Linux Audio

Check our new training course

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}
v5.4
   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 "trace.h"
  40#include "x86.h"
  41#include "cpuid.h"
  42#include "hyperv.h"
 
  43
  44#ifndef CONFIG_X86_64
  45#define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
  46#else
  47#define mod_64(x, y) ((x) % (y))
  48#endif
  49
  50#define PRId64 "d"
  51#define PRIx64 "llx"
  52#define PRIu64 "u"
  53#define PRIo64 "o"
  54
  55/* 14 is the version for Xeon and Pentium 8.4.8*/
  56#define APIC_VERSION			(0x14UL | ((KVM_APIC_LVT_NUM - 1) << 16))
  57#define LAPIC_MMIO_LENGTH		(1 << 12)
  58/* followed define is not in apicdef.h */
  59#define APIC_SHORT_MASK			0xc0000
  60#define APIC_DEST_NOSHORT		0x0
  61#define APIC_DEST_MASK			0x800
  62#define MAX_APIC_VECTOR			256
  63#define APIC_VECTORS_PER_REG		32
  64
  65#define APIC_BROADCAST			0xFF
  66#define X2APIC_BROADCAST		0xFFFFFFFFul
  67
  68static bool lapic_timer_advance_dynamic __read_mostly;
  69#define LAPIC_TIMER_ADVANCE_ADJUST_MIN	100	/* clock cycles */
  70#define LAPIC_TIMER_ADVANCE_ADJUST_MAX	10000	/* clock cycles */
  71#define LAPIC_TIMER_ADVANCE_NS_INIT	1000
  72#define LAPIC_TIMER_ADVANCE_NS_MAX     5000
  73/* step-by-step approximation to mitigate fluctuation */
  74#define LAPIC_TIMER_ADVANCE_ADJUST_STEP 8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  75
  76static inline int apic_test_vector(int vec, void *bitmap)
  77{
  78	return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
  79}
  80
  81bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
  82{
  83	struct kvm_lapic *apic = vcpu->arch.apic;
  84
  85	return apic_test_vector(vector, apic->regs + APIC_ISR) ||
  86		apic_test_vector(vector, apic->regs + APIC_IRR);
  87}
  88
  89static inline int __apic_test_and_set_vector(int vec, void *bitmap)
  90{
  91	return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
  92}
  93
  94static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
  95{
  96	return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
  97}
  98
  99struct static_key_deferred apic_hw_disabled __read_mostly;
 100struct static_key_deferred apic_sw_disabled __read_mostly;
 101
 102static inline int apic_enabled(struct kvm_lapic *apic)
 103{
 104	return kvm_apic_sw_enabled(apic) &&	kvm_apic_hw_enabled(apic);
 105}
 106
 107#define LVT_MASK	\
 108	(APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
 109
 110#define LINT_MASK	\
 111	(LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
 112	 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
 113
 114static inline u32 kvm_x2apic_id(struct kvm_lapic *apic)
 115{
 116	return apic->vcpu->vcpu_id;
 117}
 118
 119bool kvm_can_post_timer_interrupt(struct kvm_vcpu *vcpu)
 120{
 121	return pi_inject_timer && kvm_vcpu_apicv_active(vcpu);
 
 
 
 
 
 
 
 
 122}
 123EXPORT_SYMBOL_GPL(kvm_can_post_timer_interrupt);
 124
 125static bool kvm_use_posted_timer_interrupt(struct kvm_vcpu *vcpu)
 126{
 127	return kvm_can_post_timer_interrupt(vcpu) && vcpu->mode == IN_GUEST_MODE;
 128}
 129
 130static inline bool kvm_apic_map_get_logical_dest(struct kvm_apic_map *map,
 131		u32 dest_id, struct kvm_lapic ***cluster, u16 *mask) {
 132	switch (map->mode) {
 133	case KVM_APIC_MODE_X2APIC: {
 134		u32 offset = (dest_id >> 16) * 16;
 135		u32 max_apic_id = map->max_apic_id;
 136
 137		if (offset <= max_apic_id) {
 138			u8 cluster_size = min(max_apic_id - offset + 1, 16U);
 139
 140			offset = array_index_nospec(offset, map->max_apic_id + 1);
 141			*cluster = &map->phys_map[offset];
 142			*mask = dest_id & (0xffff >> (16 - cluster_size));
 143		} else {
 144			*mask = 0;
 145		}
 146
 147		return true;
 148		}
 149	case KVM_APIC_MODE_XAPIC_FLAT:
 150		*cluster = map->xapic_flat_map;
 151		*mask = dest_id & 0xff;
 152		return true;
 153	case KVM_APIC_MODE_XAPIC_CLUSTER:
 154		*cluster = map->xapic_cluster_map[(dest_id >> 4) & 0xf];
 155		*mask = dest_id & 0xf;
 156		return true;
 157	default:
 158		/* Not optimized. */
 159		return false;
 160	}
 161}
 162
 163static void kvm_apic_map_free(struct rcu_head *rcu)
 164{
 165	struct kvm_apic_map *map = container_of(rcu, struct kvm_apic_map, rcu);
 166
 167	kvfree(map);
 168}
 169
 170static void recalculate_apic_map(struct kvm *kvm)
 
 
 
 
 
 
 
 
 
 
 
 
 171{
 172	struct kvm_apic_map *new, *old = NULL;
 173	struct kvm_vcpu *vcpu;
 174	int i;
 175	u32 max_id = 255; /* enough space for any xAPIC ID */
 176
 
 
 
 
 
 
 
 177	mutex_lock(&kvm->arch.apic_map_lock);
 
 
 
 
 
 
 
 
 
 
 178
 179	kvm_for_each_vcpu(i, vcpu, kvm)
 180		if (kvm_apic_present(vcpu))
 181			max_id = max(max_id, kvm_x2apic_id(vcpu->arch.apic));
 182
 183	new = kvzalloc(sizeof(struct kvm_apic_map) +
 184	                   sizeof(struct kvm_lapic *) * ((u64)max_id + 1),
 185			   GFP_KERNEL_ACCOUNT);
 186
 187	if (!new)
 188		goto out;
 189
 190	new->max_apic_id = max_id;
 191
 192	kvm_for_each_vcpu(i, vcpu, kvm) {
 193		struct kvm_lapic *apic = vcpu->arch.apic;
 194		struct kvm_lapic **cluster;
 195		u16 mask;
 196		u32 ldr;
 197		u8 xapic_id;
 198		u32 x2apic_id;
 199
 200		if (!kvm_apic_present(vcpu))
 201			continue;
 202
 203		xapic_id = kvm_xapic_id(apic);
 204		x2apic_id = kvm_x2apic_id(apic);
 205
 206		/* Hotplug hack: see kvm_apic_match_physical_addr(), ... */
 207		if ((apic_x2apic_mode(apic) || x2apic_id > 0xff) &&
 208				x2apic_id <= new->max_apic_id)
 209			new->phys_map[x2apic_id] = apic;
 210		/*
 211		 * ... xAPIC ID of VCPUs with APIC ID > 0xff will wrap-around,
 212		 * prevent them from masking VCPUs with APIC ID <= 0xff.
 213		 */
 214		if (!apic_x2apic_mode(apic) && !new->phys_map[xapic_id])
 215			new->phys_map[xapic_id] = apic;
 216
 217		if (!kvm_apic_sw_enabled(apic))
 218			continue;
 219
 220		ldr = kvm_lapic_get_reg(apic, APIC_LDR);
 221
 222		if (apic_x2apic_mode(apic)) {
 223			new->mode |= KVM_APIC_MODE_X2APIC;
 224		} else if (ldr) {
 225			ldr = GET_APIC_LOGICAL_ID(ldr);
 226			if (kvm_lapic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
 227				new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
 228			else
 229				new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
 230		}
 231
 232		if (!kvm_apic_map_get_logical_dest(new, ldr, &cluster, &mask))
 233			continue;
 234
 235		if (mask)
 236			cluster[ffs(mask) - 1] = apic;
 237	}
 238out:
 239	old = rcu_dereference_protected(kvm->arch.apic_map,
 240			lockdep_is_held(&kvm->arch.apic_map_lock));
 241	rcu_assign_pointer(kvm->arch.apic_map, new);
 
 
 
 
 
 
 242	mutex_unlock(&kvm->arch.apic_map_lock);
 243
 244	if (old)
 245		call_rcu(&old->rcu, kvm_apic_map_free);
 246
 247	kvm_make_scan_ioapic_request(kvm);
 248}
 249
 250static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
 251{
 252	bool enabled = val & APIC_SPIV_APIC_ENABLED;
 253
 254	kvm_lapic_set_reg(apic, APIC_SPIV, val);
 255
 256	if (enabled != apic->sw_enabled) {
 257		apic->sw_enabled = enabled;
 258		if (enabled)
 259			static_key_slow_dec_deferred(&apic_sw_disabled);
 260		else
 261			static_key_slow_inc(&apic_sw_disabled.key);
 262
 263		recalculate_apic_map(apic->vcpu->kvm);
 264	}
 
 
 
 
 265}
 266
 267static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
 268{
 269	kvm_lapic_set_reg(apic, APIC_ID, id << 24);
 270	recalculate_apic_map(apic->vcpu->kvm);
 271}
 272
 273static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
 274{
 275	kvm_lapic_set_reg(apic, APIC_LDR, id);
 276	recalculate_apic_map(apic->vcpu->kvm);
 
 
 
 
 
 
 277}
 278
 279static inline u32 kvm_apic_calc_x2apic_ldr(u32 id)
 280{
 281	return ((id >> 4) << 16) | (1 << (id & 0xf));
 282}
 283
 284static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
 285{
 286	u32 ldr = kvm_apic_calc_x2apic_ldr(id);
 287
 288	WARN_ON_ONCE(id != apic->vcpu->vcpu_id);
 289
 290	kvm_lapic_set_reg(apic, APIC_ID, id);
 291	kvm_lapic_set_reg(apic, APIC_LDR, ldr);
 292	recalculate_apic_map(apic->vcpu->kvm);
 293}
 294
 295static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
 296{
 297	return !(kvm_lapic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
 298}
 299
 300static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
 301{
 302	return kvm_lapic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
 303}
 304
 305static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
 306{
 307	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
 308}
 309
 310static inline int apic_lvtt_period(struct kvm_lapic *apic)
 311{
 312	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
 313}
 314
 315static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
 316{
 317	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
 318}
 319
 320static inline int apic_lvt_nmi_mode(u32 lvt_val)
 321{
 322	return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
 323}
 324
 
 
 
 
 
 
 
 
 
 
 325void kvm_apic_set_version(struct kvm_vcpu *vcpu)
 326{
 327	struct kvm_lapic *apic = vcpu->arch.apic;
 328	struct kvm_cpuid_entry2 *feat;
 329	u32 v = APIC_VERSION;
 330
 331	if (!lapic_in_kernel(vcpu))
 332		return;
 333
 
 
 334	/*
 335	 * KVM emulates 82093AA datasheet (with in-kernel IOAPIC implementation)
 336	 * which doesn't have EOI register; Some buggy OSes (e.g. Windows with
 337	 * Hyper-V role) disable EOI broadcast in lapic not checking for IOAPIC
 338	 * version first and level-triggered interrupts never get EOIed in
 339	 * IOAPIC.
 340	 */
 341	feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
 342	if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))) &&
 343	    !ioapic_in_kernel(vcpu->kvm))
 344		v |= APIC_LVR_DIRECTED_EOI;
 345	kvm_lapic_set_reg(apic, APIC_LVR, v);
 346}
 347
 348static const unsigned int apic_lvt_mask[KVM_APIC_LVT_NUM] = {
 349	LVT_MASK ,      /* part LVTT mask, timer mode mask added at runtime */
 350	LVT_MASK | APIC_MODE_MASK,	/* LVTTHMR */
 351	LVT_MASK | APIC_MODE_MASK,	/* LVTPC */
 352	LINT_MASK, LINT_MASK,	/* LVT0-1 */
 353	LVT_MASK		/* LVTERR */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 354};
 355
 356static int find_highest_vector(void *bitmap)
 357{
 358	int vec;
 359	u32 *reg;
 360
 361	for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
 362	     vec >= 0; vec -= APIC_VECTORS_PER_REG) {
 363		reg = bitmap + REG_POS(vec);
 364		if (*reg)
 365			return __fls(*reg) + vec;
 366	}
 367
 368	return -1;
 369}
 370
 371static u8 count_vectors(void *bitmap)
 372{
 373	int vec;
 374	u32 *reg;
 375	u8 count = 0;
 376
 377	for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
 378		reg = bitmap + REG_POS(vec);
 379		count += hweight32(*reg);
 380	}
 381
 382	return count;
 383}
 384
 385bool __kvm_apic_update_irr(u32 *pir, void *regs, int *max_irr)
 386{
 387	u32 i, vec;
 388	u32 pir_val, irr_val, prev_irr_val;
 389	int max_updated_irr;
 390
 391	max_updated_irr = -1;
 392	*max_irr = -1;
 393
 394	for (i = vec = 0; i <= 7; i++, vec += 32) {
 395		pir_val = READ_ONCE(pir[i]);
 396		irr_val = *((u32 *)(regs + APIC_IRR + i * 0x10));
 397		if (pir_val) {
 398			prev_irr_val = irr_val;
 399			irr_val |= xchg(&pir[i], 0);
 400			*((u32 *)(regs + APIC_IRR + i * 0x10)) = irr_val;
 401			if (prev_irr_val != irr_val) {
 402				max_updated_irr =
 403					__fls(irr_val ^ prev_irr_val) + vec;
 404			}
 405		}
 406		if (irr_val)
 407			*max_irr = __fls(irr_val) + vec;
 408	}
 409
 410	return ((max_updated_irr != -1) &&
 411		(max_updated_irr == *max_irr));
 412}
 413EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
 414
 415bool kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir, int *max_irr)
 416{
 417	struct kvm_lapic *apic = vcpu->arch.apic;
 418
 419	return __kvm_apic_update_irr(pir, apic->regs, max_irr);
 420}
 421EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
 422
 423static inline int apic_search_irr(struct kvm_lapic *apic)
 424{
 425	return find_highest_vector(apic->regs + APIC_IRR);
 426}
 427
 428static inline int apic_find_highest_irr(struct kvm_lapic *apic)
 429{
 430	int result;
 431
 432	/*
 433	 * Note that irr_pending is just a hint. It will be always
 434	 * true with virtual interrupt delivery enabled.
 435	 */
 436	if (!apic->irr_pending)
 437		return -1;
 438
 439	result = apic_search_irr(apic);
 440	ASSERT(result == -1 || result >= 16);
 441
 442	return result;
 443}
 444
 445static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
 446{
 447	struct kvm_vcpu *vcpu;
 448
 449	vcpu = apic->vcpu;
 450
 451	if (unlikely(vcpu->arch.apicv_active)) {
 452		/* need to update RVI */
 453		kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
 454		kvm_x86_ops->hwapic_irr_update(vcpu,
 455				apic_find_highest_irr(apic));
 456	} else {
 457		apic->irr_pending = false;
 458		kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
 459		if (apic_search_irr(apic) != -1)
 460			apic->irr_pending = true;
 461	}
 462}
 463
 
 
 
 
 
 
 464static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
 465{
 466	struct kvm_vcpu *vcpu;
 467
 468	if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
 469		return;
 470
 471	vcpu = apic->vcpu;
 472
 473	/*
 474	 * With APIC virtualization enabled, all caching is disabled
 475	 * because the processor can modify ISR under the hood.  Instead
 476	 * just set SVI.
 477	 */
 478	if (unlikely(vcpu->arch.apicv_active))
 479		kvm_x86_ops->hwapic_isr_update(vcpu, vec);
 480	else {
 481		++apic->isr_count;
 482		BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
 483		/*
 484		 * ISR (in service register) bit is set when injecting an interrupt.
 485		 * The highest vector is injected. Thus the latest bit set matches
 486		 * the highest bit in ISR.
 487		 */
 488		apic->highest_isr_cache = vec;
 489	}
 490}
 491
 492static inline int apic_find_highest_isr(struct kvm_lapic *apic)
 493{
 494	int result;
 495
 496	/*
 497	 * Note that isr_count is always 1, and highest_isr_cache
 498	 * is always -1, with APIC virtualization enabled.
 499	 */
 500	if (!apic->isr_count)
 501		return -1;
 502	if (likely(apic->highest_isr_cache != -1))
 503		return apic->highest_isr_cache;
 504
 505	result = find_highest_vector(apic->regs + APIC_ISR);
 506	ASSERT(result == -1 || result >= 16);
 507
 508	return result;
 509}
 510
 511static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
 512{
 513	struct kvm_vcpu *vcpu;
 514	if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
 515		return;
 516
 517	vcpu = apic->vcpu;
 518
 519	/*
 520	 * We do get here for APIC virtualization enabled if the guest
 521	 * uses the Hyper-V APIC enlightenment.  In this case we may need
 522	 * to trigger a new interrupt delivery by writing the SVI field;
 523	 * on the other hand isr_count and highest_isr_cache are unused
 524	 * and must be left alone.
 525	 */
 526	if (unlikely(vcpu->arch.apicv_active))
 527		kvm_x86_ops->hwapic_isr_update(vcpu,
 528					       apic_find_highest_isr(apic));
 529	else {
 530		--apic->isr_count;
 531		BUG_ON(apic->isr_count < 0);
 532		apic->highest_isr_cache = -1;
 533	}
 534}
 535
 536int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
 537{
 538	/* This may race with setting of irr in __apic_accept_irq() and
 539	 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
 540	 * will cause vmexit immediately and the value will be recalculated
 541	 * on the next vmentry.
 542	 */
 543	return apic_find_highest_irr(vcpu->arch.apic);
 544}
 545EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
 546
 547static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
 548			     int vector, int level, int trig_mode,
 549			     struct dest_map *dest_map);
 550
 551int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
 552		     struct dest_map *dest_map)
 553{
 554	struct kvm_lapic *apic = vcpu->arch.apic;
 555
 556	return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
 557			irq->level, irq->trig_mode, dest_map);
 558}
 559
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 560int kvm_pv_send_ipi(struct kvm *kvm, unsigned long ipi_bitmap_low,
 561		    unsigned long ipi_bitmap_high, u32 min,
 562		    unsigned long icr, int op_64_bit)
 563{
 564	int i;
 565	struct kvm_apic_map *map;
 566	struct kvm_vcpu *vcpu;
 567	struct kvm_lapic_irq irq = {0};
 568	int cluster_size = op_64_bit ? 64 : 32;
 569	int count = 0;
 
 
 
 570
 571	irq.vector = icr & APIC_VECTOR_MASK;
 572	irq.delivery_mode = icr & APIC_MODE_MASK;
 573	irq.level = (icr & APIC_INT_ASSERT) != 0;
 574	irq.trig_mode = icr & APIC_INT_LEVELTRIG;
 575
 576	if (icr & APIC_DEST_MASK)
 577		return -KVM_EINVAL;
 578	if (icr & APIC_SHORT_MASK)
 579		return -KVM_EINVAL;
 580
 581	rcu_read_lock();
 582	map = rcu_dereference(kvm->arch.apic_map);
 583
 584	if (unlikely(!map)) {
 585		count = -EOPNOTSUPP;
 586		goto out;
 587	}
 588
 589	if (min > map->max_apic_id)
 590		goto out;
 591	/* Bits above cluster_size are masked in the caller.  */
 592	for_each_set_bit(i, &ipi_bitmap_low,
 593		min((u32)BITS_PER_LONG, (map->max_apic_id - min + 1))) {
 594		if (map->phys_map[min + i]) {
 595			vcpu = map->phys_map[min + i]->vcpu;
 596			count += kvm_apic_set_irq(vcpu, &irq, NULL);
 597		}
 598	}
 599
 600	min += cluster_size;
 601
 602	if (min > map->max_apic_id)
 603		goto out;
 604
 605	for_each_set_bit(i, &ipi_bitmap_high,
 606		min((u32)BITS_PER_LONG, (map->max_apic_id - min + 1))) {
 607		if (map->phys_map[min + i]) {
 608			vcpu = map->phys_map[min + i]->vcpu;
 609			count += kvm_apic_set_irq(vcpu, &irq, NULL);
 610		}
 611	}
 612
 613out:
 614	rcu_read_unlock();
 615	return count;
 616}
 617
 618static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
 619{
 620
 621	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
 622				      sizeof(val));
 623}
 624
 625static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
 626{
 627
 628	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
 629				      sizeof(*val));
 630}
 631
 632static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
 633{
 634	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
 635}
 636
 637static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
 638{
 639	u8 val;
 640	if (pv_eoi_get_user(vcpu, &val) < 0)
 641		printk(KERN_WARNING "Can't read EOI MSR value: 0x%llx\n",
 642			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
 643	return val & 0x1;
 644}
 645
 646static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
 647{
 648	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
 649		printk(KERN_WARNING "Can't set EOI MSR value: 0x%llx\n",
 650			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
 651		return;
 652	}
 653	__set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
 654}
 655
 656static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
 657{
 658	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
 659		printk(KERN_WARNING "Can't clear EOI MSR value: 0x%llx\n",
 660			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
 661		return;
 662	}
 
 
 
 
 
 
 
 
 
 
 663	__clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
 
 
 664}
 665
 666static int apic_has_interrupt_for_ppr(struct kvm_lapic *apic, u32 ppr)
 667{
 668	int highest_irr;
 669	if (apic->vcpu->arch.apicv_active)
 670		highest_irr = kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
 671	else
 672		highest_irr = apic_find_highest_irr(apic);
 673	if (highest_irr == -1 || (highest_irr & 0xF0) <= ppr)
 674		return -1;
 675	return highest_irr;
 676}
 677
 678static bool __apic_update_ppr(struct kvm_lapic *apic, u32 *new_ppr)
 679{
 680	u32 tpr, isrv, ppr, old_ppr;
 681	int isr;
 682
 683	old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
 684	tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
 685	isr = apic_find_highest_isr(apic);
 686	isrv = (isr != -1) ? isr : 0;
 687
 688	if ((tpr & 0xf0) >= (isrv & 0xf0))
 689		ppr = tpr & 0xff;
 690	else
 691		ppr = isrv & 0xf0;
 692
 693	*new_ppr = ppr;
 694	if (old_ppr != ppr)
 695		kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
 696
 697	return ppr < old_ppr;
 698}
 699
 700static void apic_update_ppr(struct kvm_lapic *apic)
 701{
 702	u32 ppr;
 703
 704	if (__apic_update_ppr(apic, &ppr) &&
 705	    apic_has_interrupt_for_ppr(apic, ppr) != -1)
 706		kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
 707}
 708
 709void kvm_apic_update_ppr(struct kvm_vcpu *vcpu)
 710{
 711	apic_update_ppr(vcpu->arch.apic);
 712}
 713EXPORT_SYMBOL_GPL(kvm_apic_update_ppr);
 714
 715static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
 716{
 717	kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
 718	apic_update_ppr(apic);
 719}
 720
 721static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
 722{
 723	return mda == (apic_x2apic_mode(apic) ?
 724			X2APIC_BROADCAST : APIC_BROADCAST);
 725}
 726
 727static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
 728{
 729	if (kvm_apic_broadcast(apic, mda))
 730		return true;
 731
 732	if (apic_x2apic_mode(apic))
 733		return mda == kvm_x2apic_id(apic);
 734
 735	/*
 736	 * Hotplug hack: Make LAPIC in xAPIC mode also accept interrupts as if
 737	 * it were in x2APIC mode.  Hotplugged VCPUs start in xAPIC mode and
 738	 * this allows unique addressing of VCPUs with APIC ID over 0xff.
 739	 * The 0xff condition is needed because writeable xAPIC ID.
 
 
 
 740	 */
 741	if (kvm_x2apic_id(apic) > 0xff && mda == kvm_x2apic_id(apic))
 742		return true;
 743
 744	return mda == kvm_xapic_id(apic);
 745}
 746
 747static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
 748{
 749	u32 logical_id;
 750
 751	if (kvm_apic_broadcast(apic, mda))
 752		return true;
 753
 754	logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
 755
 756	if (apic_x2apic_mode(apic))
 757		return ((logical_id >> 16) == (mda >> 16))
 758		       && (logical_id & mda & 0xffff) != 0;
 759
 760	logical_id = GET_APIC_LOGICAL_ID(logical_id);
 761
 762	switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
 763	case APIC_DFR_FLAT:
 764		return (logical_id & mda) != 0;
 765	case APIC_DFR_CLUSTER:
 766		return ((logical_id >> 4) == (mda >> 4))
 767		       && (logical_id & mda & 0xf) != 0;
 768	default:
 769		return false;
 770	}
 771}
 772
 773/* The KVM local APIC implementation has two quirks:
 774 *
 775 *  - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
 776 *    in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
 777 *    KVM doesn't do that aliasing.
 778 *
 779 *  - in-kernel IOAPIC messages have to be delivered directly to
 780 *    x2APIC, because the kernel does not support interrupt remapping.
 781 *    In order to support broadcast without interrupt remapping, x2APIC
 782 *    rewrites the destination of non-IPI messages from APIC_BROADCAST
 783 *    to X2APIC_BROADCAST.
 784 *
 785 * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API.  This is
 786 * important when userspace wants to use x2APIC-format MSIs, because
 787 * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
 788 */
 789static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
 790		struct kvm_lapic *source, struct kvm_lapic *target)
 791{
 792	bool ipi = source != NULL;
 793
 794	if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
 795	    !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
 796		return X2APIC_BROADCAST;
 797
 798	return dest_id;
 799}
 800
 801bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
 802			   int short_hand, unsigned int dest, int dest_mode)
 803{
 804	struct kvm_lapic *target = vcpu->arch.apic;
 805	u32 mda = kvm_apic_mda(vcpu, dest, source, target);
 806
 807	ASSERT(target);
 808	switch (short_hand) {
 809	case APIC_DEST_NOSHORT:
 810		if (dest_mode == APIC_DEST_PHYSICAL)
 811			return kvm_apic_match_physical_addr(target, mda);
 812		else
 813			return kvm_apic_match_logical_addr(target, mda);
 814	case APIC_DEST_SELF:
 815		return target == source;
 816	case APIC_DEST_ALLINC:
 817		return true;
 818	case APIC_DEST_ALLBUT:
 819		return target != source;
 820	default:
 821		return false;
 822	}
 823}
 824EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
 825
 826int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
 827		       const unsigned long *bitmap, u32 bitmap_size)
 828{
 829	u32 mod;
 830	int i, idx = -1;
 831
 832	mod = vector % dest_vcpus;
 833
 834	for (i = 0; i <= mod; i++) {
 835		idx = find_next_bit(bitmap, bitmap_size, idx + 1);
 836		BUG_ON(idx == bitmap_size);
 837	}
 838
 839	return idx;
 840}
 841
 842static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
 843{
 844	if (!kvm->arch.disabled_lapic_found) {
 845		kvm->arch.disabled_lapic_found = true;
 846		printk(KERN_INFO
 847		       "Disabled LAPIC found during irq injection\n");
 848	}
 849}
 850
 851static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
 852		struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
 853{
 854	if (kvm->arch.x2apic_broadcast_quirk_disabled) {
 855		if ((irq->dest_id == APIC_BROADCAST &&
 856				map->mode != KVM_APIC_MODE_X2APIC))
 857			return true;
 858		if (irq->dest_id == X2APIC_BROADCAST)
 859			return true;
 860	} else {
 861		bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
 862		if (irq->dest_id == (x2apic_ipi ?
 863		                     X2APIC_BROADCAST : APIC_BROADCAST))
 864			return true;
 865	}
 866
 867	return false;
 868}
 869
 870/* Return true if the interrupt can be handled by using *bitmap as index mask
 871 * for valid destinations in *dst array.
 872 * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
 873 * Note: we may have zero kvm_lapic destinations when we return true, which
 874 * means that the interrupt should be dropped.  In this case, *bitmap would be
 875 * zero and *dst undefined.
 876 */
 877static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
 878		struct kvm_lapic **src, struct kvm_lapic_irq *irq,
 879		struct kvm_apic_map *map, struct kvm_lapic ***dst,
 880		unsigned long *bitmap)
 881{
 882	int i, lowest;
 883
 884	if (irq->shorthand == APIC_DEST_SELF && src) {
 885		*dst = src;
 886		*bitmap = 1;
 887		return true;
 888	} else if (irq->shorthand)
 889		return false;
 890
 891	if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
 892		return false;
 893
 894	if (irq->dest_mode == APIC_DEST_PHYSICAL) {
 895		if (irq->dest_id > map->max_apic_id) {
 896			*bitmap = 0;
 897		} else {
 898			u32 dest_id = array_index_nospec(irq->dest_id, map->max_apic_id + 1);
 899			*dst = &map->phys_map[dest_id];
 900			*bitmap = 1;
 901		}
 902		return true;
 903	}
 904
 905	*bitmap = 0;
 906	if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
 907				(u16 *)bitmap))
 908		return false;
 909
 910	if (!kvm_lowest_prio_delivery(irq))
 911		return true;
 912
 913	if (!kvm_vector_hashing_enabled()) {
 914		lowest = -1;
 915		for_each_set_bit(i, bitmap, 16) {
 916			if (!(*dst)[i])
 917				continue;
 918			if (lowest < 0)
 919				lowest = i;
 920			else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
 921						(*dst)[lowest]->vcpu) < 0)
 922				lowest = i;
 923		}
 924	} else {
 925		if (!*bitmap)
 926			return true;
 927
 928		lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
 929				bitmap, 16);
 930
 931		if (!(*dst)[lowest]) {
 932			kvm_apic_disabled_lapic_found(kvm);
 933			*bitmap = 0;
 934			return true;
 935		}
 936	}
 937
 938	*bitmap = (lowest >= 0) ? 1 << lowest : 0;
 939
 940	return true;
 941}
 942
 943bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
 944		struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
 945{
 946	struct kvm_apic_map *map;
 947	unsigned long bitmap;
 948	struct kvm_lapic **dst = NULL;
 949	int i;
 950	bool ret;
 951
 952	*r = -1;
 953
 954	if (irq->shorthand == APIC_DEST_SELF) {
 
 
 
 
 955		*r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
 956		return true;
 957	}
 958
 959	rcu_read_lock();
 960	map = rcu_dereference(kvm->arch.apic_map);
 961
 962	ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
 963	if (ret) {
 964		*r = 0;
 965		for_each_set_bit(i, &bitmap, 16) {
 966			if (!dst[i])
 967				continue;
 968			*r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
 969		}
 970	}
 971
 972	rcu_read_unlock();
 973	return ret;
 974}
 975
 976/*
 977 * This routine tries to handler interrupts in posted mode, here is how
 978 * it deals with different cases:
 979 * - For single-destination interrupts, handle it in posted mode
 980 * - Else if vector hashing is enabled and it is a lowest-priority
 981 *   interrupt, handle it in posted mode and use the following mechanism
 982 *   to find the destinaiton vCPU.
 983 *	1. For lowest-priority interrupts, store all the possible
 984 *	   destination vCPUs in an array.
 985 *	2. Use "guest vector % max number of destination vCPUs" to find
 986 *	   the right destination vCPU in the array for the lowest-priority
 987 *	   interrupt.
 988 * - Otherwise, use remapped mode to inject the interrupt.
 989 */
 990bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
 991			struct kvm_vcpu **dest_vcpu)
 992{
 993	struct kvm_apic_map *map;
 994	unsigned long bitmap;
 995	struct kvm_lapic **dst = NULL;
 996	bool ret = false;
 997
 998	if (irq->shorthand)
 999		return false;
1000
1001	rcu_read_lock();
1002	map = rcu_dereference(kvm->arch.apic_map);
1003
1004	if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
1005			hweight16(bitmap) == 1) {
1006		unsigned long i = find_first_bit(&bitmap, 16);
1007
1008		if (dst[i]) {
1009			*dest_vcpu = dst[i]->vcpu;
1010			ret = true;
1011		}
1012	}
1013
1014	rcu_read_unlock();
1015	return ret;
1016}
1017
1018/*
1019 * Add a pending IRQ into lapic.
1020 * Return 1 if successfully added and 0 if discarded.
1021 */
1022static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
1023			     int vector, int level, int trig_mode,
1024			     struct dest_map *dest_map)
1025{
1026	int result = 0;
1027	struct kvm_vcpu *vcpu = apic->vcpu;
1028
1029	trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
1030				  trig_mode, vector);
1031	switch (delivery_mode) {
1032	case APIC_DM_LOWEST:
1033		vcpu->arch.apic_arb_prio++;
1034		/* fall through */
1035	case APIC_DM_FIXED:
1036		if (unlikely(trig_mode && !level))
1037			break;
1038
1039		/* FIXME add logic for vcpu on reset */
1040		if (unlikely(!apic_enabled(apic)))
1041			break;
1042
1043		result = 1;
1044
1045		if (dest_map) {
1046			__set_bit(vcpu->vcpu_id, dest_map->map);
1047			dest_map->vectors[vcpu->vcpu_id] = vector;
1048		}
1049
1050		if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
1051			if (trig_mode)
1052				kvm_lapic_set_vector(vector,
1053						     apic->regs + APIC_TMR);
1054			else
1055				kvm_lapic_clear_vector(vector,
1056						       apic->regs + APIC_TMR);
1057		}
1058
1059		if (vcpu->arch.apicv_active)
1060			kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
1061		else {
1062			kvm_lapic_set_irr(vector, apic);
1063
1064			kvm_make_request(KVM_REQ_EVENT, vcpu);
1065			kvm_vcpu_kick(vcpu);
1066		}
1067		break;
1068
1069	case APIC_DM_REMRD:
1070		result = 1;
1071		vcpu->arch.pv.pv_unhalted = 1;
1072		kvm_make_request(KVM_REQ_EVENT, vcpu);
1073		kvm_vcpu_kick(vcpu);
1074		break;
1075
1076	case APIC_DM_SMI:
1077		result = 1;
1078		kvm_make_request(KVM_REQ_SMI, vcpu);
1079		kvm_vcpu_kick(vcpu);
 
1080		break;
1081
1082	case APIC_DM_NMI:
1083		result = 1;
1084		kvm_inject_nmi(vcpu);
1085		kvm_vcpu_kick(vcpu);
1086		break;
1087
1088	case APIC_DM_INIT:
1089		if (!trig_mode || level) {
1090			result = 1;
1091			/* assumes that there are only KVM_APIC_INIT/SIPI */
1092			apic->pending_events = (1UL << KVM_APIC_INIT);
1093			/* make sure pending_events is visible before sending
1094			 * the request */
1095			smp_wmb();
1096			kvm_make_request(KVM_REQ_EVENT, vcpu);
1097			kvm_vcpu_kick(vcpu);
1098		}
1099		break;
1100
1101	case APIC_DM_STARTUP:
1102		result = 1;
1103		apic->sipi_vector = vector;
1104		/* make sure sipi_vector is visible for the receiver */
1105		smp_wmb();
1106		set_bit(KVM_APIC_SIPI, &apic->pending_events);
1107		kvm_make_request(KVM_REQ_EVENT, vcpu);
1108		kvm_vcpu_kick(vcpu);
1109		break;
1110
1111	case APIC_DM_EXTINT:
1112		/*
1113		 * Should only be called by kvm_apic_local_deliver() with LVT0,
1114		 * before NMI watchdog was enabled. Already handled by
1115		 * kvm_apic_accept_pic_intr().
1116		 */
1117		break;
1118
1119	default:
1120		printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
1121		       delivery_mode);
1122		break;
1123	}
1124	return result;
1125}
1126
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1127int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1128{
1129	return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
1130}
1131
1132static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
1133{
1134	return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
1135}
1136
1137static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1138{
1139	int trigger_mode;
1140
1141	/* Eoi the ioapic only if the ioapic doesn't own the vector. */
1142	if (!kvm_ioapic_handles_vector(apic, vector))
1143		return;
1144
1145	/* Request a KVM exit to inform the userspace IOAPIC. */
1146	if (irqchip_split(apic->vcpu->kvm)) {
1147		apic->vcpu->arch.pending_ioapic_eoi = vector;
1148		kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1149		return;
1150	}
1151
1152	if (apic_test_vector(vector, apic->regs + APIC_TMR))
1153		trigger_mode = IOAPIC_LEVEL_TRIG;
1154	else
1155		trigger_mode = IOAPIC_EDGE_TRIG;
1156
1157	kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1158}
1159
1160static int apic_set_eoi(struct kvm_lapic *apic)
1161{
1162	int vector = apic_find_highest_isr(apic);
1163
1164	trace_kvm_eoi(apic, vector);
1165
1166	/*
1167	 * Not every write EOI will has corresponding ISR,
1168	 * one example is when Kernel check timer on setup_IO_APIC
1169	 */
1170	if (vector == -1)
1171		return vector;
1172
1173	apic_clear_isr(vector, apic);
1174	apic_update_ppr(apic);
1175
1176	if (test_bit(vector, vcpu_to_synic(apic->vcpu)->vec_bitmap))
 
1177		kvm_hv_synic_send_eoi(apic->vcpu, vector);
1178
1179	kvm_ioapic_send_eoi(apic, vector);
1180	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1181	return vector;
1182}
1183
1184/*
1185 * this interface assumes a trap-like exit, which has already finished
1186 * desired side effect including vISR and vPPR update.
1187 */
1188void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1189{
1190	struct kvm_lapic *apic = vcpu->arch.apic;
1191
1192	trace_kvm_eoi(apic, vector);
1193
1194	kvm_ioapic_send_eoi(apic, vector);
1195	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1196}
1197EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1198
1199static void apic_send_ipi(struct kvm_lapic *apic, u32 icr_low, u32 icr_high)
1200{
1201	struct kvm_lapic_irq irq;
1202
 
 
 
1203	irq.vector = icr_low & APIC_VECTOR_MASK;
1204	irq.delivery_mode = icr_low & APIC_MODE_MASK;
1205	irq.dest_mode = icr_low & APIC_DEST_MASK;
1206	irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1207	irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1208	irq.shorthand = icr_low & APIC_SHORT_MASK;
1209	irq.msi_redir_hint = false;
1210	if (apic_x2apic_mode(apic))
1211		irq.dest_id = icr_high;
1212	else
1213		irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
1214
1215	trace_kvm_apic_ipi(icr_low, irq.dest_id);
1216
1217	kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1218}
 
1219
1220static u32 apic_get_tmcct(struct kvm_lapic *apic)
1221{
1222	ktime_t remaining, now;
1223	s64 ns;
1224	u32 tmcct;
1225
1226	ASSERT(apic != NULL);
1227
1228	/* if initial count is 0, current count should also be 0 */
1229	if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1230		apic->lapic_timer.period == 0)
1231		return 0;
1232
1233	now = ktime_get();
1234	remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1235	if (ktime_to_ns(remaining) < 0)
1236		remaining = 0;
1237
1238	ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1239	tmcct = div64_u64(ns,
1240			 (APIC_BUS_CYCLE_NS * apic->divide_count));
1241
1242	return tmcct;
1243}
1244
1245static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1246{
1247	struct kvm_vcpu *vcpu = apic->vcpu;
1248	struct kvm_run *run = vcpu->run;
1249
1250	kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1251	run->tpr_access.rip = kvm_rip_read(vcpu);
1252	run->tpr_access.is_write = write;
1253}
1254
1255static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1256{
1257	if (apic->vcpu->arch.tpr_access_reporting)
1258		__report_tpr_access(apic, write);
1259}
1260
1261static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1262{
1263	u32 val = 0;
1264
1265	if (offset >= LAPIC_MMIO_LENGTH)
1266		return 0;
1267
1268	switch (offset) {
1269	case APIC_ARBPRI:
1270		break;
1271
1272	case APIC_TMCCT:	/* Timer CCR */
1273		if (apic_lvtt_tscdeadline(apic))
1274			return 0;
1275
1276		val = apic_get_tmcct(apic);
1277		break;
1278	case APIC_PROCPRI:
1279		apic_update_ppr(apic);
1280		val = kvm_lapic_get_reg(apic, offset);
1281		break;
1282	case APIC_TASKPRI:
1283		report_tpr_access(apic, false);
1284		/* fall thru */
1285	default:
1286		val = kvm_lapic_get_reg(apic, offset);
1287		break;
1288	}
1289
1290	return val;
1291}
1292
1293static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1294{
1295	return container_of(dev, struct kvm_lapic, dev);
1296}
1297
1298#define APIC_REG_MASK(reg)	(1ull << ((reg) >> 4))
1299#define APIC_REGS_MASK(first, count) \
1300	(APIC_REG_MASK(first) * ((1ull << (count)) - 1))
1301
1302int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1303		void *data)
1304{
1305	unsigned char alignment = offset & 0xf;
1306	u32 result;
1307	/* this bitmask has a bit cleared for each reserved register */
1308	u64 valid_reg_mask =
1309		APIC_REG_MASK(APIC_ID) |
1310		APIC_REG_MASK(APIC_LVR) |
1311		APIC_REG_MASK(APIC_TASKPRI) |
1312		APIC_REG_MASK(APIC_PROCPRI) |
1313		APIC_REG_MASK(APIC_LDR) |
1314		APIC_REG_MASK(APIC_DFR) |
1315		APIC_REG_MASK(APIC_SPIV) |
1316		APIC_REGS_MASK(APIC_ISR, APIC_ISR_NR) |
1317		APIC_REGS_MASK(APIC_TMR, APIC_ISR_NR) |
1318		APIC_REGS_MASK(APIC_IRR, APIC_ISR_NR) |
1319		APIC_REG_MASK(APIC_ESR) |
1320		APIC_REG_MASK(APIC_ICR) |
1321		APIC_REG_MASK(APIC_ICR2) |
1322		APIC_REG_MASK(APIC_LVTT) |
1323		APIC_REG_MASK(APIC_LVTTHMR) |
1324		APIC_REG_MASK(APIC_LVTPC) |
1325		APIC_REG_MASK(APIC_LVT0) |
1326		APIC_REG_MASK(APIC_LVT1) |
1327		APIC_REG_MASK(APIC_LVTERR) |
1328		APIC_REG_MASK(APIC_TMICT) |
1329		APIC_REG_MASK(APIC_TMCCT) |
1330		APIC_REG_MASK(APIC_TDCR);
1331
1332	/* ARBPRI is not valid on x2APIC */
 
 
 
 
 
 
 
1333	if (!apic_x2apic_mode(apic))
1334		valid_reg_mask |= APIC_REG_MASK(APIC_ARBPRI);
 
 
 
 
 
 
1335
1336	if (offset > 0x3f0 || !(valid_reg_mask & APIC_REG_MASK(offset)))
1337		return 1;
1338
1339	result = __apic_read(apic, offset & ~0xf);
1340
1341	trace_kvm_apic_read(offset, result);
1342
1343	switch (len) {
1344	case 1:
1345	case 2:
1346	case 4:
1347		memcpy(data, (char *)&result + alignment, len);
1348		break;
1349	default:
1350		printk(KERN_ERR "Local APIC read with len = %x, "
1351		       "should be 1,2, or 4 instead\n", len);
1352		break;
1353	}
1354	return 0;
1355}
1356EXPORT_SYMBOL_GPL(kvm_lapic_reg_read);
1357
1358static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1359{
1360	return addr >= apic->base_address &&
1361		addr < apic->base_address + LAPIC_MMIO_LENGTH;
1362}
1363
1364static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1365			   gpa_t address, int len, void *data)
1366{
1367	struct kvm_lapic *apic = to_lapic(this);
1368	u32 offset = address - apic->base_address;
1369
1370	if (!apic_mmio_in_range(apic, address))
1371		return -EOPNOTSUPP;
1372
1373	if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
1374		if (!kvm_check_has_quirk(vcpu->kvm,
1375					 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
1376			return -EOPNOTSUPP;
1377
1378		memset(data, 0xff, len);
1379		return 0;
1380	}
1381
1382	kvm_lapic_reg_read(apic, offset, len, data);
1383
1384	return 0;
1385}
1386
1387static void update_divide_count(struct kvm_lapic *apic)
1388{
1389	u32 tmp1, tmp2, tdcr;
1390
1391	tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1392	tmp1 = tdcr & 0xf;
1393	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1394	apic->divide_count = 0x1 << (tmp2 & 0x7);
1395}
1396
1397static void limit_periodic_timer_frequency(struct kvm_lapic *apic)
1398{
1399	/*
1400	 * Do not allow the guest to program periodic timers with small
1401	 * interval, since the hrtimers are not throttled by the host
1402	 * scheduler.
1403	 */
1404	if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1405		s64 min_period = min_timer_period_us * 1000LL;
1406
1407		if (apic->lapic_timer.period < min_period) {
1408			pr_info_ratelimited(
1409			    "kvm: vcpu %i: requested %lld ns "
1410			    "lapic timer period limited to %lld ns\n",
1411			    apic->vcpu->vcpu_id,
1412			    apic->lapic_timer.period, min_period);
1413			apic->lapic_timer.period = min_period;
1414		}
1415	}
1416}
1417
 
 
 
 
 
 
 
 
 
 
 
 
1418static void apic_update_lvtt(struct kvm_lapic *apic)
1419{
1420	u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1421			apic->lapic_timer.timer_mode_mask;
1422
1423	if (apic->lapic_timer.timer_mode != timer_mode) {
1424		if (apic_lvtt_tscdeadline(apic) != (timer_mode ==
1425				APIC_LVT_TIMER_TSCDEADLINE)) {
1426			hrtimer_cancel(&apic->lapic_timer.timer);
1427			kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1428			apic->lapic_timer.period = 0;
1429			apic->lapic_timer.tscdeadline = 0;
1430		}
1431		apic->lapic_timer.timer_mode = timer_mode;
1432		limit_periodic_timer_frequency(apic);
1433	}
1434}
1435
1436/*
1437 * On APICv, this test will cause a busy wait
1438 * during a higher-priority task.
1439 */
1440
1441static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1442{
1443	struct kvm_lapic *apic = vcpu->arch.apic;
1444	u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1445
1446	if (kvm_apic_hw_enabled(apic)) {
1447		int vec = reg & APIC_VECTOR_MASK;
1448		void *bitmap = apic->regs + APIC_ISR;
1449
1450		if (vcpu->arch.apicv_active)
1451			bitmap = apic->regs + APIC_IRR;
1452
1453		if (apic_test_vector(vec, bitmap))
1454			return true;
1455	}
1456	return false;
1457}
1458
1459static inline void __wait_lapic_expire(struct kvm_vcpu *vcpu, u64 guest_cycles)
1460{
1461	u64 timer_advance_ns = vcpu->arch.apic->lapic_timer.timer_advance_ns;
1462
1463	/*
1464	 * If the guest TSC is running at a different ratio than the host, then
1465	 * convert the delay to nanoseconds to achieve an accurate delay.  Note
1466	 * that __delay() uses delay_tsc whenever the hardware has TSC, thus
1467	 * always for VMX enabled hardware.
1468	 */
1469	if (vcpu->arch.tsc_scaling_ratio == kvm_default_tsc_scaling_ratio) {
1470		__delay(min(guest_cycles,
1471			nsec_to_cycles(vcpu, timer_advance_ns)));
1472	} else {
1473		u64 delay_ns = guest_cycles * 1000000ULL;
1474		do_div(delay_ns, vcpu->arch.virtual_tsc_khz);
1475		ndelay(min_t(u32, delay_ns, timer_advance_ns));
1476	}
1477}
1478
1479static inline void adjust_lapic_timer_advance(struct kvm_vcpu *vcpu,
1480					      s64 advance_expire_delta)
1481{
1482	struct kvm_lapic *apic = vcpu->arch.apic;
1483	u32 timer_advance_ns = apic->lapic_timer.timer_advance_ns;
1484	u64 ns;
1485
1486	/* Do not adjust for tiny fluctuations or large random spikes. */
1487	if (abs(advance_expire_delta) > LAPIC_TIMER_ADVANCE_ADJUST_MAX ||
1488	    abs(advance_expire_delta) < LAPIC_TIMER_ADVANCE_ADJUST_MIN)
1489		return;
1490
1491	/* too early */
1492	if (advance_expire_delta < 0) {
1493		ns = -advance_expire_delta * 1000000ULL;
1494		do_div(ns, vcpu->arch.virtual_tsc_khz);
1495		timer_advance_ns -= ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1496	} else {
1497	/* too late */
1498		ns = advance_expire_delta * 1000000ULL;
1499		do_div(ns, vcpu->arch.virtual_tsc_khz);
1500		timer_advance_ns += ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1501	}
1502
1503	if (unlikely(timer_advance_ns > LAPIC_TIMER_ADVANCE_NS_MAX))
1504		timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
1505	apic->lapic_timer.timer_advance_ns = timer_advance_ns;
1506}
1507
1508static void __kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1509{
1510	struct kvm_lapic *apic = vcpu->arch.apic;
1511	u64 guest_tsc, tsc_deadline;
1512
1513	if (apic->lapic_timer.expired_tscdeadline == 0)
1514		return;
1515
1516	tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1517	apic->lapic_timer.expired_tscdeadline = 0;
1518	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1519	apic->lapic_timer.advance_expire_delta = guest_tsc - tsc_deadline;
 
 
 
 
 
 
 
 
 
 
 
1520
1521	if (guest_tsc < tsc_deadline)
1522		__wait_lapic_expire(vcpu, tsc_deadline - guest_tsc);
1523
1524	if (lapic_timer_advance_dynamic)
1525		adjust_lapic_timer_advance(vcpu, apic->lapic_timer.advance_expire_delta);
1526}
1527
1528void kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1529{
1530	if (lapic_timer_int_injected(vcpu))
 
 
 
1531		__kvm_wait_lapic_expire(vcpu);
1532}
1533EXPORT_SYMBOL_GPL(kvm_wait_lapic_expire);
1534
1535static void kvm_apic_inject_pending_timer_irqs(struct kvm_lapic *apic)
1536{
1537	struct kvm_timer *ktimer = &apic->lapic_timer;
1538
1539	kvm_apic_local_deliver(apic, APIC_LVTT);
1540	if (apic_lvtt_tscdeadline(apic))
1541		ktimer->tscdeadline = 0;
1542	if (apic_lvtt_oneshot(apic)) {
1543		ktimer->tscdeadline = 0;
1544		ktimer->target_expiration = 0;
1545	}
1546}
1547
1548static void apic_timer_expired(struct kvm_lapic *apic)
1549{
1550	struct kvm_vcpu *vcpu = apic->vcpu;
1551	struct kvm_timer *ktimer = &apic->lapic_timer;
1552
1553	if (atomic_read(&apic->lapic_timer.pending))
1554		return;
1555
1556	if (apic_lvtt_tscdeadline(apic) || ktimer->hv_timer_in_use)
1557		ktimer->expired_tscdeadline = ktimer->tscdeadline;
1558
 
 
 
 
 
 
1559	if (kvm_use_posted_timer_interrupt(apic->vcpu)) {
1560		if (apic->lapic_timer.timer_advance_ns)
 
 
 
 
 
 
 
 
1561			__kvm_wait_lapic_expire(vcpu);
1562		kvm_apic_inject_pending_timer_irqs(apic);
1563		return;
1564	}
1565
1566	atomic_inc(&apic->lapic_timer.pending);
1567	kvm_set_pending_timer(vcpu);
 
 
1568}
1569
1570static void start_sw_tscdeadline(struct kvm_lapic *apic)
1571{
1572	struct kvm_timer *ktimer = &apic->lapic_timer;
1573	u64 guest_tsc, tscdeadline = ktimer->tscdeadline;
1574	u64 ns = 0;
1575	ktime_t expire;
1576	struct kvm_vcpu *vcpu = apic->vcpu;
1577	unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1578	unsigned long flags;
1579	ktime_t now;
1580
1581	if (unlikely(!tscdeadline || !this_tsc_khz))
1582		return;
1583
1584	local_irq_save(flags);
1585
1586	now = ktime_get();
1587	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1588
1589	ns = (tscdeadline - guest_tsc) * 1000000ULL;
1590	do_div(ns, this_tsc_khz);
1591
1592	if (likely(tscdeadline > guest_tsc) &&
1593	    likely(ns > apic->lapic_timer.timer_advance_ns)) {
1594		expire = ktime_add_ns(now, ns);
1595		expire = ktime_sub_ns(expire, ktimer->timer_advance_ns);
1596		hrtimer_start(&ktimer->timer, expire, HRTIMER_MODE_ABS_HARD);
1597	} else
1598		apic_timer_expired(apic);
1599
1600	local_irq_restore(flags);
1601}
1602
 
 
 
 
 
1603static void update_target_expiration(struct kvm_lapic *apic, uint32_t old_divisor)
1604{
1605	ktime_t now, remaining;
1606	u64 ns_remaining_old, ns_remaining_new;
1607
1608	apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
1609		* APIC_BUS_CYCLE_NS * apic->divide_count;
1610	limit_periodic_timer_frequency(apic);
1611
1612	now = ktime_get();
1613	remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1614	if (ktime_to_ns(remaining) < 0)
1615		remaining = 0;
1616
1617	ns_remaining_old = ktime_to_ns(remaining);
1618	ns_remaining_new = mul_u64_u32_div(ns_remaining_old,
1619	                                   apic->divide_count, old_divisor);
1620
1621	apic->lapic_timer.tscdeadline +=
1622		nsec_to_cycles(apic->vcpu, ns_remaining_new) -
1623		nsec_to_cycles(apic->vcpu, ns_remaining_old);
1624	apic->lapic_timer.target_expiration = ktime_add_ns(now, ns_remaining_new);
1625}
1626
1627static bool set_target_expiration(struct kvm_lapic *apic)
1628{
1629	ktime_t now;
1630	u64 tscl = rdtsc();
 
1631
1632	now = ktime_get();
1633	apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
1634		* APIC_BUS_CYCLE_NS * apic->divide_count;
1635
1636	if (!apic->lapic_timer.period) {
1637		apic->lapic_timer.tscdeadline = 0;
1638		return false;
1639	}
1640
1641	limit_periodic_timer_frequency(apic);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1642
1643	apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1644		nsec_to_cycles(apic->vcpu, apic->lapic_timer.period);
1645	apic->lapic_timer.target_expiration = ktime_add_ns(now, apic->lapic_timer.period);
1646
1647	return true;
1648}
1649
1650static void advance_periodic_target_expiration(struct kvm_lapic *apic)
1651{
1652	ktime_t now = ktime_get();
1653	u64 tscl = rdtsc();
1654	ktime_t delta;
1655
1656	/*
1657	 * Synchronize both deadlines to the same time source or
1658	 * differences in the periods (caused by differences in the
1659	 * underlying clocks or numerical approximation errors) will
1660	 * cause the two to drift apart over time as the errors
1661	 * accumulate.
1662	 */
1663	apic->lapic_timer.target_expiration =
1664		ktime_add_ns(apic->lapic_timer.target_expiration,
1665				apic->lapic_timer.period);
1666	delta = ktime_sub(apic->lapic_timer.target_expiration, now);
1667	apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1668		nsec_to_cycles(apic->vcpu, delta);
1669}
1670
1671static void start_sw_period(struct kvm_lapic *apic)
1672{
1673	if (!apic->lapic_timer.period)
1674		return;
1675
1676	if (ktime_after(ktime_get(),
1677			apic->lapic_timer.target_expiration)) {
1678		apic_timer_expired(apic);
1679
1680		if (apic_lvtt_oneshot(apic))
1681			return;
1682
1683		advance_periodic_target_expiration(apic);
1684	}
1685
1686	hrtimer_start(&apic->lapic_timer.timer,
1687		apic->lapic_timer.target_expiration,
1688		HRTIMER_MODE_ABS);
1689}
1690
1691bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1692{
1693	if (!lapic_in_kernel(vcpu))
1694		return false;
1695
1696	return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1697}
1698EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use);
1699
1700static void cancel_hv_timer(struct kvm_lapic *apic)
1701{
1702	WARN_ON(preemptible());
1703	WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1704	kvm_x86_ops->cancel_hv_timer(apic->vcpu);
1705	apic->lapic_timer.hv_timer_in_use = false;
1706}
1707
1708static bool start_hv_timer(struct kvm_lapic *apic)
1709{
1710	struct kvm_timer *ktimer = &apic->lapic_timer;
1711	struct kvm_vcpu *vcpu = apic->vcpu;
1712	bool expired;
1713
1714	WARN_ON(preemptible());
1715	if (!kvm_x86_ops->set_hv_timer)
1716		return false;
1717
1718	if (!ktimer->tscdeadline)
1719		return false;
1720
1721	if (kvm_x86_ops->set_hv_timer(vcpu, ktimer->tscdeadline, &expired))
1722		return false;
1723
1724	ktimer->hv_timer_in_use = true;
1725	hrtimer_cancel(&ktimer->timer);
1726
1727	/*
1728	 * To simplify handling the periodic timer, leave the hv timer running
1729	 * even if the deadline timer has expired, i.e. rely on the resulting
1730	 * VM-Exit to recompute the periodic timer's target expiration.
1731	 */
1732	if (!apic_lvtt_period(apic)) {
1733		/*
1734		 * Cancel the hv timer if the sw timer fired while the hv timer
1735		 * was being programmed, or if the hv timer itself expired.
1736		 */
1737		if (atomic_read(&ktimer->pending)) {
1738			cancel_hv_timer(apic);
1739		} else if (expired) {
1740			apic_timer_expired(apic);
1741			cancel_hv_timer(apic);
1742		}
1743	}
1744
1745	trace_kvm_hv_timer_state(vcpu->vcpu_id, ktimer->hv_timer_in_use);
1746
1747	return true;
1748}
1749
1750static void start_sw_timer(struct kvm_lapic *apic)
1751{
1752	struct kvm_timer *ktimer = &apic->lapic_timer;
1753
1754	WARN_ON(preemptible());
1755	if (apic->lapic_timer.hv_timer_in_use)
1756		cancel_hv_timer(apic);
1757	if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
1758		return;
1759
1760	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1761		start_sw_period(apic);
1762	else if (apic_lvtt_tscdeadline(apic))
1763		start_sw_tscdeadline(apic);
1764	trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, false);
1765}
1766
1767static void restart_apic_timer(struct kvm_lapic *apic)
1768{
1769	preempt_disable();
1770
1771	if (!apic_lvtt_period(apic) && atomic_read(&apic->lapic_timer.pending))
1772		goto out;
1773
1774	if (!start_hv_timer(apic))
1775		start_sw_timer(apic);
1776out:
1777	preempt_enable();
1778}
1779
1780void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1781{
1782	struct kvm_lapic *apic = vcpu->arch.apic;
1783
1784	preempt_disable();
1785	/* If the preempt notifier has already run, it also called apic_timer_expired */
1786	if (!apic->lapic_timer.hv_timer_in_use)
1787		goto out;
1788	WARN_ON(swait_active(&vcpu->wq));
 
1789	cancel_hv_timer(apic);
1790	apic_timer_expired(apic);
1791
1792	if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1793		advance_periodic_target_expiration(apic);
1794		restart_apic_timer(apic);
1795	}
1796out:
1797	preempt_enable();
1798}
1799EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
1800
1801void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
1802{
1803	restart_apic_timer(vcpu->arch.apic);
1804}
1805EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer);
1806
1807void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
1808{
1809	struct kvm_lapic *apic = vcpu->arch.apic;
1810
1811	preempt_disable();
1812	/* Possibly the TSC deadline timer is not enabled yet */
1813	if (apic->lapic_timer.hv_timer_in_use)
1814		start_sw_timer(apic);
1815	preempt_enable();
1816}
1817EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer);
1818
1819void kvm_lapic_restart_hv_timer(struct kvm_vcpu *vcpu)
1820{
1821	struct kvm_lapic *apic = vcpu->arch.apic;
1822
1823	WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1824	restart_apic_timer(apic);
1825}
1826
1827static void start_apic_timer(struct kvm_lapic *apic)
1828{
1829	atomic_set(&apic->lapic_timer.pending, 0);
1830
1831	if ((apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1832	    && !set_target_expiration(apic))
1833		return;
1834
1835	restart_apic_timer(apic);
1836}
1837
 
 
 
 
 
1838static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1839{
1840	bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
1841
1842	if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1843		apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1844		if (lvt0_in_nmi_mode) {
1845			atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1846		} else
1847			atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1848	}
1849}
1850
1851int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1852{
1853	int ret = 0;
1854
1855	trace_kvm_apic_write(reg, val);
1856
1857	switch (reg) {
1858	case APIC_ID:		/* Local APIC ID */
1859		if (!apic_x2apic_mode(apic))
1860			kvm_apic_set_xapic_id(apic, val >> 24);
1861		else
 
1862			ret = 1;
 
1863		break;
1864
1865	case APIC_TASKPRI:
1866		report_tpr_access(apic, true);
1867		apic_set_tpr(apic, val & 0xff);
1868		break;
1869
1870	case APIC_EOI:
1871		apic_set_eoi(apic);
1872		break;
1873
1874	case APIC_LDR:
1875		if (!apic_x2apic_mode(apic))
1876			kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1877		else
1878			ret = 1;
1879		break;
1880
1881	case APIC_DFR:
1882		if (!apic_x2apic_mode(apic)) {
1883			kvm_lapic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1884			recalculate_apic_map(apic->vcpu->kvm);
1885		} else
1886			ret = 1;
1887		break;
1888
1889	case APIC_SPIV: {
1890		u32 mask = 0x3ff;
1891		if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1892			mask |= APIC_SPIV_DIRECTED_EOI;
1893		apic_set_spiv(apic, val & mask);
1894		if (!(val & APIC_SPIV_APIC_ENABLED)) {
1895			int i;
1896			u32 lvt_val;
1897
1898			for (i = 0; i < KVM_APIC_LVT_NUM; i++) {
1899				lvt_val = kvm_lapic_get_reg(apic,
1900						       APIC_LVTT + 0x10 * i);
1901				kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i,
1902					     lvt_val | APIC_LVT_MASKED);
1903			}
1904			apic_update_lvtt(apic);
1905			atomic_set(&apic->lapic_timer.pending, 0);
1906
1907		}
1908		break;
1909	}
1910	case APIC_ICR:
 
 
1911		/* No delay here, so we always clear the pending bit */
1912		val &= ~(1 << 12);
1913		apic_send_ipi(apic, val, kvm_lapic_get_reg(apic, APIC_ICR2));
1914		kvm_lapic_set_reg(apic, APIC_ICR, val);
1915		break;
1916
1917	case APIC_ICR2:
1918		if (!apic_x2apic_mode(apic))
1919			val &= 0xff000000;
1920		kvm_lapic_set_reg(apic, APIC_ICR2, val);
 
1921		break;
1922
1923	case APIC_LVT0:
1924		apic_manage_nmi_watchdog(apic, val);
1925		/* fall through */
1926	case APIC_LVTTHMR:
1927	case APIC_LVTPC:
1928	case APIC_LVT1:
1929	case APIC_LVTERR:
1930		/* TODO: Check vector */
 
 
 
 
 
1931		if (!kvm_apic_sw_enabled(apic))
1932			val |= APIC_LVT_MASKED;
1933
1934		val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1935		kvm_lapic_set_reg(apic, reg, val);
1936
1937		break;
 
1938
1939	case APIC_LVTT:
1940		if (!kvm_apic_sw_enabled(apic))
1941			val |= APIC_LVT_MASKED;
1942		val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1943		kvm_lapic_set_reg(apic, APIC_LVTT, val);
1944		apic_update_lvtt(apic);
1945		break;
1946
1947	case APIC_TMICT:
1948		if (apic_lvtt_tscdeadline(apic))
1949			break;
1950
1951		hrtimer_cancel(&apic->lapic_timer.timer);
1952		kvm_lapic_set_reg(apic, APIC_TMICT, val);
1953		start_apic_timer(apic);
1954		break;
1955
1956	case APIC_TDCR: {
1957		uint32_t old_divisor = apic->divide_count;
1958
1959		kvm_lapic_set_reg(apic, APIC_TDCR, val);
1960		update_divide_count(apic);
1961		if (apic->divide_count != old_divisor &&
1962				apic->lapic_timer.period) {
1963			hrtimer_cancel(&apic->lapic_timer.timer);
1964			update_target_expiration(apic, old_divisor);
1965			restart_apic_timer(apic);
1966		}
1967		break;
1968	}
1969	case APIC_ESR:
1970		if (apic_x2apic_mode(apic) && val != 0)
1971			ret = 1;
1972		break;
1973
1974	case APIC_SELF_IPI:
1975		if (apic_x2apic_mode(apic)) {
1976			kvm_lapic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1977		} else
1978			ret = 1;
1979		break;
1980	default:
1981		ret = 1;
1982		break;
1983	}
1984
 
 
 
 
 
 
 
1985	return ret;
1986}
1987EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
1988
1989static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1990			    gpa_t address, int len, const void *data)
1991{
1992	struct kvm_lapic *apic = to_lapic(this);
1993	unsigned int offset = address - apic->base_address;
1994	u32 val;
1995
1996	if (!apic_mmio_in_range(apic, address))
1997		return -EOPNOTSUPP;
1998
1999	if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
2000		if (!kvm_check_has_quirk(vcpu->kvm,
2001					 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
2002			return -EOPNOTSUPP;
2003
2004		return 0;
2005	}
2006
2007	/*
2008	 * APIC register must be aligned on 128-bits boundary.
2009	 * 32/64/128 bits registers must be accessed thru 32 bits.
2010	 * Refer SDM 8.4.1
2011	 */
2012	if (len != 4 || (offset & 0xf))
2013		return 0;
2014
2015	val = *(u32*)data;
2016
2017	kvm_lapic_reg_write(apic, offset & 0xff0, val);
2018
2019	return 0;
2020}
2021
2022void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
2023{
2024	kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
2025}
2026EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
2027
2028/* emulate APIC access in a trap manner */
2029void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
2030{
2031	u32 val = 0;
 
2032
2033	/* hw has done the conditional check and inst decode */
2034	offset &= 0xff0;
 
 
 
 
2035
2036	kvm_lapic_reg_read(vcpu->arch.apic, offset, 4, &val);
2037
2038	/* TODO: optimize to just emulate side effect w/o one more write */
2039	kvm_lapic_reg_write(vcpu->arch.apic, offset, val);
 
 
 
 
 
 
 
 
2040}
2041EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
2042
2043void kvm_free_lapic(struct kvm_vcpu *vcpu)
2044{
2045	struct kvm_lapic *apic = vcpu->arch.apic;
2046
2047	if (!vcpu->arch.apic)
2048		return;
2049
2050	hrtimer_cancel(&apic->lapic_timer.timer);
2051
2052	if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
2053		static_key_slow_dec_deferred(&apic_hw_disabled);
2054
2055	if (!apic->sw_enabled)
2056		static_key_slow_dec_deferred(&apic_sw_disabled);
2057
2058	if (apic->regs)
2059		free_page((unsigned long)apic->regs);
2060
2061	kfree(apic);
2062}
2063
2064/*
2065 *----------------------------------------------------------------------
2066 * LAPIC interface
2067 *----------------------------------------------------------------------
2068 */
2069u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
2070{
2071	struct kvm_lapic *apic = vcpu->arch.apic;
2072
2073	if (!lapic_in_kernel(vcpu) ||
2074		!apic_lvtt_tscdeadline(apic))
2075		return 0;
2076
2077	return apic->lapic_timer.tscdeadline;
2078}
2079
2080void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
2081{
2082	struct kvm_lapic *apic = vcpu->arch.apic;
2083
2084	if (!lapic_in_kernel(vcpu) || apic_lvtt_oneshot(apic) ||
2085			apic_lvtt_period(apic))
2086		return;
2087
2088	hrtimer_cancel(&apic->lapic_timer.timer);
2089	apic->lapic_timer.tscdeadline = data;
2090	start_apic_timer(apic);
2091}
2092
2093void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
2094{
2095	struct kvm_lapic *apic = vcpu->arch.apic;
2096
2097	apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
2098		     | (kvm_lapic_get_reg(apic, APIC_TASKPRI) & 4));
2099}
2100
2101u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
2102{
2103	u64 tpr;
2104
2105	tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
2106
2107	return (tpr & 0xf0) >> 4;
2108}
2109
2110void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
2111{
2112	u64 old_value = vcpu->arch.apic_base;
2113	struct kvm_lapic *apic = vcpu->arch.apic;
2114
2115	if (!apic)
2116		value |= MSR_IA32_APICBASE_BSP;
2117
2118	vcpu->arch.apic_base = value;
2119
2120	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
2121		kvm_update_cpuid(vcpu);
2122
2123	if (!apic)
2124		return;
2125
2126	/* update jump label if enable bit changes */
2127	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
2128		if (value & MSR_IA32_APICBASE_ENABLE) {
2129			kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2130			static_key_slow_dec_deferred(&apic_hw_disabled);
 
 
2131		} else {
2132			static_key_slow_inc(&apic_hw_disabled.key);
2133			recalculate_apic_map(vcpu->kvm);
2134		}
2135	}
2136
2137	if (((old_value ^ value) & X2APIC_ENABLE) && (value & X2APIC_ENABLE))
2138		kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
2139
2140	if ((old_value ^ value) & (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE))
2141		kvm_x86_ops->set_virtual_apic_mode(vcpu);
 
 
2142
2143	apic->base_address = apic->vcpu->arch.apic_base &
2144			     MSR_IA32_APICBASE_BASE;
2145
2146	if ((value & MSR_IA32_APICBASE_ENABLE) &&
2147	     apic->base_address != APIC_DEFAULT_PHYS_BASE)
2148		pr_warn_once("APIC base relocation is unsupported by KVM");
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2149}
2150
2151void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
2152{
2153	struct kvm_lapic *apic = vcpu->arch.apic;
 
2154	int i;
2155
 
 
 
 
 
 
 
2156	if (!apic)
2157		return;
2158
2159	/* Stop the timer in case it's a reset to an active apic */
2160	hrtimer_cancel(&apic->lapic_timer.timer);
2161
2162	if (!init_event) {
2163		kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
2164		                         MSR_IA32_APICBASE_ENABLE);
2165		kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2166	}
2167	kvm_apic_set_version(apic->vcpu);
2168
2169	for (i = 0; i < KVM_APIC_LVT_NUM; i++)
2170		kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
2171	apic_update_lvtt(apic);
2172	if (kvm_vcpu_is_reset_bsp(vcpu) &&
2173	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
2174		kvm_lapic_set_reg(apic, APIC_LVT0,
2175			     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
2176	apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2177
2178	kvm_lapic_set_reg(apic, APIC_DFR, 0xffffffffU);
2179	apic_set_spiv(apic, 0xff);
2180	kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
2181	if (!apic_x2apic_mode(apic))
2182		kvm_apic_set_ldr(apic, 0);
2183	kvm_lapic_set_reg(apic, APIC_ESR, 0);
2184	kvm_lapic_set_reg(apic, APIC_ICR, 0);
2185	kvm_lapic_set_reg(apic, APIC_ICR2, 0);
 
 
 
 
2186	kvm_lapic_set_reg(apic, APIC_TDCR, 0);
2187	kvm_lapic_set_reg(apic, APIC_TMICT, 0);
2188	for (i = 0; i < 8; i++) {
2189		kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
2190		kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
2191		kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
2192	}
2193	apic->irr_pending = vcpu->arch.apicv_active;
2194	apic->isr_count = vcpu->arch.apicv_active ? 1 : 0;
2195	apic->highest_isr_cache = -1;
2196	update_divide_count(apic);
2197	atomic_set(&apic->lapic_timer.pending, 0);
2198	if (kvm_vcpu_is_bsp(vcpu))
2199		kvm_lapic_set_base(vcpu,
2200				vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
2201	vcpu->arch.pv_eoi.msr_val = 0;
2202	apic_update_ppr(apic);
2203	if (vcpu->arch.apicv_active) {
2204		kvm_x86_ops->apicv_post_state_restore(vcpu);
2205		kvm_x86_ops->hwapic_irr_update(vcpu, -1);
2206		kvm_x86_ops->hwapic_isr_update(vcpu, -1);
2207	}
2208
2209	vcpu->arch.apic_arb_prio = 0;
2210	vcpu->arch.apic_attention = 0;
 
 
2211}
2212
2213/*
2214 *----------------------------------------------------------------------
2215 * timer interface
2216 *----------------------------------------------------------------------
2217 */
2218
2219static bool lapic_is_periodic(struct kvm_lapic *apic)
2220{
2221	return apic_lvtt_period(apic);
2222}
2223
2224int apic_has_pending_timer(struct kvm_vcpu *vcpu)
2225{
2226	struct kvm_lapic *apic = vcpu->arch.apic;
2227
2228	if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
2229		return atomic_read(&apic->lapic_timer.pending);
2230
2231	return 0;
2232}
2233
2234int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
2235{
2236	u32 reg = kvm_lapic_get_reg(apic, lvt_type);
2237	int vector, mode, trig_mode;
2238
2239	if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
2240		vector = reg & APIC_VECTOR_MASK;
2241		mode = reg & APIC_MODE_MASK;
2242		trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
2243		return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
2244					NULL);
2245	}
2246	return 0;
2247}
2248
2249void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
2250{
2251	struct kvm_lapic *apic = vcpu->arch.apic;
2252
2253	if (apic)
2254		kvm_apic_local_deliver(apic, APIC_LVT0);
2255}
2256
2257static const struct kvm_io_device_ops apic_mmio_ops = {
2258	.read     = apic_mmio_read,
2259	.write    = apic_mmio_write,
2260};
2261
2262static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
2263{
2264	struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
2265	struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
2266
2267	apic_timer_expired(apic);
2268
2269	if (lapic_is_periodic(apic)) {
2270		advance_periodic_target_expiration(apic);
2271		hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
2272		return HRTIMER_RESTART;
2273	} else
2274		return HRTIMER_NORESTART;
2275}
2276
2277int kvm_create_lapic(struct kvm_vcpu *vcpu, int timer_advance_ns)
2278{
2279	struct kvm_lapic *apic;
2280
2281	ASSERT(vcpu != NULL);
2282
2283	apic = kzalloc(sizeof(*apic), GFP_KERNEL_ACCOUNT);
2284	if (!apic)
2285		goto nomem;
2286
2287	vcpu->arch.apic = apic;
2288
2289	apic->regs = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
2290	if (!apic->regs) {
2291		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
2292		       vcpu->vcpu_id);
2293		goto nomem_free_apic;
2294	}
2295	apic->vcpu = vcpu;
2296
 
 
2297	hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2298		     HRTIMER_MODE_ABS_HARD);
2299	apic->lapic_timer.timer.function = apic_timer_fn;
2300	if (timer_advance_ns == -1) {
2301		apic->lapic_timer.timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
2302		lapic_timer_advance_dynamic = true;
2303	} else {
2304		apic->lapic_timer.timer_advance_ns = timer_advance_ns;
2305		lapic_timer_advance_dynamic = false;
2306	}
2307
2308	/*
2309	 * APIC is created enabled. This will prevent kvm_lapic_set_base from
2310	 * thinking that APIC state has changed.
2311	 */
2312	vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2313	static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2314	kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2315
2316	return 0;
2317nomem_free_apic:
2318	kfree(apic);
2319	vcpu->arch.apic = NULL;
2320nomem:
2321	return -ENOMEM;
2322}
2323
2324int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2325{
2326	struct kvm_lapic *apic = vcpu->arch.apic;
2327	u32 ppr;
2328
2329	if (!kvm_apic_hw_enabled(apic))
2330		return -1;
2331
2332	__apic_update_ppr(apic, &ppr);
2333	return apic_has_interrupt_for_ppr(apic, ppr);
2334}
 
2335
2336int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2337{
2338	u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2339	int r = 0;
2340
2341	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2342		r = 1;
2343	if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2344	    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2345		r = 1;
2346	return r;
2347}
2348
2349void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2350{
2351	struct kvm_lapic *apic = vcpu->arch.apic;
2352
2353	if (atomic_read(&apic->lapic_timer.pending) > 0) {
2354		kvm_apic_inject_pending_timer_irqs(apic);
2355		atomic_set(&apic->lapic_timer.pending, 0);
2356	}
2357}
2358
2359int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2360{
2361	int vector = kvm_apic_has_interrupt(vcpu);
2362	struct kvm_lapic *apic = vcpu->arch.apic;
2363	u32 ppr;
2364
2365	if (vector == -1)
2366		return -1;
2367
2368	/*
2369	 * We get here even with APIC virtualization enabled, if doing
2370	 * nested virtualization and L1 runs with the "acknowledge interrupt
2371	 * on exit" mode.  Then we cannot inject the interrupt via RVI,
2372	 * because the process would deliver it through the IDT.
2373	 */
2374
2375	apic_clear_irr(vector, apic);
2376	if (test_bit(vector, vcpu_to_synic(vcpu)->auto_eoi_bitmap)) {
2377		/*
2378		 * For auto-EOI interrupts, there might be another pending
2379		 * interrupt above PPR, so check whether to raise another
2380		 * KVM_REQ_EVENT.
2381		 */
2382		apic_update_ppr(apic);
2383	} else {
2384		/*
2385		 * For normal interrupts, PPR has been raised and there cannot
2386		 * be a higher-priority pending interrupt---except if there was
2387		 * a concurrent interrupt injection, but that would have
2388		 * triggered KVM_REQ_EVENT already.
2389		 */
2390		apic_set_isr(vector, apic);
2391		__apic_update_ppr(apic, &ppr);
2392	}
2393
2394	return vector;
2395}
2396
2397static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2398		struct kvm_lapic_state *s, bool set)
2399{
2400	if (apic_x2apic_mode(vcpu->arch.apic)) {
2401		u32 *id = (u32 *)(s->regs + APIC_ID);
2402		u32 *ldr = (u32 *)(s->regs + APIC_LDR);
 
2403
2404		if (vcpu->kvm->arch.x2apic_format) {
2405			if (*id != vcpu->vcpu_id)
2406				return -EINVAL;
2407		} else {
2408			if (set)
2409				*id >>= 24;
2410			else
2411				*id <<= 24;
2412		}
2413
2414		/* In x2APIC mode, the LDR is fixed and based on the id */
2415		if (set)
 
 
 
 
2416			*ldr = kvm_apic_calc_x2apic_ldr(*id);
 
 
 
 
 
 
 
 
2417	}
2418
2419	return 0;
2420}
2421
2422int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2423{
2424	memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
 
 
 
 
 
 
 
 
2425	return kvm_apic_state_fixup(vcpu, s, false);
2426}
2427
2428int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2429{
2430	struct kvm_lapic *apic = vcpu->arch.apic;
2431	int r;
2432
2433
2434	kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2435	/* set SPIV separately to get count of SW disabled APICs right */
2436	apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2437
2438	r = kvm_apic_state_fixup(vcpu, s, true);
2439	if (r)
 
2440		return r;
 
2441	memcpy(vcpu->arch.apic->regs, s->regs, sizeof(*s));
2442
2443	recalculate_apic_map(vcpu->kvm);
 
 
 
 
2444	kvm_apic_set_version(vcpu);
2445
2446	apic_update_ppr(apic);
2447	hrtimer_cancel(&apic->lapic_timer.timer);
 
2448	apic_update_lvtt(apic);
2449	apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2450	update_divide_count(apic);
2451	start_apic_timer(apic);
2452	apic->irr_pending = true;
2453	apic->isr_count = vcpu->arch.apicv_active ?
2454				1 : count_vectors(apic->regs + APIC_ISR);
2455	apic->highest_isr_cache = -1;
2456	if (vcpu->arch.apicv_active) {
2457		kvm_x86_ops->apicv_post_state_restore(vcpu);
2458		kvm_x86_ops->hwapic_irr_update(vcpu,
2459				apic_find_highest_irr(apic));
2460		kvm_x86_ops->hwapic_isr_update(vcpu,
2461				apic_find_highest_isr(apic));
2462	}
2463	kvm_make_request(KVM_REQ_EVENT, vcpu);
2464	if (ioapic_in_kernel(vcpu->kvm))
2465		kvm_rtc_eoi_tracking_restore_one(vcpu);
2466
2467	vcpu->arch.apic_arb_prio = 0;
2468
2469	return 0;
2470}
2471
2472void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2473{
2474	struct hrtimer *timer;
2475
2476	if (!lapic_in_kernel(vcpu) ||
2477		kvm_can_post_timer_interrupt(vcpu))
2478		return;
2479
2480	timer = &vcpu->arch.apic->lapic_timer.timer;
2481	if (hrtimer_cancel(timer))
2482		hrtimer_start_expires(timer, HRTIMER_MODE_ABS_HARD);
2483}
2484
2485/*
2486 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2487 *
2488 * Detect whether guest triggered PV EOI since the
2489 * last entry. If yes, set EOI on guests's behalf.
2490 * Clear PV EOI in guest memory in any case.
2491 */
2492static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2493					struct kvm_lapic *apic)
2494{
2495	bool pending;
2496	int vector;
2497	/*
2498	 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2499	 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2500	 *
2501	 * KVM_APIC_PV_EOI_PENDING is unset:
2502	 * 	-> host disabled PV EOI.
2503	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2504	 * 	-> host enabled PV EOI, guest did not execute EOI yet.
2505	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2506	 * 	-> host enabled PV EOI, guest executed EOI.
2507	 */
2508	BUG_ON(!pv_eoi_enabled(vcpu));
2509	pending = pv_eoi_get_pending(vcpu);
2510	/*
2511	 * Clear pending bit in any case: it will be set again on vmentry.
2512	 * While this might not be ideal from performance point of view,
2513	 * this makes sure pv eoi is only enabled when we know it's safe.
2514	 */
2515	pv_eoi_clr_pending(vcpu);
2516	if (pending)
2517		return;
2518	vector = apic_set_eoi(apic);
2519	trace_kvm_pv_eoi(apic, vector);
2520}
2521
2522void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2523{
2524	u32 data;
2525
2526	if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2527		apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2528
2529	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2530		return;
2531
2532	if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2533				  sizeof(u32)))
2534		return;
2535
2536	apic_set_tpr(vcpu->arch.apic, data & 0xff);
2537}
2538
2539/*
2540 * apic_sync_pv_eoi_to_guest - called before vmentry
2541 *
2542 * Detect whether it's safe to enable PV EOI and
2543 * if yes do so.
2544 */
2545static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2546					struct kvm_lapic *apic)
2547{
2548	if (!pv_eoi_enabled(vcpu) ||
2549	    /* IRR set or many bits in ISR: could be nested. */
2550	    apic->irr_pending ||
2551	    /* Cache not set: could be safe but we don't bother. */
2552	    apic->highest_isr_cache == -1 ||
2553	    /* Need EOI to update ioapic. */
2554	    kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2555		/*
2556		 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2557		 * so we need not do anything here.
2558		 */
2559		return;
2560	}
2561
2562	pv_eoi_set_pending(apic->vcpu);
2563}
2564
2565void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2566{
2567	u32 data, tpr;
2568	int max_irr, max_isr;
2569	struct kvm_lapic *apic = vcpu->arch.apic;
2570
2571	apic_sync_pv_eoi_to_guest(vcpu, apic);
2572
2573	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2574		return;
2575
2576	tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
2577	max_irr = apic_find_highest_irr(apic);
2578	if (max_irr < 0)
2579		max_irr = 0;
2580	max_isr = apic_find_highest_isr(apic);
2581	if (max_isr < 0)
2582		max_isr = 0;
2583	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2584
2585	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2586				sizeof(u32));
2587}
2588
2589int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2590{
2591	if (vapic_addr) {
2592		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2593					&vcpu->arch.apic->vapic_cache,
2594					vapic_addr, sizeof(u32)))
2595			return -EINVAL;
2596		__set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2597	} else {
2598		__clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2599	}
2600
2601	vcpu->arch.apic->vapic_addr = vapic_addr;
2602	return 0;
2603}
2604
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2605int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2606{
2607	struct kvm_lapic *apic = vcpu->arch.apic;
2608	u32 reg = (msr - APIC_BASE_MSR) << 4;
2609
2610	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2611		return 1;
2612
2613	if (reg == APIC_ICR2)
2614		return 1;
2615
2616	/* if this is ICR write vector before command */
2617	if (reg == APIC_ICR)
2618		kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2619	return kvm_lapic_reg_write(apic, reg, (u32)data);
2620}
2621
2622int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2623{
2624	struct kvm_lapic *apic = vcpu->arch.apic;
2625	u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2626
2627	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2628		return 1;
2629
2630	if (reg == APIC_DFR || reg == APIC_ICR2)
2631		return 1;
2632
2633	if (kvm_lapic_reg_read(apic, reg, 4, &low))
2634		return 1;
2635	if (reg == APIC_ICR)
2636		kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2637
2638	*data = (((u64)high) << 32) | low;
2639
2640	return 0;
2641}
2642
2643int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2644{
2645	struct kvm_lapic *apic = vcpu->arch.apic;
2646
2647	if (!lapic_in_kernel(vcpu))
2648		return 1;
2649
2650	/* if this is ICR write vector before command */
2651	if (reg == APIC_ICR)
2652		kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2653	return kvm_lapic_reg_write(apic, reg, (u32)data);
2654}
2655
2656int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2657{
2658	struct kvm_lapic *apic = vcpu->arch.apic;
2659	u32 low, high = 0;
2660
2661	if (!lapic_in_kernel(vcpu))
2662		return 1;
2663
2664	if (kvm_lapic_reg_read(apic, reg, 4, &low))
2665		return 1;
2666	if (reg == APIC_ICR)
2667		kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2668
2669	*data = (((u64)high) << 32) | low;
2670
2671	return 0;
2672}
2673
2674int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data, unsigned long len)
2675{
2676	u64 addr = data & ~KVM_MSR_ENABLED;
2677	struct gfn_to_hva_cache *ghc = &vcpu->arch.pv_eoi.data;
2678	unsigned long new_len;
 
2679
2680	if (!IS_ALIGNED(addr, 4))
2681		return 1;
2682
 
 
 
 
 
 
 
 
 
 
 
2683	vcpu->arch.pv_eoi.msr_val = data;
2684	if (!pv_eoi_enabled(vcpu))
2685		return 0;
2686
2687	if (addr == ghc->gpa && len <= ghc->len)
2688		new_len = ghc->len;
2689	else
2690		new_len = len;
2691
2692	return kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, addr, new_len);
2693}
2694
2695void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2696{
2697	struct kvm_lapic *apic = vcpu->arch.apic;
2698	u8 sipi_vector;
2699	unsigned long pe;
 
 
 
2700
2701	if (!lapic_in_kernel(vcpu) || !apic->pending_events)
2702		return;
 
 
 
 
 
 
 
 
2703
2704	/*
2705	 * INITs are latched while CPU is in specific states
2706	 * (SMM, VMX non-root mode, SVM with GIF=0).
2707	 * Because a CPU cannot be in these states immediately
2708	 * after it has processed an INIT signal (and thus in
2709	 * KVM_MP_STATE_INIT_RECEIVED state), just eat SIPIs
2710	 * and leave the INIT pending.
2711	 */
2712	if (is_smm(vcpu) || kvm_x86_ops->apic_init_signal_blocked(vcpu)) {
2713		WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2714		if (test_bit(KVM_APIC_SIPI, &apic->pending_events))
2715			clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2716		return;
2717	}
2718
2719	pe = xchg(&apic->pending_events, 0);
2720	if (test_bit(KVM_APIC_INIT, &pe)) {
2721		kvm_vcpu_reset(vcpu, true);
2722		if (kvm_vcpu_is_bsp(apic->vcpu))
2723			vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2724		else
2725			vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2726	}
2727	if (test_bit(KVM_APIC_SIPI, &pe) &&
2728	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2729		/* evaluate pending_events before reading the vector */
2730		smp_rmb();
2731		sipi_vector = apic->sipi_vector;
2732		kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2733		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
 
2734	}
2735}
2736
2737void kvm_lapic_init(void)
2738{
2739	/* do not patch jump label more than once per second */
2740	jump_label_rate_limit(&apic_hw_disabled, HZ);
2741	jump_label_rate_limit(&apic_sw_disabled, HZ);
2742}
2743
2744void kvm_lapic_exit(void)
2745{
2746	static_key_deferred_flush(&apic_hw_disabled);
 
2747	static_key_deferred_flush(&apic_sw_disabled);
 
2748}