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 void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u8 id)
 244 {
 245         u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
 246
 247         apic_set_reg(apic, APIC_ID, id << 24);
 248         apic_set_reg(apic, APIC_LDR, ldr);
 249         recalculate_apic_map(apic->vcpu->kvm);
 250 }
 251
 252 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
 253 {
 254         return !(kvm_apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
 255 }
 256
 257 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
 258 {
 259         return kvm_apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
 260 }
 261
 262 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
 263 {
 264         return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
 265 }
 266
 267 static inline int apic_lvtt_period(struct kvm_lapic *apic)
 268 {
 269         return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
 270 }
 271
 272 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
 273 {
 274         return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
 275 }
 276
 277 static inline int apic_lvt_nmi_mode(u32 lvt_val)
 278 {
 279         return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
 280 }
 281
 282 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
 283 {
 284         struct kvm_lapic *apic = vcpu->arch.apic;
 285         struct kvm_cpuid_entry2 *feat;
 286         u32 v = APIC_VERSION;
 287
 288         if (!kvm_vcpu_has_lapic(vcpu))
 289                 return;
 290
 291         feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
 292         if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
 293                 v |= APIC_LVR_DIRECTED_EOI;
 294         apic_set_reg(apic, APIC_LVR, v);
 295 }
 296
 297 static const unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
 298         LVT_MASK ,      /* part LVTT mask, timer mode mask added at runtime */
 299         LVT_MASK | APIC_MODE_MASK,      /* LVTTHMR */
 300         LVT_MASK | APIC_MODE_MASK,      /* LVTPC */
 301         LINT_MASK, LINT_MASK,   /* LVT0-1 */
 302         LVT_MASK                /* LVTERR */
 303 };
 304
 305 static int find_highest_vector(void *bitmap)
 306 {
 307         int vec;
 308         u32 *reg;
 309
 310         for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
 311              vec >= 0; vec -= APIC_VECTORS_PER_REG) {
 312                 reg = bitmap + REG_POS(vec);
 313                 if (*reg)
 314                         return fls(*reg) - 1 + vec;
 315         }
 316
 317         return -1;
 318 }
 319
 320 static u8 count_vectors(void *bitmap)
 321 {
 322         int vec;
 323         u32 *reg;
 324         u8 count = 0;
 325
 326         for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
 327                 reg = bitmap + REG_POS(vec);
 328                 count += hweight32(*reg);
 329         }
 330
 331         return count;
 332 }
 333
 334 void __kvm_apic_update_irr(u32 *pir, void *regs)
 335 {
 336         u32 i, pir_val;
 337
 338         for (i = 0; i <= 7; i++) {
 339                 pir_val = xchg(&pir[i], 0);
 340                 if (pir_val)
 341                         *((u32 *)(regs + APIC_IRR + i * 0x10)) |= pir_val;
 342         }
 343 }
 344 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
 345
 346 void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
 347 {
 348         struct kvm_lapic *apic = vcpu->arch.apic;
 349
 350         __kvm_apic_update_irr(pir, apic->regs);
 351 }
 352 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
 353
 354 static inline void apic_set_irr(int vec, struct kvm_lapic *apic)
 355 {
 356         apic_set_vector(vec, apic->regs + APIC_IRR);
 357         /*
 358          * irr_pending must be true if any interrupt is pending; set it after
 359          * APIC_IRR to avoid race with apic_clear_irr
 360          */
 361         apic->irr_pending = true;
 362 }
 363
 364 static inline int apic_search_irr(struct kvm_lapic *apic)
 365 {
 366         return find_highest_vector(apic->regs + APIC_IRR);
 367 }
 368
 369 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
 370 {
 371         int result;
 372
 373         /*
 374          * Note that irr_pending is just a hint. It will be always
 375          * true with virtual interrupt delivery enabled.
 376          */
 377         if (!apic->irr_pending)
 378                 return -1;
 379
 380         kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
 381         result = apic_search_irr(apic);
 382         ASSERT(result == -1 || result >= 16);
 383
 384         return result;
 385 }
 386
 387 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
 388 {
 389         struct kvm_vcpu *vcpu;
 390
 391         vcpu = apic->vcpu;
 392
 393         if (unlikely(kvm_apic_vid_enabled(vcpu->kvm))) {
 394                 /* try to update RVI */
 395                 apic_clear_vector(vec, apic->regs + APIC_IRR);
 396                 kvm_make_request(KVM_REQ_EVENT, vcpu);
 397         } else {
 398                 apic->irr_pending = false;
 399                 apic_clear_vector(vec, apic->regs + APIC_IRR);
 400                 if (apic_search_irr(apic) != -1)
 401                         apic->irr_pending = true;
 402         }
 403 }
 404
 405 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
 406 {
 407         struct kvm_vcpu *vcpu;
 408
 409         if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
 410                 return;
 411
 412         vcpu = apic->vcpu;
 413
 414         /*
 415          * With APIC virtualization enabled, all caching is disabled
 416          * because the processor can modify ISR under the hood.  Instead
 417          * just set SVI.
 418          */
 419         if (unlikely(kvm_x86_ops->hwapic_isr_update))
 420                 kvm_x86_ops->hwapic_isr_update(vcpu->kvm, vec);
 421         else {
 422                 ++apic->isr_count;
 423                 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
 424                 /*
 425                  * ISR (in service register) bit is set when injecting an interrupt.
 426                  * The highest vector is injected. Thus the latest bit set matches
 427                  * the highest bit in ISR.
 428                  */
 429                 apic->highest_isr_cache = vec;
 430         }
 431 }
 432
 433 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
 434 {
 435         int result;
 436
 437         /*
 438          * Note that isr_count is always 1, and highest_isr_cache
 439          * is always -1, with APIC virtualization enabled.
 440          */
 441         if (!apic->isr_count)
 442                 return -1;
 443         if (likely(apic->highest_isr_cache != -1))
 444                 return apic->highest_isr_cache;
 445
 446         result = find_highest_vector(apic->regs + APIC_ISR);
 447         ASSERT(result == -1 || result >= 16);
 448
 449         return result;
 450 }
 451
 452 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
 453 {
 454         struct kvm_vcpu *vcpu;
 455         if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
 456                 return;
 457
 458         vcpu = apic->vcpu;
 459
 460         /*
 461          * We do get here for APIC virtualization enabled if the guest
 462          * uses the Hyper-V APIC enlightenment.  In this case we may need
 463          * to trigger a new interrupt delivery by writing the SVI field;
 464          * on the other hand isr_count and highest_isr_cache are unused
 465          * and must be left alone.
 466          */
 467         if (unlikely(kvm_x86_ops->hwapic_isr_update))
 468                 kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
 469                                                apic_find_highest_isr(apic));
 470         else {
 471                 --apic->isr_count;
 472                 BUG_ON(apic->isr_count < 0);
 473                 apic->highest_isr_cache = -1;
 474         }
 475 }
 476
 477 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
 478 {
 479         int highest_irr;
 480
 481         /* This may race with setting of irr in __apic_accept_irq() and
 482          * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
 483          * will cause vmexit immediately and the value will be recalculated
 484          * on the next vmentry.
 485          */
 486         if (!kvm_vcpu_has_lapic(vcpu))
 487                 return 0;
 488         highest_irr = apic_find_highest_irr(vcpu->arch.apic);
 489
 490         return highest_irr;
 491 }
 492
 493 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
 494                              int vector, int level, int trig_mode,
 495                              unsigned long *dest_map);
 496
 497 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
 498                 unsigned long *dest_map)
 499 {
 500         struct kvm_lapic *apic = vcpu->arch.apic;
 501
 502         return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
 503                         irq->level, irq->trig_mode, dest_map);
 504 }
 505
 506 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
 507 {
 508
 509         return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
 510                                       sizeof(val));
 511 }
 512
 513 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
 514 {
 515
 516         return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
 517                                       sizeof(*val));
 518 }
 519
 520 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
 521 {
 522         return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
 523 }
 524
 525 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
 526 {
 527         u8 val;
 528         if (pv_eoi_get_user(vcpu, &val) < 0)
 529                 apic_debug("Can't read EOI MSR value: 0x%llx\n",
 530                            (unsigned long long)vcpu->arch.pv_eoi.msr_val);
 531         return val & 0x1;
 532 }
 533
 534 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
 535 {
 536         if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
 537                 apic_debug("Can't set EOI MSR value: 0x%llx\n",
 538                            (unsigned long long)vcpu->arch.pv_eoi.msr_val);
 539                 return;
 540         }
 541         __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
 542 }
 543
 544 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
 545 {
 546         if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
 547                 apic_debug("Can't clear EOI MSR value: 0x%llx\n",
 548                            (unsigned long long)vcpu->arch.pv_eoi.msr_val);
 549                 return;
 550         }
 551         __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
 552 }
 553
 554 void kvm_apic_update_tmr(struct kvm_vcpu *vcpu, u32 *tmr)
 555 {
 556         struct kvm_lapic *apic = vcpu->arch.apic;
 557         int i;
 558
 559         for (i = 0; i < 8; i++)
 560                 apic_set_reg(apic, APIC_TMR + 0x10 * i, tmr[i]);
 561 }
 562
 563 static void apic_update_ppr(struct kvm_lapic *apic)
 564 {
 565         u32 tpr, isrv, ppr, old_ppr;
 566         int isr;
 567
 568         old_ppr = kvm_apic_get_reg(apic, APIC_PROCPRI);
 569         tpr = kvm_apic_get_reg(apic, APIC_TASKPRI);
 570         isr = apic_find_highest_isr(apic);
 571         isrv = (isr != -1) ? isr : 0;
 572
 573         if ((tpr & 0xf0) >= (isrv & 0xf0))
 574                 ppr = tpr & 0xff;
 575         else
 576                 ppr = isrv & 0xf0;
 577
 578         apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
 579                    apic, ppr, isr, isrv);
 580
 581         if (old_ppr != ppr) {
 582                 apic_set_reg(apic, APIC_PROCPRI, ppr);
 583                 if (ppr < old_ppr)
 584                         kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
 585         }
 586 }
 587
 588 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
 589 {
 590         apic_set_reg(apic, APIC_TASKPRI, tpr);
 591         apic_update_ppr(apic);
 592 }
 593
 594 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
 595 {
 596         if (apic_x2apic_mode(apic))
 597                 return mda == X2APIC_BROADCAST;
 598
 599         return GET_APIC_DEST_FIELD(mda) == APIC_BROADCAST;
 600 }
 601
 602 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
 603 {
 604         if (kvm_apic_broadcast(apic, mda))
 605                 return true;
 606
 607         if (apic_x2apic_mode(apic))
 608                 return mda == kvm_apic_id(apic);
 609
 610         return mda == SET_APIC_DEST_FIELD(kvm_apic_id(apic));
 611 }
 612
 613 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
 614 {
 615         u32 logical_id;
 616
 617         if (kvm_apic_broadcast(apic, mda))
 618                 return true;
 619
 620         logical_id = kvm_apic_get_reg(apic, APIC_LDR);
 621
 622         if (apic_x2apic_mode(apic))
 623                 return ((logical_id >> 16) == (mda >> 16))
 624                        && (logical_id & mda & 0xffff) != 0;
 625
 626         logical_id = GET_APIC_LOGICAL_ID(logical_id);
 627         mda = GET_APIC_DEST_FIELD(mda);
 628
 629         switch (kvm_apic_get_reg(apic, APIC_DFR)) {
 630         case APIC_DFR_FLAT:
 631                 return (logical_id & mda) != 0;
 632         case APIC_DFR_CLUSTER:
 633                 return ((logical_id >> 4) == (mda >> 4))
 634                        && (logical_id & mda & 0xf) != 0;
 635         default:
 636                 apic_debug("Bad DFR vcpu %d: %08x\n",
 637                            apic->vcpu->vcpu_id, kvm_apic_get_reg(apic, APIC_DFR));
 638                 return false;
 639         }
 640 }
 641
 642 /* KVM APIC implementation has two quirks
 643  *  - dest always begins at 0 while xAPIC MDA has offset 24,
 644  *  - IOxAPIC messages have to be delivered (directly) to x2APIC.
 645  */
 646 static u32 kvm_apic_mda(unsigned int dest_id, struct kvm_lapic *source,
 647                                               struct kvm_lapic *target)
 648 {
 649         bool ipi = source != NULL;
 650         bool x2apic_mda = apic_x2apic_mode(ipi ? source : target);
 651
 652         if (!ipi && dest_id == APIC_BROADCAST && x2apic_mda)
 653                 return X2APIC_BROADCAST;
 654
 655         return x2apic_mda ? dest_id : SET_APIC_DEST_FIELD(dest_id);
 656 }
 657
 658 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
 659                            int short_hand, unsigned int dest, int dest_mode)
 660 {
 661         struct kvm_lapic *target = vcpu->arch.apic;
 662         u32 mda = kvm_apic_mda(dest, source, target);
 663
 664         apic_debug("target %p, source %p, dest 0x%x, "
 665                    "dest_mode 0x%x, short_hand 0x%x\n",
 666                    target, source, dest, dest_mode, short_hand);
 667
 668         ASSERT(target);
 669         switch (short_hand) {
 670         case APIC_DEST_NOSHORT:
 671                 if (dest_mode == APIC_DEST_PHYSICAL)
 672                         return kvm_apic_match_physical_addr(target, mda);
 673                 else
 674                         return kvm_apic_match_logical_addr(target, mda);
 675         case APIC_DEST_SELF:
 676                 return target == source;
 677         case APIC_DEST_ALLINC:
 678                 return true;
 679         case APIC_DEST_ALLBUT:
 680                 return target != source;
 681         default:
 682                 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
 683                            short_hand);
 684                 return false;
 685         }
 686 }
 687
 688 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
 689                 struct kvm_lapic_irq *irq, int *r, unsigned long *dest_map)
 690 {
 691         struct kvm_apic_map *map;
 692         unsigned long bitmap = 1;
 693         struct kvm_lapic **dst;
 694         int i;
 695         bool ret, x2apic_ipi;
 696
 697         *r = -1;
 698
 699         if (irq->shorthand == APIC_DEST_SELF) {
 700                 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
 701                 return true;
 702         }
 703
 704         if (irq->shorthand)
 705                 return false;
 706
 707         x2apic_ipi = src && apic_x2apic_mode(src);
 708         if (irq->dest_id == (x2apic_ipi ? X2APIC_BROADCAST : APIC_BROADCAST))
 709                 return false;
 710
 711         ret = true;
 712         rcu_read_lock();
 713         map = rcu_dereference(kvm->arch.apic_map);
 714
 715         if (!map) {
 716                 ret = false;
 717                 goto out;
 718         }
 719
 720         if (irq->dest_mode == APIC_DEST_PHYSICAL) {
 721                 if (irq->dest_id >= ARRAY_SIZE(map->phys_map))
 722                         goto out;
 723
 724                 dst = &map->phys_map[irq->dest_id];
 725         } else {
 726                 u16 cid;
 727
 728                 if (!kvm_apic_logical_map_valid(map)) {
 729                         ret = false;
 730                         goto out;
 731                 }
 732
 733                 apic_logical_id(map, irq->dest_id, &cid, (u16 *)&bitmap);
 734
 735                 if (cid >= ARRAY_SIZE(map->logical_map))
 736                         goto out;
 737
 738                 dst = map->logical_map[cid];
 739
 740                 if (kvm_lowest_prio_delivery(irq)) {
 741                         int l = -1;
 742                         for_each_set_bit(i, &bitmap, 16) {
 743                                 if (!dst[i])
 744                                         continue;
 745                                 if (l < 0)
 746                                         l = i;
 747                                 else if (kvm_apic_compare_prio(dst[i]->vcpu, dst[l]->vcpu) < 0)
 748                                         l = i;
 749                         }
 750
 751                         bitmap = (l >= 0) ? 1 << l : 0;
 752                 }
 753         }
 754
 755         for_each_set_bit(i, &bitmap, 16) {
 756                 if (!dst[i])
 757                         continue;
 758                 if (*r < 0)
 759                         *r = 0;
 760                 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
 761         }
 762 out:
 763         rcu_read_unlock();
 764         return ret;
 765 }
 766
 767 /*
 768  * Add a pending IRQ into lapic.
 769  * Return 1 if successfully added and 0 if discarded.
 770  */
 771 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
 772                              int vector, int level, int trig_mode,
 773                              unsigned long *dest_map)
 774 {
 775         int result = 0;
 776         struct kvm_vcpu *vcpu = apic->vcpu;
 777
 778         trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
 779                                   trig_mode, vector);
 780         switch (delivery_mode) {
 781         case APIC_DM_LOWEST:
 782                 vcpu->arch.apic_arb_prio++;
 783         case APIC_DM_FIXED:
 784                 /* FIXME add logic for vcpu on reset */
 785                 if (unlikely(!apic_enabled(apic)))
 786                         break;
 787
 788                 result = 1;
 789
 790                 if (dest_map)
 791                         __set_bit(vcpu->vcpu_id, dest_map);
 792
 793                 if (kvm_x86_ops->deliver_posted_interrupt)
 794                         kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
 795                 else {
 796                         apic_set_irr(vector, apic);
 797
 798                         kvm_make_request(KVM_REQ_EVENT, vcpu);
 799                         kvm_vcpu_kick(vcpu);
 800                 }
 801                 break;
 802
 803         case APIC_DM_REMRD:
 804                 result = 1;
 805                 vcpu->arch.pv.pv_unhalted = 1;
 806                 kvm_make_request(KVM_REQ_EVENT, vcpu);
 807                 kvm_vcpu_kick(vcpu);
 808                 break;
 809
 810         case APIC_DM_SMI:
 811                 result = 1;
 812                 kvm_make_request(KVM_REQ_SMI, vcpu);
 813                 kvm_vcpu_kick(vcpu);
 814                 break;
 815
 816         case APIC_DM_NMI:
 817                 result = 1;
 818                 kvm_inject_nmi(vcpu);
 819                 kvm_vcpu_kick(vcpu);
 820                 break;
 821
 822         case APIC_DM_INIT:
 823                 if (!trig_mode || level) {
 824                         result = 1;
 825                         /* assumes that there are only KVM_APIC_INIT/SIPI */
 826                         apic->pending_events = (1UL << KVM_APIC_INIT);
 827                         /* make sure pending_events is visible before sending
 828                          * the request */
 829                         smp_wmb();
 830                         kvm_make_request(KVM_REQ_EVENT, vcpu);
 831                         kvm_vcpu_kick(vcpu);
 832                 } else {
 833                         apic_debug("Ignoring de-assert INIT to vcpu %d\n",
 834                                    vcpu->vcpu_id);
 835                 }
 836                 break;
 837
 838         case APIC_DM_STARTUP:
 839                 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
 840                            vcpu->vcpu_id, vector);
 841                 result = 1;
 842                 apic->sipi_vector = vector;
 843                 /* make sure sipi_vector is visible for the receiver */
 844                 smp_wmb();
 845                 set_bit(KVM_APIC_SIPI, &apic->pending_events);
 846                 kvm_make_request(KVM_REQ_EVENT, vcpu);
 847                 kvm_vcpu_kick(vcpu);
 848                 break;
 849
 850         case APIC_DM_EXTINT:
 851                 /*
 852                  * Should only be called by kvm_apic_local_deliver() with LVT0,
 853                  * before NMI watchdog was enabled. Already handled by
 854                  * kvm_apic_accept_pic_intr().
 855                  */
 856                 break;
 857
 858         default:
 859                 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
 860                        delivery_mode);
 861                 break;
 862         }
 863         return result;
 864 }
 865
 866 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
 867 {
 868         return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
 869 }
 870
 871 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
 872 {
 873         if (kvm_ioapic_handles_vector(apic->vcpu->kvm, vector)) {
 874                 int trigger_mode;
 875                 if (apic_test_vector(vector, apic->regs + APIC_TMR))
 876                         trigger_mode = IOAPIC_LEVEL_TRIG;
 877                 else
 878                         trigger_mode = IOAPIC_EDGE_TRIG;
 879                 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
 880         }
 881 }
 882
 883 static int apic_set_eoi(struct kvm_lapic *apic)
 884 {
 885         int vector = apic_find_highest_isr(apic);
 886
 887         trace_kvm_eoi(apic, vector);
 888
 889         /*
 890          * Not every write EOI will has corresponding ISR,
 891          * one example is when Kernel check timer on setup_IO_APIC
 892          */
 893         if (vector == -1)
 894                 return vector;
 895
 896         apic_clear_isr(vector, apic);
 897         apic_update_ppr(apic);
 898
 899         kvm_ioapic_send_eoi(apic, vector);
 900         kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
 901         return vector;
 902 }
 903
 904 /*
 905  * this interface assumes a trap-like exit, which has already finished
 906  * desired side effect including vISR and vPPR update.
 907  */
 908 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
 909 {
 910         struct kvm_lapic *apic = vcpu->arch.apic;
 911
 912         trace_kvm_eoi(apic, vector);
 913
 914         kvm_ioapic_send_eoi(apic, vector);
 915         kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
 916 }
 917 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
 918
 919 static void apic_send_ipi(struct kvm_lapic *apic)
 920 {
 921         u32 icr_low = kvm_apic_get_reg(apic, APIC_ICR);
 922         u32 icr_high = kvm_apic_get_reg(apic, APIC_ICR2);
 923         struct kvm_lapic_irq irq;
 924
 925         irq.vector = icr_low & APIC_VECTOR_MASK;
 926         irq.delivery_mode = icr_low & APIC_MODE_MASK;
 927         irq.dest_mode = icr_low & APIC_DEST_MASK;
 928         irq.level = (icr_low & APIC_INT_ASSERT) != 0;
 929         irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
 930         irq.shorthand = icr_low & APIC_SHORT_MASK;
 931         irq.msi_redir_hint = false;
 932         if (apic_x2apic_mode(apic))
 933                 irq.dest_id = icr_high;
 934         else
 935                 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
 936
 937         trace_kvm_apic_ipi(icr_low, irq.dest_id);
 938
 939         apic_debug("icr_high 0x%x, icr_low 0x%x, "
 940                    "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
 941                    "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x, "
 942                    "msi_redir_hint 0x%x\n",
 943                    icr_high, icr_low, irq.shorthand, irq.dest_id,
 944                    irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
 945                    irq.vector, irq.msi_redir_hint);
 946
 947         kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
 948 }
 949
 950 static u32 apic_get_tmcct(struct kvm_lapic *apic)
 951 {
 952         ktime_t remaining;
 953         s64 ns;
 954         u32 tmcct;
 955
 956         ASSERT(apic != NULL);
 957
 958         /* if initial count is 0, current count should also be 0 */
 959         if (kvm_apic_get_reg(apic, APIC_TMICT) == 0 ||
 960                 apic->lapic_timer.period == 0)
 961                 return 0;
 962
 963         remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
 964         if (ktime_to_ns(remaining) < 0)
 965                 remaining = ktime_set(0, 0);
 966
 967         ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
 968         tmcct = div64_u64(ns,
 969                          (APIC_BUS_CYCLE_NS * apic->divide_count));
 970
 971         return tmcct;
 972 }
 973
 974 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
 975 {
 976         struct kvm_vcpu *vcpu = apic->vcpu;
 977         struct kvm_run *run = vcpu->run;
 978
 979         kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
 980         run->tpr_access.rip = kvm_rip_read(vcpu);
 981         run->tpr_access.is_write = write;
 982 }
 983
 984 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
 985 {
 986         if (apic->vcpu->arch.tpr_access_reporting)
 987                 __report_tpr_access(apic, write);
 988 }
 989
 990 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
 991 {
 992         u32 val = 0;
 993
 994         if (offset >= LAPIC_MMIO_LENGTH)
 995                 return 0;
 996
 997         switch (offset) {
 998         case APIC_ID:
 999                 if (apic_x2apic_mode(apic))
1000                         val = kvm_apic_id(apic);
1001                 else
1002                         val = kvm_apic_id(apic) << 24;
1003                 break;
1004         case APIC_ARBPRI:
1005                 apic_debug("Access APIC ARBPRI register which is for P6\n");
1006                 break;
1007
1008         case APIC_TMCCT:        /* Timer CCR */
1009                 if (apic_lvtt_tscdeadline(apic))
1010                         return 0;
1011
1012                 val = apic_get_tmcct(apic);
1013                 break;
1014         case APIC_PROCPRI:
1015                 apic_update_ppr(apic);
1016                 val = kvm_apic_get_reg(apic, offset);
1017                 break;
1018         case APIC_TASKPRI:
1019                 report_tpr_access(apic, false);
1020                 /* fall thru */
1021         default:
1022                 val = kvm_apic_get_reg(apic, offset);
1023                 break;
1024         }
1025
1026         return val;
1027 }
1028
1029 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1030 {
1031         return container_of(dev, struct kvm_lapic, dev);
1032 }
1033
1034 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1035                 void *data)
1036 {
1037         unsigned char alignment = offset & 0xf;
1038         u32 result;
1039         /* this bitmask has a bit cleared for each reserved register */
1040         static const u64 rmask = 0x43ff01ffffffe70cULL;
1041
1042         if ((alignment + len) > 4) {
1043                 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1044                            offset, len);
1045                 return 1;
1046         }
1047
1048         if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1049                 apic_debug("KVM_APIC_READ: read reserved register %x\n",
1050                            offset);
1051                 return 1;
1052         }
1053
1054         result = __apic_read(apic, offset & ~0xf);
1055
1056         trace_kvm_apic_read(offset, result);
1057
1058         switch (len) {
1059         case 1:
1060         case 2:
1061         case 4:
1062                 memcpy(data, (char *)&result + alignment, len);
1063                 break;
1064         default:
1065                 printk(KERN_ERR "Local APIC read with len = %x, "
1066                        "should be 1,2, or 4 instead\n", len);
1067                 break;
1068         }
1069         return 0;
1070 }
1071
1072 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1073 {
1074         return kvm_apic_hw_enabled(apic) &&
1075             addr >= apic->base_address &&
1076             addr < apic->base_address + LAPIC_MMIO_LENGTH;
1077 }
1078
1079 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1080                            gpa_t address, int len, void *data)
1081 {
1082         struct kvm_lapic *apic = to_lapic(this);
1083         u32 offset = address - apic->base_address;
1084
1085         if (!apic_mmio_in_range(apic, address))
1086                 return -EOPNOTSUPP;
1087
1088         apic_reg_read(apic, offset, len, data);
1089
1090         return 0;
1091 }
1092
1093 static void update_divide_count(struct kvm_lapic *apic)
1094 {
1095         u32 tmp1, tmp2, tdcr;
1096
1097         tdcr = kvm_apic_get_reg(apic, APIC_TDCR);
1098         tmp1 = tdcr & 0xf;
1099         tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1100         apic->divide_count = 0x1 << (tmp2 & 0x7);
1101
1102         apic_debug("timer divide count is 0x%x\n",
1103                                    apic->divide_count);
1104 }
1105
1106 static void apic_update_lvtt(struct kvm_lapic *apic)
1107 {
1108         u32 timer_mode = kvm_apic_get_reg(apic, APIC_LVTT) &
1109                         apic->lapic_timer.timer_mode_mask;
1110
1111         if (apic->lapic_timer.timer_mode != timer_mode) {
1112                 apic->lapic_timer.timer_mode = timer_mode;
1113                 hrtimer_cancel(&apic->lapic_timer.timer);
1114         }
1115 }
1116
1117 static void apic_timer_expired(struct kvm_lapic *apic)
1118 {
1119         struct kvm_vcpu *vcpu = apic->vcpu;
1120         wait_queue_head_t *q = &vcpu->wq;
1121         struct kvm_timer *ktimer = &apic->lapic_timer;
1122
1123         if (atomic_read(&apic->lapic_timer.pending))
1124                 return;
1125
1126         atomic_inc(&apic->lapic_timer.pending);
1127         kvm_set_pending_timer(vcpu);
1128
1129         if (waitqueue_active(q))
1130                 wake_up_interruptible(q);
1131
1132         if (apic_lvtt_tscdeadline(apic))
1133                 ktimer->expired_tscdeadline = ktimer->tscdeadline;
1134 }
1135
1136 /*
1137  * On APICv, this test will cause a busy wait
1138  * during a higher-priority task.
1139  */
1140
1141 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1142 {
1143         struct kvm_lapic *apic = vcpu->arch.apic;
1144         u32 reg = kvm_apic_get_reg(apic, APIC_LVTT);
1145
1146         if (kvm_apic_hw_enabled(apic)) {
1147                 int vec = reg & APIC_VECTOR_MASK;
1148                 void *bitmap = apic->regs + APIC_ISR;
1149
1150                 if (kvm_x86_ops->deliver_posted_interrupt)
1151                         bitmap = apic->regs + APIC_IRR;
1152
1153                 if (apic_test_vector(vec, bitmap))
1154                         return true;
1155         }
1156         return false;
1157 }
1158
1159 void wait_lapic_expire(struct kvm_vcpu *vcpu)
1160 {
1161         struct kvm_lapic *apic = vcpu->arch.apic;
1162         u64 guest_tsc, tsc_deadline;
1163
1164         if (!kvm_vcpu_has_lapic(vcpu))
1165                 return;
1166
1167         if (apic->lapic_timer.expired_tscdeadline == 0)
1168                 return;
1169
1170         if (!lapic_timer_int_injected(vcpu))
1171                 return;
1172
1173         tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1174         apic->lapic_timer.expired_tscdeadline = 0;
1175         guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1176         trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1177
1178         /* __delay is delay_tsc whenever the hardware has TSC, thus always.  */
1179         if (guest_tsc < tsc_deadline)
1180                 __delay(tsc_deadline - guest_tsc);
1181 }
1182
1183 static void start_apic_timer(struct kvm_lapic *apic)
1184 {
1185         ktime_t now;
1186
1187         atomic_set(&apic->lapic_timer.pending, 0);
1188
1189         if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1190                 /* lapic timer in oneshot or periodic mode */
1191                 now = apic->lapic_timer.timer.base->get_time();
1192                 apic->lapic_timer.period = (u64)kvm_apic_get_reg(apic, APIC_TMICT)
1193                             * APIC_BUS_CYCLE_NS * apic->divide_count;
1194
1195                 if (!apic->lapic_timer.period)
1196                         return;
1197                 /*
1198                  * Do not allow the guest to program periodic timers with small
1199                  * interval, since the hrtimers are not throttled by the host
1200                  * scheduler.
1201                  */
1202                 if (apic_lvtt_period(apic)) {
1203                         s64 min_period = min_timer_period_us * 1000LL;
1204
1205                         if (apic->lapic_timer.period < min_period) {
1206                                 pr_info_ratelimited(
1207                                     "kvm: vcpu %i: requested %lld ns "
1208                                     "lapic timer period limited to %lld ns\n",
1209                                     apic->vcpu->vcpu_id,
1210                                     apic->lapic_timer.period, min_period);
1211                                 apic->lapic_timer.period = min_period;
1212                         }
1213                 }
1214
1215                 hrtimer_start(&apic->lapic_timer.timer,
1216                               ktime_add_ns(now, apic->lapic_timer.period),
1217                               HRTIMER_MODE_ABS);
1218
1219                 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1220                            PRIx64 ", "
1221                            "timer initial count 0x%x, period %lldns, "
1222                            "expire @ 0x%016" PRIx64 ".\n", __func__,
1223                            APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1224                            kvm_apic_get_reg(apic, APIC_TMICT),
1225                            apic->lapic_timer.period,
1226                            ktime_to_ns(ktime_add_ns(now,
1227                                         apic->lapic_timer.period)));
1228         } else if (apic_lvtt_tscdeadline(apic)) {
1229                 /* lapic timer in tsc deadline mode */
1230                 u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1231                 u64 ns = 0;
1232                 ktime_t expire;
1233                 struct kvm_vcpu *vcpu = apic->vcpu;
1234                 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1235                 unsigned long flags;
1236
1237                 if (unlikely(!tscdeadline || !this_tsc_khz))
1238                         return;
1239
1240                 local_irq_save(flags);
1241
1242                 now = apic->lapic_timer.timer.base->get_time();
1243                 guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1244                 if (likely(tscdeadline > guest_tsc)) {
1245                         ns = (tscdeadline - guest_tsc) * 1000000ULL;
1246                         do_div(ns, this_tsc_khz);
1247                         expire = ktime_add_ns(now, ns);
1248                         expire = ktime_sub_ns(expire, lapic_timer_advance_ns);
1249                         hrtimer_start(&apic->lapic_timer.timer,
1250                                       expire, HRTIMER_MODE_ABS);
1251                 } else
1252                         apic_timer_expired(apic);
1253
1254                 local_irq_restore(flags);
1255         }
1256 }
1257
1258 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1259 {
1260         bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
1261
1262         if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1263                 apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1264                 if (lvt0_in_nmi_mode) {
1265                         apic_debug("Receive NMI setting on APIC_LVT0 "
1266                                    "for cpu %d\n", apic->vcpu->vcpu_id);
1267                         atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1268                 } else
1269                         atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1270         }
1271 }
1272
1273 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1274 {
1275         int ret = 0;
1276
1277         trace_kvm_apic_write(reg, val);
1278
1279         switch (reg) {
1280         case APIC_ID:           /* Local APIC ID */
1281                 if (!apic_x2apic_mode(apic))
1282                         kvm_apic_set_id(apic, val >> 24);
1283                 else
1284                         ret = 1;
1285                 break;
1286
1287         case APIC_TASKPRI:
1288                 report_tpr_access(apic, true);
1289                 apic_set_tpr(apic, val & 0xff);
1290                 break;
1291
1292         case APIC_EOI:
1293                 apic_set_eoi(apic);
1294                 break;
1295
1296         case APIC_LDR:
1297                 if (!apic_x2apic_mode(apic))
1298                         kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1299                 else
1300                         ret = 1;
1301                 break;
1302
1303         case APIC_DFR:
1304                 if (!apic_x2apic_mode(apic)) {
1305                         apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1306                         recalculate_apic_map(apic->vcpu->kvm);
1307                 } else
1308                         ret = 1;
1309                 break;
1310
1311         case APIC_SPIV: {
1312                 u32 mask = 0x3ff;
1313                 if (kvm_apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1314                         mask |= APIC_SPIV_DIRECTED_EOI;
1315                 apic_set_spiv(apic, val & mask);
1316                 if (!(val & APIC_SPIV_APIC_ENABLED)) {
1317                         int i;
1318                         u32 lvt_val;
1319
1320                         for (i = 0; i < APIC_LVT_NUM; i++) {
1321                                 lvt_val = kvm_apic_get_reg(apic,
1322                                                        APIC_LVTT + 0x10 * i);
1323                                 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
1324                                              lvt_val | APIC_LVT_MASKED);
1325                         }
1326                         apic_update_lvtt(apic);
1327                         atomic_set(&apic->lapic_timer.pending, 0);
1328
1329                 }
1330                 break;
1331         }
1332         case APIC_ICR:
1333                 /* No delay here, so we always clear the pending bit */
1334                 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1335                 apic_send_ipi(apic);
1336                 break;
1337
1338         case APIC_ICR2:
1339                 if (!apic_x2apic_mode(apic))
1340                         val &= 0xff000000;
1341                 apic_set_reg(apic, APIC_ICR2, val);
1342                 break;
1343
1344         case APIC_LVT0:
1345                 apic_manage_nmi_watchdog(apic, val);
1346         case APIC_LVTTHMR:
1347         case APIC_LVTPC:
1348         case APIC_LVT1:
1349         case APIC_LVTERR:
1350                 /* TODO: Check vector */
1351                 if (!kvm_apic_sw_enabled(apic))
1352                         val |= APIC_LVT_MASKED;
1353
1354                 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1355                 apic_set_reg(apic, reg, val);
1356
1357                 break;
1358
1359         case APIC_LVTT:
1360                 if (!kvm_apic_sw_enabled(apic))
1361                         val |= APIC_LVT_MASKED;
1362                 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1363                 apic_set_reg(apic, APIC_LVTT, val);
1364                 apic_update_lvtt(apic);
1365                 break;
1366
1367         case APIC_TMICT:
1368                 if (apic_lvtt_tscdeadline(apic))
1369                         break;
1370
1371                 hrtimer_cancel(&apic->lapic_timer.timer);
1372                 apic_set_reg(apic, APIC_TMICT, val);
1373                 start_apic_timer(apic);
1374                 break;
1375
1376         case APIC_TDCR:
1377                 if (val & 4)
1378                         apic_debug("KVM_WRITE:TDCR %x\n", val);
1379                 apic_set_reg(apic, APIC_TDCR, val);
1380                 update_divide_count(apic);
1381                 break;
1382
1383         case APIC_ESR:
1384                 if (apic_x2apic_mode(apic) && val != 0) {
1385                         apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1386                         ret = 1;
1387                 }
1388                 break;
1389
1390         case APIC_SELF_IPI:
1391                 if (apic_x2apic_mode(apic)) {
1392                         apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1393                 } else
1394                         ret = 1;
1395                 break;
1396         default:
1397                 ret = 1;
1398                 break;
1399         }
1400         if (ret)
1401                 apic_debug("Local APIC Write to read-only register %x\n", reg);
1402         return ret;
1403 }
1404
1405 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1406                             gpa_t address, int len, const void *data)
1407 {
1408         struct kvm_lapic *apic = to_lapic(this);
1409         unsigned int offset = address - apic->base_address;
1410         u32 val;
1411
1412         if (!apic_mmio_in_range(apic, address))
1413                 return -EOPNOTSUPP;
1414
1415         /*
1416          * APIC register must be aligned on 128-bits boundary.
1417          * 32/64/128 bits registers must be accessed thru 32 bits.
1418          * Refer SDM 8.4.1
1419          */
1420         if (len != 4 || (offset & 0xf)) {
1421                 /* Don't shout loud, $infamous_os would cause only noise. */
1422                 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1423                 return 0;
1424         }
1425
1426         val = *(u32*)data;
1427
1428         /* too common printing */
1429         if (offset != APIC_EOI)
1430                 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1431                            "0x%x\n", __func__, offset, len, val);
1432
1433         apic_reg_write(apic, offset & 0xff0, val);
1434
1435         return 0;
1436 }
1437
1438 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1439 {
1440         if (kvm_vcpu_has_lapic(vcpu))
1441                 apic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1442 }
1443 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1444
1445 /* emulate APIC access in a trap manner */
1446 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1447 {
1448         u32 val = 0;
1449
1450         /* hw has done the conditional check and inst decode */
1451         offset &= 0xff0;
1452
1453         apic_reg_read(vcpu->arch.apic, offset, 4, &val);
1454
1455         /* TODO: optimize to just emulate side effect w/o one more write */
1456         apic_reg_write(vcpu->arch.apic, offset, val);
1457 }
1458 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1459
1460 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1461 {
1462         struct kvm_lapic *apic = vcpu->arch.apic;
1463
1464         if (!vcpu->arch.apic)
1465                 return;
1466
1467         hrtimer_cancel(&apic->lapic_timer.timer);
1468
1469         if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1470                 static_key_slow_dec_deferred(&apic_hw_disabled);
1471
1472         if (!apic->sw_enabled)
1473                 static_key_slow_dec_deferred(&apic_sw_disabled);
1474
1475         if (apic->regs)
1476                 free_page((unsigned long)apic->regs);
1477
1478         kfree(apic);
1479 }
1480
1481 /*
1482  *----------------------------------------------------------------------
1483  * LAPIC interface
1484  *----------------------------------------------------------------------
1485  */
1486
1487 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1488 {
1489         struct kvm_lapic *apic = vcpu->arch.apic;
1490
1491         if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1492                         apic_lvtt_period(apic))
1493                 return 0;
1494
1495         return apic->lapic_timer.tscdeadline;
1496 }
1497
1498 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1499 {
1500         struct kvm_lapic *apic = vcpu->arch.apic;
1501
1502         if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1503                         apic_lvtt_period(apic))
1504                 return;
1505
1506         hrtimer_cancel(&apic->lapic_timer.timer);
1507         apic->lapic_timer.tscdeadline = data;
1508         start_apic_timer(apic);
1509 }
1510
1511 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1512 {
1513         struct kvm_lapic *apic = vcpu->arch.apic;
1514
1515         if (!kvm_vcpu_has_lapic(vcpu))
1516                 return;
1517
1518         apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1519                      | (kvm_apic_get_reg(apic, APIC_TASKPRI) & 4));
1520 }
1521
1522 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1523 {
1524         u64 tpr;
1525
1526         if (!kvm_vcpu_has_lapic(vcpu))
1527                 return 0;
1528
1529         tpr = (u64) kvm_apic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1530
1531         return (tpr & 0xf0) >> 4;
1532 }
1533
1534 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1535 {
1536         u64 old_value = vcpu->arch.apic_base;
1537         struct kvm_lapic *apic = vcpu->arch.apic;
1538
1539         if (!apic) {
1540                 value |= MSR_IA32_APICBASE_BSP;
1541                 vcpu->arch.apic_base = value;
1542                 return;
1543         }
1544
1545         vcpu->arch.apic_base = value;
1546
1547         /* update jump label if enable bit changes */
1548         if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1549                 if (value & MSR_IA32_APICBASE_ENABLE)
1550                         static_key_slow_dec_deferred(&apic_hw_disabled);
1551                 else
1552                         static_key_slow_inc(&apic_hw_disabled.key);
1553                 recalculate_apic_map(vcpu->kvm);
1554         }
1555
1556         if ((old_value ^ value) & X2APIC_ENABLE) {
1557                 if (value & X2APIC_ENABLE) {
1558                         kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
1559                         kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1560                 } else
1561                         kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1562         }
1563
1564         apic->base_address = apic->vcpu->arch.apic_base &
1565                              MSR_IA32_APICBASE_BASE;
1566
1567         if ((value & MSR_IA32_APICBASE_ENABLE) &&
1568              apic->base_address != APIC_DEFAULT_PHYS_BASE)
1569                 pr_warn_once("APIC base relocation is unsupported by KVM");
1570
1571         /* with FSB delivery interrupt, we can restart APIC functionality */
1572         apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1573                    "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1574
1575 }
1576
1577 void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
1578 {
1579         struct kvm_lapic *apic;
1580         int i;
1581
1582         apic_debug("%s\n", __func__);
1583
1584         ASSERT(vcpu);
1585         apic = vcpu->arch.apic;
1586         ASSERT(apic != NULL);
1587
1588         /* Stop the timer in case it's a reset to an active apic */
1589         hrtimer_cancel(&apic->lapic_timer.timer);
1590
1591         if (!init_event)
1592                 kvm_apic_set_id(apic, vcpu->vcpu_id);
1593         kvm_apic_set_version(apic->vcpu);
1594
1595         for (i = 0; i < APIC_LVT_NUM; i++)
1596                 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1597         apic_update_lvtt(apic);
1598         if (!(vcpu->kvm->arch.disabled_quirks & KVM_QUIRK_LINT0_REENABLED))
1599                 apic_set_reg(apic, APIC_LVT0,
1600                              SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1601         apic_manage_nmi_watchdog(apic, kvm_apic_get_reg(apic, APIC_LVT0));
1602
1603         apic_set_reg(apic, APIC_DFR, 0xffffffffU);
1604         apic_set_spiv(apic, 0xff);
1605         apic_set_reg(apic, APIC_TASKPRI, 0);
1606         if (!apic_x2apic_mode(apic))
1607                 kvm_apic_set_ldr(apic, 0);
1608         apic_set_reg(apic, APIC_ESR, 0);
1609         apic_set_reg(apic, APIC_ICR, 0);
1610         apic_set_reg(apic, APIC_ICR2, 0);
1611         apic_set_reg(apic, APIC_TDCR, 0);
1612         apic_set_reg(apic, APIC_TMICT, 0);
1613         for (i = 0; i < 8; i++) {
1614                 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1615                 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1616                 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1617         }
1618         apic->irr_pending = kvm_apic_vid_enabled(vcpu->kvm);
1619         apic->isr_count = kvm_x86_ops->hwapic_isr_update ? 1 : 0;
1620         apic->highest_isr_cache = -1;
1621         update_divide_count(apic);
1622         atomic_set(&apic->lapic_timer.pending, 0);
1623         if (kvm_vcpu_is_bsp(vcpu))
1624                 kvm_lapic_set_base(vcpu,
1625                                 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1626         vcpu->arch.pv_eoi.msr_val = 0;
1627         apic_update_ppr(apic);
1628
1629         vcpu->arch.apic_arb_prio = 0;
1630         vcpu->arch.apic_attention = 0;
1631
1632         apic_debug("%s: vcpu=%p, id=%d, base_msr="
1633                    "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1634                    vcpu, kvm_apic_id(apic),
1635                    vcpu->arch.apic_base, apic->base_address);
1636 }
1637
1638 /*
1639  *----------------------------------------------------------------------
1640  * timer interface
1641  *----------------------------------------------------------------------
1642  */
1643
1644 static bool lapic_is_periodic(struct kvm_lapic *apic)
1645 {
1646         return apic_lvtt_period(apic);
1647 }
1648
1649 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1650 {
1651         struct kvm_lapic *apic = vcpu->arch.apic;
1652
1653         if (kvm_vcpu_has_lapic(vcpu) && apic_enabled(apic) &&
1654                         apic_lvt_enabled(apic, APIC_LVTT))
1655                 return atomic_read(&apic->lapic_timer.pending);
1656
1657         return 0;
1658 }
1659
1660 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1661 {
1662         u32 reg = kvm_apic_get_reg(apic, lvt_type);
1663         int vector, mode, trig_mode;
1664
1665         if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1666                 vector = reg & APIC_VECTOR_MASK;
1667                 mode = reg & APIC_MODE_MASK;
1668                 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1669                 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
1670                                         NULL);
1671         }
1672         return 0;
1673 }
1674
1675 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1676 {
1677         struct kvm_lapic *apic = vcpu->arch.apic;
1678
1679         if (apic)
1680                 kvm_apic_local_deliver(apic, APIC_LVT0);
1681 }
1682
1683 static const struct kvm_io_device_ops apic_mmio_ops = {
1684         .read     = apic_mmio_read,
1685         .write    = apic_mmio_write,
1686 };
1687
1688 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1689 {
1690         struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1691         struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
1692
1693         apic_timer_expired(apic);
1694
1695         if (lapic_is_periodic(apic)) {
1696                 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1697                 return HRTIMER_RESTART;
1698         } else
1699                 return HRTIMER_NORESTART;
1700 }
1701
1702 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1703 {
1704         struct kvm_lapic *apic;
1705
1706         ASSERT(vcpu != NULL);
1707         apic_debug("apic_init %d\n", vcpu->vcpu_id);
1708
1709         apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1710         if (!apic)
1711                 goto nomem;
1712
1713         vcpu->arch.apic = apic;
1714
1715         apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1716         if (!apic->regs) {
1717                 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1718                        vcpu->vcpu_id);
1719                 goto nomem_free_apic;
1720         }
1721         apic->vcpu = vcpu;
1722
1723         hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1724                      HRTIMER_MODE_ABS);
1725         apic->lapic_timer.timer.function = apic_timer_fn;
1726
1727         /*
1728          * APIC is created enabled. This will prevent kvm_lapic_set_base from
1729          * thinking that APIC satet has changed.
1730          */
1731         vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
1732         kvm_lapic_set_base(vcpu,
1733                         APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE);
1734
1735         static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
1736         kvm_lapic_reset(vcpu, false);
1737         kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1738
1739         return 0;
1740 nomem_free_apic:
1741         kfree(apic);
1742 nomem:
1743         return -ENOMEM;
1744 }
1745
1746 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1747 {
1748         struct kvm_lapic *apic = vcpu->arch.apic;
1749         int highest_irr;
1750
1751         if (!kvm_vcpu_has_lapic(vcpu) || !apic_enabled(apic))
1752                 return -1;
1753
1754         apic_update_ppr(apic);
1755         highest_irr = apic_find_highest_irr(apic);
1756         if ((highest_irr == -1) ||
1757             ((highest_irr & 0xF0) <= kvm_apic_get_reg(apic, APIC_PROCPRI)))
1758                 return -1;
1759         return highest_irr;
1760 }
1761
1762 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1763 {
1764         u32 lvt0 = kvm_apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1765         int r = 0;
1766
1767         if (!kvm_apic_hw_enabled(vcpu->arch.apic))
1768                 r = 1;
1769         if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1770             GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1771                 r = 1;
1772         return r;
1773 }
1774
1775 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1776 {
1777         struct kvm_lapic *apic = vcpu->arch.apic;
1778
1779         if (!kvm_vcpu_has_lapic(vcpu))
1780                 return;
1781
1782         if (atomic_read(&apic->lapic_timer.pending) > 0) {
1783                 kvm_apic_local_deliver(apic, APIC_LVTT);
1784                 if (apic_lvtt_tscdeadline(apic))
1785                         apic->lapic_timer.tscdeadline = 0;
1786                 atomic_set(&apic->lapic_timer.pending, 0);
1787         }
1788 }
1789
1790 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1791 {
1792         int vector = kvm_apic_has_interrupt(vcpu);
1793         struct kvm_lapic *apic = vcpu->arch.apic;
1794
1795         if (vector == -1)
1796                 return -1;
1797
1798         /*
1799          * We get here even with APIC virtualization enabled, if doing
1800          * nested virtualization and L1 runs with the "acknowledge interrupt
1801          * on exit" mode.  Then we cannot inject the interrupt via RVI,
1802          * because the process would deliver it through the IDT.
1803          */
1804
1805         apic_set_isr(vector, apic);
1806         apic_update_ppr(apic);
1807         apic_clear_irr(vector, apic);
1808         return vector;
1809 }
1810
1811 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu,
1812                 struct kvm_lapic_state *s)
1813 {
1814         struct kvm_lapic *apic = vcpu->arch.apic;
1815
1816         kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
1817         /* set SPIV separately to get count of SW disabled APICs right */
1818         apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
1819         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1820         /* call kvm_apic_set_id() to put apic into apic_map */
1821         kvm_apic_set_id(apic, kvm_apic_id(apic));
1822         kvm_apic_set_version(vcpu);
1823
1824         apic_update_ppr(apic);
1825         hrtimer_cancel(&apic->lapic_timer.timer);
1826         apic_update_lvtt(apic);
1827         apic_manage_nmi_watchdog(apic, kvm_apic_get_reg(apic, APIC_LVT0));
1828         update_divide_count(apic);
1829         start_apic_timer(apic);
1830         apic->irr_pending = true;
1831         apic->isr_count = kvm_x86_ops->hwapic_isr_update ?
1832                                 1 : count_vectors(apic->regs + APIC_ISR);
1833         apic->highest_isr_cache = -1;
1834         if (kvm_x86_ops->hwapic_irr_update)
1835                 kvm_x86_ops->hwapic_irr_update(vcpu,
1836                                 apic_find_highest_irr(apic));
1837         if (unlikely(kvm_x86_ops->hwapic_isr_update))
1838                 kvm_x86_ops->hwapic_isr_update(vcpu->kvm,
1839                                 apic_find_highest_isr(apic));
1840         kvm_make_request(KVM_REQ_EVENT, vcpu);
1841         kvm_rtc_eoi_tracking_restore_one(vcpu);
1842 }
1843
1844 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1845 {
1846         struct hrtimer *timer;
1847
1848         if (!kvm_vcpu_has_lapic(vcpu))
1849                 return;
1850
1851         timer = &vcpu->arch.apic->lapic_timer.timer;
1852         if (hrtimer_cancel(timer))
1853                 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1854 }
1855
1856 /*
1857  * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
1858  *
1859  * Detect whether guest triggered PV EOI since the
1860  * last entry. If yes, set EOI on guests's behalf.
1861  * Clear PV EOI in guest memory in any case.
1862  */
1863 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
1864                                         struct kvm_lapic *apic)
1865 {
1866         bool pending;
1867         int vector;
1868         /*
1869          * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
1870          * and KVM_PV_EOI_ENABLED in guest memory as follows:
1871          *
1872          * KVM_APIC_PV_EOI_PENDING is unset:
1873          *      -> host disabled PV EOI.
1874          * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
1875          *      -> host enabled PV EOI, guest did not execute EOI yet.
1876          * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
1877          *      -> host enabled PV EOI, guest executed EOI.
1878          */
1879         BUG_ON(!pv_eoi_enabled(vcpu));
1880         pending = pv_eoi_get_pending(vcpu);
1881         /*
1882          * Clear pending bit in any case: it will be set again on vmentry.
1883          * While this might not be ideal from performance point of view,
1884          * this makes sure pv eoi is only enabled when we know it's safe.
1885          */
1886         pv_eoi_clr_pending(vcpu);
1887         if (pending)
1888                 return;
1889         vector = apic_set_eoi(apic);
1890         trace_kvm_pv_eoi(apic, vector);
1891 }
1892
1893 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1894 {
1895         u32 data;
1896
1897         if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
1898                 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
1899
1900         if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1901                 return;
1902
1903         kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1904                                 sizeof(u32));
1905
1906         apic_set_tpr(vcpu->arch.apic, data & 0xff);
1907 }
1908
1909 /*
1910  * apic_sync_pv_eoi_to_guest - called before vmentry
1911  *
1912  * Detect whether it's safe to enable PV EOI and
1913  * if yes do so.
1914  */
1915 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
1916                                         struct kvm_lapic *apic)
1917 {
1918         if (!pv_eoi_enabled(vcpu) ||
1919             /* IRR set or many bits in ISR: could be nested. */
1920             apic->irr_pending ||
1921             /* Cache not set: could be safe but we don't bother. */
1922             apic->highest_isr_cache == -1 ||
1923             /* Need EOI to update ioapic. */
1924             kvm_ioapic_handles_vector(vcpu->kvm, apic->highest_isr_cache)) {
1925                 /*
1926                  * PV EOI was disabled by apic_sync_pv_eoi_from_guest
1927                  * so we need not do anything here.
1928                  */
1929                 return;
1930         }
1931
1932         pv_eoi_set_pending(apic->vcpu);
1933 }
1934
1935 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1936 {
1937         u32 data, tpr;
1938         int max_irr, max_isr;
1939         struct kvm_lapic *apic = vcpu->arch.apic;
1940
1941         apic_sync_pv_eoi_to_guest(vcpu, apic);
1942
1943         if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1944                 return;
1945
1946         tpr = kvm_apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1947         max_irr = apic_find_highest_irr(apic);
1948         if (max_irr < 0)
1949                 max_irr = 0;
1950         max_isr = apic_find_highest_isr(apic);
1951         if (max_isr < 0)
1952                 max_isr = 0;
1953         data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1954
1955         kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1956                                 sizeof(u32));
1957 }
1958
1959 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1960 {
1961         if (vapic_addr) {
1962                 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
1963                                         &vcpu->arch.apic->vapic_cache,
1964                                         vapic_addr, sizeof(u32)))
1965                         return -EINVAL;
1966                 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1967         } else {
1968                 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1969         }
1970
1971         vcpu->arch.apic->vapic_addr = vapic_addr;
1972         return 0;
1973 }
1974
1975 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1976 {
1977         struct kvm_lapic *apic = vcpu->arch.apic;
1978         u32 reg = (msr - APIC_BASE_MSR) << 4;
1979
1980         if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1981                 return 1;
1982
1983         if (reg == APIC_ICR2)
1984                 return 1;
1985
1986         /* if this is ICR write vector before command */
1987         if (reg == APIC_ICR)
1988                 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1989         return apic_reg_write(apic, reg, (u32)data);
1990 }
1991
1992 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1993 {
1994         struct kvm_lapic *apic = vcpu->arch.apic;
1995         u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1996
1997         if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1998                 return 1;
1999
2000         if (reg == APIC_DFR || reg == APIC_ICR2) {
2001                 apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
2002                            reg);
2003                 return 1;
2004         }
2005
2006         if (apic_reg_read(apic, reg, 4, &low))
2007                 return 1;
2008         if (reg == APIC_ICR)
2009                 apic_reg_read(apic, APIC_ICR2, 4, &high);
2010
2011         *data = (((u64)high) << 32) | low;
2012
2013         return 0;
2014 }
2015
2016 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2017 {
2018         struct kvm_lapic *apic = vcpu->arch.apic;
2019
2020         if (!kvm_vcpu_has_lapic(vcpu))
2021                 return 1;
2022
2023         /* if this is ICR write vector before command */
2024         if (reg == APIC_ICR)
2025                 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2026         return apic_reg_write(apic, reg, (u32)data);
2027 }
2028
2029 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2030 {
2031         struct kvm_lapic *apic = vcpu->arch.apic;
2032         u32 low, high = 0;
2033
2034         if (!kvm_vcpu_has_lapic(vcpu))
2035                 return 1;
2036
2037         if (apic_reg_read(apic, reg, 4, &low))
2038                 return 1;
2039         if (reg == APIC_ICR)
2040                 apic_reg_read(apic, APIC_ICR2, 4, &high);
2041
2042         *data = (((u64)high) << 32) | low;
2043
2044         return 0;
2045 }
2046
2047 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
2048 {
2049         u64 addr = data & ~KVM_MSR_ENABLED;
2050         if (!IS_ALIGNED(addr, 4))
2051                 return 1;
2052
2053         vcpu->arch.pv_eoi.msr_val = data;
2054         if (!pv_eoi_enabled(vcpu))
2055                 return 0;
2056         return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
2057                                          addr, sizeof(u8));
2058 }
2059
2060 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2061 {
2062         struct kvm_lapic *apic = vcpu->arch.apic;
2063         u8 sipi_vector;
2064         unsigned long pe;
2065
2066         if (!kvm_vcpu_has_lapic(vcpu) || !apic->pending_events)
2067                 return;
2068
2069         /*
2070          * INITs are latched while in SMM.  Because an SMM CPU cannot
2071          * be in KVM_MP_STATE_INIT_RECEIVED state, just eat SIPIs
2072          * and delay processing of INIT until the next RSM.
2073          */
2074         if (is_smm(vcpu)) {
2075                 WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2076                 if (test_bit(KVM_APIC_SIPI, &apic->pending_events))
2077                         clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2078                 return;
2079         }
2080
2081         pe = xchg(&apic->pending_events, 0);
2082         if (test_bit(KVM_APIC_INIT, &pe)) {
2083                 kvm_lapic_reset(vcpu, true);
2084                 kvm_vcpu_reset(vcpu, true);
2085                 if (kvm_vcpu_is_bsp(apic->vcpu))
2086                         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2087                 else
2088                         vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2089         }
2090         if (test_bit(KVM_APIC_SIPI, &pe) &&
2091             vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2092                 /* evaluate pending_events before reading the vector */
2093                 smp_rmb();
2094                 sipi_vector = apic->sipi_vector;
2095                 apic_debug("vcpu %d received sipi with vector # %x\n",
2096                          vcpu->vcpu_id, sipi_vector);
2097                 kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2098                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2099         }
2100 }
2101
2102 void kvm_lapic_init(void)
2103 {
2104         /* do not patch jump label more than once per second */
2105         jump_label_rate_limit(&apic_hw_disabled, HZ);
2106         jump_label_rate_limit(&apic_sw_disabled, HZ);
2107 }