arch/x86/kvm/hyperv.c

   1 /*
   2  * KVM Microsoft Hyper-V emulation
   3  *
   4  * derived from arch/x86/kvm/x86.c
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  * Copyright (C) 2008 Qumranet, Inc.
   8  * Copyright IBM Corporation, 2008
   9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
  10  * Copyright (C) 2015 Andrey Smetanin <asmetanin@virtuozzo.com>
  11  *
  12  * Authors:
  13  *   Avi Kivity   <avi@qumranet.com>
  14  *   Yaniv Kamay  <yaniv@qumranet.com>
  15  *   Amit Shah    <amit.shah@qumranet.com>
  16  *   Ben-Ami Yassour <benami@il.ibm.com>
  17  *   Andrey Smetanin <asmetanin@virtuozzo.com>
  18  *
  19  * This work is licensed under the terms of the GNU GPL, version 2.  See
  20  * the COPYING file in the top-level directory.
  21  *
  22  */
  23
  24 #include "x86.h"
  25 #include "lapic.h"
  26 #include "ioapic.h"
  27 #include "hyperv.h"
  28
  29 #include <linux/kvm_host.h>
  30 #include <linux/highmem.h>
  31 #include <linux/sched/cputime.h>
  32
  33 #include <asm/apicdef.h>
  34 #include <trace/events/kvm.h>
  35
  36 #include "trace.h"
  37
  38 static inline u64 synic_read_sint(struct kvm_vcpu_hv_synic *synic, int sint)
  39 {
  40         return atomic64_read(&synic->sint[sint]);
  41 }
  42
  43 static inline int synic_get_sint_vector(u64 sint_value)
  44 {
  45         if (sint_value & HV_SYNIC_SINT_MASKED)
  46                 return -1;
  47         return sint_value & HV_SYNIC_SINT_VECTOR_MASK;
  48 }
  49
  50 static bool synic_has_vector_connected(struct kvm_vcpu_hv_synic *synic,
  51                                       int vector)
  52 {
  53         int i;
  54
  55         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
  56                 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
  57                         return true;
  58         }
  59         return false;
  60 }
  61
  62 static bool synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic *synic,
  63                                      int vector)
  64 {
  65         int i;
  66         u64 sint_value;
  67
  68         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
  69                 sint_value = synic_read_sint(synic, i);
  70                 if (synic_get_sint_vector(sint_value) == vector &&
  71                     sint_value & HV_SYNIC_SINT_AUTO_EOI)
  72                         return true;
  73         }
  74         return false;
  75 }
  76
  77 static int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint,
  78                           u64 data, bool host)
  79 {
  80         int vector;
  81
  82         vector = data & HV_SYNIC_SINT_VECTOR_MASK;
  83         if (vector < 16 && !host)
  84                 return 1;
  85         /*
  86          * Guest may configure multiple SINTs to use the same vector, so
  87          * we maintain a bitmap of vectors handled by synic, and a
  88          * bitmap of vectors with auto-eoi behavior.  The bitmaps are
  89          * updated here, and atomically queried on fast paths.
  90          */
  91
  92         atomic64_set(&synic->sint[sint], data);
  93
  94         if (synic_has_vector_connected(synic, vector))
  95                 __set_bit(vector, synic->vec_bitmap);
  96         else
  97                 __clear_bit(vector, synic->vec_bitmap);
  98
  99         if (synic_has_vector_auto_eoi(synic, vector))
 100                 __set_bit(vector, synic->auto_eoi_bitmap);
 101         else
 102                 __clear_bit(vector, synic->auto_eoi_bitmap);
 103
 104         /* Load SynIC vectors into EOI exit bitmap */
 105         kvm_make_request(KVM_REQ_SCAN_IOAPIC, synic_to_vcpu(synic));
 106         return 0;
 107 }
 108
 109 static struct kvm_vcpu *get_vcpu_by_vpidx(struct kvm *kvm, u32 vpidx)
 110 {
 111         struct kvm_vcpu *vcpu = NULL;
 112         int i;
 113
 114         if (vpidx < KVM_MAX_VCPUS)
 115                 vcpu = kvm_get_vcpu(kvm, vpidx);
 116         if (vcpu && vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
 117                 return vcpu;
 118         kvm_for_each_vcpu(i, vcpu, kvm)
 119                 if (vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
 120                         return vcpu;
 121         return NULL;
 122 }
 123
 124 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vpidx)
 125 {
 126         struct kvm_vcpu *vcpu;
 127         struct kvm_vcpu_hv_synic *synic;
 128
 129         vcpu = get_vcpu_by_vpidx(kvm, vpidx);
 130         if (!vcpu)
 131                 return NULL;
 132         synic = vcpu_to_synic(vcpu);
 133         return (synic->active) ? synic : NULL;
 134 }
 135
 136 static void synic_clear_sint_msg_pending(struct kvm_vcpu_hv_synic *synic,
 137                                         u32 sint)
 138 {
 139         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 140         struct page *page;
 141         gpa_t gpa;
 142         struct hv_message *msg;
 143         struct hv_message_page *msg_page;
 144
 145         gpa = synic->msg_page & PAGE_MASK;
 146         page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
 147         if (is_error_page(page)) {
 148                 vcpu_err(vcpu, "Hyper-V SynIC can't get msg page, gpa 0x%llx\n",
 149                          gpa);
 150                 return;
 151         }
 152         msg_page = kmap_atomic(page);
 153
 154         msg = &msg_page->sint_message[sint];
 155         msg->header.message_flags.msg_pending = 0;
 156
 157         kunmap_atomic(msg_page);
 158         kvm_release_page_dirty(page);
 159         kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
 160 }
 161
 162 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
 163 {
 164         struct kvm *kvm = vcpu->kvm;
 165         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
 166         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 167         struct kvm_vcpu_hv_stimer *stimer;
 168         int gsi, idx, stimers_pending;
 169
 170         trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
 171
 172         if (synic->msg_page & HV_SYNIC_SIMP_ENABLE)
 173                 synic_clear_sint_msg_pending(synic, sint);
 174
 175         /* Try to deliver pending Hyper-V SynIC timers messages */
 176         stimers_pending = 0;
 177         for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
 178                 stimer = &hv_vcpu->stimer[idx];
 179                 if (stimer->msg_pending &&
 180                     (stimer->config & HV_STIMER_ENABLE) &&
 181                     HV_STIMER_SINT(stimer->config) == sint) {
 182                         set_bit(stimer->index,
 183                                 hv_vcpu->stimer_pending_bitmap);
 184                         stimers_pending++;
 185                 }
 186         }
 187         if (stimers_pending)
 188                 kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
 189
 190         idx = srcu_read_lock(&kvm->irq_srcu);
 191         gsi = atomic_read(&synic->sint_to_gsi[sint]);
 192         if (gsi != -1)
 193                 kvm_notify_acked_gsi(kvm, gsi);
 194         srcu_read_unlock(&kvm->irq_srcu, idx);
 195 }
 196
 197 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
 198 {
 199         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 200         struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
 201
 202         hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
 203         hv_vcpu->exit.u.synic.msr = msr;
 204         hv_vcpu->exit.u.synic.control = synic->control;
 205         hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
 206         hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
 207
 208         kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
 209 }
 210
 211 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
 212                          u32 msr, u64 data, bool host)
 213 {
 214         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 215         int ret;
 216
 217         if (!synic->active)
 218                 return 1;
 219
 220         trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
 221
 222         ret = 0;
 223         switch (msr) {
 224         case HV_X64_MSR_SCONTROL:
 225                 synic->control = data;
 226                 if (!host)
 227                         synic_exit(synic, msr);
 228                 break;
 229         case HV_X64_MSR_SVERSION:
 230                 if (!host) {
 231                         ret = 1;
 232                         break;
 233                 }
 234                 synic->version = data;
 235                 break;
 236         case HV_X64_MSR_SIEFP:
 237                 if ((data & HV_SYNIC_SIEFP_ENABLE) && !host &&
 238                     !synic->dont_zero_synic_pages)
 239                         if (kvm_clear_guest(vcpu->kvm,
 240                                             data & PAGE_MASK, PAGE_SIZE)) {
 241                                 ret = 1;
 242                                 break;
 243                         }
 244                 synic->evt_page = data;
 245                 if (!host)
 246                         synic_exit(synic, msr);
 247                 break;
 248         case HV_X64_MSR_SIMP:
 249                 if ((data & HV_SYNIC_SIMP_ENABLE) && !host &&
 250                     !synic->dont_zero_synic_pages)
 251                         if (kvm_clear_guest(vcpu->kvm,
 252                                             data & PAGE_MASK, PAGE_SIZE)) {
 253                                 ret = 1;
 254                                 break;
 255                         }
 256                 synic->msg_page = data;
 257                 if (!host)
 258                         synic_exit(synic, msr);
 259                 break;
 260         case HV_X64_MSR_EOM: {
 261                 int i;
 262
 263                 if (!synic->active)
 264                         break;
 265
 266                 for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
 267                         kvm_hv_notify_acked_sint(vcpu, i);
 268                 break;
 269         }
 270         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
 271                 ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
 272                 break;
 273         default:
 274                 ret = 1;
 275                 break;
 276         }
 277         return ret;
 278 }
 279
 280 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata)
 281 {
 282         int ret;
 283
 284         if (!synic->active)
 285                 return 1;
 286
 287         ret = 0;
 288         switch (msr) {
 289         case HV_X64_MSR_SCONTROL:
 290                 *pdata = synic->control;
 291                 break;
 292         case HV_X64_MSR_SVERSION:
 293                 *pdata = synic->version;
 294                 break;
 295         case HV_X64_MSR_SIEFP:
 296                 *pdata = synic->evt_page;
 297                 break;
 298         case HV_X64_MSR_SIMP:
 299                 *pdata = synic->msg_page;
 300                 break;
 301         case HV_X64_MSR_EOM:
 302                 *pdata = 0;
 303                 break;
 304         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
 305                 *pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
 306                 break;
 307         default:
 308                 ret = 1;
 309                 break;
 310         }
 311         return ret;
 312 }
 313
 314 static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
 315 {
 316         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 317         struct kvm_lapic_irq irq;
 318         int ret, vector;
 319
 320         if (sint >= ARRAY_SIZE(synic->sint))
 321                 return -EINVAL;
 322
 323         vector = synic_get_sint_vector(synic_read_sint(synic, sint));
 324         if (vector < 0)
 325                 return -ENOENT;
 326
 327         memset(&irq, 0, sizeof(irq));
 328         irq.shorthand = APIC_DEST_SELF;
 329         irq.dest_mode = APIC_DEST_PHYSICAL;
 330         irq.delivery_mode = APIC_DM_FIXED;
 331         irq.vector = vector;
 332         irq.level = 1;
 333
 334         ret = kvm_irq_delivery_to_apic(vcpu->kvm, vcpu->arch.apic, &irq, NULL);
 335         trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
 336         return ret;
 337 }
 338
 339 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vpidx, u32 sint)
 340 {
 341         struct kvm_vcpu_hv_synic *synic;
 342
 343         synic = synic_get(kvm, vpidx);
 344         if (!synic)
 345                 return -EINVAL;
 346
 347         return synic_set_irq(synic, sint);
 348 }
 349
 350 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
 351 {
 352         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
 353         int i;
 354
 355         trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
 356
 357         for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
 358                 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
 359                         kvm_hv_notify_acked_sint(vcpu, i);
 360 }
 361
 362 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi)
 363 {
 364         struct kvm_vcpu_hv_synic *synic;
 365
 366         synic = synic_get(kvm, vpidx);
 367         if (!synic)
 368                 return -EINVAL;
 369
 370         if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
 371                 return -EINVAL;
 372
 373         atomic_set(&synic->sint_to_gsi[sint], gsi);
 374         return 0;
 375 }
 376
 377 void kvm_hv_irq_routing_update(struct kvm *kvm)
 378 {
 379         struct kvm_irq_routing_table *irq_rt;
 380         struct kvm_kernel_irq_routing_entry *e;
 381         u32 gsi;
 382
 383         irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
 384                                         lockdep_is_held(&kvm->irq_lock));
 385
 386         for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
 387                 hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
 388                         if (e->type == KVM_IRQ_ROUTING_HV_SINT)
 389                                 kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
 390                                                     e->hv_sint.sint, gsi);
 391                 }
 392         }
 393 }
 394
 395 static void synic_init(struct kvm_vcpu_hv_synic *synic)
 396 {
 397         int i;
 398
 399         memset(synic, 0, sizeof(*synic));
 400         synic->version = HV_SYNIC_VERSION_1;
 401         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
 402                 atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
 403                 atomic_set(&synic->sint_to_gsi[i], -1);
 404         }
 405 }
 406
 407 static u64 get_time_ref_counter(struct kvm *kvm)
 408 {
 409         struct kvm_hv *hv = &kvm->arch.hyperv;
 410         struct kvm_vcpu *vcpu;
 411         u64 tsc;
 412
 413         /*
 414          * The guest has not set up the TSC page or the clock isn't
 415          * stable, fall back to get_kvmclock_ns.
 416          */
 417         if (!hv->tsc_ref.tsc_sequence)
 418                 return div_u64(get_kvmclock_ns(kvm), 100);
 419
 420         vcpu = kvm_get_vcpu(kvm, 0);
 421         tsc = kvm_read_l1_tsc(vcpu, rdtsc());
 422         return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64)
 423                 + hv->tsc_ref.tsc_offset;
 424 }
 425
 426 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
 427                                 bool vcpu_kick)
 428 {
 429         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
 430
 431         set_bit(stimer->index,
 432                 vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
 433         kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
 434         if (vcpu_kick)
 435                 kvm_vcpu_kick(vcpu);
 436 }
 437
 438 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
 439 {
 440         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
 441
 442         trace_kvm_hv_stimer_cleanup(stimer_to_vcpu(stimer)->vcpu_id,
 443                                     stimer->index);
 444
 445         hrtimer_cancel(&stimer->timer);
 446         clear_bit(stimer->index,
 447                   vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
 448         stimer->msg_pending = false;
 449         stimer->exp_time = 0;
 450 }
 451
 452 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
 453 {
 454         struct kvm_vcpu_hv_stimer *stimer;
 455
 456         stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
 457         trace_kvm_hv_stimer_callback(stimer_to_vcpu(stimer)->vcpu_id,
 458                                      stimer->index);
 459         stimer_mark_pending(stimer, true);
 460
 461         return HRTIMER_NORESTART;
 462 }
 463
 464 /*
 465  * stimer_start() assumptions:
 466  * a) stimer->count is not equal to 0
 467  * b) stimer->config has HV_STIMER_ENABLE flag
 468  */
 469 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
 470 {
 471         u64 time_now;
 472         ktime_t ktime_now;
 473
 474         time_now = get_time_ref_counter(stimer_to_vcpu(stimer)->kvm);
 475         ktime_now = ktime_get();
 476
 477         if (stimer->config & HV_STIMER_PERIODIC) {
 478                 if (stimer->exp_time) {
 479                         if (time_now >= stimer->exp_time) {
 480                                 u64 remainder;
 481
 482                                 div64_u64_rem(time_now - stimer->exp_time,
 483                                               stimer->count, &remainder);
 484                                 stimer->exp_time =
 485                                         time_now + (stimer->count - remainder);
 486                         }
 487                 } else
 488                         stimer->exp_time = time_now + stimer->count;
 489
 490                 trace_kvm_hv_stimer_start_periodic(
 491                                         stimer_to_vcpu(stimer)->vcpu_id,
 492                                         stimer->index,
 493                                         time_now, stimer->exp_time);
 494
 495                 hrtimer_start(&stimer->timer,
 496                               ktime_add_ns(ktime_now,
 497                                            100 * (stimer->exp_time - time_now)),
 498                               HRTIMER_MODE_ABS);
 499                 return 0;
 500         }
 501         stimer->exp_time = stimer->count;
 502         if (time_now >= stimer->count) {
 503                 /*
 504                  * Expire timer according to Hypervisor Top-Level Functional
 505                  * specification v4(15.3.1):
 506                  * "If a one shot is enabled and the specified count is in
 507                  * the past, it will expire immediately."
 508                  */
 509                 stimer_mark_pending(stimer, false);
 510                 return 0;
 511         }
 512
 513         trace_kvm_hv_stimer_start_one_shot(stimer_to_vcpu(stimer)->vcpu_id,
 514                                            stimer->index,
 515                                            time_now, stimer->count);
 516
 517         hrtimer_start(&stimer->timer,
 518                       ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)),
 519                       HRTIMER_MODE_ABS);
 520         return 0;
 521 }
 522
 523 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
 524                              bool host)
 525 {
 526         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
 527         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
 528
 529         if (!synic->active && (!host || config))
 530                 return 1;
 531
 532         trace_kvm_hv_stimer_set_config(stimer_to_vcpu(stimer)->vcpu_id,
 533                                        stimer->index, config, host);
 534
 535         stimer_cleanup(stimer);
 536         if ((stimer->config & HV_STIMER_ENABLE) && HV_STIMER_SINT(config) == 0)
 537                 config &= ~HV_STIMER_ENABLE;
 538         stimer->config = config;
 539         stimer_mark_pending(stimer, false);
 540         return 0;
 541 }
 542
 543 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
 544                             bool host)
 545 {
 546         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
 547         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
 548
 549         if (!synic->active && (!host || count))
 550                 return 1;
 551
 552         trace_kvm_hv_stimer_set_count(stimer_to_vcpu(stimer)->vcpu_id,
 553                                       stimer->index, count, host);
 554
 555         stimer_cleanup(stimer);
 556         stimer->count = count;
 557         if (stimer->count == 0)
 558                 stimer->config &= ~HV_STIMER_ENABLE;
 559         else if (stimer->config & HV_STIMER_AUTOENABLE)
 560                 stimer->config |= HV_STIMER_ENABLE;
 561         stimer_mark_pending(stimer, false);
 562         return 0;
 563 }
 564
 565 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
 566 {
 567         *pconfig = stimer->config;
 568         return 0;
 569 }
 570
 571 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
 572 {
 573         *pcount = stimer->count;
 574         return 0;
 575 }
 576
 577 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
 578                              struct hv_message *src_msg)
 579 {
 580         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 581         struct page *page;
 582         gpa_t gpa;
 583         struct hv_message *dst_msg;
 584         int r;
 585         struct hv_message_page *msg_page;
 586
 587         if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
 588                 return -ENOENT;
 589
 590         gpa = synic->msg_page & PAGE_MASK;
 591         page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
 592         if (is_error_page(page))
 593                 return -EFAULT;
 594
 595         msg_page = kmap_atomic(page);
 596         dst_msg = &msg_page->sint_message[sint];
 597         if (sync_cmpxchg(&dst_msg->header.message_type, HVMSG_NONE,
 598                          src_msg->header.message_type) != HVMSG_NONE) {
 599                 dst_msg->header.message_flags.msg_pending = 1;
 600                 r = -EAGAIN;
 601         } else {
 602                 memcpy(&dst_msg->u.payload, &src_msg->u.payload,
 603                        src_msg->header.payload_size);
 604                 dst_msg->header.message_type = src_msg->header.message_type;
 605                 dst_msg->header.payload_size = src_msg->header.payload_size;
 606                 r = synic_set_irq(synic, sint);
 607                 if (r >= 1)
 608                         r = 0;
 609                 else if (r == 0)
 610                         r = -EFAULT;
 611         }
 612         kunmap_atomic(msg_page);
 613         kvm_release_page_dirty(page);
 614         kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
 615         return r;
 616 }
 617
 618 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
 619 {
 620         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
 621         struct hv_message *msg = &stimer->msg;
 622         struct hv_timer_message_payload *payload =
 623                         (struct hv_timer_message_payload *)&msg->u.payload;
 624
 625         payload->expiration_time = stimer->exp_time;
 626         payload->delivery_time = get_time_ref_counter(vcpu->kvm);
 627         return synic_deliver_msg(vcpu_to_synic(vcpu),
 628                                  HV_STIMER_SINT(stimer->config), msg);
 629 }
 630
 631 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
 632 {
 633         int r;
 634
 635         stimer->msg_pending = true;
 636         r = stimer_send_msg(stimer);
 637         trace_kvm_hv_stimer_expiration(stimer_to_vcpu(stimer)->vcpu_id,
 638                                        stimer->index, r);
 639         if (!r) {
 640                 stimer->msg_pending = false;
 641                 if (!(stimer->config & HV_STIMER_PERIODIC))
 642                         stimer->config &= ~HV_STIMER_ENABLE;
 643         }
 644 }
 645
 646 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
 647 {
 648         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 649         struct kvm_vcpu_hv_stimer *stimer;
 650         u64 time_now, exp_time;
 651         int i;
 652
 653         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 654                 if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
 655                         stimer = &hv_vcpu->stimer[i];
 656                         if (stimer->config & HV_STIMER_ENABLE) {
 657                                 exp_time = stimer->exp_time;
 658
 659                                 if (exp_time) {
 660                                         time_now =
 661                                                 get_time_ref_counter(vcpu->kvm);
 662                                         if (time_now >= exp_time)
 663                                                 stimer_expiration(stimer);
 664                                 }
 665
 666                                 if ((stimer->config & HV_STIMER_ENABLE) &&
 667                                     stimer->count) {
 668                                         if (!stimer->msg_pending)
 669                                                 stimer_start(stimer);
 670                                 } else
 671                                         stimer_cleanup(stimer);
 672                         }
 673                 }
 674 }
 675
 676 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
 677 {
 678         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 679         int i;
 680
 681         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 682                 stimer_cleanup(&hv_vcpu->stimer[i]);
 683 }
 684
 685 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
 686 {
 687         struct hv_message *msg = &stimer->msg;
 688         struct hv_timer_message_payload *payload =
 689                         (struct hv_timer_message_payload *)&msg->u.payload;
 690
 691         memset(&msg->header, 0, sizeof(msg->header));
 692         msg->header.message_type = HVMSG_TIMER_EXPIRED;
 693         msg->header.payload_size = sizeof(*payload);
 694
 695         payload->timer_index = stimer->index;
 696         payload->expiration_time = 0;
 697         payload->delivery_time = 0;
 698 }
 699
 700 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
 701 {
 702         memset(stimer, 0, sizeof(*stimer));
 703         stimer->index = timer_index;
 704         hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
 705         stimer->timer.function = stimer_timer_callback;
 706         stimer_prepare_msg(stimer);
 707 }
 708
 709 void kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
 710 {
 711         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 712         int i;
 713
 714         synic_init(&hv_vcpu->synic);
 715
 716         bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
 717         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 718                 stimer_init(&hv_vcpu->stimer[i], i);
 719 }
 720
 721 void kvm_hv_vcpu_postcreate(struct kvm_vcpu *vcpu)
 722 {
 723         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 724
 725         hv_vcpu->vp_index = kvm_vcpu_get_idx(vcpu);
 726 }
 727
 728 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
 729 {
 730         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
 731
 732         /*
 733          * Hyper-V SynIC auto EOI SINT's are
 734          * not compatible with APICV, so deactivate APICV
 735          */
 736         kvm_vcpu_deactivate_apicv(vcpu);
 737         synic->active = true;
 738         synic->dont_zero_synic_pages = dont_zero_synic_pages;
 739         return 0;
 740 }
 741
 742 static bool kvm_hv_msr_partition_wide(u32 msr)
 743 {
 744         bool r = false;
 745
 746         switch (msr) {
 747         case HV_X64_MSR_GUEST_OS_ID:
 748         case HV_X64_MSR_HYPERCALL:
 749         case HV_X64_MSR_REFERENCE_TSC:
 750         case HV_X64_MSR_TIME_REF_COUNT:
 751         case HV_X64_MSR_CRASH_CTL:
 752         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
 753         case HV_X64_MSR_RESET:
 754                 r = true;
 755                 break;
 756         }
 757
 758         return r;
 759 }
 760
 761 static int kvm_hv_msr_get_crash_data(struct kvm_vcpu *vcpu,
 762                                      u32 index, u64 *pdata)
 763 {
 764         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 765         size_t size = ARRAY_SIZE(hv->hv_crash_param);
 766
 767         if (WARN_ON_ONCE(index >= size))
 768                 return -EINVAL;
 769
 770         *pdata = hv->hv_crash_param[array_index_nospec(index, size)];
 771         return 0;
 772 }
 773
 774 static int kvm_hv_msr_get_crash_ctl(struct kvm_vcpu *vcpu, u64 *pdata)
 775 {
 776         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 777
 778         *pdata = hv->hv_crash_ctl;
 779         return 0;
 780 }
 781
 782 static int kvm_hv_msr_set_crash_ctl(struct kvm_vcpu *vcpu, u64 data, bool host)
 783 {
 784         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 785
 786         if (host)
 787                 hv->hv_crash_ctl = data & HV_X64_MSR_CRASH_CTL_NOTIFY;
 788
 789         if (!host && (data & HV_X64_MSR_CRASH_CTL_NOTIFY)) {
 790
 791                 vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
 792                           hv->hv_crash_param[0],
 793                           hv->hv_crash_param[1],
 794                           hv->hv_crash_param[2],
 795                           hv->hv_crash_param[3],
 796                           hv->hv_crash_param[4]);
 797
 798                 /* Send notification about crash to user space */
 799                 kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
 800         }
 801
 802         return 0;
 803 }
 804
 805 static int kvm_hv_msr_set_crash_data(struct kvm_vcpu *vcpu,
 806                                      u32 index, u64 data)
 807 {
 808         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 809         size_t size = ARRAY_SIZE(hv->hv_crash_param);
 810
 811         if (WARN_ON_ONCE(index >= size))
 812                 return -EINVAL;
 813
 814         hv->hv_crash_param[array_index_nospec(index, size)] = data;
 815         return 0;
 816 }
 817
 818 /*
 819  * The kvmclock and Hyper-V TSC page use similar formulas, and converting
 820  * between them is possible:
 821  *
 822  * kvmclock formula:
 823  *    nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
 824  *           + system_time
 825  *
 826  * Hyper-V formula:
 827  *    nsec/100 = ticks * scale / 2^64 + offset
 828  *
 829  * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
 830  * By dividing the kvmclock formula by 100 and equating what's left we get:
 831  *    ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
 832  *            scale / 2^64 =         tsc_to_system_mul * 2^(tsc_shift-32) / 100
 833  *            scale        =         tsc_to_system_mul * 2^(32+tsc_shift) / 100
 834  *
 835  * Now expand the kvmclock formula and divide by 100:
 836  *    nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
 837  *           - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
 838  *           + system_time
 839  *    nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
 840  *               - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
 841  *               + system_time / 100
 842  *
 843  * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
 844  *    nsec/100 = ticks * scale / 2^64
 845  *               - tsc_timestamp * scale / 2^64
 846  *               + system_time / 100
 847  *
 848  * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
 849  *    offset = system_time / 100 - tsc_timestamp * scale / 2^64
 850  *
 851  * These two equivalencies are implemented in this function.
 852  */
 853 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
 854                                         HV_REFERENCE_TSC_PAGE *tsc_ref)
 855 {
 856         u64 max_mul;
 857
 858         if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
 859                 return false;
 860
 861         /*
 862          * check if scale would overflow, if so we use the time ref counter
 863          *    tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
 864          *    tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
 865          *    tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
 866          */
 867         max_mul = 100ull << (32 - hv_clock->tsc_shift);
 868         if (hv_clock->tsc_to_system_mul >= max_mul)
 869                 return false;
 870
 871         /*
 872          * Otherwise compute the scale and offset according to the formulas
 873          * derived above.
 874          */
 875         tsc_ref->tsc_scale =
 876                 mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
 877                                 hv_clock->tsc_to_system_mul,
 878                                 100);
 879
 880         tsc_ref->tsc_offset = hv_clock->system_time;
 881         do_div(tsc_ref->tsc_offset, 100);
 882         tsc_ref->tsc_offset -=
 883                 mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
 884         return true;
 885 }
 886
 887 void kvm_hv_setup_tsc_page(struct kvm *kvm,
 888                            struct pvclock_vcpu_time_info *hv_clock)
 889 {
 890         struct kvm_hv *hv = &kvm->arch.hyperv;
 891         u32 tsc_seq;
 892         u64 gfn;
 893
 894         BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
 895         BUILD_BUG_ON(offsetof(HV_REFERENCE_TSC_PAGE, tsc_sequence) != 0);
 896
 897         if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
 898                 return;
 899
 900         mutex_lock(&kvm->arch.hyperv.hv_lock);
 901         if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
 902                 goto out_unlock;
 903
 904         gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
 905         /*
 906          * Because the TSC parameters only vary when there is a
 907          * change in the master clock, do not bother with caching.
 908          */
 909         if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
 910                                     &tsc_seq, sizeof(tsc_seq))))
 911                 goto out_unlock;
 912
 913         /*
 914          * While we're computing and writing the parameters, force the
 915          * guest to use the time reference count MSR.
 916          */
 917         hv->tsc_ref.tsc_sequence = 0;
 918         if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
 919                             &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
 920                 goto out_unlock;
 921
 922         if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
 923                 goto out_unlock;
 924
 925         /* Ensure sequence is zero before writing the rest of the struct.  */
 926         smp_wmb();
 927         if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
 928                 goto out_unlock;
 929
 930         /*
 931          * Now switch to the TSC page mechanism by writing the sequence.
 932          */
 933         tsc_seq++;
 934         if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
 935                 tsc_seq = 1;
 936
 937         /* Write the struct entirely before the non-zero sequence.  */
 938         smp_wmb();
 939
 940         hv->tsc_ref.tsc_sequence = tsc_seq;
 941         kvm_write_guest(kvm, gfn_to_gpa(gfn),
 942                         &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence));
 943 out_unlock:
 944         mutex_unlock(&kvm->arch.hyperv.hv_lock);
 945 }
 946
 947 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
 948                              bool host)
 949 {
 950         struct kvm *kvm = vcpu->kvm;
 951         struct kvm_hv *hv = &kvm->arch.hyperv;
 952
 953         switch (msr) {
 954         case HV_X64_MSR_GUEST_OS_ID:
 955                 hv->hv_guest_os_id = data;
 956                 /* setting guest os id to zero disables hypercall page */
 957                 if (!hv->hv_guest_os_id)
 958                         hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
 959                 break;
 960         case HV_X64_MSR_HYPERCALL: {
 961                 u64 gfn;
 962                 unsigned long addr;
 963                 u8 instructions[4];
 964
 965                 /* if guest os id is not set hypercall should remain disabled */
 966                 if (!hv->hv_guest_os_id)
 967                         break;
 968                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
 969                         hv->hv_hypercall = data;
 970                         break;
 971                 }
 972                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
 973                 addr = gfn_to_hva(kvm, gfn);
 974                 if (kvm_is_error_hva(addr))
 975                         return 1;
 976                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
 977                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
 978                 if (__copy_to_user((void __user *)addr, instructions, 4))
 979                         return 1;
 980                 hv->hv_hypercall = data;
 981                 mark_page_dirty(kvm, gfn);
 982                 break;
 983         }
 984         case HV_X64_MSR_REFERENCE_TSC:
 985                 hv->hv_tsc_page = data;
 986                 if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE)
 987                         kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
 988                 break;
 989         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
 990                 return kvm_hv_msr_set_crash_data(vcpu,
 991                                                  msr - HV_X64_MSR_CRASH_P0,
 992                                                  data);
 993         case HV_X64_MSR_CRASH_CTL:
 994                 return kvm_hv_msr_set_crash_ctl(vcpu, data, host);
 995         case HV_X64_MSR_RESET:
 996                 if (data == 1) {
 997                         vcpu_debug(vcpu, "hyper-v reset requested\n");
 998                         kvm_make_request(KVM_REQ_HV_RESET, vcpu);
 999                 }
1000                 break;
1001         default:
1002                 vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
1003                             msr, data);
1004                 return 1;
1005         }
1006         return 0;
1007 }
1008
1009 /* Calculate cpu time spent by current task in 100ns units */
1010 static u64 current_task_runtime_100ns(void)
1011 {
1012         u64 utime, stime;
1013
1014         task_cputime_adjusted(current, &utime, &stime);
1015
1016         return div_u64(utime + stime, 100);
1017 }
1018
1019 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1020 {
1021         struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
1022
1023         switch (msr) {
1024         case HV_X64_MSR_VP_INDEX:
1025                 if (!host)
1026                         return 1;
1027                 hv->vp_index = (u32)data;
1028                 break;
1029         case HV_X64_MSR_APIC_ASSIST_PAGE: {
1030                 u64 gfn;
1031                 unsigned long addr;
1032
1033                 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1034                         hv->hv_vapic = data;
1035                         if (kvm_lapic_enable_pv_eoi(vcpu, 0))
1036                                 return 1;
1037                         break;
1038                 }
1039                 gfn = data >> HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT;
1040                 addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1041                 if (kvm_is_error_hva(addr))
1042                         return 1;
1043                 if (__clear_user((void __user *)addr, PAGE_SIZE))
1044                         return 1;
1045                 hv->hv_vapic = data;
1046                 kvm_vcpu_mark_page_dirty(vcpu, gfn);
1047                 if (kvm_lapic_enable_pv_eoi(vcpu,
1048                                             gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
1049                         return 1;
1050                 break;
1051         }
1052         case HV_X64_MSR_EOI:
1053                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1054         case HV_X64_MSR_ICR:
1055                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1056         case HV_X64_MSR_TPR:
1057                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1058         case HV_X64_MSR_VP_RUNTIME:
1059                 if (!host)
1060                         return 1;
1061                 hv->runtime_offset = data - current_task_runtime_100ns();
1062                 break;
1063         case HV_X64_MSR_SCONTROL:
1064         case HV_X64_MSR_SVERSION:
1065         case HV_X64_MSR_SIEFP:
1066         case HV_X64_MSR_SIMP:
1067         case HV_X64_MSR_EOM:
1068         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1069                 return synic_set_msr(vcpu_to_synic(vcpu), msr, data, host);
1070         case HV_X64_MSR_STIMER0_CONFIG:
1071         case HV_X64_MSR_STIMER1_CONFIG:
1072         case HV_X64_MSR_STIMER2_CONFIG:
1073         case HV_X64_MSR_STIMER3_CONFIG: {
1074                 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1075
1076                 return stimer_set_config(vcpu_to_stimer(vcpu, timer_index),
1077                                          data, host);
1078         }
1079         case HV_X64_MSR_STIMER0_COUNT:
1080         case HV_X64_MSR_STIMER1_COUNT:
1081         case HV_X64_MSR_STIMER2_COUNT:
1082         case HV_X64_MSR_STIMER3_COUNT: {
1083                 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1084
1085                 return stimer_set_count(vcpu_to_stimer(vcpu, timer_index),
1086                                         data, host);
1087         }
1088         default:
1089                 vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
1090                             msr, data);
1091                 return 1;
1092         }
1093
1094         return 0;
1095 }
1096
1097 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1098 {
1099         u64 data = 0;
1100         struct kvm *kvm = vcpu->kvm;
1101         struct kvm_hv *hv = &kvm->arch.hyperv;
1102
1103         switch (msr) {
1104         case HV_X64_MSR_GUEST_OS_ID:
1105                 data = hv->hv_guest_os_id;
1106                 break;
1107         case HV_X64_MSR_HYPERCALL:
1108                 data = hv->hv_hypercall;
1109                 break;
1110         case HV_X64_MSR_TIME_REF_COUNT:
1111                 data = get_time_ref_counter(kvm);
1112                 break;
1113         case HV_X64_MSR_REFERENCE_TSC:
1114                 data = hv->hv_tsc_page;
1115                 break;
1116         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1117                 return kvm_hv_msr_get_crash_data(vcpu,
1118                                                  msr - HV_X64_MSR_CRASH_P0,
1119                                                  pdata);
1120         case HV_X64_MSR_CRASH_CTL:
1121                 return kvm_hv_msr_get_crash_ctl(vcpu, pdata);
1122         case HV_X64_MSR_RESET:
1123                 data = 0;
1124                 break;
1125         default:
1126                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1127                 return 1;
1128         }
1129
1130         *pdata = data;
1131         return 0;
1132 }
1133
1134 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1135 {
1136         u64 data = 0;
1137         struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
1138
1139         switch (msr) {
1140         case HV_X64_MSR_VP_INDEX:
1141                 data = hv->vp_index;
1142                 break;
1143         case HV_X64_MSR_EOI:
1144                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1145         case HV_X64_MSR_ICR:
1146                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1147         case HV_X64_MSR_TPR:
1148                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1149         case HV_X64_MSR_APIC_ASSIST_PAGE:
1150                 data = hv->hv_vapic;
1151                 break;
1152         case HV_X64_MSR_VP_RUNTIME:
1153                 data = current_task_runtime_100ns() + hv->runtime_offset;
1154                 break;
1155         case HV_X64_MSR_SCONTROL:
1156         case HV_X64_MSR_SVERSION:
1157         case HV_X64_MSR_SIEFP:
1158         case HV_X64_MSR_SIMP:
1159         case HV_X64_MSR_EOM:
1160         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1161                 return synic_get_msr(vcpu_to_synic(vcpu), msr, pdata);
1162         case HV_X64_MSR_STIMER0_CONFIG:
1163         case HV_X64_MSR_STIMER1_CONFIG:
1164         case HV_X64_MSR_STIMER2_CONFIG:
1165         case HV_X64_MSR_STIMER3_CONFIG: {
1166                 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1167
1168                 return stimer_get_config(vcpu_to_stimer(vcpu, timer_index),
1169                                          pdata);
1170         }
1171         case HV_X64_MSR_STIMER0_COUNT:
1172         case HV_X64_MSR_STIMER1_COUNT:
1173         case HV_X64_MSR_STIMER2_COUNT:
1174         case HV_X64_MSR_STIMER3_COUNT: {
1175                 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1176
1177                 return stimer_get_count(vcpu_to_stimer(vcpu, timer_index),
1178                                         pdata);
1179         }
1180         case HV_X64_MSR_TSC_FREQUENCY:
1181                 data = (u64)vcpu->arch.virtual_tsc_khz * 1000;
1182                 break;
1183         case HV_X64_MSR_APIC_FREQUENCY:
1184                 data = APIC_BUS_FREQUENCY;
1185                 break;
1186         default:
1187                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1188                 return 1;
1189         }
1190         *pdata = data;
1191         return 0;
1192 }
1193
1194 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1195 {
1196         if (kvm_hv_msr_partition_wide(msr)) {
1197                 int r;
1198
1199                 mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1200                 r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1201                 mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1202                 return r;
1203         } else
1204                 return kvm_hv_set_msr(vcpu, msr, data, host);
1205 }
1206
1207 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1208 {
1209         if (kvm_hv_msr_partition_wide(msr)) {
1210                 int r;
1211
1212                 mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1213                 r = kvm_hv_get_msr_pw(vcpu, msr, pdata);
1214                 mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1215                 return r;
1216         } else
1217                 return kvm_hv_get_msr(vcpu, msr, pdata);
1218 }
1219
1220 bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1221 {
1222         return READ_ONCE(kvm->arch.hyperv.hv_hypercall) & HV_X64_MSR_HYPERCALL_ENABLE;
1223 }
1224
1225 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
1226 {
1227         bool longmode;
1228
1229         longmode = is_64_bit_mode(vcpu);
1230         if (longmode)
1231                 kvm_register_write(vcpu, VCPU_REGS_RAX, result);
1232         else {
1233                 kvm_register_write(vcpu, VCPU_REGS_RDX, result >> 32);
1234                 kvm_register_write(vcpu, VCPU_REGS_RAX, result & 0xffffffff);
1235         }
1236 }
1237
1238 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
1239 {
1240         struct kvm_run *run = vcpu->run;
1241
1242         kvm_hv_hypercall_set_result(vcpu, run->hyperv.u.hcall.result);
1243         return kvm_skip_emulated_instruction(vcpu);
1244 }
1245
1246 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
1247 {
1248         u64 param, ingpa, outgpa, ret;
1249         uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
1250         bool fast, longmode;
1251
1252         /*
1253          * hypercall generates UD from non zero cpl and real mode
1254          * per HYPER-V spec
1255          */
1256         if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
1257                 kvm_queue_exception(vcpu, UD_VECTOR);
1258                 return 1;
1259         }
1260
1261         longmode = is_64_bit_mode(vcpu);
1262
1263         if (!longmode) {
1264                 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
1265                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
1266                 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
1267                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
1268                 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
1269                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
1270         }
1271 #ifdef CONFIG_X86_64
1272         else {
1273                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
1274                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
1275                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
1276         }
1277 #endif
1278
1279         code = param & 0xffff;
1280         fast = (param >> 16) & 0x1;
1281         rep_cnt = (param >> 32) & 0xfff;
1282         rep_idx = (param >> 48) & 0xfff;
1283
1284         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
1285
1286         /* Hypercall continuation is not supported yet */
1287         if (rep_cnt || rep_idx) {
1288                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
1289                 goto set_result;
1290         }
1291
1292         switch (code) {
1293         case HVCALL_NOTIFY_LONG_SPIN_WAIT:
1294                 kvm_vcpu_on_spin(vcpu, true);
1295                 break;
1296         case HVCALL_POST_MESSAGE:
1297         case HVCALL_SIGNAL_EVENT:
1298                 /* don't bother userspace if it has no way to handle it */
1299                 if (!vcpu_to_synic(vcpu)->active) {
1300                         res = HV_STATUS_INVALID_HYPERCALL_CODE;
1301                         break;
1302                 }
1303                 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
1304                 vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
1305                 vcpu->run->hyperv.u.hcall.input = param;
1306                 vcpu->run->hyperv.u.hcall.params[0] = ingpa;
1307                 vcpu->run->hyperv.u.hcall.params[1] = outgpa;
1308                 vcpu->arch.complete_userspace_io =
1309                                 kvm_hv_hypercall_complete_userspace;
1310                 return 0;
1311         default:
1312                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
1313                 break;
1314         }
1315
1316 set_result:
1317         ret = res | (((u64)rep_done & 0xfff) << 32);
1318         kvm_hv_hypercall_set_result(vcpu, ret);
1319         return 1;
1320 }