Linux Audio

Check our new training course

Linux debugging, profiling, tracing and performance analysis training

Mar 24-27, 2025, special US time zones
Register
Loading...
v3.1
   1
   2/*
   3 * Local APIC virtualization
   4 *
   5 * Copyright (C) 2006 Qumranet, Inc.
   6 * Copyright (C) 2007 Novell
   7 * Copyright (C) 2007 Intel
   8 * Copyright 2009 Red Hat, Inc. and/or its affiliates.
   9 *
  10 * Authors:
  11 *   Dor Laor <dor.laor@qumranet.com>
  12 *   Gregory Haskins <ghaskins@novell.com>
  13 *   Yaozu (Eddie) Dong <eddie.dong@intel.com>
  14 *
  15 * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
  16 *
  17 * This work is licensed under the terms of the GNU GPL, version 2.  See
  18 * the COPYING file in the top-level directory.
  19 */
  20
  21#include <linux/kvm_host.h>
  22#include <linux/kvm.h>
  23#include <linux/mm.h>
  24#include <linux/highmem.h>
  25#include <linux/smp.h>
  26#include <linux/hrtimer.h>
  27#include <linux/io.h>
  28#include <linux/module.h>
  29#include <linux/math64.h>
  30#include <linux/slab.h>
  31#include <asm/processor.h>
  32#include <asm/msr.h>
  33#include <asm/page.h>
  34#include <asm/current.h>
  35#include <asm/apicdef.h>
 
  36#include <linux/atomic.h>
 
  37#include "kvm_cache_regs.h"
  38#include "irq.h"
  39#include "trace.h"
  40#include "x86.h"
 
 
  41
  42#ifndef CONFIG_X86_64
  43#define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
  44#else
  45#define mod_64(x, y) ((x) % (y))
  46#endif
  47
  48#define PRId64 "d"
  49#define PRIx64 "llx"
  50#define PRIu64 "u"
  51#define PRIo64 "o"
  52
  53#define APIC_BUS_CYCLE_NS 1
  54
  55/* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
  56#define apic_debug(fmt, arg...)
  57
  58#define APIC_LVT_NUM			6
  59/* 14 is the version for Xeon and Pentium 8.4.8*/
  60#define APIC_VERSION			(0x14UL | ((APIC_LVT_NUM - 1) << 16))
  61#define LAPIC_MMIO_LENGTH		(1 << 12)
  62/* followed define is not in apicdef.h */
  63#define APIC_SHORT_MASK			0xc0000
  64#define APIC_DEST_NOSHORT		0x0
  65#define APIC_DEST_MASK			0x800
  66#define MAX_APIC_VECTOR			256
 
  67
  68#define VEC_POS(v) ((v) & (32 - 1))
  69#define REG_POS(v) (((v) >> 5) << 4)
  70
  71static inline u32 apic_get_reg(struct kvm_lapic *apic, int reg_off)
  72{
  73	return *((u32 *) (apic->regs + reg_off));
  74}
  75
  76static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
  77{
  78	*((u32 *) (apic->regs + reg_off)) = val;
  79}
  80
  81static inline int apic_test_and_set_vector(int vec, void *bitmap)
  82{
  83	return test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
  84}
  85
  86static inline int apic_test_and_clear_vector(int vec, void *bitmap)
  87{
  88	return test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
  89}
  90
  91static inline void apic_set_vector(int vec, void *bitmap)
  92{
  93	set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
  94}
  95
  96static inline void apic_clear_vector(int vec, void *bitmap)
  97{
  98	clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
  99}
 100
 101static inline int apic_hw_enabled(struct kvm_lapic *apic)
 102{
 103	return (apic)->vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE;
 104}
 105
 106static inline int  apic_sw_enabled(struct kvm_lapic *apic)
 107{
 108	return apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED;
 109}
 110
 
 
 
 111static inline int apic_enabled(struct kvm_lapic *apic)
 112{
 113	return apic_sw_enabled(apic) &&	apic_hw_enabled(apic);
 114}
 115
 116#define LVT_MASK	\
 117	(APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
 118
 119#define LINT_MASK	\
 120	(LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
 121	 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
 122
 123static inline int kvm_apic_id(struct kvm_lapic *apic)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 124{
 125	return (apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 126}
 127
 128static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
 129{
 130	return !(apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
 131}
 132
 133static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
 134{
 135	return apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
 
 
 
 
 
 136}
 137
 138static inline int apic_lvtt_period(struct kvm_lapic *apic)
 139{
 140	return apic_get_reg(apic, APIC_LVTT) & APIC_LVT_TIMER_PERIODIC;
 
 
 
 
 
 141}
 142
 143static inline int apic_lvt_nmi_mode(u32 lvt_val)
 144{
 145	return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
 146}
 147
 148void kvm_apic_set_version(struct kvm_vcpu *vcpu)
 149{
 150	struct kvm_lapic *apic = vcpu->arch.apic;
 151	struct kvm_cpuid_entry2 *feat;
 152	u32 v = APIC_VERSION;
 153
 154	if (!irqchip_in_kernel(vcpu->kvm))
 155		return;
 156
 157	feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
 158	if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
 159		v |= APIC_LVR_DIRECTED_EOI;
 160	apic_set_reg(apic, APIC_LVR, v);
 161}
 162
 163static inline int apic_x2apic_mode(struct kvm_lapic *apic)
 164{
 165	return apic->vcpu->arch.apic_base & X2APIC_ENABLE;
 166}
 167
 168static unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
 169	LVT_MASK | APIC_LVT_TIMER_PERIODIC,	/* LVTT */
 170	LVT_MASK | APIC_MODE_MASK,	/* LVTTHMR */
 171	LVT_MASK | APIC_MODE_MASK,	/* LVTPC */
 172	LINT_MASK, LINT_MASK,	/* LVT0-1 */
 173	LVT_MASK		/* LVTERR */
 174};
 175
 176static int find_highest_vector(void *bitmap)
 177{
 178	u32 *word = bitmap;
 179	int word_offset = MAX_APIC_VECTOR >> 5;
 180
 181	while ((word_offset != 0) && (word[(--word_offset) << 2] == 0))
 182		continue;
 
 
 
 
 183
 184	if (likely(!word_offset && !word[0]))
 185		return -1;
 186	else
 187		return fls(word[word_offset << 2]) - 1 + (word_offset << 5);
 188}
 189
 190static inline int apic_test_and_set_irr(int vec, struct kvm_lapic *apic)
 191{
 192	apic->irr_pending = true;
 193	return apic_test_and_set_vector(vec, apic->regs + APIC_IRR);
 
 
 
 
 
 
 
 
 194}
 195
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 196static inline int apic_search_irr(struct kvm_lapic *apic)
 197{
 198	return find_highest_vector(apic->regs + APIC_IRR);
 199}
 200
 201static inline int apic_find_highest_irr(struct kvm_lapic *apic)
 202{
 203	int result;
 204
 
 
 
 
 205	if (!apic->irr_pending)
 206		return -1;
 207
 
 
 208	result = apic_search_irr(apic);
 209	ASSERT(result == -1 || result >= 16);
 210
 211	return result;
 212}
 213
 214static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
 215{
 216	apic->irr_pending = false;
 217	apic_clear_vector(vec, apic->regs + APIC_IRR);
 218	if (apic_search_irr(apic) != -1)
 219		apic->irr_pending = true;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 220}
 221
 222int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
 223{
 224	struct kvm_lapic *apic = vcpu->arch.apic;
 225	int highest_irr;
 226
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 227	/* This may race with setting of irr in __apic_accept_irq() and
 228	 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
 229	 * will cause vmexit immediately and the value will be recalculated
 230	 * on the next vmentry.
 231	 */
 232	if (!apic)
 233		return 0;
 234	highest_irr = apic_find_highest_irr(apic);
 235
 236	return highest_irr;
 237}
 238
 239static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
 240			     int vector, int level, int trig_mode);
 
 241
 242int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq)
 
 243{
 244	struct kvm_lapic *apic = vcpu->arch.apic;
 245
 246	return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
 247			irq->level, irq->trig_mode);
 248}
 249
 250static inline int apic_find_highest_isr(struct kvm_lapic *apic)
 251{
 252	int result;
 253
 254	result = find_highest_vector(apic->regs + APIC_ISR);
 255	ASSERT(result == -1 || result >= 16);
 
 256
 257	return result;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 258}
 259
 260static void apic_update_ppr(struct kvm_lapic *apic)
 261{
 262	u32 tpr, isrv, ppr, old_ppr;
 263	int isr;
 264
 265	old_ppr = apic_get_reg(apic, APIC_PROCPRI);
 266	tpr = apic_get_reg(apic, APIC_TASKPRI);
 267	isr = apic_find_highest_isr(apic);
 268	isrv = (isr != -1) ? isr : 0;
 269
 270	if ((tpr & 0xf0) >= (isrv & 0xf0))
 271		ppr = tpr & 0xff;
 272	else
 273		ppr = isrv & 0xf0;
 274
 275	apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
 276		   apic, ppr, isr, isrv);
 277
 278	if (old_ppr != ppr) {
 279		apic_set_reg(apic, APIC_PROCPRI, ppr);
 280		if (ppr < old_ppr)
 281			kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
 282	}
 283}
 284
 285static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
 286{
 287	apic_set_reg(apic, APIC_TASKPRI, tpr);
 288	apic_update_ppr(apic);
 289}
 290
 291int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
 292{
 293	return dest == 0xff || kvm_apic_id(apic) == dest;
 
 
 
 294}
 295
 296int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
 
 
 
 
 
 
 
 
 
 
 
 297{
 298	int result = 0;
 299	u32 logical_id;
 300
 301	if (apic_x2apic_mode(apic)) {
 302		logical_id = apic_get_reg(apic, APIC_LDR);
 303		return logical_id & mda;
 304	}
 
 
 
 
 305
 306	logical_id = GET_APIC_LOGICAL_ID(apic_get_reg(apic, APIC_LDR));
 
 307
 308	switch (apic_get_reg(apic, APIC_DFR)) {
 309	case APIC_DFR_FLAT:
 310		if (logical_id & mda)
 311			result = 1;
 312		break;
 313	case APIC_DFR_CLUSTER:
 314		if (((logical_id >> 4) == (mda >> 0x4))
 315		    && (logical_id & mda & 0xf))
 316			result = 1;
 317		break;
 318	default:
 319		printk(KERN_WARNING "Bad DFR vcpu %d: %08x\n",
 320		       apic->vcpu->vcpu_id, apic_get_reg(apic, APIC_DFR));
 321		break;
 322	}
 
 323
 324	return result;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 325}
 326
 327int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
 328			   int short_hand, int dest, int dest_mode)
 329{
 330	int result = 0;
 331	struct kvm_lapic *target = vcpu->arch.apic;
 
 332
 333	apic_debug("target %p, source %p, dest 0x%x, "
 334		   "dest_mode 0x%x, short_hand 0x%x\n",
 335		   target, source, dest, dest_mode, short_hand);
 336
 337	ASSERT(target);
 338	switch (short_hand) {
 339	case APIC_DEST_NOSHORT:
 340		if (dest_mode == 0)
 341			/* Physical mode. */
 342			result = kvm_apic_match_physical_addr(target, dest);
 343		else
 344			/* Logical mode. */
 345			result = kvm_apic_match_logical_addr(target, dest);
 346		break;
 347	case APIC_DEST_SELF:
 348		result = (target == source);
 349		break;
 350	case APIC_DEST_ALLINC:
 351		result = 1;
 352		break;
 353	case APIC_DEST_ALLBUT:
 354		result = (target != source);
 355		break;
 356	default:
 357		printk(KERN_WARNING "Bad dest shorthand value %x\n",
 358		       short_hand);
 359		break;
 360	}
 
 
 361
 362	return result;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 363}
 364
 365/*
 366 * Add a pending IRQ into lapic.
 367 * Return 1 if successfully added and 0 if discarded.
 368 */
 369static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
 370			     int vector, int level, int trig_mode)
 
 371{
 372	int result = 0;
 373	struct kvm_vcpu *vcpu = apic->vcpu;
 374
 
 
 375	switch (delivery_mode) {
 376	case APIC_DM_LOWEST:
 377		vcpu->arch.apic_arb_prio++;
 378	case APIC_DM_FIXED:
 
 
 
 379		/* FIXME add logic for vcpu on reset */
 380		if (unlikely(!apic_enabled(apic)))
 381			break;
 382
 383		if (trig_mode) {
 384			apic_debug("level trig mode for vector %d", vector);
 385			apic_set_vector(vector, apic->regs + APIC_TMR);
 386		} else
 387			apic_clear_vector(vector, apic->regs + APIC_TMR);
 388
 389		result = !apic_test_and_set_irr(vector, apic);
 390		trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
 391					  trig_mode, vector, !result);
 392		if (!result) {
 
 
 393			if (trig_mode)
 394				apic_debug("level trig mode repeatedly for "
 395						"vector %d", vector);
 396			break;
 397		}
 398
 399		kvm_make_request(KVM_REQ_EVENT, vcpu);
 400		kvm_vcpu_kick(vcpu);
 
 
 
 
 
 
 401		break;
 402
 403	case APIC_DM_REMRD:
 404		printk(KERN_DEBUG "Ignoring delivery mode 3\n");
 
 
 
 405		break;
 406
 407	case APIC_DM_SMI:
 408		printk(KERN_DEBUG "Ignoring guest SMI\n");
 
 
 409		break;
 410
 411	case APIC_DM_NMI:
 412		result = 1;
 413		kvm_inject_nmi(vcpu);
 414		kvm_vcpu_kick(vcpu);
 415		break;
 416
 417	case APIC_DM_INIT:
 418		if (level) {
 419			result = 1;
 420			vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
 
 
 
 
 421			kvm_make_request(KVM_REQ_EVENT, vcpu);
 422			kvm_vcpu_kick(vcpu);
 423		} else {
 424			apic_debug("Ignoring de-assert INIT to vcpu %d\n",
 425				   vcpu->vcpu_id);
 426		}
 427		break;
 428
 429	case APIC_DM_STARTUP:
 430		apic_debug("SIPI to vcpu %d vector 0x%02x\n",
 431			   vcpu->vcpu_id, vector);
 432		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
 433			result = 1;
 434			vcpu->arch.sipi_vector = vector;
 435			vcpu->arch.mp_state = KVM_MP_STATE_SIPI_RECEIVED;
 436			kvm_make_request(KVM_REQ_EVENT, vcpu);
 437			kvm_vcpu_kick(vcpu);
 438		}
 439		break;
 440
 441	case APIC_DM_EXTINT:
 442		/*
 443		 * Should only be called by kvm_apic_local_deliver() with LVT0,
 444		 * before NMI watchdog was enabled. Already handled by
 445		 * kvm_apic_accept_pic_intr().
 446		 */
 447		break;
 448
 449	default:
 450		printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
 451		       delivery_mode);
 452		break;
 453	}
 454	return result;
 455}
 456
 457int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
 458{
 459	return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
 460}
 461
 462static void apic_set_eoi(struct kvm_lapic *apic)
 
 
 
 
 
 463{
 464	int vector = apic_find_highest_isr(apic);
 465	int trigger_mode;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 466	/*
 467	 * Not every write EOI will has corresponding ISR,
 468	 * one example is when Kernel check timer on setup_IO_APIC
 469	 */
 470	if (vector == -1)
 471		return;
 472
 473	apic_clear_vector(vector, apic->regs + APIC_ISR);
 474	apic_update_ppr(apic);
 475
 476	if (apic_test_and_clear_vector(vector, apic->regs + APIC_TMR))
 477		trigger_mode = IOAPIC_LEVEL_TRIG;
 478	else
 479		trigger_mode = IOAPIC_EDGE_TRIG;
 480	if (!(apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI))
 481		kvm_ioapic_update_eoi(apic->vcpu->kvm, vector, trigger_mode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 482	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
 483}
 
 484
 485static void apic_send_ipi(struct kvm_lapic *apic)
 486{
 487	u32 icr_low = apic_get_reg(apic, APIC_ICR);
 488	u32 icr_high = apic_get_reg(apic, APIC_ICR2);
 489	struct kvm_lapic_irq irq;
 490
 491	irq.vector = icr_low & APIC_VECTOR_MASK;
 492	irq.delivery_mode = icr_low & APIC_MODE_MASK;
 493	irq.dest_mode = icr_low & APIC_DEST_MASK;
 494	irq.level = icr_low & APIC_INT_ASSERT;
 495	irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
 496	irq.shorthand = icr_low & APIC_SHORT_MASK;
 
 497	if (apic_x2apic_mode(apic))
 498		irq.dest_id = icr_high;
 499	else
 500		irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
 501
 502	trace_kvm_apic_ipi(icr_low, irq.dest_id);
 503
 504	apic_debug("icr_high 0x%x, icr_low 0x%x, "
 505		   "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
 506		   "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
 
 507		   icr_high, icr_low, irq.shorthand, irq.dest_id,
 508		   irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
 509		   irq.vector);
 510
 511	kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq);
 512}
 513
 514static u32 apic_get_tmcct(struct kvm_lapic *apic)
 515{
 516	ktime_t remaining;
 517	s64 ns;
 518	u32 tmcct;
 519
 520	ASSERT(apic != NULL);
 521
 522	/* if initial count is 0, current count should also be 0 */
 523	if (apic_get_reg(apic, APIC_TMICT) == 0)
 
 524		return 0;
 525
 526	remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
 
 527	if (ktime_to_ns(remaining) < 0)
 528		remaining = ktime_set(0, 0);
 529
 530	ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
 531	tmcct = div64_u64(ns,
 532			 (APIC_BUS_CYCLE_NS * apic->divide_count));
 533
 534	return tmcct;
 535}
 536
 537static void __report_tpr_access(struct kvm_lapic *apic, bool write)
 538{
 539	struct kvm_vcpu *vcpu = apic->vcpu;
 540	struct kvm_run *run = vcpu->run;
 541
 542	kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
 543	run->tpr_access.rip = kvm_rip_read(vcpu);
 544	run->tpr_access.is_write = write;
 545}
 546
 547static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
 548{
 549	if (apic->vcpu->arch.tpr_access_reporting)
 550		__report_tpr_access(apic, write);
 551}
 552
 553static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
 554{
 555	u32 val = 0;
 556
 557	if (offset >= LAPIC_MMIO_LENGTH)
 558		return 0;
 559
 560	switch (offset) {
 561	case APIC_ID:
 562		if (apic_x2apic_mode(apic))
 563			val = kvm_apic_id(apic);
 564		else
 565			val = kvm_apic_id(apic) << 24;
 566		break;
 567	case APIC_ARBPRI:
 568		printk(KERN_WARNING "Access APIC ARBPRI register "
 569		       "which is for P6\n");
 570		break;
 571
 572	case APIC_TMCCT:	/* Timer CCR */
 
 
 
 573		val = apic_get_tmcct(apic);
 574		break;
 575
 
 
 
 576	case APIC_TASKPRI:
 577		report_tpr_access(apic, false);
 578		/* fall thru */
 579	default:
 580		apic_update_ppr(apic);
 581		val = apic_get_reg(apic, offset);
 582		break;
 583	}
 584
 585	return val;
 586}
 587
 588static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
 589{
 590	return container_of(dev, struct kvm_lapic, dev);
 591}
 592
 593static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
 594		void *data)
 595{
 596	unsigned char alignment = offset & 0xf;
 597	u32 result;
 598	/* this bitmask has a bit cleared for each reserver register */
 599	static const u64 rmask = 0x43ff01ffffffe70cULL;
 600
 601	if ((alignment + len) > 4) {
 602		apic_debug("KVM_APIC_READ: alignment error %x %d\n",
 603			   offset, len);
 604		return 1;
 605	}
 606
 607	if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
 608		apic_debug("KVM_APIC_READ: read reserved register %x\n",
 609			   offset);
 610		return 1;
 611	}
 612
 613	result = __apic_read(apic, offset & ~0xf);
 614
 615	trace_kvm_apic_read(offset, result);
 616
 617	switch (len) {
 618	case 1:
 619	case 2:
 620	case 4:
 621		memcpy(data, (char *)&result + alignment, len);
 622		break;
 623	default:
 624		printk(KERN_ERR "Local APIC read with len = %x, "
 625		       "should be 1,2, or 4 instead\n", len);
 626		break;
 627	}
 628	return 0;
 629}
 
 630
 631static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
 632{
 633	return apic_hw_enabled(apic) &&
 634	    addr >= apic->base_address &&
 635	    addr < apic->base_address + LAPIC_MMIO_LENGTH;
 636}
 637
 638static int apic_mmio_read(struct kvm_io_device *this,
 639			   gpa_t address, int len, void *data)
 640{
 641	struct kvm_lapic *apic = to_lapic(this);
 642	u32 offset = address - apic->base_address;
 643
 644	if (!apic_mmio_in_range(apic, address))
 645		return -EOPNOTSUPP;
 646
 647	apic_reg_read(apic, offset, len, data);
 648
 649	return 0;
 650}
 651
 652static void update_divide_count(struct kvm_lapic *apic)
 653{
 654	u32 tmp1, tmp2, tdcr;
 655
 656	tdcr = apic_get_reg(apic, APIC_TDCR);
 657	tmp1 = tdcr & 0xf;
 658	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
 659	apic->divide_count = 0x1 << (tmp2 & 0x7);
 660
 661	apic_debug("timer divide count is 0x%x\n",
 662				   apic->divide_count);
 663}
 664
 665static void start_apic_timer(struct kvm_lapic *apic)
 666{
 667	ktime_t now = apic->lapic_timer.timer.base->get_time();
 
 668
 669	apic->lapic_timer.period = (u64)apic_get_reg(apic, APIC_TMICT) *
 670		    APIC_BUS_CYCLE_NS * apic->divide_count;
 671	atomic_set(&apic->lapic_timer.pending, 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 672
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 673	if (!apic->lapic_timer.period)
 674		return;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 675	/*
 676	 * Do not allow the guest to program periodic timers with small
 677	 * interval, since the hrtimers are not throttled by the host
 678	 * scheduler.
 679	 */
 680	if (apic_lvtt_period(apic)) {
 681		if (apic->lapic_timer.period < NSEC_PER_MSEC/2)
 682			apic->lapic_timer.period = NSEC_PER_MSEC/2;
 683	}
 684
 685	hrtimer_start(&apic->lapic_timer.timer,
 686		      ktime_add_ns(now, apic->lapic_timer.period),
 687		      HRTIMER_MODE_ABS);
 
 
 
 
 
 
 688
 689	apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
 690			   PRIx64 ", "
 691			   "timer initial count 0x%x, period %lldns, "
 692			   "expire @ 0x%016" PRIx64 ".\n", __func__,
 693			   APIC_BUS_CYCLE_NS, ktime_to_ns(now),
 694			   apic_get_reg(apic, APIC_TMICT),
 695			   apic->lapic_timer.period,
 696			   ktime_to_ns(ktime_add_ns(now,
 697					apic->lapic_timer.period)));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 698}
 699
 700static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
 701{
 702	int nmi_wd_enabled = apic_lvt_nmi_mode(apic_get_reg(apic, APIC_LVT0));
 703
 704	if (apic_lvt_nmi_mode(lvt0_val)) {
 705		if (!nmi_wd_enabled) {
 
 706			apic_debug("Receive NMI setting on APIC_LVT0 "
 707				   "for cpu %d\n", apic->vcpu->vcpu_id);
 708			apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
 709		}
 710	} else if (nmi_wd_enabled)
 711		apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
 712}
 713
 714static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
 715{
 716	int ret = 0;
 717
 718	trace_kvm_apic_write(reg, val);
 719
 720	switch (reg) {
 721	case APIC_ID:		/* Local APIC ID */
 722		if (!apic_x2apic_mode(apic))
 723			apic_set_reg(apic, APIC_ID, val);
 724		else
 725			ret = 1;
 726		break;
 727
 728	case APIC_TASKPRI:
 729		report_tpr_access(apic, true);
 730		apic_set_tpr(apic, val & 0xff);
 731		break;
 732
 733	case APIC_EOI:
 734		apic_set_eoi(apic);
 735		break;
 736
 737	case APIC_LDR:
 738		if (!apic_x2apic_mode(apic))
 739			apic_set_reg(apic, APIC_LDR, val & APIC_LDR_MASK);
 740		else
 741			ret = 1;
 742		break;
 743
 744	case APIC_DFR:
 745		if (!apic_x2apic_mode(apic))
 746			apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
 747		else
 
 748			ret = 1;
 749		break;
 750
 751	case APIC_SPIV: {
 752		u32 mask = 0x3ff;
 753		if (apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
 754			mask |= APIC_SPIV_DIRECTED_EOI;
 755		apic_set_reg(apic, APIC_SPIV, val & mask);
 756		if (!(val & APIC_SPIV_APIC_ENABLED)) {
 757			int i;
 758			u32 lvt_val;
 759
 760			for (i = 0; i < APIC_LVT_NUM; i++) {
 761				lvt_val = apic_get_reg(apic,
 762						       APIC_LVTT + 0x10 * i);
 763				apic_set_reg(apic, APIC_LVTT + 0x10 * i,
 764					     lvt_val | APIC_LVT_MASKED);
 765			}
 
 766			atomic_set(&apic->lapic_timer.pending, 0);
 767
 768		}
 769		break;
 770	}
 771	case APIC_ICR:
 772		/* No delay here, so we always clear the pending bit */
 773		apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
 774		apic_send_ipi(apic);
 775		break;
 776
 777	case APIC_ICR2:
 778		if (!apic_x2apic_mode(apic))
 779			val &= 0xff000000;
 780		apic_set_reg(apic, APIC_ICR2, val);
 781		break;
 782
 783	case APIC_LVT0:
 784		apic_manage_nmi_watchdog(apic, val);
 785	case APIC_LVTT:
 786	case APIC_LVTTHMR:
 787	case APIC_LVTPC:
 788	case APIC_LVT1:
 789	case APIC_LVTERR:
 790		/* TODO: Check vector */
 791		if (!apic_sw_enabled(apic))
 792			val |= APIC_LVT_MASKED;
 793
 794		val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
 795		apic_set_reg(apic, reg, val);
 
 
 796
 
 
 
 
 
 
 797		break;
 798
 799	case APIC_TMICT:
 
 
 
 800		hrtimer_cancel(&apic->lapic_timer.timer);
 801		apic_set_reg(apic, APIC_TMICT, val);
 802		start_apic_timer(apic);
 803		break;
 804
 805	case APIC_TDCR:
 806		if (val & 4)
 807			printk(KERN_ERR "KVM_WRITE:TDCR %x\n", val);
 808		apic_set_reg(apic, APIC_TDCR, val);
 809		update_divide_count(apic);
 810		break;
 811
 812	case APIC_ESR:
 813		if (apic_x2apic_mode(apic) && val != 0) {
 814			printk(KERN_ERR "KVM_WRITE:ESR not zero %x\n", val);
 815			ret = 1;
 816		}
 817		break;
 818
 819	case APIC_SELF_IPI:
 820		if (apic_x2apic_mode(apic)) {
 821			apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
 822		} else
 823			ret = 1;
 824		break;
 825	default:
 826		ret = 1;
 827		break;
 828	}
 829	if (ret)
 830		apic_debug("Local APIC Write to read-only register %x\n", reg);
 831	return ret;
 832}
 
 833
 834static int apic_mmio_write(struct kvm_io_device *this,
 835			    gpa_t address, int len, const void *data)
 836{
 837	struct kvm_lapic *apic = to_lapic(this);
 838	unsigned int offset = address - apic->base_address;
 839	u32 val;
 840
 841	if (!apic_mmio_in_range(apic, address))
 842		return -EOPNOTSUPP;
 843
 844	/*
 845	 * APIC register must be aligned on 128-bits boundary.
 846	 * 32/64/128 bits registers must be accessed thru 32 bits.
 847	 * Refer SDM 8.4.1
 848	 */
 849	if (len != 4 || (offset & 0xf)) {
 850		/* Don't shout loud, $infamous_os would cause only noise. */
 851		apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
 852		return 0;
 853	}
 854
 855	val = *(u32*)data;
 856
 857	/* too common printing */
 858	if (offset != APIC_EOI)
 859		apic_debug("%s: offset 0x%x with length 0x%x, and value is "
 860			   "0x%x\n", __func__, offset, len, val);
 861
 862	apic_reg_write(apic, offset & 0xff0, val);
 863
 864	return 0;
 865}
 866
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 867void kvm_free_lapic(struct kvm_vcpu *vcpu)
 868{
 
 
 869	if (!vcpu->arch.apic)
 870		return;
 871
 872	hrtimer_cancel(&vcpu->arch.apic->lapic_timer.timer);
 
 
 
 873
 874	if (vcpu->arch.apic->regs)
 875		free_page((unsigned long)vcpu->arch.apic->regs);
 876
 877	kfree(vcpu->arch.apic);
 
 
 
 878}
 879
 880/*
 881 *----------------------------------------------------------------------
 882 * LAPIC interface
 883 *----------------------------------------------------------------------
 884 */
 
 
 
 885
 886void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
 
 
 
 
 
 
 887{
 888	struct kvm_lapic *apic = vcpu->arch.apic;
 889
 890	if (!apic)
 
 
 
 
 
 
 
 
 
 
 
 
 891		return;
 
 
 
 
 
 
 
 
 
 
 892	apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
 893		     | (apic_get_reg(apic, APIC_TASKPRI) & 4));
 894}
 895
 896u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
 897{
 898	struct kvm_lapic *apic = vcpu->arch.apic;
 899	u64 tpr;
 900
 901	if (!apic)
 902		return 0;
 903	tpr = (u64) apic_get_reg(apic, APIC_TASKPRI);
 904
 905	return (tpr & 0xf0) >> 4;
 906}
 907
 908void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
 909{
 
 910	struct kvm_lapic *apic = vcpu->arch.apic;
 911
 912	if (!apic) {
 913		value |= MSR_IA32_APICBASE_BSP;
 914		vcpu->arch.apic_base = value;
 
 
 
 
 
 
 915		return;
 916	}
 917
 918	if (!kvm_vcpu_is_bsp(apic->vcpu))
 919		value &= ~MSR_IA32_APICBASE_BSP;
 
 
 
 
 
 
 
 
 920
 921	vcpu->arch.apic_base = value;
 922	if (apic_x2apic_mode(apic)) {
 923		u32 id = kvm_apic_id(apic);
 924		u32 ldr = ((id & ~0xf) << 16) | (1 << (id & 0xf));
 925		apic_set_reg(apic, APIC_LDR, ldr);
 
 926	}
 
 927	apic->base_address = apic->vcpu->arch.apic_base &
 928			     MSR_IA32_APICBASE_BASE;
 929
 
 
 
 
 930	/* with FSB delivery interrupt, we can restart APIC functionality */
 931	apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
 932		   "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
 933
 934}
 935
 936void kvm_lapic_reset(struct kvm_vcpu *vcpu)
 937{
 938	struct kvm_lapic *apic;
 939	int i;
 940
 941	apic_debug("%s\n", __func__);
 942
 943	ASSERT(vcpu);
 944	apic = vcpu->arch.apic;
 945	ASSERT(apic != NULL);
 946
 947	/* Stop the timer in case it's a reset to an active apic */
 948	hrtimer_cancel(&apic->lapic_timer.timer);
 949
 950	apic_set_reg(apic, APIC_ID, vcpu->vcpu_id << 24);
 
 
 
 
 951	kvm_apic_set_version(apic->vcpu);
 952
 953	for (i = 0; i < APIC_LVT_NUM; i++)
 954		apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
 955	apic_set_reg(apic, APIC_LVT0,
 956		     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
 957
 958	apic_set_reg(apic, APIC_DFR, 0xffffffffU);
 959	apic_set_reg(apic, APIC_SPIV, 0xff);
 960	apic_set_reg(apic, APIC_TASKPRI, 0);
 961	apic_set_reg(apic, APIC_LDR, 0);
 962	apic_set_reg(apic, APIC_ESR, 0);
 963	apic_set_reg(apic, APIC_ICR, 0);
 964	apic_set_reg(apic, APIC_ICR2, 0);
 965	apic_set_reg(apic, APIC_TDCR, 0);
 966	apic_set_reg(apic, APIC_TMICT, 0);
 
 
 
 
 967	for (i = 0; i < 8; i++) {
 968		apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
 969		apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
 970		apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
 971	}
 972	apic->irr_pending = false;
 
 
 973	update_divide_count(apic);
 974	atomic_set(&apic->lapic_timer.pending, 0);
 975	if (kvm_vcpu_is_bsp(vcpu))
 976		vcpu->arch.apic_base |= MSR_IA32_APICBASE_BSP;
 
 
 977	apic_update_ppr(apic);
 978
 979	vcpu->arch.apic_arb_prio = 0;
 
 980
 981	apic_debug(KERN_INFO "%s: vcpu=%p, id=%d, base_msr="
 982		   "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
 983		   vcpu, kvm_apic_id(apic),
 984		   vcpu->arch.apic_base, apic->base_address);
 985}
 986
 987bool kvm_apic_present(struct kvm_vcpu *vcpu)
 988{
 989	return vcpu->arch.apic && apic_hw_enabled(vcpu->arch.apic);
 990}
 991
 992int kvm_lapic_enabled(struct kvm_vcpu *vcpu)
 993{
 994	return kvm_apic_present(vcpu) && apic_sw_enabled(vcpu->arch.apic);
 995}
 996
 997/*
 998 *----------------------------------------------------------------------
 999 * timer interface
1000 *----------------------------------------------------------------------
1001 */
1002
1003static bool lapic_is_periodic(struct kvm_timer *ktimer)
1004{
1005	struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic,
1006					      lapic_timer);
1007	return apic_lvtt_period(apic);
1008}
1009
1010int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1011{
1012	struct kvm_lapic *lapic = vcpu->arch.apic;
1013
1014	if (lapic && apic_enabled(lapic) && apic_lvt_enabled(lapic, APIC_LVTT))
1015		return atomic_read(&lapic->lapic_timer.pending);
1016
1017	return 0;
1018}
1019
1020static int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1021{
1022	u32 reg = apic_get_reg(apic, lvt_type);
1023	int vector, mode, trig_mode;
1024
1025	if (apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1026		vector = reg & APIC_VECTOR_MASK;
1027		mode = reg & APIC_MODE_MASK;
1028		trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1029		return __apic_accept_irq(apic, mode, vector, 1, trig_mode);
 
1030	}
1031	return 0;
1032}
1033
1034void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1035{
1036	struct kvm_lapic *apic = vcpu->arch.apic;
1037
1038	if (apic)
1039		kvm_apic_local_deliver(apic, APIC_LVT0);
1040}
1041
1042static struct kvm_timer_ops lapic_timer_ops = {
1043	.is_periodic = lapic_is_periodic,
1044};
1045
1046static const struct kvm_io_device_ops apic_mmio_ops = {
1047	.read     = apic_mmio_read,
1048	.write    = apic_mmio_write,
1049};
1050
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1051int kvm_create_lapic(struct kvm_vcpu *vcpu)
1052{
1053	struct kvm_lapic *apic;
1054
1055	ASSERT(vcpu != NULL);
1056	apic_debug("apic_init %d\n", vcpu->vcpu_id);
1057
1058	apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1059	if (!apic)
1060		goto nomem;
1061
1062	vcpu->arch.apic = apic;
1063
1064	apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1065	if (!apic->regs) {
1066		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1067		       vcpu->vcpu_id);
1068		goto nomem_free_apic;
1069	}
1070	apic->vcpu = vcpu;
1071
1072	hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1073		     HRTIMER_MODE_ABS);
1074	apic->lapic_timer.timer.function = kvm_timer_fn;
1075	apic->lapic_timer.t_ops = &lapic_timer_ops;
1076	apic->lapic_timer.kvm = vcpu->kvm;
1077	apic->lapic_timer.vcpu = vcpu;
1078
1079	apic->base_address = APIC_DEFAULT_PHYS_BASE;
1080	vcpu->arch.apic_base = APIC_DEFAULT_PHYS_BASE;
1081
1082	kvm_lapic_reset(vcpu);
 
 
 
 
 
 
1083	kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1084
1085	return 0;
1086nomem_free_apic:
1087	kfree(apic);
1088nomem:
1089	return -ENOMEM;
1090}
1091
1092int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1093{
1094	struct kvm_lapic *apic = vcpu->arch.apic;
1095	int highest_irr;
1096
1097	if (!apic || !apic_enabled(apic))
1098		return -1;
1099
1100	apic_update_ppr(apic);
1101	highest_irr = apic_find_highest_irr(apic);
1102	if ((highest_irr == -1) ||
1103	    ((highest_irr & 0xF0) <= apic_get_reg(apic, APIC_PROCPRI)))
1104		return -1;
1105	return highest_irr;
1106}
1107
1108int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1109{
1110	u32 lvt0 = apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1111	int r = 0;
1112
1113	if (!apic_hw_enabled(vcpu->arch.apic))
1114		r = 1;
1115	if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1116	    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1117		r = 1;
1118	return r;
1119}
1120
1121void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1122{
1123	struct kvm_lapic *apic = vcpu->arch.apic;
1124
1125	if (apic && atomic_read(&apic->lapic_timer.pending) > 0) {
1126		if (kvm_apic_local_deliver(apic, APIC_LVTT))
1127			atomic_dec(&apic->lapic_timer.pending);
 
 
 
 
 
 
1128	}
1129}
1130
1131int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1132{
1133	int vector = kvm_apic_has_interrupt(vcpu);
1134	struct kvm_lapic *apic = vcpu->arch.apic;
1135
1136	if (vector == -1)
1137		return -1;
1138
1139	apic_set_vector(vector, apic->regs + APIC_ISR);
 
 
 
 
 
 
 
1140	apic_update_ppr(apic);
1141	apic_clear_irr(vector, apic);
 
 
 
 
 
 
1142	return vector;
1143}
1144
1145void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1146{
1147	struct kvm_lapic *apic = vcpu->arch.apic;
 
1148
1149	apic->base_address = vcpu->arch.apic_base &
1150			     MSR_IA32_APICBASE_BASE;
 
 
 
 
 
 
 
 
 
1151	kvm_apic_set_version(vcpu);
1152
1153	apic_update_ppr(apic);
1154	hrtimer_cancel(&apic->lapic_timer.timer);
 
 
1155	update_divide_count(apic);
1156	start_apic_timer(apic);
1157	apic->irr_pending = true;
 
 
 
 
 
 
 
 
 
 
 
1158	kvm_make_request(KVM_REQ_EVENT, vcpu);
 
 
 
 
 
 
1159}
1160
1161void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1162{
1163	struct kvm_lapic *apic = vcpu->arch.apic;
1164	struct hrtimer *timer;
1165
1166	if (!apic)
1167		return;
1168
1169	timer = &apic->lapic_timer.timer;
1170	if (hrtimer_cancel(timer))
1171		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1172}
1173
1174void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1175{
1176	u32 data;
1177	void *vapic;
1178
1179	if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
 
 
 
1180		return;
1181
1182	vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1183	data = *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr));
1184	kunmap_atomic(vapic, KM_USER0);
1185
1186	apic_set_tpr(vcpu->arch.apic, data & 0xff);
1187}
1188
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1189void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1190{
1191	u32 data, tpr;
1192	int max_irr, max_isr;
1193	struct kvm_lapic *apic;
1194	void *vapic;
 
1195
1196	if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1197		return;
1198
1199	apic = vcpu->arch.apic;
1200	tpr = apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1201	max_irr = apic_find_highest_irr(apic);
1202	if (max_irr < 0)
1203		max_irr = 0;
1204	max_isr = apic_find_highest_isr(apic);
1205	if (max_isr < 0)
1206		max_isr = 0;
1207	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1208
1209	vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1210	*(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr)) = data;
1211	kunmap_atomic(vapic, KM_USER0);
1212}
1213
1214void kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1215{
1216	if (!irqchip_in_kernel(vcpu->kvm))
1217		return;
 
 
 
 
 
 
 
1218
1219	vcpu->arch.apic->vapic_addr = vapic_addr;
 
1220}
1221
1222int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1223{
1224	struct kvm_lapic *apic = vcpu->arch.apic;
1225	u32 reg = (msr - APIC_BASE_MSR) << 4;
1226
1227	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
 
 
 
1228		return 1;
1229
1230	/* if this is ICR write vector before command */
1231	if (msr == 0x830)
1232		apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1233	return apic_reg_write(apic, reg, (u32)data);
1234}
1235
1236int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1237{
1238	struct kvm_lapic *apic = vcpu->arch.apic;
1239	u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1240
1241	if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1242		return 1;
1243
1244	if (apic_reg_read(apic, reg, 4, &low))
 
 
1245		return 1;
1246	if (msr == 0x830)
1247		apic_reg_read(apic, APIC_ICR2, 4, &high);
 
 
 
 
1248
1249	*data = (((u64)high) << 32) | low;
1250
1251	return 0;
1252}
1253
1254int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
1255{
1256	struct kvm_lapic *apic = vcpu->arch.apic;
1257
1258	if (!irqchip_in_kernel(vcpu->kvm))
1259		return 1;
1260
1261	/* if this is ICR write vector before command */
1262	if (reg == APIC_ICR)
1263		apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1264	return apic_reg_write(apic, reg, (u32)data);
1265}
1266
1267int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
1268{
1269	struct kvm_lapic *apic = vcpu->arch.apic;
1270	u32 low, high = 0;
1271
1272	if (!irqchip_in_kernel(vcpu->kvm))
1273		return 1;
1274
1275	if (apic_reg_read(apic, reg, 4, &low))
1276		return 1;
1277	if (reg == APIC_ICR)
1278		apic_reg_read(apic, APIC_ICR2, 4, &high);
1279
1280	*data = (((u64)high) << 32) | low;
1281
1282	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1283}
v4.10.11
   1
   2/*
   3 * Local APIC virtualization
   4 *
   5 * Copyright (C) 2006 Qumranet, Inc.
   6 * Copyright (C) 2007 Novell
   7 * Copyright (C) 2007 Intel
   8 * Copyright 2009 Red Hat, Inc. and/or its affiliates.
   9 *
  10 * Authors:
  11 *   Dor Laor <dor.laor@qumranet.com>
  12 *   Gregory Haskins <ghaskins@novell.com>
  13 *   Yaozu (Eddie) Dong <eddie.dong@intel.com>
  14 *
  15 * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
  16 *
  17 * This work is licensed under the terms of the GNU GPL, version 2.  See
  18 * the COPYING file in the top-level directory.
  19 */
  20
  21#include <linux/kvm_host.h>
  22#include <linux/kvm.h>
  23#include <linux/mm.h>
  24#include <linux/highmem.h>
  25#include <linux/smp.h>
  26#include <linux/hrtimer.h>
  27#include <linux/io.h>
  28#include <linux/export.h>
  29#include <linux/math64.h>
  30#include <linux/slab.h>
  31#include <asm/processor.h>
  32#include <asm/msr.h>
  33#include <asm/page.h>
  34#include <asm/current.h>
  35#include <asm/apicdef.h>
  36#include <asm/delay.h>
  37#include <linux/atomic.h>
  38#include <linux/jump_label.h>
  39#include "kvm_cache_regs.h"
  40#include "irq.h"
  41#include "trace.h"
  42#include "x86.h"
  43#include "cpuid.h"
  44#include "hyperv.h"
  45
  46#ifndef CONFIG_X86_64
  47#define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
  48#else
  49#define mod_64(x, y) ((x) % (y))
  50#endif
  51
  52#define PRId64 "d"
  53#define PRIx64 "llx"
  54#define PRIu64 "u"
  55#define PRIo64 "o"
  56
  57#define APIC_BUS_CYCLE_NS 1
  58
  59/* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
  60#define apic_debug(fmt, arg...)
  61
 
  62/* 14 is the version for Xeon and Pentium 8.4.8*/
  63#define APIC_VERSION			(0x14UL | ((KVM_APIC_LVT_NUM - 1) << 16))
  64#define LAPIC_MMIO_LENGTH		(1 << 12)
  65/* followed define is not in apicdef.h */
  66#define APIC_SHORT_MASK			0xc0000
  67#define APIC_DEST_NOSHORT		0x0
  68#define APIC_DEST_MASK			0x800
  69#define MAX_APIC_VECTOR			256
  70#define APIC_VECTORS_PER_REG		32
  71
  72#define APIC_BROADCAST			0xFF
  73#define X2APIC_BROADCAST		0xFFFFFFFFul
  74
  75static inline int apic_test_vector(int vec, void *bitmap)
  76{
  77	return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
  78}
  79
  80bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
  81{
  82	struct kvm_lapic *apic = vcpu->arch.apic;
 
 
 
 
 
 
 
 
 
 
 
  83
  84	return apic_test_vector(vector, apic->regs + APIC_ISR) ||
  85		apic_test_vector(vector, apic->regs + APIC_IRR);
 
  86}
  87
  88static inline void apic_clear_vector(int vec, void *bitmap)
  89{
  90	clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
  91}
  92
  93static inline int __apic_test_and_set_vector(int vec, void *bitmap)
  94{
  95	return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
  96}
  97
  98static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
  99{
 100	return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
 101}
 102
 103struct static_key_deferred apic_hw_disabled __read_mostly;
 104struct static_key_deferred apic_sw_disabled __read_mostly;
 105
 106static inline int apic_enabled(struct kvm_lapic *apic)
 107{
 108	return kvm_apic_sw_enabled(apic) &&	kvm_apic_hw_enabled(apic);
 109}
 110
 111#define LVT_MASK	\
 112	(APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
 113
 114#define LINT_MASK	\
 115	(LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
 116	 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
 117
 118static inline bool kvm_apic_map_get_logical_dest(struct kvm_apic_map *map,
 119		u32 dest_id, struct kvm_lapic ***cluster, u16 *mask) {
 120	switch (map->mode) {
 121	case KVM_APIC_MODE_X2APIC: {
 122		u32 offset = (dest_id >> 16) * 16;
 123		u32 max_apic_id = map->max_apic_id;
 124
 125		if (offset <= max_apic_id) {
 126			u8 cluster_size = min(max_apic_id - offset + 1, 16U);
 127
 128			*cluster = &map->phys_map[offset];
 129			*mask = dest_id & (0xffff >> (16 - cluster_size));
 130		} else {
 131			*mask = 0;
 132		}
 133
 134		return true;
 135		}
 136	case KVM_APIC_MODE_XAPIC_FLAT:
 137		*cluster = map->xapic_flat_map;
 138		*mask = dest_id & 0xff;
 139		return true;
 140	case KVM_APIC_MODE_XAPIC_CLUSTER:
 141		*cluster = map->xapic_cluster_map[(dest_id >> 4) & 0xf];
 142		*mask = dest_id & 0xf;
 143		return true;
 144	default:
 145		/* Not optimized. */
 146		return false;
 147	}
 148}
 149
 150static void kvm_apic_map_free(struct rcu_head *rcu)
 151{
 152	struct kvm_apic_map *map = container_of(rcu, struct kvm_apic_map, rcu);
 153
 154	kvfree(map);
 155}
 156
 157static void recalculate_apic_map(struct kvm *kvm)
 158{
 159	struct kvm_apic_map *new, *old = NULL;
 160	struct kvm_vcpu *vcpu;
 161	int i;
 162	u32 max_id = 255;
 163
 164	mutex_lock(&kvm->arch.apic_map_lock);
 165
 166	kvm_for_each_vcpu(i, vcpu, kvm)
 167		if (kvm_apic_present(vcpu))
 168			max_id = max(max_id, kvm_apic_id(vcpu->arch.apic));
 169
 170	new = kvm_kvzalloc(sizeof(struct kvm_apic_map) +
 171	                   sizeof(struct kvm_lapic *) * ((u64)max_id + 1));
 172
 173	if (!new)
 174		goto out;
 175
 176	new->max_apic_id = max_id;
 177
 178	kvm_for_each_vcpu(i, vcpu, kvm) {
 179		struct kvm_lapic *apic = vcpu->arch.apic;
 180		struct kvm_lapic **cluster;
 181		u16 mask;
 182		u32 ldr, aid;
 183
 184		if (!kvm_apic_present(vcpu))
 185			continue;
 186
 187		aid = kvm_apic_id(apic);
 188		ldr = kvm_lapic_get_reg(apic, APIC_LDR);
 189
 190		if (aid <= new->max_apic_id)
 191			new->phys_map[aid] = apic;
 192
 193		if (apic_x2apic_mode(apic)) {
 194			new->mode |= KVM_APIC_MODE_X2APIC;
 195		} else if (ldr) {
 196			ldr = GET_APIC_LOGICAL_ID(ldr);
 197			if (kvm_lapic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
 198				new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
 199			else
 200				new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
 201		}
 202
 203		if (!kvm_apic_map_get_logical_dest(new, ldr, &cluster, &mask))
 204			continue;
 205
 206		if (mask)
 207			cluster[ffs(mask) - 1] = apic;
 208	}
 209out:
 210	old = rcu_dereference_protected(kvm->arch.apic_map,
 211			lockdep_is_held(&kvm->arch.apic_map_lock));
 212	rcu_assign_pointer(kvm->arch.apic_map, new);
 213	mutex_unlock(&kvm->arch.apic_map_lock);
 214
 215	if (old)
 216		call_rcu(&old->rcu, kvm_apic_map_free);
 217
 218	kvm_make_scan_ioapic_request(kvm);
 219}
 220
 221static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
 222{
 223	bool enabled = val & APIC_SPIV_APIC_ENABLED;
 224
 225	kvm_lapic_set_reg(apic, APIC_SPIV, val);
 226
 227	if (enabled != apic->sw_enabled) {
 228		apic->sw_enabled = enabled;
 229		if (enabled) {
 230			static_key_slow_dec_deferred(&apic_sw_disabled);
 231			recalculate_apic_map(apic->vcpu->kvm);
 232		} else
 233			static_key_slow_inc(&apic_sw_disabled.key);
 234	}
 235}
 236
 237static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
 238{
 239	kvm_lapic_set_reg(apic, APIC_ID, id << 24);
 240	recalculate_apic_map(apic->vcpu->kvm);
 241}
 242
 243static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
 244{
 245	kvm_lapic_set_reg(apic, APIC_LDR, id);
 246	recalculate_apic_map(apic->vcpu->kvm);
 247}
 248
 249static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
 250{
 251	u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
 252
 253	kvm_lapic_set_reg(apic, APIC_ID, id);
 254	kvm_lapic_set_reg(apic, APIC_LDR, ldr);
 255	recalculate_apic_map(apic->vcpu->kvm);
 256}
 257
 258static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
 259{
 260	return !(kvm_lapic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
 261}
 262
 263static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
 264{
 265	return kvm_lapic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
 266}
 267
 268static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
 269{
 270	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
 271}
 272
 273static inline int apic_lvtt_period(struct kvm_lapic *apic)
 274{
 275	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
 276}
 277
 278static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
 279{
 280	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
 281}
 282
 283static inline int apic_lvt_nmi_mode(u32 lvt_val)
 284{
 285	return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
 286}
 287
 288void kvm_apic_set_version(struct kvm_vcpu *vcpu)
 289{
 290	struct kvm_lapic *apic = vcpu->arch.apic;
 291	struct kvm_cpuid_entry2 *feat;
 292	u32 v = APIC_VERSION;
 293
 294	if (!lapic_in_kernel(vcpu))
 295		return;
 296
 297	feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
 298	if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
 299		v |= APIC_LVR_DIRECTED_EOI;
 300	kvm_lapic_set_reg(apic, APIC_LVR, v);
 
 
 
 
 
 301}
 302
 303static const unsigned int apic_lvt_mask[KVM_APIC_LVT_NUM] = {
 304	LVT_MASK ,      /* part LVTT mask, timer mode mask added at runtime */
 305	LVT_MASK | APIC_MODE_MASK,	/* LVTTHMR */
 306	LVT_MASK | APIC_MODE_MASK,	/* LVTPC */
 307	LINT_MASK, LINT_MASK,	/* LVT0-1 */
 308	LVT_MASK		/* LVTERR */
 309};
 310
 311static int find_highest_vector(void *bitmap)
 312{
 313	int vec;
 314	u32 *reg;
 315
 316	for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
 317	     vec >= 0; vec -= APIC_VECTORS_PER_REG) {
 318		reg = bitmap + REG_POS(vec);
 319		if (*reg)
 320			return fls(*reg) - 1 + vec;
 321	}
 322
 323	return -1;
 
 
 
 324}
 325
 326static u8 count_vectors(void *bitmap)
 327{
 328	int vec;
 329	u32 *reg;
 330	u8 count = 0;
 331
 332	for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
 333		reg = bitmap + REG_POS(vec);
 334		count += hweight32(*reg);
 335	}
 336
 337	return count;
 338}
 339
 340void __kvm_apic_update_irr(u32 *pir, void *regs)
 341{
 342	u32 i, pir_val;
 343
 344	for (i = 0; i <= 7; i++) {
 345		pir_val = READ_ONCE(pir[i]);
 346		if (pir_val) {
 347			pir_val = xchg(&pir[i], 0);
 348			*((u32 *)(regs + APIC_IRR + i * 0x10)) |= pir_val;
 349		}
 350	}
 351}
 352EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
 353
 354void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
 355{
 356	struct kvm_lapic *apic = vcpu->arch.apic;
 357
 358	__kvm_apic_update_irr(pir, apic->regs);
 359
 360	kvm_make_request(KVM_REQ_EVENT, vcpu);
 361}
 362EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
 363
 364static inline int apic_search_irr(struct kvm_lapic *apic)
 365{
 366	return find_highest_vector(apic->regs + APIC_IRR);
 367}
 368
 369static inline int apic_find_highest_irr(struct kvm_lapic *apic)
 370{
 371	int result;
 372
 373	/*
 374	 * Note that irr_pending is just a hint. It will be always
 375	 * true with virtual interrupt delivery enabled.
 376	 */
 377	if (!apic->irr_pending)
 378		return -1;
 379
 380	if (apic->vcpu->arch.apicv_active)
 381		kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
 382	result = apic_search_irr(apic);
 383	ASSERT(result == -1 || result >= 16);
 384
 385	return result;
 386}
 387
 388static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
 389{
 390	struct kvm_vcpu *vcpu;
 391
 392	vcpu = apic->vcpu;
 393
 394	if (unlikely(vcpu->arch.apicv_active)) {
 395		/* try to update RVI */
 396		apic_clear_vector(vec, apic->regs + APIC_IRR);
 397		kvm_make_request(KVM_REQ_EVENT, vcpu);
 398	} else {
 399		apic->irr_pending = false;
 400		apic_clear_vector(vec, apic->regs + APIC_IRR);
 401		if (apic_search_irr(apic) != -1)
 402			apic->irr_pending = true;
 403	}
 404}
 405
 406static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
 407{
 408	struct kvm_vcpu *vcpu;
 409
 410	if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
 411		return;
 412
 413	vcpu = apic->vcpu;
 414
 415	/*
 416	 * With APIC virtualization enabled, all caching is disabled
 417	 * because the processor can modify ISR under the hood.  Instead
 418	 * just set SVI.
 419	 */
 420	if (unlikely(vcpu->arch.apicv_active))
 421		kvm_x86_ops->hwapic_isr_update(vcpu, vec);
 422	else {
 423		++apic->isr_count;
 424		BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
 425		/*
 426		 * ISR (in service register) bit is set when injecting an interrupt.
 427		 * The highest vector is injected. Thus the latest bit set matches
 428		 * the highest bit in ISR.
 429		 */
 430		apic->highest_isr_cache = vec;
 431	}
 432}
 433
 434static inline int apic_find_highest_isr(struct kvm_lapic *apic)
 435{
 436	int result;
 
 437
 438	/*
 439	 * Note that isr_count is always 1, and highest_isr_cache
 440	 * is always -1, with APIC virtualization enabled.
 441	 */
 442	if (!apic->isr_count)
 443		return -1;
 444	if (likely(apic->highest_isr_cache != -1))
 445		return apic->highest_isr_cache;
 446
 447	result = find_highest_vector(apic->regs + APIC_ISR);
 448	ASSERT(result == -1 || result >= 16);
 449
 450	return result;
 451}
 452
 453static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
 454{
 455	struct kvm_vcpu *vcpu;
 456	if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
 457		return;
 458
 459	vcpu = apic->vcpu;
 460
 461	/*
 462	 * We do get here for APIC virtualization enabled if the guest
 463	 * uses the Hyper-V APIC enlightenment.  In this case we may need
 464	 * to trigger a new interrupt delivery by writing the SVI field;
 465	 * on the other hand isr_count and highest_isr_cache are unused
 466	 * and must be left alone.
 467	 */
 468	if (unlikely(vcpu->arch.apicv_active))
 469		kvm_x86_ops->hwapic_isr_update(vcpu,
 470					       apic_find_highest_isr(apic));
 471	else {
 472		--apic->isr_count;
 473		BUG_ON(apic->isr_count < 0);
 474		apic->highest_isr_cache = -1;
 475	}
 476}
 477
 478int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
 479{
 480	/* This may race with setting of irr in __apic_accept_irq() and
 481	 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
 482	 * will cause vmexit immediately and the value will be recalculated
 483	 * on the next vmentry.
 484	 */
 485	return apic_find_highest_irr(vcpu->arch.apic);
 
 
 
 
 486}
 487
 488static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
 489			     int vector, int level, int trig_mode,
 490			     struct dest_map *dest_map);
 491
 492int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
 493		     struct dest_map *dest_map)
 494{
 495	struct kvm_lapic *apic = vcpu->arch.apic;
 496
 497	return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
 498			irq->level, irq->trig_mode, dest_map);
 499}
 500
 501static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
 502{
 
 503
 504	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
 505				      sizeof(val));
 506}
 507
 508static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
 509{
 510
 511	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
 512				      sizeof(*val));
 513}
 514
 515static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
 516{
 517	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
 518}
 519
 520static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
 521{
 522	u8 val;
 523	if (pv_eoi_get_user(vcpu, &val) < 0)
 524		apic_debug("Can't read EOI MSR value: 0x%llx\n",
 525			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
 526	return val & 0x1;
 527}
 528
 529static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
 530{
 531	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
 532		apic_debug("Can't set EOI MSR value: 0x%llx\n",
 533			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
 534		return;
 535	}
 536	__set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
 537}
 538
 539static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
 540{
 541	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
 542		apic_debug("Can't clear EOI MSR value: 0x%llx\n",
 543			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
 544		return;
 545	}
 546	__clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
 547}
 548
 549static void apic_update_ppr(struct kvm_lapic *apic)
 550{
 551	u32 tpr, isrv, ppr, old_ppr;
 552	int isr;
 553
 554	old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
 555	tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
 556	isr = apic_find_highest_isr(apic);
 557	isrv = (isr != -1) ? isr : 0;
 558
 559	if ((tpr & 0xf0) >= (isrv & 0xf0))
 560		ppr = tpr & 0xff;
 561	else
 562		ppr = isrv & 0xf0;
 563
 564	apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
 565		   apic, ppr, isr, isrv);
 566
 567	if (old_ppr != ppr) {
 568		kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
 569		if (ppr < old_ppr)
 570			kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
 571	}
 572}
 573
 574static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
 575{
 576	kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
 577	apic_update_ppr(apic);
 578}
 579
 580static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
 581{
 582	if (apic_x2apic_mode(apic))
 583		return mda == X2APIC_BROADCAST;
 584
 585	return GET_APIC_DEST_FIELD(mda) == APIC_BROADCAST;
 586}
 587
 588static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
 589{
 590	if (kvm_apic_broadcast(apic, mda))
 591		return true;
 592
 593	if (apic_x2apic_mode(apic))
 594		return mda == kvm_apic_id(apic);
 595
 596	return mda == SET_APIC_DEST_FIELD(kvm_apic_id(apic));
 597}
 598
 599static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
 600{
 
 601	u32 logical_id;
 602
 603	if (kvm_apic_broadcast(apic, mda))
 604		return true;
 605
 606	logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
 607
 608	if (apic_x2apic_mode(apic))
 609		return ((logical_id >> 16) == (mda >> 16))
 610		       && (logical_id & mda & 0xffff) != 0;
 611
 612	logical_id = GET_APIC_LOGICAL_ID(logical_id);
 613	mda = GET_APIC_DEST_FIELD(mda);
 614
 615	switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
 616	case APIC_DFR_FLAT:
 617		return (logical_id & mda) != 0;
 
 
 618	case APIC_DFR_CLUSTER:
 619		return ((logical_id >> 4) == (mda >> 4))
 620		       && (logical_id & mda & 0xf) != 0;
 
 
 621	default:
 622		apic_debug("Bad DFR vcpu %d: %08x\n",
 623			   apic->vcpu->vcpu_id, kvm_lapic_get_reg(apic, APIC_DFR));
 624		return false;
 625	}
 626}
 627
 628/* The KVM local APIC implementation has two quirks:
 629 *
 630 *  - the xAPIC MDA stores the destination at bits 24-31, while this
 631 *    is not true of struct kvm_lapic_irq's dest_id field.  This is
 632 *    just a quirk in the API and is not problematic.
 633 *
 634 *  - in-kernel IOAPIC messages have to be delivered directly to
 635 *    x2APIC, because the kernel does not support interrupt remapping.
 636 *    In order to support broadcast without interrupt remapping, x2APIC
 637 *    rewrites the destination of non-IPI messages from APIC_BROADCAST
 638 *    to X2APIC_BROADCAST.
 639 *
 640 * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API.  This is
 641 * important when userspace wants to use x2APIC-format MSIs, because
 642 * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
 643 */
 644static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
 645		struct kvm_lapic *source, struct kvm_lapic *target)
 646{
 647	bool ipi = source != NULL;
 648	bool x2apic_mda = apic_x2apic_mode(ipi ? source : target);
 649
 650	if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
 651	    !ipi && dest_id == APIC_BROADCAST && x2apic_mda)
 652		return X2APIC_BROADCAST;
 653
 654	return x2apic_mda ? dest_id : SET_APIC_DEST_FIELD(dest_id);
 655}
 656
 657bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
 658			   int short_hand, unsigned int dest, int dest_mode)
 659{
 
 660	struct kvm_lapic *target = vcpu->arch.apic;
 661	u32 mda = kvm_apic_mda(vcpu, dest, source, target);
 662
 663	apic_debug("target %p, source %p, dest 0x%x, "
 664		   "dest_mode 0x%x, short_hand 0x%x\n",
 665		   target, source, dest, dest_mode, short_hand);
 666
 667	ASSERT(target);
 668	switch (short_hand) {
 669	case APIC_DEST_NOSHORT:
 670		if (dest_mode == APIC_DEST_PHYSICAL)
 671			return kvm_apic_match_physical_addr(target, mda);
 
 672		else
 673			return kvm_apic_match_logical_addr(target, mda);
 
 
 674	case APIC_DEST_SELF:
 675		return target == source;
 
 676	case APIC_DEST_ALLINC:
 677		return true;
 
 678	case APIC_DEST_ALLBUT:
 679		return target != source;
 
 680	default:
 681		apic_debug("kvm: apic: Bad dest shorthand value %x\n",
 682			   short_hand);
 683		return false;
 684	}
 685}
 686EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
 687
 688int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
 689		       const unsigned long *bitmap, u32 bitmap_size)
 690{
 691	u32 mod;
 692	int i, idx = -1;
 693
 694	mod = vector % dest_vcpus;
 695
 696	for (i = 0; i <= mod; i++) {
 697		idx = find_next_bit(bitmap, bitmap_size, idx + 1);
 698		BUG_ON(idx == bitmap_size);
 699	}
 700
 701	return idx;
 702}
 703
 704static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
 705{
 706	if (!kvm->arch.disabled_lapic_found) {
 707		kvm->arch.disabled_lapic_found = true;
 708		printk(KERN_INFO
 709		       "Disabled LAPIC found during irq injection\n");
 710	}
 711}
 712
 713static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
 714		struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
 715{
 716	if (kvm->arch.x2apic_broadcast_quirk_disabled) {
 717		if ((irq->dest_id == APIC_BROADCAST &&
 718				map->mode != KVM_APIC_MODE_X2APIC))
 719			return true;
 720		if (irq->dest_id == X2APIC_BROADCAST)
 721			return true;
 722	} else {
 723		bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
 724		if (irq->dest_id == (x2apic_ipi ?
 725		                     X2APIC_BROADCAST : APIC_BROADCAST))
 726			return true;
 727	}
 728
 729	return false;
 730}
 731
 732/* Return true if the interrupt can be handled by using *bitmap as index mask
 733 * for valid destinations in *dst array.
 734 * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
 735 * Note: we may have zero kvm_lapic destinations when we return true, which
 736 * means that the interrupt should be dropped.  In this case, *bitmap would be
 737 * zero and *dst undefined.
 738 */
 739static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
 740		struct kvm_lapic **src, struct kvm_lapic_irq *irq,
 741		struct kvm_apic_map *map, struct kvm_lapic ***dst,
 742		unsigned long *bitmap)
 743{
 744	int i, lowest;
 745
 746	if (irq->shorthand == APIC_DEST_SELF && src) {
 747		*dst = src;
 748		*bitmap = 1;
 749		return true;
 750	} else if (irq->shorthand)
 751		return false;
 752
 753	if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
 754		return false;
 755
 756	if (irq->dest_mode == APIC_DEST_PHYSICAL) {
 757		if (irq->dest_id > map->max_apic_id) {
 758			*bitmap = 0;
 759		} else {
 760			*dst = &map->phys_map[irq->dest_id];
 761			*bitmap = 1;
 762		}
 763		return true;
 764	}
 765
 766	*bitmap = 0;
 767	if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
 768				(u16 *)bitmap))
 769		return false;
 770
 771	if (!kvm_lowest_prio_delivery(irq))
 772		return true;
 773
 774	if (!kvm_vector_hashing_enabled()) {
 775		lowest = -1;
 776		for_each_set_bit(i, bitmap, 16) {
 777			if (!(*dst)[i])
 778				continue;
 779			if (lowest < 0)
 780				lowest = i;
 781			else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
 782						(*dst)[lowest]->vcpu) < 0)
 783				lowest = i;
 784		}
 785	} else {
 786		if (!*bitmap)
 787			return true;
 788
 789		lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
 790				bitmap, 16);
 791
 792		if (!(*dst)[lowest]) {
 793			kvm_apic_disabled_lapic_found(kvm);
 794			*bitmap = 0;
 795			return true;
 796		}
 797	}
 798
 799	*bitmap = (lowest >= 0) ? 1 << lowest : 0;
 800
 801	return true;
 802}
 803
 804bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
 805		struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
 806{
 807	struct kvm_apic_map *map;
 808	unsigned long bitmap;
 809	struct kvm_lapic **dst = NULL;
 810	int i;
 811	bool ret;
 812
 813	*r = -1;
 814
 815	if (irq->shorthand == APIC_DEST_SELF) {
 816		*r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
 817		return true;
 818	}
 819
 820	rcu_read_lock();
 821	map = rcu_dereference(kvm->arch.apic_map);
 822
 823	ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
 824	if (ret)
 825		for_each_set_bit(i, &bitmap, 16) {
 826			if (!dst[i])
 827				continue;
 828			if (*r < 0)
 829				*r = 0;
 830			*r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
 831		}
 832
 833	rcu_read_unlock();
 834	return ret;
 835}
 836
 837/*
 838 * This routine tries to handler interrupts in posted mode, here is how
 839 * it deals with different cases:
 840 * - For single-destination interrupts, handle it in posted mode
 841 * - Else if vector hashing is enabled and it is a lowest-priority
 842 *   interrupt, handle it in posted mode and use the following mechanism
 843 *   to find the destinaiton vCPU.
 844 *	1. For lowest-priority interrupts, store all the possible
 845 *	   destination vCPUs in an array.
 846 *	2. Use "guest vector % max number of destination vCPUs" to find
 847 *	   the right destination vCPU in the array for the lowest-priority
 848 *	   interrupt.
 849 * - Otherwise, use remapped mode to inject the interrupt.
 850 */
 851bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
 852			struct kvm_vcpu **dest_vcpu)
 853{
 854	struct kvm_apic_map *map;
 855	unsigned long bitmap;
 856	struct kvm_lapic **dst = NULL;
 857	bool ret = false;
 858
 859	if (irq->shorthand)
 860		return false;
 861
 862	rcu_read_lock();
 863	map = rcu_dereference(kvm->arch.apic_map);
 864
 865	if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
 866			hweight16(bitmap) == 1) {
 867		unsigned long i = find_first_bit(&bitmap, 16);
 868
 869		if (dst[i]) {
 870			*dest_vcpu = dst[i]->vcpu;
 871			ret = true;
 872		}
 873	}
 874
 875	rcu_read_unlock();
 876	return ret;
 877}
 878
 879/*
 880 * Add a pending IRQ into lapic.
 881 * Return 1 if successfully added and 0 if discarded.
 882 */
 883static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
 884			     int vector, int level, int trig_mode,
 885			     struct dest_map *dest_map)
 886{
 887	int result = 0;
 888	struct kvm_vcpu *vcpu = apic->vcpu;
 889
 890	trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
 891				  trig_mode, vector);
 892	switch (delivery_mode) {
 893	case APIC_DM_LOWEST:
 894		vcpu->arch.apic_arb_prio++;
 895	case APIC_DM_FIXED:
 896		if (unlikely(trig_mode && !level))
 897			break;
 898
 899		/* FIXME add logic for vcpu on reset */
 900		if (unlikely(!apic_enabled(apic)))
 901			break;
 902
 903		result = 1;
 
 
 
 
 904
 905		if (dest_map) {
 906			__set_bit(vcpu->vcpu_id, dest_map->map);
 907			dest_map->vectors[vcpu->vcpu_id] = vector;
 908		}
 909
 910		if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
 911			if (trig_mode)
 912				kvm_lapic_set_vector(vector, apic->regs + APIC_TMR);
 913			else
 914				apic_clear_vector(vector, apic->regs + APIC_TMR);
 915		}
 916
 917		if (vcpu->arch.apicv_active)
 918			kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
 919		else {
 920			kvm_lapic_set_irr(vector, apic);
 921
 922			kvm_make_request(KVM_REQ_EVENT, vcpu);
 923			kvm_vcpu_kick(vcpu);
 924		}
 925		break;
 926
 927	case APIC_DM_REMRD:
 928		result = 1;
 929		vcpu->arch.pv.pv_unhalted = 1;
 930		kvm_make_request(KVM_REQ_EVENT, vcpu);
 931		kvm_vcpu_kick(vcpu);
 932		break;
 933
 934	case APIC_DM_SMI:
 935		result = 1;
 936		kvm_make_request(KVM_REQ_SMI, vcpu);
 937		kvm_vcpu_kick(vcpu);
 938		break;
 939
 940	case APIC_DM_NMI:
 941		result = 1;
 942		kvm_inject_nmi(vcpu);
 943		kvm_vcpu_kick(vcpu);
 944		break;
 945
 946	case APIC_DM_INIT:
 947		if (!trig_mode || level) {
 948			result = 1;
 949			/* assumes that there are only KVM_APIC_INIT/SIPI */
 950			apic->pending_events = (1UL << KVM_APIC_INIT);
 951			/* make sure pending_events is visible before sending
 952			 * the request */
 953			smp_wmb();
 954			kvm_make_request(KVM_REQ_EVENT, vcpu);
 955			kvm_vcpu_kick(vcpu);
 956		} else {
 957			apic_debug("Ignoring de-assert INIT to vcpu %d\n",
 958				   vcpu->vcpu_id);
 959		}
 960		break;
 961
 962	case APIC_DM_STARTUP:
 963		apic_debug("SIPI to vcpu %d vector 0x%02x\n",
 964			   vcpu->vcpu_id, vector);
 965		result = 1;
 966		apic->sipi_vector = vector;
 967		/* make sure sipi_vector is visible for the receiver */
 968		smp_wmb();
 969		set_bit(KVM_APIC_SIPI, &apic->pending_events);
 970		kvm_make_request(KVM_REQ_EVENT, vcpu);
 971		kvm_vcpu_kick(vcpu);
 972		break;
 973
 974	case APIC_DM_EXTINT:
 975		/*
 976		 * Should only be called by kvm_apic_local_deliver() with LVT0,
 977		 * before NMI watchdog was enabled. Already handled by
 978		 * kvm_apic_accept_pic_intr().
 979		 */
 980		break;
 981
 982	default:
 983		printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
 984		       delivery_mode);
 985		break;
 986	}
 987	return result;
 988}
 989
 990int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
 991{
 992	return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
 993}
 994
 995static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
 996{
 997	return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
 998}
 999
1000static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1001{
 
1002	int trigger_mode;
1003
1004	/* Eoi the ioapic only if the ioapic doesn't own the vector. */
1005	if (!kvm_ioapic_handles_vector(apic, vector))
1006		return;
1007
1008	/* Request a KVM exit to inform the userspace IOAPIC. */
1009	if (irqchip_split(apic->vcpu->kvm)) {
1010		apic->vcpu->arch.pending_ioapic_eoi = vector;
1011		kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1012		return;
1013	}
1014
1015	if (apic_test_vector(vector, apic->regs + APIC_TMR))
1016		trigger_mode = IOAPIC_LEVEL_TRIG;
1017	else
1018		trigger_mode = IOAPIC_EDGE_TRIG;
1019
1020	kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1021}
1022
1023static int apic_set_eoi(struct kvm_lapic *apic)
1024{
1025	int vector = apic_find_highest_isr(apic);
1026
1027	trace_kvm_eoi(apic, vector);
1028
1029	/*
1030	 * Not every write EOI will has corresponding ISR,
1031	 * one example is when Kernel check timer on setup_IO_APIC
1032	 */
1033	if (vector == -1)
1034		return vector;
1035
1036	apic_clear_isr(vector, apic);
1037	apic_update_ppr(apic);
1038
1039	if (test_bit(vector, vcpu_to_synic(apic->vcpu)->vec_bitmap))
1040		kvm_hv_synic_send_eoi(apic->vcpu, vector);
1041
1042	kvm_ioapic_send_eoi(apic, vector);
1043	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1044	return vector;
1045}
1046
1047/*
1048 * this interface assumes a trap-like exit, which has already finished
1049 * desired side effect including vISR and vPPR update.
1050 */
1051void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1052{
1053	struct kvm_lapic *apic = vcpu->arch.apic;
1054
1055	trace_kvm_eoi(apic, vector);
1056
1057	kvm_ioapic_send_eoi(apic, vector);
1058	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1059}
1060EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1061
1062static void apic_send_ipi(struct kvm_lapic *apic)
1063{
1064	u32 icr_low = kvm_lapic_get_reg(apic, APIC_ICR);
1065	u32 icr_high = kvm_lapic_get_reg(apic, APIC_ICR2);
1066	struct kvm_lapic_irq irq;
1067
1068	irq.vector = icr_low & APIC_VECTOR_MASK;
1069	irq.delivery_mode = icr_low & APIC_MODE_MASK;
1070	irq.dest_mode = icr_low & APIC_DEST_MASK;
1071	irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1072	irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1073	irq.shorthand = icr_low & APIC_SHORT_MASK;
1074	irq.msi_redir_hint = false;
1075	if (apic_x2apic_mode(apic))
1076		irq.dest_id = icr_high;
1077	else
1078		irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
1079
1080	trace_kvm_apic_ipi(icr_low, irq.dest_id);
1081
1082	apic_debug("icr_high 0x%x, icr_low 0x%x, "
1083		   "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
1084		   "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x, "
1085		   "msi_redir_hint 0x%x\n",
1086		   icr_high, icr_low, irq.shorthand, irq.dest_id,
1087		   irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
1088		   irq.vector, irq.msi_redir_hint);
1089
1090	kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1091}
1092
1093static u32 apic_get_tmcct(struct kvm_lapic *apic)
1094{
1095	ktime_t remaining, now;
1096	s64 ns;
1097	u32 tmcct;
1098
1099	ASSERT(apic != NULL);
1100
1101	/* if initial count is 0, current count should also be 0 */
1102	if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1103		apic->lapic_timer.period == 0)
1104		return 0;
1105
1106	now = ktime_get();
1107	remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1108	if (ktime_to_ns(remaining) < 0)
1109		remaining = 0;
1110
1111	ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1112	tmcct = div64_u64(ns,
1113			 (APIC_BUS_CYCLE_NS * apic->divide_count));
1114
1115	return tmcct;
1116}
1117
1118static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1119{
1120	struct kvm_vcpu *vcpu = apic->vcpu;
1121	struct kvm_run *run = vcpu->run;
1122
1123	kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1124	run->tpr_access.rip = kvm_rip_read(vcpu);
1125	run->tpr_access.is_write = write;
1126}
1127
1128static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1129{
1130	if (apic->vcpu->arch.tpr_access_reporting)
1131		__report_tpr_access(apic, write);
1132}
1133
1134static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1135{
1136	u32 val = 0;
1137
1138	if (offset >= LAPIC_MMIO_LENGTH)
1139		return 0;
1140
1141	switch (offset) {
 
 
 
 
 
 
1142	case APIC_ARBPRI:
1143		apic_debug("Access APIC ARBPRI register which is for P6\n");
 
1144		break;
1145
1146	case APIC_TMCCT:	/* Timer CCR */
1147		if (apic_lvtt_tscdeadline(apic))
1148			return 0;
1149
1150		val = apic_get_tmcct(apic);
1151		break;
1152	case APIC_PROCPRI:
1153		apic_update_ppr(apic);
1154		val = kvm_lapic_get_reg(apic, offset);
1155		break;
1156	case APIC_TASKPRI:
1157		report_tpr_access(apic, false);
1158		/* fall thru */
1159	default:
1160		val = kvm_lapic_get_reg(apic, offset);
 
1161		break;
1162	}
1163
1164	return val;
1165}
1166
1167static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1168{
1169	return container_of(dev, struct kvm_lapic, dev);
1170}
1171
1172int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1173		void *data)
1174{
1175	unsigned char alignment = offset & 0xf;
1176	u32 result;
1177	/* this bitmask has a bit cleared for each reserved register */
1178	static const u64 rmask = 0x43ff01ffffffe70cULL;
1179
1180	if ((alignment + len) > 4) {
1181		apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1182			   offset, len);
1183		return 1;
1184	}
1185
1186	if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1187		apic_debug("KVM_APIC_READ: read reserved register %x\n",
1188			   offset);
1189		return 1;
1190	}
1191
1192	result = __apic_read(apic, offset & ~0xf);
1193
1194	trace_kvm_apic_read(offset, result);
1195
1196	switch (len) {
1197	case 1:
1198	case 2:
1199	case 4:
1200		memcpy(data, (char *)&result + alignment, len);
1201		break;
1202	default:
1203		printk(KERN_ERR "Local APIC read with len = %x, "
1204		       "should be 1,2, or 4 instead\n", len);
1205		break;
1206	}
1207	return 0;
1208}
1209EXPORT_SYMBOL_GPL(kvm_lapic_reg_read);
1210
1211static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1212{
1213	return kvm_apic_hw_enabled(apic) &&
1214	    addr >= apic->base_address &&
1215	    addr < apic->base_address + LAPIC_MMIO_LENGTH;
1216}
1217
1218static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1219			   gpa_t address, int len, void *data)
1220{
1221	struct kvm_lapic *apic = to_lapic(this);
1222	u32 offset = address - apic->base_address;
1223
1224	if (!apic_mmio_in_range(apic, address))
1225		return -EOPNOTSUPP;
1226
1227	kvm_lapic_reg_read(apic, offset, len, data);
1228
1229	return 0;
1230}
1231
1232static void update_divide_count(struct kvm_lapic *apic)
1233{
1234	u32 tmp1, tmp2, tdcr;
1235
1236	tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1237	tmp1 = tdcr & 0xf;
1238	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1239	apic->divide_count = 0x1 << (tmp2 & 0x7);
1240
1241	apic_debug("timer divide count is 0x%x\n",
1242				   apic->divide_count);
1243}
1244
1245static void apic_update_lvtt(struct kvm_lapic *apic)
1246{
1247	u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1248			apic->lapic_timer.timer_mode_mask;
1249
1250	if (apic->lapic_timer.timer_mode != timer_mode) {
1251		apic->lapic_timer.timer_mode = timer_mode;
1252		hrtimer_cancel(&apic->lapic_timer.timer);
1253	}
1254}
1255
1256static void apic_timer_expired(struct kvm_lapic *apic)
1257{
1258	struct kvm_vcpu *vcpu = apic->vcpu;
1259	struct swait_queue_head *q = &vcpu->wq;
1260	struct kvm_timer *ktimer = &apic->lapic_timer;
1261
1262	if (atomic_read(&apic->lapic_timer.pending))
1263		return;
1264
1265	atomic_inc(&apic->lapic_timer.pending);
1266	kvm_set_pending_timer(vcpu);
1267
1268	if (swait_active(q))
1269		swake_up(q);
1270
1271	if (apic_lvtt_tscdeadline(apic))
1272		ktimer->expired_tscdeadline = ktimer->tscdeadline;
1273}
1274
1275/*
1276 * On APICv, this test will cause a busy wait
1277 * during a higher-priority task.
1278 */
1279
1280static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1281{
1282	struct kvm_lapic *apic = vcpu->arch.apic;
1283	u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1284
1285	if (kvm_apic_hw_enabled(apic)) {
1286		int vec = reg & APIC_VECTOR_MASK;
1287		void *bitmap = apic->regs + APIC_ISR;
1288
1289		if (vcpu->arch.apicv_active)
1290			bitmap = apic->regs + APIC_IRR;
1291
1292		if (apic_test_vector(vec, bitmap))
1293			return true;
1294	}
1295	return false;
1296}
1297
1298void wait_lapic_expire(struct kvm_vcpu *vcpu)
1299{
1300	struct kvm_lapic *apic = vcpu->arch.apic;
1301	u64 guest_tsc, tsc_deadline;
1302
1303	if (!lapic_in_kernel(vcpu))
1304		return;
1305
1306	if (apic->lapic_timer.expired_tscdeadline == 0)
1307		return;
1308
1309	if (!lapic_timer_int_injected(vcpu))
1310		return;
1311
1312	tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1313	apic->lapic_timer.expired_tscdeadline = 0;
1314	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1315	trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1316
1317	/* __delay is delay_tsc whenever the hardware has TSC, thus always.  */
1318	if (guest_tsc < tsc_deadline)
1319		__delay(min(tsc_deadline - guest_tsc,
1320			nsec_to_cycles(vcpu, lapic_timer_advance_ns)));
1321}
1322
1323static void start_sw_tscdeadline(struct kvm_lapic *apic)
1324{
1325	u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1326	u64 ns = 0;
1327	ktime_t expire;
1328	struct kvm_vcpu *vcpu = apic->vcpu;
1329	unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1330	unsigned long flags;
1331	ktime_t now;
1332
1333	if (unlikely(!tscdeadline || !this_tsc_khz))
1334		return;
1335
1336	local_irq_save(flags);
1337
1338	now = ktime_get();
1339	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1340	if (likely(tscdeadline > guest_tsc)) {
1341		ns = (tscdeadline - guest_tsc) * 1000000ULL;
1342		do_div(ns, this_tsc_khz);
1343		expire = ktime_add_ns(now, ns);
1344		expire = ktime_sub_ns(expire, lapic_timer_advance_ns);
1345		hrtimer_start(&apic->lapic_timer.timer,
1346				expire, HRTIMER_MODE_ABS_PINNED);
1347	} else
1348		apic_timer_expired(apic);
1349
1350	local_irq_restore(flags);
1351}
1352
1353static void start_sw_period(struct kvm_lapic *apic)
1354{
1355	if (!apic->lapic_timer.period)
1356		return;
1357
1358	if (apic_lvtt_oneshot(apic) &&
1359	    ktime_after(ktime_get(),
1360			apic->lapic_timer.target_expiration)) {
1361		apic_timer_expired(apic);
1362		return;
1363	}
1364
1365	hrtimer_start(&apic->lapic_timer.timer,
1366		apic->lapic_timer.target_expiration,
1367		HRTIMER_MODE_ABS_PINNED);
1368}
1369
1370static bool set_target_expiration(struct kvm_lapic *apic)
1371{
1372	ktime_t now;
1373	u64 tscl = rdtsc();
1374
1375	now = ktime_get();
1376	apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
1377		* APIC_BUS_CYCLE_NS * apic->divide_count;
1378
1379	if (!apic->lapic_timer.period)
1380		return false;
1381
1382	/*
1383	 * Do not allow the guest to program periodic timers with small
1384	 * interval, since the hrtimers are not throttled by the host
1385	 * scheduler.
1386	 */
1387	if (apic_lvtt_period(apic)) {
1388		s64 min_period = min_timer_period_us * 1000LL;
 
 
1389
1390		if (apic->lapic_timer.period < min_period) {
1391			pr_info_ratelimited(
1392			    "kvm: vcpu %i: requested %lld ns "
1393			    "lapic timer period limited to %lld ns\n",
1394			    apic->vcpu->vcpu_id,
1395			    apic->lapic_timer.period, min_period);
1396			apic->lapic_timer.period = min_period;
1397		}
1398	}
1399
1400	apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1401		   PRIx64 ", "
1402		   "timer initial count 0x%x, period %lldns, "
1403		   "expire @ 0x%016" PRIx64 ".\n", __func__,
1404		   APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1405		   kvm_lapic_get_reg(apic, APIC_TMICT),
1406		   apic->lapic_timer.period,
1407		   ktime_to_ns(ktime_add_ns(now,
1408				apic->lapic_timer.period)));
1409
1410	apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1411		nsec_to_cycles(apic->vcpu, apic->lapic_timer.period);
1412	apic->lapic_timer.target_expiration = ktime_add_ns(now, apic->lapic_timer.period);
1413
1414	return true;
1415}
1416
1417static void advance_periodic_target_expiration(struct kvm_lapic *apic)
1418{
1419	apic->lapic_timer.tscdeadline +=
1420		nsec_to_cycles(apic->vcpu, apic->lapic_timer.period);
1421	apic->lapic_timer.target_expiration =
1422		ktime_add_ns(apic->lapic_timer.target_expiration,
1423				apic->lapic_timer.period);
1424}
1425
1426bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1427{
1428	if (!lapic_in_kernel(vcpu))
1429		return false;
1430
1431	return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1432}
1433EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use);
1434
1435static void cancel_hv_timer(struct kvm_lapic *apic)
1436{
1437	kvm_x86_ops->cancel_hv_timer(apic->vcpu);
1438	apic->lapic_timer.hv_timer_in_use = false;
1439}
1440
1441static bool start_hv_timer(struct kvm_lapic *apic)
1442{
1443	u64 tscdeadline = apic->lapic_timer.tscdeadline;
1444
1445	if ((atomic_read(&apic->lapic_timer.pending) &&
1446		!apic_lvtt_period(apic)) ||
1447		kvm_x86_ops->set_hv_timer(apic->vcpu, tscdeadline)) {
1448		if (apic->lapic_timer.hv_timer_in_use)
1449			cancel_hv_timer(apic);
1450	} else {
1451		apic->lapic_timer.hv_timer_in_use = true;
1452		hrtimer_cancel(&apic->lapic_timer.timer);
1453
1454		/* In case the sw timer triggered in the window */
1455		if (atomic_read(&apic->lapic_timer.pending) &&
1456			!apic_lvtt_period(apic))
1457			cancel_hv_timer(apic);
1458	}
1459	trace_kvm_hv_timer_state(apic->vcpu->vcpu_id,
1460			apic->lapic_timer.hv_timer_in_use);
1461	return apic->lapic_timer.hv_timer_in_use;
1462}
1463
1464void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1465{
1466	struct kvm_lapic *apic = vcpu->arch.apic;
1467
1468	WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1469	WARN_ON(swait_active(&vcpu->wq));
1470	cancel_hv_timer(apic);
1471	apic_timer_expired(apic);
1472
1473	if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1474		advance_periodic_target_expiration(apic);
1475		if (!start_hv_timer(apic))
1476			start_sw_period(apic);
1477	}
1478}
1479EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
1480
1481void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
1482{
1483	struct kvm_lapic *apic = vcpu->arch.apic;
1484
1485	WARN_ON(apic->lapic_timer.hv_timer_in_use);
1486
1487	start_hv_timer(apic);
1488}
1489EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer);
1490
1491void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
1492{
1493	struct kvm_lapic *apic = vcpu->arch.apic;
1494
1495	/* Possibly the TSC deadline timer is not enabled yet */
1496	if (!apic->lapic_timer.hv_timer_in_use)
1497		return;
1498
1499	cancel_hv_timer(apic);
1500
1501	if (atomic_read(&apic->lapic_timer.pending))
1502		return;
1503
1504	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1505		start_sw_period(apic);
1506	else if (apic_lvtt_tscdeadline(apic))
1507		start_sw_tscdeadline(apic);
1508}
1509EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer);
1510
1511static void start_apic_timer(struct kvm_lapic *apic)
1512{
1513	atomic_set(&apic->lapic_timer.pending, 0);
1514
1515	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1516		if (set_target_expiration(apic) &&
1517			!(kvm_x86_ops->set_hv_timer && start_hv_timer(apic)))
1518			start_sw_period(apic);
1519	} else if (apic_lvtt_tscdeadline(apic)) {
1520		if (!(kvm_x86_ops->set_hv_timer && start_hv_timer(apic)))
1521			start_sw_tscdeadline(apic);
1522	}
1523}
1524
1525static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1526{
1527	bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
1528
1529	if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1530		apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1531		if (lvt0_in_nmi_mode) {
1532			apic_debug("Receive NMI setting on APIC_LVT0 "
1533				   "for cpu %d\n", apic->vcpu->vcpu_id);
1534			atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1535		} else
1536			atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1537	}
1538}
1539
1540int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1541{
1542	int ret = 0;
1543
1544	trace_kvm_apic_write(reg, val);
1545
1546	switch (reg) {
1547	case APIC_ID:		/* Local APIC ID */
1548		if (!apic_x2apic_mode(apic))
1549			kvm_apic_set_xapic_id(apic, val >> 24);
1550		else
1551			ret = 1;
1552		break;
1553
1554	case APIC_TASKPRI:
1555		report_tpr_access(apic, true);
1556		apic_set_tpr(apic, val & 0xff);
1557		break;
1558
1559	case APIC_EOI:
1560		apic_set_eoi(apic);
1561		break;
1562
1563	case APIC_LDR:
1564		if (!apic_x2apic_mode(apic))
1565			kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1566		else
1567			ret = 1;
1568		break;
1569
1570	case APIC_DFR:
1571		if (!apic_x2apic_mode(apic)) {
1572			kvm_lapic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1573			recalculate_apic_map(apic->vcpu->kvm);
1574		} else
1575			ret = 1;
1576		break;
1577
1578	case APIC_SPIV: {
1579		u32 mask = 0x3ff;
1580		if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1581			mask |= APIC_SPIV_DIRECTED_EOI;
1582		apic_set_spiv(apic, val & mask);
1583		if (!(val & APIC_SPIV_APIC_ENABLED)) {
1584			int i;
1585			u32 lvt_val;
1586
1587			for (i = 0; i < KVM_APIC_LVT_NUM; i++) {
1588				lvt_val = kvm_lapic_get_reg(apic,
1589						       APIC_LVTT + 0x10 * i);
1590				kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i,
1591					     lvt_val | APIC_LVT_MASKED);
1592			}
1593			apic_update_lvtt(apic);
1594			atomic_set(&apic->lapic_timer.pending, 0);
1595
1596		}
1597		break;
1598	}
1599	case APIC_ICR:
1600		/* No delay here, so we always clear the pending bit */
1601		kvm_lapic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1602		apic_send_ipi(apic);
1603		break;
1604
1605	case APIC_ICR2:
1606		if (!apic_x2apic_mode(apic))
1607			val &= 0xff000000;
1608		kvm_lapic_set_reg(apic, APIC_ICR2, val);
1609		break;
1610
1611	case APIC_LVT0:
1612		apic_manage_nmi_watchdog(apic, val);
 
1613	case APIC_LVTTHMR:
1614	case APIC_LVTPC:
1615	case APIC_LVT1:
1616	case APIC_LVTERR:
1617		/* TODO: Check vector */
1618		if (!kvm_apic_sw_enabled(apic))
1619			val |= APIC_LVT_MASKED;
1620
1621		val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1622		kvm_lapic_set_reg(apic, reg, val);
1623
1624		break;
1625
1626	case APIC_LVTT:
1627		if (!kvm_apic_sw_enabled(apic))
1628			val |= APIC_LVT_MASKED;
1629		val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1630		kvm_lapic_set_reg(apic, APIC_LVTT, val);
1631		apic_update_lvtt(apic);
1632		break;
1633
1634	case APIC_TMICT:
1635		if (apic_lvtt_tscdeadline(apic))
1636			break;
1637
1638		hrtimer_cancel(&apic->lapic_timer.timer);
1639		kvm_lapic_set_reg(apic, APIC_TMICT, val);
1640		start_apic_timer(apic);
1641		break;
1642
1643	case APIC_TDCR:
1644		if (val & 4)
1645			apic_debug("KVM_WRITE:TDCR %x\n", val);
1646		kvm_lapic_set_reg(apic, APIC_TDCR, val);
1647		update_divide_count(apic);
1648		break;
1649
1650	case APIC_ESR:
1651		if (apic_x2apic_mode(apic) && val != 0) {
1652			apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1653			ret = 1;
1654		}
1655		break;
1656
1657	case APIC_SELF_IPI:
1658		if (apic_x2apic_mode(apic)) {
1659			kvm_lapic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1660		} else
1661			ret = 1;
1662		break;
1663	default:
1664		ret = 1;
1665		break;
1666	}
1667	if (ret)
1668		apic_debug("Local APIC Write to read-only register %x\n", reg);
1669	return ret;
1670}
1671EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
1672
1673static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1674			    gpa_t address, int len, const void *data)
1675{
1676	struct kvm_lapic *apic = to_lapic(this);
1677	unsigned int offset = address - apic->base_address;
1678	u32 val;
1679
1680	if (!apic_mmio_in_range(apic, address))
1681		return -EOPNOTSUPP;
1682
1683	/*
1684	 * APIC register must be aligned on 128-bits boundary.
1685	 * 32/64/128 bits registers must be accessed thru 32 bits.
1686	 * Refer SDM 8.4.1
1687	 */
1688	if (len != 4 || (offset & 0xf)) {
1689		/* Don't shout loud, $infamous_os would cause only noise. */
1690		apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1691		return 0;
1692	}
1693
1694	val = *(u32*)data;
1695
1696	/* too common printing */
1697	if (offset != APIC_EOI)
1698		apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1699			   "0x%x\n", __func__, offset, len, val);
1700
1701	kvm_lapic_reg_write(apic, offset & 0xff0, val);
1702
1703	return 0;
1704}
1705
1706void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1707{
1708	kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1709}
1710EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1711
1712/* emulate APIC access in a trap manner */
1713void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1714{
1715	u32 val = 0;
1716
1717	/* hw has done the conditional check and inst decode */
1718	offset &= 0xff0;
1719
1720	kvm_lapic_reg_read(vcpu->arch.apic, offset, 4, &val);
1721
1722	/* TODO: optimize to just emulate side effect w/o one more write */
1723	kvm_lapic_reg_write(vcpu->arch.apic, offset, val);
1724}
1725EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1726
1727void kvm_free_lapic(struct kvm_vcpu *vcpu)
1728{
1729	struct kvm_lapic *apic = vcpu->arch.apic;
1730
1731	if (!vcpu->arch.apic)
1732		return;
1733
1734	hrtimer_cancel(&apic->lapic_timer.timer);
1735
1736	if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1737		static_key_slow_dec_deferred(&apic_hw_disabled);
1738
1739	if (!apic->sw_enabled)
1740		static_key_slow_dec_deferred(&apic_sw_disabled);
1741
1742	if (apic->regs)
1743		free_page((unsigned long)apic->regs);
1744
1745	kfree(apic);
1746}
1747
1748/*
1749 *----------------------------------------------------------------------
1750 * LAPIC interface
1751 *----------------------------------------------------------------------
1752 */
1753u64 kvm_get_lapic_target_expiration_tsc(struct kvm_vcpu *vcpu)
1754{
1755	struct kvm_lapic *apic = vcpu->arch.apic;
1756
1757	if (!lapic_in_kernel(vcpu))
1758		return 0;
1759
1760	return apic->lapic_timer.tscdeadline;
1761}
1762
1763u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1764{
1765	struct kvm_lapic *apic = vcpu->arch.apic;
1766
1767	if (!lapic_in_kernel(vcpu) ||
1768		!apic_lvtt_tscdeadline(apic))
1769		return 0;
1770
1771	return apic->lapic_timer.tscdeadline;
1772}
1773
1774void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1775{
1776	struct kvm_lapic *apic = vcpu->arch.apic;
1777
1778	if (!lapic_in_kernel(vcpu) || apic_lvtt_oneshot(apic) ||
1779			apic_lvtt_period(apic))
1780		return;
1781
1782	hrtimer_cancel(&apic->lapic_timer.timer);
1783	apic->lapic_timer.tscdeadline = data;
1784	start_apic_timer(apic);
1785}
1786
1787void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1788{
1789	struct kvm_lapic *apic = vcpu->arch.apic;
1790
1791	apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1792		     | (kvm_lapic_get_reg(apic, APIC_TASKPRI) & 4));
1793}
1794
1795u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1796{
 
1797	u64 tpr;
1798
1799	tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
 
 
1800
1801	return (tpr & 0xf0) >> 4;
1802}
1803
1804void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1805{
1806	u64 old_value = vcpu->arch.apic_base;
1807	struct kvm_lapic *apic = vcpu->arch.apic;
1808
1809	if (!apic)
1810		value |= MSR_IA32_APICBASE_BSP;
1811
1812	vcpu->arch.apic_base = value;
1813
1814	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
1815		kvm_update_cpuid(vcpu);
1816
1817	if (!apic)
1818		return;
 
1819
1820	/* update jump label if enable bit changes */
1821	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1822		if (value & MSR_IA32_APICBASE_ENABLE) {
1823			kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
1824			static_key_slow_dec_deferred(&apic_hw_disabled);
1825		} else {
1826			static_key_slow_inc(&apic_hw_disabled.key);
1827			recalculate_apic_map(vcpu->kvm);
1828		}
1829	}
1830
1831	if ((old_value ^ value) & X2APIC_ENABLE) {
1832		if (value & X2APIC_ENABLE) {
1833			kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
1834			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1835		} else
1836			kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1837	}
1838
1839	apic->base_address = apic->vcpu->arch.apic_base &
1840			     MSR_IA32_APICBASE_BASE;
1841
1842	if ((value & MSR_IA32_APICBASE_ENABLE) &&
1843	     apic->base_address != APIC_DEFAULT_PHYS_BASE)
1844		pr_warn_once("APIC base relocation is unsupported by KVM");
1845
1846	/* with FSB delivery interrupt, we can restart APIC functionality */
1847	apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1848		   "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1849
1850}
1851
1852void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
1853{
1854	struct kvm_lapic *apic;
1855	int i;
1856
1857	apic_debug("%s\n", __func__);
1858
1859	ASSERT(vcpu);
1860	apic = vcpu->arch.apic;
1861	ASSERT(apic != NULL);
1862
1863	/* Stop the timer in case it's a reset to an active apic */
1864	hrtimer_cancel(&apic->lapic_timer.timer);
1865
1866	if (!init_event) {
1867		kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
1868		                         MSR_IA32_APICBASE_ENABLE);
1869		kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
1870	}
1871	kvm_apic_set_version(apic->vcpu);
1872
1873	for (i = 0; i < KVM_APIC_LVT_NUM; i++)
1874		kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1875	apic_update_lvtt(apic);
1876	if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
1877		kvm_lapic_set_reg(apic, APIC_LVT0,
1878			     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1879	apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
1880
1881	kvm_lapic_set_reg(apic, APIC_DFR, 0xffffffffU);
1882	apic_set_spiv(apic, 0xff);
1883	kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
1884	if (!apic_x2apic_mode(apic))
1885		kvm_apic_set_ldr(apic, 0);
1886	kvm_lapic_set_reg(apic, APIC_ESR, 0);
1887	kvm_lapic_set_reg(apic, APIC_ICR, 0);
1888	kvm_lapic_set_reg(apic, APIC_ICR2, 0);
1889	kvm_lapic_set_reg(apic, APIC_TDCR, 0);
1890	kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1891	for (i = 0; i < 8; i++) {
1892		kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1893		kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1894		kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1895	}
1896	apic->irr_pending = vcpu->arch.apicv_active;
1897	apic->isr_count = vcpu->arch.apicv_active ? 1 : 0;
1898	apic->highest_isr_cache = -1;
1899	update_divide_count(apic);
1900	atomic_set(&apic->lapic_timer.pending, 0);
1901	if (kvm_vcpu_is_bsp(vcpu))
1902		kvm_lapic_set_base(vcpu,
1903				vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1904	vcpu->arch.pv_eoi.msr_val = 0;
1905	apic_update_ppr(apic);
1906
1907	vcpu->arch.apic_arb_prio = 0;
1908	vcpu->arch.apic_attention = 0;
1909
1910	apic_debug("%s: vcpu=%p, id=%d, base_msr="
1911		   "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1912		   vcpu, kvm_apic_id(apic),
1913		   vcpu->arch.apic_base, apic->base_address);
1914}
1915
 
 
 
 
 
 
 
 
 
 
1916/*
1917 *----------------------------------------------------------------------
1918 * timer interface
1919 *----------------------------------------------------------------------
1920 */
1921
1922static bool lapic_is_periodic(struct kvm_lapic *apic)
1923{
 
 
1924	return apic_lvtt_period(apic);
1925}
1926
1927int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1928{
1929	struct kvm_lapic *apic = vcpu->arch.apic;
1930
1931	if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
1932		return atomic_read(&apic->lapic_timer.pending);
1933
1934	return 0;
1935}
1936
1937int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1938{
1939	u32 reg = kvm_lapic_get_reg(apic, lvt_type);
1940	int vector, mode, trig_mode;
1941
1942	if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1943		vector = reg & APIC_VECTOR_MASK;
1944		mode = reg & APIC_MODE_MASK;
1945		trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1946		return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
1947					NULL);
1948	}
1949	return 0;
1950}
1951
1952void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1953{
1954	struct kvm_lapic *apic = vcpu->arch.apic;
1955
1956	if (apic)
1957		kvm_apic_local_deliver(apic, APIC_LVT0);
1958}
1959
 
 
 
 
1960static const struct kvm_io_device_ops apic_mmio_ops = {
1961	.read     = apic_mmio_read,
1962	.write    = apic_mmio_write,
1963};
1964
1965static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1966{
1967	struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1968	struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
1969
1970	apic_timer_expired(apic);
1971
1972	if (lapic_is_periodic(apic)) {
1973		advance_periodic_target_expiration(apic);
1974		hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1975		return HRTIMER_RESTART;
1976	} else
1977		return HRTIMER_NORESTART;
1978}
1979
1980int kvm_create_lapic(struct kvm_vcpu *vcpu)
1981{
1982	struct kvm_lapic *apic;
1983
1984	ASSERT(vcpu != NULL);
1985	apic_debug("apic_init %d\n", vcpu->vcpu_id);
1986
1987	apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1988	if (!apic)
1989		goto nomem;
1990
1991	vcpu->arch.apic = apic;
1992
1993	apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1994	if (!apic->regs) {
1995		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1996		       vcpu->vcpu_id);
1997		goto nomem_free_apic;
1998	}
1999	apic->vcpu = vcpu;
2000
2001	hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2002		     HRTIMER_MODE_ABS_PINNED);
2003	apic->lapic_timer.timer.function = apic_timer_fn;
 
 
 
 
 
 
2004
2005	/*
2006	 * APIC is created enabled. This will prevent kvm_lapic_set_base from
2007	 * thinking that APIC satet has changed.
2008	 */
2009	vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2010	static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2011	kvm_lapic_reset(vcpu, false);
2012	kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2013
2014	return 0;
2015nomem_free_apic:
2016	kfree(apic);
2017nomem:
2018	return -ENOMEM;
2019}
2020
2021int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2022{
2023	struct kvm_lapic *apic = vcpu->arch.apic;
2024	int highest_irr;
2025
2026	if (!apic_enabled(apic))
2027		return -1;
2028
2029	apic_update_ppr(apic);
2030	highest_irr = apic_find_highest_irr(apic);
2031	if ((highest_irr == -1) ||
2032	    ((highest_irr & 0xF0) <= kvm_lapic_get_reg(apic, APIC_PROCPRI)))
2033		return -1;
2034	return highest_irr;
2035}
2036
2037int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2038{
2039	u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2040	int r = 0;
2041
2042	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2043		r = 1;
2044	if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2045	    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2046		r = 1;
2047	return r;
2048}
2049
2050void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2051{
2052	struct kvm_lapic *apic = vcpu->arch.apic;
2053
2054	if (atomic_read(&apic->lapic_timer.pending) > 0) {
2055		kvm_apic_local_deliver(apic, APIC_LVTT);
2056		if (apic_lvtt_tscdeadline(apic))
2057			apic->lapic_timer.tscdeadline = 0;
2058		if (apic_lvtt_oneshot(apic)) {
2059			apic->lapic_timer.tscdeadline = 0;
2060			apic->lapic_timer.target_expiration = 0;
2061		}
2062		atomic_set(&apic->lapic_timer.pending, 0);
2063	}
2064}
2065
2066int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2067{
2068	int vector = kvm_apic_has_interrupt(vcpu);
2069	struct kvm_lapic *apic = vcpu->arch.apic;
2070
2071	if (vector == -1)
2072		return -1;
2073
2074	/*
2075	 * We get here even with APIC virtualization enabled, if doing
2076	 * nested virtualization and L1 runs with the "acknowledge interrupt
2077	 * on exit" mode.  Then we cannot inject the interrupt via RVI,
2078	 * because the process would deliver it through the IDT.
2079	 */
2080
2081	apic_set_isr(vector, apic);
2082	apic_update_ppr(apic);
2083	apic_clear_irr(vector, apic);
2084
2085	if (test_bit(vector, vcpu_to_synic(vcpu)->auto_eoi_bitmap)) {
2086		apic_clear_isr(vector, apic);
2087		apic_update_ppr(apic);
2088	}
2089
2090	return vector;
2091}
2092
2093static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2094		struct kvm_lapic_state *s, bool set)
2095{
2096	if (apic_x2apic_mode(vcpu->arch.apic)) {
2097		u32 *id = (u32 *)(s->regs + APIC_ID);
2098
2099		if (vcpu->kvm->arch.x2apic_format) {
2100			if (*id != vcpu->vcpu_id)
2101				return -EINVAL;
2102		} else {
2103			if (set)
2104				*id >>= 24;
2105			else
2106				*id <<= 24;
2107		}
2108	}
2109
2110	return 0;
2111}
2112
2113int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2114{
2115	memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2116	return kvm_apic_state_fixup(vcpu, s, false);
2117}
2118
2119int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2120{
2121	struct kvm_lapic *apic = vcpu->arch.apic;
2122	int r;
2123
2124
2125	kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2126	/* set SPIV separately to get count of SW disabled APICs right */
2127	apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2128
2129	r = kvm_apic_state_fixup(vcpu, s, true);
2130	if (r)
2131		return r;
2132	memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2133
2134	recalculate_apic_map(vcpu->kvm);
2135	kvm_apic_set_version(vcpu);
2136
2137	apic_update_ppr(apic);
2138	hrtimer_cancel(&apic->lapic_timer.timer);
2139	apic_update_lvtt(apic);
2140	apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2141	update_divide_count(apic);
2142	start_apic_timer(apic);
2143	apic->irr_pending = true;
2144	apic->isr_count = vcpu->arch.apicv_active ?
2145				1 : count_vectors(apic->regs + APIC_ISR);
2146	apic->highest_isr_cache = -1;
2147	if (vcpu->arch.apicv_active) {
2148		if (kvm_x86_ops->apicv_post_state_restore)
2149			kvm_x86_ops->apicv_post_state_restore(vcpu);
2150		kvm_x86_ops->hwapic_irr_update(vcpu,
2151				apic_find_highest_irr(apic));
2152		kvm_x86_ops->hwapic_isr_update(vcpu,
2153				apic_find_highest_isr(apic));
2154	}
2155	kvm_make_request(KVM_REQ_EVENT, vcpu);
2156	if (ioapic_in_kernel(vcpu->kvm))
2157		kvm_rtc_eoi_tracking_restore_one(vcpu);
2158
2159	vcpu->arch.apic_arb_prio = 0;
2160
2161	return 0;
2162}
2163
2164void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2165{
 
2166	struct hrtimer *timer;
2167
2168	if (!lapic_in_kernel(vcpu))
2169		return;
2170
2171	timer = &vcpu->arch.apic->lapic_timer.timer;
2172	if (hrtimer_cancel(timer))
2173		hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
2174}
2175
2176/*
2177 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2178 *
2179 * Detect whether guest triggered PV EOI since the
2180 * last entry. If yes, set EOI on guests's behalf.
2181 * Clear PV EOI in guest memory in any case.
2182 */
2183static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2184					struct kvm_lapic *apic)
2185{
2186	bool pending;
2187	int vector;
2188	/*
2189	 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2190	 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2191	 *
2192	 * KVM_APIC_PV_EOI_PENDING is unset:
2193	 * 	-> host disabled PV EOI.
2194	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2195	 * 	-> host enabled PV EOI, guest did not execute EOI yet.
2196	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2197	 * 	-> host enabled PV EOI, guest executed EOI.
2198	 */
2199	BUG_ON(!pv_eoi_enabled(vcpu));
2200	pending = pv_eoi_get_pending(vcpu);
2201	/*
2202	 * Clear pending bit in any case: it will be set again on vmentry.
2203	 * While this might not be ideal from performance point of view,
2204	 * this makes sure pv eoi is only enabled when we know it's safe.
2205	 */
2206	pv_eoi_clr_pending(vcpu);
2207	if (pending)
2208		return;
2209	vector = apic_set_eoi(apic);
2210	trace_kvm_pv_eoi(apic, vector);
2211}
2212
2213void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2214{
2215	u32 data;
 
2216
2217	if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2218		apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2219
2220	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2221		return;
2222
2223	if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2224				  sizeof(u32)))
2225		return;
2226
2227	apic_set_tpr(vcpu->arch.apic, data & 0xff);
2228}
2229
2230/*
2231 * apic_sync_pv_eoi_to_guest - called before vmentry
2232 *
2233 * Detect whether it's safe to enable PV EOI and
2234 * if yes do so.
2235 */
2236static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2237					struct kvm_lapic *apic)
2238{
2239	if (!pv_eoi_enabled(vcpu) ||
2240	    /* IRR set or many bits in ISR: could be nested. */
2241	    apic->irr_pending ||
2242	    /* Cache not set: could be safe but we don't bother. */
2243	    apic->highest_isr_cache == -1 ||
2244	    /* Need EOI to update ioapic. */
2245	    kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2246		/*
2247		 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2248		 * so we need not do anything here.
2249		 */
2250		return;
2251	}
2252
2253	pv_eoi_set_pending(apic->vcpu);
2254}
2255
2256void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2257{
2258	u32 data, tpr;
2259	int max_irr, max_isr;
2260	struct kvm_lapic *apic = vcpu->arch.apic;
2261
2262	apic_sync_pv_eoi_to_guest(vcpu, apic);
2263
2264	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2265		return;
2266
2267	tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
 
2268	max_irr = apic_find_highest_irr(apic);
2269	if (max_irr < 0)
2270		max_irr = 0;
2271	max_isr = apic_find_highest_isr(apic);
2272	if (max_isr < 0)
2273		max_isr = 0;
2274	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2275
2276	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2277				sizeof(u32));
 
2278}
2279
2280int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2281{
2282	if (vapic_addr) {
2283		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2284					&vcpu->arch.apic->vapic_cache,
2285					vapic_addr, sizeof(u32)))
2286			return -EINVAL;
2287		__set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2288	} else {
2289		__clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2290	}
2291
2292	vcpu->arch.apic->vapic_addr = vapic_addr;
2293	return 0;
2294}
2295
2296int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2297{
2298	struct kvm_lapic *apic = vcpu->arch.apic;
2299	u32 reg = (msr - APIC_BASE_MSR) << 4;
2300
2301	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2302		return 1;
2303
2304	if (reg == APIC_ICR2)
2305		return 1;
2306
2307	/* if this is ICR write vector before command */
2308	if (reg == APIC_ICR)
2309		kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2310	return kvm_lapic_reg_write(apic, reg, (u32)data);
2311}
2312
2313int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2314{
2315	struct kvm_lapic *apic = vcpu->arch.apic;
2316	u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2317
2318	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2319		return 1;
2320
2321	if (reg == APIC_DFR || reg == APIC_ICR2) {
2322		apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
2323			   reg);
2324		return 1;
2325	}
2326
2327	if (kvm_lapic_reg_read(apic, reg, 4, &low))
2328		return 1;
2329	if (reg == APIC_ICR)
2330		kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2331
2332	*data = (((u64)high) << 32) | low;
2333
2334	return 0;
2335}
2336
2337int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2338{
2339	struct kvm_lapic *apic = vcpu->arch.apic;
2340
2341	if (!lapic_in_kernel(vcpu))
2342		return 1;
2343
2344	/* if this is ICR write vector before command */
2345	if (reg == APIC_ICR)
2346		kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2347	return kvm_lapic_reg_write(apic, reg, (u32)data);
2348}
2349
2350int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2351{
2352	struct kvm_lapic *apic = vcpu->arch.apic;
2353	u32 low, high = 0;
2354
2355	if (!lapic_in_kernel(vcpu))
2356		return 1;
2357
2358	if (kvm_lapic_reg_read(apic, reg, 4, &low))
2359		return 1;
2360	if (reg == APIC_ICR)
2361		kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2362
2363	*data = (((u64)high) << 32) | low;
2364
2365	return 0;
2366}
2367
2368int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
2369{
2370	u64 addr = data & ~KVM_MSR_ENABLED;
2371	if (!IS_ALIGNED(addr, 4))
2372		return 1;
2373
2374	vcpu->arch.pv_eoi.msr_val = data;
2375	if (!pv_eoi_enabled(vcpu))
2376		return 0;
2377	return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
2378					 addr, sizeof(u8));
2379}
2380
2381void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2382{
2383	struct kvm_lapic *apic = vcpu->arch.apic;
2384	u8 sipi_vector;
2385	unsigned long pe;
2386
2387	if (!lapic_in_kernel(vcpu) || !apic->pending_events)
2388		return;
2389
2390	/*
2391	 * INITs are latched while in SMM.  Because an SMM CPU cannot
2392	 * be in KVM_MP_STATE_INIT_RECEIVED state, just eat SIPIs
2393	 * and delay processing of INIT until the next RSM.
2394	 */
2395	if (is_smm(vcpu)) {
2396		WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2397		if (test_bit(KVM_APIC_SIPI, &apic->pending_events))
2398			clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2399		return;
2400	}
2401
2402	pe = xchg(&apic->pending_events, 0);
2403	if (test_bit(KVM_APIC_INIT, &pe)) {
2404		kvm_lapic_reset(vcpu, true);
2405		kvm_vcpu_reset(vcpu, true);
2406		if (kvm_vcpu_is_bsp(apic->vcpu))
2407			vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2408		else
2409			vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2410	}
2411	if (test_bit(KVM_APIC_SIPI, &pe) &&
2412	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2413		/* evaluate pending_events before reading the vector */
2414		smp_rmb();
2415		sipi_vector = apic->sipi_vector;
2416		apic_debug("vcpu %d received sipi with vector # %x\n",
2417			 vcpu->vcpu_id, sipi_vector);
2418		kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2419		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2420	}
2421}
2422
2423void kvm_lapic_init(void)
2424{
2425	/* do not patch jump label more than once per second */
2426	jump_label_rate_limit(&apic_hw_disabled, HZ);
2427	jump_label_rate_limit(&apic_sw_disabled, HZ);
2428}
2429
2430void kvm_lapic_exit(void)
2431{
2432	static_key_deferred_flush(&apic_hw_disabled);
2433	static_key_deferred_flush(&apic_sw_disabled);
2434}