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