kernel/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
  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 #define APIC_BUS_CYCLE_NS 1
  57
  58 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
  59 #define apic_debug(fmt, arg...)
  60
  61 #define APIC_LVT_NUM                    6
  62 /* 14 is the version for Xeon and Pentium 8.4.8*/
  63 #define APIC_VERSION                    (0x14UL | ((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 #define VEC_POS(v) ((v) & (32 - 1))
  76 #define REG_POS(v) (((v) >> 5) << 4)
  77
  78 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
  79 {
  80         *((u32 *) (apic->regs + reg_off)) = val;
  81 }
  82
  83 static inline int apic_test_vector(int vec, void *bitmap)
  84 {
  85         return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
  86 }
  87
  88 bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
  89 {
  90         struct kvm_lapic *apic = vcpu->arch.apic;
  91
  92         return apic_test_vector(vector, apic->regs + APIC_ISR) ||
  93                 apic_test_vector(vector, apic->regs + APIC_IRR);
  94 }
  95
  96 static inline void apic_set_vector(int vec, void *bitmap)
  97 {
  98         set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
  99 }
 100
 101 static inline void apic_clear_vector(int vec, void *bitmap)
 102 {
 103         clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
 104 }
 105
 106 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
 107 {
 108         return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
 109 }
 110
 111 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
 112 {
 113         return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
 114 }
 115
 116 struct static_key_deferred apic_hw_disabled __read_mostly;
 117 struct static_key_deferred apic_sw_disabled __read_mostly;
 118
 119 static inline int apic_enabled(struct kvm_lapic *apic)
 120 {
 121         return kvm_apic_sw_enabled(apic) &&     kvm_apic_hw_enabled(apic);
 122 }
 123
 124 #define LVT_MASK        \
 125         (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
 126
 127 #define LINT_MASK       \
 128         (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
 129          APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
 130
 131 static inline int kvm_apic_id(struct kvm_lapic *apic)
 132 {
 133         return (kvm_apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
 134 }
 135
 136 /* The logical map is definitely wrong if we have multiple
 137  * modes at the same time.  (Physical map is always right.)
 138  */
 139 static inline bool kvm_apic_logical_map_valid(struct kvm_apic_map *map)
 140 {
 141         return !(map->mode & (map->mode - 1));
 142 }
 143
 144 static inline void
 145 apic_logical_id(struct kvm_apic_map *map, u32 dest_id, u16 *cid, u16 *lid)
 146 {
 147         unsigned lid_bits;
 148
 149         BUILD_BUG_ON(KVM_APIC_MODE_XAPIC_CLUSTER !=  4);
 150         BUILD_BUG_ON(KVM_APIC_MODE_XAPIC_FLAT    !=  8);
 151         BUILD_BUG_ON(KVM_APIC_MODE_X2APIC        != 16);
 152         lid_bits = map->mode;
 153
 154         *cid = dest_id >> lid_bits;
 155         *lid = dest_id & ((1 << lid_bits) - 1);
 156 }
 157
 158 static void recalculate_apic_map(struct kvm *kvm)
 159 {
 160         struct kvm_apic_map *new, *old = NULL;
 161         struct kvm_vcpu *vcpu;
 162         int i;
 163
 164         new = kzalloc(sizeof(struct kvm_apic_map), GFP_KERNEL);
 165
 166         mutex_lock(&kvm->arch.apic_map_lock);
 167
 168         if (!new)
 169                 goto out;
 170
 171         kvm_for_each_vcpu(i, vcpu, kvm) {
 172                 struct kvm_lapic *apic = vcpu->arch.apic;
 173                 u16 cid, lid;
 174                 u32 ldr, aid;
 175
 176                 if (!kvm_apic_present(vcpu))
 177                         continue;
 178
 179                 aid = kvm_apic_id(apic);
 180                 ldr = kvm_apic_get_reg(apic, APIC_LDR);
 181
 182                 if (aid < ARRAY_SIZE(new->phys_map))
 183                         new->phys_map[aid] = apic;
 184
 185                 if (apic_x2apic_mode(apic)) {
 186                         new->mode |= KVM_APIC_MODE_X2APIC;
 187                 } else if (ldr) {
 188                         ldr = GET_APIC_LOGICAL_ID(ldr);
 189                         if (kvm_apic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
 190                                 new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
 191                         else
 192                                 new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
 193                 }
 194
 195                 if (!kvm_apic_logical_map_valid(new))
 196                         continue;
 197
 198                 apic_logical_id(new, ldr, &cid, &lid);
 199
 200                 if (lid && cid < ARRAY_SIZE(new->logical_map))
 201                         new->logical_map[cid][ffs(lid) - 1] = apic;
 202         }
 203 out:
 204         old = rcu_dereference_protected(kvm->arch.apic_map,
 205                         lockdep_is_held(&kvm->arch.apic_map_lock));
 206         rcu_assign_pointer(kvm->arch.apic_map, new);
 207         mutex_unlock(&kvm->arch.apic_map_lock);
 208
 209         if (old)
 210                 kfree_rcu(old, rcu);
 211
 212         kvm_vcpu_request_scan_ioapic(kvm);
 213 }
 214
 215 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
 216 {
 217         bool enabled = val & APIC_SPIV_APIC_ENABLED;
 218
 219         apic_set_reg(apic, APIC_SPIV, val);
 220
 221         if (enabled != apic->sw_enabled) {
 222                 apic->sw_enabled = enabled;
 223                 if (enabled) {
 224                         static_key_slow_dec_deferred(&apic_sw_disabled);
 225                         recalculate_apic_map(apic->vcpu->kvm);
 226                 } else
 227                         static_key_slow_inc(&apic_sw_disabled.key);
 228         }
 229 }
 230
 231 static inline void kvm_apic_set_id(struct kvm_lapic *apic, u8 id)
 232 {
 233         apic_set_reg(apic, APIC_ID, id << 24);
 234         recalculate_apic_map(apic->vcpu->kvm);
 235 }
 236
 237 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
 238 {
 239         apic_set_reg(apic, APIC_LDR, id);
 240         recalculate_apic_map(apic->vcpu->kvm);
 241 }
 242
 243 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
 244 {
 245         return !(kvm_apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
 246 }
 247
 248 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
 249 {
 250         return kvm_apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
 251 }
 252
 253 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
 254 {
 255         return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
 256 }
 257
 258 static inline int apic_lvtt_period(struct kvm_lapic *apic)
 259 {
 260         return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
 261 }
 262
 263 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
 264 {
 265         return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
 266 }
 267
 268 static inline int apic_lvt_nmi_mode(u32 lvt_val)
 269 {
 270         return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
 271 }
 272
 273 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
 274 {
 275         struct kvm_lapic *apic = vcpu->arch.apic;
 276         struct kvm_cpuid_entry2 *feat;
 277         u32 v = APIC_VERSION;
 278
 279         if (!kvm_vcpu_has_lapic(vcpu))
 280                 return;
 281
 282         feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
 283         if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
 284                 v |= APIC_LVR_DIRECTED_EOI;
 285         apic_set_reg(apic, APIC_LVR, v);
 286 }
 287
 288 static const unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
 289         LVT_MASK ,      /* part LVTT mask, timer mode mask added at runtime */
 290         LVT_MASK | APIC_MODE_MASK,      /* LVTTHMR */
 291         LVT_MASK | APIC_MODE_MASK,      /* LVTPC */
 292         LINT_MASK, LINT_MASK,   /* LVT0-1 */
 293         LVT_MASK                /* LVTERR */
 294 };
 295
 296 static int find_highest_vector(void *bitmap)
 297 {
 298         int vec;
 299         u32 *reg;
 300
 301         for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
 302              vec >= 0; vec -= APIC_VECTORS_PER_REG) {
 303                 reg = bitmap + REG_POS(vec);
 304                 if (*reg)
 305                         return fls(*reg) - 1 + vec;
 306         }
 307
 308         return -1;
 309 }
 310
 311 static u8 count_vectors(void *bitmap)
 312 {
 313         int vec;
 314         u32 *reg;
 315         u8 count = 0;
 316
 317         for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
 318                 reg = bitmap + REG_POS(vec);
 319                 count += hweight32(*reg);
 320         }
 321
 322         return count;
 323 }
 324
 325 void __kvm_apic_update_irr(u32 *pir, void *regs)
 326 {
 327         u32 i, pir_val;
 328
 329         for (i = 0; i <= 7; i++) {
 330                 pir_val = xchg(&pir[i], 0);
 331                 if (pir_val)
 332                         *((u32 *)(regs + APIC_IRR + i * 0x10)) |= pir_val;
 333         }
 334 }
 335 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
 336
 337 void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
 338 {
 339         struct kvm_lapic *apic = vcpu->arch.apic;
 340
 341         __kvm_apic_update_irr(pir, apic->regs);
 342 }
 343 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
 344
 345 static inline void apic_set_irr(int vec, struct kvm_lapic *apic)
 346 {
 347         apic_set_vector(vec, apic->regs + APIC_IRR);
 348         /*
 349          * irr_pending must be true if any interrupt is pending; set it after
 350          * APIC_IRR to avoid race with apic_clear_irr
 351          */
 352         apic->irr_pending = true;
 353 }
 354
 355 static inline int apic_search_irr(struct kvm_lapic *apic)
 356 {
 357         return find_highest_vector(apic->regs + APIC_IRR);
 358 }
 359
 360 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
 361 {
 362         int result;
 363
 364         /*
 365          * Note that irr_pending is just a hint. It will be always
 366          * true with virtual interrupt delivery enabled.
 367          */
 368         if (!apic->irr_pending)
 369                 return -1;
 370
 371         kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
 372         result = apic_search_irr(apic);
 373         ASSERT(result == -1 || result >= 16);
 374
 375         return result;
 376 }
 377
 378 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
 379 {
 380         struct kvm_vcpu *vcpu;
 381
 382         vcpu = apic->vcpu;
 383
 384         if (unlikely(kvm_apic_vid_enabled(vcpu->kvm))) {
 385                 /* try to update RVI */
 386                 apic_clear_vector(vec, apic->regs + APIC_IRR);
 387                 kvm_make_request(KVM_REQ_EVENT, vcpu);
 388         } else {
 389                 apic->irr_pending = false;
 390                 apic_clear_vector(vec, apic->regs + APIC_IRR);
 391                 if (apic_search_irr(apic) != -1)
 392                         apic->irr_pending = true;
 393         }
 394 }
 395
 396 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
 397 {
 398         struct kvm_vcpu *vcpu;
 399
 400         if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
 401                 return;
 402
 403         vcpu = apic->vcpu;
 404
 405         /*
 406          * With APIC virtualization enabled, all caching is disabled
 407          * because the processor can modify ISR under the hood.  Instead
 408          * just set SVI.
 409          */
 410         if (unlikely(kvm_x86_ops->hwapic_isr_update))
 411                 kvm_x86_ops->hwapic_isr_update(vcpu->kvm, vec);
 412         else {
 413                 ++apic->isr_count;
 414                 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
 415                 /*
 416                  * ISR (in service register) bit is set when injecting an interrupt.
 417                  * The highest vector is injected. Thus the latest bit set matches
 418                  * the highest bit in ISR.
 419                  */
 420                 apic->highest_isr_cache = vec;
 421         }
 422 }
 423
 424 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
 425 {
 426         int result;
 427
 428         /*
 429          * Note that isr_count is always 1, and highest_isr_cache
 430          * is always -1, with APIC virtualization enabled.
 431          */
 432         if (!apic->isr_count)
 433                 return -1;
 434         if (likely(apic->highest_isr_cache != -1))
 435                 return apic->highest_isr_cache;
 436
 437         result = find_highest_vector(apic->regs + APIC_ISR);
 438         ASSERT(result == -1 || result >= 16);
 439
 440         return result;
 441 }
 442
 443 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
 444 {
 445         struct kvm_vcpu *vcpu;
 446         if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
 447                 return;
 448
 449         vcpu = apic->vcpu;
 450
 451         /*
 452          * We do get here for APIC virtualization enabled if the guest
 453          * uses the Hyper-V APIC enlightenment.  In this case we may need
 454          * to trigger a new interrupt delivery by writing the SVI field;
 455          * on the other hand isr_count and highest_isr_cache are unused
 456          * and must be left alone.
 457          */
 458         if (unlikely(kvm_x86_ops->hwapic_isr_update))
 459                 kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
 460                                                apic_find_highest_isr(apic));
 461         else {
 462                 --apic->isr_count;
 463                 BUG_ON(apic->isr_count < 0);
 464                 apic->highest_isr_cache = -1;
 465         }
 466 }
 467
 468 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
 469 {
 470         int highest_irr;
 471
 472         /* This may race with setting of irr in __apic_accept_irq() and
 473          * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
 474          * will cause vmexit immediately and the value will be recalculated
 475          * on the next vmentry.
 476          */
 477         if (!kvm_vcpu_has_lapic(vcpu))
 478                 return 0;
 479         highest_irr = apic_find_highest_irr(vcpu->arch.apic);
 480
 481         return highest_irr;
 482 }
 483
 484 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
 485                              int vector, int level, int trig_mode,
 486                              unsigned long *dest_map);
 487
 488 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
 489                 unsigned long *dest_map)
 490 {
 491         struct kvm_lapic *apic = vcpu->arch.apic;
 492
 493         return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
 494                         irq->level, irq->trig_mode, dest_map);
 495 }
 496
 497 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
 498 {
 499
 500         return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
 501                                       sizeof(val));
 502 }
 503
 504 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
 505 {
 506
 507         return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
 508                                       sizeof(*val));
 509 }
 510
 511 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
 512 {
 513         return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
 514 }
 515
 516 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
 517 {
 518         u8 val;
 519         if (pv_eoi_get_user(vcpu, &val) < 0)
 520                 apic_debug("Can't read EOI MSR value: 0x%llx\n",
 521                            (unsigned long long)vcpu->arch.pv_eoi.msr_val);
 522         return val & 0x1;
 523 }
 524
 525 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
 526 {
 527         if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
 528                 apic_debug("Can't set EOI MSR value: 0x%llx\n",
 529                            (unsigned long long)vcpu->arch.pv_eoi.msr_val);
 530                 return;
 531         }
 532         __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
 533 }
 534
 535 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
 536 {
 537         if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
 538                 apic_debug("Can't clear EOI MSR value: 0x%llx\n",
 539                            (unsigned long long)vcpu->arch.pv_eoi.msr_val);
 540                 return;
 541         }
 542         __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
 543 }
 544
 545 void kvm_apic_update_tmr(struct kvm_vcpu *vcpu, u32 *tmr)
 546 {
 547         struct kvm_lapic *apic = vcpu->arch.apic;
 548         int i;
 549
 550         for (i = 0; i < 8; i++)
 551                 apic_set_reg(apic, APIC_TMR + 0x10 * i, tmr[i]);
 552 }
 553
 554 static void apic_update_ppr(struct kvm_lapic *apic)
 555 {
 556         u32 tpr, isrv, ppr, old_ppr;
 557         int isr;
 558
 559         old_ppr = kvm_apic_get_reg(apic, APIC_PROCPRI);
 560         tpr = kvm_apic_get_reg(apic, APIC_TASKPRI);
 561         isr = apic_find_highest_isr(apic);
 562         isrv = (isr != -1) ? isr : 0;
 563
 564         if ((tpr & 0xf0) >= (isrv & 0xf0))
 565                 ppr = tpr & 0xff;
 566         else
 567                 ppr = isrv & 0xf0;
 568
 569         apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
 570                    apic, ppr, isr, isrv);
 571
 572         if (old_ppr != ppr) {
 573                 apic_set_reg(apic, APIC_PROCPRI, ppr);
 574                 if (ppr < old_ppr)
 575                         kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
 576         }
 577 }
 578
 579 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
 580 {
 581         apic_set_reg(apic, APIC_TASKPRI, tpr);
 582         apic_update_ppr(apic);
 583 }
 584
 585 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
 586 {
 587         if (apic_x2apic_mode(apic))
 588                 return mda == X2APIC_BROADCAST;
 589
 590         return GET_APIC_DEST_FIELD(mda) == APIC_BROADCAST;
 591 }
 592
 593 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
 594 {
 595         if (kvm_apic_broadcast(apic, mda))
 596                 return true;
 597
 598         if (apic_x2apic_mode(apic))
 599                 return mda == kvm_apic_id(apic);
 600
 601         return mda == SET_APIC_DEST_FIELD(kvm_apic_id(apic));
 602 }
 603
 604 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
 605 {
 606         u32 logical_id;
 607
 608         if (kvm_apic_broadcast(apic, mda))
 609                 return true;
 610
 611         logical_id = kvm_apic_get_reg(apic, APIC_LDR);
 612
 613         if (apic_x2apic_mode(apic))
 614                 return ((logical_id >> 16) == (mda >> 16))
 615                        && (logical_id & mda & 0xffff) != 0;
 616
 617         logical_id = GET_APIC_LOGICAL_ID(logical_id);
 618         mda = GET_APIC_DEST_FIELD(mda);
 619
 620         switch (kvm_apic_get_reg(apic, APIC_DFR)) {
 621         case APIC_DFR_FLAT:
 622                 return (logical_id & mda) != 0;
 623         case APIC_DFR_CLUSTER:
 624                 return ((logical_id >> 4) == (mda >> 4))
 625                        && (logical_id & mda & 0xf) != 0;
 626         default:
 627                 apic_debug("Bad DFR vcpu %d: %08x\n",
 628                            apic->vcpu->vcpu_id, kvm_apic_get_reg(apic, APIC_DFR));
 629                 return false;
 630         }
 631 }
 632
 633 /* KVM APIC implementation has two quirks
 634  *  - dest always begins at 0 while xAPIC MDA has offset 24,
 635  *  - IOxAPIC messages have to be delivered (directly) to x2APIC.
 636  */
 637 static u32 kvm_apic_mda(unsigned int dest_id, struct kvm_lapic *source,
 638                                               struct kvm_lapic *target)
 639 {
 640         bool ipi = source != NULL;
 641         bool x2apic_mda = apic_x2apic_mode(ipi ? source : target);
 642
 643         if (!ipi && dest_id == APIC_BROADCAST && x2apic_mda)
 644                 return X2APIC_BROADCAST;
 645
 646         return x2apic_mda ? dest_id : SET_APIC_DEST_FIELD(dest_id);
 647 }
 648
 649 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
 650                            int short_hand, unsigned int dest, int dest_mode)
 651 {
 652         struct kvm_lapic *target = vcpu->arch.apic;
 653         u32 mda = kvm_apic_mda(dest, source, target);
 654
 655         apic_debug("target %p, source %p, dest 0x%x, "
 656                    "dest_mode 0x%x, short_hand 0x%x\n",
 657                    target, source, dest, dest_mode, short_hand);
 658
 659         ASSERT(target);
 660         switch (short_hand) {
 661         case APIC_DEST_NOSHORT:
 662                 if (dest_mode == APIC_DEST_PHYSICAL)
 663                         return kvm_apic_match_physical_addr(target, mda);
 664                 else
 665                         return kvm_apic_match_logical_addr(target, mda);
 666         case APIC_DEST_SELF:
 667                 return target == source;
 668         case APIC_DEST_ALLINC:
 669                 return true;
 670         case APIC_DEST_ALLBUT:
 671                 return target != source;
 672         default:
 673                 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
 674                            short_hand);
 675                 return false;
 676         }
 677 }
 678
 679 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
 680                 struct kvm_lapic_irq *irq, int *r, unsigned long *dest_map)
 681 {
 682         struct kvm_apic_map *map;
 683         unsigned long bitmap = 1;
 684         struct kvm_lapic **dst;
 685         int i;
 686         bool ret, x2apic_ipi;
 687
 688         *r = -1;
 689
 690         if (irq->shorthand == APIC_DEST_SELF) {
 691                 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
 692                 return true;
 693         }
 694
 695         if (irq->shorthand)
 696                 return false;
 697
 698         x2apic_ipi = src && apic_x2apic_mode(src);
 699         if (irq->dest_id == (x2apic_ipi ? X2APIC_BROADCAST : APIC_BROADCAST))
 700                 return false;
 701
 702         ret = true;
 703         rcu_read_lock();
 704         map = rcu_dereference(kvm->arch.apic_map);
 705
 706         if (!map) {
 707                 ret = false;
 708                 goto out;
 709         }
 710
 711         if (irq->dest_mode == APIC_DEST_PHYSICAL) {
 712                 if (irq->dest_id >= ARRAY_SIZE(map->phys_map))
 713                         goto out;
 714
 715                 dst = &map->phys_map[irq->dest_id];
 716         } else {
 717                 u16 cid;
 718
 719                 if (!kvm_apic_logical_map_valid(map)) {
 720                         ret = false;
 721                         goto out;
 722                 }
 723
 724                 apic_logical_id(map, irq->dest_id, &cid, (u16 *)&bitmap);
 725
 726                 if (cid >= ARRAY_SIZE(map->logical_map))
 727                         goto out;
 728
 729                 dst = map->logical_map[cid];
 730
 731                 if (irq->delivery_mode == APIC_DM_LOWEST) {
 732                         int l = -1;
 733                         for_each_set_bit(i, &bitmap, 16) {
 734                                 if (!dst[i])
 735                                         continue;
 736                                 if (l < 0)
 737                                         l = i;
 738                                 else if (kvm_apic_compare_prio(dst[i]->vcpu, dst[l]->vcpu) < 0)
 739                                         l = i;
 740                         }
 741
 742                         bitmap = (l >= 0) ? 1 << l : 0;
 743                 }
 744         }
 745
 746         for_each_set_bit(i, &bitmap, 16) {
 747                 if (!dst[i])
 748                         continue;
 749                 if (*r < 0)
 750                         *r = 0;
 751                 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
 752         }
 753 out:
 754         rcu_read_unlock();
 755         return ret;
 756 }
 757
 758 /*
 759  * Add a pending IRQ into lapic.
 760  * Return 1 if successfully added and 0 if discarded.
 761  */
 762 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
 763                              int vector, int level, int trig_mode,
 764                              unsigned long *dest_map)
 765 {
 766         int result = 0;
 767         struct kvm_vcpu *vcpu = apic->vcpu;
 768
 769         trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
 770                                   trig_mode, vector);
 771         switch (delivery_mode) {
 772         case APIC_DM_LOWEST:
 773                 vcpu->arch.apic_arb_prio++;
 774         case APIC_DM_FIXED:
 775                 /* FIXME add logic for vcpu on reset */
 776                 if (unlikely(!apic_enabled(apic)))
 777                         break;
 778
 779                 result = 1;
 780
 781                 if (dest_map)
 782                         __set_bit(vcpu->vcpu_id, dest_map);
 783
 784                 if (kvm_x86_ops->deliver_posted_interrupt)
 785                         kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
 786                 else {
 787                         apic_set_irr(vector, apic);
 788
 789                         kvm_make_request(KVM_REQ_EVENT, vcpu);
 790                         kvm_vcpu_kick(vcpu);
 791                 }
 792                 break;
 793
 794         case APIC_DM_REMRD:
 795                 result = 1;
 796                 vcpu->arch.pv.pv_unhalted = 1;
 797                 kvm_make_request(KVM_REQ_EVENT, vcpu);
 798                 kvm_vcpu_kick(vcpu);
 799                 break;
 800
 801         case APIC_DM_SMI:
 802                 apic_debug("Ignoring guest SMI\n");
 803                 break;
 804
 805         case APIC_DM_NMI:
 806                 result = 1;
 807                 kvm_inject_nmi(vcpu);
 808                 kvm_vcpu_kick(vcpu);
 809                 break;
 810
 811         case APIC_DM_INIT:
 812                 if (!trig_mode || level) {
 813                         result = 1;
 814                         /* assumes that there are only KVM_APIC_INIT/SIPI */
 815                         apic->pending_events = (1UL << KVM_APIC_INIT);
 816                         /* make sure pending_events is visible before sending
 817                          * the request */
 818                         smp_wmb();
 819                         kvm_make_request(KVM_REQ_EVENT, vcpu);
 820                         kvm_vcpu_kick(vcpu);
 821                 } else {
 822                         apic_debug("Ignoring de-assert INIT to vcpu %d\n",
 823                                    vcpu->vcpu_id);
 824                 }
 825                 break;
 826
 827         case APIC_DM_STARTUP:
 828                 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
 829                            vcpu->vcpu_id, vector);
 830                 result = 1;
 831                 apic->sipi_vector = vector;
 832                 /* make sure sipi_vector is visible for the receiver */
 833                 smp_wmb();
 834                 set_bit(KVM_APIC_SIPI, &apic->pending_events);
 835                 kvm_make_request(KVM_REQ_EVENT, vcpu);
 836                 kvm_vcpu_kick(vcpu);
 837                 break;
 838
 839         case APIC_DM_EXTINT:
 840                 /*
 841                  * Should only be called by kvm_apic_local_deliver() with LVT0,
 842                  * before NMI watchdog was enabled. Already handled by
 843                  * kvm_apic_accept_pic_intr().
 844                  */
 845                 break;
 846
 847         default:
 848                 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
 849                        delivery_mode);
 850                 break;
 851         }
 852         return result;
 853 }
 854
 855 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
 856 {
 857         return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
 858 }
 859
 860 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
 861 {
 862         if (kvm_ioapic_handles_vector(apic->vcpu->kvm, vector)) {
 863                 int trigger_mode;
 864                 if (apic_test_vector(vector, apic->regs + APIC_TMR))
 865                         trigger_mode = IOAPIC_LEVEL_TRIG;
 866                 else
 867                         trigger_mode = IOAPIC_EDGE_TRIG;
 868                 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
 869         }
 870 }
 871
 872 static int apic_set_eoi(struct kvm_lapic *apic)
 873 {
 874         int vector = apic_find_highest_isr(apic);
 875
 876         trace_kvm_eoi(apic, vector);
 877
 878         /*
 879          * Not every write EOI will has corresponding ISR,
 880          * one example is when Kernel check timer on setup_IO_APIC
 881          */
 882         if (vector == -1)
 883                 return vector;
 884
 885         apic_clear_isr(vector, apic);
 886         apic_update_ppr(apic);
 887
 888         kvm_ioapic_send_eoi(apic, vector);
 889         kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
 890         return vector;
 891 }
 892
 893 /*
 894  * this interface assumes a trap-like exit, which has already finished
 895  * desired side effect including vISR and vPPR update.
 896  */
 897 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
 898 {
 899         struct kvm_lapic *apic = vcpu->arch.apic;
 900
 901         trace_kvm_eoi(apic, vector);
 902
 903         kvm_ioapic_send_eoi(apic, vector);
 904         kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
 905 }
 906 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
 907
 908 static void apic_send_ipi(struct kvm_lapic *apic)
 909 {
 910         u32 icr_low = kvm_apic_get_reg(apic, APIC_ICR);
 911         u32 icr_high = kvm_apic_get_reg(apic, APIC_ICR2);
 912         struct kvm_lapic_irq irq;
 913
 914         irq.vector = icr_low & APIC_VECTOR_MASK;
 915         irq.delivery_mode = icr_low & APIC_MODE_MASK;
 916         irq.dest_mode = icr_low & APIC_DEST_MASK;
 917         irq.level = icr_low & APIC_INT_ASSERT;
 918         irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
 919         irq.shorthand = icr_low & APIC_SHORT_MASK;
 920         if (apic_x2apic_mode(apic))
 921                 irq.dest_id = icr_high;
 922         else
 923                 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
 924
 925         trace_kvm_apic_ipi(icr_low, irq.dest_id);
 926
 927         apic_debug("icr_high 0x%x, icr_low 0x%x, "
 928                    "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
 929                    "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
 930                    icr_high, icr_low, irq.shorthand, irq.dest_id,
 931                    irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
 932                    irq.vector);
 933
 934         kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
 935 }
 936
 937 static u32 apic_get_tmcct(struct kvm_lapic *apic)
 938 {
 939         ktime_t remaining;
 940         s64 ns;
 941         u32 tmcct;
 942
 943         ASSERT(apic != NULL);
 944
 945         /* if initial count is 0, current count should also be 0 */
 946         if (kvm_apic_get_reg(apic, APIC_TMICT) == 0 ||
 947                 apic->lapic_timer.period == 0)
 948                 return 0;
 949
 950         remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
 951         if (ktime_to_ns(remaining) < 0)
 952                 remaining = ktime_set(0, 0);
 953
 954         ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
 955         tmcct = div64_u64(ns,
 956                          (APIC_BUS_CYCLE_NS * apic->divide_count));
 957
 958         return tmcct;
 959 }
 960
 961 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
 962 {
 963         struct kvm_vcpu *vcpu = apic->vcpu;
 964         struct kvm_run *run = vcpu->run;
 965
 966         kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
 967         run->tpr_access.rip = kvm_rip_read(vcpu);
 968         run->tpr_access.is_write = write;
 969 }
 970
 971 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
 972 {
 973         if (apic->vcpu->arch.tpr_access_reporting)
 974                 __report_tpr_access(apic, write);
 975 }
 976
 977 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
 978 {
 979         u32 val = 0;
 980
 981         if (offset >= LAPIC_MMIO_LENGTH)
 982                 return 0;
 983
 984         switch (offset) {
 985         case APIC_ID:
 986                 if (apic_x2apic_mode(apic))
 987                         val = kvm_apic_id(apic);
 988                 else
 989                         val = kvm_apic_id(apic) << 24;
 990                 break;
 991         case APIC_ARBPRI:
 992                 apic_debug("Access APIC ARBPRI register which is for P6\n");
 993                 break;
 994
 995         case APIC_TMCCT:        /* Timer CCR */
 996                 if (apic_lvtt_tscdeadline(apic))
 997                         return 0;
 998
 999                 val = apic_get_tmcct(apic);
1000                 break;
1001         case APIC_PROCPRI:
1002                 apic_update_ppr(apic);
1003                 val = kvm_apic_get_reg(apic, offset);
1004                 break;
1005         case APIC_TASKPRI:
1006                 report_tpr_access(apic, false);
1007                 /* fall thru */
1008         default:
1009                 val = kvm_apic_get_reg(apic, offset);
1010                 break;
1011         }
1012
1013         return val;
1014 }
1015
1016 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1017 {
1018         return container_of(dev, struct kvm_lapic, dev);
1019 }
1020
1021 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1022                 void *data)
1023 {
1024         unsigned char alignment = offset & 0xf;
1025         u32 result;
1026         /* this bitmask has a bit cleared for each reserved register */
1027         static const u64 rmask = 0x43ff01ffffffe70cULL;
1028
1029         if ((alignment + len) > 4) {
1030                 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1031                            offset, len);
1032                 return 1;
1033         }
1034
1035         if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1036                 apic_debug("KVM_APIC_READ: read reserved register %x\n",
1037                            offset);
1038                 return 1;
1039         }
1040
1041         result = __apic_read(apic, offset & ~0xf);
1042
1043         trace_kvm_apic_read(offset, result);
1044
1045         switch (len) {
1046         case 1:
1047         case 2:
1048         case 4:
1049                 memcpy(data, (char *)&result + alignment, len);
1050                 break;
1051         default:
1052                 printk(KERN_ERR "Local APIC read with len = %x, "
1053                        "should be 1,2, or 4 instead\n", len);
1054                 break;
1055         }
1056         return 0;
1057 }
1058
1059 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1060 {
1061         return kvm_apic_hw_enabled(apic) &&
1062             addr >= apic->base_address &&
1063             addr < apic->base_address + LAPIC_MMIO_LENGTH;
1064 }
1065
1066 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1067                            gpa_t address, int len, void *data)
1068 {
1069         struct kvm_lapic *apic = to_lapic(this);
1070         u32 offset = address - apic->base_address;
1071
1072         if (!apic_mmio_in_range(apic, address))
1073                 return -EOPNOTSUPP;
1074
1075         apic_reg_read(apic, offset, len, data);
1076
1077         return 0;
1078 }
1079
1080 static void update_divide_count(struct kvm_lapic *apic)
1081 {
1082         u32 tmp1, tmp2, tdcr;
1083
1084         tdcr = kvm_apic_get_reg(apic, APIC_TDCR);
1085         tmp1 = tdcr & 0xf;
1086         tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1087         apic->divide_count = 0x1 << (tmp2 & 0x7);
1088
1089         apic_debug("timer divide count is 0x%x\n",
1090                                    apic->divide_count);
1091 }
1092
1093 static void apic_update_lvtt(struct kvm_lapic *apic)
1094 {
1095         u32 timer_mode = kvm_apic_get_reg(apic, APIC_LVTT) &
1096                         apic->lapic_timer.timer_mode_mask;
1097
1098         if (apic->lapic_timer.timer_mode != timer_mode) {
1099                 apic->lapic_timer.timer_mode = timer_mode;
1100                 hrtimer_cancel(&apic->lapic_timer.timer);
1101         }
1102 }
1103
1104 static void apic_timer_expired(struct kvm_lapic *apic)
1105 {
1106         struct kvm_vcpu *vcpu = apic->vcpu;
1107         struct swait_head *q = &vcpu->wq;
1108         struct kvm_timer *ktimer = &apic->lapic_timer;
1109
1110         if (atomic_read(&apic->lapic_timer.pending))
1111                 return;
1112
1113         atomic_inc(&apic->lapic_timer.pending);
1114         kvm_set_pending_timer(vcpu);
1115
1116         if (swaitqueue_active(q))
1117                 swait_wake_interruptible(q);
1118
1119         if (apic_lvtt_tscdeadline(apic))
1120                 ktimer->expired_tscdeadline = ktimer->tscdeadline;
1121 }
1122
1123 /*
1124  * On APICv, this test will cause a busy wait
1125  * during a higher-priority task.
1126  */
1127
1128 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1129 {
1130         struct kvm_lapic *apic = vcpu->arch.apic;
1131         u32 reg = kvm_apic_get_reg(apic, APIC_LVTT);
1132
1133         if (kvm_apic_hw_enabled(apic)) {
1134                 int vec = reg & APIC_VECTOR_MASK;
1135                 void *bitmap = apic->regs + APIC_ISR;
1136
1137                 if (kvm_x86_ops->deliver_posted_interrupt)
1138                         bitmap = apic->regs + APIC_IRR;
1139
1140                 if (apic_test_vector(vec, bitmap))
1141                         return true;
1142         }
1143         return false;
1144 }
1145
1146 void wait_lapic_expire(struct kvm_vcpu *vcpu)
1147 {
1148         struct kvm_lapic *apic = vcpu->arch.apic;
1149         u64 guest_tsc, tsc_deadline;
1150
1151         if (!kvm_vcpu_has_lapic(vcpu))
1152                 return;
1153
1154         if (apic->lapic_timer.expired_tscdeadline == 0)
1155                 return;
1156
1157         if (!lapic_timer_int_injected(vcpu))
1158                 return;
1159
1160         tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1161         apic->lapic_timer.expired_tscdeadline = 0;
1162         guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1163         trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1164
1165         /* __delay is delay_tsc whenever the hardware has TSC, thus always.  */
1166         if (guest_tsc < tsc_deadline)
1167                 __delay(tsc_deadline - guest_tsc);
1168 }
1169
1170 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data);
1171
1172 static void __apic_timer_expired(struct hrtimer *data)
1173 {
1174         int ret, i = 0;
1175         enum hrtimer_restart r;
1176         struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1177
1178         r = apic_timer_fn(data);
1179
1180         if (r == HRTIMER_RESTART) {
1181                 do {
1182                         ret = hrtimer_start_expires(data, HRTIMER_MODE_ABS);
1183                         if (ret == -ETIME)
1184                                 hrtimer_add_expires_ns(&ktimer->timer,
1185                                                        ktimer->period);
1186                         i++;
1187                 } while (ret == -ETIME && i < 10);
1188
1189                 if (ret == -ETIME) {
1190                         printk_once(KERN_ERR "%s: failed to reprogram timer\n",
1191                                     __func__);
1192                         WARN_ON_ONCE(1);
1193                 }
1194         }
1195 }
1196
1197 static void start_apic_timer(struct kvm_lapic *apic)
1198 {
1199         int ret;
1200         ktime_t now;
1201
1202         atomic_set(&apic->lapic_timer.pending, 0);
1203
1204         if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1205                 /* lapic timer in oneshot or periodic mode */
1206                 now = apic->lapic_timer.timer.base->get_time();
1207                 apic->lapic_timer.period = (u64)kvm_apic_get_reg(apic, APIC_TMICT)
1208                             * APIC_BUS_CYCLE_NS * apic->divide_count;
1209
1210                 if (!apic->lapic_timer.period)
1211                         return;
1212                 /*
1213                  * Do not allow the guest to program periodic timers with small
1214                  * interval, since the hrtimers are not throttled by the host
1215                  * scheduler.
1216                  */
1217                 if (apic_lvtt_period(apic)) {
1218                         s64 min_period = min_timer_period_us * 1000LL;
1219
1220                         if (apic->lapic_timer.period < min_period) {
1221                                 pr_info_ratelimited(
1222                                     "kvm: vcpu %i: requested %lld ns "
1223                                     "lapic timer period limited to %lld ns\n",
1224                                     apic->vcpu->vcpu_id,
1225                                     apic->lapic_timer.period, min_period);
1226                                 apic->lapic_timer.period = min_period;
1227                         }
1228                 }
1229
1230                 ret = hrtimer_start(&apic->lapic_timer.timer,
1231                               ktime_add_ns(now, apic->lapic_timer.period),
1232                               HRTIMER_MODE_ABS);
1233                 if (ret == -ETIME)
1234                         __apic_timer_expired(&apic->lapic_timer.timer);
1235
1236                 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1237                            PRIx64 ", "
1238                            "timer initial count 0x%x, period %lldns, "
1239                            "expire @ 0x%016" PRIx64 ".\n", __func__,
1240                            APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1241                            kvm_apic_get_reg(apic, APIC_TMICT),
1242                            apic->lapic_timer.period,
1243                            ktime_to_ns(ktime_add_ns(now,
1244                                         apic->lapic_timer.period)));
1245         } else if (apic_lvtt_tscdeadline(apic)) {
1246                 /* lapic timer in tsc deadline mode */
1247                 u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1248                 u64 ns = 0;
1249                 ktime_t expire;
1250                 struct kvm_vcpu *vcpu = apic->vcpu;
1251                 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1252                 unsigned long flags;
1253
1254                 if (unlikely(!tscdeadline || !this_tsc_khz))
1255                         return;
1256
1257                 local_irq_save(flags);
1258
1259                 now = apic->lapic_timer.timer.base->get_time();
1260                 guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1261                 if (likely(tscdeadline > guest_tsc)) {
1262                         ns = (tscdeadline - guest_tsc) * 1000000ULL;
1263                         do_div(ns, this_tsc_khz);
1264                         expire = ktime_add_ns(now, ns);
1265                         expire = ktime_sub_ns(expire, lapic_timer_advance_ns);
1266                         ret = hrtimer_start(&apic->lapic_timer.timer,
1267                                       expire, HRTIMER_MODE_ABS);
1268                         if (ret == -ETIME)
1269                                 __apic_timer_expired(&apic->lapic_timer.timer);
1270                 } else
1271                         apic_timer_expired(apic);
1272
1273                 local_irq_restore(flags);
1274         }
1275 }
1276
1277 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1278 {
1279         int nmi_wd_enabled = apic_lvt_nmi_mode(kvm_apic_get_reg(apic, APIC_LVT0));
1280
1281         if (apic_lvt_nmi_mode(lvt0_val)) {
1282                 if (!nmi_wd_enabled) {
1283                         apic_debug("Receive NMI setting on APIC_LVT0 "
1284                                    "for cpu %d\n", apic->vcpu->vcpu_id);
1285                         atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1286                 }
1287         } else if (nmi_wd_enabled)
1288                 atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1289 }
1290
1291 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1292 {
1293         int ret = 0;
1294
1295         trace_kvm_apic_write(reg, val);
1296
1297         switch (reg) {
1298         case APIC_ID:           /* Local APIC ID */
1299                 if (!apic_x2apic_mode(apic))
1300                         kvm_apic_set_id(apic, val >> 24);
1301                 else
1302                         ret = 1;
1303                 break;
1304
1305         case APIC_TASKPRI:
1306                 report_tpr_access(apic, true);
1307                 apic_set_tpr(apic, val & 0xff);
1308                 break;
1309
1310         case APIC_EOI:
1311                 apic_set_eoi(apic);
1312                 break;
1313
1314         case APIC_LDR:
1315                 if (!apic_x2apic_mode(apic))
1316                         kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1317                 else
1318                         ret = 1;
1319                 break;
1320
1321         case APIC_DFR:
1322                 if (!apic_x2apic_mode(apic)) {
1323                         apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1324                         recalculate_apic_map(apic->vcpu->kvm);
1325                 } else
1326                         ret = 1;
1327                 break;
1328
1329         case APIC_SPIV: {
1330                 u32 mask = 0x3ff;
1331                 if (kvm_apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1332                         mask |= APIC_SPIV_DIRECTED_EOI;
1333                 apic_set_spiv(apic, val & mask);
1334                 if (!(val & APIC_SPIV_APIC_ENABLED)) {
1335                         int i;
1336                         u32 lvt_val;
1337
1338                         for (i = 0; i < APIC_LVT_NUM; i++) {
1339                                 lvt_val = kvm_apic_get_reg(apic,
1340                                                        APIC_LVTT + 0x10 * i);
1341                                 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
1342                                              lvt_val | APIC_LVT_MASKED);
1343                         }
1344                         apic_update_lvtt(apic);
1345                         atomic_set(&apic->lapic_timer.pending, 0);
1346
1347                 }
1348                 break;
1349         }
1350         case APIC_ICR:
1351                 /* No delay here, so we always clear the pending bit */
1352                 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1353                 apic_send_ipi(apic);
1354                 break;
1355
1356         case APIC_ICR2:
1357                 if (!apic_x2apic_mode(apic))
1358                         val &= 0xff000000;
1359                 apic_set_reg(apic, APIC_ICR2, val);
1360                 break;
1361
1362         case APIC_LVT0:
1363                 apic_manage_nmi_watchdog(apic, val);
1364         case APIC_LVTTHMR:
1365         case APIC_LVTPC:
1366         case APIC_LVT1:
1367         case APIC_LVTERR:
1368                 /* TODO: Check vector */
1369                 if (!kvm_apic_sw_enabled(apic))
1370                         val |= APIC_LVT_MASKED;
1371
1372                 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1373                 apic_set_reg(apic, reg, val);
1374
1375                 break;
1376
1377         case APIC_LVTT:
1378                 if (!kvm_apic_sw_enabled(apic))
1379                         val |= APIC_LVT_MASKED;
1380                 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1381                 apic_set_reg(apic, APIC_LVTT, val);
1382                 apic_update_lvtt(apic);
1383                 break;
1384
1385         case APIC_TMICT:
1386                 if (apic_lvtt_tscdeadline(apic))
1387                         break;
1388
1389                 hrtimer_cancel(&apic->lapic_timer.timer);
1390                 apic_set_reg(apic, APIC_TMICT, val);
1391                 start_apic_timer(apic);
1392                 break;
1393
1394         case APIC_TDCR:
1395                 if (val & 4)
1396                         apic_debug("KVM_WRITE:TDCR %x\n", val);
1397                 apic_set_reg(apic, APIC_TDCR, val);
1398                 update_divide_count(apic);
1399                 break;
1400
1401         case APIC_ESR:
1402                 if (apic_x2apic_mode(apic) && val != 0) {
1403                         apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1404                         ret = 1;
1405                 }
1406                 break;
1407
1408         case APIC_SELF_IPI:
1409                 if (apic_x2apic_mode(apic)) {
1410                         apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1411                 } else
1412                         ret = 1;
1413                 break;
1414         default:
1415                 ret = 1;
1416                 break;
1417         }
1418         if (ret)
1419                 apic_debug("Local APIC Write to read-only register %x\n", reg);
1420         return ret;
1421 }
1422
1423 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1424                             gpa_t address, int len, const void *data)
1425 {
1426         struct kvm_lapic *apic = to_lapic(this);
1427         unsigned int offset = address - apic->base_address;
1428         u32 val;
1429
1430         if (!apic_mmio_in_range(apic, address))
1431                 return -EOPNOTSUPP;
1432
1433         /*
1434          * APIC register must be aligned on 128-bits boundary.
1435          * 32/64/128 bits registers must be accessed thru 32 bits.
1436          * Refer SDM 8.4.1
1437          */
1438         if (len != 4 || (offset & 0xf)) {
1439                 /* Don't shout loud, $infamous_os would cause only noise. */
1440                 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1441                 return 0;
1442         }
1443
1444         val = *(u32*)data;
1445
1446         /* too common printing */
1447         if (offset != APIC_EOI)
1448                 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1449                            "0x%x\n", __func__, offset, len, val);
1450
1451         apic_reg_write(apic, offset & 0xff0, val);
1452
1453         return 0;
1454 }
1455
1456 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1457 {
1458         if (kvm_vcpu_has_lapic(vcpu))
1459                 apic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1460 }
1461 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1462
1463 /* emulate APIC access in a trap manner */
1464 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1465 {
1466         u32 val = 0;
1467
1468         /* hw has done the conditional check and inst decode */
1469         offset &= 0xff0;
1470
1471         apic_reg_read(vcpu->arch.apic, offset, 4, &val);
1472
1473         /* TODO: optimize to just emulate side effect w/o one more write */
1474         apic_reg_write(vcpu->arch.apic, offset, val);
1475 }
1476 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1477
1478 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1479 {
1480         struct kvm_lapic *apic = vcpu->arch.apic;
1481
1482         if (!vcpu->arch.apic)
1483                 return;
1484
1485         hrtimer_cancel(&apic->lapic_timer.timer);
1486
1487         if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1488                 static_key_slow_dec_deferred(&apic_hw_disabled);
1489
1490         if (!apic->sw_enabled)
1491                 static_key_slow_dec_deferred(&apic_sw_disabled);
1492
1493         if (apic->regs)
1494                 free_page((unsigned long)apic->regs);
1495
1496         kfree(apic);
1497 }
1498
1499 /*
1500  *----------------------------------------------------------------------
1501  * LAPIC interface
1502  *----------------------------------------------------------------------
1503  */
1504
1505 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1506 {
1507         struct kvm_lapic *apic = vcpu->arch.apic;
1508
1509         if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1510                         apic_lvtt_period(apic))
1511                 return 0;
1512
1513         return apic->lapic_timer.tscdeadline;
1514 }
1515
1516 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1517 {
1518         struct kvm_lapic *apic = vcpu->arch.apic;
1519
1520         if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1521                         apic_lvtt_period(apic))
1522                 return;
1523
1524         hrtimer_cancel(&apic->lapic_timer.timer);
1525         apic->lapic_timer.tscdeadline = data;
1526         start_apic_timer(apic);
1527 }
1528
1529 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1530 {
1531         struct kvm_lapic *apic = vcpu->arch.apic;
1532
1533         if (!kvm_vcpu_has_lapic(vcpu))
1534                 return;
1535
1536         apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1537                      | (kvm_apic_get_reg(apic, APIC_TASKPRI) & 4));
1538 }
1539
1540 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1541 {
1542         u64 tpr;
1543
1544         if (!kvm_vcpu_has_lapic(vcpu))
1545                 return 0;
1546
1547         tpr = (u64) kvm_apic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1548
1549         return (tpr & 0xf0) >> 4;
1550 }
1551
1552 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1553 {
1554         u64 old_value = vcpu->arch.apic_base;
1555         struct kvm_lapic *apic = vcpu->arch.apic;
1556
1557         if (!apic) {
1558                 value |= MSR_IA32_APICBASE_BSP;
1559                 vcpu->arch.apic_base = value;
1560                 return;
1561         }
1562
1563         vcpu->arch.apic_base = value;
1564
1565         /* update jump label if enable bit changes */
1566         if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1567                 if (value & MSR_IA32_APICBASE_ENABLE)
1568                         static_key_slow_dec_deferred(&apic_hw_disabled);
1569                 else
1570                         static_key_slow_inc(&apic_hw_disabled.key);
1571                 recalculate_apic_map(vcpu->kvm);
1572         }
1573
1574         if ((old_value ^ value) & X2APIC_ENABLE) {
1575                 if (value & X2APIC_ENABLE) {
1576                         u32 id = kvm_apic_id(apic);
1577                         u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
1578                         kvm_apic_set_ldr(apic, ldr);
1579                         kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1580                 } else
1581                         kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1582         }
1583
1584         apic->base_address = apic->vcpu->arch.apic_base &
1585                              MSR_IA32_APICBASE_BASE;
1586
1587         if ((value & MSR_IA32_APICBASE_ENABLE) &&
1588              apic->base_address != APIC_DEFAULT_PHYS_BASE)
1589                 pr_warn_once("APIC base relocation is unsupported by KVM");
1590
1591         /* with FSB delivery interrupt, we can restart APIC functionality */
1592         apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1593                    "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1594
1595 }
1596
1597 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
1598 {
1599         struct kvm_lapic *apic;
1600         int i;
1601
1602         apic_debug("%s\n", __func__);
1603
1604         ASSERT(vcpu);
1605         apic = vcpu->arch.apic;
1606         ASSERT(apic != NULL);
1607
1608         /* Stop the timer in case it's a reset to an active apic */
1609         hrtimer_cancel(&apic->lapic_timer.timer);
1610
1611         kvm_apic_set_id(apic, vcpu->vcpu_id);
1612         kvm_apic_set_version(apic->vcpu);
1613
1614         for (i = 0; i < APIC_LVT_NUM; i++)
1615                 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1616         apic_update_lvtt(apic);
1617         apic_set_reg(apic, APIC_LVT0,
1618                      SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1619
1620         apic_set_reg(apic, APIC_DFR, 0xffffffffU);
1621         apic_set_spiv(apic, 0xff);
1622         apic_set_reg(apic, APIC_TASKPRI, 0);
1623         kvm_apic_set_ldr(apic, 0);
1624         apic_set_reg(apic, APIC_ESR, 0);
1625         apic_set_reg(apic, APIC_ICR, 0);
1626         apic_set_reg(apic, APIC_ICR2, 0);
1627         apic_set_reg(apic, APIC_TDCR, 0);
1628         apic_set_reg(apic, APIC_TMICT, 0);
1629         for (i = 0; i < 8; i++) {
1630                 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1631                 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1632                 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1633         }
1634         apic->irr_pending = kvm_apic_vid_enabled(vcpu->kvm);
1635         apic->isr_count = kvm_x86_ops->hwapic_isr_update ? 1 : 0;
1636         apic->highest_isr_cache = -1;
1637         update_divide_count(apic);
1638         atomic_set(&apic->lapic_timer.pending, 0);
1639         if (kvm_vcpu_is_bsp(vcpu))
1640                 kvm_lapic_set_base(vcpu,
1641                                 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1642         vcpu->arch.pv_eoi.msr_val = 0;
1643         apic_update_ppr(apic);
1644
1645         vcpu->arch.apic_arb_prio = 0;
1646         vcpu->arch.apic_attention = 0;
1647
1648         apic_debug("%s: vcpu=%p, id=%d, base_msr="
1649                    "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1650                    vcpu, kvm_apic_id(apic),
1651                    vcpu->arch.apic_base, apic->base_address);
1652 }
1653
1654 /*
1655  *----------------------------------------------------------------------
1656  * timer interface
1657  *----------------------------------------------------------------------
1658  */
1659
1660 static bool lapic_is_periodic(struct kvm_lapic *apic)
1661 {
1662         return apic_lvtt_period(apic);
1663 }
1664
1665 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1666 {
1667         struct kvm_lapic *apic = vcpu->arch.apic;
1668
1669         if (kvm_vcpu_has_lapic(vcpu) && apic_enabled(apic) &&
1670                         apic_lvt_enabled(apic, APIC_LVTT))
1671                 return atomic_read(&apic->lapic_timer.pending);
1672
1673         return 0;
1674 }
1675
1676 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1677 {
1678         u32 reg = kvm_apic_get_reg(apic, lvt_type);
1679         int vector, mode, trig_mode;
1680
1681         if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1682                 vector = reg & APIC_VECTOR_MASK;
1683                 mode = reg & APIC_MODE_MASK;
1684                 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1685                 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
1686                                         NULL);
1687         }
1688         return 0;
1689 }
1690
1691 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1692 {
1693         struct kvm_lapic *apic = vcpu->arch.apic;
1694
1695         if (apic)
1696                 kvm_apic_local_deliver(apic, APIC_LVT0);
1697 }
1698
1699 static const struct kvm_io_device_ops apic_mmio_ops = {
1700         .read     = apic_mmio_read,
1701         .write    = apic_mmio_write,
1702 };
1703
1704 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1705 {
1706         struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1707         struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
1708
1709         apic_timer_expired(apic);
1710
1711         if (lapic_is_periodic(apic)) {
1712                 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1713                 return HRTIMER_RESTART;
1714         } else
1715                 return HRTIMER_NORESTART;
1716 }
1717
1718 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1719 {
1720         struct kvm_lapic *apic;
1721
1722         ASSERT(vcpu != NULL);
1723         apic_debug("apic_init %d\n", vcpu->vcpu_id);
1724
1725         apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1726         if (!apic)
1727                 goto nomem;
1728
1729         vcpu->arch.apic = apic;
1730
1731         apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1732         if (!apic->regs) {
1733                 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1734                        vcpu->vcpu_id);
1735                 goto nomem_free_apic;
1736         }
1737         apic->vcpu = vcpu;
1738
1739         hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1740                      HRTIMER_MODE_ABS);
1741         apic->lapic_timer.timer.function = apic_timer_fn;
1742         apic->lapic_timer.timer.irqsafe = 1;
1743
1744         /*
1745          * APIC is created enabled. This will prevent kvm_lapic_set_base from
1746          * thinking that APIC satet has changed.
1747          */
1748         vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
1749         kvm_lapic_set_base(vcpu,
1750                         APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE);
1751
1752         static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
1753         kvm_lapic_reset(vcpu);
1754         kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1755
1756         return 0;
1757 nomem_free_apic:
1758         kfree(apic);
1759 nomem:
1760         return -ENOMEM;
1761 }
1762
1763 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1764 {
1765         struct kvm_lapic *apic = vcpu->arch.apic;
1766         int highest_irr;
1767
1768         if (!kvm_vcpu_has_lapic(vcpu) || !apic_enabled(apic))
1769                 return -1;
1770
1771         apic_update_ppr(apic);
1772         highest_irr = apic_find_highest_irr(apic);
1773         if ((highest_irr == -1) ||
1774             ((highest_irr & 0xF0) <= kvm_apic_get_reg(apic, APIC_PROCPRI)))
1775                 return -1;
1776         return highest_irr;
1777 }
1778
1779 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1780 {
1781         u32 lvt0 = kvm_apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1782         int r = 0;
1783
1784         if (!kvm_apic_hw_enabled(vcpu->arch.apic))
1785                 r = 1;
1786         if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1787             GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1788                 r = 1;
1789         return r;
1790 }
1791
1792 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1793 {
1794         struct kvm_lapic *apic = vcpu->arch.apic;
1795
1796         if (!kvm_vcpu_has_lapic(vcpu))
1797                 return;
1798
1799         if (atomic_read(&apic->lapic_timer.pending) > 0) {
1800                 kvm_apic_local_deliver(apic, APIC_LVTT);
1801                 if (apic_lvtt_tscdeadline(apic))
1802                         apic->lapic_timer.tscdeadline = 0;
1803                 atomic_set(&apic->lapic_timer.pending, 0);
1804         }
1805 }
1806
1807 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1808 {
1809         int vector = kvm_apic_has_interrupt(vcpu);
1810         struct kvm_lapic *apic = vcpu->arch.apic;
1811
1812         if (vector == -1)
1813                 return -1;
1814
1815         /*
1816          * We get here even with APIC virtualization enabled, if doing
1817          * nested virtualization and L1 runs with the "acknowledge interrupt
1818          * on exit" mode.  Then we cannot inject the interrupt via RVI,
1819          * because the process would deliver it through the IDT.
1820          */
1821
1822         apic_set_isr(vector, apic);
1823         apic_update_ppr(apic);
1824         apic_clear_irr(vector, apic);
1825         return vector;
1826 }
1827
1828 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu,
1829                 struct kvm_lapic_state *s)
1830 {
1831         struct kvm_lapic *apic = vcpu->arch.apic;
1832
1833         kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
1834         /* set SPIV separately to get count of SW disabled APICs right */
1835         apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
1836         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1837         /* call kvm_apic_set_id() to put apic into apic_map */
1838         kvm_apic_set_id(apic, kvm_apic_id(apic));
1839         kvm_apic_set_version(vcpu);
1840
1841         apic_update_ppr(apic);
1842         hrtimer_cancel(&apic->lapic_timer.timer);
1843         apic_update_lvtt(apic);
1844         apic_manage_nmi_watchdog(apic, kvm_apic_get_reg(apic, APIC_LVT0));
1845         update_divide_count(apic);
1846         start_apic_timer(apic);
1847         apic->irr_pending = true;
1848         apic->isr_count = kvm_x86_ops->hwapic_isr_update ?
1849                                 1 : count_vectors(apic->regs + APIC_ISR);
1850         apic->highest_isr_cache = -1;
1851         if (kvm_x86_ops->hwapic_irr_update)
1852                 kvm_x86_ops->hwapic_irr_update(vcpu,
1853                                 apic_find_highest_irr(apic));
1854         if (unlikely(kvm_x86_ops->hwapic_isr_update))
1855                 kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
1856                                 apic_find_highest_isr(apic));
1857         kvm_make_request(KVM_REQ_EVENT, vcpu);
1858         kvm_rtc_eoi_tracking_restore_one(vcpu);
1859 }
1860
1861 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1862 {
1863         struct hrtimer *timer;
1864
1865         if (!kvm_vcpu_has_lapic(vcpu))
1866                 return;
1867
1868         timer = &vcpu->arch.apic->lapic_timer.timer;
1869         if (hrtimer_cancel(timer))
1870                 if (hrtimer_start_expires(timer, HRTIMER_MODE_ABS) == -ETIME)
1871                         __apic_timer_expired(timer);
1872 }
1873
1874 /*
1875  * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
1876  *
1877  * Detect whether guest triggered PV EOI since the
1878  * last entry. If yes, set EOI on guests's behalf.
1879  * Clear PV EOI in guest memory in any case.
1880  */
1881 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
1882                                         struct kvm_lapic *apic)
1883 {
1884         bool pending;
1885         int vector;
1886         /*
1887          * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
1888          * and KVM_PV_EOI_ENABLED in guest memory as follows:
1889          *
1890          * KVM_APIC_PV_EOI_PENDING is unset:
1891          *      -> host disabled PV EOI.
1892          * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
1893          *      -> host enabled PV EOI, guest did not execute EOI yet.
1894          * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
1895          *      -> host enabled PV EOI, guest executed EOI.
1896          */
1897         BUG_ON(!pv_eoi_enabled(vcpu));
1898         pending = pv_eoi_get_pending(vcpu);
1899         /*
1900          * Clear pending bit in any case: it will be set again on vmentry.
1901          * While this might not be ideal from performance point of view,
1902          * this makes sure pv eoi is only enabled when we know it's safe.
1903          */
1904         pv_eoi_clr_pending(vcpu);
1905         if (pending)
1906                 return;
1907         vector = apic_set_eoi(apic);
1908         trace_kvm_pv_eoi(apic, vector);
1909 }
1910
1911 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1912 {
1913         u32 data;
1914
1915         if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
1916                 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
1917
1918         if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1919                 return;
1920
1921         kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1922                                 sizeof(u32));
1923
1924         apic_set_tpr(vcpu->arch.apic, data & 0xff);
1925 }
1926
1927 /*
1928  * apic_sync_pv_eoi_to_guest - called before vmentry
1929  *
1930  * Detect whether it's safe to enable PV EOI and
1931  * if yes do so.
1932  */
1933 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
1934                                         struct kvm_lapic *apic)
1935 {
1936         if (!pv_eoi_enabled(vcpu) ||
1937             /* IRR set or many bits in ISR: could be nested. */
1938             apic->irr_pending ||
1939             /* Cache not set: could be safe but we don't bother. */
1940             apic->highest_isr_cache == -1 ||
1941             /* Need EOI to update ioapic. */
1942             kvm_ioapic_handles_vector(vcpu->kvm, apic->highest_isr_cache)) {
1943                 /*
1944                  * PV EOI was disabled by apic_sync_pv_eoi_from_guest
1945                  * so we need not do anything here.
1946                  */
1947                 return;
1948         }
1949
1950         pv_eoi_set_pending(apic->vcpu);
1951 }
1952
1953 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1954 {
1955         u32 data, tpr;
1956         int max_irr, max_isr;
1957         struct kvm_lapic *apic = vcpu->arch.apic;
1958
1959         apic_sync_pv_eoi_to_guest(vcpu, apic);
1960
1961         if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1962                 return;
1963
1964         tpr = kvm_apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1965         max_irr = apic_find_highest_irr(apic);
1966         if (max_irr < 0)
1967                 max_irr = 0;
1968         max_isr = apic_find_highest_isr(apic);
1969         if (max_isr < 0)
1970                 max_isr = 0;
1971         data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1972
1973         kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1974                                 sizeof(u32));
1975 }
1976
1977 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1978 {
1979         if (vapic_addr) {
1980                 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
1981                                         &vcpu->arch.apic->vapic_cache,
1982                                         vapic_addr, sizeof(u32)))
1983                         return -EINVAL;
1984                 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1985         } else {
1986                 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1987         }
1988
1989         vcpu->arch.apic->vapic_addr = vapic_addr;
1990         return 0;
1991 }
1992
1993 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1994 {
1995         struct kvm_lapic *apic = vcpu->arch.apic;
1996         u32 reg = (msr - APIC_BASE_MSR) << 4;
1997
1998         if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1999                 return 1;
2000
2001         if (reg == APIC_ICR2)
2002                 return 1;
2003
2004         /* if this is ICR write vector before command */
2005         if (reg == APIC_ICR)
2006                 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2007         return apic_reg_write(apic, reg, (u32)data);
2008 }
2009
2010 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2011 {
2012         struct kvm_lapic *apic = vcpu->arch.apic;
2013         u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2014
2015         if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
2016                 return 1;
2017
2018         if (reg == APIC_DFR || reg == APIC_ICR2) {
2019                 apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
2020                            reg);
2021                 return 1;
2022         }
2023
2024         if (apic_reg_read(apic, reg, 4, &low))
2025                 return 1;
2026         if (reg == APIC_ICR)
2027                 apic_reg_read(apic, APIC_ICR2, 4, &high);
2028
2029         *data = (((u64)high) << 32) | low;
2030
2031         return 0;
2032 }
2033
2034 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2035 {
2036         struct kvm_lapic *apic = vcpu->arch.apic;
2037
2038         if (!kvm_vcpu_has_lapic(vcpu))
2039                 return 1;
2040
2041         /* if this is ICR write vector before command */
2042         if (reg == APIC_ICR)
2043                 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2044         return apic_reg_write(apic, reg, (u32)data);
2045 }
2046
2047 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2048 {
2049         struct kvm_lapic *apic = vcpu->arch.apic;
2050         u32 low, high = 0;
2051
2052         if (!kvm_vcpu_has_lapic(vcpu))
2053                 return 1;
2054
2055         if (apic_reg_read(apic, reg, 4, &low))
2056                 return 1;
2057         if (reg == APIC_ICR)
2058                 apic_reg_read(apic, APIC_ICR2, 4, &high);
2059
2060         *data = (((u64)high) << 32) | low;
2061
2062         return 0;
2063 }
2064
2065 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
2066 {
2067         u64 addr = data & ~KVM_MSR_ENABLED;
2068         if (!IS_ALIGNED(addr, 4))
2069                 return 1;
2070
2071         vcpu->arch.pv_eoi.msr_val = data;
2072         if (!pv_eoi_enabled(vcpu))
2073                 return 0;
2074         return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
2075                                          addr, sizeof(u8));
2076 }
2077
2078 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2079 {
2080         struct kvm_lapic *apic = vcpu->arch.apic;
2081         u8 sipi_vector;
2082         unsigned long pe;
2083
2084         if (!kvm_vcpu_has_lapic(vcpu) || !apic->pending_events)
2085                 return;
2086
2087         pe = xchg(&apic->pending_events, 0);
2088
2089         if (test_bit(KVM_APIC_INIT, &pe)) {
2090                 kvm_lapic_reset(vcpu);
2091                 kvm_vcpu_reset(vcpu);
2092                 if (kvm_vcpu_is_bsp(apic->vcpu))
2093                         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2094                 else
2095                         vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2096         }
2097         if (test_bit(KVM_APIC_SIPI, &pe) &&
2098             vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2099                 /* evaluate pending_events before reading the vector */
2100                 smp_rmb();
2101                 sipi_vector = apic->sipi_vector;
2102                 apic_debug("vcpu %d received sipi with vector # %x\n",
2103                          vcpu->vcpu_id, sipi_vector);
2104                 kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2105                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2106         }
2107 }
2108
2109 void kvm_lapic_init(void)
2110 {
2111         /* do not patch jump label more than once per second */
2112         jump_label_rate_limit(&apic_hw_disabled, HZ);
2113         jump_label_rate_limit(&apic_sw_disabled, HZ);
2114 }