arch/x86/kvm/lapic.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2
   3 /*
   4  * Local APIC virtualization
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  * Copyright (C) 2007 Novell
   8  * Copyright (C) 2007 Intel
   9  * Copyright 2009 Red Hat, Inc. and/or its affiliates.
  10  *
  11  * Authors:
  12  *   Dor Laor <dor.laor@qumranet.com>
  13  *   Gregory Haskins <ghaskins@novell.com>
  14  *   Yaozu (Eddie) Dong <eddie.dong@intel.com>
  15  *
  16  * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
  17  */
  18
  19 #include <linux/kvm_host.h>
  20 #include <linux/kvm.h>
  21 #include <linux/mm.h>
  22 #include <linux/highmem.h>
  23 #include <linux/smp.h>
  24 #include <linux/hrtimer.h>
  25 #include <linux/io.h>
  26 #include <linux/export.h>
  27 #include <linux/math64.h>
  28 #include <linux/slab.h>
  29 #include <asm/processor.h>
  30 #include <asm/msr.h>
  31 #include <asm/page.h>
  32 #include <asm/current.h>
  33 #include <asm/apicdef.h>
  34 #include <asm/delay.h>
  35 #include <linux/atomic.h>
  36 #include <linux/jump_label.h>
  37 #include "kvm_cache_regs.h"
  38 #include "irq.h"
  39 #include "ioapic.h"
  40 #include "trace.h"
  41 #include "x86.h"
  42 #include "cpuid.h"
  43 #include "hyperv.h"
  44
  45 #ifndef CONFIG_X86_64
  46 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
  47 #else
  48 #define mod_64(x, y) ((x) % (y))
  49 #endif
  50
  51 #define PRId64 "d"
  52 #define PRIx64 "llx"
  53 #define PRIu64 "u"
  54 #define PRIo64 "o"
  55
  56 /* 14 is the version for Xeon and Pentium 8.4.8*/
  57 #define APIC_VERSION                    (0x14UL | ((KVM_APIC_LVT_NUM - 1) << 16))
  58 #define LAPIC_MMIO_LENGTH               (1 << 12)
  59 /* followed define is not in apicdef.h */
  60 #define MAX_APIC_VECTOR                 256
  61 #define APIC_VECTORS_PER_REG            32
  62
  63 static bool lapic_timer_advance_dynamic __read_mostly;
  64 #define LAPIC_TIMER_ADVANCE_ADJUST_MIN  100     /* clock cycles */
  65 #define LAPIC_TIMER_ADVANCE_ADJUST_MAX  10000   /* clock cycles */
  66 #define LAPIC_TIMER_ADVANCE_NS_INIT     1000
  67 #define LAPIC_TIMER_ADVANCE_NS_MAX     5000
  68 /* step-by-step approximation to mitigate fluctuation */
  69 #define LAPIC_TIMER_ADVANCE_ADJUST_STEP 8
  70
  71 static inline int apic_test_vector(int vec, void *bitmap)
  72 {
  73         return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
  74 }
  75
  76 bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
  77 {
  78         struct kvm_lapic *apic = vcpu->arch.apic;
  79
  80         return apic_test_vector(vector, apic->regs + APIC_ISR) ||
  81                 apic_test_vector(vector, apic->regs + APIC_IRR);
  82 }
  83
  84 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
  85 {
  86         return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
  87 }
  88
  89 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
  90 {
  91         return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
  92 }
  93
  94 __read_mostly DEFINE_STATIC_KEY_DEFERRED_FALSE(apic_hw_disabled, HZ);
  95 __read_mostly DEFINE_STATIC_KEY_DEFERRED_FALSE(apic_sw_disabled, HZ);
  96
  97 static inline int apic_enabled(struct kvm_lapic *apic)
  98 {
  99         return kvm_apic_sw_enabled(apic) &&     kvm_apic_hw_enabled(apic);
 100 }
 101
 102 #define LVT_MASK        \
 103         (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
 104
 105 #define LINT_MASK       \
 106         (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
 107          APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
 108
 109 static inline u32 kvm_x2apic_id(struct kvm_lapic *apic)
 110 {
 111         return apic->vcpu->vcpu_id;
 112 }
 113
 114 static bool kvm_can_post_timer_interrupt(struct kvm_vcpu *vcpu)
 115 {
 116         return pi_inject_timer && kvm_vcpu_apicv_active(vcpu);
 117 }
 118
 119 bool kvm_can_use_hv_timer(struct kvm_vcpu *vcpu)
 120 {
 121         return kvm_x86_ops.set_hv_timer
 122                && !(kvm_mwait_in_guest(vcpu->kvm) ||
 123                     kvm_can_post_timer_interrupt(vcpu));
 124 }
 125 EXPORT_SYMBOL_GPL(kvm_can_use_hv_timer);
 126
 127 static bool kvm_use_posted_timer_interrupt(struct kvm_vcpu *vcpu)
 128 {
 129         return kvm_can_post_timer_interrupt(vcpu) && vcpu->mode == IN_GUEST_MODE;
 130 }
 131
 132 static inline bool kvm_apic_map_get_logical_dest(struct kvm_apic_map *map,
 133                 u32 dest_id, struct kvm_lapic ***cluster, u16 *mask) {
 134         switch (map->mode) {
 135         case KVM_APIC_MODE_X2APIC: {
 136                 u32 offset = (dest_id >> 16) * 16;
 137                 u32 max_apic_id = map->max_apic_id;
 138
 139                 if (offset <= max_apic_id) {
 140                         u8 cluster_size = min(max_apic_id - offset + 1, 16U);
 141
 142                         offset = array_index_nospec(offset, map->max_apic_id + 1);
 143                         *cluster = &map->phys_map[offset];
 144                         *mask = dest_id & (0xffff >> (16 - cluster_size));
 145                 } else {
 146                         *mask = 0;
 147                 }
 148
 149                 return true;
 150                 }
 151         case KVM_APIC_MODE_XAPIC_FLAT:
 152                 *cluster = map->xapic_flat_map;
 153                 *mask = dest_id & 0xff;
 154                 return true;
 155         case KVM_APIC_MODE_XAPIC_CLUSTER:
 156                 *cluster = map->xapic_cluster_map[(dest_id >> 4) & 0xf];
 157                 *mask = dest_id & 0xf;
 158                 return true;
 159         default:
 160                 /* Not optimized. */
 161                 return false;
 162         }
 163 }
 164
 165 static void kvm_apic_map_free(struct rcu_head *rcu)
 166 {
 167         struct kvm_apic_map *map = container_of(rcu, struct kvm_apic_map, rcu);
 168
 169         kvfree(map);
 170 }
 171
 172 /*
 173  * CLEAN -> DIRTY and UPDATE_IN_PROGRESS -> DIRTY changes happen without a lock.
 174  *
 175  * DIRTY -> UPDATE_IN_PROGRESS and UPDATE_IN_PROGRESS -> CLEAN happen with
 176  * apic_map_lock_held.
 177  */
 178 enum {
 179         CLEAN,
 180         UPDATE_IN_PROGRESS,
 181         DIRTY
 182 };
 183
 184 void kvm_recalculate_apic_map(struct kvm *kvm)
 185 {
 186         struct kvm_apic_map *new, *old = NULL;
 187         struct kvm_vcpu *vcpu;
 188         int i;
 189         u32 max_id = 255; /* enough space for any xAPIC ID */
 190
 191         /* Read kvm->arch.apic_map_dirty before kvm->arch.apic_map.  */
 192         if (atomic_read_acquire(&kvm->arch.apic_map_dirty) == CLEAN)
 193                 return;
 194
 195         mutex_lock(&kvm->arch.apic_map_lock);
 196         /*
 197          * Read kvm->arch.apic_map_dirty before kvm->arch.apic_map
 198          * (if clean) or the APIC registers (if dirty).
 199          */
 200         if (atomic_cmpxchg_acquire(&kvm->arch.apic_map_dirty,
 201                                    DIRTY, UPDATE_IN_PROGRESS) == CLEAN) {
 202                 /* Someone else has updated the map. */
 203                 mutex_unlock(&kvm->arch.apic_map_lock);
 204                 return;
 205         }
 206
 207         kvm_for_each_vcpu(i, vcpu, kvm)
 208                 if (kvm_apic_present(vcpu))
 209                         max_id = max(max_id, kvm_x2apic_id(vcpu->arch.apic));
 210
 211         new = kvzalloc(sizeof(struct kvm_apic_map) +
 212                            sizeof(struct kvm_lapic *) * ((u64)max_id + 1),
 213                            GFP_KERNEL_ACCOUNT);
 214
 215         if (!new)
 216                 goto out;
 217
 218         new->max_apic_id = max_id;
 219
 220         kvm_for_each_vcpu(i, vcpu, kvm) {
 221                 struct kvm_lapic *apic = vcpu->arch.apic;
 222                 struct kvm_lapic **cluster;
 223                 u16 mask;
 224                 u32 ldr;
 225                 u8 xapic_id;
 226                 u32 x2apic_id;
 227
 228                 if (!kvm_apic_present(vcpu))
 229                         continue;
 230
 231                 xapic_id = kvm_xapic_id(apic);
 232                 x2apic_id = kvm_x2apic_id(apic);
 233
 234                 /* Hotplug hack: see kvm_apic_match_physical_addr(), ... */
 235                 if ((apic_x2apic_mode(apic) || x2apic_id > 0xff) &&
 236                                 x2apic_id <= new->max_apic_id)
 237                         new->phys_map[x2apic_id] = apic;
 238                 /*
 239                  * ... xAPIC ID of VCPUs with APIC ID > 0xff will wrap-around,
 240                  * prevent them from masking VCPUs with APIC ID <= 0xff.
 241                  */
 242                 if (!apic_x2apic_mode(apic) && !new->phys_map[xapic_id])
 243                         new->phys_map[xapic_id] = apic;
 244
 245                 if (!kvm_apic_sw_enabled(apic))
 246                         continue;
 247
 248                 ldr = kvm_lapic_get_reg(apic, APIC_LDR);
 249
 250                 if (apic_x2apic_mode(apic)) {
 251                         new->mode |= KVM_APIC_MODE_X2APIC;
 252                 } else if (ldr) {
 253                         ldr = GET_APIC_LOGICAL_ID(ldr);
 254                         if (kvm_lapic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
 255                                 new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
 256                         else
 257                                 new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
 258                 }
 259
 260                 if (!kvm_apic_map_get_logical_dest(new, ldr, &cluster, &mask))
 261                         continue;
 262
 263                 if (mask)
 264                         cluster[ffs(mask) - 1] = apic;
 265         }
 266 out:
 267         old = rcu_dereference_protected(kvm->arch.apic_map,
 268                         lockdep_is_held(&kvm->arch.apic_map_lock));
 269         rcu_assign_pointer(kvm->arch.apic_map, new);
 270         /*
 271          * Write kvm->arch.apic_map before clearing apic->apic_map_dirty.
 272          * If another update has come in, leave it DIRTY.
 273          */
 274         atomic_cmpxchg_release(&kvm->arch.apic_map_dirty,
 275                                UPDATE_IN_PROGRESS, CLEAN);
 276         mutex_unlock(&kvm->arch.apic_map_lock);
 277
 278         if (old)
 279                 call_rcu(&old->rcu, kvm_apic_map_free);
 280
 281         kvm_make_scan_ioapic_request(kvm);
 282 }
 283
 284 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
 285 {
 286         bool enabled = val & APIC_SPIV_APIC_ENABLED;
 287
 288         kvm_lapic_set_reg(apic, APIC_SPIV, val);
 289
 290         if (enabled != apic->sw_enabled) {
 291                 apic->sw_enabled = enabled;
 292                 if (enabled)
 293                         static_branch_slow_dec_deferred(&apic_sw_disabled);
 294                 else
 295                         static_branch_inc(&apic_sw_disabled.key);
 296
 297                 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
 298         }
 299
 300         /* Check if there are APF page ready requests pending */
 301         if (enabled)
 302                 kvm_make_request(KVM_REQ_APF_READY, apic->vcpu);
 303 }
 304
 305 static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
 306 {
 307         kvm_lapic_set_reg(apic, APIC_ID, id << 24);
 308         atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
 309 }
 310
 311 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
 312 {
 313         kvm_lapic_set_reg(apic, APIC_LDR, id);
 314         atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
 315 }
 316
 317 static inline void kvm_apic_set_dfr(struct kvm_lapic *apic, u32 val)
 318 {
 319         kvm_lapic_set_reg(apic, APIC_DFR, val);
 320         atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
 321 }
 322
 323 static inline u32 kvm_apic_calc_x2apic_ldr(u32 id)
 324 {
 325         return ((id >> 4) << 16) | (1 << (id & 0xf));
 326 }
 327
 328 static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
 329 {
 330         u32 ldr = kvm_apic_calc_x2apic_ldr(id);
 331
 332         WARN_ON_ONCE(id != apic->vcpu->vcpu_id);
 333
 334         kvm_lapic_set_reg(apic, APIC_ID, id);
 335         kvm_lapic_set_reg(apic, APIC_LDR, ldr);
 336         atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
 337 }
 338
 339 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
 340 {
 341         return !(kvm_lapic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
 342 }
 343
 344 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
 345 {
 346         return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
 347 }
 348
 349 static inline int apic_lvtt_period(struct kvm_lapic *apic)
 350 {
 351         return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
 352 }
 353
 354 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
 355 {
 356         return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
 357 }
 358
 359 static inline int apic_lvt_nmi_mode(u32 lvt_val)
 360 {
 361         return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
 362 }
 363
 364 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
 365 {
 366         struct kvm_lapic *apic = vcpu->arch.apic;
 367         u32 v = APIC_VERSION;
 368
 369         if (!lapic_in_kernel(vcpu))
 370                 return;
 371
 372         /*
 373          * KVM emulates 82093AA datasheet (with in-kernel IOAPIC implementation)
 374          * which doesn't have EOI register; Some buggy OSes (e.g. Windows with
 375          * Hyper-V role) disable EOI broadcast in lapic not checking for IOAPIC
 376          * version first and level-triggered interrupts never get EOIed in
 377          * IOAPIC.
 378          */
 379         if (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) &&
 380             !ioapic_in_kernel(vcpu->kvm))
 381                 v |= APIC_LVR_DIRECTED_EOI;
 382         kvm_lapic_set_reg(apic, APIC_LVR, v);
 383 }
 384
 385 static const unsigned int apic_lvt_mask[KVM_APIC_LVT_NUM] = {
 386         LVT_MASK ,      /* part LVTT mask, timer mode mask added at runtime */
 387         LVT_MASK | APIC_MODE_MASK,      /* LVTTHMR */
 388         LVT_MASK | APIC_MODE_MASK,      /* LVTPC */
 389         LINT_MASK, LINT_MASK,   /* LVT0-1 */
 390         LVT_MASK                /* LVTERR */
 391 };
 392
 393 static int find_highest_vector(void *bitmap)
 394 {
 395         int vec;
 396         u32 *reg;
 397
 398         for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
 399              vec >= 0; vec -= APIC_VECTORS_PER_REG) {
 400                 reg = bitmap + REG_POS(vec);
 401                 if (*reg)
 402                         return __fls(*reg) + vec;
 403         }
 404
 405         return -1;
 406 }
 407
 408 static u8 count_vectors(void *bitmap)
 409 {
 410         int vec;
 411         u32 *reg;
 412         u8 count = 0;
 413
 414         for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
 415                 reg = bitmap + REG_POS(vec);
 416                 count += hweight32(*reg);
 417         }
 418
 419         return count;
 420 }
 421
 422 bool __kvm_apic_update_irr(u32 *pir, void *regs, int *max_irr)
 423 {
 424         u32 i, vec;
 425         u32 pir_val, irr_val, prev_irr_val;
 426         int max_updated_irr;
 427
 428         max_updated_irr = -1;
 429         *max_irr = -1;
 430
 431         for (i = vec = 0; i <= 7; i++, vec += 32) {
 432                 pir_val = READ_ONCE(pir[i]);
 433                 irr_val = *((u32 *)(regs + APIC_IRR + i * 0x10));
 434                 if (pir_val) {
 435                         prev_irr_val = irr_val;
 436                         irr_val |= xchg(&pir[i], 0);
 437                         *((u32 *)(regs + APIC_IRR + i * 0x10)) = irr_val;
 438                         if (prev_irr_val != irr_val) {
 439                                 max_updated_irr =
 440                                         __fls(irr_val ^ prev_irr_val) + vec;
 441                         }
 442                 }
 443                 if (irr_val)
 444                         *max_irr = __fls(irr_val) + vec;
 445         }
 446
 447         return ((max_updated_irr != -1) &&
 448                 (max_updated_irr == *max_irr));
 449 }
 450 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
 451
 452 bool kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir, int *max_irr)
 453 {
 454         struct kvm_lapic *apic = vcpu->arch.apic;
 455
 456         return __kvm_apic_update_irr(pir, apic->regs, max_irr);
 457 }
 458 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
 459
 460 static inline int apic_search_irr(struct kvm_lapic *apic)
 461 {
 462         return find_highest_vector(apic->regs + APIC_IRR);
 463 }
 464
 465 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
 466 {
 467         int result;
 468
 469         /*
 470          * Note that irr_pending is just a hint. It will be always
 471          * true with virtual interrupt delivery enabled.
 472          */
 473         if (!apic->irr_pending)
 474                 return -1;
 475
 476         result = apic_search_irr(apic);
 477         ASSERT(result == -1 || result >= 16);
 478
 479         return result;
 480 }
 481
 482 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
 483 {
 484         struct kvm_vcpu *vcpu;
 485
 486         vcpu = apic->vcpu;
 487
 488         if (unlikely(vcpu->arch.apicv_active)) {
 489                 /* need to update RVI */
 490                 kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
 491                 static_call(kvm_x86_hwapic_irr_update)(vcpu,
 492                                 apic_find_highest_irr(apic));
 493         } else {
 494                 apic->irr_pending = false;
 495                 kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
 496                 if (apic_search_irr(apic) != -1)
 497                         apic->irr_pending = true;
 498         }
 499 }
 500
 501 void kvm_apic_clear_irr(struct kvm_vcpu *vcpu, int vec)
 502 {
 503         apic_clear_irr(vec, vcpu->arch.apic);
 504 }
 505 EXPORT_SYMBOL_GPL(kvm_apic_clear_irr);
 506
 507 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
 508 {
 509         struct kvm_vcpu *vcpu;
 510
 511         if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
 512                 return;
 513
 514         vcpu = apic->vcpu;
 515
 516         /*
 517          * With APIC virtualization enabled, all caching is disabled
 518          * because the processor can modify ISR under the hood.  Instead
 519          * just set SVI.
 520          */
 521         if (unlikely(vcpu->arch.apicv_active))
 522                 static_call(kvm_x86_hwapic_isr_update)(vcpu, vec);
 523         else {
 524                 ++apic->isr_count;
 525                 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
 526                 /*
 527                  * ISR (in service register) bit is set when injecting an interrupt.
 528                  * The highest vector is injected. Thus the latest bit set matches
 529                  * the highest bit in ISR.
 530                  */
 531                 apic->highest_isr_cache = vec;
 532         }
 533 }
 534
 535 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
 536 {
 537         int result;
 538
 539         /*
 540          * Note that isr_count is always 1, and highest_isr_cache
 541          * is always -1, with APIC virtualization enabled.
 542          */
 543         if (!apic->isr_count)
 544                 return -1;
 545         if (likely(apic->highest_isr_cache != -1))
 546                 return apic->highest_isr_cache;
 547
 548         result = find_highest_vector(apic->regs + APIC_ISR);
 549         ASSERT(result == -1 || result >= 16);
 550
 551         return result;
 552 }
 553
 554 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
 555 {
 556         struct kvm_vcpu *vcpu;
 557         if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
 558                 return;
 559
 560         vcpu = apic->vcpu;
 561
 562         /*
 563          * We do get here for APIC virtualization enabled if the guest
 564          * uses the Hyper-V APIC enlightenment.  In this case we may need
 565          * to trigger a new interrupt delivery by writing the SVI field;
 566          * on the other hand isr_count and highest_isr_cache are unused
 567          * and must be left alone.
 568          */
 569         if (unlikely(vcpu->arch.apicv_active))
 570                 static_call(kvm_x86_hwapic_isr_update)(vcpu,
 571                                                 apic_find_highest_isr(apic));
 572         else {
 573                 --apic->isr_count;
 574                 BUG_ON(apic->isr_count < 0);
 575                 apic->highest_isr_cache = -1;
 576         }
 577 }
 578
 579 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
 580 {
 581         /* This may race with setting of irr in __apic_accept_irq() and
 582          * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
 583          * will cause vmexit immediately and the value will be recalculated
 584          * on the next vmentry.
 585          */
 586         return apic_find_highest_irr(vcpu->arch.apic);
 587 }
 588 EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
 589
 590 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
 591                              int vector, int level, int trig_mode,
 592                              struct dest_map *dest_map);
 593
 594 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
 595                      struct dest_map *dest_map)
 596 {
 597         struct kvm_lapic *apic = vcpu->arch.apic;
 598
 599         return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
 600                         irq->level, irq->trig_mode, dest_map);
 601 }
 602
 603 static int __pv_send_ipi(unsigned long *ipi_bitmap, struct kvm_apic_map *map,
 604                          struct kvm_lapic_irq *irq, u32 min)
 605 {
 606         int i, count = 0;
 607         struct kvm_vcpu *vcpu;
 608
 609         if (min > map->max_apic_id)
 610                 return 0;
 611
 612         for_each_set_bit(i, ipi_bitmap,
 613                 min((u32)BITS_PER_LONG, (map->max_apic_id - min + 1))) {
 614                 if (map->phys_map[min + i]) {
 615                         vcpu = map->phys_map[min + i]->vcpu;
 616                         count += kvm_apic_set_irq(vcpu, irq, NULL);
 617                 }
 618         }
 619
 620         return count;
 621 }
 622
 623 int kvm_pv_send_ipi(struct kvm *kvm, unsigned long ipi_bitmap_low,
 624                     unsigned long ipi_bitmap_high, u32 min,
 625                     unsigned long icr, int op_64_bit)
 626 {
 627         struct kvm_apic_map *map;
 628         struct kvm_lapic_irq irq = {0};
 629         int cluster_size = op_64_bit ? 64 : 32;
 630         int count;
 631
 632         if (icr & (APIC_DEST_MASK | APIC_SHORT_MASK))
 633                 return -KVM_EINVAL;
 634
 635         irq.vector = icr & APIC_VECTOR_MASK;
 636         irq.delivery_mode = icr & APIC_MODE_MASK;
 637         irq.level = (icr & APIC_INT_ASSERT) != 0;
 638         irq.trig_mode = icr & APIC_INT_LEVELTRIG;
 639
 640         rcu_read_lock();
 641         map = rcu_dereference(kvm->arch.apic_map);
 642
 643         count = -EOPNOTSUPP;
 644         if (likely(map)) {
 645                 count = __pv_send_ipi(&ipi_bitmap_low, map, &irq, min);
 646                 min += cluster_size;
 647                 count += __pv_send_ipi(&ipi_bitmap_high, map, &irq, min);
 648         }
 649
 650         rcu_read_unlock();
 651         return count;
 652 }
 653
 654 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
 655 {
 656
 657         return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
 658                                       sizeof(val));
 659 }
 660
 661 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
 662 {
 663
 664         return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
 665                                       sizeof(*val));
 666 }
 667
 668 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
 669 {
 670         return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
 671 }
 672
 673 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
 674 {
 675         u8 val;
 676         if (pv_eoi_get_user(vcpu, &val) < 0) {
 677                 printk(KERN_WARNING "Can't read EOI MSR value: 0x%llx\n",
 678                            (unsigned long long)vcpu->arch.pv_eoi.msr_val);
 679                 return false;
 680         }
 681         return val & KVM_PV_EOI_ENABLED;
 682 }
 683
 684 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
 685 {
 686         if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
 687                 printk(KERN_WARNING "Can't set EOI MSR value: 0x%llx\n",
 688                            (unsigned long long)vcpu->arch.pv_eoi.msr_val);
 689                 return;
 690         }
 691         __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
 692 }
 693
 694 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
 695 {
 696         if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
 697                 printk(KERN_WARNING "Can't clear EOI MSR value: 0x%llx\n",
 698                            (unsigned long long)vcpu->arch.pv_eoi.msr_val);
 699                 return;
 700         }
 701         __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
 702 }
 703
 704 static int apic_has_interrupt_for_ppr(struct kvm_lapic *apic, u32 ppr)
 705 {
 706         int highest_irr;
 707         if (apic->vcpu->arch.apicv_active)
 708                 highest_irr = static_call(kvm_x86_sync_pir_to_irr)(apic->vcpu);
 709         else
 710                 highest_irr = apic_find_highest_irr(apic);
 711         if (highest_irr == -1 || (highest_irr & 0xF0) <= ppr)
 712                 return -1;
 713         return highest_irr;
 714 }
 715
 716 static bool __apic_update_ppr(struct kvm_lapic *apic, u32 *new_ppr)
 717 {
 718         u32 tpr, isrv, ppr, old_ppr;
 719         int isr;
 720
 721         old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
 722         tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
 723         isr = apic_find_highest_isr(apic);
 724         isrv = (isr != -1) ? isr : 0;
 725
 726         if ((tpr & 0xf0) >= (isrv & 0xf0))
 727                 ppr = tpr & 0xff;
 728         else
 729                 ppr = isrv & 0xf0;
 730
 731         *new_ppr = ppr;
 732         if (old_ppr != ppr)
 733                 kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
 734
 735         return ppr < old_ppr;
 736 }
 737
 738 static void apic_update_ppr(struct kvm_lapic *apic)
 739 {
 740         u32 ppr;
 741
 742         if (__apic_update_ppr(apic, &ppr) &&
 743             apic_has_interrupt_for_ppr(apic, ppr) != -1)
 744                 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
 745 }
 746
 747 void kvm_apic_update_ppr(struct kvm_vcpu *vcpu)
 748 {
 749         apic_update_ppr(vcpu->arch.apic);
 750 }
 751 EXPORT_SYMBOL_GPL(kvm_apic_update_ppr);
 752
 753 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
 754 {
 755         kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
 756         apic_update_ppr(apic);
 757 }
 758
 759 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
 760 {
 761         return mda == (apic_x2apic_mode(apic) ?
 762                         X2APIC_BROADCAST : APIC_BROADCAST);
 763 }
 764
 765 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
 766 {
 767         if (kvm_apic_broadcast(apic, mda))
 768                 return true;
 769
 770         if (apic_x2apic_mode(apic))
 771                 return mda == kvm_x2apic_id(apic);
 772
 773         /*
 774          * Hotplug hack: Make LAPIC in xAPIC mode also accept interrupts as if
 775          * it were in x2APIC mode.  Hotplugged VCPUs start in xAPIC mode and
 776          * this allows unique addressing of VCPUs with APIC ID over 0xff.
 777          * The 0xff condition is needed because writeable xAPIC ID.
 778          */
 779         if (kvm_x2apic_id(apic) > 0xff && mda == kvm_x2apic_id(apic))
 780                 return true;
 781
 782         return mda == kvm_xapic_id(apic);
 783 }
 784
 785 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
 786 {
 787         u32 logical_id;
 788
 789         if (kvm_apic_broadcast(apic, mda))
 790                 return true;
 791
 792         logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
 793
 794         if (apic_x2apic_mode(apic))
 795                 return ((logical_id >> 16) == (mda >> 16))
 796                        && (logical_id & mda & 0xffff) != 0;
 797
 798         logical_id = GET_APIC_LOGICAL_ID(logical_id);
 799
 800         switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
 801         case APIC_DFR_FLAT:
 802                 return (logical_id & mda) != 0;
 803         case APIC_DFR_CLUSTER:
 804                 return ((logical_id >> 4) == (mda >> 4))
 805                        && (logical_id & mda & 0xf) != 0;
 806         default:
 807                 return false;
 808         }
 809 }
 810
 811 /* The KVM local APIC implementation has two quirks:
 812  *
 813  *  - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
 814  *    in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
 815  *    KVM doesn't do that aliasing.
 816  *
 817  *  - in-kernel IOAPIC messages have to be delivered directly to
 818  *    x2APIC, because the kernel does not support interrupt remapping.
 819  *    In order to support broadcast without interrupt remapping, x2APIC
 820  *    rewrites the destination of non-IPI messages from APIC_BROADCAST
 821  *    to X2APIC_BROADCAST.
 822  *
 823  * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API.  This is
 824  * important when userspace wants to use x2APIC-format MSIs, because
 825  * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
 826  */
 827 static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
 828                 struct kvm_lapic *source, struct kvm_lapic *target)
 829 {
 830         bool ipi = source != NULL;
 831
 832         if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
 833             !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
 834                 return X2APIC_BROADCAST;
 835
 836         return dest_id;
 837 }
 838
 839 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
 840                            int shorthand, unsigned int dest, int dest_mode)
 841 {
 842         struct kvm_lapic *target = vcpu->arch.apic;
 843         u32 mda = kvm_apic_mda(vcpu, dest, source, target);
 844
 845         ASSERT(target);
 846         switch (shorthand) {
 847         case APIC_DEST_NOSHORT:
 848                 if (dest_mode == APIC_DEST_PHYSICAL)
 849                         return kvm_apic_match_physical_addr(target, mda);
 850                 else
 851                         return kvm_apic_match_logical_addr(target, mda);
 852         case APIC_DEST_SELF:
 853                 return target == source;
 854         case APIC_DEST_ALLINC:
 855                 return true;
 856         case APIC_DEST_ALLBUT:
 857                 return target != source;
 858         default:
 859                 return false;
 860         }
 861 }
 862 EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
 863
 864 int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
 865                        const unsigned long *bitmap, u32 bitmap_size)
 866 {
 867         u32 mod;
 868         int i, idx = -1;
 869
 870         mod = vector % dest_vcpus;
 871
 872         for (i = 0; i <= mod; i++) {
 873                 idx = find_next_bit(bitmap, bitmap_size, idx + 1);
 874                 BUG_ON(idx == bitmap_size);
 875         }
 876
 877         return idx;
 878 }
 879
 880 static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
 881 {
 882         if (!kvm->arch.disabled_lapic_found) {
 883                 kvm->arch.disabled_lapic_found = true;
 884                 printk(KERN_INFO
 885                        "Disabled LAPIC found during irq injection\n");
 886         }
 887 }
 888
 889 static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
 890                 struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
 891 {
 892         if (kvm->arch.x2apic_broadcast_quirk_disabled) {
 893                 if ((irq->dest_id == APIC_BROADCAST &&
 894                                 map->mode != KVM_APIC_MODE_X2APIC))
 895                         return true;
 896                 if (irq->dest_id == X2APIC_BROADCAST)
 897                         return true;
 898         } else {
 899                 bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
 900                 if (irq->dest_id == (x2apic_ipi ?
 901                                      X2APIC_BROADCAST : APIC_BROADCAST))
 902                         return true;
 903         }
 904
 905         return false;
 906 }
 907
 908 /* Return true if the interrupt can be handled by using *bitmap as index mask
 909  * for valid destinations in *dst array.
 910  * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
 911  * Note: we may have zero kvm_lapic destinations when we return true, which
 912  * means that the interrupt should be dropped.  In this case, *bitmap would be
 913  * zero and *dst undefined.
 914  */
 915 static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
 916                 struct kvm_lapic **src, struct kvm_lapic_irq *irq,
 917                 struct kvm_apic_map *map, struct kvm_lapic ***dst,
 918                 unsigned long *bitmap)
 919 {
 920         int i, lowest;
 921
 922         if (irq->shorthand == APIC_DEST_SELF && src) {
 923                 *dst = src;
 924                 *bitmap = 1;
 925                 return true;
 926         } else if (irq->shorthand)
 927                 return false;
 928
 929         if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
 930                 return false;
 931
 932         if (irq->dest_mode == APIC_DEST_PHYSICAL) {
 933                 if (irq->dest_id > map->max_apic_id) {
 934                         *bitmap = 0;
 935                 } else {
 936                         u32 dest_id = array_index_nospec(irq->dest_id, map->max_apic_id + 1);
 937                         *dst = &map->phys_map[dest_id];
 938                         *bitmap = 1;
 939                 }
 940                 return true;
 941         }
 942
 943         *bitmap = 0;
 944         if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
 945                                 (u16 *)bitmap))
 946                 return false;
 947
 948         if (!kvm_lowest_prio_delivery(irq))
 949                 return true;
 950
 951         if (!kvm_vector_hashing_enabled()) {
 952                 lowest = -1;
 953                 for_each_set_bit(i, bitmap, 16) {
 954                         if (!(*dst)[i])
 955                                 continue;
 956                         if (lowest < 0)
 957                                 lowest = i;
 958                         else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
 959                                                 (*dst)[lowest]->vcpu) < 0)
 960                                 lowest = i;
 961                 }
 962         } else {
 963                 if (!*bitmap)
 964                         return true;
 965
 966                 lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
 967                                 bitmap, 16);
 968
 969                 if (!(*dst)[lowest]) {
 970                         kvm_apic_disabled_lapic_found(kvm);
 971                         *bitmap = 0;
 972                         return true;
 973                 }
 974         }
 975
 976         *bitmap = (lowest >= 0) ? 1 << lowest : 0;
 977
 978         return true;
 979 }
 980
 981 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
 982                 struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
 983 {
 984         struct kvm_apic_map *map;
 985         unsigned long bitmap;
 986         struct kvm_lapic **dst = NULL;
 987         int i;
 988         bool ret;
 989
 990         *r = -1;
 991
 992         if (irq->shorthand == APIC_DEST_SELF) {
 993                 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
 994                 return true;
 995         }
 996
 997         rcu_read_lock();
 998         map = rcu_dereference(kvm->arch.apic_map);
 999
1000         ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
1001         if (ret) {
1002                 *r = 0;
1003                 for_each_set_bit(i, &bitmap, 16) {
1004                         if (!dst[i])
1005                                 continue;
1006                         *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
1007                 }
1008         }
1009
1010         rcu_read_unlock();
1011         return ret;
1012 }
1013
1014 /*
1015  * This routine tries to handle interrupts in posted mode, here is how
1016  * it deals with different cases:
1017  * - For single-destination interrupts, handle it in posted mode
1018  * - Else if vector hashing is enabled and it is a lowest-priority
1019  *   interrupt, handle it in posted mode and use the following mechanism
1020  *   to find the destination vCPU.
1021  *      1. For lowest-priority interrupts, store all the possible
1022  *         destination vCPUs in an array.
1023  *      2. Use "guest vector % max number of destination vCPUs" to find
1024  *         the right destination vCPU in the array for the lowest-priority
1025  *         interrupt.
1026  * - Otherwise, use remapped mode to inject the interrupt.
1027  */
1028 bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
1029                         struct kvm_vcpu **dest_vcpu)
1030 {
1031         struct kvm_apic_map *map;
1032         unsigned long bitmap;
1033         struct kvm_lapic **dst = NULL;
1034         bool ret = false;
1035
1036         if (irq->shorthand)
1037                 return false;
1038
1039         rcu_read_lock();
1040         map = rcu_dereference(kvm->arch.apic_map);
1041
1042         if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
1043                         hweight16(bitmap) == 1) {
1044                 unsigned long i = find_first_bit(&bitmap, 16);
1045
1046                 if (dst[i]) {
1047                         *dest_vcpu = dst[i]->vcpu;
1048                         ret = true;
1049                 }
1050         }
1051
1052         rcu_read_unlock();
1053         return ret;
1054 }
1055
1056 /*
1057  * Add a pending IRQ into lapic.
1058  * Return 1 if successfully added and 0 if discarded.
1059  */
1060 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
1061                              int vector, int level, int trig_mode,
1062                              struct dest_map *dest_map)
1063 {
1064         int result = 0;
1065         struct kvm_vcpu *vcpu = apic->vcpu;
1066
1067         trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
1068                                   trig_mode, vector);
1069         switch (delivery_mode) {
1070         case APIC_DM_LOWEST:
1071                 vcpu->arch.apic_arb_prio++;
1072                 fallthrough;
1073         case APIC_DM_FIXED:
1074                 if (unlikely(trig_mode && !level))
1075                         break;
1076
1077                 /* FIXME add logic for vcpu on reset */
1078                 if (unlikely(!apic_enabled(apic)))
1079                         break;
1080
1081                 result = 1;
1082
1083                 if (dest_map) {
1084                         __set_bit(vcpu->vcpu_id, dest_map->map);
1085                         dest_map->vectors[vcpu->vcpu_id] = vector;
1086                 }
1087
1088                 if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
1089                         if (trig_mode)
1090                                 kvm_lapic_set_vector(vector,
1091                                                      apic->regs + APIC_TMR);
1092                         else
1093                                 kvm_lapic_clear_vector(vector,
1094                                                        apic->regs + APIC_TMR);
1095                 }
1096
1097                 if (static_call(kvm_x86_deliver_posted_interrupt)(vcpu, vector)) {
1098                         kvm_lapic_set_irr(vector, apic);
1099                         kvm_make_request(KVM_REQ_EVENT, vcpu);
1100                         kvm_vcpu_kick(vcpu);
1101                 }
1102                 break;
1103
1104         case APIC_DM_REMRD:
1105                 result = 1;
1106                 vcpu->arch.pv.pv_unhalted = 1;
1107                 kvm_make_request(KVM_REQ_EVENT, vcpu);
1108                 kvm_vcpu_kick(vcpu);
1109                 break;
1110
1111         case APIC_DM_SMI:
1112                 result = 1;
1113                 kvm_make_request(KVM_REQ_SMI, vcpu);
1114                 kvm_vcpu_kick(vcpu);
1115                 break;
1116
1117         case APIC_DM_NMI:
1118                 result = 1;
1119                 kvm_inject_nmi(vcpu);
1120                 kvm_vcpu_kick(vcpu);
1121                 break;
1122
1123         case APIC_DM_INIT:
1124                 if (!trig_mode || level) {
1125                         result = 1;
1126                         /* assumes that there are only KVM_APIC_INIT/SIPI */
1127                         apic->pending_events = (1UL << KVM_APIC_INIT);
1128                         kvm_make_request(KVM_REQ_EVENT, vcpu);
1129                         kvm_vcpu_kick(vcpu);
1130                 }
1131                 break;
1132
1133         case APIC_DM_STARTUP:
1134                 result = 1;
1135                 apic->sipi_vector = vector;
1136                 /* make sure sipi_vector is visible for the receiver */
1137                 smp_wmb();
1138                 set_bit(KVM_APIC_SIPI, &apic->pending_events);
1139                 kvm_make_request(KVM_REQ_EVENT, vcpu);
1140                 kvm_vcpu_kick(vcpu);
1141                 break;
1142
1143         case APIC_DM_EXTINT:
1144                 /*
1145                  * Should only be called by kvm_apic_local_deliver() with LVT0,
1146                  * before NMI watchdog was enabled. Already handled by
1147                  * kvm_apic_accept_pic_intr().
1148                  */
1149                 break;
1150
1151         default:
1152                 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
1153                        delivery_mode);
1154                 break;
1155         }
1156         return result;
1157 }
1158
1159 /*
1160  * This routine identifies the destination vcpus mask meant to receive the
1161  * IOAPIC interrupts. It either uses kvm_apic_map_get_dest_lapic() to find
1162  * out the destination vcpus array and set the bitmap or it traverses to
1163  * each available vcpu to identify the same.
1164  */
1165 void kvm_bitmap_or_dest_vcpus(struct kvm *kvm, struct kvm_lapic_irq *irq,
1166                               unsigned long *vcpu_bitmap)
1167 {
1168         struct kvm_lapic **dest_vcpu = NULL;
1169         struct kvm_lapic *src = NULL;
1170         struct kvm_apic_map *map;
1171         struct kvm_vcpu *vcpu;
1172         unsigned long bitmap;
1173         int i, vcpu_idx;
1174         bool ret;
1175
1176         rcu_read_lock();
1177         map = rcu_dereference(kvm->arch.apic_map);
1178
1179         ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dest_vcpu,
1180                                           &bitmap);
1181         if (ret) {
1182                 for_each_set_bit(i, &bitmap, 16) {
1183                         if (!dest_vcpu[i])
1184                                 continue;
1185                         vcpu_idx = dest_vcpu[i]->vcpu->vcpu_idx;
1186                         __set_bit(vcpu_idx, vcpu_bitmap);
1187                 }
1188         } else {
1189                 kvm_for_each_vcpu(i, vcpu, kvm) {
1190                         if (!kvm_apic_present(vcpu))
1191                                 continue;
1192                         if (!kvm_apic_match_dest(vcpu, NULL,
1193                                                  irq->shorthand,
1194                                                  irq->dest_id,
1195                                                  irq->dest_mode))
1196                                 continue;
1197                         __set_bit(i, vcpu_bitmap);
1198                 }
1199         }
1200         rcu_read_unlock();
1201 }
1202
1203 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1204 {
1205         return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
1206 }
1207
1208 static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
1209 {
1210         return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
1211 }
1212
1213 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1214 {
1215         int trigger_mode;
1216
1217         /* Eoi the ioapic only if the ioapic doesn't own the vector. */
1218         if (!kvm_ioapic_handles_vector(apic, vector))
1219                 return;
1220
1221         /* Request a KVM exit to inform the userspace IOAPIC. */
1222         if (irqchip_split(apic->vcpu->kvm)) {
1223                 apic->vcpu->arch.pending_ioapic_eoi = vector;
1224                 kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1225                 return;
1226         }
1227
1228         if (apic_test_vector(vector, apic->regs + APIC_TMR))
1229                 trigger_mode = IOAPIC_LEVEL_TRIG;
1230         else
1231                 trigger_mode = IOAPIC_EDGE_TRIG;
1232
1233         kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1234 }
1235
1236 static int apic_set_eoi(struct kvm_lapic *apic)
1237 {
1238         int vector = apic_find_highest_isr(apic);
1239
1240         trace_kvm_eoi(apic, vector);
1241
1242         /*
1243          * Not every write EOI will has corresponding ISR,
1244          * one example is when Kernel check timer on setup_IO_APIC
1245          */
1246         if (vector == -1)
1247                 return vector;
1248
1249         apic_clear_isr(vector, apic);
1250         apic_update_ppr(apic);
1251
1252         if (to_hv_vcpu(apic->vcpu) &&
1253             test_bit(vector, to_hv_synic(apic->vcpu)->vec_bitmap))
1254                 kvm_hv_synic_send_eoi(apic->vcpu, vector);
1255
1256         kvm_ioapic_send_eoi(apic, vector);
1257         kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1258         return vector;
1259 }
1260
1261 /*
1262  * this interface assumes a trap-like exit, which has already finished
1263  * desired side effect including vISR and vPPR update.
1264  */
1265 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1266 {
1267         struct kvm_lapic *apic = vcpu->arch.apic;
1268
1269         trace_kvm_eoi(apic, vector);
1270
1271         kvm_ioapic_send_eoi(apic, vector);
1272         kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1273 }
1274 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1275
1276 void kvm_apic_send_ipi(struct kvm_lapic *apic, u32 icr_low, u32 icr_high)
1277 {
1278         struct kvm_lapic_irq irq;
1279
1280         irq.vector = icr_low & APIC_VECTOR_MASK;
1281         irq.delivery_mode = icr_low & APIC_MODE_MASK;
1282         irq.dest_mode = icr_low & APIC_DEST_MASK;
1283         irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1284         irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1285         irq.shorthand = icr_low & APIC_SHORT_MASK;
1286         irq.msi_redir_hint = false;
1287         if (apic_x2apic_mode(apic))
1288                 irq.dest_id = icr_high;
1289         else
1290                 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
1291
1292         trace_kvm_apic_ipi(icr_low, irq.dest_id);
1293
1294         kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1295 }
1296
1297 static u32 apic_get_tmcct(struct kvm_lapic *apic)
1298 {
1299         ktime_t remaining, now;
1300         s64 ns;
1301         u32 tmcct;
1302
1303         ASSERT(apic != NULL);
1304
1305         /* if initial count is 0, current count should also be 0 */
1306         if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1307                 apic->lapic_timer.period == 0)
1308                 return 0;
1309
1310         now = ktime_get();
1311         remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1312         if (ktime_to_ns(remaining) < 0)
1313                 remaining = 0;
1314
1315         ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1316         tmcct = div64_u64(ns,
1317                          (APIC_BUS_CYCLE_NS * apic->divide_count));
1318
1319         return tmcct;
1320 }
1321
1322 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1323 {
1324         struct kvm_vcpu *vcpu = apic->vcpu;
1325         struct kvm_run *run = vcpu->run;
1326
1327         kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1328         run->tpr_access.rip = kvm_rip_read(vcpu);
1329         run->tpr_access.is_write = write;
1330 }
1331
1332 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1333 {
1334         if (apic->vcpu->arch.tpr_access_reporting)
1335                 __report_tpr_access(apic, write);
1336 }
1337
1338 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1339 {
1340         u32 val = 0;
1341
1342         if (offset >= LAPIC_MMIO_LENGTH)
1343                 return 0;
1344
1345         switch (offset) {
1346         case APIC_ARBPRI:
1347                 break;
1348
1349         case APIC_TMCCT:        /* Timer CCR */
1350                 if (apic_lvtt_tscdeadline(apic))
1351                         return 0;
1352
1353                 val = apic_get_tmcct(apic);
1354                 break;
1355         case APIC_PROCPRI:
1356                 apic_update_ppr(apic);
1357                 val = kvm_lapic_get_reg(apic, offset);
1358                 break;
1359         case APIC_TASKPRI:
1360                 report_tpr_access(apic, false);
1361                 fallthrough;
1362         default:
1363                 val = kvm_lapic_get_reg(apic, offset);
1364                 break;
1365         }
1366
1367         return val;
1368 }
1369
1370 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1371 {
1372         return container_of(dev, struct kvm_lapic, dev);
1373 }
1374
1375 #define APIC_REG_MASK(reg)      (1ull << ((reg) >> 4))
1376 #define APIC_REGS_MASK(first, count) \
1377         (APIC_REG_MASK(first) * ((1ull << (count)) - 1))
1378
1379 int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1380                 void *data)
1381 {
1382         unsigned char alignment = offset & 0xf;
1383         u32 result;
1384         /* this bitmask has a bit cleared for each reserved register */
1385         u64 valid_reg_mask =
1386                 APIC_REG_MASK(APIC_ID) |
1387                 APIC_REG_MASK(APIC_LVR) |
1388                 APIC_REG_MASK(APIC_TASKPRI) |
1389                 APIC_REG_MASK(APIC_PROCPRI) |
1390                 APIC_REG_MASK(APIC_LDR) |
1391                 APIC_REG_MASK(APIC_DFR) |
1392                 APIC_REG_MASK(APIC_SPIV) |
1393                 APIC_REGS_MASK(APIC_ISR, APIC_ISR_NR) |
1394                 APIC_REGS_MASK(APIC_TMR, APIC_ISR_NR) |
1395                 APIC_REGS_MASK(APIC_IRR, APIC_ISR_NR) |
1396                 APIC_REG_MASK(APIC_ESR) |
1397                 APIC_REG_MASK(APIC_ICR) |
1398                 APIC_REG_MASK(APIC_ICR2) |
1399                 APIC_REG_MASK(APIC_LVTT) |
1400                 APIC_REG_MASK(APIC_LVTTHMR) |
1401                 APIC_REG_MASK(APIC_LVTPC) |
1402                 APIC_REG_MASK(APIC_LVT0) |
1403                 APIC_REG_MASK(APIC_LVT1) |
1404                 APIC_REG_MASK(APIC_LVTERR) |
1405                 APIC_REG_MASK(APIC_TMICT) |
1406                 APIC_REG_MASK(APIC_TMCCT) |
1407                 APIC_REG_MASK(APIC_TDCR);
1408
1409         /* ARBPRI is not valid on x2APIC */
1410         if (!apic_x2apic_mode(apic))
1411                 valid_reg_mask |= APIC_REG_MASK(APIC_ARBPRI);
1412
1413         if (offset > 0x3f0 || !(valid_reg_mask & APIC_REG_MASK(offset)))
1414                 return 1;
1415
1416         result = __apic_read(apic, offset & ~0xf);
1417
1418         trace_kvm_apic_read(offset, result);
1419
1420         switch (len) {
1421         case 1:
1422         case 2:
1423         case 4:
1424                 memcpy(data, (char *)&result + alignment, len);
1425                 break;
1426         default:
1427                 printk(KERN_ERR "Local APIC read with len = %x, "
1428                        "should be 1,2, or 4 instead\n", len);
1429                 break;
1430         }
1431         return 0;
1432 }
1433 EXPORT_SYMBOL_GPL(kvm_lapic_reg_read);
1434
1435 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1436 {
1437         return addr >= apic->base_address &&
1438                 addr < apic->base_address + LAPIC_MMIO_LENGTH;
1439 }
1440
1441 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1442                            gpa_t address, int len, void *data)
1443 {
1444         struct kvm_lapic *apic = to_lapic(this);
1445         u32 offset = address - apic->base_address;
1446
1447         if (!apic_mmio_in_range(apic, address))
1448                 return -EOPNOTSUPP;
1449
1450         if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
1451                 if (!kvm_check_has_quirk(vcpu->kvm,
1452                                          KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
1453                         return -EOPNOTSUPP;
1454
1455                 memset(data, 0xff, len);
1456                 return 0;
1457         }
1458
1459         kvm_lapic_reg_read(apic, offset, len, data);
1460
1461         return 0;
1462 }
1463
1464 static void update_divide_count(struct kvm_lapic *apic)
1465 {
1466         u32 tmp1, tmp2, tdcr;
1467
1468         tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1469         tmp1 = tdcr & 0xf;
1470         tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1471         apic->divide_count = 0x1 << (tmp2 & 0x7);
1472 }
1473
1474 static void limit_periodic_timer_frequency(struct kvm_lapic *apic)
1475 {
1476         /*
1477          * Do not allow the guest to program periodic timers with small
1478          * interval, since the hrtimers are not throttled by the host
1479          * scheduler.
1480          */
1481         if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1482                 s64 min_period = min_timer_period_us * 1000LL;
1483
1484                 if (apic->lapic_timer.period < min_period) {
1485                         pr_info_ratelimited(
1486                             "kvm: vcpu %i: requested %lld ns "
1487                             "lapic timer period limited to %lld ns\n",
1488                             apic->vcpu->vcpu_id,
1489                             apic->lapic_timer.period, min_period);
1490                         apic->lapic_timer.period = min_period;
1491                 }
1492         }
1493 }
1494
1495 static void cancel_hv_timer(struct kvm_lapic *apic);
1496
1497 static void cancel_apic_timer(struct kvm_lapic *apic)
1498 {
1499         hrtimer_cancel(&apic->lapic_timer.timer);
1500         preempt_disable();
1501         if (apic->lapic_timer.hv_timer_in_use)
1502                 cancel_hv_timer(apic);
1503         preempt_enable();
1504 }
1505
1506 static void apic_update_lvtt(struct kvm_lapic *apic)
1507 {
1508         u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1509                         apic->lapic_timer.timer_mode_mask;
1510
1511         if (apic->lapic_timer.timer_mode != timer_mode) {
1512                 if (apic_lvtt_tscdeadline(apic) != (timer_mode ==
1513                                 APIC_LVT_TIMER_TSCDEADLINE)) {
1514                         cancel_apic_timer(apic);
1515                         kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1516                         apic->lapic_timer.period = 0;
1517                         apic->lapic_timer.tscdeadline = 0;
1518                 }
1519                 apic->lapic_timer.timer_mode = timer_mode;
1520                 limit_periodic_timer_frequency(apic);
1521         }
1522 }
1523
1524 /*
1525  * On APICv, this test will cause a busy wait
1526  * during a higher-priority task.
1527  */
1528
1529 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1530 {
1531         struct kvm_lapic *apic = vcpu->arch.apic;
1532         u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1533
1534         if (kvm_apic_hw_enabled(apic)) {
1535                 int vec = reg & APIC_VECTOR_MASK;
1536                 void *bitmap = apic->regs + APIC_ISR;
1537
1538                 if (vcpu->arch.apicv_active)
1539                         bitmap = apic->regs + APIC_IRR;
1540
1541                 if (apic_test_vector(vec, bitmap))
1542                         return true;
1543         }
1544         return false;
1545 }
1546
1547 static inline void __wait_lapic_expire(struct kvm_vcpu *vcpu, u64 guest_cycles)
1548 {
1549         u64 timer_advance_ns = vcpu->arch.apic->lapic_timer.timer_advance_ns;
1550
1551         /*
1552          * If the guest TSC is running at a different ratio than the host, then
1553          * convert the delay to nanoseconds to achieve an accurate delay.  Note
1554          * that __delay() uses delay_tsc whenever the hardware has TSC, thus
1555          * always for VMX enabled hardware.
1556          */
1557         if (vcpu->arch.tsc_scaling_ratio == kvm_default_tsc_scaling_ratio) {
1558                 __delay(min(guest_cycles,
1559                         nsec_to_cycles(vcpu, timer_advance_ns)));
1560         } else {
1561                 u64 delay_ns = guest_cycles * 1000000ULL;
1562                 do_div(delay_ns, vcpu->arch.virtual_tsc_khz);
1563                 ndelay(min_t(u32, delay_ns, timer_advance_ns));
1564         }
1565 }
1566
1567 static inline void adjust_lapic_timer_advance(struct kvm_vcpu *vcpu,
1568                                               s64 advance_expire_delta)
1569 {
1570         struct kvm_lapic *apic = vcpu->arch.apic;
1571         u32 timer_advance_ns = apic->lapic_timer.timer_advance_ns;
1572         u64 ns;
1573
1574         /* Do not adjust for tiny fluctuations or large random spikes. */
1575         if (abs(advance_expire_delta) > LAPIC_TIMER_ADVANCE_ADJUST_MAX ||
1576             abs(advance_expire_delta) < LAPIC_TIMER_ADVANCE_ADJUST_MIN)
1577                 return;
1578
1579         /* too early */
1580         if (advance_expire_delta < 0) {
1581                 ns = -advance_expire_delta * 1000000ULL;
1582                 do_div(ns, vcpu->arch.virtual_tsc_khz);
1583                 timer_advance_ns -= ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1584         } else {
1585         /* too late */
1586                 ns = advance_expire_delta * 1000000ULL;
1587                 do_div(ns, vcpu->arch.virtual_tsc_khz);
1588                 timer_advance_ns += ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1589         }
1590
1591         if (unlikely(timer_advance_ns > LAPIC_TIMER_ADVANCE_NS_MAX))
1592                 timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
1593         apic->lapic_timer.timer_advance_ns = timer_advance_ns;
1594 }
1595
1596 static void __kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1597 {
1598         struct kvm_lapic *apic = vcpu->arch.apic;
1599         u64 guest_tsc, tsc_deadline;
1600
1601         tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1602         apic->lapic_timer.expired_tscdeadline = 0;
1603         guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1604         apic->lapic_timer.advance_expire_delta = guest_tsc - tsc_deadline;
1605
1606         if (lapic_timer_advance_dynamic) {
1607                 adjust_lapic_timer_advance(vcpu, apic->lapic_timer.advance_expire_delta);
1608                 /*
1609                  * If the timer fired early, reread the TSC to account for the
1610                  * overhead of the above adjustment to avoid waiting longer
1611                  * than is necessary.
1612                  */
1613                 if (guest_tsc < tsc_deadline)
1614                         guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1615         }
1616
1617         if (guest_tsc < tsc_deadline)
1618                 __wait_lapic_expire(vcpu, tsc_deadline - guest_tsc);
1619 }
1620
1621 void kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1622 {
1623         if (lapic_in_kernel(vcpu) &&
1624             vcpu->arch.apic->lapic_timer.expired_tscdeadline &&
1625             vcpu->arch.apic->lapic_timer.timer_advance_ns &&
1626             lapic_timer_int_injected(vcpu))
1627                 __kvm_wait_lapic_expire(vcpu);
1628 }
1629 EXPORT_SYMBOL_GPL(kvm_wait_lapic_expire);
1630
1631 static void kvm_apic_inject_pending_timer_irqs(struct kvm_lapic *apic)
1632 {
1633         struct kvm_timer *ktimer = &apic->lapic_timer;
1634
1635         kvm_apic_local_deliver(apic, APIC_LVTT);
1636         if (apic_lvtt_tscdeadline(apic)) {
1637                 ktimer->tscdeadline = 0;
1638         } else if (apic_lvtt_oneshot(apic)) {
1639                 ktimer->tscdeadline = 0;
1640                 ktimer->target_expiration = 0;
1641         }
1642 }
1643
1644 static void apic_timer_expired(struct kvm_lapic *apic, bool from_timer_fn)
1645 {
1646         struct kvm_vcpu *vcpu = apic->vcpu;
1647         struct kvm_timer *ktimer = &apic->lapic_timer;
1648
1649         if (atomic_read(&apic->lapic_timer.pending))
1650                 return;
1651
1652         if (apic_lvtt_tscdeadline(apic) || ktimer->hv_timer_in_use)
1653                 ktimer->expired_tscdeadline = ktimer->tscdeadline;
1654
1655         if (!from_timer_fn && vcpu->arch.apicv_active) {
1656                 WARN_ON(kvm_get_running_vcpu() != vcpu);
1657                 kvm_apic_inject_pending_timer_irqs(apic);
1658                 return;
1659         }
1660
1661         if (kvm_use_posted_timer_interrupt(apic->vcpu)) {
1662                 /*
1663                  * Ensure the guest's timer has truly expired before posting an
1664                  * interrupt.  Open code the relevant checks to avoid querying
1665                  * lapic_timer_int_injected(), which will be false since the
1666                  * interrupt isn't yet injected.  Waiting until after injecting
1667                  * is not an option since that won't help a posted interrupt.
1668                  */
1669                 if (vcpu->arch.apic->lapic_timer.expired_tscdeadline &&
1670                     vcpu->arch.apic->lapic_timer.timer_advance_ns)
1671                         __kvm_wait_lapic_expire(vcpu);
1672                 kvm_apic_inject_pending_timer_irqs(apic);
1673                 return;
1674         }
1675
1676         atomic_inc(&apic->lapic_timer.pending);
1677         kvm_make_request(KVM_REQ_UNBLOCK, vcpu);
1678         if (from_timer_fn)
1679                 kvm_vcpu_kick(vcpu);
1680 }
1681
1682 static void start_sw_tscdeadline(struct kvm_lapic *apic)
1683 {
1684         struct kvm_timer *ktimer = &apic->lapic_timer;
1685         u64 guest_tsc, tscdeadline = ktimer->tscdeadline;
1686         u64 ns = 0;
1687         ktime_t expire;
1688         struct kvm_vcpu *vcpu = apic->vcpu;
1689         unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1690         unsigned long flags;
1691         ktime_t now;
1692
1693         if (unlikely(!tscdeadline || !this_tsc_khz))
1694                 return;
1695
1696         local_irq_save(flags);
1697
1698         now = ktime_get();
1699         guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1700
1701         ns = (tscdeadline - guest_tsc) * 1000000ULL;
1702         do_div(ns, this_tsc_khz);
1703
1704         if (likely(tscdeadline > guest_tsc) &&
1705             likely(ns > apic->lapic_timer.timer_advance_ns)) {
1706                 expire = ktime_add_ns(now, ns);
1707                 expire = ktime_sub_ns(expire, ktimer->timer_advance_ns);
1708                 hrtimer_start(&ktimer->timer, expire, HRTIMER_MODE_ABS_HARD);
1709         } else
1710                 apic_timer_expired(apic, false);
1711
1712         local_irq_restore(flags);
1713 }
1714
1715 static inline u64 tmict_to_ns(struct kvm_lapic *apic, u32 tmict)
1716 {
1717         return (u64)tmict * APIC_BUS_CYCLE_NS * (u64)apic->divide_count;
1718 }
1719
1720 static void update_target_expiration(struct kvm_lapic *apic, uint32_t old_divisor)
1721 {
1722         ktime_t now, remaining;
1723         u64 ns_remaining_old, ns_remaining_new;
1724
1725         apic->lapic_timer.period =
1726                         tmict_to_ns(apic, kvm_lapic_get_reg(apic, APIC_TMICT));
1727         limit_periodic_timer_frequency(apic);
1728
1729         now = ktime_get();
1730         remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1731         if (ktime_to_ns(remaining) < 0)
1732                 remaining = 0;
1733
1734         ns_remaining_old = ktime_to_ns(remaining);
1735         ns_remaining_new = mul_u64_u32_div(ns_remaining_old,
1736                                            apic->divide_count, old_divisor);
1737
1738         apic->lapic_timer.tscdeadline +=
1739                 nsec_to_cycles(apic->vcpu, ns_remaining_new) -
1740                 nsec_to_cycles(apic->vcpu, ns_remaining_old);
1741         apic->lapic_timer.target_expiration = ktime_add_ns(now, ns_remaining_new);
1742 }
1743
1744 static bool set_target_expiration(struct kvm_lapic *apic, u32 count_reg)
1745 {
1746         ktime_t now;
1747         u64 tscl = rdtsc();
1748         s64 deadline;
1749
1750         now = ktime_get();
1751         apic->lapic_timer.period =
1752                         tmict_to_ns(apic, kvm_lapic_get_reg(apic, APIC_TMICT));
1753
1754         if (!apic->lapic_timer.period) {
1755                 apic->lapic_timer.tscdeadline = 0;
1756                 return false;
1757         }
1758
1759         limit_periodic_timer_frequency(apic);
1760         deadline = apic->lapic_timer.period;
1761
1762         if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1763                 if (unlikely(count_reg != APIC_TMICT)) {
1764                         deadline = tmict_to_ns(apic,
1765                                      kvm_lapic_get_reg(apic, count_reg));
1766                         if (unlikely(deadline <= 0))
1767                                 deadline = apic->lapic_timer.period;
1768                         else if (unlikely(deadline > apic->lapic_timer.period)) {
1769                                 pr_info_ratelimited(
1770                                     "kvm: vcpu %i: requested lapic timer restore with "
1771                                     "starting count register %#x=%u (%lld ns) > initial count (%lld ns). "
1772                                     "Using initial count to start timer.\n",
1773                                     apic->vcpu->vcpu_id,
1774                                     count_reg,
1775                                     kvm_lapic_get_reg(apic, count_reg),
1776                                     deadline, apic->lapic_timer.period);
1777                                 kvm_lapic_set_reg(apic, count_reg, 0);
1778                                 deadline = apic->lapic_timer.period;
1779                         }
1780                 }
1781         }
1782
1783         apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1784                 nsec_to_cycles(apic->vcpu, deadline);
1785         apic->lapic_timer.target_expiration = ktime_add_ns(now, deadline);
1786
1787         return true;
1788 }
1789
1790 static void advance_periodic_target_expiration(struct kvm_lapic *apic)
1791 {
1792         ktime_t now = ktime_get();
1793         u64 tscl = rdtsc();
1794         ktime_t delta;
1795
1796         /*
1797          * Synchronize both deadlines to the same time source or
1798          * differences in the periods (caused by differences in the
1799          * underlying clocks or numerical approximation errors) will
1800          * cause the two to drift apart over time as the errors
1801          * accumulate.
1802          */
1803         apic->lapic_timer.target_expiration =
1804                 ktime_add_ns(apic->lapic_timer.target_expiration,
1805                                 apic->lapic_timer.period);
1806         delta = ktime_sub(apic->lapic_timer.target_expiration, now);
1807         apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1808                 nsec_to_cycles(apic->vcpu, delta);
1809 }
1810
1811 static void start_sw_period(struct kvm_lapic *apic)
1812 {
1813         if (!apic->lapic_timer.period)
1814                 return;
1815
1816         if (ktime_after(ktime_get(),
1817                         apic->lapic_timer.target_expiration)) {
1818                 apic_timer_expired(apic, false);
1819
1820                 if (apic_lvtt_oneshot(apic))
1821                         return;
1822
1823                 advance_periodic_target_expiration(apic);
1824         }
1825
1826         hrtimer_start(&apic->lapic_timer.timer,
1827                 apic->lapic_timer.target_expiration,
1828                 HRTIMER_MODE_ABS_HARD);
1829 }
1830
1831 bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1832 {
1833         if (!lapic_in_kernel(vcpu))
1834                 return false;
1835
1836         return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1837 }
1838 EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use);
1839
1840 static void cancel_hv_timer(struct kvm_lapic *apic)
1841 {
1842         WARN_ON(preemptible());
1843         WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1844         static_call(kvm_x86_cancel_hv_timer)(apic->vcpu);
1845         apic->lapic_timer.hv_timer_in_use = false;
1846 }
1847
1848 static bool start_hv_timer(struct kvm_lapic *apic)
1849 {
1850         struct kvm_timer *ktimer = &apic->lapic_timer;
1851         struct kvm_vcpu *vcpu = apic->vcpu;
1852         bool expired;
1853
1854         WARN_ON(preemptible());
1855         if (!kvm_can_use_hv_timer(vcpu))
1856                 return false;
1857
1858         if (!ktimer->tscdeadline)
1859                 return false;
1860
1861         if (static_call(kvm_x86_set_hv_timer)(vcpu, ktimer->tscdeadline, &expired))
1862                 return false;
1863
1864         ktimer->hv_timer_in_use = true;
1865         hrtimer_cancel(&ktimer->timer);
1866
1867         /*
1868          * To simplify handling the periodic timer, leave the hv timer running
1869          * even if the deadline timer has expired, i.e. rely on the resulting
1870          * VM-Exit to recompute the periodic timer's target expiration.
1871          */
1872         if (!apic_lvtt_period(apic)) {
1873                 /*
1874                  * Cancel the hv timer if the sw timer fired while the hv timer
1875                  * was being programmed, or if the hv timer itself expired.
1876                  */
1877                 if (atomic_read(&ktimer->pending)) {
1878                         cancel_hv_timer(apic);
1879                 } else if (expired) {
1880                         apic_timer_expired(apic, false);
1881                         cancel_hv_timer(apic);
1882                 }
1883         }
1884
1885         trace_kvm_hv_timer_state(vcpu->vcpu_id, ktimer->hv_timer_in_use);
1886
1887         return true;
1888 }
1889
1890 static void start_sw_timer(struct kvm_lapic *apic)
1891 {
1892         struct kvm_timer *ktimer = &apic->lapic_timer;
1893
1894         WARN_ON(preemptible());
1895         if (apic->lapic_timer.hv_timer_in_use)
1896                 cancel_hv_timer(apic);
1897         if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
1898                 return;
1899
1900         if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1901                 start_sw_period(apic);
1902         else if (apic_lvtt_tscdeadline(apic))
1903                 start_sw_tscdeadline(apic);
1904         trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, false);
1905 }
1906
1907 static void restart_apic_timer(struct kvm_lapic *apic)
1908 {
1909         preempt_disable();
1910
1911         if (!apic_lvtt_period(apic) && atomic_read(&apic->lapic_timer.pending))
1912                 goto out;
1913
1914         if (!start_hv_timer(apic))
1915                 start_sw_timer(apic);
1916 out:
1917         preempt_enable();
1918 }
1919
1920 void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1921 {
1922         struct kvm_lapic *apic = vcpu->arch.apic;
1923
1924         preempt_disable();
1925         /* If the preempt notifier has already run, it also called apic_timer_expired */
1926         if (!apic->lapic_timer.hv_timer_in_use)
1927                 goto out;
1928         WARN_ON(rcuwait_active(&vcpu->wait));
1929         apic_timer_expired(apic, false);
1930         cancel_hv_timer(apic);
1931
1932         if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1933                 advance_periodic_target_expiration(apic);
1934                 restart_apic_timer(apic);
1935         }
1936 out:
1937         preempt_enable();
1938 }
1939 EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
1940
1941 void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
1942 {
1943         restart_apic_timer(vcpu->arch.apic);
1944 }
1945 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer);
1946
1947 void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
1948 {
1949         struct kvm_lapic *apic = vcpu->arch.apic;
1950
1951         preempt_disable();
1952         /* Possibly the TSC deadline timer is not enabled yet */
1953         if (apic->lapic_timer.hv_timer_in_use)
1954                 start_sw_timer(apic);
1955         preempt_enable();
1956 }
1957 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer);
1958
1959 void kvm_lapic_restart_hv_timer(struct kvm_vcpu *vcpu)
1960 {
1961         struct kvm_lapic *apic = vcpu->arch.apic;
1962
1963         WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1964         restart_apic_timer(apic);
1965 }
1966
1967 static void __start_apic_timer(struct kvm_lapic *apic, u32 count_reg)
1968 {
1969         atomic_set(&apic->lapic_timer.pending, 0);
1970
1971         if ((apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1972             && !set_target_expiration(apic, count_reg))
1973                 return;
1974
1975         restart_apic_timer(apic);
1976 }
1977
1978 static void start_apic_timer(struct kvm_lapic *apic)
1979 {
1980         __start_apic_timer(apic, APIC_TMICT);
1981 }
1982
1983 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1984 {
1985         bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
1986
1987         if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1988                 apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1989                 if (lvt0_in_nmi_mode) {
1990                         atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1991                 } else
1992                         atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1993         }
1994 }
1995
1996 int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1997 {
1998         int ret = 0;
1999
2000         trace_kvm_apic_write(reg, val);
2001
2002         switch (reg) {
2003         case APIC_ID:           /* Local APIC ID */
2004                 if (!apic_x2apic_mode(apic))
2005                         kvm_apic_set_xapic_id(apic, val >> 24);
2006                 else
2007                         ret = 1;
2008                 break;
2009
2010         case APIC_TASKPRI:
2011                 report_tpr_access(apic, true);
2012                 apic_set_tpr(apic, val & 0xff);
2013                 break;
2014
2015         case APIC_EOI:
2016                 apic_set_eoi(apic);
2017                 break;
2018
2019         case APIC_LDR:
2020                 if (!apic_x2apic_mode(apic))
2021                         kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
2022                 else
2023                         ret = 1;
2024                 break;
2025
2026         case APIC_DFR:
2027                 if (!apic_x2apic_mode(apic))
2028                         kvm_apic_set_dfr(apic, val | 0x0FFFFFFF);
2029                 else
2030                         ret = 1;
2031                 break;
2032
2033         case APIC_SPIV: {
2034                 u32 mask = 0x3ff;
2035                 if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
2036                         mask |= APIC_SPIV_DIRECTED_EOI;
2037                 apic_set_spiv(apic, val & mask);
2038                 if (!(val & APIC_SPIV_APIC_ENABLED)) {
2039                         int i;
2040                         u32 lvt_val;
2041
2042                         for (i = 0; i < KVM_APIC_LVT_NUM; i++) {
2043                                 lvt_val = kvm_lapic_get_reg(apic,
2044                                                        APIC_LVTT + 0x10 * i);
2045                                 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i,
2046                                              lvt_val | APIC_LVT_MASKED);
2047                         }
2048                         apic_update_lvtt(apic);
2049                         atomic_set(&apic->lapic_timer.pending, 0);
2050
2051                 }
2052                 break;
2053         }
2054         case APIC_ICR:
2055                 /* No delay here, so we always clear the pending bit */
2056                 val &= ~(1 << 12);
2057                 kvm_apic_send_ipi(apic, val, kvm_lapic_get_reg(apic, APIC_ICR2));
2058                 kvm_lapic_set_reg(apic, APIC_ICR, val);
2059                 break;
2060
2061         case APIC_ICR2:
2062                 if (!apic_x2apic_mode(apic))
2063                         val &= 0xff000000;
2064                 kvm_lapic_set_reg(apic, APIC_ICR2, val);
2065                 break;
2066
2067         case APIC_LVT0:
2068                 apic_manage_nmi_watchdog(apic, val);
2069                 fallthrough;
2070         case APIC_LVTTHMR:
2071         case APIC_LVTPC:
2072         case APIC_LVT1:
2073         case APIC_LVTERR: {
2074                 /* TODO: Check vector */
2075                 size_t size;
2076                 u32 index;
2077
2078                 if (!kvm_apic_sw_enabled(apic))
2079                         val |= APIC_LVT_MASKED;
2080                 size = ARRAY_SIZE(apic_lvt_mask);
2081                 index = array_index_nospec(
2082                                 (reg - APIC_LVTT) >> 4, size);
2083                 val &= apic_lvt_mask[index];
2084                 kvm_lapic_set_reg(apic, reg, val);
2085                 break;
2086         }
2087
2088         case APIC_LVTT:
2089                 if (!kvm_apic_sw_enabled(apic))
2090                         val |= APIC_LVT_MASKED;
2091                 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
2092                 kvm_lapic_set_reg(apic, APIC_LVTT, val);
2093                 apic_update_lvtt(apic);
2094                 break;
2095
2096         case APIC_TMICT:
2097                 if (apic_lvtt_tscdeadline(apic))
2098                         break;
2099
2100                 cancel_apic_timer(apic);
2101                 kvm_lapic_set_reg(apic, APIC_TMICT, val);
2102                 start_apic_timer(apic);
2103                 break;
2104
2105         case APIC_TDCR: {
2106                 uint32_t old_divisor = apic->divide_count;
2107
2108                 kvm_lapic_set_reg(apic, APIC_TDCR, val & 0xb);
2109                 update_divide_count(apic);
2110                 if (apic->divide_count != old_divisor &&
2111                                 apic->lapic_timer.period) {
2112                         hrtimer_cancel(&apic->lapic_timer.timer);
2113                         update_target_expiration(apic, old_divisor);
2114                         restart_apic_timer(apic);
2115                 }
2116                 break;
2117         }
2118         case APIC_ESR:
2119                 if (apic_x2apic_mode(apic) && val != 0)
2120                         ret = 1;
2121                 break;
2122
2123         case APIC_SELF_IPI:
2124                 if (apic_x2apic_mode(apic)) {
2125                         kvm_lapic_reg_write(apic, APIC_ICR,
2126                                             APIC_DEST_SELF | (val & APIC_VECTOR_MASK));
2127                 } else
2128                         ret = 1;
2129                 break;
2130         default:
2131                 ret = 1;
2132                 break;
2133         }
2134
2135         kvm_recalculate_apic_map(apic->vcpu->kvm);
2136
2137         return ret;
2138 }
2139 EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
2140
2141 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
2142                             gpa_t address, int len, const void *data)
2143 {
2144         struct kvm_lapic *apic = to_lapic(this);
2145         unsigned int offset = address - apic->base_address;
2146         u32 val;
2147
2148         if (!apic_mmio_in_range(apic, address))
2149                 return -EOPNOTSUPP;
2150
2151         if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
2152                 if (!kvm_check_has_quirk(vcpu->kvm,
2153                                          KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
2154                         return -EOPNOTSUPP;
2155
2156                 return 0;
2157         }
2158
2159         /*
2160          * APIC register must be aligned on 128-bits boundary.
2161          * 32/64/128 bits registers must be accessed thru 32 bits.
2162          * Refer SDM 8.4.1
2163          */
2164         if (len != 4 || (offset & 0xf))
2165                 return 0;
2166
2167         val = *(u32*)data;
2168
2169         kvm_lapic_reg_write(apic, offset & 0xff0, val);
2170
2171         return 0;
2172 }
2173
2174 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
2175 {
2176         kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
2177 }
2178 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
2179
2180 /* emulate APIC access in a trap manner */
2181 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
2182 {
2183         u32 val = 0;
2184
2185         /* hw has done the conditional check and inst decode */
2186         offset &= 0xff0;
2187
2188         kvm_lapic_reg_read(vcpu->arch.apic, offset, 4, &val);
2189
2190         /* TODO: optimize to just emulate side effect w/o one more write */
2191         kvm_lapic_reg_write(vcpu->arch.apic, offset, val);
2192 }
2193 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
2194
2195 void kvm_free_lapic(struct kvm_vcpu *vcpu)
2196 {
2197         struct kvm_lapic *apic = vcpu->arch.apic;
2198
2199         if (!vcpu->arch.apic)
2200                 return;
2201
2202         hrtimer_cancel(&apic->lapic_timer.timer);
2203
2204         if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
2205                 static_branch_slow_dec_deferred(&apic_hw_disabled);
2206
2207         if (!apic->sw_enabled)
2208                 static_branch_slow_dec_deferred(&apic_sw_disabled);
2209
2210         if (apic->regs)
2211                 free_page((unsigned long)apic->regs);
2212
2213         kfree(apic);
2214 }
2215
2216 /*
2217  *----------------------------------------------------------------------
2218  * LAPIC interface
2219  *----------------------------------------------------------------------
2220  */
2221 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
2222 {
2223         struct kvm_lapic *apic = vcpu->arch.apic;
2224
2225         if (!kvm_apic_present(vcpu) || !apic_lvtt_tscdeadline(apic))
2226                 return 0;
2227
2228         return apic->lapic_timer.tscdeadline;
2229 }
2230
2231 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
2232 {
2233         struct kvm_lapic *apic = vcpu->arch.apic;
2234
2235         if (!kvm_apic_present(vcpu) || !apic_lvtt_tscdeadline(apic))
2236                 return;
2237
2238         hrtimer_cancel(&apic->lapic_timer.timer);
2239         apic->lapic_timer.tscdeadline = data;
2240         start_apic_timer(apic);
2241 }
2242
2243 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
2244 {
2245         struct kvm_lapic *apic = vcpu->arch.apic;
2246
2247         apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
2248                      | (kvm_lapic_get_reg(apic, APIC_TASKPRI) & 4));
2249 }
2250
2251 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
2252 {
2253         u64 tpr;
2254
2255         tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
2256
2257         return (tpr & 0xf0) >> 4;
2258 }
2259
2260 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
2261 {
2262         u64 old_value = vcpu->arch.apic_base;
2263         struct kvm_lapic *apic = vcpu->arch.apic;
2264
2265         if (!apic)
2266                 value |= MSR_IA32_APICBASE_BSP;
2267
2268         vcpu->arch.apic_base = value;
2269
2270         if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
2271                 kvm_update_cpuid_runtime(vcpu);
2272
2273         if (!apic)
2274                 return;
2275
2276         /* update jump label if enable bit changes */
2277         if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
2278                 if (value & MSR_IA32_APICBASE_ENABLE) {
2279                         kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2280                         static_branch_slow_dec_deferred(&apic_hw_disabled);
2281                         /* Check if there are APF page ready requests pending */
2282                         kvm_make_request(KVM_REQ_APF_READY, vcpu);
2283                 } else {
2284                         static_branch_inc(&apic_hw_disabled.key);
2285                         atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
2286                 }
2287         }
2288
2289         if (((old_value ^ value) & X2APIC_ENABLE) && (value & X2APIC_ENABLE))
2290                 kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
2291
2292         if ((old_value ^ value) & (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE))
2293                 static_call(kvm_x86_set_virtual_apic_mode)(vcpu);
2294
2295         apic->base_address = apic->vcpu->arch.apic_base &
2296                              MSR_IA32_APICBASE_BASE;
2297
2298         if ((value & MSR_IA32_APICBASE_ENABLE) &&
2299              apic->base_address != APIC_DEFAULT_PHYS_BASE)
2300                 pr_warn_once("APIC base relocation is unsupported by KVM");
2301 }
2302
2303 void kvm_apic_update_apicv(struct kvm_vcpu *vcpu)
2304 {
2305         struct kvm_lapic *apic = vcpu->arch.apic;
2306
2307         if (vcpu->arch.apicv_active) {
2308                 /* irr_pending is always true when apicv is activated. */
2309                 apic->irr_pending = true;
2310                 apic->isr_count = 1;
2311         } else {
2312                 apic->irr_pending = (apic_search_irr(apic) != -1);
2313                 apic->isr_count = count_vectors(apic->regs + APIC_ISR);
2314         }
2315 }
2316 EXPORT_SYMBOL_GPL(kvm_apic_update_apicv);
2317
2318 void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
2319 {
2320         struct kvm_lapic *apic = vcpu->arch.apic;
2321         int i;
2322
2323         if (!apic)
2324                 return;
2325
2326         /* Stop the timer in case it's a reset to an active apic */
2327         hrtimer_cancel(&apic->lapic_timer.timer);
2328
2329         if (!init_event) {
2330                 kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
2331                                          MSR_IA32_APICBASE_ENABLE);
2332                 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2333         }
2334         kvm_apic_set_version(apic->vcpu);
2335
2336         for (i = 0; i < KVM_APIC_LVT_NUM; i++)
2337                 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
2338         apic_update_lvtt(apic);
2339         if (kvm_vcpu_is_reset_bsp(vcpu) &&
2340             kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
2341                 kvm_lapic_set_reg(apic, APIC_LVT0,
2342                              SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
2343         apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2344
2345         kvm_apic_set_dfr(apic, 0xffffffffU);
2346         apic_set_spiv(apic, 0xff);
2347         kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
2348         if (!apic_x2apic_mode(apic))
2349                 kvm_apic_set_ldr(apic, 0);
2350         kvm_lapic_set_reg(apic, APIC_ESR, 0);
2351         kvm_lapic_set_reg(apic, APIC_ICR, 0);
2352         kvm_lapic_set_reg(apic, APIC_ICR2, 0);
2353         kvm_lapic_set_reg(apic, APIC_TDCR, 0);
2354         kvm_lapic_set_reg(apic, APIC_TMICT, 0);
2355         for (i = 0; i < 8; i++) {
2356                 kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
2357                 kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
2358                 kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
2359         }
2360         kvm_apic_update_apicv(vcpu);
2361         apic->highest_isr_cache = -1;
2362         update_divide_count(apic);
2363         atomic_set(&apic->lapic_timer.pending, 0);
2364         if (kvm_vcpu_is_bsp(vcpu))
2365                 kvm_lapic_set_base(vcpu,
2366                                 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
2367         vcpu->arch.pv_eoi.msr_val = 0;
2368         apic_update_ppr(apic);
2369         if (vcpu->arch.apicv_active) {
2370                 static_call(kvm_x86_apicv_post_state_restore)(vcpu);
2371                 static_call(kvm_x86_hwapic_irr_update)(vcpu, -1);
2372                 static_call(kvm_x86_hwapic_isr_update)(vcpu, -1);
2373         }
2374
2375         vcpu->arch.apic_arb_prio = 0;
2376         vcpu->arch.apic_attention = 0;
2377
2378         kvm_recalculate_apic_map(vcpu->kvm);
2379 }
2380
2381 /*
2382  *----------------------------------------------------------------------
2383  * timer interface
2384  *----------------------------------------------------------------------
2385  */
2386
2387 static bool lapic_is_periodic(struct kvm_lapic *apic)
2388 {
2389         return apic_lvtt_period(apic);
2390 }
2391
2392 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
2393 {
2394         struct kvm_lapic *apic = vcpu->arch.apic;
2395
2396         if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
2397                 return atomic_read(&apic->lapic_timer.pending);
2398
2399         return 0;
2400 }
2401
2402 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
2403 {
2404         u32 reg = kvm_lapic_get_reg(apic, lvt_type);
2405         int vector, mode, trig_mode;
2406
2407         if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
2408                 vector = reg & APIC_VECTOR_MASK;
2409                 mode = reg & APIC_MODE_MASK;
2410                 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
2411                 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
2412                                         NULL);
2413         }
2414         return 0;
2415 }
2416
2417 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
2418 {
2419         struct kvm_lapic *apic = vcpu->arch.apic;
2420
2421         if (apic)
2422                 kvm_apic_local_deliver(apic, APIC_LVT0);
2423 }
2424
2425 static const struct kvm_io_device_ops apic_mmio_ops = {
2426         .read     = apic_mmio_read,
2427         .write    = apic_mmio_write,
2428 };
2429
2430 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
2431 {
2432         struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
2433         struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
2434
2435         apic_timer_expired(apic, true);
2436
2437         if (lapic_is_periodic(apic)) {
2438                 advance_periodic_target_expiration(apic);
2439                 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
2440                 return HRTIMER_RESTART;
2441         } else
2442                 return HRTIMER_NORESTART;
2443 }
2444
2445 int kvm_create_lapic(struct kvm_vcpu *vcpu, int timer_advance_ns)
2446 {
2447         struct kvm_lapic *apic;
2448
2449         ASSERT(vcpu != NULL);
2450
2451         apic = kzalloc(sizeof(*apic), GFP_KERNEL_ACCOUNT);
2452         if (!apic)
2453                 goto nomem;
2454
2455         vcpu->arch.apic = apic;
2456
2457         apic->regs = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
2458         if (!apic->regs) {
2459                 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
2460                        vcpu->vcpu_id);
2461                 goto nomem_free_apic;
2462         }
2463         apic->vcpu = vcpu;
2464
2465         hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2466                      HRTIMER_MODE_ABS_HARD);
2467         apic->lapic_timer.timer.function = apic_timer_fn;
2468         if (timer_advance_ns == -1) {
2469                 apic->lapic_timer.timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
2470                 lapic_timer_advance_dynamic = true;
2471         } else {
2472                 apic->lapic_timer.timer_advance_ns = timer_advance_ns;
2473                 lapic_timer_advance_dynamic = false;
2474         }
2475
2476         /*
2477          * APIC is created enabled. This will prevent kvm_lapic_set_base from
2478          * thinking that APIC state has changed.
2479          */
2480         vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2481         static_branch_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2482         kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2483
2484         return 0;
2485 nomem_free_apic:
2486         kfree(apic);
2487         vcpu->arch.apic = NULL;
2488 nomem:
2489         return -ENOMEM;
2490 }
2491
2492 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2493 {
2494         struct kvm_lapic *apic = vcpu->arch.apic;
2495         u32 ppr;
2496
2497         if (!kvm_apic_present(vcpu))
2498                 return -1;
2499
2500         __apic_update_ppr(apic, &ppr);
2501         return apic_has_interrupt_for_ppr(apic, ppr);
2502 }
2503 EXPORT_SYMBOL_GPL(kvm_apic_has_interrupt);
2504
2505 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2506 {
2507         u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2508
2509         if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2510                 return 1;
2511         if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2512             GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2513                 return 1;
2514         return 0;
2515 }
2516
2517 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2518 {
2519         struct kvm_lapic *apic = vcpu->arch.apic;
2520
2521         if (atomic_read(&apic->lapic_timer.pending) > 0) {
2522                 kvm_apic_inject_pending_timer_irqs(apic);
2523                 atomic_set(&apic->lapic_timer.pending, 0);
2524         }
2525 }
2526
2527 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2528 {
2529         int vector = kvm_apic_has_interrupt(vcpu);
2530         struct kvm_lapic *apic = vcpu->arch.apic;
2531         u32 ppr;
2532
2533         if (vector == -1)
2534                 return -1;
2535
2536         /*
2537          * We get here even with APIC virtualization enabled, if doing
2538          * nested virtualization and L1 runs with the "acknowledge interrupt
2539          * on exit" mode.  Then we cannot inject the interrupt via RVI,
2540          * because the process would deliver it through the IDT.
2541          */
2542
2543         apic_clear_irr(vector, apic);
2544         if (to_hv_vcpu(vcpu) && test_bit(vector, to_hv_synic(vcpu)->auto_eoi_bitmap)) {
2545                 /*
2546                  * For auto-EOI interrupts, there might be another pending
2547                  * interrupt above PPR, so check whether to raise another
2548                  * KVM_REQ_EVENT.
2549                  */
2550                 apic_update_ppr(apic);
2551         } else {
2552                 /*
2553                  * For normal interrupts, PPR has been raised and there cannot
2554                  * be a higher-priority pending interrupt---except if there was
2555                  * a concurrent interrupt injection, but that would have
2556                  * triggered KVM_REQ_EVENT already.
2557                  */
2558                 apic_set_isr(vector, apic);
2559                 __apic_update_ppr(apic, &ppr);
2560         }
2561
2562         return vector;
2563 }
2564
2565 static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2566                 struct kvm_lapic_state *s, bool set)
2567 {
2568         if (apic_x2apic_mode(vcpu->arch.apic)) {
2569                 u32 *id = (u32 *)(s->regs + APIC_ID);
2570                 u32 *ldr = (u32 *)(s->regs + APIC_LDR);
2571
2572                 if (vcpu->kvm->arch.x2apic_format) {
2573                         if (*id != vcpu->vcpu_id)
2574                                 return -EINVAL;
2575                 } else {
2576                         if (set)
2577                                 *id >>= 24;
2578                         else
2579                                 *id <<= 24;
2580                 }
2581
2582                 /* In x2APIC mode, the LDR is fixed and based on the id */
2583                 if (set)
2584                         *ldr = kvm_apic_calc_x2apic_ldr(*id);
2585         }
2586
2587         return 0;
2588 }
2589
2590 int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2591 {
2592         memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2593
2594         /*
2595          * Get calculated timer current count for remaining timer period (if
2596          * any) and store it in the returned register set.
2597          */
2598         __kvm_lapic_set_reg(s->regs, APIC_TMCCT,
2599                             __apic_read(vcpu->arch.apic, APIC_TMCCT));
2600
2601         return kvm_apic_state_fixup(vcpu, s, false);
2602 }
2603
2604 int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2605 {
2606         struct kvm_lapic *apic = vcpu->arch.apic;
2607         int r;
2608
2609         kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2610         /* set SPIV separately to get count of SW disabled APICs right */
2611         apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2612
2613         r = kvm_apic_state_fixup(vcpu, s, true);
2614         if (r) {
2615                 kvm_recalculate_apic_map(vcpu->kvm);
2616                 return r;
2617         }
2618         memcpy(vcpu->arch.apic->regs, s->regs, sizeof(*s));
2619
2620         atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
2621         kvm_recalculate_apic_map(vcpu->kvm);
2622         kvm_apic_set_version(vcpu);
2623
2624         apic_update_ppr(apic);
2625         hrtimer_cancel(&apic->lapic_timer.timer);
2626         apic->lapic_timer.expired_tscdeadline = 0;
2627         apic_update_lvtt(apic);
2628         apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2629         update_divide_count(apic);
2630         __start_apic_timer(apic, APIC_TMCCT);
2631         kvm_apic_update_apicv(vcpu);
2632         apic->highest_isr_cache = -1;
2633         if (vcpu->arch.apicv_active) {
2634                 static_call(kvm_x86_apicv_post_state_restore)(vcpu);
2635                 static_call(kvm_x86_hwapic_irr_update)(vcpu,
2636                                 apic_find_highest_irr(apic));
2637                 static_call(kvm_x86_hwapic_isr_update)(vcpu,
2638                                 apic_find_highest_isr(apic));
2639         }
2640         kvm_make_request(KVM_REQ_EVENT, vcpu);
2641         if (ioapic_in_kernel(vcpu->kvm))
2642                 kvm_rtc_eoi_tracking_restore_one(vcpu);
2643
2644         vcpu->arch.apic_arb_prio = 0;
2645
2646         return 0;
2647 }
2648
2649 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2650 {
2651         struct hrtimer *timer;
2652
2653         if (!lapic_in_kernel(vcpu) ||
2654                 kvm_can_post_timer_interrupt(vcpu))
2655                 return;
2656
2657         timer = &vcpu->arch.apic->lapic_timer.timer;
2658         if (hrtimer_cancel(timer))
2659                 hrtimer_start_expires(timer, HRTIMER_MODE_ABS_HARD);
2660 }
2661
2662 /*
2663  * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2664  *
2665  * Detect whether guest triggered PV EOI since the
2666  * last entry. If yes, set EOI on guests's behalf.
2667  * Clear PV EOI in guest memory in any case.
2668  */
2669 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2670                                         struct kvm_lapic *apic)
2671 {
2672         bool pending;
2673         int vector;
2674         /*
2675          * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2676          * and KVM_PV_EOI_ENABLED in guest memory as follows:
2677          *
2678          * KVM_APIC_PV_EOI_PENDING is unset:
2679          *      -> host disabled PV EOI.
2680          * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2681          *      -> host enabled PV EOI, guest did not execute EOI yet.
2682          * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2683          *      -> host enabled PV EOI, guest executed EOI.
2684          */
2685         BUG_ON(!pv_eoi_enabled(vcpu));
2686         pending = pv_eoi_get_pending(vcpu);
2687         /*
2688          * Clear pending bit in any case: it will be set again on vmentry.
2689          * While this might not be ideal from performance point of view,
2690          * this makes sure pv eoi is only enabled when we know it's safe.
2691          */
2692         pv_eoi_clr_pending(vcpu);
2693         if (pending)
2694                 return;
2695         vector = apic_set_eoi(apic);
2696         trace_kvm_pv_eoi(apic, vector);
2697 }
2698
2699 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2700 {
2701         u32 data;
2702
2703         if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2704                 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2705
2706         if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2707                 return;
2708
2709         if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2710                                   sizeof(u32)))
2711                 return;
2712
2713         apic_set_tpr(vcpu->arch.apic, data & 0xff);
2714 }
2715
2716 /*
2717  * apic_sync_pv_eoi_to_guest - called before vmentry
2718  *
2719  * Detect whether it's safe to enable PV EOI and
2720  * if yes do so.
2721  */
2722 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2723                                         struct kvm_lapic *apic)
2724 {
2725         if (!pv_eoi_enabled(vcpu) ||
2726             /* IRR set or many bits in ISR: could be nested. */
2727             apic->irr_pending ||
2728             /* Cache not set: could be safe but we don't bother. */
2729             apic->highest_isr_cache == -1 ||
2730             /* Need EOI to update ioapic. */
2731             kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2732                 /*
2733                  * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2734                  * so we need not do anything here.
2735                  */
2736                 return;
2737         }
2738
2739         pv_eoi_set_pending(apic->vcpu);
2740 }
2741
2742 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2743 {
2744         u32 data, tpr;
2745         int max_irr, max_isr;
2746         struct kvm_lapic *apic = vcpu->arch.apic;
2747
2748         apic_sync_pv_eoi_to_guest(vcpu, apic);
2749
2750         if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2751                 return;
2752
2753         tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
2754         max_irr = apic_find_highest_irr(apic);
2755         if (max_irr < 0)
2756                 max_irr = 0;
2757         max_isr = apic_find_highest_isr(apic);
2758         if (max_isr < 0)
2759                 max_isr = 0;
2760         data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2761
2762         kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2763                                 sizeof(u32));
2764 }
2765
2766 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2767 {
2768         if (vapic_addr) {
2769                 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2770                                         &vcpu->arch.apic->vapic_cache,
2771                                         vapic_addr, sizeof(u32)))
2772                         return -EINVAL;
2773                 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2774         } else {
2775                 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2776         }
2777
2778         vcpu->arch.apic->vapic_addr = vapic_addr;
2779         return 0;
2780 }
2781
2782 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2783 {
2784         struct kvm_lapic *apic = vcpu->arch.apic;
2785         u32 reg = (msr - APIC_BASE_MSR) << 4;
2786
2787         if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2788                 return 1;
2789
2790         if (reg == APIC_ICR2)
2791                 return 1;
2792
2793         /* if this is ICR write vector before command */
2794         if (reg == APIC_ICR)
2795                 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2796         return kvm_lapic_reg_write(apic, reg, (u32)data);
2797 }
2798
2799 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2800 {
2801         struct kvm_lapic *apic = vcpu->arch.apic;
2802         u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2803
2804         if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2805                 return 1;
2806
2807         if (reg == APIC_DFR || reg == APIC_ICR2)
2808                 return 1;
2809
2810         if (kvm_lapic_reg_read(apic, reg, 4, &low))
2811                 return 1;
2812         if (reg == APIC_ICR)
2813                 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2814
2815         *data = (((u64)high) << 32) | low;
2816
2817         return 0;
2818 }
2819
2820 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2821 {
2822         struct kvm_lapic *apic = vcpu->arch.apic;
2823
2824         if (!lapic_in_kernel(vcpu))
2825                 return 1;
2826
2827         /* if this is ICR write vector before command */
2828         if (reg == APIC_ICR)
2829                 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2830         return kvm_lapic_reg_write(apic, reg, (u32)data);
2831 }
2832
2833 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2834 {
2835         struct kvm_lapic *apic = vcpu->arch.apic;
2836         u32 low, high = 0;
2837
2838         if (!lapic_in_kernel(vcpu))
2839                 return 1;
2840
2841         if (kvm_lapic_reg_read(apic, reg, 4, &low))
2842                 return 1;
2843         if (reg == APIC_ICR)
2844                 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2845
2846         *data = (((u64)high) << 32) | low;
2847
2848         return 0;
2849 }
2850
2851 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data, unsigned long len)
2852 {
2853         u64 addr = data & ~KVM_MSR_ENABLED;
2854         struct gfn_to_hva_cache *ghc = &vcpu->arch.pv_eoi.data;
2855         unsigned long new_len;
2856
2857         if (!IS_ALIGNED(addr, 4))
2858                 return 1;
2859
2860         vcpu->arch.pv_eoi.msr_val = data;
2861         if (!pv_eoi_enabled(vcpu))
2862                 return 0;
2863
2864         if (addr == ghc->gpa && len <= ghc->len)
2865                 new_len = ghc->len;
2866         else
2867                 new_len = len;
2868
2869         return kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, addr, new_len);
2870 }
2871
2872 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2873 {
2874         struct kvm_lapic *apic = vcpu->arch.apic;
2875         u8 sipi_vector;
2876         int r;
2877         unsigned long pe;
2878
2879         if (!lapic_in_kernel(vcpu))
2880                 return;
2881
2882         /*
2883          * Read pending events before calling the check_events
2884          * callback.
2885          */
2886         pe = smp_load_acquire(&apic->pending_events);
2887         if (!pe)
2888                 return;
2889
2890         if (is_guest_mode(vcpu)) {
2891                 r = kvm_check_nested_events(vcpu);
2892                 if (r < 0)
2893                         return;
2894                 /*
2895                  * If an event has happened and caused a vmexit,
2896                  * we know INITs are latched and therefore
2897                  * we will not incorrectly deliver an APIC
2898                  * event instead of a vmexit.
2899                  */
2900         }
2901
2902         /*
2903          * INITs are latched while CPU is in specific states
2904          * (SMM, VMX root mode, SVM with GIF=0).
2905          * Because a CPU cannot be in these states immediately
2906          * after it has processed an INIT signal (and thus in
2907          * KVM_MP_STATE_INIT_RECEIVED state), just eat SIPIs
2908          * and leave the INIT pending.
2909          */
2910         if (kvm_vcpu_latch_init(vcpu)) {
2911                 WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2912                 if (test_bit(KVM_APIC_SIPI, &pe))
2913                         clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2914                 return;
2915         }
2916
2917         if (test_bit(KVM_APIC_INIT, &pe)) {
2918                 clear_bit(KVM_APIC_INIT, &apic->pending_events);
2919                 kvm_vcpu_reset(vcpu, true);
2920                 if (kvm_vcpu_is_bsp(apic->vcpu))
2921                         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2922                 else
2923                         vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2924         }
2925         if (test_bit(KVM_APIC_SIPI, &pe)) {
2926                 clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2927                 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2928                         /* evaluate pending_events before reading the vector */
2929                         smp_rmb();
2930                         sipi_vector = apic->sipi_vector;
2931                         kvm_x86_ops.vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2932                         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2933                 }
2934         }
2935 }
2936
2937 void kvm_lapic_exit(void)
2938 {
2939         static_key_deferred_flush(&apic_hw_disabled);
2940         static_key_deferred_flush(&apic_sw_disabled);
2941 }