to store mapping between logical CPU IDs and APIC IDs.
2187 */
2188static int cpuid_to_apicid[] = {
2189	[0 ... NR_CPUS - 1] = -1,
2190};
2191
2192/*
2193 * Should use this API to allocate logical CPU IDs to keep nr_logical_cpuids
2194 * and cpuid_to_apicid[] synchronized.
2195 */
2196static int allocate_logical_cpuid(int apicid)
2197{
2198	int i;
2199
2200	/*
2201	 * cpuid <-> apicid mapping is persistent, so when a cpu is up,
2202	 * check if the kernel has allocated a cpuid for it.
2203	 */
2204	for (i = 0; i < nr_logical_cpuids; i++) {
2205		if (cpuid_to_apicid[i] == apicid)
2206			return i;
2207	}
2208
2209	/* Allocate a new cpuid. */
2210	if (nr_logical_cpuids >= nr_cpu_ids) {
2211		WARN_ONCE(1, "APIC: NR_CPUS/possible_cpus limit of %u reached. "
2212			     "Processor %d/0x%x and the rest are ignored.\n",
2213			     nr_cpu_ids, nr_logical_cpuids, apicid);
2214		return -EINVAL;
2215	}
2216
2217	cpuid_to_apicid[nr_logical_cpuids] = apicid;
2218	return nr_logical_cpuids++;
 
 
 
 
2219}
2220
2221int generic_processor_info(int apicid, int version)
2222{
2223	int cpu, max = nr_cpu_ids;
2224	bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
2225				phys_cpu_present_map);
2226
2227	/*
2228	 * boot_cpu_physical_apicid is designed to have the apicid
2229	 * returned by read_apic_id(), i.e, the apicid of the
2230	 * currently booting-up processor. However, on some platforms,
2231	 * it is temporarily modified by the apicid reported as BSP
2232	 * through MP table. Concretely:
2233	 *
2234	 * - arch/x86/kernel/mpparse.c: MP_processor_info()
2235	 * - arch/x86/mm/amdtopology.c: amd_numa_init()
2236	 *
2237	 * This function is executed with the modified
2238	 * boot_cpu_physical_apicid. So, disabled_cpu_apicid kernel
2239	 * parameter doesn't work to disable APs on kdump 2nd kernel.
2240	 *
2241	 * Since fixing handling of boot_cpu_physical_apicid requires
2242	 * another discussion and tests on each platform, we leave it
2243	 * for now and here we use read_apic_id() directly in this
2244	 * function, generic_processor_info().
2245	 */
2246	if (disabled_cpu_apicid != BAD_APICID &&
2247	    disabled_cpu_apicid != read_apic_id() &&
2248	    disabled_cpu_apicid == apicid) {
2249		int thiscpu = num_processors + disabled_cpus;
2250
2251		pr_warning("APIC: Disabling requested cpu."
2252			   " Processor %d/0x%x ignored.\n",
2253			   thiscpu, apicid);
2254
2255		disabled_cpus++;
2256		return -ENODEV;
2257	}
2258
2259	/*
2260	 * If boot cpu has not been detected yet, then only allow upto
2261	 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2262	 */
2263	if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2264	    apicid != boot_cpu_physical_apicid) {
2265		int thiscpu = max + disabled_cpus - 1;
2266
2267		pr_warning(
2268			"APIC: NR_CPUS/possible_cpus limit of %i almost"
2269			" reached. Keeping one slot for boot cpu."
2270			"  Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2271
2272		disabled_cpus++;
2273		return -ENODEV;
2274	}
2275
2276	if (num_processors >= nr_cpu_ids) {
2277		int thiscpu = max + disabled_cpus;
2278
2279		pr_warning("APIC: NR_CPUS/possible_cpus limit of %i "
2280			   "reached. Processor %d/0x%x ignored.\n",
2281			   max, thiscpu, apicid);
2282
2283		disabled_cpus++;
2284		return -EINVAL;
2285	}
2286
2287	if (apicid == boot_cpu_physical_apicid) {
2288		/*
2289		 * x86_bios_cpu_apicid is required to have processors listed
2290		 * in same order as logical cpu numbers. Hence the first
2291		 * entry is BSP, and so on.
2292		 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2293		 * for BSP.
2294		 */
2295		cpu = 0;
2296
2297		/* Logical cpuid 0 is reserved for BSP. */
2298		cpuid_to_apicid[0] = apicid;
2299	} else {
2300		cpu = allocate_logical_cpuid(apicid);
2301		if (cpu < 0) {
2302			disabled_cpus++;
2303			return -EINVAL;
2304		}
2305	}
2306
2307	/*
2308	 * Validate version
2309	 */
2310	if (version == 0x0) {
2311		pr_warning("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2312			   cpu, apicid);
2313		version = 0x10;
2314	}
2315
2316	if (version != boot_cpu_apic_version) {
2317		pr_warning("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2318			boot_cpu_apic_version, cpu, version);
2319	}
2320
2321	if (apicid > max_physical_apicid)
2322		max_physical_apicid = apicid;
2323
2324#if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2325	early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2326	early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2327#endif
2328#ifdef CONFIG_X86_32
2329	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2330		apic->x86_32_early_logical_apicid(cpu);
2331#endif
2332	set_cpu_possible(cpu, true);
2333	physid_set(apicid, phys_cpu_present_map);
2334	set_cpu_present(cpu, true);
2335	num_processors++;
2336
2337	return cpu;
2338}
2339
2340int hard_smp_processor_id(void)
2341{
2342	return read_apic_id();
2343}
2344
2345/*
2346 * Override the generic EOI implementation with an optimized version.
2347 * Only called during early boot when only one CPU is active and with
2348 * interrupts disabled, so we know this does not race with actual APIC driver
2349 * use.
2350 */
2351void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2352{
2353	struct apic **drv;
2354
2355	for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2356		/* Should happen once for each apic */
2357		WARN_ON((*drv)->eoi_write == eoi_write);
2358		(*drv)->native_eoi_write = (*drv)->eoi_write;
2359		(*drv)->eoi_write = eoi_write;
2360	}
2361}
 
2362
2363static void __init apic_bsp_up_setup(void)
2364{
2365#ifdef CONFIG_X86_64
2366	apic_write(APIC_ID, apic->set_apic_id(boot_cpu_physical_apicid));
2367#else
2368	/*
2369	 * Hack: In case of kdump, after a crash, kernel might be booting
2370	 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
2371	 * might be zero if read from MP tables. Get it from LAPIC.
2372	 */
2373# ifdef CONFIG_CRASH_DUMP
2374	boot_cpu_physical_apicid = read_apic_id();
2375# endif
2376#endif
2377	physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
2378}
2379
2380/**
2381 * apic_bsp_setup - Setup function for local apic and io-apic
2382 * @upmode:		Force UP mode (for APIC_init_uniprocessor)
2383 *
2384 * Returns:
2385 * apic_id of BSP APIC
2386 */
2387void __init apic_bsp_setup(bool upmode)
2388{
2389	connect_bsp_APIC();
2390	if (upmode)
2391		apic_bsp_up_setup();
2392	setup_local_APIC();
2393
2394	enable_IO_APIC();
2395	end_local_APIC_setup();
2396	irq_remap_enable_fault_handling();
2397	setup_IO_APIC();
 
2398}
2399
2400#ifdef CONFIG_UP_LATE_INIT
2401void __init up_late_init(void)
2402{
2403	if (apic_intr_mode == APIC_PIC)
2404		return;
2405
2406	/* Setup local timer */
2407	x86_init.timers.setup_percpu_clockev();
2408}
2409#endif
2410
2411/*
2412 * Power management
2413 */
2414#ifdef CONFIG_PM
2415
2416static struct {
2417	/*
2418	 * 'active' is true if the local APIC was enabled by us and
2419	 * not the BIOS; this signifies that we are also responsible
2420	 * for disabling it before entering apm/acpi suspend
2421	 */
2422	int active;
2423	/* r/w apic fields */
2424	unsigned int apic_id;
2425	unsigned int apic_taskpri;
2426	unsigned int apic_ldr;
2427	unsigned int apic_dfr;
2428	unsigned int apic_spiv;
2429	unsigned int apic_lvtt;
2430	unsigned int apic_lvtpc;
2431	unsigned int apic_lvt0;
2432	unsigned int apic_lvt1;
2433	unsigned int apic_lvterr;
2434	unsigned int apic_tmict;
2435	unsigned int apic_tdcr;
2436	unsigned int apic_thmr;
2437	unsigned int apic_cmci;
2438} apic_pm_state;
2439
2440static int lapic_suspend(void)
2441{
2442	unsigned long flags;
2443	int maxlvt;
2444
2445	if (!apic_pm_state.active)
2446		return 0;
2447
2448	maxlvt = lapic_get_maxlvt();
2449
2450	apic_pm_state.apic_id = apic_read(APIC_ID);
2451	apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2452	apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2453	apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2454	apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2455	apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2456	if (maxlvt >= 4)
2457		apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2458	apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2459	apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2460	apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2461	apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2462	apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2463#ifdef CONFIG_X86_THERMAL_VECTOR
2464	if (maxlvt >= 5)
2465		apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2466#endif
2467#ifdef CONFIG_X86_MCE_INTEL
2468	if (maxlvt >= 6)
2469		apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI);
2470#endif
2471
2472	local_irq_save(flags);
 
 
 
 
 
 
 
2473	disable_local_APIC();
2474
2475	irq_remapping_disable();
2476
2477	local_irq_restore(flags);
2478	return 0;
2479}
2480
2481static void lapic_resume(void)
2482{
2483	unsigned int l, h;
2484	unsigned long flags;
2485	int maxlvt;
2486
2487	if (!apic_pm_state.active)
2488		return;
2489
2490	local_irq_save(flags);
2491
2492	/*
2493	 * IO-APIC and PIC have their own resume routines.
2494	 * We just mask them here to make sure the interrupt
2495	 * subsystem is completely quiet while we enable x2apic
2496	 * and interrupt-remapping.
2497	 */
2498	mask_ioapic_entries();
2499	legacy_pic->mask_all();
2500
2501	if (x2apic_mode) {
2502		__x2apic_enable();
2503	} else {
2504		/*
2505		 * Make sure the APICBASE points to the right address
2506		 *
2507		 * FIXME! This will be wrong if we ever support suspend on
2508		 * SMP! We'll need to do this as part of the CPU restore!
2509		 */
2510		if (boot_cpu_data.x86 >= 6) {
2511			rdmsr(MSR_IA32_APICBASE, l, h);
2512			l &= ~MSR_IA32_APICBASE_BASE;
2513			l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2514			wrmsr(MSR_IA32_APICBASE, l, h);
2515		}
2516	}
2517
2518	maxlvt = lapic_get_maxlvt();
2519	apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2520	apic_write(APIC_ID, apic_pm_state.apic_id);
2521	apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2522	apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2523	apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2524	apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2525	apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2526	apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2527#ifdef CONFIG_X86_THERMAL_VECTOR
2528	if (maxlvt >= 5)
2529		apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2530#endif
2531#ifdef CONFIG_X86_MCE_INTEL
2532	if (maxlvt >= 6)
2533		apic_write(APIC_LVTCMCI, apic_pm_state.apic_cmci);
2534#endif
2535	if (maxlvt >= 4)
2536		apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2537	apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2538	apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2539	apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2540	apic_write(APIC_ESR, 0);
2541	apic_read(APIC_ESR);
2542	apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2543	apic_write(APIC_ESR, 0);
2544	apic_read(APIC_ESR);
2545
2546	irq_remapping_reenable(x2apic_mode);
2547
2548	local_irq_restore(flags);
2549}
2550
2551/*
2552 * This device has no shutdown method - fully functioning local APICs
2553 * are needed on every CPU up until machine_halt/restart/poweroff.
2554 */
2555
2556static struct syscore_ops lapic_syscore_ops = {
2557	.resume		= lapic_resume,
2558	.suspend	= lapic_suspend,
2559};
2560
2561static void apic_pm_activate(void)
2562{
2563	apic_pm_state.active = 1;
2564}
2565
2566static int __init init_lapic_sysfs(void)
2567{
2568	/* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2569	if (boot_cpu_has(X86_FEATURE_APIC))
2570		register_syscore_ops(&lapic_syscore_ops);
2571
2572	return 0;
2573}
2574
2575/* local apic needs to resume before other devices access its registers. */
2576core_initcall(init_lapic_sysfs);
2577
2578#else	/* CONFIG_PM */
2579
2580static void apic_pm_activate(void) { }
2581
2582#endif	/* CONFIG_PM */
2583
2584#ifdef CONFIG_X86_64
2585
2586static int multi_checked;
2587static int multi;
2588
2589static int set_multi(const struct dmi_system_id *d)
2590{
2591	if (multi)
2592		return 0;
2593	pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2594	multi = 1;
2595	return 0;
2596}
2597
2598static const struct dmi_system_id multi_dmi_table[] = {
2599	{
2600		.callback = set_multi,
2601		.ident = "IBM System Summit2",
2602		.matches = {
2603			DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2604			DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2605		},
2606	},
2607	{}
2608};
2609
2610static void dmi_check_multi(void)
2611{
2612	if (multi_checked)
2613		return;
2614
2615	dmi_check_system(multi_dmi_table);
2616	multi_checked = 1;
2617}
2618
2619/*
2620 * apic_is_clustered_box() -- Check if we can expect good TSC
2621 *
2622 * Thus far, the major user of this is IBM's Summit2 series:
2623 * Clustered boxes may have unsynced TSC problems if they are
2624 * multi-chassis.
2625 * Use DMI to check them
2626 */
2627int apic_is_clustered_box(void)
2628{
2629	dmi_check_multi();
2630	return multi;
2631}
2632#endif
2633
2634/*
2635 * APIC command line parameters
2636 */
2637static int __init setup_disableapic(char *arg)
2638{
2639	disable_apic = 1;
2640	setup_clear_cpu_cap(X86_FEATURE_APIC);
2641	return 0;
2642}
2643early_param("disableapic", setup_disableapic);
2644
2645/* same as disableapic, for compatibility */
2646static int __init setup_nolapic(char *arg)
2647{
2648	return setup_disableapic(arg);
2649}
2650early_param("nolapic", setup_nolapic);
2651
2652static int __init parse_lapic_timer_c2_ok(char *arg)
2653{
2654	local_apic_timer_c2_ok = 1;
2655	return 0;
2656}
2657early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2658
2659static int __init parse_disable_apic_timer(char *arg)
2660{
2661	disable_apic_timer = 1;
2662	return 0;
2663}
2664early_param("noapictimer", parse_disable_apic_timer);
2665
2666static int __init parse_nolapic_timer(char *arg)
2667{
2668	disable_apic_timer = 1;
2669	return 0;
2670}
2671early_param("nolapic_timer", parse_nolapic_timer);
2672
2673static int __init apic_set_verbosity(char *arg)
2674{
2675	if (!arg)  {
2676#ifdef CONFIG_X86_64
2677		skip_ioapic_setup = 0;
 
 
2678		return 0;
2679#endif
2680		return -EINVAL;
2681	}
2682
2683	if (strcmp("debug", arg) == 0)
2684		apic_verbosity = APIC_DEBUG;
2685	else if (strcmp("verbose", arg) == 0)
2686		apic_verbosity = APIC_VERBOSE;
2687#ifdef CONFIG_X86_64
2688	else {
2689		pr_warning("APIC Verbosity level %s not recognised"
2690			" use apic=verbose or apic=debug\n", arg);
2691		return -EINVAL;
2692	}
2693#endif
2694
2695	return 0;
2696}
2697early_param("apic", apic_set_verbosity);
2698
2699static int __init lapic_insert_resource(void)
2700{
2701	if (!apic_phys)
2702		return -1;
2703
2704	/* Put local APIC into the resource map. */
2705	lapic_resource.start = apic_phys;
2706	lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2707	insert_resource(&iomem_resource, &lapic_resource);
2708
2709	return 0;
2710}
2711
2712/*
2713 * need call insert after e820__reserve_resources()
2714 * that is using request_resource
2715 */
2716late_initcall(lapic_insert_resource);
2717
2718static int __init apic_set_disabled_cpu_apicid(char *arg)
2719{
2720	if (!arg || !get_option(&arg, &disabled_cpu_apicid))
2721		return -EINVAL;
2722
2723	return 0;
2724}
2725early_param("disable_cpu_apicid", apic_set_disabled_cpu_apicid);
2726
2727static int __init apic_set_extnmi(char *arg)
2728{
2729	if (!arg)
2730		return -EINVAL;
2731
2732	if (!strncmp("all", arg, 3))
2733		apic_extnmi = APIC_EXTNMI_ALL;
2734	else if (!strncmp("none", arg, 4))
2735		apic_extnmi = APIC_EXTNMI_NONE;
2736	else if (!strncmp("bsp", arg, 3))
2737		apic_extnmi = APIC_EXTNMI_BSP;
2738	else {
2739		pr_warn("Unknown external NMI delivery mode `%s' ignored\n", arg);
2740		return -EINVAL;
2741	}
2742
2743	return 0;
2744}
2745early_param("apic_extnmi", apic_set_extnmi);