Linux Audio

Check our new training course

Linux kernel drivers training

Mar 31-Apr 9, 2025, special US time zones
Register
Loading...
v6.2
   1// SPDX-License-Identifier: GPL-2.0-only
   2
   3/*
   4 * Local APIC virtualization
   5 *
   6 * Copyright (C) 2006 Qumranet, Inc.
   7 * Copyright (C) 2007 Novell
   8 * Copyright (C) 2007 Intel
   9 * Copyright 2009 Red Hat, Inc. and/or its affiliates.
  10 *
  11 * Authors:
  12 *   Dor Laor <dor.laor@qumranet.com>
  13 *   Gregory Haskins <ghaskins@novell.com>
  14 *   Yaozu (Eddie) Dong <eddie.dong@intel.com>
  15 *
  16 * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
  17 */
  18
  19#include <linux/kvm_host.h>
  20#include <linux/kvm.h>
  21#include <linux/mm.h>
  22#include <linux/highmem.h>
  23#include <linux/smp.h>
  24#include <linux/hrtimer.h>
  25#include <linux/io.h>
  26#include <linux/export.h>
  27#include <linux/math64.h>
  28#include <linux/slab.h>
  29#include <asm/processor.h>
  30#include <asm/mce.h>
  31#include <asm/msr.h>
  32#include <asm/page.h>
  33#include <asm/current.h>
  34#include <asm/apicdef.h>
  35#include <asm/delay.h>
  36#include <linux/atomic.h>
  37#include <linux/jump_label.h>
  38#include "kvm_cache_regs.h"
  39#include "irq.h"
  40#include "ioapic.h"
  41#include "trace.h"
  42#include "x86.h"
  43#include "cpuid.h"
  44#include "hyperv.h"
  45#include "smm.h"
  46
  47#ifndef CONFIG_X86_64
  48#define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
  49#else
  50#define mod_64(x, y) ((x) % (y))
  51#endif
  52
  53#define PRId64 "d"
  54#define PRIx64 "llx"
  55#define PRIu64 "u"
  56#define PRIo64 "o"
  57
  58/* 14 is the version for Xeon and Pentium 8.4.8*/
  59#define APIC_VERSION			0x14UL
  60#define LAPIC_MMIO_LENGTH		(1 << 12)
  61/* followed define is not in apicdef.h */
  62#define MAX_APIC_VECTOR			256
  63#define APIC_VECTORS_PER_REG		32
  64
  65static bool lapic_timer_advance_dynamic __read_mostly;
  66#define LAPIC_TIMER_ADVANCE_ADJUST_MIN	100	/* clock cycles */
  67#define LAPIC_TIMER_ADVANCE_ADJUST_MAX	10000	/* clock cycles */
  68#define LAPIC_TIMER_ADVANCE_NS_INIT	1000
  69#define LAPIC_TIMER_ADVANCE_NS_MAX     5000
  70/* step-by-step approximation to mitigate fluctuation */
  71#define LAPIC_TIMER_ADVANCE_ADJUST_STEP 8
  72static int kvm_lapic_msr_read(struct kvm_lapic *apic, u32 reg, u64 *data);
  73static int kvm_lapic_msr_write(struct kvm_lapic *apic, u32 reg, u64 data);
  74
  75static inline void __kvm_lapic_set_reg(char *regs, int reg_off, u32 val)
  76{
  77	*((u32 *) (regs + reg_off)) = val;
  78}
  79
  80static inline void kvm_lapic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
  81{
  82	__kvm_lapic_set_reg(apic->regs, reg_off, val);
  83}
  84
  85static __always_inline u64 __kvm_lapic_get_reg64(char *regs, int reg)
  86{
  87	BUILD_BUG_ON(reg != APIC_ICR);
  88	return *((u64 *) (regs + reg));
  89}
  90
  91static __always_inline u64 kvm_lapic_get_reg64(struct kvm_lapic *apic, int reg)
  92{
  93	return __kvm_lapic_get_reg64(apic->regs, reg);
  94}
  95
  96static __always_inline void __kvm_lapic_set_reg64(char *regs, int reg, u64 val)
  97{
  98	BUILD_BUG_ON(reg != APIC_ICR);
  99	*((u64 *) (regs + reg)) = val;
 100}
 101
 102static __always_inline void kvm_lapic_set_reg64(struct kvm_lapic *apic,
 103						int reg, u64 val)
 104{
 105	__kvm_lapic_set_reg64(apic->regs, reg, val);
 106}
 107
 108static inline int apic_test_vector(int vec, void *bitmap)
 109{
 110	return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
 111}
 112
 113bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
 114{
 115	struct kvm_lapic *apic = vcpu->arch.apic;
 116
 117	return apic_test_vector(vector, apic->regs + APIC_ISR) ||
 118		apic_test_vector(vector, apic->regs + APIC_IRR);
 119}
 120
 121static inline int __apic_test_and_set_vector(int vec, void *bitmap)
 122{
 123	return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
 124}
 125
 126static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
 127{
 128	return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
 129}
 130
 131__read_mostly DEFINE_STATIC_KEY_DEFERRED_FALSE(apic_hw_disabled, HZ);
 132__read_mostly DEFINE_STATIC_KEY_DEFERRED_FALSE(apic_sw_disabled, HZ);
 133
 134static inline int apic_enabled(struct kvm_lapic *apic)
 135{
 136	return kvm_apic_sw_enabled(apic) &&	kvm_apic_hw_enabled(apic);
 137}
 138
 139#define LVT_MASK	\
 140	(APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
 141
 142#define LINT_MASK	\
 143	(LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
 144	 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
 145
 146static inline u32 kvm_x2apic_id(struct kvm_lapic *apic)
 147{
 148	return apic->vcpu->vcpu_id;
 149}
 150
 151static bool kvm_can_post_timer_interrupt(struct kvm_vcpu *vcpu)
 152{
 153	return pi_inject_timer && kvm_vcpu_apicv_active(vcpu) &&
 154		(kvm_mwait_in_guest(vcpu->kvm) || kvm_hlt_in_guest(vcpu->kvm));
 155}
 156
 157bool kvm_can_use_hv_timer(struct kvm_vcpu *vcpu)
 158{
 159	return kvm_x86_ops.set_hv_timer
 160	       && !(kvm_mwait_in_guest(vcpu->kvm) ||
 161		    kvm_can_post_timer_interrupt(vcpu));
 162}
 
 163
 164static bool kvm_use_posted_timer_interrupt(struct kvm_vcpu *vcpu)
 165{
 166	return kvm_can_post_timer_interrupt(vcpu) && vcpu->mode == IN_GUEST_MODE;
 167}
 168
 169static inline bool kvm_apic_map_get_logical_dest(struct kvm_apic_map *map,
 170		u32 dest_id, struct kvm_lapic ***cluster, u16 *mask) {
 171	switch (map->mode) {
 172	case KVM_APIC_MODE_X2APIC: {
 173		u32 offset = (dest_id >> 16) * 16;
 174		u32 max_apic_id = map->max_apic_id;
 175
 176		if (offset <= max_apic_id) {
 177			u8 cluster_size = min(max_apic_id - offset + 1, 16U);
 178
 179			offset = array_index_nospec(offset, map->max_apic_id + 1);
 180			*cluster = &map->phys_map[offset];
 181			*mask = dest_id & (0xffff >> (16 - cluster_size));
 182		} else {
 183			*mask = 0;
 184		}
 185
 186		return true;
 187		}
 188	case KVM_APIC_MODE_XAPIC_FLAT:
 189		*cluster = map->xapic_flat_map;
 190		*mask = dest_id & 0xff;
 191		return true;
 192	case KVM_APIC_MODE_XAPIC_CLUSTER:
 193		*cluster = map->xapic_cluster_map[(dest_id >> 4) & 0xf];
 194		*mask = dest_id & 0xf;
 195		return true;
 196	default:
 197		/* Not optimized. */
 198		return false;
 199	}
 200}
 201
 202static void kvm_apic_map_free(struct rcu_head *rcu)
 203{
 204	struct kvm_apic_map *map = container_of(rcu, struct kvm_apic_map, rcu);
 205
 206	kvfree(map);
 207}
 208
 209/*
 210 * CLEAN -> DIRTY and UPDATE_IN_PROGRESS -> DIRTY changes happen without a lock.
 211 *
 212 * DIRTY -> UPDATE_IN_PROGRESS and UPDATE_IN_PROGRESS -> CLEAN happen with
 213 * apic_map_lock_held.
 214 */
 215enum {
 216	CLEAN,
 217	UPDATE_IN_PROGRESS,
 218	DIRTY
 219};
 220
 221void kvm_recalculate_apic_map(struct kvm *kvm)
 222{
 223	struct kvm_apic_map *new, *old = NULL;
 224	struct kvm_vcpu *vcpu;
 225	unsigned long i;
 226	u32 max_id = 255; /* enough space for any xAPIC ID */
 227
 228	/* Read kvm->arch.apic_map_dirty before kvm->arch.apic_map.  */
 229	if (atomic_read_acquire(&kvm->arch.apic_map_dirty) == CLEAN)
 230		return;
 231
 232	WARN_ONCE(!irqchip_in_kernel(kvm),
 233		  "Dirty APIC map without an in-kernel local APIC");
 234
 235	mutex_lock(&kvm->arch.apic_map_lock);
 236	/*
 237	 * Read kvm->arch.apic_map_dirty before kvm->arch.apic_map
 238	 * (if clean) or the APIC registers (if dirty).
 239	 */
 240	if (atomic_cmpxchg_acquire(&kvm->arch.apic_map_dirty,
 241				   DIRTY, UPDATE_IN_PROGRESS) == CLEAN) {
 242		/* Someone else has updated the map. */
 243		mutex_unlock(&kvm->arch.apic_map_lock);
 244		return;
 245	}
 246
 247	kvm_for_each_vcpu(i, vcpu, kvm)
 248		if (kvm_apic_present(vcpu))
 249			max_id = max(max_id, kvm_x2apic_id(vcpu->arch.apic));
 250
 251	new = kvzalloc(sizeof(struct kvm_apic_map) +
 252	                   sizeof(struct kvm_lapic *) * ((u64)max_id + 1),
 253			   GFP_KERNEL_ACCOUNT);
 254
 255	if (!new)
 256		goto out;
 257
 258	new->max_apic_id = max_id;
 259
 260	kvm_for_each_vcpu(i, vcpu, kvm) {
 261		struct kvm_lapic *apic = vcpu->arch.apic;
 262		struct kvm_lapic **cluster;
 263		u16 mask;
 264		u32 ldr;
 265		u8 xapic_id;
 266		u32 x2apic_id;
 267
 268		if (!kvm_apic_present(vcpu))
 269			continue;
 270
 271		xapic_id = kvm_xapic_id(apic);
 272		x2apic_id = kvm_x2apic_id(apic);
 273
 274		/* Hotplug hack: see kvm_apic_match_physical_addr(), ... */
 275		if ((apic_x2apic_mode(apic) || x2apic_id > 0xff) &&
 276				x2apic_id <= new->max_apic_id)
 277			new->phys_map[x2apic_id] = apic;
 278		/*
 279		 * ... xAPIC ID of VCPUs with APIC ID > 0xff will wrap-around,
 280		 * prevent them from masking VCPUs with APIC ID <= 0xff.
 281		 */
 282		if (!apic_x2apic_mode(apic) && !new->phys_map[xapic_id])
 283			new->phys_map[xapic_id] = apic;
 284
 285		if (!kvm_apic_sw_enabled(apic))
 286			continue;
 287
 288		ldr = kvm_lapic_get_reg(apic, APIC_LDR);
 289
 290		if (apic_x2apic_mode(apic)) {
 291			new->mode |= KVM_APIC_MODE_X2APIC;
 292		} else if (ldr) {
 293			ldr = GET_APIC_LOGICAL_ID(ldr);
 294			if (kvm_lapic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
 295				new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
 296			else
 297				new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
 298		}
 299
 300		if (!kvm_apic_map_get_logical_dest(new, ldr, &cluster, &mask))
 301			continue;
 302
 303		if (mask)
 304			cluster[ffs(mask) - 1] = apic;
 305	}
 306out:
 307	old = rcu_dereference_protected(kvm->arch.apic_map,
 308			lockdep_is_held(&kvm->arch.apic_map_lock));
 309	rcu_assign_pointer(kvm->arch.apic_map, new);
 310	/*
 311	 * Write kvm->arch.apic_map before clearing apic->apic_map_dirty.
 312	 * If another update has come in, leave it DIRTY.
 313	 */
 314	atomic_cmpxchg_release(&kvm->arch.apic_map_dirty,
 315			       UPDATE_IN_PROGRESS, CLEAN);
 316	mutex_unlock(&kvm->arch.apic_map_lock);
 317
 318	if (old)
 319		call_rcu(&old->rcu, kvm_apic_map_free);
 320
 321	kvm_make_scan_ioapic_request(kvm);
 322}
 323
 324static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
 325{
 326	bool enabled = val & APIC_SPIV_APIC_ENABLED;
 327
 328	kvm_lapic_set_reg(apic, APIC_SPIV, val);
 329
 330	if (enabled != apic->sw_enabled) {
 331		apic->sw_enabled = enabled;
 332		if (enabled)
 333			static_branch_slow_dec_deferred(&apic_sw_disabled);
 334		else
 335			static_branch_inc(&apic_sw_disabled.key);
 336
 337		atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
 338	}
 339
 340	/* Check if there are APF page ready requests pending */
 341	if (enabled)
 342		kvm_make_request(KVM_REQ_APF_READY, apic->vcpu);
 343}
 344
 345static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
 346{
 347	kvm_lapic_set_reg(apic, APIC_ID, id << 24);
 348	atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
 349}
 350
 351static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
 352{
 353	kvm_lapic_set_reg(apic, APIC_LDR, id);
 354	atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
 355}
 356
 357static inline void kvm_apic_set_dfr(struct kvm_lapic *apic, u32 val)
 358{
 359	kvm_lapic_set_reg(apic, APIC_DFR, val);
 360	atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
 361}
 362
 363static inline u32 kvm_apic_calc_x2apic_ldr(u32 id)
 364{
 365	return ((id >> 4) << 16) | (1 << (id & 0xf));
 366}
 367
 368static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
 369{
 370	u32 ldr = kvm_apic_calc_x2apic_ldr(id);
 371
 372	WARN_ON_ONCE(id != apic->vcpu->vcpu_id);
 373
 374	kvm_lapic_set_reg(apic, APIC_ID, id);
 375	kvm_lapic_set_reg(apic, APIC_LDR, ldr);
 376	atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
 377}
 378
 379static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
 380{
 381	return !(kvm_lapic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
 382}
 383
 384static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
 385{
 386	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
 387}
 388
 389static inline int apic_lvtt_period(struct kvm_lapic *apic)
 390{
 391	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
 392}
 393
 394static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
 395{
 396	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
 397}
 398
 399static inline int apic_lvt_nmi_mode(u32 lvt_val)
 400{
 401	return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
 402}
 403
 404static inline bool kvm_lapic_lvt_supported(struct kvm_lapic *apic, int lvt_index)
 405{
 406	return apic->nr_lvt_entries > lvt_index;
 407}
 408
 409static inline int kvm_apic_calc_nr_lvt_entries(struct kvm_vcpu *vcpu)
 410{
 411	return KVM_APIC_MAX_NR_LVT_ENTRIES - !(vcpu->arch.mcg_cap & MCG_CMCI_P);
 412}
 413
 414void kvm_apic_set_version(struct kvm_vcpu *vcpu)
 415{
 416	struct kvm_lapic *apic = vcpu->arch.apic;
 417	u32 v = 0;
 418
 419	if (!lapic_in_kernel(vcpu))
 420		return;
 421
 422	v = APIC_VERSION | ((apic->nr_lvt_entries - 1) << 16);
 423
 424	/*
 425	 * KVM emulates 82093AA datasheet (with in-kernel IOAPIC implementation)
 426	 * which doesn't have EOI register; Some buggy OSes (e.g. Windows with
 427	 * Hyper-V role) disable EOI broadcast in lapic not checking for IOAPIC
 428	 * version first and level-triggered interrupts never get EOIed in
 429	 * IOAPIC.
 430	 */
 431	if (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) &&
 432	    !ioapic_in_kernel(vcpu->kvm))
 433		v |= APIC_LVR_DIRECTED_EOI;
 434	kvm_lapic_set_reg(apic, APIC_LVR, v);
 435}
 436
 437void kvm_apic_after_set_mcg_cap(struct kvm_vcpu *vcpu)
 438{
 439	int nr_lvt_entries = kvm_apic_calc_nr_lvt_entries(vcpu);
 440	struct kvm_lapic *apic = vcpu->arch.apic;
 441	int i;
 442
 443	if (!lapic_in_kernel(vcpu) || nr_lvt_entries == apic->nr_lvt_entries)
 444		return;
 445
 446	/* Initialize/mask any "new" LVT entries. */
 447	for (i = apic->nr_lvt_entries; i < nr_lvt_entries; i++)
 448		kvm_lapic_set_reg(apic, APIC_LVTx(i), APIC_LVT_MASKED);
 449
 450	apic->nr_lvt_entries = nr_lvt_entries;
 451
 452	/* The number of LVT entries is reflected in the version register. */
 453	kvm_apic_set_version(vcpu);
 454}
 455
 456static const unsigned int apic_lvt_mask[KVM_APIC_MAX_NR_LVT_ENTRIES] = {
 457	[LVT_TIMER] = LVT_MASK,      /* timer mode mask added at runtime */
 458	[LVT_THERMAL_MONITOR] = LVT_MASK | APIC_MODE_MASK,
 459	[LVT_PERFORMANCE_COUNTER] = LVT_MASK | APIC_MODE_MASK,
 460	[LVT_LINT0] = LINT_MASK,
 461	[LVT_LINT1] = LINT_MASK,
 462	[LVT_ERROR] = LVT_MASK,
 463	[LVT_CMCI] = LVT_MASK | APIC_MODE_MASK
 464};
 465
 466static int find_highest_vector(void *bitmap)
 467{
 468	int vec;
 469	u32 *reg;
 470
 471	for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
 472	     vec >= 0; vec -= APIC_VECTORS_PER_REG) {
 473		reg = bitmap + REG_POS(vec);
 474		if (*reg)
 475			return __fls(*reg) + vec;
 476	}
 477
 478	return -1;
 479}
 480
 481static u8 count_vectors(void *bitmap)
 482{
 483	int vec;
 484	u32 *reg;
 485	u8 count = 0;
 486
 487	for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
 488		reg = bitmap + REG_POS(vec);
 489		count += hweight32(*reg);
 490	}
 491
 492	return count;
 493}
 494
 495bool __kvm_apic_update_irr(u32 *pir, void *regs, int *max_irr)
 496{
 497	u32 i, vec;
 498	u32 pir_val, irr_val, prev_irr_val;
 499	int max_updated_irr;
 500
 501	max_updated_irr = -1;
 502	*max_irr = -1;
 503
 504	for (i = vec = 0; i <= 7; i++, vec += 32) {
 505		pir_val = READ_ONCE(pir[i]);
 506		irr_val = *((u32 *)(regs + APIC_IRR + i * 0x10));
 507		if (pir_val) {
 508			prev_irr_val = irr_val;
 509			irr_val |= xchg(&pir[i], 0);
 510			*((u32 *)(regs + APIC_IRR + i * 0x10)) = irr_val;
 511			if (prev_irr_val != irr_val) {
 512				max_updated_irr =
 513					__fls(irr_val ^ prev_irr_val) + vec;
 514			}
 515		}
 516		if (irr_val)
 517			*max_irr = __fls(irr_val) + vec;
 518	}
 519
 520	return ((max_updated_irr != -1) &&
 521		(max_updated_irr == *max_irr));
 522}
 523EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
 524
 525bool kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir, int *max_irr)
 526{
 527	struct kvm_lapic *apic = vcpu->arch.apic;
 528
 529	return __kvm_apic_update_irr(pir, apic->regs, max_irr);
 530}
 531EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
 532
 533static inline int apic_search_irr(struct kvm_lapic *apic)
 534{
 535	return find_highest_vector(apic->regs + APIC_IRR);
 536}
 537
 538static inline int apic_find_highest_irr(struct kvm_lapic *apic)
 539{
 540	int result;
 541
 542	/*
 543	 * Note that irr_pending is just a hint. It will be always
 544	 * true with virtual interrupt delivery enabled.
 545	 */
 546	if (!apic->irr_pending)
 547		return -1;
 548
 549	result = apic_search_irr(apic);
 550	ASSERT(result == -1 || result >= 16);
 551
 552	return result;
 553}
 554
 555static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
 556{
 557	if (unlikely(apic->apicv_active)) {
 
 
 
 
 558		/* need to update RVI */
 559		kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
 560		static_call_cond(kvm_x86_hwapic_irr_update)(apic->vcpu,
 561							    apic_find_highest_irr(apic));
 562	} else {
 563		apic->irr_pending = false;
 564		kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
 565		if (apic_search_irr(apic) != -1)
 566			apic->irr_pending = true;
 567	}
 568}
 569
 570void kvm_apic_clear_irr(struct kvm_vcpu *vcpu, int vec)
 571{
 572	apic_clear_irr(vec, vcpu->arch.apic);
 573}
 574EXPORT_SYMBOL_GPL(kvm_apic_clear_irr);
 575
 576static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
 577{
 
 
 578	if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
 579		return;
 580
 
 
 581	/*
 582	 * With APIC virtualization enabled, all caching is disabled
 583	 * because the processor can modify ISR under the hood.  Instead
 584	 * just set SVI.
 585	 */
 586	if (unlikely(apic->apicv_active))
 587		static_call_cond(kvm_x86_hwapic_isr_update)(vec);
 588	else {
 589		++apic->isr_count;
 590		BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
 591		/*
 592		 * ISR (in service register) bit is set when injecting an interrupt.
 593		 * The highest vector is injected. Thus the latest bit set matches
 594		 * the highest bit in ISR.
 595		 */
 596		apic->highest_isr_cache = vec;
 597	}
 598}
 599
 600static inline int apic_find_highest_isr(struct kvm_lapic *apic)
 601{
 602	int result;
 603
 604	/*
 605	 * Note that isr_count is always 1, and highest_isr_cache
 606	 * is always -1, with APIC virtualization enabled.
 607	 */
 608	if (!apic->isr_count)
 609		return -1;
 610	if (likely(apic->highest_isr_cache != -1))
 611		return apic->highest_isr_cache;
 612
 613	result = find_highest_vector(apic->regs + APIC_ISR);
 614	ASSERT(result == -1 || result >= 16);
 615
 616	return result;
 617}
 618
 619static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
 620{
 
 621	if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
 622		return;
 623
 
 
 624	/*
 625	 * We do get here for APIC virtualization enabled if the guest
 626	 * uses the Hyper-V APIC enlightenment.  In this case we may need
 627	 * to trigger a new interrupt delivery by writing the SVI field;
 628	 * on the other hand isr_count and highest_isr_cache are unused
 629	 * and must be left alone.
 630	 */
 631	if (unlikely(apic->apicv_active))
 632		static_call_cond(kvm_x86_hwapic_isr_update)(apic_find_highest_isr(apic));
 
 633	else {
 634		--apic->isr_count;
 635		BUG_ON(apic->isr_count < 0);
 636		apic->highest_isr_cache = -1;
 637	}
 638}
 639
 640int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
 641{
 642	/* This may race with setting of irr in __apic_accept_irq() and
 643	 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
 644	 * will cause vmexit immediately and the value will be recalculated
 645	 * on the next vmentry.
 646	 */
 647	return apic_find_highest_irr(vcpu->arch.apic);
 648}
 649EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
 650
 651static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
 652			     int vector, int level, int trig_mode,
 653			     struct dest_map *dest_map);
 654
 655int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
 656		     struct dest_map *dest_map)
 657{
 658	struct kvm_lapic *apic = vcpu->arch.apic;
 659
 660	return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
 661			irq->level, irq->trig_mode, dest_map);
 662}
 663
 664static int __pv_send_ipi(unsigned long *ipi_bitmap, struct kvm_apic_map *map,
 665			 struct kvm_lapic_irq *irq, u32 min)
 666{
 667	int i, count = 0;
 668	struct kvm_vcpu *vcpu;
 669
 670	if (min > map->max_apic_id)
 671		return 0;
 672
 673	for_each_set_bit(i, ipi_bitmap,
 674		min((u32)BITS_PER_LONG, (map->max_apic_id - min + 1))) {
 675		if (map->phys_map[min + i]) {
 676			vcpu = map->phys_map[min + i]->vcpu;
 677			count += kvm_apic_set_irq(vcpu, irq, NULL);
 678		}
 679	}
 680
 681	return count;
 682}
 683
 684int kvm_pv_send_ipi(struct kvm *kvm, unsigned long ipi_bitmap_low,
 685		    unsigned long ipi_bitmap_high, u32 min,
 686		    unsigned long icr, int op_64_bit)
 687{
 688	struct kvm_apic_map *map;
 689	struct kvm_lapic_irq irq = {0};
 690	int cluster_size = op_64_bit ? 64 : 32;
 691	int count;
 692
 693	if (icr & (APIC_DEST_MASK | APIC_SHORT_MASK))
 694		return -KVM_EINVAL;
 695
 696	irq.vector = icr & APIC_VECTOR_MASK;
 697	irq.delivery_mode = icr & APIC_MODE_MASK;
 698	irq.level = (icr & APIC_INT_ASSERT) != 0;
 699	irq.trig_mode = icr & APIC_INT_LEVELTRIG;
 700
 701	rcu_read_lock();
 702	map = rcu_dereference(kvm->arch.apic_map);
 703
 704	count = -EOPNOTSUPP;
 705	if (likely(map)) {
 706		count = __pv_send_ipi(&ipi_bitmap_low, map, &irq, min);
 707		min += cluster_size;
 708		count += __pv_send_ipi(&ipi_bitmap_high, map, &irq, min);
 709	}
 710
 711	rcu_read_unlock();
 712	return count;
 713}
 714
 715static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
 716{
 717
 718	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
 719				      sizeof(val));
 720}
 721
 722static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
 723{
 724
 725	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
 726				      sizeof(*val));
 727}
 728
 729static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
 730{
 731	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
 732}
 733
 
 
 
 
 
 
 
 
 
 
 
 734static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
 735{
 736	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0)
 
 
 737		return;
 738
 739	__set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
 740}
 741
 742static bool pv_eoi_test_and_clr_pending(struct kvm_vcpu *vcpu)
 743{
 744	u8 val;
 745
 746	if (pv_eoi_get_user(vcpu, &val) < 0)
 747		return false;
 748
 749	val &= KVM_PV_EOI_ENABLED;
 750
 751	if (val && pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0)
 752		return false;
 753
 754	/*
 755	 * Clear pending bit in any case: it will be set again on vmentry.
 756	 * While this might not be ideal from performance point of view,
 757	 * this makes sure pv eoi is only enabled when we know it's safe.
 758	 */
 759	__clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
 760
 761	return val;
 762}
 763
 764static int apic_has_interrupt_for_ppr(struct kvm_lapic *apic, u32 ppr)
 765{
 766	int highest_irr;
 767	if (kvm_x86_ops.sync_pir_to_irr)
 768		highest_irr = static_call(kvm_x86_sync_pir_to_irr)(apic->vcpu);
 769	else
 770		highest_irr = apic_find_highest_irr(apic);
 771	if (highest_irr == -1 || (highest_irr & 0xF0) <= ppr)
 772		return -1;
 773	return highest_irr;
 774}
 775
 776static bool __apic_update_ppr(struct kvm_lapic *apic, u32 *new_ppr)
 777{
 778	u32 tpr, isrv, ppr, old_ppr;
 779	int isr;
 780
 781	old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
 782	tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
 783	isr = apic_find_highest_isr(apic);
 784	isrv = (isr != -1) ? isr : 0;
 785
 786	if ((tpr & 0xf0) >= (isrv & 0xf0))
 787		ppr = tpr & 0xff;
 788	else
 789		ppr = isrv & 0xf0;
 790
 791	*new_ppr = ppr;
 792	if (old_ppr != ppr)
 793		kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
 794
 795	return ppr < old_ppr;
 796}
 797
 798static void apic_update_ppr(struct kvm_lapic *apic)
 799{
 800	u32 ppr;
 801
 802	if (__apic_update_ppr(apic, &ppr) &&
 803	    apic_has_interrupt_for_ppr(apic, ppr) != -1)
 804		kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
 805}
 806
 807void kvm_apic_update_ppr(struct kvm_vcpu *vcpu)
 808{
 809	apic_update_ppr(vcpu->arch.apic);
 810}
 811EXPORT_SYMBOL_GPL(kvm_apic_update_ppr);
 812
 813static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
 814{
 815	kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
 816	apic_update_ppr(apic);
 817}
 818
 819static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
 820{
 821	return mda == (apic_x2apic_mode(apic) ?
 822			X2APIC_BROADCAST : APIC_BROADCAST);
 823}
 824
 825static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
 826{
 827	if (kvm_apic_broadcast(apic, mda))
 828		return true;
 829
 
 
 
 830	/*
 831	 * Hotplug hack: Accept interrupts for vCPUs in xAPIC mode as if they
 832	 * were in x2APIC mode if the target APIC ID can't be encoded as an
 833	 * xAPIC ID.  This allows unique addressing of hotplugged vCPUs (which
 834	 * start in xAPIC mode) with an APIC ID that is unaddressable in xAPIC
 835	 * mode.  Match the x2APIC ID if and only if the target APIC ID can't
 836	 * be encoded in xAPIC to avoid spurious matches against a vCPU that
 837	 * changed its (addressable) xAPIC ID (which is writable).
 838	 */
 839	if (apic_x2apic_mode(apic) || mda > 0xff)
 840		return mda == kvm_x2apic_id(apic);
 841
 842	return mda == kvm_xapic_id(apic);
 843}
 844
 845static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
 846{
 847	u32 logical_id;
 848
 849	if (kvm_apic_broadcast(apic, mda))
 850		return true;
 851
 852	logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
 853
 854	if (apic_x2apic_mode(apic))
 855		return ((logical_id >> 16) == (mda >> 16))
 856		       && (logical_id & mda & 0xffff) != 0;
 857
 858	logical_id = GET_APIC_LOGICAL_ID(logical_id);
 859
 860	switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
 861	case APIC_DFR_FLAT:
 862		return (logical_id & mda) != 0;
 863	case APIC_DFR_CLUSTER:
 864		return ((logical_id >> 4) == (mda >> 4))
 865		       && (logical_id & mda & 0xf) != 0;
 866	default:
 867		return false;
 868	}
 869}
 870
 871/* The KVM local APIC implementation has two quirks:
 872 *
 873 *  - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
 874 *    in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
 875 *    KVM doesn't do that aliasing.
 876 *
 877 *  - in-kernel IOAPIC messages have to be delivered directly to
 878 *    x2APIC, because the kernel does not support interrupt remapping.
 879 *    In order to support broadcast without interrupt remapping, x2APIC
 880 *    rewrites the destination of non-IPI messages from APIC_BROADCAST
 881 *    to X2APIC_BROADCAST.
 882 *
 883 * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API.  This is
 884 * important when userspace wants to use x2APIC-format MSIs, because
 885 * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
 886 */
 887static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
 888		struct kvm_lapic *source, struct kvm_lapic *target)
 889{
 890	bool ipi = source != NULL;
 891
 892	if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
 893	    !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
 894		return X2APIC_BROADCAST;
 895
 896	return dest_id;
 897}
 898
 899bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
 900			   int shorthand, unsigned int dest, int dest_mode)
 901{
 902	struct kvm_lapic *target = vcpu->arch.apic;
 903	u32 mda = kvm_apic_mda(vcpu, dest, source, target);
 904
 905	ASSERT(target);
 906	switch (shorthand) {
 907	case APIC_DEST_NOSHORT:
 908		if (dest_mode == APIC_DEST_PHYSICAL)
 909			return kvm_apic_match_physical_addr(target, mda);
 910		else
 911			return kvm_apic_match_logical_addr(target, mda);
 912	case APIC_DEST_SELF:
 913		return target == source;
 914	case APIC_DEST_ALLINC:
 915		return true;
 916	case APIC_DEST_ALLBUT:
 917		return target != source;
 918	default:
 919		return false;
 920	}
 921}
 922EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
 923
 924int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
 925		       const unsigned long *bitmap, u32 bitmap_size)
 926{
 927	u32 mod;
 928	int i, idx = -1;
 929
 930	mod = vector % dest_vcpus;
 931
 932	for (i = 0; i <= mod; i++) {
 933		idx = find_next_bit(bitmap, bitmap_size, idx + 1);
 934		BUG_ON(idx == bitmap_size);
 935	}
 936
 937	return idx;
 938}
 939
 940static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
 941{
 942	if (!kvm->arch.disabled_lapic_found) {
 943		kvm->arch.disabled_lapic_found = true;
 944		printk(KERN_INFO
 945		       "Disabled LAPIC found during irq injection\n");
 946	}
 947}
 948
 949static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
 950		struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
 951{
 952	if (kvm->arch.x2apic_broadcast_quirk_disabled) {
 953		if ((irq->dest_id == APIC_BROADCAST &&
 954				map->mode != KVM_APIC_MODE_X2APIC))
 955			return true;
 956		if (irq->dest_id == X2APIC_BROADCAST)
 957			return true;
 958	} else {
 959		bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
 960		if (irq->dest_id == (x2apic_ipi ?
 961		                     X2APIC_BROADCAST : APIC_BROADCAST))
 962			return true;
 963	}
 964
 965	return false;
 966}
 967
 968/* Return true if the interrupt can be handled by using *bitmap as index mask
 969 * for valid destinations in *dst array.
 970 * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
 971 * Note: we may have zero kvm_lapic destinations when we return true, which
 972 * means that the interrupt should be dropped.  In this case, *bitmap would be
 973 * zero and *dst undefined.
 974 */
 975static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
 976		struct kvm_lapic **src, struct kvm_lapic_irq *irq,
 977		struct kvm_apic_map *map, struct kvm_lapic ***dst,
 978		unsigned long *bitmap)
 979{
 980	int i, lowest;
 981
 982	if (irq->shorthand == APIC_DEST_SELF && src) {
 983		*dst = src;
 984		*bitmap = 1;
 985		return true;
 986	} else if (irq->shorthand)
 987		return false;
 988
 989	if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
 990		return false;
 991
 992	if (irq->dest_mode == APIC_DEST_PHYSICAL) {
 993		if (irq->dest_id > map->max_apic_id) {
 994			*bitmap = 0;
 995		} else {
 996			u32 dest_id = array_index_nospec(irq->dest_id, map->max_apic_id + 1);
 997			*dst = &map->phys_map[dest_id];
 998			*bitmap = 1;
 999		}
1000		return true;
1001	}
1002
1003	*bitmap = 0;
1004	if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
1005				(u16 *)bitmap))
1006		return false;
1007
1008	if (!kvm_lowest_prio_delivery(irq))
1009		return true;
1010
1011	if (!kvm_vector_hashing_enabled()) {
1012		lowest = -1;
1013		for_each_set_bit(i, bitmap, 16) {
1014			if (!(*dst)[i])
1015				continue;
1016			if (lowest < 0)
1017				lowest = i;
1018			else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
1019						(*dst)[lowest]->vcpu) < 0)
1020				lowest = i;
1021		}
1022	} else {
1023		if (!*bitmap)
1024			return true;
1025
1026		lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
1027				bitmap, 16);
1028
1029		if (!(*dst)[lowest]) {
1030			kvm_apic_disabled_lapic_found(kvm);
1031			*bitmap = 0;
1032			return true;
1033		}
1034	}
1035
1036	*bitmap = (lowest >= 0) ? 1 << lowest : 0;
1037
1038	return true;
1039}
1040
1041bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
1042		struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
1043{
1044	struct kvm_apic_map *map;
1045	unsigned long bitmap;
1046	struct kvm_lapic **dst = NULL;
1047	int i;
1048	bool ret;
1049
1050	*r = -1;
1051
1052	if (irq->shorthand == APIC_DEST_SELF) {
1053		if (KVM_BUG_ON(!src, kvm)) {
1054			*r = 0;
1055			return true;
1056		}
1057		*r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
1058		return true;
1059	}
1060
1061	rcu_read_lock();
1062	map = rcu_dereference(kvm->arch.apic_map);
1063
1064	ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
1065	if (ret) {
1066		*r = 0;
1067		for_each_set_bit(i, &bitmap, 16) {
1068			if (!dst[i])
1069				continue;
1070			*r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
1071		}
1072	}
1073
1074	rcu_read_unlock();
1075	return ret;
1076}
1077
1078/*
1079 * This routine tries to handle interrupts in posted mode, here is how
1080 * it deals with different cases:
1081 * - For single-destination interrupts, handle it in posted mode
1082 * - Else if vector hashing is enabled and it is a lowest-priority
1083 *   interrupt, handle it in posted mode and use the following mechanism
1084 *   to find the destination vCPU.
1085 *	1. For lowest-priority interrupts, store all the possible
1086 *	   destination vCPUs in an array.
1087 *	2. Use "guest vector % max number of destination vCPUs" to find
1088 *	   the right destination vCPU in the array for the lowest-priority
1089 *	   interrupt.
1090 * - Otherwise, use remapped mode to inject the interrupt.
1091 */
1092bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
1093			struct kvm_vcpu **dest_vcpu)
1094{
1095	struct kvm_apic_map *map;
1096	unsigned long bitmap;
1097	struct kvm_lapic **dst = NULL;
1098	bool ret = false;
1099
1100	if (irq->shorthand)
1101		return false;
1102
1103	rcu_read_lock();
1104	map = rcu_dereference(kvm->arch.apic_map);
1105
1106	if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
1107			hweight16(bitmap) == 1) {
1108		unsigned long i = find_first_bit(&bitmap, 16);
1109
1110		if (dst[i]) {
1111			*dest_vcpu = dst[i]->vcpu;
1112			ret = true;
1113		}
1114	}
1115
1116	rcu_read_unlock();
1117	return ret;
1118}
1119
1120/*
1121 * Add a pending IRQ into lapic.
1122 * Return 1 if successfully added and 0 if discarded.
1123 */
1124static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
1125			     int vector, int level, int trig_mode,
1126			     struct dest_map *dest_map)
1127{
1128	int result = 0;
1129	struct kvm_vcpu *vcpu = apic->vcpu;
1130
1131	trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
1132				  trig_mode, vector);
1133	switch (delivery_mode) {
1134	case APIC_DM_LOWEST:
1135		vcpu->arch.apic_arb_prio++;
1136		fallthrough;
1137	case APIC_DM_FIXED:
1138		if (unlikely(trig_mode && !level))
1139			break;
1140
1141		/* FIXME add logic for vcpu on reset */
1142		if (unlikely(!apic_enabled(apic)))
1143			break;
1144
1145		result = 1;
1146
1147		if (dest_map) {
1148			__set_bit(vcpu->vcpu_id, dest_map->map);
1149			dest_map->vectors[vcpu->vcpu_id] = vector;
1150		}
1151
1152		if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
1153			if (trig_mode)
1154				kvm_lapic_set_vector(vector,
1155						     apic->regs + APIC_TMR);
1156			else
1157				kvm_lapic_clear_vector(vector,
1158						       apic->regs + APIC_TMR);
1159		}
1160
1161		static_call(kvm_x86_deliver_interrupt)(apic, delivery_mode,
1162						       trig_mode, vector);
 
 
 
1163		break;
1164
1165	case APIC_DM_REMRD:
1166		result = 1;
1167		vcpu->arch.pv.pv_unhalted = 1;
1168		kvm_make_request(KVM_REQ_EVENT, vcpu);
1169		kvm_vcpu_kick(vcpu);
1170		break;
1171
1172	case APIC_DM_SMI:
1173		if (!kvm_inject_smi(vcpu)) {
1174			kvm_vcpu_kick(vcpu);
1175			result = 1;
1176		}
1177		break;
1178
1179	case APIC_DM_NMI:
1180		result = 1;
1181		kvm_inject_nmi(vcpu);
1182		kvm_vcpu_kick(vcpu);
1183		break;
1184
1185	case APIC_DM_INIT:
1186		if (!trig_mode || level) {
1187			result = 1;
1188			/* assumes that there are only KVM_APIC_INIT/SIPI */
1189			apic->pending_events = (1UL << KVM_APIC_INIT);
1190			kvm_make_request(KVM_REQ_EVENT, vcpu);
1191			kvm_vcpu_kick(vcpu);
1192		}
1193		break;
1194
1195	case APIC_DM_STARTUP:
1196		result = 1;
1197		apic->sipi_vector = vector;
1198		/* make sure sipi_vector is visible for the receiver */
1199		smp_wmb();
1200		set_bit(KVM_APIC_SIPI, &apic->pending_events);
1201		kvm_make_request(KVM_REQ_EVENT, vcpu);
1202		kvm_vcpu_kick(vcpu);
1203		break;
1204
1205	case APIC_DM_EXTINT:
1206		/*
1207		 * Should only be called by kvm_apic_local_deliver() with LVT0,
1208		 * before NMI watchdog was enabled. Already handled by
1209		 * kvm_apic_accept_pic_intr().
1210		 */
1211		break;
1212
1213	default:
1214		printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
1215		       delivery_mode);
1216		break;
1217	}
1218	return result;
1219}
1220
1221/*
1222 * This routine identifies the destination vcpus mask meant to receive the
1223 * IOAPIC interrupts. It either uses kvm_apic_map_get_dest_lapic() to find
1224 * out the destination vcpus array and set the bitmap or it traverses to
1225 * each available vcpu to identify the same.
1226 */
1227void kvm_bitmap_or_dest_vcpus(struct kvm *kvm, struct kvm_lapic_irq *irq,
1228			      unsigned long *vcpu_bitmap)
1229{
1230	struct kvm_lapic **dest_vcpu = NULL;
1231	struct kvm_lapic *src = NULL;
1232	struct kvm_apic_map *map;
1233	struct kvm_vcpu *vcpu;
1234	unsigned long bitmap, i;
1235	int vcpu_idx;
1236	bool ret;
1237
1238	rcu_read_lock();
1239	map = rcu_dereference(kvm->arch.apic_map);
1240
1241	ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dest_vcpu,
1242					  &bitmap);
1243	if (ret) {
1244		for_each_set_bit(i, &bitmap, 16) {
1245			if (!dest_vcpu[i])
1246				continue;
1247			vcpu_idx = dest_vcpu[i]->vcpu->vcpu_idx;
1248			__set_bit(vcpu_idx, vcpu_bitmap);
1249		}
1250	} else {
1251		kvm_for_each_vcpu(i, vcpu, kvm) {
1252			if (!kvm_apic_present(vcpu))
1253				continue;
1254			if (!kvm_apic_match_dest(vcpu, NULL,
1255						 irq->shorthand,
1256						 irq->dest_id,
1257						 irq->dest_mode))
1258				continue;
1259			__set_bit(i, vcpu_bitmap);
1260		}
1261	}
1262	rcu_read_unlock();
1263}
1264
1265int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1266{
1267	return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
1268}
1269
1270static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
1271{
1272	return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
1273}
1274
1275static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1276{
1277	int trigger_mode;
1278
1279	/* Eoi the ioapic only if the ioapic doesn't own the vector. */
1280	if (!kvm_ioapic_handles_vector(apic, vector))
1281		return;
1282
1283	/* Request a KVM exit to inform the userspace IOAPIC. */
1284	if (irqchip_split(apic->vcpu->kvm)) {
1285		apic->vcpu->arch.pending_ioapic_eoi = vector;
1286		kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1287		return;
1288	}
1289
1290	if (apic_test_vector(vector, apic->regs + APIC_TMR))
1291		trigger_mode = IOAPIC_LEVEL_TRIG;
1292	else
1293		trigger_mode = IOAPIC_EDGE_TRIG;
1294
1295	kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1296}
1297
1298static int apic_set_eoi(struct kvm_lapic *apic)
1299{
1300	int vector = apic_find_highest_isr(apic);
1301
1302	trace_kvm_eoi(apic, vector);
1303
1304	/*
1305	 * Not every write EOI will has corresponding ISR,
1306	 * one example is when Kernel check timer on setup_IO_APIC
1307	 */
1308	if (vector == -1)
1309		return vector;
1310
1311	apic_clear_isr(vector, apic);
1312	apic_update_ppr(apic);
1313
1314	if (to_hv_vcpu(apic->vcpu) &&
1315	    test_bit(vector, to_hv_synic(apic->vcpu)->vec_bitmap))
1316		kvm_hv_synic_send_eoi(apic->vcpu, vector);
1317
1318	kvm_ioapic_send_eoi(apic, vector);
1319	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1320	return vector;
1321}
1322
1323/*
1324 * this interface assumes a trap-like exit, which has already finished
1325 * desired side effect including vISR and vPPR update.
1326 */
1327void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1328{
1329	struct kvm_lapic *apic = vcpu->arch.apic;
1330
1331	trace_kvm_eoi(apic, vector);
1332
1333	kvm_ioapic_send_eoi(apic, vector);
1334	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1335}
1336EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1337
1338void kvm_apic_send_ipi(struct kvm_lapic *apic, u32 icr_low, u32 icr_high)
1339{
1340	struct kvm_lapic_irq irq;
1341
1342	/* KVM has no delay and should always clear the BUSY/PENDING flag. */
1343	WARN_ON_ONCE(icr_low & APIC_ICR_BUSY);
1344
1345	irq.vector = icr_low & APIC_VECTOR_MASK;
1346	irq.delivery_mode = icr_low & APIC_MODE_MASK;
1347	irq.dest_mode = icr_low & APIC_DEST_MASK;
1348	irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1349	irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1350	irq.shorthand = icr_low & APIC_SHORT_MASK;
1351	irq.msi_redir_hint = false;
1352	if (apic_x2apic_mode(apic))
1353		irq.dest_id = icr_high;
1354	else
1355		irq.dest_id = GET_XAPIC_DEST_FIELD(icr_high);
1356
1357	trace_kvm_apic_ipi(icr_low, irq.dest_id);
1358
1359	kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1360}
1361EXPORT_SYMBOL_GPL(kvm_apic_send_ipi);
1362
1363static u32 apic_get_tmcct(struct kvm_lapic *apic)
1364{
1365	ktime_t remaining, now;
1366	s64 ns;
1367	u32 tmcct;
1368
1369	ASSERT(apic != NULL);
1370
1371	/* if initial count is 0, current count should also be 0 */
1372	if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1373		apic->lapic_timer.period == 0)
1374		return 0;
1375
1376	now = ktime_get();
1377	remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1378	if (ktime_to_ns(remaining) < 0)
1379		remaining = 0;
1380
1381	ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1382	tmcct = div64_u64(ns,
1383			 (APIC_BUS_CYCLE_NS * apic->divide_count));
1384
1385	return tmcct;
1386}
1387
1388static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1389{
1390	struct kvm_vcpu *vcpu = apic->vcpu;
1391	struct kvm_run *run = vcpu->run;
1392
1393	kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1394	run->tpr_access.rip = kvm_rip_read(vcpu);
1395	run->tpr_access.is_write = write;
1396}
1397
1398static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1399{
1400	if (apic->vcpu->arch.tpr_access_reporting)
1401		__report_tpr_access(apic, write);
1402}
1403
1404static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1405{
1406	u32 val = 0;
1407
1408	if (offset >= LAPIC_MMIO_LENGTH)
1409		return 0;
1410
1411	switch (offset) {
1412	case APIC_ARBPRI:
1413		break;
1414
1415	case APIC_TMCCT:	/* Timer CCR */
1416		if (apic_lvtt_tscdeadline(apic))
1417			return 0;
1418
1419		val = apic_get_tmcct(apic);
1420		break;
1421	case APIC_PROCPRI:
1422		apic_update_ppr(apic);
1423		val = kvm_lapic_get_reg(apic, offset);
1424		break;
1425	case APIC_TASKPRI:
1426		report_tpr_access(apic, false);
1427		fallthrough;
1428	default:
1429		val = kvm_lapic_get_reg(apic, offset);
1430		break;
1431	}
1432
1433	return val;
1434}
1435
1436static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1437{
1438	return container_of(dev, struct kvm_lapic, dev);
1439}
1440
1441#define APIC_REG_MASK(reg)	(1ull << ((reg) >> 4))
1442#define APIC_REGS_MASK(first, count) \
1443	(APIC_REG_MASK(first) * ((1ull << (count)) - 1))
1444
1445static int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1446			      void *data)
1447{
1448	unsigned char alignment = offset & 0xf;
1449	u32 result;
1450	/* this bitmask has a bit cleared for each reserved register */
1451	u64 valid_reg_mask =
1452		APIC_REG_MASK(APIC_ID) |
1453		APIC_REG_MASK(APIC_LVR) |
1454		APIC_REG_MASK(APIC_TASKPRI) |
1455		APIC_REG_MASK(APIC_PROCPRI) |
1456		APIC_REG_MASK(APIC_LDR) |
1457		APIC_REG_MASK(APIC_DFR) |
1458		APIC_REG_MASK(APIC_SPIV) |
1459		APIC_REGS_MASK(APIC_ISR, APIC_ISR_NR) |
1460		APIC_REGS_MASK(APIC_TMR, APIC_ISR_NR) |
1461		APIC_REGS_MASK(APIC_IRR, APIC_ISR_NR) |
1462		APIC_REG_MASK(APIC_ESR) |
1463		APIC_REG_MASK(APIC_ICR) |
 
1464		APIC_REG_MASK(APIC_LVTT) |
1465		APIC_REG_MASK(APIC_LVTTHMR) |
1466		APIC_REG_MASK(APIC_LVTPC) |
1467		APIC_REG_MASK(APIC_LVT0) |
1468		APIC_REG_MASK(APIC_LVT1) |
1469		APIC_REG_MASK(APIC_LVTERR) |
1470		APIC_REG_MASK(APIC_TMICT) |
1471		APIC_REG_MASK(APIC_TMCCT) |
1472		APIC_REG_MASK(APIC_TDCR);
1473
1474	if (kvm_lapic_lvt_supported(apic, LVT_CMCI))
1475		valid_reg_mask |= APIC_REG_MASK(APIC_LVTCMCI);
1476
1477	/*
1478	 * ARBPRI and ICR2 are not valid in x2APIC mode.  WARN if KVM reads ICR
1479	 * in x2APIC mode as it's an 8-byte register in x2APIC and needs to be
1480	 * manually handled by the caller.
1481	 */
1482	if (!apic_x2apic_mode(apic))
1483		valid_reg_mask |= APIC_REG_MASK(APIC_ARBPRI) |
1484				  APIC_REG_MASK(APIC_ICR2);
1485	else
1486		WARN_ON_ONCE(offset == APIC_ICR);
1487
1488	if (alignment + len > 4)
1489		return 1;
1490
1491	if (offset > 0x3f0 || !(valid_reg_mask & APIC_REG_MASK(offset)))
1492		return 1;
1493
1494	result = __apic_read(apic, offset & ~0xf);
1495
1496	trace_kvm_apic_read(offset, result);
1497
1498	switch (len) {
1499	case 1:
1500	case 2:
1501	case 4:
1502		memcpy(data, (char *)&result + alignment, len);
1503		break;
1504	default:
1505		printk(KERN_ERR "Local APIC read with len = %x, "
1506		       "should be 1,2, or 4 instead\n", len);
1507		break;
1508	}
1509	return 0;
1510}
 
1511
1512static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1513{
1514	return addr >= apic->base_address &&
1515		addr < apic->base_address + LAPIC_MMIO_LENGTH;
1516}
1517
1518static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1519			   gpa_t address, int len, void *data)
1520{
1521	struct kvm_lapic *apic = to_lapic(this);
1522	u32 offset = address - apic->base_address;
1523
1524	if (!apic_mmio_in_range(apic, address))
1525		return -EOPNOTSUPP;
1526
1527	if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
1528		if (!kvm_check_has_quirk(vcpu->kvm,
1529					 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
1530			return -EOPNOTSUPP;
1531
1532		memset(data, 0xff, len);
1533		return 0;
1534	}
1535
1536	kvm_lapic_reg_read(apic, offset, len, data);
1537
1538	return 0;
1539}
1540
1541static void update_divide_count(struct kvm_lapic *apic)
1542{
1543	u32 tmp1, tmp2, tdcr;
1544
1545	tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1546	tmp1 = tdcr & 0xf;
1547	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1548	apic->divide_count = 0x1 << (tmp2 & 0x7);
1549}
1550
1551static void limit_periodic_timer_frequency(struct kvm_lapic *apic)
1552{
1553	/*
1554	 * Do not allow the guest to program periodic timers with small
1555	 * interval, since the hrtimers are not throttled by the host
1556	 * scheduler.
1557	 */
1558	if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1559		s64 min_period = min_timer_period_us * 1000LL;
1560
1561		if (apic->lapic_timer.period < min_period) {
1562			pr_info_ratelimited(
1563			    "kvm: vcpu %i: requested %lld ns "
1564			    "lapic timer period limited to %lld ns\n",
1565			    apic->vcpu->vcpu_id,
1566			    apic->lapic_timer.period, min_period);
1567			apic->lapic_timer.period = min_period;
1568		}
1569	}
1570}
1571
1572static void cancel_hv_timer(struct kvm_lapic *apic);
1573
1574static void cancel_apic_timer(struct kvm_lapic *apic)
1575{
1576	hrtimer_cancel(&apic->lapic_timer.timer);
1577	preempt_disable();
1578	if (apic->lapic_timer.hv_timer_in_use)
1579		cancel_hv_timer(apic);
1580	preempt_enable();
1581	atomic_set(&apic->lapic_timer.pending, 0);
1582}
1583
1584static void apic_update_lvtt(struct kvm_lapic *apic)
1585{
1586	u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1587			apic->lapic_timer.timer_mode_mask;
1588
1589	if (apic->lapic_timer.timer_mode != timer_mode) {
1590		if (apic_lvtt_tscdeadline(apic) != (timer_mode ==
1591				APIC_LVT_TIMER_TSCDEADLINE)) {
1592			cancel_apic_timer(apic);
 
 
 
 
1593			kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1594			apic->lapic_timer.period = 0;
1595			apic->lapic_timer.tscdeadline = 0;
1596		}
1597		apic->lapic_timer.timer_mode = timer_mode;
1598		limit_periodic_timer_frequency(apic);
1599	}
1600}
1601
1602/*
1603 * On APICv, this test will cause a busy wait
1604 * during a higher-priority task.
1605 */
1606
1607static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1608{
1609	struct kvm_lapic *apic = vcpu->arch.apic;
1610	u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1611
1612	if (kvm_apic_hw_enabled(apic)) {
1613		int vec = reg & APIC_VECTOR_MASK;
1614		void *bitmap = apic->regs + APIC_ISR;
1615
1616		if (apic->apicv_active)
1617			bitmap = apic->regs + APIC_IRR;
1618
1619		if (apic_test_vector(vec, bitmap))
1620			return true;
1621	}
1622	return false;
1623}
1624
1625static inline void __wait_lapic_expire(struct kvm_vcpu *vcpu, u64 guest_cycles)
1626{
1627	u64 timer_advance_ns = vcpu->arch.apic->lapic_timer.timer_advance_ns;
1628
1629	/*
1630	 * If the guest TSC is running at a different ratio than the host, then
1631	 * convert the delay to nanoseconds to achieve an accurate delay.  Note
1632	 * that __delay() uses delay_tsc whenever the hardware has TSC, thus
1633	 * always for VMX enabled hardware.
1634	 */
1635	if (vcpu->arch.tsc_scaling_ratio == kvm_caps.default_tsc_scaling_ratio) {
1636		__delay(min(guest_cycles,
1637			nsec_to_cycles(vcpu, timer_advance_ns)));
1638	} else {
1639		u64 delay_ns = guest_cycles * 1000000ULL;
1640		do_div(delay_ns, vcpu->arch.virtual_tsc_khz);
1641		ndelay(min_t(u32, delay_ns, timer_advance_ns));
1642	}
1643}
1644
1645static inline void adjust_lapic_timer_advance(struct kvm_vcpu *vcpu,
1646					      s64 advance_expire_delta)
1647{
1648	struct kvm_lapic *apic = vcpu->arch.apic;
1649	u32 timer_advance_ns = apic->lapic_timer.timer_advance_ns;
1650	u64 ns;
1651
1652	/* Do not adjust for tiny fluctuations or large random spikes. */
1653	if (abs(advance_expire_delta) > LAPIC_TIMER_ADVANCE_ADJUST_MAX ||
1654	    abs(advance_expire_delta) < LAPIC_TIMER_ADVANCE_ADJUST_MIN)
1655		return;
1656
1657	/* too early */
1658	if (advance_expire_delta < 0) {
1659		ns = -advance_expire_delta * 1000000ULL;
1660		do_div(ns, vcpu->arch.virtual_tsc_khz);
1661		timer_advance_ns -= ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1662	} else {
1663	/* too late */
1664		ns = advance_expire_delta * 1000000ULL;
1665		do_div(ns, vcpu->arch.virtual_tsc_khz);
1666		timer_advance_ns += ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1667	}
1668
1669	if (unlikely(timer_advance_ns > LAPIC_TIMER_ADVANCE_NS_MAX))
1670		timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
1671	apic->lapic_timer.timer_advance_ns = timer_advance_ns;
1672}
1673
1674static void __kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1675{
1676	struct kvm_lapic *apic = vcpu->arch.apic;
1677	u64 guest_tsc, tsc_deadline;
1678
 
 
 
1679	tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1680	apic->lapic_timer.expired_tscdeadline = 0;
1681	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1682	trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1683
1684	if (lapic_timer_advance_dynamic) {
1685		adjust_lapic_timer_advance(vcpu, guest_tsc - tsc_deadline);
1686		/*
1687		 * If the timer fired early, reread the TSC to account for the
1688		 * overhead of the above adjustment to avoid waiting longer
1689		 * than is necessary.
1690		 */
1691		if (guest_tsc < tsc_deadline)
1692			guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1693	}
1694
1695	if (guest_tsc < tsc_deadline)
1696		__wait_lapic_expire(vcpu, tsc_deadline - guest_tsc);
 
 
 
1697}
1698
1699void kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1700{
1701	if (lapic_in_kernel(vcpu) &&
1702	    vcpu->arch.apic->lapic_timer.expired_tscdeadline &&
1703	    vcpu->arch.apic->lapic_timer.timer_advance_ns &&
1704	    lapic_timer_int_injected(vcpu))
1705		__kvm_wait_lapic_expire(vcpu);
1706}
1707EXPORT_SYMBOL_GPL(kvm_wait_lapic_expire);
1708
1709static void kvm_apic_inject_pending_timer_irqs(struct kvm_lapic *apic)
1710{
1711	struct kvm_timer *ktimer = &apic->lapic_timer;
1712
1713	kvm_apic_local_deliver(apic, APIC_LVTT);
1714	if (apic_lvtt_tscdeadline(apic)) {
1715		ktimer->tscdeadline = 0;
1716	} else if (apic_lvtt_oneshot(apic)) {
1717		ktimer->tscdeadline = 0;
1718		ktimer->target_expiration = 0;
1719	}
1720}
1721
1722static void apic_timer_expired(struct kvm_lapic *apic, bool from_timer_fn)
1723{
1724	struct kvm_vcpu *vcpu = apic->vcpu;
1725	struct kvm_timer *ktimer = &apic->lapic_timer;
1726
1727	if (atomic_read(&apic->lapic_timer.pending))
1728		return;
1729
1730	if (apic_lvtt_tscdeadline(apic) || ktimer->hv_timer_in_use)
1731		ktimer->expired_tscdeadline = ktimer->tscdeadline;
1732
1733	if (!from_timer_fn && apic->apicv_active) {
1734		WARN_ON(kvm_get_running_vcpu() != vcpu);
1735		kvm_apic_inject_pending_timer_irqs(apic);
1736		return;
1737	}
1738
1739	if (kvm_use_posted_timer_interrupt(apic->vcpu)) {
1740		/*
1741		 * Ensure the guest's timer has truly expired before posting an
1742		 * interrupt.  Open code the relevant checks to avoid querying
1743		 * lapic_timer_int_injected(), which will be false since the
1744		 * interrupt isn't yet injected.  Waiting until after injecting
1745		 * is not an option since that won't help a posted interrupt.
1746		 */
1747		if (vcpu->arch.apic->lapic_timer.expired_tscdeadline &&
1748		    vcpu->arch.apic->lapic_timer.timer_advance_ns)
1749			__kvm_wait_lapic_expire(vcpu);
1750		kvm_apic_inject_pending_timer_irqs(apic);
1751		return;
1752	}
1753
1754	atomic_inc(&apic->lapic_timer.pending);
1755	kvm_make_request(KVM_REQ_UNBLOCK, vcpu);
1756	if (from_timer_fn)
1757		kvm_vcpu_kick(vcpu);
1758}
1759
1760static void start_sw_tscdeadline(struct kvm_lapic *apic)
1761{
1762	struct kvm_timer *ktimer = &apic->lapic_timer;
1763	u64 guest_tsc, tscdeadline = ktimer->tscdeadline;
1764	u64 ns = 0;
1765	ktime_t expire;
1766	struct kvm_vcpu *vcpu = apic->vcpu;
1767	unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1768	unsigned long flags;
1769	ktime_t now;
1770
1771	if (unlikely(!tscdeadline || !this_tsc_khz))
1772		return;
1773
1774	local_irq_save(flags);
1775
1776	now = ktime_get();
1777	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1778
1779	ns = (tscdeadline - guest_tsc) * 1000000ULL;
1780	do_div(ns, this_tsc_khz);
1781
1782	if (likely(tscdeadline > guest_tsc) &&
1783	    likely(ns > apic->lapic_timer.timer_advance_ns)) {
1784		expire = ktime_add_ns(now, ns);
1785		expire = ktime_sub_ns(expire, ktimer->timer_advance_ns);
1786		hrtimer_start(&ktimer->timer, expire, HRTIMER_MODE_ABS_HARD);
1787	} else
1788		apic_timer_expired(apic, false);
1789
1790	local_irq_restore(flags);
1791}
1792
1793static inline u64 tmict_to_ns(struct kvm_lapic *apic, u32 tmict)
1794{
1795	return (u64)tmict * APIC_BUS_CYCLE_NS * (u64)apic->divide_count;
1796}
1797
1798static void update_target_expiration(struct kvm_lapic *apic, uint32_t old_divisor)
1799{
1800	ktime_t now, remaining;
1801	u64 ns_remaining_old, ns_remaining_new;
1802
1803	apic->lapic_timer.period =
1804			tmict_to_ns(apic, kvm_lapic_get_reg(apic, APIC_TMICT));
1805	limit_periodic_timer_frequency(apic);
1806
1807	now = ktime_get();
1808	remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1809	if (ktime_to_ns(remaining) < 0)
1810		remaining = 0;
1811
1812	ns_remaining_old = ktime_to_ns(remaining);
1813	ns_remaining_new = mul_u64_u32_div(ns_remaining_old,
1814	                                   apic->divide_count, old_divisor);
1815
1816	apic->lapic_timer.tscdeadline +=
1817		nsec_to_cycles(apic->vcpu, ns_remaining_new) -
1818		nsec_to_cycles(apic->vcpu, ns_remaining_old);
1819	apic->lapic_timer.target_expiration = ktime_add_ns(now, ns_remaining_new);
1820}
1821
1822static bool set_target_expiration(struct kvm_lapic *apic, u32 count_reg)
1823{
1824	ktime_t now;
1825	u64 tscl = rdtsc();
1826	s64 deadline;
1827
1828	now = ktime_get();
1829	apic->lapic_timer.period =
1830			tmict_to_ns(apic, kvm_lapic_get_reg(apic, APIC_TMICT));
1831
1832	if (!apic->lapic_timer.period) {
1833		apic->lapic_timer.tscdeadline = 0;
1834		return false;
1835	}
1836
1837	limit_periodic_timer_frequency(apic);
1838	deadline = apic->lapic_timer.period;
1839
1840	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1841		if (unlikely(count_reg != APIC_TMICT)) {
1842			deadline = tmict_to_ns(apic,
1843				     kvm_lapic_get_reg(apic, count_reg));
1844			if (unlikely(deadline <= 0))
1845				deadline = apic->lapic_timer.period;
1846			else if (unlikely(deadline > apic->lapic_timer.period)) {
1847				pr_info_ratelimited(
1848				    "kvm: vcpu %i: requested lapic timer restore with "
1849				    "starting count register %#x=%u (%lld ns) > initial count (%lld ns). "
1850				    "Using initial count to start timer.\n",
1851				    apic->vcpu->vcpu_id,
1852				    count_reg,
1853				    kvm_lapic_get_reg(apic, count_reg),
1854				    deadline, apic->lapic_timer.period);
1855				kvm_lapic_set_reg(apic, count_reg, 0);
1856				deadline = apic->lapic_timer.period;
1857			}
1858		}
1859	}
1860
1861	apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1862		nsec_to_cycles(apic->vcpu, deadline);
1863	apic->lapic_timer.target_expiration = ktime_add_ns(now, deadline);
1864
1865	return true;
1866}
1867
1868static void advance_periodic_target_expiration(struct kvm_lapic *apic)
1869{
1870	ktime_t now = ktime_get();
1871	u64 tscl = rdtsc();
1872	ktime_t delta;
1873
1874	/*
1875	 * Synchronize both deadlines to the same time source or
1876	 * differences in the periods (caused by differences in the
1877	 * underlying clocks or numerical approximation errors) will
1878	 * cause the two to drift apart over time as the errors
1879	 * accumulate.
1880	 */
1881	apic->lapic_timer.target_expiration =
1882		ktime_add_ns(apic->lapic_timer.target_expiration,
1883				apic->lapic_timer.period);
1884	delta = ktime_sub(apic->lapic_timer.target_expiration, now);
1885	apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1886		nsec_to_cycles(apic->vcpu, delta);
1887}
1888
1889static void start_sw_period(struct kvm_lapic *apic)
1890{
1891	if (!apic->lapic_timer.period)
1892		return;
1893
1894	if (ktime_after(ktime_get(),
1895			apic->lapic_timer.target_expiration)) {
1896		apic_timer_expired(apic, false);
1897
1898		if (apic_lvtt_oneshot(apic))
1899			return;
1900
1901		advance_periodic_target_expiration(apic);
1902	}
1903
1904	hrtimer_start(&apic->lapic_timer.timer,
1905		apic->lapic_timer.target_expiration,
1906		HRTIMER_MODE_ABS_HARD);
1907}
1908
1909bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1910{
1911	if (!lapic_in_kernel(vcpu))
1912		return false;
1913
1914	return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1915}
 
1916
1917static void cancel_hv_timer(struct kvm_lapic *apic)
1918{
1919	WARN_ON(preemptible());
1920	WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1921	static_call(kvm_x86_cancel_hv_timer)(apic->vcpu);
1922	apic->lapic_timer.hv_timer_in_use = false;
1923}
1924
1925static bool start_hv_timer(struct kvm_lapic *apic)
1926{
1927	struct kvm_timer *ktimer = &apic->lapic_timer;
1928	struct kvm_vcpu *vcpu = apic->vcpu;
1929	bool expired;
1930
1931	WARN_ON(preemptible());
1932	if (!kvm_can_use_hv_timer(vcpu))
1933		return false;
1934
1935	if (!ktimer->tscdeadline)
1936		return false;
1937
1938	if (static_call(kvm_x86_set_hv_timer)(vcpu, ktimer->tscdeadline, &expired))
1939		return false;
1940
1941	ktimer->hv_timer_in_use = true;
1942	hrtimer_cancel(&ktimer->timer);
1943
1944	/*
1945	 * To simplify handling the periodic timer, leave the hv timer running
1946	 * even if the deadline timer has expired, i.e. rely on the resulting
1947	 * VM-Exit to recompute the periodic timer's target expiration.
1948	 */
1949	if (!apic_lvtt_period(apic)) {
1950		/*
1951		 * Cancel the hv timer if the sw timer fired while the hv timer
1952		 * was being programmed, or if the hv timer itself expired.
1953		 */
1954		if (atomic_read(&ktimer->pending)) {
1955			cancel_hv_timer(apic);
1956		} else if (expired) {
1957			apic_timer_expired(apic, false);
1958			cancel_hv_timer(apic);
1959		}
1960	}
1961
1962	trace_kvm_hv_timer_state(vcpu->vcpu_id, ktimer->hv_timer_in_use);
1963
1964	return true;
1965}
1966
1967static void start_sw_timer(struct kvm_lapic *apic)
1968{
1969	struct kvm_timer *ktimer = &apic->lapic_timer;
1970
1971	WARN_ON(preemptible());
1972	if (apic->lapic_timer.hv_timer_in_use)
1973		cancel_hv_timer(apic);
1974	if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
1975		return;
1976
1977	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1978		start_sw_period(apic);
1979	else if (apic_lvtt_tscdeadline(apic))
1980		start_sw_tscdeadline(apic);
1981	trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, false);
1982}
1983
1984static void restart_apic_timer(struct kvm_lapic *apic)
1985{
1986	preempt_disable();
1987
1988	if (!apic_lvtt_period(apic) && atomic_read(&apic->lapic_timer.pending))
1989		goto out;
1990
1991	if (!start_hv_timer(apic))
1992		start_sw_timer(apic);
1993out:
1994	preempt_enable();
1995}
1996
1997void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1998{
1999	struct kvm_lapic *apic = vcpu->arch.apic;
2000
2001	preempt_disable();
2002	/* If the preempt notifier has already run, it also called apic_timer_expired */
2003	if (!apic->lapic_timer.hv_timer_in_use)
2004		goto out;
2005	WARN_ON(kvm_vcpu_is_blocking(vcpu));
2006	apic_timer_expired(apic, false);
2007	cancel_hv_timer(apic);
 
2008
2009	if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
2010		advance_periodic_target_expiration(apic);
2011		restart_apic_timer(apic);
2012	}
2013out:
2014	preempt_enable();
2015}
2016EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
2017
2018void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
2019{
2020	restart_apic_timer(vcpu->arch.apic);
2021}
 
2022
2023void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
2024{
2025	struct kvm_lapic *apic = vcpu->arch.apic;
2026
2027	preempt_disable();
2028	/* Possibly the TSC deadline timer is not enabled yet */
2029	if (apic->lapic_timer.hv_timer_in_use)
2030		start_sw_timer(apic);
2031	preempt_enable();
2032}
 
2033
2034void kvm_lapic_restart_hv_timer(struct kvm_vcpu *vcpu)
2035{
2036	struct kvm_lapic *apic = vcpu->arch.apic;
2037
2038	WARN_ON(!apic->lapic_timer.hv_timer_in_use);
2039	restart_apic_timer(apic);
2040}
2041
2042static void __start_apic_timer(struct kvm_lapic *apic, u32 count_reg)
2043{
2044	atomic_set(&apic->lapic_timer.pending, 0);
2045
2046	if ((apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
2047	    && !set_target_expiration(apic, count_reg))
2048		return;
2049
2050	restart_apic_timer(apic);
2051}
2052
2053static void start_apic_timer(struct kvm_lapic *apic)
2054{
2055	__start_apic_timer(apic, APIC_TMICT);
2056}
2057
2058static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
2059{
2060	bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
2061
2062	if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
2063		apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
2064		if (lvt0_in_nmi_mode) {
2065			atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
2066		} else
2067			atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
2068	}
2069}
2070
2071static void kvm_lapic_xapic_id_updated(struct kvm_lapic *apic)
2072{
2073	struct kvm *kvm = apic->vcpu->kvm;
2074
2075	if (KVM_BUG_ON(apic_x2apic_mode(apic), kvm))
2076		return;
2077
2078	if (kvm_xapic_id(apic) == apic->vcpu->vcpu_id)
2079		return;
2080
2081	kvm_set_apicv_inhibit(apic->vcpu->kvm, APICV_INHIBIT_REASON_APIC_ID_MODIFIED);
2082}
2083
2084static int get_lvt_index(u32 reg)
2085{
2086	if (reg == APIC_LVTCMCI)
2087		return LVT_CMCI;
2088	if (reg < APIC_LVTT || reg > APIC_LVTERR)
2089		return -1;
2090	return array_index_nospec(
2091			(reg - APIC_LVTT) >> 4, KVM_APIC_MAX_NR_LVT_ENTRIES);
2092}
2093
2094static int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
2095{
2096	int ret = 0;
2097
2098	trace_kvm_apic_write(reg, val);
2099
2100	switch (reg) {
2101	case APIC_ID:		/* Local APIC ID */
2102		if (!apic_x2apic_mode(apic)) {
2103			kvm_apic_set_xapic_id(apic, val >> 24);
2104			kvm_lapic_xapic_id_updated(apic);
2105		} else {
2106			ret = 1;
2107		}
2108		break;
2109
2110	case APIC_TASKPRI:
2111		report_tpr_access(apic, true);
2112		apic_set_tpr(apic, val & 0xff);
2113		break;
2114
2115	case APIC_EOI:
2116		apic_set_eoi(apic);
2117		break;
2118
2119	case APIC_LDR:
2120		if (!apic_x2apic_mode(apic))
2121			kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
2122		else
2123			ret = 1;
2124		break;
2125
2126	case APIC_DFR:
2127		if (!apic_x2apic_mode(apic))
2128			kvm_apic_set_dfr(apic, val | 0x0FFFFFFF);
2129		else
 
2130			ret = 1;
2131		break;
2132
2133	case APIC_SPIV: {
2134		u32 mask = 0x3ff;
2135		if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
2136			mask |= APIC_SPIV_DIRECTED_EOI;
2137		apic_set_spiv(apic, val & mask);
2138		if (!(val & APIC_SPIV_APIC_ENABLED)) {
2139			int i;
 
2140
2141			for (i = 0; i < apic->nr_lvt_entries; i++) {
2142				kvm_lapic_set_reg(apic, APIC_LVTx(i),
2143					kvm_lapic_get_reg(apic, APIC_LVTx(i)) | APIC_LVT_MASKED);
 
 
2144			}
2145			apic_update_lvtt(apic);
2146			atomic_set(&apic->lapic_timer.pending, 0);
2147
2148		}
2149		break;
2150	}
2151	case APIC_ICR:
2152		WARN_ON_ONCE(apic_x2apic_mode(apic));
2153
2154		/* No delay here, so we always clear the pending bit */
2155		val &= ~APIC_ICR_BUSY;
2156		kvm_apic_send_ipi(apic, val, kvm_lapic_get_reg(apic, APIC_ICR2));
2157		kvm_lapic_set_reg(apic, APIC_ICR, val);
2158		break;
 
2159	case APIC_ICR2:
2160		if (apic_x2apic_mode(apic))
2161			ret = 1;
2162		else
2163			kvm_lapic_set_reg(apic, APIC_ICR2, val & 0xff000000);
2164		break;
2165
2166	case APIC_LVT0:
2167		apic_manage_nmi_watchdog(apic, val);
2168		fallthrough;
2169	case APIC_LVTTHMR:
2170	case APIC_LVTPC:
2171	case APIC_LVT1:
2172	case APIC_LVTERR:
2173	case APIC_LVTCMCI: {
2174		u32 index = get_lvt_index(reg);
2175		if (!kvm_lapic_lvt_supported(apic, index)) {
2176			ret = 1;
2177			break;
2178		}
2179		if (!kvm_apic_sw_enabled(apic))
2180			val |= APIC_LVT_MASKED;
 
 
 
2181		val &= apic_lvt_mask[index];
2182		kvm_lapic_set_reg(apic, reg, val);
2183		break;
2184	}
2185
2186	case APIC_LVTT:
2187		if (!kvm_apic_sw_enabled(apic))
2188			val |= APIC_LVT_MASKED;
2189		val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
2190		kvm_lapic_set_reg(apic, APIC_LVTT, val);
2191		apic_update_lvtt(apic);
2192		break;
2193
2194	case APIC_TMICT:
2195		if (apic_lvtt_tscdeadline(apic))
2196			break;
2197
2198		cancel_apic_timer(apic);
2199		kvm_lapic_set_reg(apic, APIC_TMICT, val);
2200		start_apic_timer(apic);
2201		break;
2202
2203	case APIC_TDCR: {
2204		uint32_t old_divisor = apic->divide_count;
2205
2206		kvm_lapic_set_reg(apic, APIC_TDCR, val & 0xb);
2207		update_divide_count(apic);
2208		if (apic->divide_count != old_divisor &&
2209				apic->lapic_timer.period) {
2210			hrtimer_cancel(&apic->lapic_timer.timer);
2211			update_target_expiration(apic, old_divisor);
2212			restart_apic_timer(apic);
2213		}
2214		break;
2215	}
2216	case APIC_ESR:
2217		if (apic_x2apic_mode(apic) && val != 0)
2218			ret = 1;
2219		break;
2220
2221	case APIC_SELF_IPI:
2222		if (apic_x2apic_mode(apic))
2223			kvm_apic_send_ipi(apic, APIC_DEST_SELF | (val & APIC_VECTOR_MASK), 0);
2224		else
 
2225			ret = 1;
2226		break;
2227	default:
2228		ret = 1;
2229		break;
2230	}
2231
2232	/*
2233	 * Recalculate APIC maps if necessary, e.g. if the software enable bit
2234	 * was toggled, the APIC ID changed, etc...   The maps are marked dirty
2235	 * on relevant changes, i.e. this is a nop for most writes.
2236	 */
2237	kvm_recalculate_apic_map(apic->vcpu->kvm);
2238
2239	return ret;
2240}
 
2241
2242static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
2243			    gpa_t address, int len, const void *data)
2244{
2245	struct kvm_lapic *apic = to_lapic(this);
2246	unsigned int offset = address - apic->base_address;
2247	u32 val;
2248
2249	if (!apic_mmio_in_range(apic, address))
2250		return -EOPNOTSUPP;
2251
2252	if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
2253		if (!kvm_check_has_quirk(vcpu->kvm,
2254					 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
2255			return -EOPNOTSUPP;
2256
2257		return 0;
2258	}
2259
2260	/*
2261	 * APIC register must be aligned on 128-bits boundary.
2262	 * 32/64/128 bits registers must be accessed thru 32 bits.
2263	 * Refer SDM 8.4.1
2264	 */
2265	if (len != 4 || (offset & 0xf))
2266		return 0;
2267
2268	val = *(u32*)data;
2269
2270	kvm_lapic_reg_write(apic, offset & 0xff0, val);
2271
2272	return 0;
2273}
2274
2275void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
2276{
2277	kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
2278}
2279EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
2280
2281/* emulate APIC access in a trap manner */
2282void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
2283{
2284	struct kvm_lapic *apic = vcpu->arch.apic;
2285	u64 val;
2286
2287	if (apic_x2apic_mode(apic)) {
2288		if (KVM_BUG_ON(kvm_lapic_msr_read(apic, offset, &val), vcpu->kvm))
2289			return;
2290	} else {
2291		val = kvm_lapic_get_reg(apic, offset);
2292	}
2293
2294	/*
2295	 * ICR is a single 64-bit register when x2APIC is enabled.  For legacy
2296	 * xAPIC, ICR writes need to go down the common (slightly slower) path
2297	 * to get the upper half from ICR2.
2298	 */
2299	if (apic_x2apic_mode(apic) && offset == APIC_ICR) {
2300		kvm_apic_send_ipi(apic, (u32)val, (u32)(val >> 32));
2301		trace_kvm_apic_write(APIC_ICR, val);
2302	} else {
2303		/* TODO: optimize to just emulate side effect w/o one more write */
2304		kvm_lapic_reg_write(apic, offset, (u32)val);
2305	}
2306}
2307EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
2308
2309void kvm_free_lapic(struct kvm_vcpu *vcpu)
2310{
2311	struct kvm_lapic *apic = vcpu->arch.apic;
2312
2313	if (!vcpu->arch.apic)
2314		return;
2315
2316	hrtimer_cancel(&apic->lapic_timer.timer);
2317
2318	if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
2319		static_branch_slow_dec_deferred(&apic_hw_disabled);
2320
2321	if (!apic->sw_enabled)
2322		static_branch_slow_dec_deferred(&apic_sw_disabled);
2323
2324	if (apic->regs)
2325		free_page((unsigned long)apic->regs);
2326
2327	kfree(apic);
2328}
2329
2330/*
2331 *----------------------------------------------------------------------
2332 * LAPIC interface
2333 *----------------------------------------------------------------------
2334 */
2335u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
2336{
2337	struct kvm_lapic *apic = vcpu->arch.apic;
2338
2339	if (!kvm_apic_present(vcpu) || !apic_lvtt_tscdeadline(apic))
 
2340		return 0;
2341
2342	return apic->lapic_timer.tscdeadline;
2343}
2344
2345void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
2346{
2347	struct kvm_lapic *apic = vcpu->arch.apic;
2348
2349	if (!kvm_apic_present(vcpu) || !apic_lvtt_tscdeadline(apic))
 
2350		return;
2351
2352	hrtimer_cancel(&apic->lapic_timer.timer);
2353	apic->lapic_timer.tscdeadline = data;
2354	start_apic_timer(apic);
2355}
2356
2357void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
2358{
2359	apic_set_tpr(vcpu->arch.apic, (cr8 & 0x0f) << 4);
 
 
 
2360}
2361
2362u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
2363{
2364	u64 tpr;
2365
2366	tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
2367
2368	return (tpr & 0xf0) >> 4;
2369}
2370
2371void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
2372{
2373	u64 old_value = vcpu->arch.apic_base;
2374	struct kvm_lapic *apic = vcpu->arch.apic;
2375
 
 
 
2376	vcpu->arch.apic_base = value;
2377
2378	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
2379		kvm_update_cpuid_runtime(vcpu);
2380
2381	if (!apic)
2382		return;
2383
2384	/* update jump label if enable bit changes */
2385	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
2386		if (value & MSR_IA32_APICBASE_ENABLE) {
2387			kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2388			static_branch_slow_dec_deferred(&apic_hw_disabled);
2389			/* Check if there are APF page ready requests pending */
2390			kvm_make_request(KVM_REQ_APF_READY, vcpu);
2391		} else {
2392			static_branch_inc(&apic_hw_disabled.key);
2393			atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
2394		}
2395	}
2396
2397	if (((old_value ^ value) & X2APIC_ENABLE) && (value & X2APIC_ENABLE))
2398		kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
2399
2400	if ((old_value ^ value) & (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) {
2401		kvm_vcpu_update_apicv(vcpu);
2402		static_call_cond(kvm_x86_set_virtual_apic_mode)(vcpu);
2403	}
2404
2405	apic->base_address = apic->vcpu->arch.apic_base &
2406			     MSR_IA32_APICBASE_BASE;
2407
2408	if ((value & MSR_IA32_APICBASE_ENABLE) &&
2409	     apic->base_address != APIC_DEFAULT_PHYS_BASE) {
2410		kvm_set_apicv_inhibit(apic->vcpu->kvm,
2411				      APICV_INHIBIT_REASON_APIC_BASE_MODIFIED);
2412	}
2413}
2414
2415void kvm_apic_update_apicv(struct kvm_vcpu *vcpu)
2416{
2417	struct kvm_lapic *apic = vcpu->arch.apic;
2418
2419	if (apic->apicv_active) {
2420		/* irr_pending is always true when apicv is activated. */
2421		apic->irr_pending = true;
2422		apic->isr_count = 1;
2423	} else {
2424		/*
2425		 * Don't clear irr_pending, searching the IRR can race with
2426		 * updates from the CPU as APICv is still active from hardware's
2427		 * perspective.  The flag will be cleared as appropriate when
2428		 * KVM injects the interrupt.
2429		 */
2430		apic->isr_count = count_vectors(apic->regs + APIC_ISR);
2431	}
2432}
 
2433
2434void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
2435{
2436	struct kvm_lapic *apic = vcpu->arch.apic;
2437	u64 msr_val;
2438	int i;
2439
2440	if (!init_event) {
2441		msr_val = APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE;
2442		if (kvm_vcpu_is_reset_bsp(vcpu))
2443			msr_val |= MSR_IA32_APICBASE_BSP;
2444		kvm_lapic_set_base(vcpu, msr_val);
2445	}
2446
2447	if (!apic)
2448		return;
2449
2450	/* Stop the timer in case it's a reset to an active apic */
2451	hrtimer_cancel(&apic->lapic_timer.timer);
2452
2453	/* The xAPIC ID is set at RESET even if the APIC was already enabled. */
2454	if (!init_event)
 
2455		kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
 
2456	kvm_apic_set_version(apic->vcpu);
2457
2458	for (i = 0; i < apic->nr_lvt_entries; i++)
2459		kvm_lapic_set_reg(apic, APIC_LVTx(i), APIC_LVT_MASKED);
2460	apic_update_lvtt(apic);
2461	if (kvm_vcpu_is_reset_bsp(vcpu) &&
2462	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
2463		kvm_lapic_set_reg(apic, APIC_LVT0,
2464			     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
2465	apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2466
2467	kvm_apic_set_dfr(apic, 0xffffffffU);
2468	apic_set_spiv(apic, 0xff);
2469	kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
2470	if (!apic_x2apic_mode(apic))
2471		kvm_apic_set_ldr(apic, 0);
2472	kvm_lapic_set_reg(apic, APIC_ESR, 0);
2473	if (!apic_x2apic_mode(apic)) {
2474		kvm_lapic_set_reg(apic, APIC_ICR, 0);
2475		kvm_lapic_set_reg(apic, APIC_ICR2, 0);
2476	} else {
2477		kvm_lapic_set_reg64(apic, APIC_ICR, 0);
2478	}
2479	kvm_lapic_set_reg(apic, APIC_TDCR, 0);
2480	kvm_lapic_set_reg(apic, APIC_TMICT, 0);
2481	for (i = 0; i < 8; i++) {
2482		kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
2483		kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
2484		kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
2485	}
2486	kvm_apic_update_apicv(vcpu);
2487	apic->highest_isr_cache = -1;
2488	update_divide_count(apic);
2489	atomic_set(&apic->lapic_timer.pending, 0);
2490
 
 
2491	vcpu->arch.pv_eoi.msr_val = 0;
2492	apic_update_ppr(apic);
2493	if (apic->apicv_active) {
2494		static_call_cond(kvm_x86_apicv_post_state_restore)(vcpu);
2495		static_call_cond(kvm_x86_hwapic_irr_update)(vcpu, -1);
2496		static_call_cond(kvm_x86_hwapic_isr_update)(-1);
2497	}
2498
2499	vcpu->arch.apic_arb_prio = 0;
2500	vcpu->arch.apic_attention = 0;
2501
2502	kvm_recalculate_apic_map(vcpu->kvm);
2503}
2504
2505/*
2506 *----------------------------------------------------------------------
2507 * timer interface
2508 *----------------------------------------------------------------------
2509 */
2510
2511static bool lapic_is_periodic(struct kvm_lapic *apic)
2512{
2513	return apic_lvtt_period(apic);
2514}
2515
2516int apic_has_pending_timer(struct kvm_vcpu *vcpu)
2517{
2518	struct kvm_lapic *apic = vcpu->arch.apic;
2519
2520	if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
2521		return atomic_read(&apic->lapic_timer.pending);
2522
2523	return 0;
2524}
2525
2526int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
2527{
2528	u32 reg = kvm_lapic_get_reg(apic, lvt_type);
2529	int vector, mode, trig_mode;
2530
2531	if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
2532		vector = reg & APIC_VECTOR_MASK;
2533		mode = reg & APIC_MODE_MASK;
2534		trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
2535		return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
2536					NULL);
2537	}
2538	return 0;
2539}
2540
2541void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
2542{
2543	struct kvm_lapic *apic = vcpu->arch.apic;
2544
2545	if (apic)
2546		kvm_apic_local_deliver(apic, APIC_LVT0);
2547}
2548
2549static const struct kvm_io_device_ops apic_mmio_ops = {
2550	.read     = apic_mmio_read,
2551	.write    = apic_mmio_write,
2552};
2553
2554static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
2555{
2556	struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
2557	struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
2558
2559	apic_timer_expired(apic, true);
2560
2561	if (lapic_is_periodic(apic)) {
2562		advance_periodic_target_expiration(apic);
2563		hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
2564		return HRTIMER_RESTART;
2565	} else
2566		return HRTIMER_NORESTART;
2567}
2568
2569int kvm_create_lapic(struct kvm_vcpu *vcpu, int timer_advance_ns)
2570{
2571	struct kvm_lapic *apic;
2572
2573	ASSERT(vcpu != NULL);
2574
2575	apic = kzalloc(sizeof(*apic), GFP_KERNEL_ACCOUNT);
2576	if (!apic)
2577		goto nomem;
2578
2579	vcpu->arch.apic = apic;
2580
2581	apic->regs = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
2582	if (!apic->regs) {
2583		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
2584		       vcpu->vcpu_id);
2585		goto nomem_free_apic;
2586	}
2587	apic->vcpu = vcpu;
2588
2589	apic->nr_lvt_entries = kvm_apic_calc_nr_lvt_entries(vcpu);
2590
2591	hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2592		     HRTIMER_MODE_ABS_HARD);
2593	apic->lapic_timer.timer.function = apic_timer_fn;
2594	if (timer_advance_ns == -1) {
2595		apic->lapic_timer.timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
2596		lapic_timer_advance_dynamic = true;
2597	} else {
2598		apic->lapic_timer.timer_advance_ns = timer_advance_ns;
2599		lapic_timer_advance_dynamic = false;
2600	}
2601
2602	/*
2603	 * Stuff the APIC ENABLE bit in lieu of temporarily incrementing
2604	 * apic_hw_disabled; the full RESET value is set by kvm_lapic_reset().
2605	 */
2606	vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2607	static_branch_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2608	kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2609
2610	return 0;
2611nomem_free_apic:
2612	kfree(apic);
2613	vcpu->arch.apic = NULL;
2614nomem:
2615	return -ENOMEM;
2616}
2617
2618int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2619{
2620	struct kvm_lapic *apic = vcpu->arch.apic;
2621	u32 ppr;
2622
2623	if (!kvm_apic_present(vcpu))
2624		return -1;
2625
2626	__apic_update_ppr(apic, &ppr);
2627	return apic_has_interrupt_for_ppr(apic, ppr);
2628}
2629EXPORT_SYMBOL_GPL(kvm_apic_has_interrupt);
2630
2631int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2632{
2633	u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2634
2635	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2636		return 1;
2637	if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2638	    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2639		return 1;
2640	return 0;
2641}
2642
2643void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2644{
2645	struct kvm_lapic *apic = vcpu->arch.apic;
2646
2647	if (atomic_read(&apic->lapic_timer.pending) > 0) {
2648		kvm_apic_inject_pending_timer_irqs(apic);
2649		atomic_set(&apic->lapic_timer.pending, 0);
2650	}
2651}
2652
2653int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2654{
2655	int vector = kvm_apic_has_interrupt(vcpu);
2656	struct kvm_lapic *apic = vcpu->arch.apic;
2657	u32 ppr;
2658
2659	if (vector == -1)
2660		return -1;
2661
2662	/*
2663	 * We get here even with APIC virtualization enabled, if doing
2664	 * nested virtualization and L1 runs with the "acknowledge interrupt
2665	 * on exit" mode.  Then we cannot inject the interrupt via RVI,
2666	 * because the process would deliver it through the IDT.
2667	 */
2668
2669	apic_clear_irr(vector, apic);
2670	if (to_hv_vcpu(vcpu) && test_bit(vector, to_hv_synic(vcpu)->auto_eoi_bitmap)) {
2671		/*
2672		 * For auto-EOI interrupts, there might be another pending
2673		 * interrupt above PPR, so check whether to raise another
2674		 * KVM_REQ_EVENT.
2675		 */
2676		apic_update_ppr(apic);
2677	} else {
2678		/*
2679		 * For normal interrupts, PPR has been raised and there cannot
2680		 * be a higher-priority pending interrupt---except if there was
2681		 * a concurrent interrupt injection, but that would have
2682		 * triggered KVM_REQ_EVENT already.
2683		 */
2684		apic_set_isr(vector, apic);
2685		__apic_update_ppr(apic, &ppr);
2686	}
2687
2688	return vector;
2689}
2690
2691static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2692		struct kvm_lapic_state *s, bool set)
2693{
2694	if (apic_x2apic_mode(vcpu->arch.apic)) {
2695		u32 *id = (u32 *)(s->regs + APIC_ID);
2696		u32 *ldr = (u32 *)(s->regs + APIC_LDR);
2697		u64 icr;
2698
2699		if (vcpu->kvm->arch.x2apic_format) {
2700			if (*id != vcpu->vcpu_id)
2701				return -EINVAL;
2702		} else {
2703			if (set)
2704				*id >>= 24;
2705			else
2706				*id <<= 24;
2707		}
2708
2709		/*
2710		 * In x2APIC mode, the LDR is fixed and based on the id.  And
2711		 * ICR is internally a single 64-bit register, but needs to be
2712		 * split to ICR+ICR2 in userspace for backwards compatibility.
2713		 */
2714		if (set) {
2715			*ldr = kvm_apic_calc_x2apic_ldr(*id);
2716
2717			icr = __kvm_lapic_get_reg(s->regs, APIC_ICR) |
2718			      (u64)__kvm_lapic_get_reg(s->regs, APIC_ICR2) << 32;
2719			__kvm_lapic_set_reg64(s->regs, APIC_ICR, icr);
2720		} else {
2721			icr = __kvm_lapic_get_reg64(s->regs, APIC_ICR);
2722			__kvm_lapic_set_reg(s->regs, APIC_ICR2, icr >> 32);
2723		}
2724	}
2725
2726	return 0;
2727}
2728
2729int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2730{
2731	memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2732
2733	/*
2734	 * Get calculated timer current count for remaining timer period (if
2735	 * any) and store it in the returned register set.
2736	 */
2737	__kvm_lapic_set_reg(s->regs, APIC_TMCCT,
2738			    __apic_read(vcpu->arch.apic, APIC_TMCCT));
2739
2740	return kvm_apic_state_fixup(vcpu, s, false);
2741}
2742
2743int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2744{
2745	struct kvm_lapic *apic = vcpu->arch.apic;
2746	int r;
2747
2748	kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2749	/* set SPIV separately to get count of SW disabled APICs right */
2750	apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2751
2752	r = kvm_apic_state_fixup(vcpu, s, true);
2753	if (r) {
2754		kvm_recalculate_apic_map(vcpu->kvm);
2755		return r;
2756	}
2757	memcpy(vcpu->arch.apic->regs, s->regs, sizeof(*s));
2758
2759	if (!apic_x2apic_mode(apic))
2760		kvm_lapic_xapic_id_updated(apic);
2761
2762	atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
2763	kvm_recalculate_apic_map(vcpu->kvm);
2764	kvm_apic_set_version(vcpu);
2765
2766	apic_update_ppr(apic);
2767	cancel_apic_timer(apic);
2768	apic->lapic_timer.expired_tscdeadline = 0;
2769	apic_update_lvtt(apic);
2770	apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2771	update_divide_count(apic);
2772	__start_apic_timer(apic, APIC_TMCCT);
2773	kvm_lapic_set_reg(apic, APIC_TMCCT, 0);
2774	kvm_apic_update_apicv(vcpu);
2775	apic->highest_isr_cache = -1;
2776	if (apic->apicv_active) {
2777		static_call_cond(kvm_x86_apicv_post_state_restore)(vcpu);
2778		static_call_cond(kvm_x86_hwapic_irr_update)(vcpu, apic_find_highest_irr(apic));
2779		static_call_cond(kvm_x86_hwapic_isr_update)(apic_find_highest_isr(apic));
 
 
2780	}
2781	kvm_make_request(KVM_REQ_EVENT, vcpu);
2782	if (ioapic_in_kernel(vcpu->kvm))
2783		kvm_rtc_eoi_tracking_restore_one(vcpu);
2784
2785	vcpu->arch.apic_arb_prio = 0;
2786
2787	return 0;
2788}
2789
2790void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2791{
2792	struct hrtimer *timer;
2793
2794	if (!lapic_in_kernel(vcpu) ||
2795		kvm_can_post_timer_interrupt(vcpu))
2796		return;
2797
2798	timer = &vcpu->arch.apic->lapic_timer.timer;
2799	if (hrtimer_cancel(timer))
2800		hrtimer_start_expires(timer, HRTIMER_MODE_ABS_HARD);
2801}
2802
2803/*
2804 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2805 *
2806 * Detect whether guest triggered PV EOI since the
2807 * last entry. If yes, set EOI on guests's behalf.
2808 * Clear PV EOI in guest memory in any case.
2809 */
2810static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2811					struct kvm_lapic *apic)
2812{
 
2813	int vector;
2814	/*
2815	 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2816	 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2817	 *
2818	 * KVM_APIC_PV_EOI_PENDING is unset:
2819	 * 	-> host disabled PV EOI.
2820	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2821	 * 	-> host enabled PV EOI, guest did not execute EOI yet.
2822	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2823	 * 	-> host enabled PV EOI, guest executed EOI.
2824	 */
2825	BUG_ON(!pv_eoi_enabled(vcpu));
2826
2827	if (pv_eoi_test_and_clr_pending(vcpu))
 
 
 
 
 
 
2828		return;
2829	vector = apic_set_eoi(apic);
2830	trace_kvm_pv_eoi(apic, vector);
2831}
2832
2833void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2834{
2835	u32 data;
2836
2837	if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2838		apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2839
2840	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2841		return;
2842
2843	if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2844				  sizeof(u32)))
2845		return;
2846
2847	apic_set_tpr(vcpu->arch.apic, data & 0xff);
2848}
2849
2850/*
2851 * apic_sync_pv_eoi_to_guest - called before vmentry
2852 *
2853 * Detect whether it's safe to enable PV EOI and
2854 * if yes do so.
2855 */
2856static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2857					struct kvm_lapic *apic)
2858{
2859	if (!pv_eoi_enabled(vcpu) ||
2860	    /* IRR set or many bits in ISR: could be nested. */
2861	    apic->irr_pending ||
2862	    /* Cache not set: could be safe but we don't bother. */
2863	    apic->highest_isr_cache == -1 ||
2864	    /* Need EOI to update ioapic. */
2865	    kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2866		/*
2867		 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2868		 * so we need not do anything here.
2869		 */
2870		return;
2871	}
2872
2873	pv_eoi_set_pending(apic->vcpu);
2874}
2875
2876void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2877{
2878	u32 data, tpr;
2879	int max_irr, max_isr;
2880	struct kvm_lapic *apic = vcpu->arch.apic;
2881
2882	apic_sync_pv_eoi_to_guest(vcpu, apic);
2883
2884	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2885		return;
2886
2887	tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
2888	max_irr = apic_find_highest_irr(apic);
2889	if (max_irr < 0)
2890		max_irr = 0;
2891	max_isr = apic_find_highest_isr(apic);
2892	if (max_isr < 0)
2893		max_isr = 0;
2894	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2895
2896	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2897				sizeof(u32));
2898}
2899
2900int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2901{
2902	if (vapic_addr) {
2903		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2904					&vcpu->arch.apic->vapic_cache,
2905					vapic_addr, sizeof(u32)))
2906			return -EINVAL;
2907		__set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2908	} else {
2909		__clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2910	}
2911
2912	vcpu->arch.apic->vapic_addr = vapic_addr;
2913	return 0;
2914}
2915
2916int kvm_x2apic_icr_write(struct kvm_lapic *apic, u64 data)
2917{
2918	data &= ~APIC_ICR_BUSY;
2919
2920	kvm_apic_send_ipi(apic, (u32)data, (u32)(data >> 32));
2921	kvm_lapic_set_reg64(apic, APIC_ICR, data);
2922	trace_kvm_apic_write(APIC_ICR, data);
2923	return 0;
2924}
2925
2926static int kvm_lapic_msr_read(struct kvm_lapic *apic, u32 reg, u64 *data)
2927{
2928	u32 low;
2929
2930	if (reg == APIC_ICR) {
2931		*data = kvm_lapic_get_reg64(apic, APIC_ICR);
2932		return 0;
2933	}
2934
2935	if (kvm_lapic_reg_read(apic, reg, 4, &low))
2936		return 1;
2937
2938	*data = low;
2939
2940	return 0;
2941}
2942
2943static int kvm_lapic_msr_write(struct kvm_lapic *apic, u32 reg, u64 data)
2944{
2945	/*
2946	 * ICR is a 64-bit register in x2APIC mode (and Hyper'v PV vAPIC) and
2947	 * can be written as such, all other registers remain accessible only
2948	 * through 32-bit reads/writes.
2949	 */
2950	if (reg == APIC_ICR)
2951		return kvm_x2apic_icr_write(apic, data);
2952
2953	return kvm_lapic_reg_write(apic, reg, (u32)data);
2954}
2955
2956int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2957{
2958	struct kvm_lapic *apic = vcpu->arch.apic;
2959	u32 reg = (msr - APIC_BASE_MSR) << 4;
2960
2961	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2962		return 1;
2963
2964	return kvm_lapic_msr_write(apic, reg, data);
 
 
 
 
 
 
2965}
2966
2967int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2968{
2969	struct kvm_lapic *apic = vcpu->arch.apic;
2970	u32 reg = (msr - APIC_BASE_MSR) << 4;
2971
2972	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2973		return 1;
2974
2975	if (reg == APIC_DFR)
 
 
 
2976		return 1;
 
 
 
 
2977
2978	return kvm_lapic_msr_read(apic, reg, data);
2979}
2980
2981int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2982{
 
 
2983	if (!lapic_in_kernel(vcpu))
2984		return 1;
2985
2986	return kvm_lapic_msr_write(vcpu->arch.apic, reg, data);
 
 
 
2987}
2988
2989int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2990{
 
 
 
2991	if (!lapic_in_kernel(vcpu))
2992		return 1;
2993
2994	return kvm_lapic_msr_read(vcpu->arch.apic, reg, data);
 
 
 
 
 
 
 
2995}
2996
2997int kvm_lapic_set_pv_eoi(struct kvm_vcpu *vcpu, u64 data, unsigned long len)
2998{
2999	u64 addr = data & ~KVM_MSR_ENABLED;
3000	struct gfn_to_hva_cache *ghc = &vcpu->arch.pv_eoi.data;
3001	unsigned long new_len;
3002	int ret;
3003
3004	if (!IS_ALIGNED(addr, 4))
3005		return 1;
3006
3007	if (data & KVM_MSR_ENABLED) {
3008		if (addr == ghc->gpa && len <= ghc->len)
3009			new_len = ghc->len;
3010		else
3011			new_len = len;
3012
3013		ret = kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, addr, new_len);
3014		if (ret)
3015			return ret;
3016	}
3017
3018	vcpu->arch.pv_eoi.msr_val = data;
 
 
3019
3020	return 0;
 
 
 
 
 
3021}
3022
3023int kvm_apic_accept_events(struct kvm_vcpu *vcpu)
3024{
3025	struct kvm_lapic *apic = vcpu->arch.apic;
3026	u8 sipi_vector;
3027	int r;
3028
3029	if (!kvm_apic_has_pending_init_or_sipi(vcpu))
3030		return 0;
3031
3032	if (is_guest_mode(vcpu)) {
3033		r = kvm_check_nested_events(vcpu);
3034		if (r < 0)
3035			return r == -EBUSY ? 0 : r;
3036		/*
3037		 * Continue processing INIT/SIPI even if a nested VM-Exit
3038		 * occurred, e.g. pending SIPIs should be dropped if INIT+SIPI
3039		 * are blocked as a result of transitioning to VMX root mode.
3040		 */
3041	}
3042
3043	/*
3044	 * INITs are blocked while CPU is in specific states (SMM, VMX root
3045	 * mode, SVM with GIF=0), while SIPIs are dropped if the CPU isn't in
3046	 * wait-for-SIPI (WFS).
 
 
 
3047	 */
3048	if (!kvm_apic_init_sipi_allowed(vcpu)) {
3049		WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
3050		clear_bit(KVM_APIC_SIPI, &apic->pending_events);
3051		return 0;
 
3052	}
3053
3054	if (test_and_clear_bit(KVM_APIC_INIT, &apic->pending_events)) {
 
3055		kvm_vcpu_reset(vcpu, true);
3056		if (kvm_vcpu_is_bsp(apic->vcpu))
3057			vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3058		else
3059			vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
3060	}
3061	if (test_and_clear_bit(KVM_APIC_SIPI, &apic->pending_events)) {
3062		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
3063			/* evaluate pending_events before reading the vector */
3064			smp_rmb();
3065			sipi_vector = apic->sipi_vector;
3066			static_call(kvm_x86_vcpu_deliver_sipi_vector)(vcpu, sipi_vector);
3067			vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3068		}
3069	}
3070	return 0;
 
 
 
 
 
 
3071}
3072
3073void kvm_lapic_exit(void)
3074{
3075	static_key_deferred_flush(&apic_hw_disabled);
3076	WARN_ON(static_branch_unlikely(&apic_hw_disabled.key));
3077	static_key_deferred_flush(&apic_sw_disabled);
3078	WARN_ON(static_branch_unlikely(&apic_sw_disabled.key));
3079}
v5.9
   1// SPDX-License-Identifier: GPL-2.0-only
   2
   3/*
   4 * Local APIC virtualization
   5 *
   6 * Copyright (C) 2006 Qumranet, Inc.
   7 * Copyright (C) 2007 Novell
   8 * Copyright (C) 2007 Intel
   9 * Copyright 2009 Red Hat, Inc. and/or its affiliates.
  10 *
  11 * Authors:
  12 *   Dor Laor <dor.laor@qumranet.com>
  13 *   Gregory Haskins <ghaskins@novell.com>
  14 *   Yaozu (Eddie) Dong <eddie.dong@intel.com>
  15 *
  16 * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
  17 */
  18
  19#include <linux/kvm_host.h>
  20#include <linux/kvm.h>
  21#include <linux/mm.h>
  22#include <linux/highmem.h>
  23#include <linux/smp.h>
  24#include <linux/hrtimer.h>
  25#include <linux/io.h>
  26#include <linux/export.h>
  27#include <linux/math64.h>
  28#include <linux/slab.h>
  29#include <asm/processor.h>
 
  30#include <asm/msr.h>
  31#include <asm/page.h>
  32#include <asm/current.h>
  33#include <asm/apicdef.h>
  34#include <asm/delay.h>
  35#include <linux/atomic.h>
  36#include <linux/jump_label.h>
  37#include "kvm_cache_regs.h"
  38#include "irq.h"
  39#include "ioapic.h"
  40#include "trace.h"
  41#include "x86.h"
  42#include "cpuid.h"
  43#include "hyperv.h"
 
  44
  45#ifndef CONFIG_X86_64
  46#define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
  47#else
  48#define mod_64(x, y) ((x) % (y))
  49#endif
  50
  51#define PRId64 "d"
  52#define PRIx64 "llx"
  53#define PRIu64 "u"
  54#define PRIo64 "o"
  55
  56/* 14 is the version for Xeon and Pentium 8.4.8*/
  57#define APIC_VERSION			(0x14UL | ((KVM_APIC_LVT_NUM - 1) << 16))
  58#define LAPIC_MMIO_LENGTH		(1 << 12)
  59/* followed define is not in apicdef.h */
  60#define MAX_APIC_VECTOR			256
  61#define APIC_VECTORS_PER_REG		32
  62
  63static bool lapic_timer_advance_dynamic __read_mostly;
  64#define LAPIC_TIMER_ADVANCE_ADJUST_MIN	100	/* clock cycles */
  65#define LAPIC_TIMER_ADVANCE_ADJUST_MAX	10000	/* clock cycles */
  66#define LAPIC_TIMER_ADVANCE_NS_INIT	1000
  67#define LAPIC_TIMER_ADVANCE_NS_MAX     5000
  68/* step-by-step approximation to mitigate fluctuation */
  69#define LAPIC_TIMER_ADVANCE_ADJUST_STEP 8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  70
  71static inline int apic_test_vector(int vec, void *bitmap)
  72{
  73	return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
  74}
  75
  76bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
  77{
  78	struct kvm_lapic *apic = vcpu->arch.apic;
  79
  80	return apic_test_vector(vector, apic->regs + APIC_ISR) ||
  81		apic_test_vector(vector, apic->regs + APIC_IRR);
  82}
  83
  84static inline int __apic_test_and_set_vector(int vec, void *bitmap)
  85{
  86	return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
  87}
  88
  89static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
  90{
  91	return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
  92}
  93
  94struct static_key_deferred apic_hw_disabled __read_mostly;
  95struct static_key_deferred apic_sw_disabled __read_mostly;
  96
  97static inline int apic_enabled(struct kvm_lapic *apic)
  98{
  99	return kvm_apic_sw_enabled(apic) &&	kvm_apic_hw_enabled(apic);
 100}
 101
 102#define LVT_MASK	\
 103	(APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
 104
 105#define LINT_MASK	\
 106	(LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
 107	 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
 108
 109static inline u32 kvm_x2apic_id(struct kvm_lapic *apic)
 110{
 111	return apic->vcpu->vcpu_id;
 112}
 113
 114static bool kvm_can_post_timer_interrupt(struct kvm_vcpu *vcpu)
 115{
 116	return pi_inject_timer && kvm_vcpu_apicv_active(vcpu);
 
 117}
 118
 119bool kvm_can_use_hv_timer(struct kvm_vcpu *vcpu)
 120{
 121	return kvm_x86_ops.set_hv_timer
 122	       && !(kvm_mwait_in_guest(vcpu->kvm) ||
 123		    kvm_can_post_timer_interrupt(vcpu));
 124}
 125EXPORT_SYMBOL_GPL(kvm_can_use_hv_timer);
 126
 127static bool kvm_use_posted_timer_interrupt(struct kvm_vcpu *vcpu)
 128{
 129	return kvm_can_post_timer_interrupt(vcpu) && vcpu->mode == IN_GUEST_MODE;
 130}
 131
 132static inline bool kvm_apic_map_get_logical_dest(struct kvm_apic_map *map,
 133		u32 dest_id, struct kvm_lapic ***cluster, u16 *mask) {
 134	switch (map->mode) {
 135	case KVM_APIC_MODE_X2APIC: {
 136		u32 offset = (dest_id >> 16) * 16;
 137		u32 max_apic_id = map->max_apic_id;
 138
 139		if (offset <= max_apic_id) {
 140			u8 cluster_size = min(max_apic_id - offset + 1, 16U);
 141
 142			offset = array_index_nospec(offset, map->max_apic_id + 1);
 143			*cluster = &map->phys_map[offset];
 144			*mask = dest_id & (0xffff >> (16 - cluster_size));
 145		} else {
 146			*mask = 0;
 147		}
 148
 149		return true;
 150		}
 151	case KVM_APIC_MODE_XAPIC_FLAT:
 152		*cluster = map->xapic_flat_map;
 153		*mask = dest_id & 0xff;
 154		return true;
 155	case KVM_APIC_MODE_XAPIC_CLUSTER:
 156		*cluster = map->xapic_cluster_map[(dest_id >> 4) & 0xf];
 157		*mask = dest_id & 0xf;
 158		return true;
 159	default:
 160		/* Not optimized. */
 161		return false;
 162	}
 163}
 164
 165static void kvm_apic_map_free(struct rcu_head *rcu)
 166{
 167	struct kvm_apic_map *map = container_of(rcu, struct kvm_apic_map, rcu);
 168
 169	kvfree(map);
 170}
 171
 172/*
 173 * CLEAN -> DIRTY and UPDATE_IN_PROGRESS -> DIRTY changes happen without a lock.
 174 *
 175 * DIRTY -> UPDATE_IN_PROGRESS and UPDATE_IN_PROGRESS -> CLEAN happen with
 176 * apic_map_lock_held.
 177 */
 178enum {
 179	CLEAN,
 180	UPDATE_IN_PROGRESS,
 181	DIRTY
 182};
 183
 184void kvm_recalculate_apic_map(struct kvm *kvm)
 185{
 186	struct kvm_apic_map *new, *old = NULL;
 187	struct kvm_vcpu *vcpu;
 188	int i;
 189	u32 max_id = 255; /* enough space for any xAPIC ID */
 190
 191	/* Read kvm->arch.apic_map_dirty before kvm->arch.apic_map.  */
 192	if (atomic_read_acquire(&kvm->arch.apic_map_dirty) == CLEAN)
 193		return;
 194
 
 
 
 195	mutex_lock(&kvm->arch.apic_map_lock);
 196	/*
 197	 * Read kvm->arch.apic_map_dirty before kvm->arch.apic_map
 198	 * (if clean) or the APIC registers (if dirty).
 199	 */
 200	if (atomic_cmpxchg_acquire(&kvm->arch.apic_map_dirty,
 201				   DIRTY, UPDATE_IN_PROGRESS) == CLEAN) {
 202		/* Someone else has updated the map. */
 203		mutex_unlock(&kvm->arch.apic_map_lock);
 204		return;
 205	}
 206
 207	kvm_for_each_vcpu(i, vcpu, kvm)
 208		if (kvm_apic_present(vcpu))
 209			max_id = max(max_id, kvm_x2apic_id(vcpu->arch.apic));
 210
 211	new = kvzalloc(sizeof(struct kvm_apic_map) +
 212	                   sizeof(struct kvm_lapic *) * ((u64)max_id + 1),
 213			   GFP_KERNEL_ACCOUNT);
 214
 215	if (!new)
 216		goto out;
 217
 218	new->max_apic_id = max_id;
 219
 220	kvm_for_each_vcpu(i, vcpu, kvm) {
 221		struct kvm_lapic *apic = vcpu->arch.apic;
 222		struct kvm_lapic **cluster;
 223		u16 mask;
 224		u32 ldr;
 225		u8 xapic_id;
 226		u32 x2apic_id;
 227
 228		if (!kvm_apic_present(vcpu))
 229			continue;
 230
 231		xapic_id = kvm_xapic_id(apic);
 232		x2apic_id = kvm_x2apic_id(apic);
 233
 234		/* Hotplug hack: see kvm_apic_match_physical_addr(), ... */
 235		if ((apic_x2apic_mode(apic) || x2apic_id > 0xff) &&
 236				x2apic_id <= new->max_apic_id)
 237			new->phys_map[x2apic_id] = apic;
 238		/*
 239		 * ... xAPIC ID of VCPUs with APIC ID > 0xff will wrap-around,
 240		 * prevent them from masking VCPUs with APIC ID <= 0xff.
 241		 */
 242		if (!apic_x2apic_mode(apic) && !new->phys_map[xapic_id])
 243			new->phys_map[xapic_id] = apic;
 244
 245		if (!kvm_apic_sw_enabled(apic))
 246			continue;
 247
 248		ldr = kvm_lapic_get_reg(apic, APIC_LDR);
 249
 250		if (apic_x2apic_mode(apic)) {
 251			new->mode |= KVM_APIC_MODE_X2APIC;
 252		} else if (ldr) {
 253			ldr = GET_APIC_LOGICAL_ID(ldr);
 254			if (kvm_lapic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
 255				new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
 256			else
 257				new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
 258		}
 259
 260		if (!kvm_apic_map_get_logical_dest(new, ldr, &cluster, &mask))
 261			continue;
 262
 263		if (mask)
 264			cluster[ffs(mask) - 1] = apic;
 265	}
 266out:
 267	old = rcu_dereference_protected(kvm->arch.apic_map,
 268			lockdep_is_held(&kvm->arch.apic_map_lock));
 269	rcu_assign_pointer(kvm->arch.apic_map, new);
 270	/*
 271	 * Write kvm->arch.apic_map before clearing apic->apic_map_dirty.
 272	 * If another update has come in, leave it DIRTY.
 273	 */
 274	atomic_cmpxchg_release(&kvm->arch.apic_map_dirty,
 275			       UPDATE_IN_PROGRESS, CLEAN);
 276	mutex_unlock(&kvm->arch.apic_map_lock);
 277
 278	if (old)
 279		call_rcu(&old->rcu, kvm_apic_map_free);
 280
 281	kvm_make_scan_ioapic_request(kvm);
 282}
 283
 284static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
 285{
 286	bool enabled = val & APIC_SPIV_APIC_ENABLED;
 287
 288	kvm_lapic_set_reg(apic, APIC_SPIV, val);
 289
 290	if (enabled != apic->sw_enabled) {
 291		apic->sw_enabled = enabled;
 292		if (enabled)
 293			static_key_slow_dec_deferred(&apic_sw_disabled);
 294		else
 295			static_key_slow_inc(&apic_sw_disabled.key);
 296
 297		atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
 298	}
 
 
 
 
 299}
 300
 301static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
 302{
 303	kvm_lapic_set_reg(apic, APIC_ID, id << 24);
 304	atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
 305}
 306
 307static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
 308{
 309	kvm_lapic_set_reg(apic, APIC_LDR, id);
 310	atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
 311}
 312
 
 
 
 
 
 
 313static inline u32 kvm_apic_calc_x2apic_ldr(u32 id)
 314{
 315	return ((id >> 4) << 16) | (1 << (id & 0xf));
 316}
 317
 318static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
 319{
 320	u32 ldr = kvm_apic_calc_x2apic_ldr(id);
 321
 322	WARN_ON_ONCE(id != apic->vcpu->vcpu_id);
 323
 324	kvm_lapic_set_reg(apic, APIC_ID, id);
 325	kvm_lapic_set_reg(apic, APIC_LDR, ldr);
 326	atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
 327}
 328
 329static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
 330{
 331	return !(kvm_lapic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
 332}
 333
 334static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
 335{
 336	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
 337}
 338
 339static inline int apic_lvtt_period(struct kvm_lapic *apic)
 340{
 341	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
 342}
 343
 344static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
 345{
 346	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
 347}
 348
 349static inline int apic_lvt_nmi_mode(u32 lvt_val)
 350{
 351	return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
 352}
 353
 
 
 
 
 
 
 
 
 
 
 354void kvm_apic_set_version(struct kvm_vcpu *vcpu)
 355{
 356	struct kvm_lapic *apic = vcpu->arch.apic;
 357	u32 v = APIC_VERSION;
 358
 359	if (!lapic_in_kernel(vcpu))
 360		return;
 361
 
 
 362	/*
 363	 * KVM emulates 82093AA datasheet (with in-kernel IOAPIC implementation)
 364	 * which doesn't have EOI register; Some buggy OSes (e.g. Windows with
 365	 * Hyper-V role) disable EOI broadcast in lapic not checking for IOAPIC
 366	 * version first and level-triggered interrupts never get EOIed in
 367	 * IOAPIC.
 368	 */
 369	if (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) &&
 370	    !ioapic_in_kernel(vcpu->kvm))
 371		v |= APIC_LVR_DIRECTED_EOI;
 372	kvm_lapic_set_reg(apic, APIC_LVR, v);
 373}
 374
 375static const unsigned int apic_lvt_mask[KVM_APIC_LVT_NUM] = {
 376	LVT_MASK ,      /* part LVTT mask, timer mode mask added at runtime */
 377	LVT_MASK | APIC_MODE_MASK,	/* LVTTHMR */
 378	LVT_MASK | APIC_MODE_MASK,	/* LVTPC */
 379	LINT_MASK, LINT_MASK,	/* LVT0-1 */
 380	LVT_MASK		/* LVTERR */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 381};
 382
 383static int find_highest_vector(void *bitmap)
 384{
 385	int vec;
 386	u32 *reg;
 387
 388	for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
 389	     vec >= 0; vec -= APIC_VECTORS_PER_REG) {
 390		reg = bitmap + REG_POS(vec);
 391		if (*reg)
 392			return __fls(*reg) + vec;
 393	}
 394
 395	return -1;
 396}
 397
 398static u8 count_vectors(void *bitmap)
 399{
 400	int vec;
 401	u32 *reg;
 402	u8 count = 0;
 403
 404	for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
 405		reg = bitmap + REG_POS(vec);
 406		count += hweight32(*reg);
 407	}
 408
 409	return count;
 410}
 411
 412bool __kvm_apic_update_irr(u32 *pir, void *regs, int *max_irr)
 413{
 414	u32 i, vec;
 415	u32 pir_val, irr_val, prev_irr_val;
 416	int max_updated_irr;
 417
 418	max_updated_irr = -1;
 419	*max_irr = -1;
 420
 421	for (i = vec = 0; i <= 7; i++, vec += 32) {
 422		pir_val = READ_ONCE(pir[i]);
 423		irr_val = *((u32 *)(regs + APIC_IRR + i * 0x10));
 424		if (pir_val) {
 425			prev_irr_val = irr_val;
 426			irr_val |= xchg(&pir[i], 0);
 427			*((u32 *)(regs + APIC_IRR + i * 0x10)) = irr_val;
 428			if (prev_irr_val != irr_val) {
 429				max_updated_irr =
 430					__fls(irr_val ^ prev_irr_val) + vec;
 431			}
 432		}
 433		if (irr_val)
 434			*max_irr = __fls(irr_val) + vec;
 435	}
 436
 437	return ((max_updated_irr != -1) &&
 438		(max_updated_irr == *max_irr));
 439}
 440EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
 441
 442bool kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir, int *max_irr)
 443{
 444	struct kvm_lapic *apic = vcpu->arch.apic;
 445
 446	return __kvm_apic_update_irr(pir, apic->regs, max_irr);
 447}
 448EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
 449
 450static inline int apic_search_irr(struct kvm_lapic *apic)
 451{
 452	return find_highest_vector(apic->regs + APIC_IRR);
 453}
 454
 455static inline int apic_find_highest_irr(struct kvm_lapic *apic)
 456{
 457	int result;
 458
 459	/*
 460	 * Note that irr_pending is just a hint. It will be always
 461	 * true with virtual interrupt delivery enabled.
 462	 */
 463	if (!apic->irr_pending)
 464		return -1;
 465
 466	result = apic_search_irr(apic);
 467	ASSERT(result == -1 || result >= 16);
 468
 469	return result;
 470}
 471
 472static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
 473{
 474	struct kvm_vcpu *vcpu;
 475
 476	vcpu = apic->vcpu;
 477
 478	if (unlikely(vcpu->arch.apicv_active)) {
 479		/* need to update RVI */
 480		kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
 481		kvm_x86_ops.hwapic_irr_update(vcpu,
 482				apic_find_highest_irr(apic));
 483	} else {
 484		apic->irr_pending = false;
 485		kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
 486		if (apic_search_irr(apic) != -1)
 487			apic->irr_pending = true;
 488	}
 489}
 490
 
 
 
 
 
 
 491static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
 492{
 493	struct kvm_vcpu *vcpu;
 494
 495	if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
 496		return;
 497
 498	vcpu = apic->vcpu;
 499
 500	/*
 501	 * With APIC virtualization enabled, all caching is disabled
 502	 * because the processor can modify ISR under the hood.  Instead
 503	 * just set SVI.
 504	 */
 505	if (unlikely(vcpu->arch.apicv_active))
 506		kvm_x86_ops.hwapic_isr_update(vcpu, vec);
 507	else {
 508		++apic->isr_count;
 509		BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
 510		/*
 511		 * ISR (in service register) bit is set when injecting an interrupt.
 512		 * The highest vector is injected. Thus the latest bit set matches
 513		 * the highest bit in ISR.
 514		 */
 515		apic->highest_isr_cache = vec;
 516	}
 517}
 518
 519static inline int apic_find_highest_isr(struct kvm_lapic *apic)
 520{
 521	int result;
 522
 523	/*
 524	 * Note that isr_count is always 1, and highest_isr_cache
 525	 * is always -1, with APIC virtualization enabled.
 526	 */
 527	if (!apic->isr_count)
 528		return -1;
 529	if (likely(apic->highest_isr_cache != -1))
 530		return apic->highest_isr_cache;
 531
 532	result = find_highest_vector(apic->regs + APIC_ISR);
 533	ASSERT(result == -1 || result >= 16);
 534
 535	return result;
 536}
 537
 538static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
 539{
 540	struct kvm_vcpu *vcpu;
 541	if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
 542		return;
 543
 544	vcpu = apic->vcpu;
 545
 546	/*
 547	 * We do get here for APIC virtualization enabled if the guest
 548	 * uses the Hyper-V APIC enlightenment.  In this case we may need
 549	 * to trigger a new interrupt delivery by writing the SVI field;
 550	 * on the other hand isr_count and highest_isr_cache are unused
 551	 * and must be left alone.
 552	 */
 553	if (unlikely(vcpu->arch.apicv_active))
 554		kvm_x86_ops.hwapic_isr_update(vcpu,
 555					       apic_find_highest_isr(apic));
 556	else {
 557		--apic->isr_count;
 558		BUG_ON(apic->isr_count < 0);
 559		apic->highest_isr_cache = -1;
 560	}
 561}
 562
 563int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
 564{
 565	/* This may race with setting of irr in __apic_accept_irq() and
 566	 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
 567	 * will cause vmexit immediately and the value will be recalculated
 568	 * on the next vmentry.
 569	 */
 570	return apic_find_highest_irr(vcpu->arch.apic);
 571}
 572EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
 573
 574static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
 575			     int vector, int level, int trig_mode,
 576			     struct dest_map *dest_map);
 577
 578int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
 579		     struct dest_map *dest_map)
 580{
 581	struct kvm_lapic *apic = vcpu->arch.apic;
 582
 583	return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
 584			irq->level, irq->trig_mode, dest_map);
 585}
 586
 587static int __pv_send_ipi(unsigned long *ipi_bitmap, struct kvm_apic_map *map,
 588			 struct kvm_lapic_irq *irq, u32 min)
 589{
 590	int i, count = 0;
 591	struct kvm_vcpu *vcpu;
 592
 593	if (min > map->max_apic_id)
 594		return 0;
 595
 596	for_each_set_bit(i, ipi_bitmap,
 597		min((u32)BITS_PER_LONG, (map->max_apic_id - min + 1))) {
 598		if (map->phys_map[min + i]) {
 599			vcpu = map->phys_map[min + i]->vcpu;
 600			count += kvm_apic_set_irq(vcpu, irq, NULL);
 601		}
 602	}
 603
 604	return count;
 605}
 606
 607int kvm_pv_send_ipi(struct kvm *kvm, unsigned long ipi_bitmap_low,
 608		    unsigned long ipi_bitmap_high, u32 min,
 609		    unsigned long icr, int op_64_bit)
 610{
 611	struct kvm_apic_map *map;
 612	struct kvm_lapic_irq irq = {0};
 613	int cluster_size = op_64_bit ? 64 : 32;
 614	int count;
 615
 616	if (icr & (APIC_DEST_MASK | APIC_SHORT_MASK))
 617		return -KVM_EINVAL;
 618
 619	irq.vector = icr & APIC_VECTOR_MASK;
 620	irq.delivery_mode = icr & APIC_MODE_MASK;
 621	irq.level = (icr & APIC_INT_ASSERT) != 0;
 622	irq.trig_mode = icr & APIC_INT_LEVELTRIG;
 623
 624	rcu_read_lock();
 625	map = rcu_dereference(kvm->arch.apic_map);
 626
 627	count = -EOPNOTSUPP;
 628	if (likely(map)) {
 629		count = __pv_send_ipi(&ipi_bitmap_low, map, &irq, min);
 630		min += cluster_size;
 631		count += __pv_send_ipi(&ipi_bitmap_high, map, &irq, min);
 632	}
 633
 634	rcu_read_unlock();
 635	return count;
 636}
 637
 638static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
 639{
 640
 641	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
 642				      sizeof(val));
 643}
 644
 645static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
 646{
 647
 648	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
 649				      sizeof(*val));
 650}
 651
 652static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
 653{
 654	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
 655}
 656
 657static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
 658{
 659	u8 val;
 660	if (pv_eoi_get_user(vcpu, &val) < 0) {
 661		printk(KERN_WARNING "Can't read EOI MSR value: 0x%llx\n",
 662			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
 663		return false;
 664	}
 665	return val & 0x1;
 666}
 667
 668static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
 669{
 670	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
 671		printk(KERN_WARNING "Can't set EOI MSR value: 0x%llx\n",
 672			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
 673		return;
 674	}
 675	__set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
 676}
 677
 678static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
 679{
 680	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
 681		printk(KERN_WARNING "Can't clear EOI MSR value: 0x%llx\n",
 682			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
 683		return;
 684	}
 
 
 
 
 
 
 
 
 
 
 685	__clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
 
 
 686}
 687
 688static int apic_has_interrupt_for_ppr(struct kvm_lapic *apic, u32 ppr)
 689{
 690	int highest_irr;
 691	if (apic->vcpu->arch.apicv_active)
 692		highest_irr = kvm_x86_ops.sync_pir_to_irr(apic->vcpu);
 693	else
 694		highest_irr = apic_find_highest_irr(apic);
 695	if (highest_irr == -1 || (highest_irr & 0xF0) <= ppr)
 696		return -1;
 697	return highest_irr;
 698}
 699
 700static bool __apic_update_ppr(struct kvm_lapic *apic, u32 *new_ppr)
 701{
 702	u32 tpr, isrv, ppr, old_ppr;
 703	int isr;
 704
 705	old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
 706	tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
 707	isr = apic_find_highest_isr(apic);
 708	isrv = (isr != -1) ? isr : 0;
 709
 710	if ((tpr & 0xf0) >= (isrv & 0xf0))
 711		ppr = tpr & 0xff;
 712	else
 713		ppr = isrv & 0xf0;
 714
 715	*new_ppr = ppr;
 716	if (old_ppr != ppr)
 717		kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
 718
 719	return ppr < old_ppr;
 720}
 721
 722static void apic_update_ppr(struct kvm_lapic *apic)
 723{
 724	u32 ppr;
 725
 726	if (__apic_update_ppr(apic, &ppr) &&
 727	    apic_has_interrupt_for_ppr(apic, ppr) != -1)
 728		kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
 729}
 730
 731void kvm_apic_update_ppr(struct kvm_vcpu *vcpu)
 732{
 733	apic_update_ppr(vcpu->arch.apic);
 734}
 735EXPORT_SYMBOL_GPL(kvm_apic_update_ppr);
 736
 737static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
 738{
 739	kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
 740	apic_update_ppr(apic);
 741}
 742
 743static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
 744{
 745	return mda == (apic_x2apic_mode(apic) ?
 746			X2APIC_BROADCAST : APIC_BROADCAST);
 747}
 748
 749static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
 750{
 751	if (kvm_apic_broadcast(apic, mda))
 752		return true;
 753
 754	if (apic_x2apic_mode(apic))
 755		return mda == kvm_x2apic_id(apic);
 756
 757	/*
 758	 * Hotplug hack: Make LAPIC in xAPIC mode also accept interrupts as if
 759	 * it were in x2APIC mode.  Hotplugged VCPUs start in xAPIC mode and
 760	 * this allows unique addressing of VCPUs with APIC ID over 0xff.
 761	 * The 0xff condition is needed because writeable xAPIC ID.
 
 
 
 762	 */
 763	if (kvm_x2apic_id(apic) > 0xff && mda == kvm_x2apic_id(apic))
 764		return true;
 765
 766	return mda == kvm_xapic_id(apic);
 767}
 768
 769static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
 770{
 771	u32 logical_id;
 772
 773	if (kvm_apic_broadcast(apic, mda))
 774		return true;
 775
 776	logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
 777
 778	if (apic_x2apic_mode(apic))
 779		return ((logical_id >> 16) == (mda >> 16))
 780		       && (logical_id & mda & 0xffff) != 0;
 781
 782	logical_id = GET_APIC_LOGICAL_ID(logical_id);
 783
 784	switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
 785	case APIC_DFR_FLAT:
 786		return (logical_id & mda) != 0;
 787	case APIC_DFR_CLUSTER:
 788		return ((logical_id >> 4) == (mda >> 4))
 789		       && (logical_id & mda & 0xf) != 0;
 790	default:
 791		return false;
 792	}
 793}
 794
 795/* The KVM local APIC implementation has two quirks:
 796 *
 797 *  - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
 798 *    in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
 799 *    KVM doesn't do that aliasing.
 800 *
 801 *  - in-kernel IOAPIC messages have to be delivered directly to
 802 *    x2APIC, because the kernel does not support interrupt remapping.
 803 *    In order to support broadcast without interrupt remapping, x2APIC
 804 *    rewrites the destination of non-IPI messages from APIC_BROADCAST
 805 *    to X2APIC_BROADCAST.
 806 *
 807 * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API.  This is
 808 * important when userspace wants to use x2APIC-format MSIs, because
 809 * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
 810 */
 811static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
 812		struct kvm_lapic *source, struct kvm_lapic *target)
 813{
 814	bool ipi = source != NULL;
 815
 816	if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
 817	    !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
 818		return X2APIC_BROADCAST;
 819
 820	return dest_id;
 821}
 822
 823bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
 824			   int shorthand, unsigned int dest, int dest_mode)
 825{
 826	struct kvm_lapic *target = vcpu->arch.apic;
 827	u32 mda = kvm_apic_mda(vcpu, dest, source, target);
 828
 829	ASSERT(target);
 830	switch (shorthand) {
 831	case APIC_DEST_NOSHORT:
 832		if (dest_mode == APIC_DEST_PHYSICAL)
 833			return kvm_apic_match_physical_addr(target, mda);
 834		else
 835			return kvm_apic_match_logical_addr(target, mda);
 836	case APIC_DEST_SELF:
 837		return target == source;
 838	case APIC_DEST_ALLINC:
 839		return true;
 840	case APIC_DEST_ALLBUT:
 841		return target != source;
 842	default:
 843		return false;
 844	}
 845}
 846EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
 847
 848int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
 849		       const unsigned long *bitmap, u32 bitmap_size)
 850{
 851	u32 mod;
 852	int i, idx = -1;
 853
 854	mod = vector % dest_vcpus;
 855
 856	for (i = 0; i <= mod; i++) {
 857		idx = find_next_bit(bitmap, bitmap_size, idx + 1);
 858		BUG_ON(idx == bitmap_size);
 859	}
 860
 861	return idx;
 862}
 863
 864static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
 865{
 866	if (!kvm->arch.disabled_lapic_found) {
 867		kvm->arch.disabled_lapic_found = true;
 868		printk(KERN_INFO
 869		       "Disabled LAPIC found during irq injection\n");
 870	}
 871}
 872
 873static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
 874		struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
 875{
 876	if (kvm->arch.x2apic_broadcast_quirk_disabled) {
 877		if ((irq->dest_id == APIC_BROADCAST &&
 878				map->mode != KVM_APIC_MODE_X2APIC))
 879			return true;
 880		if (irq->dest_id == X2APIC_BROADCAST)
 881			return true;
 882	} else {
 883		bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
 884		if (irq->dest_id == (x2apic_ipi ?
 885		                     X2APIC_BROADCAST : APIC_BROADCAST))
 886			return true;
 887	}
 888
 889	return false;
 890}
 891
 892/* Return true if the interrupt can be handled by using *bitmap as index mask
 893 * for valid destinations in *dst array.
 894 * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
 895 * Note: we may have zero kvm_lapic destinations when we return true, which
 896 * means that the interrupt should be dropped.  In this case, *bitmap would be
 897 * zero and *dst undefined.
 898 */
 899static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
 900		struct kvm_lapic **src, struct kvm_lapic_irq *irq,
 901		struct kvm_apic_map *map, struct kvm_lapic ***dst,
 902		unsigned long *bitmap)
 903{
 904	int i, lowest;
 905
 906	if (irq->shorthand == APIC_DEST_SELF && src) {
 907		*dst = src;
 908		*bitmap = 1;
 909		return true;
 910	} else if (irq->shorthand)
 911		return false;
 912
 913	if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
 914		return false;
 915
 916	if (irq->dest_mode == APIC_DEST_PHYSICAL) {
 917		if (irq->dest_id > map->max_apic_id) {
 918			*bitmap = 0;
 919		} else {
 920			u32 dest_id = array_index_nospec(irq->dest_id, map->max_apic_id + 1);
 921			*dst = &map->phys_map[dest_id];
 922			*bitmap = 1;
 923		}
 924		return true;
 925	}
 926
 927	*bitmap = 0;
 928	if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
 929				(u16 *)bitmap))
 930		return false;
 931
 932	if (!kvm_lowest_prio_delivery(irq))
 933		return true;
 934
 935	if (!kvm_vector_hashing_enabled()) {
 936		lowest = -1;
 937		for_each_set_bit(i, bitmap, 16) {
 938			if (!(*dst)[i])
 939				continue;
 940			if (lowest < 0)
 941				lowest = i;
 942			else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
 943						(*dst)[lowest]->vcpu) < 0)
 944				lowest = i;
 945		}
 946	} else {
 947		if (!*bitmap)
 948			return true;
 949
 950		lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
 951				bitmap, 16);
 952
 953		if (!(*dst)[lowest]) {
 954			kvm_apic_disabled_lapic_found(kvm);
 955			*bitmap = 0;
 956			return true;
 957		}
 958	}
 959
 960	*bitmap = (lowest >= 0) ? 1 << lowest : 0;
 961
 962	return true;
 963}
 964
 965bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
 966		struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
 967{
 968	struct kvm_apic_map *map;
 969	unsigned long bitmap;
 970	struct kvm_lapic **dst = NULL;
 971	int i;
 972	bool ret;
 973
 974	*r = -1;
 975
 976	if (irq->shorthand == APIC_DEST_SELF) {
 
 
 
 
 977		*r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
 978		return true;
 979	}
 980
 981	rcu_read_lock();
 982	map = rcu_dereference(kvm->arch.apic_map);
 983
 984	ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
 985	if (ret) {
 986		*r = 0;
 987		for_each_set_bit(i, &bitmap, 16) {
 988			if (!dst[i])
 989				continue;
 990			*r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
 991		}
 992	}
 993
 994	rcu_read_unlock();
 995	return ret;
 996}
 997
 998/*
 999 * This routine tries to handle interrupts in posted mode, here is how
1000 * it deals with different cases:
1001 * - For single-destination interrupts, handle it in posted mode
1002 * - Else if vector hashing is enabled and it is a lowest-priority
1003 *   interrupt, handle it in posted mode and use the following mechanism
1004 *   to find the destination vCPU.
1005 *	1. For lowest-priority interrupts, store all the possible
1006 *	   destination vCPUs in an array.
1007 *	2. Use "guest vector % max number of destination vCPUs" to find
1008 *	   the right destination vCPU in the array for the lowest-priority
1009 *	   interrupt.
1010 * - Otherwise, use remapped mode to inject the interrupt.
1011 */
1012bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
1013			struct kvm_vcpu **dest_vcpu)
1014{
1015	struct kvm_apic_map *map;
1016	unsigned long bitmap;
1017	struct kvm_lapic **dst = NULL;
1018	bool ret = false;
1019
1020	if (irq->shorthand)
1021		return false;
1022
1023	rcu_read_lock();
1024	map = rcu_dereference(kvm->arch.apic_map);
1025
1026	if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
1027			hweight16(bitmap) == 1) {
1028		unsigned long i = find_first_bit(&bitmap, 16);
1029
1030		if (dst[i]) {
1031			*dest_vcpu = dst[i]->vcpu;
1032			ret = true;
1033		}
1034	}
1035
1036	rcu_read_unlock();
1037	return ret;
1038}
1039
1040/*
1041 * Add a pending IRQ into lapic.
1042 * Return 1 if successfully added and 0 if discarded.
1043 */
1044static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
1045			     int vector, int level, int trig_mode,
1046			     struct dest_map *dest_map)
1047{
1048	int result = 0;
1049	struct kvm_vcpu *vcpu = apic->vcpu;
1050
1051	trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
1052				  trig_mode, vector);
1053	switch (delivery_mode) {
1054	case APIC_DM_LOWEST:
1055		vcpu->arch.apic_arb_prio++;
1056		fallthrough;
1057	case APIC_DM_FIXED:
1058		if (unlikely(trig_mode && !level))
1059			break;
1060
1061		/* FIXME add logic for vcpu on reset */
1062		if (unlikely(!apic_enabled(apic)))
1063			break;
1064
1065		result = 1;
1066
1067		if (dest_map) {
1068			__set_bit(vcpu->vcpu_id, dest_map->map);
1069			dest_map->vectors[vcpu->vcpu_id] = vector;
1070		}
1071
1072		if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
1073			if (trig_mode)
1074				kvm_lapic_set_vector(vector,
1075						     apic->regs + APIC_TMR);
1076			else
1077				kvm_lapic_clear_vector(vector,
1078						       apic->regs + APIC_TMR);
1079		}
1080
1081		if (kvm_x86_ops.deliver_posted_interrupt(vcpu, vector)) {
1082			kvm_lapic_set_irr(vector, apic);
1083			kvm_make_request(KVM_REQ_EVENT, vcpu);
1084			kvm_vcpu_kick(vcpu);
1085		}
1086		break;
1087
1088	case APIC_DM_REMRD:
1089		result = 1;
1090		vcpu->arch.pv.pv_unhalted = 1;
1091		kvm_make_request(KVM_REQ_EVENT, vcpu);
1092		kvm_vcpu_kick(vcpu);
1093		break;
1094
1095	case APIC_DM_SMI:
1096		result = 1;
1097		kvm_make_request(KVM_REQ_SMI, vcpu);
1098		kvm_vcpu_kick(vcpu);
 
1099		break;
1100
1101	case APIC_DM_NMI:
1102		result = 1;
1103		kvm_inject_nmi(vcpu);
1104		kvm_vcpu_kick(vcpu);
1105		break;
1106
1107	case APIC_DM_INIT:
1108		if (!trig_mode || level) {
1109			result = 1;
1110			/* assumes that there are only KVM_APIC_INIT/SIPI */
1111			apic->pending_events = (1UL << KVM_APIC_INIT);
1112			kvm_make_request(KVM_REQ_EVENT, vcpu);
1113			kvm_vcpu_kick(vcpu);
1114		}
1115		break;
1116
1117	case APIC_DM_STARTUP:
1118		result = 1;
1119		apic->sipi_vector = vector;
1120		/* make sure sipi_vector is visible for the receiver */
1121		smp_wmb();
1122		set_bit(KVM_APIC_SIPI, &apic->pending_events);
1123		kvm_make_request(KVM_REQ_EVENT, vcpu);
1124		kvm_vcpu_kick(vcpu);
1125		break;
1126
1127	case APIC_DM_EXTINT:
1128		/*
1129		 * Should only be called by kvm_apic_local_deliver() with LVT0,
1130		 * before NMI watchdog was enabled. Already handled by
1131		 * kvm_apic_accept_pic_intr().
1132		 */
1133		break;
1134
1135	default:
1136		printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
1137		       delivery_mode);
1138		break;
1139	}
1140	return result;
1141}
1142
1143/*
1144 * This routine identifies the destination vcpus mask meant to receive the
1145 * IOAPIC interrupts. It either uses kvm_apic_map_get_dest_lapic() to find
1146 * out the destination vcpus array and set the bitmap or it traverses to
1147 * each available vcpu to identify the same.
1148 */
1149void kvm_bitmap_or_dest_vcpus(struct kvm *kvm, struct kvm_lapic_irq *irq,
1150			      unsigned long *vcpu_bitmap)
1151{
1152	struct kvm_lapic **dest_vcpu = NULL;
1153	struct kvm_lapic *src = NULL;
1154	struct kvm_apic_map *map;
1155	struct kvm_vcpu *vcpu;
1156	unsigned long bitmap;
1157	int i, vcpu_idx;
1158	bool ret;
1159
1160	rcu_read_lock();
1161	map = rcu_dereference(kvm->arch.apic_map);
1162
1163	ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dest_vcpu,
1164					  &bitmap);
1165	if (ret) {
1166		for_each_set_bit(i, &bitmap, 16) {
1167			if (!dest_vcpu[i])
1168				continue;
1169			vcpu_idx = dest_vcpu[i]->vcpu->vcpu_idx;
1170			__set_bit(vcpu_idx, vcpu_bitmap);
1171		}
1172	} else {
1173		kvm_for_each_vcpu(i, vcpu, kvm) {
1174			if (!kvm_apic_present(vcpu))
1175				continue;
1176			if (!kvm_apic_match_dest(vcpu, NULL,
1177						 irq->shorthand,
1178						 irq->dest_id,
1179						 irq->dest_mode))
1180				continue;
1181			__set_bit(i, vcpu_bitmap);
1182		}
1183	}
1184	rcu_read_unlock();
1185}
1186
1187int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1188{
1189	return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
1190}
1191
1192static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
1193{
1194	return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
1195}
1196
1197static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1198{
1199	int trigger_mode;
1200
1201	/* Eoi the ioapic only if the ioapic doesn't own the vector. */
1202	if (!kvm_ioapic_handles_vector(apic, vector))
1203		return;
1204
1205	/* Request a KVM exit to inform the userspace IOAPIC. */
1206	if (irqchip_split(apic->vcpu->kvm)) {
1207		apic->vcpu->arch.pending_ioapic_eoi = vector;
1208		kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1209		return;
1210	}
1211
1212	if (apic_test_vector(vector, apic->regs + APIC_TMR))
1213		trigger_mode = IOAPIC_LEVEL_TRIG;
1214	else
1215		trigger_mode = IOAPIC_EDGE_TRIG;
1216
1217	kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1218}
1219
1220static int apic_set_eoi(struct kvm_lapic *apic)
1221{
1222	int vector = apic_find_highest_isr(apic);
1223
1224	trace_kvm_eoi(apic, vector);
1225
1226	/*
1227	 * Not every write EOI will has corresponding ISR,
1228	 * one example is when Kernel check timer on setup_IO_APIC
1229	 */
1230	if (vector == -1)
1231		return vector;
1232
1233	apic_clear_isr(vector, apic);
1234	apic_update_ppr(apic);
1235
1236	if (test_bit(vector, vcpu_to_synic(apic->vcpu)->vec_bitmap))
 
1237		kvm_hv_synic_send_eoi(apic->vcpu, vector);
1238
1239	kvm_ioapic_send_eoi(apic, vector);
1240	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1241	return vector;
1242}
1243
1244/*
1245 * this interface assumes a trap-like exit, which has already finished
1246 * desired side effect including vISR and vPPR update.
1247 */
1248void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1249{
1250	struct kvm_lapic *apic = vcpu->arch.apic;
1251
1252	trace_kvm_eoi(apic, vector);
1253
1254	kvm_ioapic_send_eoi(apic, vector);
1255	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1256}
1257EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1258
1259void kvm_apic_send_ipi(struct kvm_lapic *apic, u32 icr_low, u32 icr_high)
1260{
1261	struct kvm_lapic_irq irq;
1262
 
 
 
1263	irq.vector = icr_low & APIC_VECTOR_MASK;
1264	irq.delivery_mode = icr_low & APIC_MODE_MASK;
1265	irq.dest_mode = icr_low & APIC_DEST_MASK;
1266	irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1267	irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1268	irq.shorthand = icr_low & APIC_SHORT_MASK;
1269	irq.msi_redir_hint = false;
1270	if (apic_x2apic_mode(apic))
1271		irq.dest_id = icr_high;
1272	else
1273		irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
1274
1275	trace_kvm_apic_ipi(icr_low, irq.dest_id);
1276
1277	kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1278}
 
1279
1280static u32 apic_get_tmcct(struct kvm_lapic *apic)
1281{
1282	ktime_t remaining, now;
1283	s64 ns;
1284	u32 tmcct;
1285
1286	ASSERT(apic != NULL);
1287
1288	/* if initial count is 0, current count should also be 0 */
1289	if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1290		apic->lapic_timer.period == 0)
1291		return 0;
1292
1293	now = ktime_get();
1294	remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1295	if (ktime_to_ns(remaining) < 0)
1296		remaining = 0;
1297
1298	ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1299	tmcct = div64_u64(ns,
1300			 (APIC_BUS_CYCLE_NS * apic->divide_count));
1301
1302	return tmcct;
1303}
1304
1305static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1306{
1307	struct kvm_vcpu *vcpu = apic->vcpu;
1308	struct kvm_run *run = vcpu->run;
1309
1310	kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1311	run->tpr_access.rip = kvm_rip_read(vcpu);
1312	run->tpr_access.is_write = write;
1313}
1314
1315static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1316{
1317	if (apic->vcpu->arch.tpr_access_reporting)
1318		__report_tpr_access(apic, write);
1319}
1320
1321static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1322{
1323	u32 val = 0;
1324
1325	if (offset >= LAPIC_MMIO_LENGTH)
1326		return 0;
1327
1328	switch (offset) {
1329	case APIC_ARBPRI:
1330		break;
1331
1332	case APIC_TMCCT:	/* Timer CCR */
1333		if (apic_lvtt_tscdeadline(apic))
1334			return 0;
1335
1336		val = apic_get_tmcct(apic);
1337		break;
1338	case APIC_PROCPRI:
1339		apic_update_ppr(apic);
1340		val = kvm_lapic_get_reg(apic, offset);
1341		break;
1342	case APIC_TASKPRI:
1343		report_tpr_access(apic, false);
1344		fallthrough;
1345	default:
1346		val = kvm_lapic_get_reg(apic, offset);
1347		break;
1348	}
1349
1350	return val;
1351}
1352
1353static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1354{
1355	return container_of(dev, struct kvm_lapic, dev);
1356}
1357
1358#define APIC_REG_MASK(reg)	(1ull << ((reg) >> 4))
1359#define APIC_REGS_MASK(first, count) \
1360	(APIC_REG_MASK(first) * ((1ull << (count)) - 1))
1361
1362int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1363		void *data)
1364{
1365	unsigned char alignment = offset & 0xf;
1366	u32 result;
1367	/* this bitmask has a bit cleared for each reserved register */
1368	u64 valid_reg_mask =
1369		APIC_REG_MASK(APIC_ID) |
1370		APIC_REG_MASK(APIC_LVR) |
1371		APIC_REG_MASK(APIC_TASKPRI) |
1372		APIC_REG_MASK(APIC_PROCPRI) |
1373		APIC_REG_MASK(APIC_LDR) |
1374		APIC_REG_MASK(APIC_DFR) |
1375		APIC_REG_MASK(APIC_SPIV) |
1376		APIC_REGS_MASK(APIC_ISR, APIC_ISR_NR) |
1377		APIC_REGS_MASK(APIC_TMR, APIC_ISR_NR) |
1378		APIC_REGS_MASK(APIC_IRR, APIC_ISR_NR) |
1379		APIC_REG_MASK(APIC_ESR) |
1380		APIC_REG_MASK(APIC_ICR) |
1381		APIC_REG_MASK(APIC_ICR2) |
1382		APIC_REG_MASK(APIC_LVTT) |
1383		APIC_REG_MASK(APIC_LVTTHMR) |
1384		APIC_REG_MASK(APIC_LVTPC) |
1385		APIC_REG_MASK(APIC_LVT0) |
1386		APIC_REG_MASK(APIC_LVT1) |
1387		APIC_REG_MASK(APIC_LVTERR) |
1388		APIC_REG_MASK(APIC_TMICT) |
1389		APIC_REG_MASK(APIC_TMCCT) |
1390		APIC_REG_MASK(APIC_TDCR);
1391
1392	/* ARBPRI is not valid on x2APIC */
 
 
 
 
 
 
 
1393	if (!apic_x2apic_mode(apic))
1394		valid_reg_mask |= APIC_REG_MASK(APIC_ARBPRI);
 
 
 
 
 
 
1395
1396	if (offset > 0x3f0 || !(valid_reg_mask & APIC_REG_MASK(offset)))
1397		return 1;
1398
1399	result = __apic_read(apic, offset & ~0xf);
1400
1401	trace_kvm_apic_read(offset, result);
1402
1403	switch (len) {
1404	case 1:
1405	case 2:
1406	case 4:
1407		memcpy(data, (char *)&result + alignment, len);
1408		break;
1409	default:
1410		printk(KERN_ERR "Local APIC read with len = %x, "
1411		       "should be 1,2, or 4 instead\n", len);
1412		break;
1413	}
1414	return 0;
1415}
1416EXPORT_SYMBOL_GPL(kvm_lapic_reg_read);
1417
1418static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1419{
1420	return addr >= apic->base_address &&
1421		addr < apic->base_address + LAPIC_MMIO_LENGTH;
1422}
1423
1424static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1425			   gpa_t address, int len, void *data)
1426{
1427	struct kvm_lapic *apic = to_lapic(this);
1428	u32 offset = address - apic->base_address;
1429
1430	if (!apic_mmio_in_range(apic, address))
1431		return -EOPNOTSUPP;
1432
1433	if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
1434		if (!kvm_check_has_quirk(vcpu->kvm,
1435					 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
1436			return -EOPNOTSUPP;
1437
1438		memset(data, 0xff, len);
1439		return 0;
1440	}
1441
1442	kvm_lapic_reg_read(apic, offset, len, data);
1443
1444	return 0;
1445}
1446
1447static void update_divide_count(struct kvm_lapic *apic)
1448{
1449	u32 tmp1, tmp2, tdcr;
1450
1451	tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1452	tmp1 = tdcr & 0xf;
1453	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1454	apic->divide_count = 0x1 << (tmp2 & 0x7);
1455}
1456
1457static void limit_periodic_timer_frequency(struct kvm_lapic *apic)
1458{
1459	/*
1460	 * Do not allow the guest to program periodic timers with small
1461	 * interval, since the hrtimers are not throttled by the host
1462	 * scheduler.
1463	 */
1464	if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1465		s64 min_period = min_timer_period_us * 1000LL;
1466
1467		if (apic->lapic_timer.period < min_period) {
1468			pr_info_ratelimited(
1469			    "kvm: vcpu %i: requested %lld ns "
1470			    "lapic timer period limited to %lld ns\n",
1471			    apic->vcpu->vcpu_id,
1472			    apic->lapic_timer.period, min_period);
1473			apic->lapic_timer.period = min_period;
1474		}
1475	}
1476}
1477
1478static void cancel_hv_timer(struct kvm_lapic *apic);
1479
 
 
 
 
 
 
 
 
 
 
1480static void apic_update_lvtt(struct kvm_lapic *apic)
1481{
1482	u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1483			apic->lapic_timer.timer_mode_mask;
1484
1485	if (apic->lapic_timer.timer_mode != timer_mode) {
1486		if (apic_lvtt_tscdeadline(apic) != (timer_mode ==
1487				APIC_LVT_TIMER_TSCDEADLINE)) {
1488			hrtimer_cancel(&apic->lapic_timer.timer);
1489			preempt_disable();
1490			if (apic->lapic_timer.hv_timer_in_use)
1491				cancel_hv_timer(apic);
1492			preempt_enable();
1493			kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1494			apic->lapic_timer.period = 0;
1495			apic->lapic_timer.tscdeadline = 0;
1496		}
1497		apic->lapic_timer.timer_mode = timer_mode;
1498		limit_periodic_timer_frequency(apic);
1499	}
1500}
1501
1502/*
1503 * On APICv, this test will cause a busy wait
1504 * during a higher-priority task.
1505 */
1506
1507static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1508{
1509	struct kvm_lapic *apic = vcpu->arch.apic;
1510	u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1511
1512	if (kvm_apic_hw_enabled(apic)) {
1513		int vec = reg & APIC_VECTOR_MASK;
1514		void *bitmap = apic->regs + APIC_ISR;
1515
1516		if (vcpu->arch.apicv_active)
1517			bitmap = apic->regs + APIC_IRR;
1518
1519		if (apic_test_vector(vec, bitmap))
1520			return true;
1521	}
1522	return false;
1523}
1524
1525static inline void __wait_lapic_expire(struct kvm_vcpu *vcpu, u64 guest_cycles)
1526{
1527	u64 timer_advance_ns = vcpu->arch.apic->lapic_timer.timer_advance_ns;
1528
1529	/*
1530	 * If the guest TSC is running at a different ratio than the host, then
1531	 * convert the delay to nanoseconds to achieve an accurate delay.  Note
1532	 * that __delay() uses delay_tsc whenever the hardware has TSC, thus
1533	 * always for VMX enabled hardware.
1534	 */
1535	if (vcpu->arch.tsc_scaling_ratio == kvm_default_tsc_scaling_ratio) {
1536		__delay(min(guest_cycles,
1537			nsec_to_cycles(vcpu, timer_advance_ns)));
1538	} else {
1539		u64 delay_ns = guest_cycles * 1000000ULL;
1540		do_div(delay_ns, vcpu->arch.virtual_tsc_khz);
1541		ndelay(min_t(u32, delay_ns, timer_advance_ns));
1542	}
1543}
1544
1545static inline void adjust_lapic_timer_advance(struct kvm_vcpu *vcpu,
1546					      s64 advance_expire_delta)
1547{
1548	struct kvm_lapic *apic = vcpu->arch.apic;
1549	u32 timer_advance_ns = apic->lapic_timer.timer_advance_ns;
1550	u64 ns;
1551
1552	/* Do not adjust for tiny fluctuations or large random spikes. */
1553	if (abs(advance_expire_delta) > LAPIC_TIMER_ADVANCE_ADJUST_MAX ||
1554	    abs(advance_expire_delta) < LAPIC_TIMER_ADVANCE_ADJUST_MIN)
1555		return;
1556
1557	/* too early */
1558	if (advance_expire_delta < 0) {
1559		ns = -advance_expire_delta * 1000000ULL;
1560		do_div(ns, vcpu->arch.virtual_tsc_khz);
1561		timer_advance_ns -= ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1562	} else {
1563	/* too late */
1564		ns = advance_expire_delta * 1000000ULL;
1565		do_div(ns, vcpu->arch.virtual_tsc_khz);
1566		timer_advance_ns += ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1567	}
1568
1569	if (unlikely(timer_advance_ns > LAPIC_TIMER_ADVANCE_NS_MAX))
1570		timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
1571	apic->lapic_timer.timer_advance_ns = timer_advance_ns;
1572}
1573
1574static void __kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1575{
1576	struct kvm_lapic *apic = vcpu->arch.apic;
1577	u64 guest_tsc, tsc_deadline;
1578
1579	if (apic->lapic_timer.expired_tscdeadline == 0)
1580		return;
1581
1582	tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1583	apic->lapic_timer.expired_tscdeadline = 0;
1584	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1585	apic->lapic_timer.advance_expire_delta = guest_tsc - tsc_deadline;
 
 
 
 
 
 
 
 
 
 
 
1586
1587	if (guest_tsc < tsc_deadline)
1588		__wait_lapic_expire(vcpu, tsc_deadline - guest_tsc);
1589
1590	if (lapic_timer_advance_dynamic)
1591		adjust_lapic_timer_advance(vcpu, apic->lapic_timer.advance_expire_delta);
1592}
1593
1594void kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1595{
1596	if (lapic_timer_int_injected(vcpu))
 
 
 
1597		__kvm_wait_lapic_expire(vcpu);
1598}
1599EXPORT_SYMBOL_GPL(kvm_wait_lapic_expire);
1600
1601static void kvm_apic_inject_pending_timer_irqs(struct kvm_lapic *apic)
1602{
1603	struct kvm_timer *ktimer = &apic->lapic_timer;
1604
1605	kvm_apic_local_deliver(apic, APIC_LVTT);
1606	if (apic_lvtt_tscdeadline(apic)) {
1607		ktimer->tscdeadline = 0;
1608	} else if (apic_lvtt_oneshot(apic)) {
1609		ktimer->tscdeadline = 0;
1610		ktimer->target_expiration = 0;
1611	}
1612}
1613
1614static void apic_timer_expired(struct kvm_lapic *apic, bool from_timer_fn)
1615{
1616	struct kvm_vcpu *vcpu = apic->vcpu;
1617	struct kvm_timer *ktimer = &apic->lapic_timer;
1618
1619	if (atomic_read(&apic->lapic_timer.pending))
1620		return;
1621
1622	if (apic_lvtt_tscdeadline(apic) || ktimer->hv_timer_in_use)
1623		ktimer->expired_tscdeadline = ktimer->tscdeadline;
1624
1625	if (!from_timer_fn && vcpu->arch.apicv_active) {
1626		WARN_ON(kvm_get_running_vcpu() != vcpu);
1627		kvm_apic_inject_pending_timer_irqs(apic);
1628		return;
1629	}
1630
1631	if (kvm_use_posted_timer_interrupt(apic->vcpu)) {
1632		if (apic->lapic_timer.timer_advance_ns)
 
 
 
 
 
 
 
 
1633			__kvm_wait_lapic_expire(vcpu);
1634		kvm_apic_inject_pending_timer_irqs(apic);
1635		return;
1636	}
1637
1638	atomic_inc(&apic->lapic_timer.pending);
1639	kvm_set_pending_timer(vcpu);
 
 
1640}
1641
1642static void start_sw_tscdeadline(struct kvm_lapic *apic)
1643{
1644	struct kvm_timer *ktimer = &apic->lapic_timer;
1645	u64 guest_tsc, tscdeadline = ktimer->tscdeadline;
1646	u64 ns = 0;
1647	ktime_t expire;
1648	struct kvm_vcpu *vcpu = apic->vcpu;
1649	unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1650	unsigned long flags;
1651	ktime_t now;
1652
1653	if (unlikely(!tscdeadline || !this_tsc_khz))
1654		return;
1655
1656	local_irq_save(flags);
1657
1658	now = ktime_get();
1659	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1660
1661	ns = (tscdeadline - guest_tsc) * 1000000ULL;
1662	do_div(ns, this_tsc_khz);
1663
1664	if (likely(tscdeadline > guest_tsc) &&
1665	    likely(ns > apic->lapic_timer.timer_advance_ns)) {
1666		expire = ktime_add_ns(now, ns);
1667		expire = ktime_sub_ns(expire, ktimer->timer_advance_ns);
1668		hrtimer_start(&ktimer->timer, expire, HRTIMER_MODE_ABS_HARD);
1669	} else
1670		apic_timer_expired(apic, false);
1671
1672	local_irq_restore(flags);
1673}
1674
1675static inline u64 tmict_to_ns(struct kvm_lapic *apic, u32 tmict)
1676{
1677	return (u64)tmict * APIC_BUS_CYCLE_NS * (u64)apic->divide_count;
1678}
1679
1680static void update_target_expiration(struct kvm_lapic *apic, uint32_t old_divisor)
1681{
1682	ktime_t now, remaining;
1683	u64 ns_remaining_old, ns_remaining_new;
1684
1685	apic->lapic_timer.period =
1686			tmict_to_ns(apic, kvm_lapic_get_reg(apic, APIC_TMICT));
1687	limit_periodic_timer_frequency(apic);
1688
1689	now = ktime_get();
1690	remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1691	if (ktime_to_ns(remaining) < 0)
1692		remaining = 0;
1693
1694	ns_remaining_old = ktime_to_ns(remaining);
1695	ns_remaining_new = mul_u64_u32_div(ns_remaining_old,
1696	                                   apic->divide_count, old_divisor);
1697
1698	apic->lapic_timer.tscdeadline +=
1699		nsec_to_cycles(apic->vcpu, ns_remaining_new) -
1700		nsec_to_cycles(apic->vcpu, ns_remaining_old);
1701	apic->lapic_timer.target_expiration = ktime_add_ns(now, ns_remaining_new);
1702}
1703
1704static bool set_target_expiration(struct kvm_lapic *apic, u32 count_reg)
1705{
1706	ktime_t now;
1707	u64 tscl = rdtsc();
1708	s64 deadline;
1709
1710	now = ktime_get();
1711	apic->lapic_timer.period =
1712			tmict_to_ns(apic, kvm_lapic_get_reg(apic, APIC_TMICT));
1713
1714	if (!apic->lapic_timer.period) {
1715		apic->lapic_timer.tscdeadline = 0;
1716		return false;
1717	}
1718
1719	limit_periodic_timer_frequency(apic);
1720	deadline = apic->lapic_timer.period;
1721
1722	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1723		if (unlikely(count_reg != APIC_TMICT)) {
1724			deadline = tmict_to_ns(apic,
1725				     kvm_lapic_get_reg(apic, count_reg));
1726			if (unlikely(deadline <= 0))
1727				deadline = apic->lapic_timer.period;
1728			else if (unlikely(deadline > apic->lapic_timer.period)) {
1729				pr_info_ratelimited(
1730				    "kvm: vcpu %i: requested lapic timer restore with "
1731				    "starting count register %#x=%u (%lld ns) > initial count (%lld ns). "
1732				    "Using initial count to start timer.\n",
1733				    apic->vcpu->vcpu_id,
1734				    count_reg,
1735				    kvm_lapic_get_reg(apic, count_reg),
1736				    deadline, apic->lapic_timer.period);
1737				kvm_lapic_set_reg(apic, count_reg, 0);
1738				deadline = apic->lapic_timer.period;
1739			}
1740		}
1741	}
1742
1743	apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1744		nsec_to_cycles(apic->vcpu, deadline);
1745	apic->lapic_timer.target_expiration = ktime_add_ns(now, deadline);
1746
1747	return true;
1748}
1749
1750static void advance_periodic_target_expiration(struct kvm_lapic *apic)
1751{
1752	ktime_t now = ktime_get();
1753	u64 tscl = rdtsc();
1754	ktime_t delta;
1755
1756	/*
1757	 * Synchronize both deadlines to the same time source or
1758	 * differences in the periods (caused by differences in the
1759	 * underlying clocks or numerical approximation errors) will
1760	 * cause the two to drift apart over time as the errors
1761	 * accumulate.
1762	 */
1763	apic->lapic_timer.target_expiration =
1764		ktime_add_ns(apic->lapic_timer.target_expiration,
1765				apic->lapic_timer.period);
1766	delta = ktime_sub(apic->lapic_timer.target_expiration, now);
1767	apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1768		nsec_to_cycles(apic->vcpu, delta);
1769}
1770
1771static void start_sw_period(struct kvm_lapic *apic)
1772{
1773	if (!apic->lapic_timer.period)
1774		return;
1775
1776	if (ktime_after(ktime_get(),
1777			apic->lapic_timer.target_expiration)) {
1778		apic_timer_expired(apic, false);
1779
1780		if (apic_lvtt_oneshot(apic))
1781			return;
1782
1783		advance_periodic_target_expiration(apic);
1784	}
1785
1786	hrtimer_start(&apic->lapic_timer.timer,
1787		apic->lapic_timer.target_expiration,
1788		HRTIMER_MODE_ABS_HARD);
1789}
1790
1791bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1792{
1793	if (!lapic_in_kernel(vcpu))
1794		return false;
1795
1796	return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1797}
1798EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use);
1799
1800static void cancel_hv_timer(struct kvm_lapic *apic)
1801{
1802	WARN_ON(preemptible());
1803	WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1804	kvm_x86_ops.cancel_hv_timer(apic->vcpu);
1805	apic->lapic_timer.hv_timer_in_use = false;
1806}
1807
1808static bool start_hv_timer(struct kvm_lapic *apic)
1809{
1810	struct kvm_timer *ktimer = &apic->lapic_timer;
1811	struct kvm_vcpu *vcpu = apic->vcpu;
1812	bool expired;
1813
1814	WARN_ON(preemptible());
1815	if (!kvm_can_use_hv_timer(vcpu))
1816		return false;
1817
1818	if (!ktimer->tscdeadline)
1819		return false;
1820
1821	if (kvm_x86_ops.set_hv_timer(vcpu, ktimer->tscdeadline, &expired))
1822		return false;
1823
1824	ktimer->hv_timer_in_use = true;
1825	hrtimer_cancel(&ktimer->timer);
1826
1827	/*
1828	 * To simplify handling the periodic timer, leave the hv timer running
1829	 * even if the deadline timer has expired, i.e. rely on the resulting
1830	 * VM-Exit to recompute the periodic timer's target expiration.
1831	 */
1832	if (!apic_lvtt_period(apic)) {
1833		/*
1834		 * Cancel the hv timer if the sw timer fired while the hv timer
1835		 * was being programmed, or if the hv timer itself expired.
1836		 */
1837		if (atomic_read(&ktimer->pending)) {
1838			cancel_hv_timer(apic);
1839		} else if (expired) {
1840			apic_timer_expired(apic, false);
1841			cancel_hv_timer(apic);
1842		}
1843	}
1844
1845	trace_kvm_hv_timer_state(vcpu->vcpu_id, ktimer->hv_timer_in_use);
1846
1847	return true;
1848}
1849
1850static void start_sw_timer(struct kvm_lapic *apic)
1851{
1852	struct kvm_timer *ktimer = &apic->lapic_timer;
1853
1854	WARN_ON(preemptible());
1855	if (apic->lapic_timer.hv_timer_in_use)
1856		cancel_hv_timer(apic);
1857	if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
1858		return;
1859
1860	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1861		start_sw_period(apic);
1862	else if (apic_lvtt_tscdeadline(apic))
1863		start_sw_tscdeadline(apic);
1864	trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, false);
1865}
1866
1867static void restart_apic_timer(struct kvm_lapic *apic)
1868{
1869	preempt_disable();
1870
1871	if (!apic_lvtt_period(apic) && atomic_read(&apic->lapic_timer.pending))
1872		goto out;
1873
1874	if (!start_hv_timer(apic))
1875		start_sw_timer(apic);
1876out:
1877	preempt_enable();
1878}
1879
1880void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1881{
1882	struct kvm_lapic *apic = vcpu->arch.apic;
1883
1884	preempt_disable();
1885	/* If the preempt notifier has already run, it also called apic_timer_expired */
1886	if (!apic->lapic_timer.hv_timer_in_use)
1887		goto out;
1888	WARN_ON(rcuwait_active(&vcpu->wait));
 
1889	cancel_hv_timer(apic);
1890	apic_timer_expired(apic, false);
1891
1892	if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1893		advance_periodic_target_expiration(apic);
1894		restart_apic_timer(apic);
1895	}
1896out:
1897	preempt_enable();
1898}
1899EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
1900
1901void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
1902{
1903	restart_apic_timer(vcpu->arch.apic);
1904}
1905EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer);
1906
1907void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
1908{
1909	struct kvm_lapic *apic = vcpu->arch.apic;
1910
1911	preempt_disable();
1912	/* Possibly the TSC deadline timer is not enabled yet */
1913	if (apic->lapic_timer.hv_timer_in_use)
1914		start_sw_timer(apic);
1915	preempt_enable();
1916}
1917EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer);
1918
1919void kvm_lapic_restart_hv_timer(struct kvm_vcpu *vcpu)
1920{
1921	struct kvm_lapic *apic = vcpu->arch.apic;
1922
1923	WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1924	restart_apic_timer(apic);
1925}
1926
1927static void __start_apic_timer(struct kvm_lapic *apic, u32 count_reg)
1928{
1929	atomic_set(&apic->lapic_timer.pending, 0);
1930
1931	if ((apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1932	    && !set_target_expiration(apic, count_reg))
1933		return;
1934
1935	restart_apic_timer(apic);
1936}
1937
1938static void start_apic_timer(struct kvm_lapic *apic)
1939{
1940	__start_apic_timer(apic, APIC_TMICT);
1941}
1942
1943static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1944{
1945	bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
1946
1947	if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1948		apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1949		if (lvt0_in_nmi_mode) {
1950			atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1951		} else
1952			atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1953	}
1954}
1955
1956int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1957{
1958	int ret = 0;
1959
1960	trace_kvm_apic_write(reg, val);
1961
1962	switch (reg) {
1963	case APIC_ID:		/* Local APIC ID */
1964		if (!apic_x2apic_mode(apic))
1965			kvm_apic_set_xapic_id(apic, val >> 24);
1966		else
 
1967			ret = 1;
 
1968		break;
1969
1970	case APIC_TASKPRI:
1971		report_tpr_access(apic, true);
1972		apic_set_tpr(apic, val & 0xff);
1973		break;
1974
1975	case APIC_EOI:
1976		apic_set_eoi(apic);
1977		break;
1978
1979	case APIC_LDR:
1980		if (!apic_x2apic_mode(apic))
1981			kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1982		else
1983			ret = 1;
1984		break;
1985
1986	case APIC_DFR:
1987		if (!apic_x2apic_mode(apic)) {
1988			kvm_lapic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1989			atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
1990		} else
1991			ret = 1;
1992		break;
1993
1994	case APIC_SPIV: {
1995		u32 mask = 0x3ff;
1996		if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1997			mask |= APIC_SPIV_DIRECTED_EOI;
1998		apic_set_spiv(apic, val & mask);
1999		if (!(val & APIC_SPIV_APIC_ENABLED)) {
2000			int i;
2001			u32 lvt_val;
2002
2003			for (i = 0; i < KVM_APIC_LVT_NUM; i++) {
2004				lvt_val = kvm_lapic_get_reg(apic,
2005						       APIC_LVTT + 0x10 * i);
2006				kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i,
2007					     lvt_val | APIC_LVT_MASKED);
2008			}
2009			apic_update_lvtt(apic);
2010			atomic_set(&apic->lapic_timer.pending, 0);
2011
2012		}
2013		break;
2014	}
2015	case APIC_ICR:
 
 
2016		/* No delay here, so we always clear the pending bit */
2017		val &= ~(1 << 12);
2018		kvm_apic_send_ipi(apic, val, kvm_lapic_get_reg(apic, APIC_ICR2));
2019		kvm_lapic_set_reg(apic, APIC_ICR, val);
2020		break;
2021
2022	case APIC_ICR2:
2023		if (!apic_x2apic_mode(apic))
2024			val &= 0xff000000;
2025		kvm_lapic_set_reg(apic, APIC_ICR2, val);
 
2026		break;
2027
2028	case APIC_LVT0:
2029		apic_manage_nmi_watchdog(apic, val);
2030		fallthrough;
2031	case APIC_LVTTHMR:
2032	case APIC_LVTPC:
2033	case APIC_LVT1:
2034	case APIC_LVTERR: {
2035		/* TODO: Check vector */
2036		size_t size;
2037		u32 index;
2038
 
 
2039		if (!kvm_apic_sw_enabled(apic))
2040			val |= APIC_LVT_MASKED;
2041		size = ARRAY_SIZE(apic_lvt_mask);
2042		index = array_index_nospec(
2043				(reg - APIC_LVTT) >> 4, size);
2044		val &= apic_lvt_mask[index];
2045		kvm_lapic_set_reg(apic, reg, val);
2046		break;
2047	}
2048
2049	case APIC_LVTT:
2050		if (!kvm_apic_sw_enabled(apic))
2051			val |= APIC_LVT_MASKED;
2052		val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
2053		kvm_lapic_set_reg(apic, APIC_LVTT, val);
2054		apic_update_lvtt(apic);
2055		break;
2056
2057	case APIC_TMICT:
2058		if (apic_lvtt_tscdeadline(apic))
2059			break;
2060
2061		hrtimer_cancel(&apic->lapic_timer.timer);
2062		kvm_lapic_set_reg(apic, APIC_TMICT, val);
2063		start_apic_timer(apic);
2064		break;
2065
2066	case APIC_TDCR: {
2067		uint32_t old_divisor = apic->divide_count;
2068
2069		kvm_lapic_set_reg(apic, APIC_TDCR, val & 0xb);
2070		update_divide_count(apic);
2071		if (apic->divide_count != old_divisor &&
2072				apic->lapic_timer.period) {
2073			hrtimer_cancel(&apic->lapic_timer.timer);
2074			update_target_expiration(apic, old_divisor);
2075			restart_apic_timer(apic);
2076		}
2077		break;
2078	}
2079	case APIC_ESR:
2080		if (apic_x2apic_mode(apic) && val != 0)
2081			ret = 1;
2082		break;
2083
2084	case APIC_SELF_IPI:
2085		if (apic_x2apic_mode(apic)) {
2086			kvm_lapic_reg_write(apic, APIC_ICR,
2087					    APIC_DEST_SELF | (val & APIC_VECTOR_MASK));
2088		} else
2089			ret = 1;
2090		break;
2091	default:
2092		ret = 1;
2093		break;
2094	}
2095
 
 
 
 
 
2096	kvm_recalculate_apic_map(apic->vcpu->kvm);
2097
2098	return ret;
2099}
2100EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
2101
2102static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
2103			    gpa_t address, int len, const void *data)
2104{
2105	struct kvm_lapic *apic = to_lapic(this);
2106	unsigned int offset = address - apic->base_address;
2107	u32 val;
2108
2109	if (!apic_mmio_in_range(apic, address))
2110		return -EOPNOTSUPP;
2111
2112	if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
2113		if (!kvm_check_has_quirk(vcpu->kvm,
2114					 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
2115			return -EOPNOTSUPP;
2116
2117		return 0;
2118	}
2119
2120	/*
2121	 * APIC register must be aligned on 128-bits boundary.
2122	 * 32/64/128 bits registers must be accessed thru 32 bits.
2123	 * Refer SDM 8.4.1
2124	 */
2125	if (len != 4 || (offset & 0xf))
2126		return 0;
2127
2128	val = *(u32*)data;
2129
2130	kvm_lapic_reg_write(apic, offset & 0xff0, val);
2131
2132	return 0;
2133}
2134
2135void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
2136{
2137	kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
2138}
2139EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
2140
2141/* emulate APIC access in a trap manner */
2142void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
2143{
2144	u32 val = 0;
 
2145
2146	/* hw has done the conditional check and inst decode */
2147	offset &= 0xff0;
 
 
 
 
2148
2149	kvm_lapic_reg_read(vcpu->arch.apic, offset, 4, &val);
2150
2151	/* TODO: optimize to just emulate side effect w/o one more write */
2152	kvm_lapic_reg_write(vcpu->arch.apic, offset, val);
 
 
 
 
 
 
 
 
2153}
2154EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
2155
2156void kvm_free_lapic(struct kvm_vcpu *vcpu)
2157{
2158	struct kvm_lapic *apic = vcpu->arch.apic;
2159
2160	if (!vcpu->arch.apic)
2161		return;
2162
2163	hrtimer_cancel(&apic->lapic_timer.timer);
2164
2165	if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
2166		static_key_slow_dec_deferred(&apic_hw_disabled);
2167
2168	if (!apic->sw_enabled)
2169		static_key_slow_dec_deferred(&apic_sw_disabled);
2170
2171	if (apic->regs)
2172		free_page((unsigned long)apic->regs);
2173
2174	kfree(apic);
2175}
2176
2177/*
2178 *----------------------------------------------------------------------
2179 * LAPIC interface
2180 *----------------------------------------------------------------------
2181 */
2182u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
2183{
2184	struct kvm_lapic *apic = vcpu->arch.apic;
2185
2186	if (!lapic_in_kernel(vcpu) ||
2187		!apic_lvtt_tscdeadline(apic))
2188		return 0;
2189
2190	return apic->lapic_timer.tscdeadline;
2191}
2192
2193void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
2194{
2195	struct kvm_lapic *apic = vcpu->arch.apic;
2196
2197	if (!kvm_apic_present(vcpu) || apic_lvtt_oneshot(apic) ||
2198			apic_lvtt_period(apic))
2199		return;
2200
2201	hrtimer_cancel(&apic->lapic_timer.timer);
2202	apic->lapic_timer.tscdeadline = data;
2203	start_apic_timer(apic);
2204}
2205
2206void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
2207{
2208	struct kvm_lapic *apic = vcpu->arch.apic;
2209
2210	apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
2211		     | (kvm_lapic_get_reg(apic, APIC_TASKPRI) & 4));
2212}
2213
2214u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
2215{
2216	u64 tpr;
2217
2218	tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
2219
2220	return (tpr & 0xf0) >> 4;
2221}
2222
2223void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
2224{
2225	u64 old_value = vcpu->arch.apic_base;
2226	struct kvm_lapic *apic = vcpu->arch.apic;
2227
2228	if (!apic)
2229		value |= MSR_IA32_APICBASE_BSP;
2230
2231	vcpu->arch.apic_base = value;
2232
2233	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
2234		kvm_update_cpuid_runtime(vcpu);
2235
2236	if (!apic)
2237		return;
2238
2239	/* update jump label if enable bit changes */
2240	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
2241		if (value & MSR_IA32_APICBASE_ENABLE) {
2242			kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2243			static_key_slow_dec_deferred(&apic_hw_disabled);
 
 
2244		} else {
2245			static_key_slow_inc(&apic_hw_disabled.key);
2246			atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
2247		}
2248	}
2249
2250	if (((old_value ^ value) & X2APIC_ENABLE) && (value & X2APIC_ENABLE))
2251		kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
2252
2253	if ((old_value ^ value) & (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE))
2254		kvm_x86_ops.set_virtual_apic_mode(vcpu);
 
 
2255
2256	apic->base_address = apic->vcpu->arch.apic_base &
2257			     MSR_IA32_APICBASE_BASE;
2258
2259	if ((value & MSR_IA32_APICBASE_ENABLE) &&
2260	     apic->base_address != APIC_DEFAULT_PHYS_BASE)
2261		pr_warn_once("APIC base relocation is unsupported by KVM");
 
 
2262}
2263
2264void kvm_apic_update_apicv(struct kvm_vcpu *vcpu)
2265{
2266	struct kvm_lapic *apic = vcpu->arch.apic;
2267
2268	if (vcpu->arch.apicv_active) {
2269		/* irr_pending is always true when apicv is activated. */
2270		apic->irr_pending = true;
2271		apic->isr_count = 1;
2272	} else {
2273		apic->irr_pending = (apic_search_irr(apic) != -1);
 
 
 
 
 
2274		apic->isr_count = count_vectors(apic->regs + APIC_ISR);
2275	}
2276}
2277EXPORT_SYMBOL_GPL(kvm_apic_update_apicv);
2278
2279void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
2280{
2281	struct kvm_lapic *apic = vcpu->arch.apic;
 
2282	int i;
2283
 
 
 
 
 
 
 
2284	if (!apic)
2285		return;
2286
2287	/* Stop the timer in case it's a reset to an active apic */
2288	hrtimer_cancel(&apic->lapic_timer.timer);
2289
2290	if (!init_event) {
2291		kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
2292		                         MSR_IA32_APICBASE_ENABLE);
2293		kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2294	}
2295	kvm_apic_set_version(apic->vcpu);
2296
2297	for (i = 0; i < KVM_APIC_LVT_NUM; i++)
2298		kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
2299	apic_update_lvtt(apic);
2300	if (kvm_vcpu_is_reset_bsp(vcpu) &&
2301	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
2302		kvm_lapic_set_reg(apic, APIC_LVT0,
2303			     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
2304	apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2305
2306	kvm_lapic_set_reg(apic, APIC_DFR, 0xffffffffU);
2307	apic_set_spiv(apic, 0xff);
2308	kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
2309	if (!apic_x2apic_mode(apic))
2310		kvm_apic_set_ldr(apic, 0);
2311	kvm_lapic_set_reg(apic, APIC_ESR, 0);
2312	kvm_lapic_set_reg(apic, APIC_ICR, 0);
2313	kvm_lapic_set_reg(apic, APIC_ICR2, 0);
 
 
 
 
2314	kvm_lapic_set_reg(apic, APIC_TDCR, 0);
2315	kvm_lapic_set_reg(apic, APIC_TMICT, 0);
2316	for (i = 0; i < 8; i++) {
2317		kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
2318		kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
2319		kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
2320	}
2321	kvm_apic_update_apicv(vcpu);
2322	apic->highest_isr_cache = -1;
2323	update_divide_count(apic);
2324	atomic_set(&apic->lapic_timer.pending, 0);
2325	if (kvm_vcpu_is_bsp(vcpu))
2326		kvm_lapic_set_base(vcpu,
2327				vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
2328	vcpu->arch.pv_eoi.msr_val = 0;
2329	apic_update_ppr(apic);
2330	if (vcpu->arch.apicv_active) {
2331		kvm_x86_ops.apicv_post_state_restore(vcpu);
2332		kvm_x86_ops.hwapic_irr_update(vcpu, -1);
2333		kvm_x86_ops.hwapic_isr_update(vcpu, -1);
2334	}
2335
2336	vcpu->arch.apic_arb_prio = 0;
2337	vcpu->arch.apic_attention = 0;
2338
2339	kvm_recalculate_apic_map(vcpu->kvm);
2340}
2341
2342/*
2343 *----------------------------------------------------------------------
2344 * timer interface
2345 *----------------------------------------------------------------------
2346 */
2347
2348static bool lapic_is_periodic(struct kvm_lapic *apic)
2349{
2350	return apic_lvtt_period(apic);
2351}
2352
2353int apic_has_pending_timer(struct kvm_vcpu *vcpu)
2354{
2355	struct kvm_lapic *apic = vcpu->arch.apic;
2356
2357	if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
2358		return atomic_read(&apic->lapic_timer.pending);
2359
2360	return 0;
2361}
2362
2363int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
2364{
2365	u32 reg = kvm_lapic_get_reg(apic, lvt_type);
2366	int vector, mode, trig_mode;
2367
2368	if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
2369		vector = reg & APIC_VECTOR_MASK;
2370		mode = reg & APIC_MODE_MASK;
2371		trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
2372		return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
2373					NULL);
2374	}
2375	return 0;
2376}
2377
2378void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
2379{
2380	struct kvm_lapic *apic = vcpu->arch.apic;
2381
2382	if (apic)
2383		kvm_apic_local_deliver(apic, APIC_LVT0);
2384}
2385
2386static const struct kvm_io_device_ops apic_mmio_ops = {
2387	.read     = apic_mmio_read,
2388	.write    = apic_mmio_write,
2389};
2390
2391static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
2392{
2393	struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
2394	struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
2395
2396	apic_timer_expired(apic, true);
2397
2398	if (lapic_is_periodic(apic)) {
2399		advance_periodic_target_expiration(apic);
2400		hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
2401		return HRTIMER_RESTART;
2402	} else
2403		return HRTIMER_NORESTART;
2404}
2405
2406int kvm_create_lapic(struct kvm_vcpu *vcpu, int timer_advance_ns)
2407{
2408	struct kvm_lapic *apic;
2409
2410	ASSERT(vcpu != NULL);
2411
2412	apic = kzalloc(sizeof(*apic), GFP_KERNEL_ACCOUNT);
2413	if (!apic)
2414		goto nomem;
2415
2416	vcpu->arch.apic = apic;
2417
2418	apic->regs = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
2419	if (!apic->regs) {
2420		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
2421		       vcpu->vcpu_id);
2422		goto nomem_free_apic;
2423	}
2424	apic->vcpu = vcpu;
2425
 
 
2426	hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2427		     HRTIMER_MODE_ABS_HARD);
2428	apic->lapic_timer.timer.function = apic_timer_fn;
2429	if (timer_advance_ns == -1) {
2430		apic->lapic_timer.timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
2431		lapic_timer_advance_dynamic = true;
2432	} else {
2433		apic->lapic_timer.timer_advance_ns = timer_advance_ns;
2434		lapic_timer_advance_dynamic = false;
2435	}
2436
2437	/*
2438	 * APIC is created enabled. This will prevent kvm_lapic_set_base from
2439	 * thinking that APIC state has changed.
2440	 */
2441	vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2442	static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2443	kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2444
2445	return 0;
2446nomem_free_apic:
2447	kfree(apic);
2448	vcpu->arch.apic = NULL;
2449nomem:
2450	return -ENOMEM;
2451}
2452
2453int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2454{
2455	struct kvm_lapic *apic = vcpu->arch.apic;
2456	u32 ppr;
2457
2458	if (!kvm_apic_hw_enabled(apic))
2459		return -1;
2460
2461	__apic_update_ppr(apic, &ppr);
2462	return apic_has_interrupt_for_ppr(apic, ppr);
2463}
 
2464
2465int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2466{
2467	u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2468
2469	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2470		return 1;
2471	if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2472	    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2473		return 1;
2474	return 0;
2475}
2476
2477void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2478{
2479	struct kvm_lapic *apic = vcpu->arch.apic;
2480
2481	if (atomic_read(&apic->lapic_timer.pending) > 0) {
2482		kvm_apic_inject_pending_timer_irqs(apic);
2483		atomic_set(&apic->lapic_timer.pending, 0);
2484	}
2485}
2486
2487int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2488{
2489	int vector = kvm_apic_has_interrupt(vcpu);
2490	struct kvm_lapic *apic = vcpu->arch.apic;
2491	u32 ppr;
2492
2493	if (vector == -1)
2494		return -1;
2495
2496	/*
2497	 * We get here even with APIC virtualization enabled, if doing
2498	 * nested virtualization and L1 runs with the "acknowledge interrupt
2499	 * on exit" mode.  Then we cannot inject the interrupt via RVI,
2500	 * because the process would deliver it through the IDT.
2501	 */
2502
2503	apic_clear_irr(vector, apic);
2504	if (test_bit(vector, vcpu_to_synic(vcpu)->auto_eoi_bitmap)) {
2505		/*
2506		 * For auto-EOI interrupts, there might be another pending
2507		 * interrupt above PPR, so check whether to raise another
2508		 * KVM_REQ_EVENT.
2509		 */
2510		apic_update_ppr(apic);
2511	} else {
2512		/*
2513		 * For normal interrupts, PPR has been raised and there cannot
2514		 * be a higher-priority pending interrupt---except if there was
2515		 * a concurrent interrupt injection, but that would have
2516		 * triggered KVM_REQ_EVENT already.
2517		 */
2518		apic_set_isr(vector, apic);
2519		__apic_update_ppr(apic, &ppr);
2520	}
2521
2522	return vector;
2523}
2524
2525static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2526		struct kvm_lapic_state *s, bool set)
2527{
2528	if (apic_x2apic_mode(vcpu->arch.apic)) {
2529		u32 *id = (u32 *)(s->regs + APIC_ID);
2530		u32 *ldr = (u32 *)(s->regs + APIC_LDR);
 
2531
2532		if (vcpu->kvm->arch.x2apic_format) {
2533			if (*id != vcpu->vcpu_id)
2534				return -EINVAL;
2535		} else {
2536			if (set)
2537				*id >>= 24;
2538			else
2539				*id <<= 24;
2540		}
2541
2542		/* In x2APIC mode, the LDR is fixed and based on the id */
2543		if (set)
 
 
 
 
2544			*ldr = kvm_apic_calc_x2apic_ldr(*id);
 
 
 
 
 
 
 
 
2545	}
2546
2547	return 0;
2548}
2549
2550int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2551{
2552	memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2553
2554	/*
2555	 * Get calculated timer current count for remaining timer period (if
2556	 * any) and store it in the returned register set.
2557	 */
2558	__kvm_lapic_set_reg(s->regs, APIC_TMCCT,
2559			    __apic_read(vcpu->arch.apic, APIC_TMCCT));
2560
2561	return kvm_apic_state_fixup(vcpu, s, false);
2562}
2563
2564int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2565{
2566	struct kvm_lapic *apic = vcpu->arch.apic;
2567	int r;
2568
2569	kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2570	/* set SPIV separately to get count of SW disabled APICs right */
2571	apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2572
2573	r = kvm_apic_state_fixup(vcpu, s, true);
2574	if (r) {
2575		kvm_recalculate_apic_map(vcpu->kvm);
2576		return r;
2577	}
2578	memcpy(vcpu->arch.apic->regs, s->regs, sizeof(*s));
2579
 
 
 
2580	atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
2581	kvm_recalculate_apic_map(vcpu->kvm);
2582	kvm_apic_set_version(vcpu);
2583
2584	apic_update_ppr(apic);
2585	hrtimer_cancel(&apic->lapic_timer.timer);
 
2586	apic_update_lvtt(apic);
2587	apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2588	update_divide_count(apic);
2589	__start_apic_timer(apic, APIC_TMCCT);
 
2590	kvm_apic_update_apicv(vcpu);
2591	apic->highest_isr_cache = -1;
2592	if (vcpu->arch.apicv_active) {
2593		kvm_x86_ops.apicv_post_state_restore(vcpu);
2594		kvm_x86_ops.hwapic_irr_update(vcpu,
2595				apic_find_highest_irr(apic));
2596		kvm_x86_ops.hwapic_isr_update(vcpu,
2597				apic_find_highest_isr(apic));
2598	}
2599	kvm_make_request(KVM_REQ_EVENT, vcpu);
2600	if (ioapic_in_kernel(vcpu->kvm))
2601		kvm_rtc_eoi_tracking_restore_one(vcpu);
2602
2603	vcpu->arch.apic_arb_prio = 0;
2604
2605	return 0;
2606}
2607
2608void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2609{
2610	struct hrtimer *timer;
2611
2612	if (!lapic_in_kernel(vcpu) ||
2613		kvm_can_post_timer_interrupt(vcpu))
2614		return;
2615
2616	timer = &vcpu->arch.apic->lapic_timer.timer;
2617	if (hrtimer_cancel(timer))
2618		hrtimer_start_expires(timer, HRTIMER_MODE_ABS_HARD);
2619}
2620
2621/*
2622 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2623 *
2624 * Detect whether guest triggered PV EOI since the
2625 * last entry. If yes, set EOI on guests's behalf.
2626 * Clear PV EOI in guest memory in any case.
2627 */
2628static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2629					struct kvm_lapic *apic)
2630{
2631	bool pending;
2632	int vector;
2633	/*
2634	 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2635	 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2636	 *
2637	 * KVM_APIC_PV_EOI_PENDING is unset:
2638	 * 	-> host disabled PV EOI.
2639	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2640	 * 	-> host enabled PV EOI, guest did not execute EOI yet.
2641	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2642	 * 	-> host enabled PV EOI, guest executed EOI.
2643	 */
2644	BUG_ON(!pv_eoi_enabled(vcpu));
2645	pending = pv_eoi_get_pending(vcpu);
2646	/*
2647	 * Clear pending bit in any case: it will be set again on vmentry.
2648	 * While this might not be ideal from performance point of view,
2649	 * this makes sure pv eoi is only enabled when we know it's safe.
2650	 */
2651	pv_eoi_clr_pending(vcpu);
2652	if (pending)
2653		return;
2654	vector = apic_set_eoi(apic);
2655	trace_kvm_pv_eoi(apic, vector);
2656}
2657
2658void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2659{
2660	u32 data;
2661
2662	if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2663		apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2664
2665	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2666		return;
2667
2668	if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2669				  sizeof(u32)))
2670		return;
2671
2672	apic_set_tpr(vcpu->arch.apic, data & 0xff);
2673}
2674
2675/*
2676 * apic_sync_pv_eoi_to_guest - called before vmentry
2677 *
2678 * Detect whether it's safe to enable PV EOI and
2679 * if yes do so.
2680 */
2681static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2682					struct kvm_lapic *apic)
2683{
2684	if (!pv_eoi_enabled(vcpu) ||
2685	    /* IRR set or many bits in ISR: could be nested. */
2686	    apic->irr_pending ||
2687	    /* Cache not set: could be safe but we don't bother. */
2688	    apic->highest_isr_cache == -1 ||
2689	    /* Need EOI to update ioapic. */
2690	    kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2691		/*
2692		 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2693		 * so we need not do anything here.
2694		 */
2695		return;
2696	}
2697
2698	pv_eoi_set_pending(apic->vcpu);
2699}
2700
2701void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2702{
2703	u32 data, tpr;
2704	int max_irr, max_isr;
2705	struct kvm_lapic *apic = vcpu->arch.apic;
2706
2707	apic_sync_pv_eoi_to_guest(vcpu, apic);
2708
2709	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2710		return;
2711
2712	tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
2713	max_irr = apic_find_highest_irr(apic);
2714	if (max_irr < 0)
2715		max_irr = 0;
2716	max_isr = apic_find_highest_isr(apic);
2717	if (max_isr < 0)
2718		max_isr = 0;
2719	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2720
2721	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2722				sizeof(u32));
2723}
2724
2725int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2726{
2727	if (vapic_addr) {
2728		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2729					&vcpu->arch.apic->vapic_cache,
2730					vapic_addr, sizeof(u32)))
2731			return -EINVAL;
2732		__set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2733	} else {
2734		__clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2735	}
2736
2737	vcpu->arch.apic->vapic_addr = vapic_addr;
2738	return 0;
2739}
2740
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2741int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2742{
2743	struct kvm_lapic *apic = vcpu->arch.apic;
2744	u32 reg = (msr - APIC_BASE_MSR) << 4;
2745
2746	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2747		return 1;
2748
2749	if (reg == APIC_ICR2)
2750		return 1;
2751
2752	/* if this is ICR write vector before command */
2753	if (reg == APIC_ICR)
2754		kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2755	return kvm_lapic_reg_write(apic, reg, (u32)data);
2756}
2757
2758int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2759{
2760	struct kvm_lapic *apic = vcpu->arch.apic;
2761	u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2762
2763	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2764		return 1;
2765
2766	if (reg == APIC_DFR || reg == APIC_ICR2)
2767		return 1;
2768
2769	if (kvm_lapic_reg_read(apic, reg, 4, &low))
2770		return 1;
2771	if (reg == APIC_ICR)
2772		kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2773
2774	*data = (((u64)high) << 32) | low;
2775
2776	return 0;
2777}
2778
2779int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2780{
2781	struct kvm_lapic *apic = vcpu->arch.apic;
2782
2783	if (!lapic_in_kernel(vcpu))
2784		return 1;
2785
2786	/* if this is ICR write vector before command */
2787	if (reg == APIC_ICR)
2788		kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2789	return kvm_lapic_reg_write(apic, reg, (u32)data);
2790}
2791
2792int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2793{
2794	struct kvm_lapic *apic = vcpu->arch.apic;
2795	u32 low, high = 0;
2796
2797	if (!lapic_in_kernel(vcpu))
2798		return 1;
2799
2800	if (kvm_lapic_reg_read(apic, reg, 4, &low))
2801		return 1;
2802	if (reg == APIC_ICR)
2803		kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2804
2805	*data = (((u64)high) << 32) | low;
2806
2807	return 0;
2808}
2809
2810int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data, unsigned long len)
2811{
2812	u64 addr = data & ~KVM_MSR_ENABLED;
2813	struct gfn_to_hva_cache *ghc = &vcpu->arch.pv_eoi.data;
2814	unsigned long new_len;
 
2815
2816	if (!IS_ALIGNED(addr, 4))
2817		return 1;
2818
 
 
 
 
 
 
 
 
 
 
 
2819	vcpu->arch.pv_eoi.msr_val = data;
2820	if (!pv_eoi_enabled(vcpu))
2821		return 0;
2822
2823	if (addr == ghc->gpa && len <= ghc->len)
2824		new_len = ghc->len;
2825	else
2826		new_len = len;
2827
2828	return kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, addr, new_len);
2829}
2830
2831void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2832{
2833	struct kvm_lapic *apic = vcpu->arch.apic;
2834	u8 sipi_vector;
2835	unsigned long pe;
 
 
 
2836
2837	if (!lapic_in_kernel(vcpu) || !apic->pending_events)
2838		return;
 
 
 
 
 
 
 
 
2839
2840	/*
2841	 * INITs are latched while CPU is in specific states
2842	 * (SMM, VMX non-root mode, SVM with GIF=0).
2843	 * Because a CPU cannot be in these states immediately
2844	 * after it has processed an INIT signal (and thus in
2845	 * KVM_MP_STATE_INIT_RECEIVED state), just eat SIPIs
2846	 * and leave the INIT pending.
2847	 */
2848	if (kvm_vcpu_latch_init(vcpu)) {
2849		WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2850		if (test_bit(KVM_APIC_SIPI, &apic->pending_events))
2851			clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2852		return;
2853	}
2854
2855	pe = xchg(&apic->pending_events, 0);
2856	if (test_bit(KVM_APIC_INIT, &pe)) {
2857		kvm_vcpu_reset(vcpu, true);
2858		if (kvm_vcpu_is_bsp(apic->vcpu))
2859			vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2860		else
2861			vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2862	}
2863	if (test_bit(KVM_APIC_SIPI, &pe) &&
2864	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2865		/* evaluate pending_events before reading the vector */
2866		smp_rmb();
2867		sipi_vector = apic->sipi_vector;
2868		kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2869		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
 
2870	}
2871}
2872
2873void kvm_lapic_init(void)
2874{
2875	/* do not patch jump label more than once per second */
2876	jump_label_rate_limit(&apic_hw_disabled, HZ);
2877	jump_label_rate_limit(&apic_sw_disabled, HZ);
2878}
2879
2880void kvm_lapic_exit(void)
2881{
2882	static_key_deferred_flush(&apic_hw_disabled);
 
2883	static_key_deferred_flush(&apic_sw_disabled);
 
2884}