arch/x86/kvm/lapic.c

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