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 <[email protected]>
  12  *   Gregory Haskins <[email protected]>
  13  *   Yaozu (Eddie) Dong <[email protected]>
  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/export.h>
  29 #include <linux/math64.h>
  30 #include <linux/slab.h>
  31 #include <asm/processor.h>
  32 #include <asm/msr.h>
  33 #include <asm/page.h>
  34 #include <asm/current.h>
  35 #include <asm/apicdef.h>
  36 #include <asm/delay.h>
  37 #include <linux/atomic.h>
  38 #include <linux/jump_label.h>
  39 #include "kvm_cache_regs.h"
  40 #include "irq.h"
  41 #include "trace.h"
  42 #include "x86.h"
  43 #include "cpuid.h"
  44 #include "hyperv.h"
  45
  46 #ifndef CONFIG_X86_64
  47 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
  48 #else
  49 #define mod_64(x, y) ((x) % (y))
  50 #endif
  51
  52 #define PRId64 "d"
  53 #define PRIx64 "llx"
  54 #define PRIu64 "u"
  55 #define PRIo64 "o"
  56
  57 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
  58 #define apic_debug(fmt, arg...)
  59
  60 /* 14 is the version for Xeon and Pentium 8.4.8*/
  61 #define APIC_VERSION                    (0x14UL | ((KVM_APIC_LVT_NUM - 1) << 16))
  62 #define LAPIC_MMIO_LENGTH               (1 << 12)
  63 /* followed define is not in apicdef.h */
  64 #define APIC_SHORT_MASK                 0xc0000
  65 #define APIC_DEST_NOSHORT               0x0
  66 #define APIC_DEST_MASK                  0x800
  67 #define MAX_APIC_VECTOR                 256
  68 #define APIC_VECTORS_PER_REG            32
  69
  70 #define APIC_BROADCAST                  0xFF
  71 #define X2APIC_BROADCAST                0xFFFFFFFFul
  72
  73 static inline int apic_test_vector(int vec, void *bitmap)
  74 {
  75         return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
  76 }
  77
  78 bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
  79 {
  80         struct kvm_lapic *apic = vcpu->arch.apic;
  81
  82         return apic_test_vector(vector, apic->regs + APIC_ISR) ||
  83                 apic_test_vector(vector, apic->regs + APIC_IRR);
  84 }
  85
  86 static inline void apic_clear_vector(int vec, void *bitmap)
  87 {
  88         clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
  89 }
  90
  91 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
  92 {
  93         return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
  94 }
  95
  96 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
  97 {
  98         return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
  99 }
 100
 101 struct static_key_deferred apic_hw_disabled __read_mostly;
 102 struct static_key_deferred apic_sw_disabled __read_mostly;
 103
 104 static inline int apic_enabled(struct kvm_lapic *apic)
 105 {
 106         return kvm_apic_sw_enabled(apic) &&     kvm_apic_hw_enabled(apic);
 107 }
 108
 109 #define LVT_MASK        \
 110         (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
 111
 112 #define LINT_MASK       \
 113         (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
 114          APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
 115
 116 static inline u8 kvm_xapic_id(struct kvm_lapic *apic)
 117 {
 118         return kvm_lapic_get_reg(apic, APIC_ID) >> 24;
 119 }
 120
 121 static inline u32 kvm_x2apic_id(struct kvm_lapic *apic)
 122 {
 123         return apic->vcpu->vcpu_id;
 124 }
 125
 126 static inline bool kvm_apic_map_get_logical_dest(struct kvm_apic_map *map,
 127                 u32 dest_id, struct kvm_lapic ***cluster, u16 *mask) {
 128         switch (map->mode) {
 129         case KVM_APIC_MODE_X2APIC: {
 130                 u32 offset = (dest_id >> 16) * 16;
 131                 u32 max_apic_id = map->max_apic_id;
 132
 133                 if (offset <= max_apic_id) {
 134                         u8 cluster_size = min(max_apic_id - offset + 1, 16U);
 135
 136                         *cluster = &map->phys_map[offset];
 137                         *mask = dest_id & (0xffff >> (16 - cluster_size));
 138                 } else {
 139                         *mask = 0;
 140                 }
 141
 142                 return true;
 143                 }
 144         case KVM_APIC_MODE_XAPIC_FLAT:
 145                 *cluster = map->xapic_flat_map;
 146                 *mask = dest_id & 0xff;
 147                 return true;
 148         case KVM_APIC_MODE_XAPIC_CLUSTER:
 149                 *cluster = map->xapic_cluster_map[(dest_id >> 4) & 0xf];
 150                 *mask = dest_id & 0xf;
 151                 return true;
 152         default:
 153                 /* Not optimized. */
 154                 return false;
 155         }
 156 }
 157
 158 static void kvm_apic_map_free(struct rcu_head *rcu)
 159 {
 160         struct kvm_apic_map *map = container_of(rcu, struct kvm_apic_map, rcu);
 161
 162         kvfree(map);
 163 }
 164
 165 static void recalculate_apic_map(struct kvm *kvm)
 166 {
 167         struct kvm_apic_map *new, *old = NULL;
 168         struct kvm_vcpu *vcpu;
 169         int i;
 170         u32 max_id = 255; /* enough space for any xAPIC ID */
 171
 172         mutex_lock(&kvm->arch.apic_map_lock);
 173
 174         kvm_for_each_vcpu(i, vcpu, kvm)
 175                 if (kvm_apic_present(vcpu))
 176                         max_id = max(max_id, kvm_x2apic_id(vcpu->arch.apic));
 177
 178         new = kvzalloc(sizeof(struct kvm_apic_map) +
 179                            sizeof(struct kvm_lapic *) * ((u64)max_id + 1), GFP_KERNEL);
 180
 181         if (!new)
 182                 goto out;
 183
 184         new->max_apic_id = max_id;
 185
 186         kvm_for_each_vcpu(i, vcpu, kvm) {
 187                 struct kvm_lapic *apic = vcpu->arch.apic;
 188                 struct kvm_lapic **cluster;
 189                 u16 mask;
 190                 u32 ldr;
 191                 u8 xapic_id;
 192                 u32 x2apic_id;
 193
 194                 if (!kvm_apic_present(vcpu))
 195                         continue;
 196
 197                 xapic_id = kvm_xapic_id(apic);
 198                 x2apic_id = kvm_x2apic_id(apic);
 199
 200                 /* Hotplug hack: see kvm_apic_match_physical_addr(), ... */
 201                 if ((apic_x2apic_mode(apic) || x2apic_id > 0xff) &&
 202                                 x2apic_id <= new->max_apic_id)
 203                         new->phys_map[x2apic_id] = apic;
 204                 /*
 205                  * ... xAPIC ID of VCPUs with APIC ID > 0xff will wrap-around,
 206                  * prevent them from masking VCPUs with APIC ID <= 0xff.
 207                  */
 208                 if (!apic_x2apic_mode(apic) && !new->phys_map[xapic_id])
 209                         new->phys_map[xapic_id] = apic;
 210
 211                 ldr = kvm_lapic_get_reg(apic, APIC_LDR);
 212
 213                 if (apic_x2apic_mode(apic)) {
 214                         new->mode |= KVM_APIC_MODE_X2APIC;
 215                 } else if (ldr) {
 216                         ldr = GET_APIC_LOGICAL_ID(ldr);
 217                         if (kvm_lapic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
 218                                 new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
 219                         else
 220                                 new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
 221                 }
 222
 223                 if (!kvm_apic_map_get_logical_dest(new, ldr, &cluster, &mask))
 224                         continue;
 225
 226                 if (mask)
 227                         cluster[ffs(mask) - 1] = apic;
 228         }
 229 out:
 230         old = rcu_dereference_protected(kvm->arch.apic_map,
 231                         lockdep_is_held(&kvm->arch.apic_map_lock));
 232         rcu_assign_pointer(kvm->arch.apic_map, new);
 233         mutex_unlock(&kvm->arch.apic_map_lock);
 234
 235         if (old)
 236                 call_rcu(&old->rcu, kvm_apic_map_free);
 237
 238         kvm_make_scan_ioapic_request(kvm);
 239 }
 240
 241 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
 242 {
 243         bool enabled = val & APIC_SPIV_APIC_ENABLED;
 244
 245         kvm_lapic_set_reg(apic, APIC_SPIV, val);
 246
 247         if (enabled != apic->sw_enabled) {
 248                 apic->sw_enabled = enabled;
 249                 if (enabled) {
 250                         static_key_slow_dec_deferred(&apic_sw_disabled);
 251                         recalculate_apic_map(apic->vcpu->kvm);
 252                 } else
 253                         static_key_slow_inc(&apic_sw_disabled.key);
 254         }
 255 }
 256
 257 static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
 258 {
 259         kvm_lapic_set_reg(apic, APIC_ID, id << 24);
 260         recalculate_apic_map(apic->vcpu->kvm);
 261 }
 262
 263 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
 264 {
 265         kvm_lapic_set_reg(apic, APIC_LDR, id);
 266         recalculate_apic_map(apic->vcpu->kvm);
 267 }
 268
 269 static inline u32 kvm_apic_calc_x2apic_ldr(u32 id)
 270 {
 271         return ((id >> 4) << 16) | (1 << (id & 0xf));
 272 }
 273
 274 static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
 275 {
 276         u32 ldr = kvm_apic_calc_x2apic_ldr(id);
 277
 278         WARN_ON_ONCE(id != apic->vcpu->vcpu_id);
 279
 280         kvm_lapic_set_reg(apic, APIC_ID, id);
 281         kvm_lapic_set_reg(apic, APIC_LDR, ldr);
 282         recalculate_apic_map(apic->vcpu->kvm);
 283 }
 284
 285 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
 286 {
 287         return !(kvm_lapic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
 288 }
 289
 290 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
 291 {
 292         return kvm_lapic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
 293 }
 294
 295 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
 296 {
 297         return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
 298 }
 299
 300 static inline int apic_lvtt_period(struct kvm_lapic *apic)
 301 {
 302         return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
 303 }
 304
 305 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
 306 {
 307         return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
 308 }
 309
 310 static inline int apic_lvt_nmi_mode(u32 lvt_val)
 311 {
 312         return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
 313 }
 314
 315 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
 316 {
 317         struct kvm_lapic *apic = vcpu->arch.apic;
 318         struct kvm_cpuid_entry2 *feat;
 319         u32 v = APIC_VERSION;
 320
 321         if (!lapic_in_kernel(vcpu))
 322                 return;
 323
 324         /*
 325          * KVM emulates 82093AA datasheet (with in-kernel IOAPIC implementation)
 326          * which doesn't have EOI register; Some buggy OSes (e.g. Windows with
 327          * Hyper-V role) disable EOI broadcast in lapic not checking for IOAPIC
 328          * version first and level-triggered interrupts never get EOIed in
 329          * IOAPIC.
 330          */
 331         feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
 332         if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))) &&
 333             !ioapic_in_kernel(vcpu->kvm))
 334                 v |= APIC_LVR_DIRECTED_EOI;
 335         kvm_lapic_set_reg(apic, APIC_LVR, v);
 336 }
 337
 338 static const unsigned int apic_lvt_mask[KVM_APIC_LVT_NUM] = {
 339         LVT_MASK ,      /* part LVTT mask, timer mode mask added at runtime */
 340         LVT_MASK | APIC_MODE_MASK,      /* LVTTHMR */
 341         LVT_MASK | APIC_MODE_MASK,      /* LVTPC */
 342         LINT_MASK, LINT_MASK,   /* LVT0-1 */
 343         LVT_MASK                /* LVTERR */
 344 };
 345
 346 static int find_highest_vector(void *bitmap)
 347 {
 348         int vec;
 349         u32 *reg;
 350
 351         for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
 352              vec >= 0; vec -= APIC_VECTORS_PER_REG) {
 353                 reg = bitmap + REG_POS(vec);
 354                 if (*reg)
 355                         return __fls(*reg) + vec;
 356         }
 357
 358         return -1;
 359 }
 360
 361 static u8 count_vectors(void *bitmap)
 362 {
 363         int vec;
 364         u32 *reg;
 365         u8 count = 0;
 366
 367         for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
 368                 reg = bitmap + REG_POS(vec);
 369                 count += hweight32(*reg);
 370         }
 371
 372         return count;
 373 }
 374
 375 bool __kvm_apic_update_irr(u32 *pir, void *regs, int *max_irr)
 376 {
 377         u32 i, vec;
 378         u32 pir_val, irr_val, prev_irr_val;
 379         int max_updated_irr;
 380
 381         max_updated_irr = -1;
 382         *max_irr = -1;
 383
 384         for (i = vec = 0; i <= 7; i++, vec += 32) {
 385                 pir_val = READ_ONCE(pir[i]);
 386                 irr_val = *((u32 *)(regs + APIC_IRR + i * 0x10));
 387                 if (pir_val) {
 388                         prev_irr_val = irr_val;
 389                         irr_val |= xchg(&pir[i], 0);
 390                         *((u32 *)(regs + APIC_IRR + i * 0x10)) = irr_val;
 391                         if (prev_irr_val != irr_val) {
 392                                 max_updated_irr =
 393                                         __fls(irr_val ^ prev_irr_val) + vec;
 394                         }
 395                 }
 396                 if (irr_val)
 397                         *max_irr = __fls(irr_val) + vec;
 398         }
 399
 400         return ((max_updated_irr != -1) &&
 401                 (max_updated_irr == *max_irr));
 402 }
 403 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
 404
 405 bool kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir, int *max_irr)
 406 {
 407         struct kvm_lapic *apic = vcpu->arch.apic;
 408
 409         return __kvm_apic_update_irr(pir, apic->regs, max_irr);
 410 }
 411 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
 412
 413 static inline int apic_search_irr(struct kvm_lapic *apic)
 414 {
 415         return find_highest_vector(apic->regs + APIC_IRR);
 416 }
 417
 418 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
 419 {
 420         int result;
 421
 422         /*
 423          * Note that irr_pending is just a hint. It will be always
 424          * true with virtual interrupt delivery enabled.
 425          */
 426         if (!apic->irr_pending)
 427                 return -1;
 428
 429         result = apic_search_irr(apic);
 430         ASSERT(result == -1 || result >= 16);
 431
 432         return result;
 433 }
 434
 435 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
 436 {
 437         struct kvm_vcpu *vcpu;
 438
 439         vcpu = apic->vcpu;
 440
 441         if (unlikely(vcpu->arch.apicv_active)) {
 442                 /* need to update RVI */
 443                 apic_clear_vector(vec, apic->regs + APIC_IRR);
 444                 kvm_x86_ops->hwapic_irr_update(vcpu,
 445                                 apic_find_highest_irr(apic));
 446         } else {
 447                 apic->irr_pending = false;
 448                 apic_clear_vector(vec, apic->regs + APIC_IRR);
 449                 if (apic_search_irr(apic) != -1)
 450                         apic->irr_pending = true;
 451         }
 452 }
 453
 454 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
 455 {
 456         struct kvm_vcpu *vcpu;
 457
 458         if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
 459                 return;
 460
 461         vcpu = apic->vcpu;
 462
 463         /*
 464          * With APIC virtualization enabled, all caching is disabled
 465          * because the processor can modify ISR under the hood.  Instead
 466          * just set SVI.
 467          */
 468         if (unlikely(vcpu->arch.apicv_active))
 469                 kvm_x86_ops->hwapic_isr_update(vcpu, vec);
 470         else {
 471                 ++apic->isr_count;
 472                 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
 473                 /*
 474                  * ISR (in service register) bit is set when injecting an interrupt.
 475                  * The highest vector is injected. Thus the latest bit set matches
 476                  * the highest bit in ISR.
 477                  */
 478                 apic->highest_isr_cache = vec;
 479         }
 480 }
 481
 482 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
 483 {
 484         int result;
 485
 486         /*
 487          * Note that isr_count is always 1, and highest_isr_cache
 488          * is always -1, with APIC virtualization enabled.
 489          */
 490         if (!apic->isr_count)
 491                 return -1;
 492         if (likely(apic->highest_isr_cache != -1))
 493                 return apic->highest_isr_cache;
 494
 495         result = find_highest_vector(apic->regs + APIC_ISR);
 496         ASSERT(result == -1 || result >= 16);
 497
 498         return result;
 499 }
 500
 501 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
 502 {
 503         struct kvm_vcpu *vcpu;
 504         if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
 505                 return;
 506
 507         vcpu = apic->vcpu;
 508
 509         /*
 510          * We do get here for APIC virtualization enabled if the guest
 511          * uses the Hyper-V APIC enlightenment.  In this case we may need
 512          * to trigger a new interrupt delivery by writing the SVI field;
 513          * on the other hand isr_count and highest_isr_cache are unused
 514          * and must be left alone.
 515          */
 516         if (unlikely(vcpu->arch.apicv_active))
 517                 kvm_x86_ops->hwapic_isr_update(vcpu,
 518                                                apic_find_highest_isr(apic));
 519         else {
 520                 --apic->isr_count;
 521                 BUG_ON(apic->isr_count < 0);
 522                 apic->highest_isr_cache = -1;
 523         }
 524 }
 525
 526 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
 527 {
 528         /* This may race with setting of irr in __apic_accept_irq() and
 529          * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
 530          * will cause vmexit immediately and the value will be recalculated
 531          * on the next vmentry.
 532          */
 533         return apic_find_highest_irr(vcpu->arch.apic);
 534 }
 535 EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
 536
 537 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
 538                              int vector, int level, int trig_mode,
 539                              struct dest_map *dest_map);
 540
 541 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
 542                      struct dest_map *dest_map)
 543 {
 544         struct kvm_lapic *apic = vcpu->arch.apic;
 545
 546         return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
 547                         irq->level, irq->trig_mode, dest_map);
 548 }
 549
 550 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
 551 {
 552
 553         return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
 554                                       sizeof(val));
 555 }
 556
 557 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
 558 {
 559
 560         return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
 561                                       sizeof(*val));
 562 }
 563
 564 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
 565 {
 566         return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
 567 }
 568
 569 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
 570 {
 571         u8 val;
 572         if (pv_eoi_get_user(vcpu, &val) < 0)
 573                 apic_debug("Can't read EOI MSR value: 0x%llx\n",
 574                            (unsigned long long)vcpu->arch.pv_eoi.msr_val);
 575         return val & 0x1;
 576 }
 577
 578 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
 579 {
 580         if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
 581                 apic_debug("Can't set EOI MSR value: 0x%llx\n",
 582                            (unsigned long long)vcpu->arch.pv_eoi.msr_val);
 583                 return;
 584         }
 585         __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
 586 }
 587
 588 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
 589 {
 590         if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
 591                 apic_debug("Can't clear EOI MSR value: 0x%llx\n",
 592                            (unsigned long long)vcpu->arch.pv_eoi.msr_val);
 593                 return;
 594         }
 595         __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
 596 }
 597
 598 static int apic_has_interrupt_for_ppr(struct kvm_lapic *apic, u32 ppr)
 599 {
 600         int highest_irr;
 601         if (apic->vcpu->arch.apicv_active)
 602                 highest_irr = kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
 603         else
 604                 highest_irr = apic_find_highest_irr(apic);
 605         if (highest_irr == -1 || (highest_irr & 0xF0) <= ppr)
 606                 return -1;
 607         return highest_irr;
 608 }
 609
 610 static bool __apic_update_ppr(struct kvm_lapic *apic, u32 *new_ppr)
 611 {
 612         u32 tpr, isrv, ppr, old_ppr;
 613         int isr;
 614
 615         old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
 616         tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
 617         isr = apic_find_highest_isr(apic);
 618         isrv = (isr != -1) ? isr : 0;
 619
 620         if ((tpr & 0xf0) >= (isrv & 0xf0))
 621                 ppr = tpr & 0xff;
 622         else
 623                 ppr = isrv & 0xf0;
 624
 625         apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
 626                    apic, ppr, isr, isrv);
 627
 628         *new_ppr = ppr;
 629         if (old_ppr != ppr)
 630                 kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
 631
 632         return ppr < old_ppr;
 633 }
 634
 635 static void apic_update_ppr(struct kvm_lapic *apic)
 636 {
 637         u32 ppr;
 638
 639         if (__apic_update_ppr(apic, &ppr) &&
 640             apic_has_interrupt_for_ppr(apic, ppr) != -1)
 641                 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
 642 }
 643
 644 void kvm_apic_update_ppr(struct kvm_vcpu *vcpu)
 645 {
 646         apic_update_ppr(vcpu->arch.apic);
 647 }
 648 EXPORT_SYMBOL_GPL(kvm_apic_update_ppr);
 649
 650 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
 651 {
 652         kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
 653         apic_update_ppr(apic);
 654 }
 655
 656 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
 657 {
 658         return mda == (apic_x2apic_mode(apic) ?
 659                         X2APIC_BROADCAST : APIC_BROADCAST);
 660 }
 661
 662 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
 663 {
 664         if (kvm_apic_broadcast(apic, mda))
 665                 return true;
 666
 667         if (apic_x2apic_mode(apic))
 668                 return mda == kvm_x2apic_id(apic);
 669
 670         /*
 671          * Hotplug hack: Make LAPIC in xAPIC mode also accept interrupts as if
 672          * it were in x2APIC mode.  Hotplugged VCPUs start in xAPIC mode and
 673          * this allows unique addressing of VCPUs with APIC ID over 0xff.
 674          * The 0xff condition is needed because writeable xAPIC ID.
 675          */
 676         if (kvm_x2apic_id(apic) > 0xff && mda == kvm_x2apic_id(apic))
 677                 return true;
 678
 679         return mda == kvm_xapic_id(apic);
 680 }
 681
 682 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
 683 {
 684         u32 logical_id;
 685
 686         if (kvm_apic_broadcast(apic, mda))
 687                 return true;
 688
 689         logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
 690
 691         if (apic_x2apic_mode(apic))
 692                 return ((logical_id >> 16) == (mda >> 16))
 693                        && (logical_id & mda & 0xffff) != 0;
 694
 695         logical_id = GET_APIC_LOGICAL_ID(logical_id);
 696
 697         switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
 698         case APIC_DFR_FLAT:
 699                 return (logical_id & mda) != 0;
 700         case APIC_DFR_CLUSTER:
 701                 return ((logical_id >> 4) == (mda >> 4))
 702                        && (logical_id & mda & 0xf) != 0;
 703         default:
 704                 apic_debug("Bad DFR vcpu %d: %08x\n",
 705                            apic->vcpu->vcpu_id, kvm_lapic_get_reg(apic, APIC_DFR));
 706                 return false;
 707         }
 708 }
 709
 710 /* The KVM local APIC implementation has two quirks:
 711  *
 712  *  - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
 713  *    in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
 714  *    KVM doesn't do that aliasing.
 715  *
 716  *  - in-kernel IOAPIC messages have to be delivered directly to
 717  *    x2APIC, because the kernel does not support interrupt remapping.
 718  *    In order to support broadcast without interrupt remapping, x2APIC
 719  *    rewrites the destination of non-IPI messages from APIC_BROADCAST
 720  *    to X2APIC_BROADCAST.
 721  *
 722  * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API.  This is
 723  * important when userspace wants to use x2APIC-format MSIs, because
 724  * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
 725  */
 726 static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
 727                 struct kvm_lapic *source, struct kvm_lapic *target)
 728 {
 729         bool ipi = source != NULL;
 730
 731         if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
 732             !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
 733                 return X2APIC_BROADCAST;
 734
 735         return dest_id;
 736 }
 737
 738 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
 739                            int short_hand, unsigned int dest, int dest_mode)
 740 {
 741         struct kvm_lapic *target = vcpu->arch.apic;
 742         u32 mda = kvm_apic_mda(vcpu, dest, source, target);
 743
 744         apic_debug("target %p, source %p, dest 0x%x, "
 745                    "dest_mode 0x%x, short_hand 0x%x\n",
 746                    target, source, dest, dest_mode, short_hand);
 747
 748         ASSERT(target);
 749         switch (short_hand) {
 750         case APIC_DEST_NOSHORT:
 751                 if (dest_mode == APIC_DEST_PHYSICAL)
 752                         return kvm_apic_match_physical_addr(target, mda);
 753                 else
 754                         return kvm_apic_match_logical_addr(target, mda);
 755         case APIC_DEST_SELF:
 756                 return target == source;
 757         case APIC_DEST_ALLINC:
 758                 return true;
 759         case APIC_DEST_ALLBUT:
 760                 return target != source;
 761         default:
 762                 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
 763                            short_hand);
 764                 return false;
 765         }
 766 }
 767 EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
 768
 769 int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
 770                        const unsigned long *bitmap, u32 bitmap_size)
 771 {
 772         u32 mod;
 773         int i, idx = -1;
 774
 775         mod = vector % dest_vcpus;
 776
 777         for (i = 0; i <= mod; i++) {
 778                 idx = find_next_bit(bitmap, bitmap_size, idx + 1);
 779                 BUG_ON(idx == bitmap_size);
 780         }
 781
 782         return idx;
 783 }
 784
 785 static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
 786 {
 787         if (!kvm->arch.disabled_lapic_found) {
 788                 kvm->arch.disabled_lapic_found = true;
 789                 printk(KERN_INFO
 790                        "Disabled LAPIC found during irq injection\n");
 791         }
 792 }
 793
 794 static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
 795                 struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
 796 {
 797         if (kvm->arch.x2apic_broadcast_quirk_disabled) {
 798                 if ((irq->dest_id == APIC_BROADCAST &&
 799                                 map->mode != KVM_APIC_MODE_X2APIC))
 800                         return true;
 801                 if (irq->dest_id == X2APIC_BROADCAST)
 802                         return true;
 803         } else {
 804                 bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
 805                 if (irq->dest_id == (x2apic_ipi ?
 806                                      X2APIC_BROADCAST : APIC_BROADCAST))
 807                         return true;
 808         }
 809
 810         return false;
 811 }
 812
 813 /* Return true if the interrupt can be handled by using *bitmap as index mask
 814  * for valid destinations in *dst array.
 815  * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
 816  * Note: we may have zero kvm_lapic destinations when we return true, which
 817  * means that the interrupt should be dropped.  In this case, *bitmap would be
 818  * zero and *dst undefined.
 819  */
 820 static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
 821                 struct kvm_lapic **src, struct kvm_lapic_irq *irq,
 822                 struct kvm_apic_map *map, struct kvm_lapic ***dst,
 823                 unsigned long *bitmap)
 824 {
 825         int i, lowest;
 826
 827         if (irq->shorthand == APIC_DEST_SELF && src) {
 828                 *dst = src;
 829                 *bitmap = 1;
 830                 return true;
 831         } else if (irq->shorthand)
 832                 return false;
 833
 834         if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
 835                 return false;
 836
 837         if (irq->dest_mode == APIC_DEST_PHYSICAL) {
 838                 if (irq->dest_id > map->max_apic_id) {
 839                         *bitmap = 0;
 840                 } else {
 841                         *dst = &map->phys_map[irq->dest_id];
 842                         *bitmap = 1;
 843                 }
 844                 return true;
 845         }
 846
 847         *bitmap = 0;
 848         if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
 849                                 (u16 *)bitmap))
 850                 return false;
 851
 852         if (!kvm_lowest_prio_delivery(irq))
 853                 return true;
 854
 855         if (!kvm_vector_hashing_enabled()) {
 856                 lowest = -1;
 857                 for_each_set_bit(i, bitmap, 16) {
 858                         if (!(*dst)[i])
 859                                 continue;
 860                         if (lowest < 0)
 861                                 lowest = i;
 862                         else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
 863                                                 (*dst)[lowest]->vcpu) < 0)
 864                                 lowest = i;
 865                 }
 866         } else {
 867                 if (!*bitmap)
 868                         return true;
 869
 870                 lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
 871                                 bitmap, 16);
 872
 873                 if (!(*dst)[lowest]) {
 874                         kvm_apic_disabled_lapic_found(kvm);
 875                         *bitmap = 0;
 876                         return true;
 877                 }
 878         }
 879
 880         *bitmap = (lowest >= 0) ? 1 << lowest : 0;
 881
 882         return true;
 883 }
 884
 885 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
 886                 struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
 887 {
 888         struct kvm_apic_map *map;
 889         unsigned long bitmap;
 890         struct kvm_lapic **dst = NULL;
 891         int i;
 892         bool ret;
 893
 894         *r = -1;
 895
 896         if (irq->shorthand == APIC_DEST_SELF) {
 897                 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
 898                 return true;
 899         }
 900
 901         rcu_read_lock();
 902         map = rcu_dereference(kvm->arch.apic_map);
 903
 904         ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
 905         if (ret)
 906                 for_each_set_bit(i, &bitmap, 16) {
 907                         if (!dst[i])
 908                                 continue;
 909                         if (*r < 0)
 910                                 *r = 0;
 911                         *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
 912                 }
 913
 914         rcu_read_unlock();
 915         return ret;
 916 }
 917
 918 /*
 919  * This routine tries to handler interrupts in posted mode, here is how
 920  * it deals with different cases:
 921  * - For single-destination interrupts, handle it in posted mode
 922  * - Else if vector hashing is enabled and it is a lowest-priority
 923  *   interrupt, handle it in posted mode and use the following mechanism
 924  *   to find the destinaiton vCPU.
 925  *      1. For lowest-priority interrupts, store all the possible
 926  *         destination vCPUs in an array.
 927  *      2. Use "guest vector % max number of destination vCPUs" to find
 928  *         the right destination vCPU in the array for the lowest-priority
 929  *         interrupt.
 930  * - Otherwise, use remapped mode to inject the interrupt.
 931  */
 932 bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
 933                         struct kvm_vcpu **dest_vcpu)
 934 {
 935         struct kvm_apic_map *map;
 936         unsigned long bitmap;
 937         struct kvm_lapic **dst = NULL;
 938         bool ret = false;
 939
 940         if (irq->shorthand)
 941                 return false;
 942
 943         rcu_read_lock();
 944         map = rcu_dereference(kvm->arch.apic_map);
 945
 946         if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
 947                         hweight16(bitmap) == 1) {
 948                 unsigned long i = find_first_bit(&bitmap, 16);
 949
 950                 if (dst[i]) {
 951                         *dest_vcpu = dst[i]->vcpu;
 952                         ret = true;
 953                 }
 954         }
 955
 956         rcu_read_unlock();
 957         return ret;
 958 }
 959
 960 /*
 961  * Add a pending IRQ into lapic.
 962  * Return 1 if successfully added and 0 if discarded.
 963  */
 964 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
 965                              int vector, int level, int trig_mode,
 966                              struct dest_map *dest_map)
 967 {
 968         int result = 0;
 969         struct kvm_vcpu *vcpu = apic->vcpu;
 970
 971         trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
 972                                   trig_mode, vector);
 973         switch (delivery_mode) {
 974         case APIC_DM_LOWEST:
 975                 vcpu->arch.apic_arb_prio++;
 976         case APIC_DM_FIXED:
 977                 if (unlikely(trig_mode && !level))
 978                         break;
 979
 980                 /* FIXME add logic for vcpu on reset */
 981                 if (unlikely(!apic_enabled(apic)))
 982                         break;
 983
 984                 result = 1;
 985
 986                 if (dest_map) {
 987                         __set_bit(vcpu->vcpu_id, dest_map->map);
 988                         dest_map->vectors[vcpu->vcpu_id] = vector;
 989                 }
 990
 991                 if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
 992                         if (trig_mode)
 993                                 kvm_lapic_set_vector(vector, apic->regs + APIC_TMR);
 994                         else
 995                                 apic_clear_vector(vector, apic->regs + APIC_TMR);
 996                 }
 997
 998                 if (vcpu->arch.apicv_active)
 999                         kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
1000                 else {
1001                         kvm_lapic_set_irr(vector, apic);
1002
1003                         kvm_make_request(KVM_REQ_EVENT, vcpu);
1004                         kvm_vcpu_kick(vcpu);
1005                 }
1006                 break;
1007
1008         case APIC_DM_REMRD:
1009                 result = 1;
1010                 vcpu->arch.pv.pv_unhalted = 1;
1011                 kvm_make_request(KVM_REQ_EVENT, vcpu);
1012                 kvm_vcpu_kick(vcpu);
1013                 break;
1014
1015         case APIC_DM_SMI:
1016                 result = 1;
1017                 kvm_make_request(KVM_REQ_SMI, vcpu);
1018                 kvm_vcpu_kick(vcpu);
1019                 break;
1020
1021         case APIC_DM_NMI:
1022                 result = 1;
1023                 kvm_inject_nmi(vcpu);
1024                 kvm_vcpu_kick(vcpu);
1025                 break;
1026
1027         case APIC_DM_INIT:
1028                 if (!trig_mode || level) {
1029                         result = 1;
1030                         /* assumes that there are only KVM_APIC_INIT/SIPI */
1031                         apic->pending_events = (1UL << KVM_APIC_INIT);
1032                         /* make sure pending_events is visible before sending
1033                          * the request */
1034                         smp_wmb();
1035                         kvm_make_request(KVM_REQ_EVENT, vcpu);
1036                         kvm_vcpu_kick(vcpu);
1037                 } else {
1038                         apic_debug("Ignoring de-assert INIT to vcpu %d\n",
1039                                    vcpu->vcpu_id);
1040                 }
1041                 break;
1042
1043         case APIC_DM_STARTUP:
1044                 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
1045                            vcpu->vcpu_id, vector);
1046                 result = 1;
1047                 apic->sipi_vector = vector;
1048                 /* make sure sipi_vector is visible for the receiver */
1049                 smp_wmb();
1050                 set_bit(KVM_APIC_SIPI, &apic->pending_events);
1051                 kvm_make_request(KVM_REQ_EVENT, vcpu);
1052                 kvm_vcpu_kick(vcpu);
1053                 break;
1054
1055         case APIC_DM_EXTINT:
1056                 /*
1057                  * Should only be called by kvm_apic_local_deliver() with LVT0,
1058                  * before NMI watchdog was enabled. Already handled by
1059                  * kvm_apic_accept_pic_intr().
1060                  */
1061                 break;
1062
1063         default:
1064                 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
1065                        delivery_mode);
1066                 break;
1067         }
1068         return result;
1069 }
1070
1071 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1072 {
1073         return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
1074 }
1075
1076 static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
1077 {
1078         return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
1079 }
1080
1081 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1082 {
1083         int trigger_mode;
1084
1085         /* Eoi the ioapic only if the ioapic doesn't own the vector. */
1086         if (!kvm_ioapic_handles_vector(apic, vector))
1087                 return;
1088
1089         /* Request a KVM exit to inform the userspace IOAPIC. */
1090         if (irqchip_split(apic->vcpu->kvm)) {
1091                 apic->vcpu->arch.pending_ioapic_eoi = vector;
1092                 kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1093                 return;
1094         }
1095
1096         if (apic_test_vector(vector, apic->regs + APIC_TMR))
1097                 trigger_mode = IOAPIC_LEVEL_TRIG;
1098         else
1099                 trigger_mode = IOAPIC_EDGE_TRIG;
1100
1101         kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1102 }
1103
1104 static int apic_set_eoi(struct kvm_lapic *apic)
1105 {
1106         int vector = apic_find_highest_isr(apic);
1107
1108         trace_kvm_eoi(apic, vector);
1109
1110         /*
1111          * Not every write EOI will has corresponding ISR,
1112          * one example is when Kernel check timer on setup_IO_APIC
1113          */
1114         if (vector == -1)
1115                 return vector;
1116
1117         apic_clear_isr(vector, apic);
1118         apic_update_ppr(apic);
1119
1120         if (test_bit(vector, vcpu_to_synic(apic->vcpu)->vec_bitmap))
1121                 kvm_hv_synic_send_eoi(apic->vcpu, vector);
1122
1123         kvm_ioapic_send_eoi(apic, vector);
1124         kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1125         return vector;
1126 }
1127
1128 /*
1129  * this interface assumes a trap-like exit, which has already finished
1130  * desired side effect including vISR and vPPR update.
1131  */
1132 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1133 {
1134         struct kvm_lapic *apic = vcpu->arch.apic;
1135
1136         trace_kvm_eoi(apic, vector);
1137
1138         kvm_ioapic_send_eoi(apic, vector);
1139         kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1140 }
1141 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1142
1143 static void apic_send_ipi(struct kvm_lapic *apic)
1144 {
1145         u32 icr_low = kvm_lapic_get_reg(apic, APIC_ICR);
1146         u32 icr_high = kvm_lapic_get_reg(apic, APIC_ICR2);
1147         struct kvm_lapic_irq irq;
1148
1149         irq.vector = icr_low & APIC_VECTOR_MASK;
1150         irq.delivery_mode = icr_low & APIC_MODE_MASK;
1151         irq.dest_mode = icr_low & APIC_DEST_MASK;
1152         irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1153         irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1154         irq.shorthand = icr_low & APIC_SHORT_MASK;
1155         irq.msi_redir_hint = false;
1156         if (apic_x2apic_mode(apic))
1157                 irq.dest_id = icr_high;
1158         else
1159                 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
1160
1161         trace_kvm_apic_ipi(icr_low, irq.dest_id);
1162
1163         apic_debug("icr_high 0x%x, icr_low 0x%x, "
1164                    "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
1165                    "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x, "
1166                    "msi_redir_hint 0x%x\n",
1167                    icr_high, icr_low, irq.shorthand, irq.dest_id,
1168                    irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
1169                    irq.vector, irq.msi_redir_hint);
1170
1171         kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1172 }
1173
1174 static u32 apic_get_tmcct(struct kvm_lapic *apic)
1175 {
1176         ktime_t remaining, now;
1177         s64 ns;
1178         u32 tmcct;
1179
1180         ASSERT(apic != NULL);
1181
1182         /* if initial count is 0, current count should also be 0 */
1183         if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1184                 apic->lapic_timer.period == 0)
1185                 return 0;
1186
1187         now = ktime_get();
1188         remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1189         if (ktime_to_ns(remaining) < 0)
1190                 remaining = 0;
1191
1192         ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1193         tmcct = div64_u64(ns,
1194                          (APIC_BUS_CYCLE_NS * apic->divide_count));
1195
1196         return tmcct;
1197 }
1198
1199 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1200 {
1201         struct kvm_vcpu *vcpu = apic->vcpu;
1202         struct kvm_run *run = vcpu->run;
1203
1204         kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1205         run->tpr_access.rip = kvm_rip_read(vcpu);
1206         run->tpr_access.is_write = write;
1207 }
1208
1209 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1210 {
1211         if (apic->vcpu->arch.tpr_access_reporting)
1212                 __report_tpr_access(apic, write);
1213 }
1214
1215 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1216 {
1217         u32 val = 0;
1218
1219         if (offset >= LAPIC_MMIO_LENGTH)
1220                 return 0;
1221
1222         switch (offset) {
1223         case APIC_ARBPRI:
1224                 apic_debug("Access APIC ARBPRI register which is for P6\n");
1225                 break;
1226
1227         case APIC_TMCCT:        /* Timer CCR */
1228                 if (apic_lvtt_tscdeadline(apic))
1229                         return 0;
1230
1231                 val = apic_get_tmcct(apic);
1232                 break;
1233         case APIC_PROCPRI:
1234                 apic_update_ppr(apic);
1235                 val = kvm_lapic_get_reg(apic, offset);
1236                 break;
1237         case APIC_TASKPRI:
1238                 report_tpr_access(apic, false);
1239                 /* fall thru */
1240         default:
1241                 val = kvm_lapic_get_reg(apic, offset);
1242                 break;
1243         }
1244
1245         return val;
1246 }
1247
1248 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1249 {
1250         return container_of(dev, struct kvm_lapic, dev);
1251 }
1252
1253 int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1254                 void *data)
1255 {
1256         unsigned char alignment = offset & 0xf;
1257         u32 result;
1258         /* this bitmask has a bit cleared for each reserved register */
1259         static const u64 rmask = 0x43ff01ffffffe70cULL;
1260
1261         if ((alignment + len) > 4) {
1262                 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1263                            offset, len);
1264                 return 1;
1265         }
1266
1267         if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1268                 apic_debug("KVM_APIC_READ: read reserved register %x\n",
1269                            offset);
1270                 return 1;
1271         }
1272
1273         result = __apic_read(apic, offset & ~0xf);
1274
1275         trace_kvm_apic_read(offset, result);
1276
1277         switch (len) {
1278         case 1:
1279         case 2:
1280         case 4:
1281                 memcpy(data, (char *)&result + alignment, len);
1282                 break;
1283         default:
1284                 printk(KERN_ERR "Local APIC read with len = %x, "
1285                        "should be 1,2, or 4 instead\n", len);
1286                 break;
1287         }
1288         return 0;
1289 }
1290 EXPORT_SYMBOL_GPL(kvm_lapic_reg_read);
1291
1292 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1293 {
1294         return kvm_apic_hw_enabled(apic) &&
1295             addr >= apic->base_address &&
1296             addr < apic->base_address + LAPIC_MMIO_LENGTH;
1297 }
1298
1299 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1300                            gpa_t address, int len, void *data)
1301 {
1302         struct kvm_lapic *apic = to_lapic(this);
1303         u32 offset = address - apic->base_address;
1304
1305         if (!apic_mmio_in_range(apic, address))
1306                 return -EOPNOTSUPP;
1307
1308         kvm_lapic_reg_read(apic, offset, len, data);
1309
1310         return 0;
1311 }
1312
1313 static void update_divide_count(struct kvm_lapic *apic)
1314 {
1315         u32 tmp1, tmp2, tdcr;
1316
1317         tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1318         tmp1 = tdcr & 0xf;
1319         tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1320         apic->divide_count = 0x1 << (tmp2 & 0x7);
1321
1322         apic_debug("timer divide count is 0x%x\n",
1323                                    apic->divide_count);
1324 }
1325
1326 static void limit_periodic_timer_frequency(struct kvm_lapic *apic)
1327 {
1328         /*
1329          * Do not allow the guest to program periodic timers with small
1330          * interval, since the hrtimers are not throttled by the host
1331          * scheduler.
1332          */
1333         if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1334                 s64 min_period = min_timer_period_us * 1000LL;
1335
1336                 if (apic->lapic_timer.period < min_period) {
1337                         pr_info_ratelimited(
1338                             "kvm: vcpu %i: requested %lld ns "
1339                             "lapic timer period limited to %lld ns\n",
1340                             apic->vcpu->vcpu_id,
1341                             apic->lapic_timer.period, min_period);
1342                         apic->lapic_timer.period = min_period;
1343                 }
1344         }
1345 }
1346
1347 static void apic_update_lvtt(struct kvm_lapic *apic)
1348 {
1349         u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1350                         apic->lapic_timer.timer_mode_mask;
1351
1352         if (apic->lapic_timer.timer_mode != timer_mode) {
1353                 if (apic_lvtt_tscdeadline(apic) != (timer_mode ==
1354                                 APIC_LVT_TIMER_TSCDEADLINE)) {
1355                         hrtimer_cancel(&apic->lapic_timer.timer);
1356                         kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1357                         apic->lapic_timer.period = 0;
1358                         apic->lapic_timer.tscdeadline = 0;
1359                 }
1360                 apic->lapic_timer.timer_mode = timer_mode;
1361                 limit_periodic_timer_frequency(apic);
1362         }
1363 }
1364
1365 static void apic_timer_expired(struct kvm_lapic *apic)
1366 {
1367         struct kvm_vcpu *vcpu = apic->vcpu;
1368         struct swait_queue_head *q = &vcpu->wq;
1369         struct kvm_timer *ktimer = &apic->lapic_timer;
1370
1371         if (atomic_read(&apic->lapic_timer.pending))
1372                 return;
1373
1374         atomic_inc(&apic->lapic_timer.pending);
1375         kvm_set_pending_timer(vcpu);
1376
1377         /*
1378          * For x86, the atomic_inc() is serialized, thus
1379          * using swait_active() is safe.
1380          */
1381         if (swait_active(q))
1382                 swake_up(q);
1383
1384         if (apic_lvtt_tscdeadline(apic))
1385                 ktimer->expired_tscdeadline = ktimer->tscdeadline;
1386 }
1387
1388 /*
1389  * On APICv, this test will cause a busy wait
1390  * during a higher-priority task.
1391  */
1392
1393 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1394 {
1395         struct kvm_lapic *apic = vcpu->arch.apic;
1396         u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1397
1398         if (kvm_apic_hw_enabled(apic)) {
1399                 int vec = reg & APIC_VECTOR_MASK;
1400                 void *bitmap = apic->regs + APIC_ISR;
1401
1402                 if (vcpu->arch.apicv_active)
1403                         bitmap = apic->regs + APIC_IRR;
1404
1405                 if (apic_test_vector(vec, bitmap))
1406                         return true;
1407         }
1408         return false;
1409 }
1410
1411 void wait_lapic_expire(struct kvm_vcpu *vcpu)
1412 {
1413         struct kvm_lapic *apic = vcpu->arch.apic;
1414         u64 guest_tsc, tsc_deadline;
1415
1416         if (!lapic_in_kernel(vcpu))
1417                 return;
1418
1419         if (apic->lapic_timer.expired_tscdeadline == 0)
1420                 return;
1421
1422         if (!lapic_timer_int_injected(vcpu))
1423                 return;
1424
1425         tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1426         apic->lapic_timer.expired_tscdeadline = 0;
1427         guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1428         trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1429
1430         /* __delay is delay_tsc whenever the hardware has TSC, thus always.  */
1431         if (guest_tsc < tsc_deadline)
1432                 __delay(min(tsc_deadline - guest_tsc,
1433                         nsec_to_cycles(vcpu, lapic_timer_advance_ns)));
1434 }
1435
1436 static void start_sw_tscdeadline(struct kvm_lapic *apic)
1437 {
1438         u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1439         u64 ns = 0;
1440         ktime_t expire;
1441         struct kvm_vcpu *vcpu = apic->vcpu;
1442         unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1443         unsigned long flags;
1444         ktime_t now;
1445
1446         if (unlikely(!tscdeadline || !this_tsc_khz))
1447                 return;
1448
1449         local_irq_save(flags);
1450
1451         now = ktime_get();
1452         guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1453         if (likely(tscdeadline > guest_tsc)) {
1454                 ns = (tscdeadline - guest_tsc) * 1000000ULL;
1455                 do_div(ns, this_tsc_khz);
1456                 expire = ktime_add_ns(now, ns);
1457                 expire = ktime_sub_ns(expire, lapic_timer_advance_ns);
1458                 hrtimer_start(&apic->lapic_timer.timer,
1459                                 expire, HRTIMER_MODE_ABS_PINNED);
1460         } else
1461                 apic_timer_expired(apic);
1462
1463         local_irq_restore(flags);
1464 }
1465
1466 static void update_target_expiration(struct kvm_lapic *apic, uint32_t old_divisor)
1467 {
1468         ktime_t now, remaining;
1469         u64 ns_remaining_old, ns_remaining_new;
1470
1471         apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
1472                 * APIC_BUS_CYCLE_NS * apic->divide_count;
1473         limit_periodic_timer_frequency(apic);
1474
1475         now = ktime_get();
1476         remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1477         if (ktime_to_ns(remaining) < 0)
1478                 remaining = 0;
1479
1480         ns_remaining_old = ktime_to_ns(remaining);
1481         ns_remaining_new = mul_u64_u32_div(ns_remaining_old,
1482                                            apic->divide_count, old_divisor);
1483
1484         apic->lapic_timer.tscdeadline +=
1485                 nsec_to_cycles(apic->vcpu, ns_remaining_new) -
1486                 nsec_to_cycles(apic->vcpu, ns_remaining_old);
1487         apic->lapic_timer.target_expiration = ktime_add_ns(now, ns_remaining_new);
1488 }
1489
1490 static bool set_target_expiration(struct kvm_lapic *apic)
1491 {
1492         ktime_t now;
1493         u64 tscl = rdtsc();
1494
1495         now = ktime_get();
1496         apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
1497                 * APIC_BUS_CYCLE_NS * apic->divide_count;
1498
1499         if (!apic->lapic_timer.period) {
1500                 apic->lapic_timer.tscdeadline = 0;
1501                 return false;
1502         }
1503
1504         limit_periodic_timer_frequency(apic);
1505
1506         apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1507                    PRIx64 ", "
1508                    "timer initial count 0x%x, period %lldns, "
1509                    "expire @ 0x%016" PRIx64 ".\n", __func__,
1510                    APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1511                    kvm_lapic_get_reg(apic, APIC_TMICT),
1512                    apic->lapic_timer.period,
1513                    ktime_to_ns(ktime_add_ns(now,
1514                                 apic->lapic_timer.period)));
1515
1516         apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1517                 nsec_to_cycles(apic->vcpu, apic->lapic_timer.period);
1518         apic->lapic_timer.target_expiration = ktime_add_ns(now, apic->lapic_timer.period);
1519
1520         return true;
1521 }
1522
1523 static void advance_periodic_target_expiration(struct kvm_lapic *apic)
1524 {
1525         apic->lapic_timer.tscdeadline +=
1526                 nsec_to_cycles(apic->vcpu, apic->lapic_timer.period);
1527         apic->lapic_timer.target_expiration =
1528                 ktime_add_ns(apic->lapic_timer.target_expiration,
1529                                 apic->lapic_timer.period);
1530 }
1531
1532 static void start_sw_period(struct kvm_lapic *apic)
1533 {
1534         if (!apic->lapic_timer.period)
1535                 return;
1536
1537         if (ktime_after(ktime_get(),
1538                         apic->lapic_timer.target_expiration)) {
1539                 apic_timer_expired(apic);
1540
1541                 if (apic_lvtt_oneshot(apic))
1542                         return;
1543
1544                 advance_periodic_target_expiration(apic);
1545         }
1546
1547         hrtimer_start(&apic->lapic_timer.timer,
1548                 apic->lapic_timer.target_expiration,
1549                 HRTIMER_MODE_ABS_PINNED);
1550 }
1551
1552 bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1553 {
1554         if (!lapic_in_kernel(vcpu))
1555                 return false;
1556
1557         return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1558 }
1559 EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use);
1560
1561 static void cancel_hv_timer(struct kvm_lapic *apic)
1562 {
1563         WARN_ON(preemptible());
1564         WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1565         kvm_x86_ops->cancel_hv_timer(apic->vcpu);
1566         apic->lapic_timer.hv_timer_in_use = false;
1567 }
1568
1569 static bool start_hv_timer(struct kvm_lapic *apic)
1570 {
1571         struct kvm_timer *ktimer = &apic->lapic_timer;
1572         int r;
1573
1574         WARN_ON(preemptible());
1575         if (!kvm_x86_ops->set_hv_timer)
1576                 return false;
1577
1578         if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
1579                 return false;
1580
1581         if (!ktimer->tscdeadline)
1582                 return false;
1583
1584         r = kvm_x86_ops->set_hv_timer(apic->vcpu, ktimer->tscdeadline);
1585         if (r < 0)
1586                 return false;
1587
1588         ktimer->hv_timer_in_use = true;
1589         hrtimer_cancel(&ktimer->timer);
1590
1591         /*
1592          * Also recheck ktimer->pending, in case the sw timer triggered in
1593          * the window.  For periodic timer, leave the hv timer running for
1594          * simplicity, and the deadline will be recomputed on the next vmexit.
1595          */
1596         if (!apic_lvtt_period(apic) && (r || atomic_read(&ktimer->pending))) {
1597                 if (r)
1598                         apic_timer_expired(apic);
1599                 return false;
1600         }
1601
1602         trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, true);
1603         return true;
1604 }
1605
1606 static void start_sw_timer(struct kvm_lapic *apic)
1607 {
1608         struct kvm_timer *ktimer = &apic->lapic_timer;
1609
1610         WARN_ON(preemptible());
1611         if (apic->lapic_timer.hv_timer_in_use)
1612                 cancel_hv_timer(apic);
1613         if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
1614                 return;
1615
1616         if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1617                 start_sw_period(apic);
1618         else if (apic_lvtt_tscdeadline(apic))
1619                 start_sw_tscdeadline(apic);
1620         trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, false);
1621 }
1622
1623 static void restart_apic_timer(struct kvm_lapic *apic)
1624 {
1625         preempt_disable();
1626         if (!start_hv_timer(apic))
1627                 start_sw_timer(apic);
1628         preempt_enable();
1629 }
1630
1631 void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1632 {
1633         struct kvm_lapic *apic = vcpu->arch.apic;
1634
1635         preempt_disable();
1636         /* If the preempt notifier has already run, it also called apic_timer_expired */
1637         if (!apic->lapic_timer.hv_timer_in_use)
1638                 goto out;
1639         WARN_ON(swait_active(&vcpu->wq));
1640         cancel_hv_timer(apic);
1641         apic_timer_expired(apic);
1642
1643         if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1644                 advance_periodic_target_expiration(apic);
1645                 restart_apic_timer(apic);
1646         }
1647 out:
1648         preempt_enable();
1649 }
1650 EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
1651
1652 void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
1653 {
1654         restart_apic_timer(vcpu->arch.apic);
1655 }
1656 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer);
1657
1658 void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
1659 {
1660         struct kvm_lapic *apic = vcpu->arch.apic;
1661
1662         preempt_disable();
1663         /* Possibly the TSC deadline timer is not enabled yet */
1664         if (apic->lapic_timer.hv_timer_in_use)
1665                 start_sw_timer(apic);
1666         preempt_enable();
1667 }
1668 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer);
1669
1670 void kvm_lapic_restart_hv_timer(struct kvm_vcpu *vcpu)
1671 {
1672         struct kvm_lapic *apic = vcpu->arch.apic;
1673
1674         WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1675         restart_apic_timer(apic);
1676 }
1677
1678 static void start_apic_timer(struct kvm_lapic *apic)
1679 {
1680         atomic_set(&apic->lapic_timer.pending, 0);
1681
1682         if ((apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1683             && !set_target_expiration(apic))
1684                 return;
1685
1686         restart_apic_timer(apic);
1687 }
1688
1689 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1690 {
1691         bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
1692
1693         if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1694                 apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1695                 if (lvt0_in_nmi_mode) {
1696                         apic_debug("Receive NMI setting on APIC_LVT0 "
1697                                    "for cpu %d\n", apic->vcpu->vcpu_id);
1698                         atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1699                 } else
1700                         atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1701         }
1702 }
1703
1704 int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1705 {
1706         int ret = 0;
1707
1708         trace_kvm_apic_write(reg, val);
1709
1710         switch (reg) {
1711         case APIC_ID:           /* Local APIC ID */
1712                 if (!apic_x2apic_mode(apic))
1713                         kvm_apic_set_xapic_id(apic, val >> 24);
1714                 else
1715                         ret = 1;
1716                 break;
1717
1718         case APIC_TASKPRI:
1719                 report_tpr_access(apic, true);
1720                 apic_set_tpr(apic, val & 0xff);
1721                 break;
1722
1723         case APIC_EOI:
1724                 apic_set_eoi(apic);
1725                 break;
1726
1727         case APIC_LDR:
1728                 if (!apic_x2apic_mode(apic))
1729                         kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1730                 else
1731                         ret = 1;
1732                 break;
1733
1734         case APIC_DFR:
1735                 if (!apic_x2apic_mode(apic)) {
1736                         kvm_lapic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1737                         recalculate_apic_map(apic->vcpu->kvm);
1738                 } else
1739                         ret = 1;
1740                 break;
1741
1742         case APIC_SPIV: {
1743                 u32 mask = 0x3ff;
1744                 if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1745                         mask |= APIC_SPIV_DIRECTED_EOI;
1746                 apic_set_spiv(apic, val & mask);
1747                 if (!(val & APIC_SPIV_APIC_ENABLED)) {
1748                         int i;
1749                         u32 lvt_val;
1750
1751                         for (i = 0; i < KVM_APIC_LVT_NUM; i++) {
1752                                 lvt_val = kvm_lapic_get_reg(apic,
1753                                                        APIC_LVTT + 0x10 * i);
1754                                 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i,
1755                                              lvt_val | APIC_LVT_MASKED);
1756                         }
1757                         apic_update_lvtt(apic);
1758                         atomic_set(&apic->lapic_timer.pending, 0);
1759
1760                 }
1761                 break;
1762         }
1763         case APIC_ICR:
1764                 /* No delay here, so we always clear the pending bit */
1765                 kvm_lapic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1766                 apic_send_ipi(apic);
1767                 break;
1768
1769         case APIC_ICR2:
1770                 if (!apic_x2apic_mode(apic))
1771                         val &= 0xff000000;
1772                 kvm_lapic_set_reg(apic, APIC_ICR2, val);
1773                 break;
1774
1775         case APIC_LVT0:
1776                 apic_manage_nmi_watchdog(apic, val);
1777         case APIC_LVTTHMR:
1778         case APIC_LVTPC:
1779         case APIC_LVT1:
1780         case APIC_LVTERR:
1781                 /* TODO: Check vector */
1782                 if (!kvm_apic_sw_enabled(apic))
1783                         val |= APIC_LVT_MASKED;
1784
1785                 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1786                 kvm_lapic_set_reg(apic, reg, val);
1787
1788                 break;
1789
1790         case APIC_LVTT:
1791                 if (!kvm_apic_sw_enabled(apic))
1792                         val |= APIC_LVT_MASKED;
1793                 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1794                 kvm_lapic_set_reg(apic, APIC_LVTT, val);
1795                 apic_update_lvtt(apic);
1796                 break;
1797
1798         case APIC_TMICT:
1799                 if (apic_lvtt_tscdeadline(apic))
1800                         break;
1801
1802                 hrtimer_cancel(&apic->lapic_timer.timer);
1803                 kvm_lapic_set_reg(apic, APIC_TMICT, val);
1804                 start_apic_timer(apic);
1805                 break;
1806
1807         case APIC_TDCR: {
1808                 uint32_t old_divisor = apic->divide_count;
1809
1810                 if (val & 4)
1811                         apic_debug("KVM_WRITE:TDCR %x\n", val);
1812                 kvm_lapic_set_reg(apic, APIC_TDCR, val);
1813                 update_divide_count(apic);
1814                 if (apic->divide_count != old_divisor &&
1815                                 apic->lapic_timer.period) {
1816                         hrtimer_cancel(&apic->lapic_timer.timer);
1817                         update_target_expiration(apic, old_divisor);
1818                         restart_apic_timer(apic);
1819                 }
1820                 break;
1821         }
1822         case APIC_ESR:
1823                 if (apic_x2apic_mode(apic) && val != 0) {
1824                         apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1825                         ret = 1;
1826                 }
1827                 break;
1828
1829         case APIC_SELF_IPI:
1830                 if (apic_x2apic_mode(apic)) {
1831                         kvm_lapic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1832                 } else
1833                         ret = 1;
1834                 break;
1835         default:
1836                 ret = 1;
1837                 break;
1838         }
1839         if (ret)
1840                 apic_debug("Local APIC Write to read-only register %x\n", reg);
1841         return ret;
1842 }
1843 EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
1844
1845 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1846                             gpa_t address, int len, const void *data)
1847 {
1848         struct kvm_lapic *apic = to_lapic(this);
1849         unsigned int offset = address - apic->base_address;
1850         u32 val;
1851
1852         if (!apic_mmio_in_range(apic, address))
1853                 return -EOPNOTSUPP;
1854
1855         /*
1856          * APIC register must be aligned on 128-bits boundary.
1857          * 32/64/128 bits registers must be accessed thru 32 bits.
1858          * Refer SDM 8.4.1
1859          */
1860         if (len != 4 || (offset & 0xf)) {
1861                 /* Don't shout loud, $infamous_os would cause only noise. */
1862                 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1863                 return 0;
1864         }
1865
1866         val = *(u32*)data;
1867
1868         /* too common printing */
1869         if (offset != APIC_EOI)
1870                 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1871                            "0x%x\n", __func__, offset, len, val);
1872
1873         kvm_lapic_reg_write(apic, offset & 0xff0, val);
1874
1875         return 0;
1876 }
1877
1878 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1879 {
1880         kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1881 }
1882 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1883
1884 /* emulate APIC access in a trap manner */
1885 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1886 {
1887         u32 val = 0;
1888
1889         /* hw has done the conditional check and inst decode */
1890         offset &= 0xff0;
1891
1892         kvm_lapic_reg_read(vcpu->arch.apic, offset, 4, &val);
1893
1894         /* TODO: optimize to just emulate side effect w/o one more write */
1895         kvm_lapic_reg_write(vcpu->arch.apic, offset, val);
1896 }
1897 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1898
1899 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1900 {
1901         struct kvm_lapic *apic = vcpu->arch.apic;
1902
1903         if (!vcpu->arch.apic)
1904                 return;
1905
1906         hrtimer_cancel(&apic->lapic_timer.timer);
1907
1908         if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1909                 static_key_slow_dec_deferred(&apic_hw_disabled);
1910
1911         if (!apic->sw_enabled)
1912                 static_key_slow_dec_deferred(&apic_sw_disabled);
1913
1914         if (apic->regs)
1915                 free_page((unsigned long)apic->regs);
1916
1917         kfree(apic);
1918 }
1919
1920 /*
1921  *----------------------------------------------------------------------
1922  * LAPIC interface
1923  *----------------------------------------------------------------------
1924  */
1925 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1926 {
1927         struct kvm_lapic *apic = vcpu->arch.apic;
1928
1929         if (!lapic_in_kernel(vcpu) ||
1930                 !apic_lvtt_tscdeadline(apic))
1931                 return 0;
1932
1933         return apic->lapic_timer.tscdeadline;
1934 }
1935
1936 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1937 {
1938         struct kvm_lapic *apic = vcpu->arch.apic;
1939
1940         if (!lapic_in_kernel(vcpu) || apic_lvtt_oneshot(apic) ||
1941                         apic_lvtt_period(apic))
1942                 return;
1943
1944         hrtimer_cancel(&apic->lapic_timer.timer);
1945         apic->lapic_timer.tscdeadline = data;
1946         start_apic_timer(apic);
1947 }
1948
1949 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1950 {
1951         struct kvm_lapic *apic = vcpu->arch.apic;
1952
1953         apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1954                      | (kvm_lapic_get_reg(apic, APIC_TASKPRI) & 4));
1955 }
1956
1957 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1958 {
1959         u64 tpr;
1960
1961         tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1962
1963         return (tpr & 0xf0) >> 4;
1964 }
1965
1966 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1967 {
1968         u64 old_value = vcpu->arch.apic_base;
1969         struct kvm_lapic *apic = vcpu->arch.apic;
1970
1971         if (!apic)
1972                 value |= MSR_IA32_APICBASE_BSP;
1973
1974         vcpu->arch.apic_base = value;
1975
1976         if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
1977                 kvm_update_cpuid(vcpu);
1978
1979         if (!apic)
1980                 return;
1981
1982         /* update jump label if enable bit changes */
1983         if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1984                 if (value & MSR_IA32_APICBASE_ENABLE) {
1985                         kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
1986                         static_key_slow_dec_deferred(&apic_hw_disabled);
1987                 } else {
1988                         static_key_slow_inc(&apic_hw_disabled.key);
1989                         recalculate_apic_map(vcpu->kvm);
1990                 }
1991         }
1992
1993         if (((old_value ^ value) & X2APIC_ENABLE) && (value & X2APIC_ENABLE))
1994                 kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
1995
1996         if ((old_value ^ value) & (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE))
1997                 kvm_x86_ops->set_virtual_apic_mode(vcpu);
1998
1999         apic->base_address = apic->vcpu->arch.apic_base &
2000                              MSR_IA32_APICBASE_BASE;
2001
2002         if ((value & MSR_IA32_APICBASE_ENABLE) &&
2003              apic->base_address != APIC_DEFAULT_PHYS_BASE)
2004                 pr_warn_once("APIC base relocation is unsupported by KVM");
2005
2006         /* with FSB delivery interrupt, we can restart APIC functionality */
2007         apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
2008                    "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
2009
2010 }
2011
2012 void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
2013 {
2014         struct kvm_lapic *apic = vcpu->arch.apic;
2015         int i;
2016
2017         if (!apic)
2018                 return;
2019
2020         apic_debug("%s\n", __func__);
2021
2022         /* Stop the timer in case it's a reset to an active apic */
2023         hrtimer_cancel(&apic->lapic_timer.timer);
2024
2025         if (!init_event) {
2026                 kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
2027                                          MSR_IA32_APICBASE_ENABLE);
2028                 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2029         }
2030         kvm_apic_set_version(apic->vcpu);
2031
2032         for (i = 0; i < KVM_APIC_LVT_NUM; i++)
2033                 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
2034         apic_update_lvtt(apic);
2035         if (kvm_vcpu_is_reset_bsp(vcpu) &&
2036             kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
2037                 kvm_lapic_set_reg(apic, APIC_LVT0,
2038                              SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
2039         apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2040
2041         kvm_lapic_set_reg(apic, APIC_DFR, 0xffffffffU);
2042         apic_set_spiv(apic, 0xff);
2043         kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
2044         if (!apic_x2apic_mode(apic))
2045                 kvm_apic_set_ldr(apic, 0);
2046         kvm_lapic_set_reg(apic, APIC_ESR, 0);
2047         kvm_lapic_set_reg(apic, APIC_ICR, 0);
2048         kvm_lapic_set_reg(apic, APIC_ICR2, 0);
2049         kvm_lapic_set_reg(apic, APIC_TDCR, 0);
2050         kvm_lapic_set_reg(apic, APIC_TMICT, 0);
2051         for (i = 0; i < 8; i++) {
2052                 kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
2053                 kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
2054                 kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
2055         }
2056         apic->irr_pending = vcpu->arch.apicv_active;
2057         apic->isr_count = vcpu->arch.apicv_active ? 1 : 0;
2058         apic->highest_isr_cache = -1;
2059         update_divide_count(apic);
2060         atomic_set(&apic->lapic_timer.pending, 0);
2061         if (kvm_vcpu_is_bsp(vcpu))
2062                 kvm_lapic_set_base(vcpu,
2063                                 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
2064         vcpu->arch.pv_eoi.msr_val = 0;
2065         apic_update_ppr(apic);
2066         if (vcpu->arch.apicv_active) {
2067                 kvm_x86_ops->apicv_post_state_restore(vcpu);
2068                 kvm_x86_ops->hwapic_irr_update(vcpu, -1);
2069                 kvm_x86_ops->hwapic_isr_update(vcpu, -1);
2070         }
2071
2072         vcpu->arch.apic_arb_prio = 0;
2073         vcpu->arch.apic_attention = 0;
2074
2075         apic_debug("%s: vcpu=%p, id=0x%x, base_msr="
2076                    "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
2077                    vcpu, kvm_lapic_get_reg(apic, APIC_ID),
2078                    vcpu->arch.apic_base, apic->base_address);
2079 }
2080
2081 /*
2082  *----------------------------------------------------------------------
2083  * timer interface
2084  *----------------------------------------------------------------------
2085  */
2086
2087 static bool lapic_is_periodic(struct kvm_lapic *apic)
2088 {
2089         return apic_lvtt_period(apic);
2090 }
2091
2092 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
2093 {
2094         struct kvm_lapic *apic = vcpu->arch.apic;
2095
2096         if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
2097                 return atomic_read(&apic->lapic_timer.pending);
2098
2099         return 0;
2100 }
2101
2102 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
2103 {
2104         u32 reg = kvm_lapic_get_reg(apic, lvt_type);
2105         int vector, mode, trig_mode;
2106
2107         if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
2108                 vector = reg & APIC_VECTOR_MASK;
2109                 mode = reg & APIC_MODE_MASK;
2110                 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
2111                 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
2112                                         NULL);
2113         }
2114         return 0;
2115 }
2116
2117 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
2118 {
2119         struct kvm_lapic *apic = vcpu->arch.apic;
2120
2121         if (apic)
2122                 kvm_apic_local_deliver(apic, APIC_LVT0);
2123 }
2124
2125 static const struct kvm_io_device_ops apic_mmio_ops = {
2126         .read     = apic_mmio_read,
2127         .write    = apic_mmio_write,
2128 };
2129
2130 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
2131 {
2132         struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
2133         struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
2134
2135         apic_timer_expired(apic);
2136
2137         if (lapic_is_periodic(apic)) {
2138                 advance_periodic_target_expiration(apic);
2139                 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
2140                 return HRTIMER_RESTART;
2141         } else
2142                 return HRTIMER_NORESTART;
2143 }
2144
2145 int kvm_create_lapic(struct kvm_vcpu *vcpu)
2146 {
2147         struct kvm_lapic *apic;
2148
2149         ASSERT(vcpu != NULL);
2150         apic_debug("apic_init %d\n", vcpu->vcpu_id);
2151
2152         apic = kzalloc(sizeof(*apic), GFP_KERNEL);
2153         if (!apic)
2154                 goto nomem;
2155
2156         vcpu->arch.apic = apic;
2157
2158         apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
2159         if (!apic->regs) {
2160                 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
2161                        vcpu->vcpu_id);
2162                 goto nomem_free_apic;
2163         }
2164         apic->vcpu = vcpu;
2165
2166         hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2167                      HRTIMER_MODE_ABS_PINNED);
2168         apic->lapic_timer.timer.function = apic_timer_fn;
2169
2170         /*
2171          * APIC is created enabled. This will prevent kvm_lapic_set_base from
2172          * thinking that APIC satet has changed.
2173          */
2174         vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2175         static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2176         kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2177
2178         return 0;
2179 nomem_free_apic:
2180         kfree(apic);
2181 nomem:
2182         return -ENOMEM;
2183 }
2184
2185 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2186 {
2187         struct kvm_lapic *apic = vcpu->arch.apic;
2188         u32 ppr;
2189
2190         if (!apic_enabled(apic))
2191                 return -1;
2192
2193         __apic_update_ppr(apic, &ppr);
2194         return apic_has_interrupt_for_ppr(apic, ppr);
2195 }
2196
2197 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2198 {
2199         u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2200         int r = 0;
2201
2202         if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2203                 r = 1;
2204         if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2205             GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2206                 r = 1;
2207         return r;
2208 }
2209
2210 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2211 {
2212         struct kvm_lapic *apic = vcpu->arch.apic;
2213
2214         if (atomic_read(&apic->lapic_timer.pending) > 0) {
2215                 kvm_apic_local_deliver(apic, APIC_LVTT);
2216                 if (apic_lvtt_tscdeadline(apic))
2217                         apic->lapic_timer.tscdeadline = 0;
2218                 if (apic_lvtt_oneshot(apic)) {
2219                         apic->lapic_timer.tscdeadline = 0;
2220                         apic->lapic_timer.target_expiration = 0;
2221                 }
2222                 atomic_set(&apic->lapic_timer.pending, 0);
2223         }
2224 }
2225
2226 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2227 {
2228         int vector = kvm_apic_has_interrupt(vcpu);
2229         struct kvm_lapic *apic = vcpu->arch.apic;
2230         u32 ppr;
2231
2232         if (vector == -1)
2233                 return -1;
2234
2235         /*
2236          * We get here even with APIC virtualization enabled, if doing
2237          * nested virtualization and L1 runs with the "acknowledge interrupt
2238          * on exit" mode.  Then we cannot inject the interrupt via RVI,
2239          * because the process would deliver it through the IDT.
2240          */
2241
2242         apic_clear_irr(vector, apic);
2243         if (test_bit(vector, vcpu_to_synic(vcpu)->auto_eoi_bitmap)) {
2244                 /*
2245                  * For auto-EOI interrupts, there might be another pending
2246                  * interrupt above PPR, so check whether to raise another
2247                  * KVM_REQ_EVENT.
2248                  */
2249                 apic_update_ppr(apic);
2250         } else {
2251                 /*
2252                  * For normal interrupts, PPR has been raised and there cannot
2253                  * be a higher-priority pending interrupt---except if there was
2254                  * a concurrent interrupt injection, but that would have
2255                  * triggered KVM_REQ_EVENT already.
2256                  */
2257                 apic_set_isr(vector, apic);
2258                 __apic_update_ppr(apic, &ppr);
2259         }
2260
2261         return vector;
2262 }
2263
2264 static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2265                 struct kvm_lapic_state *s, bool set)
2266 {
2267         if (apic_x2apic_mode(vcpu->arch.apic)) {
2268                 u32 *id = (u32 *)(s->regs + APIC_ID);
2269                 u32 *ldr = (u32 *)(s->regs + APIC_LDR);
2270
2271                 if (vcpu->kvm->arch.x2apic_format) {
2272                         if (*id != vcpu->vcpu_id)
2273                                 return -EINVAL;
2274                 } else {
2275                         if (set)
2276                                 *id >>= 24;
2277                         else
2278                                 *id <<= 24;
2279                 }
2280
2281                 /* In x2APIC mode, the LDR is fixed and based on the id */
2282                 if (set)
2283                         *ldr = kvm_apic_calc_x2apic_ldr(*id);
2284         }
2285
2286         return 0;
2287 }
2288
2289 int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2290 {
2291         memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2292         return kvm_apic_state_fixup(vcpu, s, false);
2293 }
2294
2295 int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2296 {
2297         struct kvm_lapic *apic = vcpu->arch.apic;
2298         int r;
2299
2300
2301         kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2302         /* set SPIV separately to get count of SW disabled APICs right */
2303         apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2304
2305         r = kvm_apic_state_fixup(vcpu, s, true);
2306         if (r)
2307                 return r;
2308         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2309
2310         recalculate_apic_map(vcpu->kvm);
2311         kvm_apic_set_version(vcpu);
2312
2313         apic_update_ppr(apic);
2314         hrtimer_cancel(&apic->lapic_timer.timer);
2315         apic_update_lvtt(apic);
2316         apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2317         update_divide_count(apic);
2318         start_apic_timer(apic);
2319         apic->irr_pending = true;
2320         apic->isr_count = vcpu->arch.apicv_active ?
2321                                 1 : count_vectors(apic->regs + APIC_ISR);
2322         apic->highest_isr_cache = -1;
2323         if (vcpu->arch.apicv_active) {
2324                 kvm_x86_ops->apicv_post_state_restore(vcpu);
2325                 kvm_x86_ops->hwapic_irr_update(vcpu,
2326                                 apic_find_highest_irr(apic));
2327                 kvm_x86_ops->hwapic_isr_update(vcpu,
2328                                 apic_find_highest_isr(apic));
2329         }
2330         kvm_make_request(KVM_REQ_EVENT, vcpu);
2331         if (ioapic_in_kernel(vcpu->kvm))
2332                 kvm_rtc_eoi_tracking_restore_one(vcpu);
2333
2334         vcpu->arch.apic_arb_prio = 0;
2335
2336         return 0;
2337 }
2338
2339 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2340 {
2341         struct hrtimer *timer;
2342
2343         if (!lapic_in_kernel(vcpu))
2344                 return;
2345
2346         timer = &vcpu->arch.apic->lapic_timer.timer;
2347         if (hrtimer_cancel(timer))
2348                 hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
2349 }
2350
2351 /*
2352  * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2353  *
2354  * Detect whether guest triggered PV EOI since the
2355  * last entry. If yes, set EOI on guests's behalf.
2356  * Clear PV EOI in guest memory in any case.
2357  */
2358 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2359                                         struct kvm_lapic *apic)
2360 {
2361         bool pending;
2362         int vector;
2363         /*
2364          * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2365          * and KVM_PV_EOI_ENABLED in guest memory as follows:
2366          *
2367          * KVM_APIC_PV_EOI_PENDING is unset:
2368          *      -> host disabled PV EOI.
2369          * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2370          *      -> host enabled PV EOI, guest did not execute EOI yet.
2371          * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2372          *      -> host enabled PV EOI, guest executed EOI.
2373          */
2374         BUG_ON(!pv_eoi_enabled(vcpu));
2375         pending = pv_eoi_get_pending(vcpu);
2376         /*
2377          * Clear pending bit in any case: it will be set again on vmentry.
2378          * While this might not be ideal from performance point of view,
2379          * this makes sure pv eoi is only enabled when we know it's safe.
2380          */
2381         pv_eoi_clr_pending(vcpu);
2382         if (pending)
2383                 return;
2384         vector = apic_set_eoi(apic);
2385         trace_kvm_pv_eoi(apic, vector);
2386 }
2387
2388 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2389 {
2390         u32 data;
2391
2392         if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2393                 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2394
2395         if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2396                 return;
2397
2398         if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2399                                   sizeof(u32)))
2400                 return;
2401
2402         apic_set_tpr(vcpu->arch.apic, data & 0xff);
2403 }
2404
2405 /*
2406  * apic_sync_pv_eoi_to_guest - called before vmentry
2407  *
2408  * Detect whether it's safe to enable PV EOI and
2409  * if yes do so.
2410  */
2411 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2412                                         struct kvm_lapic *apic)
2413 {
2414         if (!pv_eoi_enabled(vcpu) ||
2415             /* IRR set or many bits in ISR: could be nested. */
2416             apic->irr_pending ||
2417             /* Cache not set: could be safe but we don't bother. */
2418             apic->highest_isr_cache == -1 ||
2419             /* Need EOI to update ioapic. */
2420             kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2421                 /*
2422                  * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2423                  * so we need not do anything here.
2424                  */
2425                 return;
2426         }
2427
2428         pv_eoi_set_pending(apic->vcpu);
2429 }
2430
2431 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2432 {
2433         u32 data, tpr;
2434         int max_irr, max_isr;
2435         struct kvm_lapic *apic = vcpu->arch.apic;
2436
2437         apic_sync_pv_eoi_to_guest(vcpu, apic);
2438
2439         if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2440                 return;
2441
2442         tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
2443         max_irr = apic_find_highest_irr(apic);
2444         if (max_irr < 0)
2445                 max_irr = 0;
2446         max_isr = apic_find_highest_isr(apic);
2447         if (max_isr < 0)
2448                 max_isr = 0;
2449         data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2450
2451         kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2452                                 sizeof(u32));
2453 }
2454
2455 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2456 {
2457         if (vapic_addr) {
2458                 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2459                                         &vcpu->arch.apic->vapic_cache,
2460                                         vapic_addr, sizeof(u32)))
2461                         return -EINVAL;
2462                 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2463         } else {
2464                 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2465         }
2466
2467         vcpu->arch.apic->vapic_addr = vapic_addr;
2468         return 0;
2469 }
2470
2471 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2472 {
2473         struct kvm_lapic *apic = vcpu->arch.apic;
2474         u32 reg = (msr - APIC_BASE_MSR) << 4;
2475
2476         if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2477                 return 1;
2478
2479         if (reg == APIC_ICR2)
2480                 return 1;
2481
2482         /* if this is ICR write vector before command */
2483         if (reg == APIC_ICR)
2484                 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2485         return kvm_lapic_reg_write(apic, reg, (u32)data);
2486 }
2487
2488 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2489 {
2490         struct kvm_lapic *apic = vcpu->arch.apic;
2491         u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2492
2493         if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2494                 return 1;
2495
2496         if (reg == APIC_DFR || reg == APIC_ICR2) {
2497                 apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
2498                            reg);
2499                 return 1;
2500         }
2501
2502         if (kvm_lapic_reg_read(apic, reg, 4, &low))
2503                 return 1;
2504         if (reg == APIC_ICR)
2505                 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2506
2507         *data = (((u64)high) << 32) | low;
2508
2509         return 0;
2510 }
2511
2512 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2513 {
2514         struct kvm_lapic *apic = vcpu->arch.apic;
2515
2516         if (!lapic_in_kernel(vcpu))
2517                 return 1;
2518
2519         /* if this is ICR write vector before command */
2520         if (reg == APIC_ICR)
2521                 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2522         return kvm_lapic_reg_write(apic, reg, (u32)data);
2523 }
2524
2525 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2526 {
2527         struct kvm_lapic *apic = vcpu->arch.apic;
2528         u32 low, high = 0;
2529
2530         if (!lapic_in_kernel(vcpu))
2531                 return 1;
2532
2533         if (kvm_lapic_reg_read(apic, reg, 4, &low))
2534                 return 1;
2535         if (reg == APIC_ICR)
2536                 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2537
2538         *data = (((u64)high) << 32) | low;
2539
2540         return 0;
2541 }
2542
2543 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
2544 {
2545         u64 addr = data & ~KVM_MSR_ENABLED;
2546         if (!IS_ALIGNED(addr, 4))
2547                 return 1;
2548
2549         vcpu->arch.pv_eoi.msr_val = data;
2550         if (!pv_eoi_enabled(vcpu))
2551                 return 0;
2552         return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
2553                                          addr, sizeof(u8));
2554 }
2555
2556 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2557 {
2558         struct kvm_lapic *apic = vcpu->arch.apic;
2559         u8 sipi_vector;
2560         unsigned long pe;
2561
2562         if (!lapic_in_kernel(vcpu) || !apic->pending_events)
2563                 return;
2564
2565         /*
2566          * INITs are latched while in SMM.  Because an SMM CPU cannot
2567          * be in KVM_MP_STATE_INIT_RECEIVED state, just eat SIPIs
2568          * and delay processing of INIT until the next RSM.
2569          */
2570         if (is_smm(vcpu)) {
2571                 WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2572                 if (test_bit(KVM_APIC_SIPI, &apic->pending_events))
2573                         clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2574                 return;
2575         }
2576
2577         pe = xchg(&apic->pending_events, 0);
2578         if (test_bit(KVM_APIC_INIT, &pe)) {
2579                 kvm_vcpu_reset(vcpu, true);
2580                 if (kvm_vcpu_is_bsp(apic->vcpu))
2581                         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2582                 else
2583                         vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2584         }
2585         if (test_bit(KVM_APIC_SIPI, &pe) &&
2586             vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2587                 /* evaluate pending_events before reading the vector */
2588                 smp_rmb();
2589                 sipi_vector = apic->sipi_vector;
2590                 apic_debug("vcpu %d received sipi with vector # %x\n",
2591                          vcpu->vcpu_id, sipi_vector);
2592                 kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2593                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2594         }
2595 }
2596
2597 void kvm_lapic_init(void)
2598 {
2599         /* do not patch jump label more than once per second */
2600         jump_label_rate_limit(&apic_hw_disabled, HZ);
2601         jump_label_rate_limit(&apic_sw_disabled, HZ);
2602 }
2603
2604 void kvm_lapic_exit(void)
2605 {
2606         static_key_deferred_flush(&apic_hw_disabled);
2607         static_key_deferred_flush(&apic_sw_disabled);
2608 }