arch/x86/kvm/hyperv.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * KVM Microsoft Hyper-V emulation
   4  *
   5  * derived from arch/x86/kvm/x86.c
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright (C) 2008 Qumranet, Inc.
   9  * Copyright IBM Corporation, 2008
  10  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
  11  * Copyright (C) 2015 Andrey Smetanin <asmetanin@virtuozzo.com>
  12  *
  13  * Authors:
  14  *   Avi Kivity   <avi@qumranet.com>
  15  *   Yaniv Kamay  <yaniv@qumranet.com>
  16  *   Amit Shah    <amit.shah@qumranet.com>
  17  *   Ben-Ami Yassour <benami@il.ibm.com>
  18  *   Andrey Smetanin <asmetanin@virtuozzo.com>
  19  */
  20
  21 #include "x86.h"
  22 #include "lapic.h"
  23 #include "ioapic.h"
  24 #include "cpuid.h"
  25 #include "hyperv.h"
  26 #include "xen.h"
  27
  28 #include <linux/cpu.h>
  29 #include <linux/kvm_host.h>
  30 #include <linux/highmem.h>
  31 #include <linux/sched/cputime.h>
  32 #include <linux/eventfd.h>
  33
  34 #include <asm/apicdef.h>
  35 #include <trace/events/kvm.h>
  36
  37 #include "trace.h"
  38 #include "irq.h"
  39 #include "fpu.h"
  40
  41 /* "Hv#1" signature */
  42 #define HYPERV_CPUID_SIGNATURE_EAX 0x31237648
  43
  44 #define KVM_HV_MAX_SPARSE_VCPU_SET_BITS DIV_ROUND_UP(KVM_MAX_VCPUS, 64)
  45
  46 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
  47                                 bool vcpu_kick);
  48
  49 static inline u64 synic_read_sint(struct kvm_vcpu_hv_synic *synic, int sint)
  50 {
  51         return atomic64_read(&synic->sint[sint]);
  52 }
  53
  54 static inline int synic_get_sint_vector(u64 sint_value)
  55 {
  56         if (sint_value & HV_SYNIC_SINT_MASKED)
  57                 return -1;
  58         return sint_value & HV_SYNIC_SINT_VECTOR_MASK;
  59 }
  60
  61 static bool synic_has_vector_connected(struct kvm_vcpu_hv_synic *synic,
  62                                       int vector)
  63 {
  64         int i;
  65
  66         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
  67                 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
  68                         return true;
  69         }
  70         return false;
  71 }
  72
  73 static bool synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic *synic,
  74                                      int vector)
  75 {
  76         int i;
  77         u64 sint_value;
  78
  79         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
  80                 sint_value = synic_read_sint(synic, i);
  81                 if (synic_get_sint_vector(sint_value) == vector &&
  82                     sint_value & HV_SYNIC_SINT_AUTO_EOI)
  83                         return true;
  84         }
  85         return false;
  86 }
  87
  88 static void synic_update_vector(struct kvm_vcpu_hv_synic *synic,
  89                                 int vector)
  90 {
  91         struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
  92         struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
  93         int auto_eoi_old, auto_eoi_new;
  94
  95         if (vector < HV_SYNIC_FIRST_VALID_VECTOR)
  96                 return;
  97
  98         if (synic_has_vector_connected(synic, vector))
  99                 __set_bit(vector, synic->vec_bitmap);
 100         else
 101                 __clear_bit(vector, synic->vec_bitmap);
 102
 103         auto_eoi_old = bitmap_weight(synic->auto_eoi_bitmap, 256);
 104
 105         if (synic_has_vector_auto_eoi(synic, vector))
 106                 __set_bit(vector, synic->auto_eoi_bitmap);
 107         else
 108                 __clear_bit(vector, synic->auto_eoi_bitmap);
 109
 110         auto_eoi_new = bitmap_weight(synic->auto_eoi_bitmap, 256);
 111
 112         if (!!auto_eoi_old == !!auto_eoi_new)
 113                 return;
 114
 115         down_write(&vcpu->kvm->arch.apicv_update_lock);
 116
 117         if (auto_eoi_new)
 118                 hv->synic_auto_eoi_used++;
 119         else
 120                 hv->synic_auto_eoi_used--;
 121
 122         __kvm_request_apicv_update(vcpu->kvm,
 123                                    !hv->synic_auto_eoi_used,
 124                                    APICV_INHIBIT_REASON_HYPERV);
 125
 126         up_write(&vcpu->kvm->arch.apicv_update_lock);
 127 }
 128
 129 static int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint,
 130                           u64 data, bool host)
 131 {
 132         int vector, old_vector;
 133         bool masked;
 134
 135         vector = data & HV_SYNIC_SINT_VECTOR_MASK;
 136         masked = data & HV_SYNIC_SINT_MASKED;
 137
 138         /*
 139          * Valid vectors are 16-255, however, nested Hyper-V attempts to write
 140          * default '0x10000' value on boot and this should not #GP. We need to
 141          * allow zero-initing the register from host as well.
 142          */
 143         if (vector < HV_SYNIC_FIRST_VALID_VECTOR && !host && !masked)
 144                 return 1;
 145         /*
 146          * Guest may configure multiple SINTs to use the same vector, so
 147          * we maintain a bitmap of vectors handled by synic, and a
 148          * bitmap of vectors with auto-eoi behavior.  The bitmaps are
 149          * updated here, and atomically queried on fast paths.
 150          */
 151         old_vector = synic_read_sint(synic, sint) & HV_SYNIC_SINT_VECTOR_MASK;
 152
 153         atomic64_set(&synic->sint[sint], data);
 154
 155         synic_update_vector(synic, old_vector);
 156
 157         synic_update_vector(synic, vector);
 158
 159         /* Load SynIC vectors into EOI exit bitmap */
 160         kvm_make_request(KVM_REQ_SCAN_IOAPIC, hv_synic_to_vcpu(synic));
 161         return 0;
 162 }
 163
 164 static struct kvm_vcpu *get_vcpu_by_vpidx(struct kvm *kvm, u32 vpidx)
 165 {
 166         struct kvm_vcpu *vcpu = NULL;
 167         unsigned long i;
 168
 169         if (vpidx >= KVM_MAX_VCPUS)
 170                 return NULL;
 171
 172         vcpu = kvm_get_vcpu(kvm, vpidx);
 173         if (vcpu && kvm_hv_get_vpindex(vcpu) == vpidx)
 174                 return vcpu;
 175         kvm_for_each_vcpu(i, vcpu, kvm)
 176                 if (kvm_hv_get_vpindex(vcpu) == vpidx)
 177                         return vcpu;
 178         return NULL;
 179 }
 180
 181 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vpidx)
 182 {
 183         struct kvm_vcpu *vcpu;
 184         struct kvm_vcpu_hv_synic *synic;
 185
 186         vcpu = get_vcpu_by_vpidx(kvm, vpidx);
 187         if (!vcpu || !to_hv_vcpu(vcpu))
 188                 return NULL;
 189         synic = to_hv_synic(vcpu);
 190         return (synic->active) ? synic : NULL;
 191 }
 192
 193 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
 194 {
 195         struct kvm *kvm = vcpu->kvm;
 196         struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
 197         struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
 198         struct kvm_vcpu_hv_stimer *stimer;
 199         int gsi, idx;
 200
 201         trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
 202
 203         /* Try to deliver pending Hyper-V SynIC timers messages */
 204         for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
 205                 stimer = &hv_vcpu->stimer[idx];
 206                 if (stimer->msg_pending && stimer->config.enable &&
 207                     !stimer->config.direct_mode &&
 208                     stimer->config.sintx == sint)
 209                         stimer_mark_pending(stimer, false);
 210         }
 211
 212         idx = srcu_read_lock(&kvm->irq_srcu);
 213         gsi = atomic_read(&synic->sint_to_gsi[sint]);
 214         if (gsi != -1)
 215                 kvm_notify_acked_gsi(kvm, gsi);
 216         srcu_read_unlock(&kvm->irq_srcu, idx);
 217 }
 218
 219 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
 220 {
 221         struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
 222         struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
 223
 224         hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
 225         hv_vcpu->exit.u.synic.msr = msr;
 226         hv_vcpu->exit.u.synic.control = synic->control;
 227         hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
 228         hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
 229
 230         kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
 231 }
 232
 233 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
 234                          u32 msr, u64 data, bool host)
 235 {
 236         struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
 237         int ret;
 238
 239         if (!synic->active && !host)
 240                 return 1;
 241
 242         trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
 243
 244         ret = 0;
 245         switch (msr) {
 246         case HV_X64_MSR_SCONTROL:
 247                 synic->control = data;
 248                 if (!host)
 249                         synic_exit(synic, msr);
 250                 break;
 251         case HV_X64_MSR_SVERSION:
 252                 if (!host) {
 253                         ret = 1;
 254                         break;
 255                 }
 256                 synic->version = data;
 257                 break;
 258         case HV_X64_MSR_SIEFP:
 259                 if ((data & HV_SYNIC_SIEFP_ENABLE) && !host &&
 260                     !synic->dont_zero_synic_pages)
 261                         if (kvm_clear_guest(vcpu->kvm,
 262                                             data & PAGE_MASK, PAGE_SIZE)) {
 263                                 ret = 1;
 264                                 break;
 265                         }
 266                 synic->evt_page = data;
 267                 if (!host)
 268                         synic_exit(synic, msr);
 269                 break;
 270         case HV_X64_MSR_SIMP:
 271                 if ((data & HV_SYNIC_SIMP_ENABLE) && !host &&
 272                     !synic->dont_zero_synic_pages)
 273                         if (kvm_clear_guest(vcpu->kvm,
 274                                             data & PAGE_MASK, PAGE_SIZE)) {
 275                                 ret = 1;
 276                                 break;
 277                         }
 278                 synic->msg_page = data;
 279                 if (!host)
 280                         synic_exit(synic, msr);
 281                 break;
 282         case HV_X64_MSR_EOM: {
 283                 int i;
 284
 285                 for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
 286                         kvm_hv_notify_acked_sint(vcpu, i);
 287                 break;
 288         }
 289         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
 290                 ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
 291                 break;
 292         default:
 293                 ret = 1;
 294                 break;
 295         }
 296         return ret;
 297 }
 298
 299 static bool kvm_hv_is_syndbg_enabled(struct kvm_vcpu *vcpu)
 300 {
 301         struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
 302
 303         return hv_vcpu->cpuid_cache.syndbg_cap_eax &
 304                 HV_X64_SYNDBG_CAP_ALLOW_KERNEL_DEBUGGING;
 305 }
 306
 307 static int kvm_hv_syndbg_complete_userspace(struct kvm_vcpu *vcpu)
 308 {
 309         struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
 310
 311         if (vcpu->run->hyperv.u.syndbg.msr == HV_X64_MSR_SYNDBG_CONTROL)
 312                 hv->hv_syndbg.control.status =
 313                         vcpu->run->hyperv.u.syndbg.status;
 314         return 1;
 315 }
 316
 317 static void syndbg_exit(struct kvm_vcpu *vcpu, u32 msr)
 318 {
 319         struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
 320         struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
 321
 322         hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNDBG;
 323         hv_vcpu->exit.u.syndbg.msr = msr;
 324         hv_vcpu->exit.u.syndbg.control = syndbg->control.control;
 325         hv_vcpu->exit.u.syndbg.send_page = syndbg->control.send_page;
 326         hv_vcpu->exit.u.syndbg.recv_page = syndbg->control.recv_page;
 327         hv_vcpu->exit.u.syndbg.pending_page = syndbg->control.pending_page;
 328         vcpu->arch.complete_userspace_io =
 329                         kvm_hv_syndbg_complete_userspace;
 330
 331         kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
 332 }
 333
 334 static int syndbg_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
 335 {
 336         struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
 337
 338         if (!kvm_hv_is_syndbg_enabled(vcpu) && !host)
 339                 return 1;
 340
 341         trace_kvm_hv_syndbg_set_msr(vcpu->vcpu_id,
 342                                     to_hv_vcpu(vcpu)->vp_index, msr, data);
 343         switch (msr) {
 344         case HV_X64_MSR_SYNDBG_CONTROL:
 345                 syndbg->control.control = data;
 346                 if (!host)
 347                         syndbg_exit(vcpu, msr);
 348                 break;
 349         case HV_X64_MSR_SYNDBG_STATUS:
 350                 syndbg->control.status = data;
 351                 break;
 352         case HV_X64_MSR_SYNDBG_SEND_BUFFER:
 353                 syndbg->control.send_page = data;
 354                 break;
 355         case HV_X64_MSR_SYNDBG_RECV_BUFFER:
 356                 syndbg->control.recv_page = data;
 357                 break;
 358         case HV_X64_MSR_SYNDBG_PENDING_BUFFER:
 359                 syndbg->control.pending_page = data;
 360                 if (!host)
 361                         syndbg_exit(vcpu, msr);
 362                 break;
 363         case HV_X64_MSR_SYNDBG_OPTIONS:
 364                 syndbg->options = data;
 365                 break;
 366         default:
 367                 break;
 368         }
 369
 370         return 0;
 371 }
 372
 373 static int syndbg_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
 374 {
 375         struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
 376
 377         if (!kvm_hv_is_syndbg_enabled(vcpu) && !host)
 378                 return 1;
 379
 380         switch (msr) {
 381         case HV_X64_MSR_SYNDBG_CONTROL:
 382                 *pdata = syndbg->control.control;
 383                 break;
 384         case HV_X64_MSR_SYNDBG_STATUS:
 385                 *pdata = syndbg->control.status;
 386                 break;
 387         case HV_X64_MSR_SYNDBG_SEND_BUFFER:
 388                 *pdata = syndbg->control.send_page;
 389                 break;
 390         case HV_X64_MSR_SYNDBG_RECV_BUFFER:
 391                 *pdata = syndbg->control.recv_page;
 392                 break;
 393         case HV_X64_MSR_SYNDBG_PENDING_BUFFER:
 394                 *pdata = syndbg->control.pending_page;
 395                 break;
 396         case HV_X64_MSR_SYNDBG_OPTIONS:
 397                 *pdata = syndbg->options;
 398                 break;
 399         default:
 400                 break;
 401         }
 402
 403         trace_kvm_hv_syndbg_get_msr(vcpu->vcpu_id, kvm_hv_get_vpindex(vcpu), msr, *pdata);
 404
 405         return 0;
 406 }
 407
 408 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata,
 409                          bool host)
 410 {
 411         int ret;
 412
 413         if (!synic->active && !host)
 414                 return 1;
 415
 416         ret = 0;
 417         switch (msr) {
 418         case HV_X64_MSR_SCONTROL:
 419                 *pdata = synic->control;
 420                 break;
 421         case HV_X64_MSR_SVERSION:
 422                 *pdata = synic->version;
 423                 break;
 424         case HV_X64_MSR_SIEFP:
 425                 *pdata = synic->evt_page;
 426                 break;
 427         case HV_X64_MSR_SIMP:
 428                 *pdata = synic->msg_page;
 429                 break;
 430         case HV_X64_MSR_EOM:
 431                 *pdata = 0;
 432                 break;
 433         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
 434                 *pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
 435                 break;
 436         default:
 437                 ret = 1;
 438                 break;
 439         }
 440         return ret;
 441 }
 442
 443 static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
 444 {
 445         struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
 446         struct kvm_lapic_irq irq;
 447         int ret, vector;
 448
 449         if (sint >= ARRAY_SIZE(synic->sint))
 450                 return -EINVAL;
 451
 452         vector = synic_get_sint_vector(synic_read_sint(synic, sint));
 453         if (vector < 0)
 454                 return -ENOENT;
 455
 456         memset(&irq, 0, sizeof(irq));
 457         irq.shorthand = APIC_DEST_SELF;
 458         irq.dest_mode = APIC_DEST_PHYSICAL;
 459         irq.delivery_mode = APIC_DM_FIXED;
 460         irq.vector = vector;
 461         irq.level = 1;
 462
 463         ret = kvm_irq_delivery_to_apic(vcpu->kvm, vcpu->arch.apic, &irq, NULL);
 464         trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
 465         return ret;
 466 }
 467
 468 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vpidx, u32 sint)
 469 {
 470         struct kvm_vcpu_hv_synic *synic;
 471
 472         synic = synic_get(kvm, vpidx);
 473         if (!synic)
 474                 return -EINVAL;
 475
 476         return synic_set_irq(synic, sint);
 477 }
 478
 479 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
 480 {
 481         struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
 482         int i;
 483
 484         trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
 485
 486         for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
 487                 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
 488                         kvm_hv_notify_acked_sint(vcpu, i);
 489 }
 490
 491 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi)
 492 {
 493         struct kvm_vcpu_hv_synic *synic;
 494
 495         synic = synic_get(kvm, vpidx);
 496         if (!synic)
 497                 return -EINVAL;
 498
 499         if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
 500                 return -EINVAL;
 501
 502         atomic_set(&synic->sint_to_gsi[sint], gsi);
 503         return 0;
 504 }
 505
 506 void kvm_hv_irq_routing_update(struct kvm *kvm)
 507 {
 508         struct kvm_irq_routing_table *irq_rt;
 509         struct kvm_kernel_irq_routing_entry *e;
 510         u32 gsi;
 511
 512         irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
 513                                         lockdep_is_held(&kvm->irq_lock));
 514
 515         for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
 516                 hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
 517                         if (e->type == KVM_IRQ_ROUTING_HV_SINT)
 518                                 kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
 519                                                     e->hv_sint.sint, gsi);
 520                 }
 521         }
 522 }
 523
 524 static void synic_init(struct kvm_vcpu_hv_synic *synic)
 525 {
 526         int i;
 527
 528         memset(synic, 0, sizeof(*synic));
 529         synic->version = HV_SYNIC_VERSION_1;
 530         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
 531                 atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
 532                 atomic_set(&synic->sint_to_gsi[i], -1);
 533         }
 534 }
 535
 536 static u64 get_time_ref_counter(struct kvm *kvm)
 537 {
 538         struct kvm_hv *hv = to_kvm_hv(kvm);
 539         struct kvm_vcpu *vcpu;
 540         u64 tsc;
 541
 542         /*
 543          * Fall back to get_kvmclock_ns() when TSC page hasn't been set up,
 544          * is broken, disabled or being updated.
 545          */
 546         if (hv->hv_tsc_page_status != HV_TSC_PAGE_SET)
 547                 return div_u64(get_kvmclock_ns(kvm), 100);
 548
 549         vcpu = kvm_get_vcpu(kvm, 0);
 550         tsc = kvm_read_l1_tsc(vcpu, rdtsc());
 551         return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64)
 552                 + hv->tsc_ref.tsc_offset;
 553 }
 554
 555 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
 556                                 bool vcpu_kick)
 557 {
 558         struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
 559
 560         set_bit(stimer->index,
 561                 to_hv_vcpu(vcpu)->stimer_pending_bitmap);
 562         kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
 563         if (vcpu_kick)
 564                 kvm_vcpu_kick(vcpu);
 565 }
 566
 567 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
 568 {
 569         struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
 570
 571         trace_kvm_hv_stimer_cleanup(hv_stimer_to_vcpu(stimer)->vcpu_id,
 572                                     stimer->index);
 573
 574         hrtimer_cancel(&stimer->timer);
 575         clear_bit(stimer->index,
 576                   to_hv_vcpu(vcpu)->stimer_pending_bitmap);
 577         stimer->msg_pending = false;
 578         stimer->exp_time = 0;
 579 }
 580
 581 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
 582 {
 583         struct kvm_vcpu_hv_stimer *stimer;
 584
 585         stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
 586         trace_kvm_hv_stimer_callback(hv_stimer_to_vcpu(stimer)->vcpu_id,
 587                                      stimer->index);
 588         stimer_mark_pending(stimer, true);
 589
 590         return HRTIMER_NORESTART;
 591 }
 592
 593 /*
 594  * stimer_start() assumptions:
 595  * a) stimer->count is not equal to 0
 596  * b) stimer->config has HV_STIMER_ENABLE flag
 597  */
 598 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
 599 {
 600         u64 time_now;
 601         ktime_t ktime_now;
 602
 603         time_now = get_time_ref_counter(hv_stimer_to_vcpu(stimer)->kvm);
 604         ktime_now = ktime_get();
 605
 606         if (stimer->config.periodic) {
 607                 if (stimer->exp_time) {
 608                         if (time_now >= stimer->exp_time) {
 609                                 u64 remainder;
 610
 611                                 div64_u64_rem(time_now - stimer->exp_time,
 612                                               stimer->count, &remainder);
 613                                 stimer->exp_time =
 614                                         time_now + (stimer->count - remainder);
 615                         }
 616                 } else
 617                         stimer->exp_time = time_now + stimer->count;
 618
 619                 trace_kvm_hv_stimer_start_periodic(
 620                                         hv_stimer_to_vcpu(stimer)->vcpu_id,
 621                                         stimer->index,
 622                                         time_now, stimer->exp_time);
 623
 624                 hrtimer_start(&stimer->timer,
 625                               ktime_add_ns(ktime_now,
 626                                            100 * (stimer->exp_time - time_now)),
 627                               HRTIMER_MODE_ABS);
 628                 return 0;
 629         }
 630         stimer->exp_time = stimer->count;
 631         if (time_now >= stimer->count) {
 632                 /*
 633                  * Expire timer according to Hypervisor Top-Level Functional
 634                  * specification v4(15.3.1):
 635                  * "If a one shot is enabled and the specified count is in
 636                  * the past, it will expire immediately."
 637                  */
 638                 stimer_mark_pending(stimer, false);
 639                 return 0;
 640         }
 641
 642         trace_kvm_hv_stimer_start_one_shot(hv_stimer_to_vcpu(stimer)->vcpu_id,
 643                                            stimer->index,
 644                                            time_now, stimer->count);
 645
 646         hrtimer_start(&stimer->timer,
 647                       ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)),
 648                       HRTIMER_MODE_ABS);
 649         return 0;
 650 }
 651
 652 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
 653                              bool host)
 654 {
 655         union hv_stimer_config new_config = {.as_uint64 = config},
 656                 old_config = {.as_uint64 = stimer->config.as_uint64};
 657         struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
 658         struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
 659         struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
 660
 661         if (!synic->active && !host)
 662                 return 1;
 663
 664         if (unlikely(!host && hv_vcpu->enforce_cpuid && new_config.direct_mode &&
 665                      !(hv_vcpu->cpuid_cache.features_edx &
 666                        HV_STIMER_DIRECT_MODE_AVAILABLE)))
 667                 return 1;
 668
 669         trace_kvm_hv_stimer_set_config(hv_stimer_to_vcpu(stimer)->vcpu_id,
 670                                        stimer->index, config, host);
 671
 672         stimer_cleanup(stimer);
 673         if (old_config.enable &&
 674             !new_config.direct_mode && new_config.sintx == 0)
 675                 new_config.enable = 0;
 676         stimer->config.as_uint64 = new_config.as_uint64;
 677
 678         if (stimer->config.enable)
 679                 stimer_mark_pending(stimer, false);
 680
 681         return 0;
 682 }
 683
 684 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
 685                             bool host)
 686 {
 687         struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
 688         struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
 689
 690         if (!synic->active && !host)
 691                 return 1;
 692
 693         trace_kvm_hv_stimer_set_count(hv_stimer_to_vcpu(stimer)->vcpu_id,
 694                                       stimer->index, count, host);
 695
 696         stimer_cleanup(stimer);
 697         stimer->count = count;
 698         if (stimer->count == 0)
 699                 stimer->config.enable = 0;
 700         else if (stimer->config.auto_enable)
 701                 stimer->config.enable = 1;
 702
 703         if (stimer->config.enable)
 704                 stimer_mark_pending(stimer, false);
 705
 706         return 0;
 707 }
 708
 709 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
 710 {
 711         *pconfig = stimer->config.as_uint64;
 712         return 0;
 713 }
 714
 715 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
 716 {
 717         *pcount = stimer->count;
 718         return 0;
 719 }
 720
 721 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
 722                              struct hv_message *src_msg, bool no_retry)
 723 {
 724         struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
 725         int msg_off = offsetof(struct hv_message_page, sint_message[sint]);
 726         gfn_t msg_page_gfn;
 727         struct hv_message_header hv_hdr;
 728         int r;
 729
 730         if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
 731                 return -ENOENT;
 732
 733         msg_page_gfn = synic->msg_page >> PAGE_SHIFT;
 734
 735         /*
 736          * Strictly following the spec-mandated ordering would assume setting
 737          * .msg_pending before checking .message_type.  However, this function
 738          * is only called in vcpu context so the entire update is atomic from
 739          * guest POV and thus the exact order here doesn't matter.
 740          */
 741         r = kvm_vcpu_read_guest_page(vcpu, msg_page_gfn, &hv_hdr.message_type,
 742                                      msg_off + offsetof(struct hv_message,
 743                                                         header.message_type),
 744                                      sizeof(hv_hdr.message_type));
 745         if (r < 0)
 746                 return r;
 747
 748         if (hv_hdr.message_type != HVMSG_NONE) {
 749                 if (no_retry)
 750                         return 0;
 751
 752                 hv_hdr.message_flags.msg_pending = 1;
 753                 r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn,
 754                                               &hv_hdr.message_flags,
 755                                               msg_off +
 756                                               offsetof(struct hv_message,
 757                                                        header.message_flags),
 758                                               sizeof(hv_hdr.message_flags));
 759                 if (r < 0)
 760                         return r;
 761                 return -EAGAIN;
 762         }
 763
 764         r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn, src_msg, msg_off,
 765                                       sizeof(src_msg->header) +
 766                                       src_msg->header.payload_size);
 767         if (r < 0)
 768                 return r;
 769
 770         r = synic_set_irq(synic, sint);
 771         if (r < 0)
 772                 return r;
 773         if (r == 0)
 774                 return -EFAULT;
 775         return 0;
 776 }
 777
 778 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
 779 {
 780         struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
 781         struct hv_message *msg = &stimer->msg;
 782         struct hv_timer_message_payload *payload =
 783                         (struct hv_timer_message_payload *)&msg->u.payload;
 784
 785         /*
 786          * To avoid piling up periodic ticks, don't retry message
 787          * delivery for them (within "lazy" lost ticks policy).
 788          */
 789         bool no_retry = stimer->config.periodic;
 790
 791         payload->expiration_time = stimer->exp_time;
 792         payload->delivery_time = get_time_ref_counter(vcpu->kvm);
 793         return synic_deliver_msg(to_hv_synic(vcpu),
 794                                  stimer->config.sintx, msg,
 795                                  no_retry);
 796 }
 797
 798 static int stimer_notify_direct(struct kvm_vcpu_hv_stimer *stimer)
 799 {
 800         struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
 801         struct kvm_lapic_irq irq = {
 802                 .delivery_mode = APIC_DM_FIXED,
 803                 .vector = stimer->config.apic_vector
 804         };
 805
 806         if (lapic_in_kernel(vcpu))
 807                 return !kvm_apic_set_irq(vcpu, &irq, NULL);
 808         return 0;
 809 }
 810
 811 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
 812 {
 813         int r, direct = stimer->config.direct_mode;
 814
 815         stimer->msg_pending = true;
 816         if (!direct)
 817                 r = stimer_send_msg(stimer);
 818         else
 819                 r = stimer_notify_direct(stimer);
 820         trace_kvm_hv_stimer_expiration(hv_stimer_to_vcpu(stimer)->vcpu_id,
 821                                        stimer->index, direct, r);
 822         if (!r) {
 823                 stimer->msg_pending = false;
 824                 if (!(stimer->config.periodic))
 825                         stimer->config.enable = 0;
 826         }
 827 }
 828
 829 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
 830 {
 831         struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
 832         struct kvm_vcpu_hv_stimer *stimer;
 833         u64 time_now, exp_time;
 834         int i;
 835
 836         if (!hv_vcpu)
 837                 return;
 838
 839         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 840                 if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
 841                         stimer = &hv_vcpu->stimer[i];
 842                         if (stimer->config.enable) {
 843                                 exp_time = stimer->exp_time;
 844
 845                                 if (exp_time) {
 846                                         time_now =
 847                                                 get_time_ref_counter(vcpu->kvm);
 848                                         if (time_now >= exp_time)
 849                                                 stimer_expiration(stimer);
 850                                 }
 851
 852                                 if ((stimer->config.enable) &&
 853                                     stimer->count) {
 854                                         if (!stimer->msg_pending)
 855                                                 stimer_start(stimer);
 856                                 } else
 857                                         stimer_cleanup(stimer);
 858                         }
 859                 }
 860 }
 861
 862 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
 863 {
 864         struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
 865         int i;
 866
 867         if (!hv_vcpu)
 868                 return;
 869
 870         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 871                 stimer_cleanup(&hv_vcpu->stimer[i]);
 872
 873         kfree(hv_vcpu);
 874         vcpu->arch.hyperv = NULL;
 875 }
 876
 877 bool kvm_hv_assist_page_enabled(struct kvm_vcpu *vcpu)
 878 {
 879         struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
 880
 881         if (!hv_vcpu)
 882                 return false;
 883
 884         if (!(hv_vcpu->hv_vapic & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE))
 885                 return false;
 886         return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
 887 }
 888 EXPORT_SYMBOL_GPL(kvm_hv_assist_page_enabled);
 889
 890 bool kvm_hv_get_assist_page(struct kvm_vcpu *vcpu,
 891                             struct hv_vp_assist_page *assist_page)
 892 {
 893         if (!kvm_hv_assist_page_enabled(vcpu))
 894                 return false;
 895         return !kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data,
 896                                       assist_page, sizeof(*assist_page));
 897 }
 898 EXPORT_SYMBOL_GPL(kvm_hv_get_assist_page);
 899
 900 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
 901 {
 902         struct hv_message *msg = &stimer->msg;
 903         struct hv_timer_message_payload *payload =
 904                         (struct hv_timer_message_payload *)&msg->u.payload;
 905
 906         memset(&msg->header, 0, sizeof(msg->header));
 907         msg->header.message_type = HVMSG_TIMER_EXPIRED;
 908         msg->header.payload_size = sizeof(*payload);
 909
 910         payload->timer_index = stimer->index;
 911         payload->expiration_time = 0;
 912         payload->delivery_time = 0;
 913 }
 914
 915 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
 916 {
 917         memset(stimer, 0, sizeof(*stimer));
 918         stimer->index = timer_index;
 919         hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
 920         stimer->timer.function = stimer_timer_callback;
 921         stimer_prepare_msg(stimer);
 922 }
 923
 924 static int kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
 925 {
 926         struct kvm_vcpu_hv *hv_vcpu;
 927         int i;
 928
 929         hv_vcpu = kzalloc(sizeof(struct kvm_vcpu_hv), GFP_KERNEL_ACCOUNT);
 930         if (!hv_vcpu)
 931                 return -ENOMEM;
 932
 933         vcpu->arch.hyperv = hv_vcpu;
 934         hv_vcpu->vcpu = vcpu;
 935
 936         synic_init(&hv_vcpu->synic);
 937
 938         bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
 939         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 940                 stimer_init(&hv_vcpu->stimer[i], i);
 941
 942         hv_vcpu->vp_index = vcpu->vcpu_idx;
 943
 944         return 0;
 945 }
 946
 947 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
 948 {
 949         struct kvm_vcpu_hv_synic *synic;
 950         int r;
 951
 952         if (!to_hv_vcpu(vcpu)) {
 953                 r = kvm_hv_vcpu_init(vcpu);
 954                 if (r)
 955                         return r;
 956         }
 957
 958         synic = to_hv_synic(vcpu);
 959
 960         synic->active = true;
 961         synic->dont_zero_synic_pages = dont_zero_synic_pages;
 962         synic->control = HV_SYNIC_CONTROL_ENABLE;
 963         return 0;
 964 }
 965
 966 static bool kvm_hv_msr_partition_wide(u32 msr)
 967 {
 968         bool r = false;
 969
 970         switch (msr) {
 971         case HV_X64_MSR_GUEST_OS_ID:
 972         case HV_X64_MSR_HYPERCALL:
 973         case HV_X64_MSR_REFERENCE_TSC:
 974         case HV_X64_MSR_TIME_REF_COUNT:
 975         case HV_X64_MSR_CRASH_CTL:
 976         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
 977         case HV_X64_MSR_RESET:
 978         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
 979         case HV_X64_MSR_TSC_EMULATION_CONTROL:
 980         case HV_X64_MSR_TSC_EMULATION_STATUS:
 981         case HV_X64_MSR_SYNDBG_OPTIONS:
 982         case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
 983                 r = true;
 984                 break;
 985         }
 986
 987         return r;
 988 }
 989
 990 static int kvm_hv_msr_get_crash_data(struct kvm *kvm, u32 index, u64 *pdata)
 991 {
 992         struct kvm_hv *hv = to_kvm_hv(kvm);
 993         size_t size = ARRAY_SIZE(hv->hv_crash_param);
 994
 995         if (WARN_ON_ONCE(index >= size))
 996                 return -EINVAL;
 997
 998         *pdata = hv->hv_crash_param[array_index_nospec(index, size)];
 999         return 0;
1000 }
1001
1002 static int kvm_hv_msr_get_crash_ctl(struct kvm *kvm, u64 *pdata)
1003 {
1004         struct kvm_hv *hv = to_kvm_hv(kvm);
1005
1006         *pdata = hv->hv_crash_ctl;
1007         return 0;
1008 }
1009
1010 static int kvm_hv_msr_set_crash_ctl(struct kvm *kvm, u64 data)
1011 {
1012         struct kvm_hv *hv = to_kvm_hv(kvm);
1013
1014         hv->hv_crash_ctl = data & HV_CRASH_CTL_CRASH_NOTIFY;
1015
1016         return 0;
1017 }
1018
1019 static int kvm_hv_msr_set_crash_data(struct kvm *kvm, u32 index, u64 data)
1020 {
1021         struct kvm_hv *hv = to_kvm_hv(kvm);
1022         size_t size = ARRAY_SIZE(hv->hv_crash_param);
1023
1024         if (WARN_ON_ONCE(index >= size))
1025                 return -EINVAL;
1026
1027         hv->hv_crash_param[array_index_nospec(index, size)] = data;
1028         return 0;
1029 }
1030
1031 /*
1032  * The kvmclock and Hyper-V TSC page use similar formulas, and converting
1033  * between them is possible:
1034  *
1035  * kvmclock formula:
1036  *    nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
1037  *           + system_time
1038  *
1039  * Hyper-V formula:
1040  *    nsec/100 = ticks * scale / 2^64 + offset
1041  *
1042  * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
1043  * By dividing the kvmclock formula by 100 and equating what's left we get:
1044  *    ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1045  *            scale / 2^64 =         tsc_to_system_mul * 2^(tsc_shift-32) / 100
1046  *            scale        =         tsc_to_system_mul * 2^(32+tsc_shift) / 100
1047  *
1048  * Now expand the kvmclock formula and divide by 100:
1049  *    nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
1050  *           - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
1051  *           + system_time
1052  *    nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1053  *               - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1054  *               + system_time / 100
1055  *
1056  * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
1057  *    nsec/100 = ticks * scale / 2^64
1058  *               - tsc_timestamp * scale / 2^64
1059  *               + system_time / 100
1060  *
1061  * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
1062  *    offset = system_time / 100 - tsc_timestamp * scale / 2^64
1063  *
1064  * These two equivalencies are implemented in this function.
1065  */
1066 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
1067                                         struct ms_hyperv_tsc_page *tsc_ref)
1068 {
1069         u64 max_mul;
1070
1071         if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
1072                 return false;
1073
1074         /*
1075          * check if scale would overflow, if so we use the time ref counter
1076          *    tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
1077          *    tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
1078          *    tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
1079          */
1080         max_mul = 100ull << (32 - hv_clock->tsc_shift);
1081         if (hv_clock->tsc_to_system_mul >= max_mul)
1082                 return false;
1083
1084         /*
1085          * Otherwise compute the scale and offset according to the formulas
1086          * derived above.
1087          */
1088         tsc_ref->tsc_scale =
1089                 mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
1090                                 hv_clock->tsc_to_system_mul,
1091                                 100);
1092
1093         tsc_ref->tsc_offset = hv_clock->system_time;
1094         do_div(tsc_ref->tsc_offset, 100);
1095         tsc_ref->tsc_offset -=
1096                 mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
1097         return true;
1098 }
1099
1100 /*
1101  * Don't touch TSC page values if the guest has opted for TSC emulation after
1102  * migration. KVM doesn't fully support reenlightenment notifications and TSC
1103  * access emulation and Hyper-V is known to expect the values in TSC page to
1104  * stay constant before TSC access emulation is disabled from guest side
1105  * (HV_X64_MSR_TSC_EMULATION_STATUS). KVM userspace is expected to preserve TSC
1106  * frequency and guest visible TSC value across migration (and prevent it when
1107  * TSC scaling is unsupported).
1108  */
1109 static inline bool tsc_page_update_unsafe(struct kvm_hv *hv)
1110 {
1111         return (hv->hv_tsc_page_status != HV_TSC_PAGE_GUEST_CHANGED) &&
1112                 hv->hv_tsc_emulation_control;
1113 }
1114
1115 void kvm_hv_setup_tsc_page(struct kvm *kvm,
1116                            struct pvclock_vcpu_time_info *hv_clock)
1117 {
1118         struct kvm_hv *hv = to_kvm_hv(kvm);
1119         u32 tsc_seq;
1120         u64 gfn;
1121
1122         BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
1123         BUILD_BUG_ON(offsetof(struct ms_hyperv_tsc_page, tsc_sequence) != 0);
1124
1125         if (hv->hv_tsc_page_status == HV_TSC_PAGE_BROKEN ||
1126             hv->hv_tsc_page_status == HV_TSC_PAGE_UNSET)
1127                 return;
1128
1129         mutex_lock(&hv->hv_lock);
1130         if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1131                 goto out_unlock;
1132
1133         gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
1134         /*
1135          * Because the TSC parameters only vary when there is a
1136          * change in the master clock, do not bother with caching.
1137          */
1138         if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
1139                                     &tsc_seq, sizeof(tsc_seq))))
1140                 goto out_err;
1141
1142         if (tsc_seq && tsc_page_update_unsafe(hv)) {
1143                 if (kvm_read_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
1144                         goto out_err;
1145
1146                 hv->hv_tsc_page_status = HV_TSC_PAGE_SET;
1147                 goto out_unlock;
1148         }
1149
1150         /*
1151          * While we're computing and writing the parameters, force the
1152          * guest to use the time reference count MSR.
1153          */
1154         hv->tsc_ref.tsc_sequence = 0;
1155         if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
1156                             &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
1157                 goto out_err;
1158
1159         if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
1160                 goto out_err;
1161
1162         /* Ensure sequence is zero before writing the rest of the struct.  */
1163         smp_wmb();
1164         if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
1165                 goto out_err;
1166
1167         /*
1168          * Now switch to the TSC page mechanism by writing the sequence.
1169          */
1170         tsc_seq++;
1171         if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
1172                 tsc_seq = 1;
1173
1174         /* Write the struct entirely before the non-zero sequence.  */
1175         smp_wmb();
1176
1177         hv->tsc_ref.tsc_sequence = tsc_seq;
1178         if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
1179                             &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
1180                 goto out_err;
1181
1182         hv->hv_tsc_page_status = HV_TSC_PAGE_SET;
1183         goto out_unlock;
1184
1185 out_err:
1186         hv->hv_tsc_page_status = HV_TSC_PAGE_BROKEN;
1187 out_unlock:
1188         mutex_unlock(&hv->hv_lock);
1189 }
1190
1191 void kvm_hv_invalidate_tsc_page(struct kvm *kvm)
1192 {
1193         struct kvm_hv *hv = to_kvm_hv(kvm);
1194         u64 gfn;
1195         int idx;
1196
1197         if (hv->hv_tsc_page_status == HV_TSC_PAGE_BROKEN ||
1198             hv->hv_tsc_page_status == HV_TSC_PAGE_UNSET ||
1199             tsc_page_update_unsafe(hv))
1200                 return;
1201
1202         mutex_lock(&hv->hv_lock);
1203
1204         if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1205                 goto out_unlock;
1206
1207         /* Preserve HV_TSC_PAGE_GUEST_CHANGED/HV_TSC_PAGE_HOST_CHANGED states */
1208         if (hv->hv_tsc_page_status == HV_TSC_PAGE_SET)
1209                 hv->hv_tsc_page_status = HV_TSC_PAGE_UPDATING;
1210
1211         gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
1212
1213         hv->tsc_ref.tsc_sequence = 0;
1214
1215         /*
1216          * Take the srcu lock as memslots will be accessed to check the gfn
1217          * cache generation against the memslots generation.
1218          */
1219         idx = srcu_read_lock(&kvm->srcu);
1220         if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
1221                             &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
1222                 hv->hv_tsc_page_status = HV_TSC_PAGE_BROKEN;
1223         srcu_read_unlock(&kvm->srcu, idx);
1224
1225 out_unlock:
1226         mutex_unlock(&hv->hv_lock);
1227 }
1228
1229
1230 static bool hv_check_msr_access(struct kvm_vcpu_hv *hv_vcpu, u32 msr)
1231 {
1232         if (!hv_vcpu->enforce_cpuid)
1233                 return true;
1234
1235         switch (msr) {
1236         case HV_X64_MSR_GUEST_OS_ID:
1237         case HV_X64_MSR_HYPERCALL:
1238                 return hv_vcpu->cpuid_cache.features_eax &
1239                         HV_MSR_HYPERCALL_AVAILABLE;
1240         case HV_X64_MSR_VP_RUNTIME:
1241                 return hv_vcpu->cpuid_cache.features_eax &
1242                         HV_MSR_VP_RUNTIME_AVAILABLE;
1243         case HV_X64_MSR_TIME_REF_COUNT:
1244                 return hv_vcpu->cpuid_cache.features_eax &
1245                         HV_MSR_TIME_REF_COUNT_AVAILABLE;
1246         case HV_X64_MSR_VP_INDEX:
1247                 return hv_vcpu->cpuid_cache.features_eax &
1248                         HV_MSR_VP_INDEX_AVAILABLE;
1249         case HV_X64_MSR_RESET:
1250                 return hv_vcpu->cpuid_cache.features_eax &
1251                         HV_MSR_RESET_AVAILABLE;
1252         case HV_X64_MSR_REFERENCE_TSC:
1253                 return hv_vcpu->cpuid_cache.features_eax &
1254                         HV_MSR_REFERENCE_TSC_AVAILABLE;
1255         case HV_X64_MSR_SCONTROL:
1256         case HV_X64_MSR_SVERSION:
1257         case HV_X64_MSR_SIEFP:
1258         case HV_X64_MSR_SIMP:
1259         case HV_X64_MSR_EOM:
1260         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1261                 return hv_vcpu->cpuid_cache.features_eax &
1262                         HV_MSR_SYNIC_AVAILABLE;
1263         case HV_X64_MSR_STIMER0_CONFIG:
1264         case HV_X64_MSR_STIMER1_CONFIG:
1265         case HV_X64_MSR_STIMER2_CONFIG:
1266         case HV_X64_MSR_STIMER3_CONFIG:
1267         case HV_X64_MSR_STIMER0_COUNT:
1268         case HV_X64_MSR_STIMER1_COUNT:
1269         case HV_X64_MSR_STIMER2_COUNT:
1270         case HV_X64_MSR_STIMER3_COUNT:
1271                 return hv_vcpu->cpuid_cache.features_eax &
1272                         HV_MSR_SYNTIMER_AVAILABLE;
1273         case HV_X64_MSR_EOI:
1274         case HV_X64_MSR_ICR:
1275         case HV_X64_MSR_TPR:
1276         case HV_X64_MSR_VP_ASSIST_PAGE:
1277                 return hv_vcpu->cpuid_cache.features_eax &
1278                         HV_MSR_APIC_ACCESS_AVAILABLE;
1279                 break;
1280         case HV_X64_MSR_TSC_FREQUENCY:
1281         case HV_X64_MSR_APIC_FREQUENCY:
1282                 return hv_vcpu->cpuid_cache.features_eax &
1283                         HV_ACCESS_FREQUENCY_MSRS;
1284         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1285         case HV_X64_MSR_TSC_EMULATION_CONTROL:
1286         case HV_X64_MSR_TSC_EMULATION_STATUS:
1287                 return hv_vcpu->cpuid_cache.features_eax &
1288                         HV_ACCESS_REENLIGHTENMENT;
1289         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1290         case HV_X64_MSR_CRASH_CTL:
1291                 return hv_vcpu->cpuid_cache.features_edx &
1292                         HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
1293         case HV_X64_MSR_SYNDBG_OPTIONS:
1294         case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1295                 return hv_vcpu->cpuid_cache.features_edx &
1296                         HV_FEATURE_DEBUG_MSRS_AVAILABLE;
1297         default:
1298                 break;
1299         }
1300
1301         return false;
1302 }
1303
1304 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
1305                              bool host)
1306 {
1307         struct kvm *kvm = vcpu->kvm;
1308         struct kvm_hv *hv = to_kvm_hv(kvm);
1309
1310         if (unlikely(!host && !hv_check_msr_access(to_hv_vcpu(vcpu), msr)))
1311                 return 1;
1312
1313         switch (msr) {
1314         case HV_X64_MSR_GUEST_OS_ID:
1315                 hv->hv_guest_os_id = data;
1316                 /* setting guest os id to zero disables hypercall page */
1317                 if (!hv->hv_guest_os_id)
1318                         hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1319                 break;
1320         case HV_X64_MSR_HYPERCALL: {
1321                 u8 instructions[9];
1322                 int i = 0;
1323                 u64 addr;
1324
1325                 /* if guest os id is not set hypercall should remain disabled */
1326                 if (!hv->hv_guest_os_id)
1327                         break;
1328                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1329                         hv->hv_hypercall = data;
1330                         break;
1331                 }
1332
1333                 /*
1334                  * If Xen and Hyper-V hypercalls are both enabled, disambiguate
1335                  * the same way Xen itself does, by setting the bit 31 of EAX
1336                  * which is RsvdZ in the 32-bit Hyper-V hypercall ABI and just
1337                  * going to be clobbered on 64-bit.
1338                  */
1339                 if (kvm_xen_hypercall_enabled(kvm)) {
1340                         /* orl $0x80000000, %eax */
1341                         instructions[i++] = 0x0d;
1342                         instructions[i++] = 0x00;
1343                         instructions[i++] = 0x00;
1344                         instructions[i++] = 0x00;
1345                         instructions[i++] = 0x80;
1346                 }
1347
1348                 /* vmcall/vmmcall */
1349                 static_call(kvm_x86_patch_hypercall)(vcpu, instructions + i);
1350                 i += 3;
1351
1352                 /* ret */
1353                 ((unsigned char *)instructions)[i++] = 0xc3;
1354
1355                 addr = data & HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_MASK;
1356                 if (kvm_vcpu_write_guest(vcpu, addr, instructions, i))
1357                         return 1;
1358                 hv->hv_hypercall = data;
1359                 break;
1360         }
1361         case HV_X64_MSR_REFERENCE_TSC:
1362                 hv->hv_tsc_page = data;
1363                 if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE) {
1364                         if (!host)
1365                                 hv->hv_tsc_page_status = HV_TSC_PAGE_GUEST_CHANGED;
1366                         else
1367                                 hv->hv_tsc_page_status = HV_TSC_PAGE_HOST_CHANGED;
1368                         kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1369                 } else {
1370                         hv->hv_tsc_page_status = HV_TSC_PAGE_UNSET;
1371                 }
1372                 break;
1373         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1374                 return kvm_hv_msr_set_crash_data(kvm,
1375                                                  msr - HV_X64_MSR_CRASH_P0,
1376                                                  data);
1377         case HV_X64_MSR_CRASH_CTL:
1378                 if (host)
1379                         return kvm_hv_msr_set_crash_ctl(kvm, data);
1380
1381                 if (data & HV_CRASH_CTL_CRASH_NOTIFY) {
1382                         vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
1383                                    hv->hv_crash_param[0],
1384                                    hv->hv_crash_param[1],
1385                                    hv->hv_crash_param[2],
1386                                    hv->hv_crash_param[3],
1387                                    hv->hv_crash_param[4]);
1388
1389                         /* Send notification about crash to user space */
1390                         kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
1391                 }
1392                 break;
1393         case HV_X64_MSR_RESET:
1394                 if (data == 1) {
1395                         vcpu_debug(vcpu, "hyper-v reset requested\n");
1396                         kvm_make_request(KVM_REQ_HV_RESET, vcpu);
1397                 }
1398                 break;
1399         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1400                 hv->hv_reenlightenment_control = data;
1401                 break;
1402         case HV_X64_MSR_TSC_EMULATION_CONTROL:
1403                 hv->hv_tsc_emulation_control = data;
1404                 break;
1405         case HV_X64_MSR_TSC_EMULATION_STATUS:
1406                 if (data && !host)
1407                         return 1;
1408
1409                 hv->hv_tsc_emulation_status = data;
1410                 break;
1411         case HV_X64_MSR_TIME_REF_COUNT:
1412                 /* read-only, but still ignore it if host-initiated */
1413                 if (!host)
1414                         return 1;
1415                 break;
1416         case HV_X64_MSR_SYNDBG_OPTIONS:
1417         case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1418                 return syndbg_set_msr(vcpu, msr, data, host);
1419         default:
1420                 vcpu_unimpl(vcpu, "Hyper-V unhandled wrmsr: 0x%x data 0x%llx\n",
1421                             msr, data);
1422                 return 1;
1423         }
1424         return 0;
1425 }
1426
1427 /* Calculate cpu time spent by current task in 100ns units */
1428 static u64 current_task_runtime_100ns(void)
1429 {
1430         u64 utime, stime;
1431
1432         task_cputime_adjusted(current, &utime, &stime);
1433
1434         return div_u64(utime + stime, 100);
1435 }
1436
1437 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1438 {
1439         struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
1440
1441         if (unlikely(!host && !hv_check_msr_access(hv_vcpu, msr)))
1442                 return 1;
1443
1444         switch (msr) {
1445         case HV_X64_MSR_VP_INDEX: {
1446                 struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1447                 u32 new_vp_index = (u32)data;
1448
1449                 if (!host || new_vp_index >= KVM_MAX_VCPUS)
1450                         return 1;
1451
1452                 if (new_vp_index == hv_vcpu->vp_index)
1453                         return 0;
1454
1455                 /*
1456                  * The VP index is initialized to vcpu_index by
1457                  * kvm_hv_vcpu_postcreate so they initially match.  Now the
1458                  * VP index is changing, adjust num_mismatched_vp_indexes if
1459                  * it now matches or no longer matches vcpu_idx.
1460                  */
1461                 if (hv_vcpu->vp_index == vcpu->vcpu_idx)
1462                         atomic_inc(&hv->num_mismatched_vp_indexes);
1463                 else if (new_vp_index == vcpu->vcpu_idx)
1464                         atomic_dec(&hv->num_mismatched_vp_indexes);
1465
1466                 hv_vcpu->vp_index = new_vp_index;
1467                 break;
1468         }
1469         case HV_X64_MSR_VP_ASSIST_PAGE: {
1470                 u64 gfn;
1471                 unsigned long addr;
1472
1473                 if (!(data & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE)) {
1474                         hv_vcpu->hv_vapic = data;
1475                         if (kvm_lapic_set_pv_eoi(vcpu, 0, 0))
1476                                 return 1;
1477                         break;
1478                 }
1479                 gfn = data >> HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT;
1480                 addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1481                 if (kvm_is_error_hva(addr))
1482                         return 1;
1483
1484                 /*
1485                  * Clear apic_assist portion of struct hv_vp_assist_page
1486                  * only, there can be valuable data in the rest which needs
1487                  * to be preserved e.g. on migration.
1488                  */
1489                 if (__put_user(0, (u32 __user *)addr))
1490                         return 1;
1491                 hv_vcpu->hv_vapic = data;
1492                 kvm_vcpu_mark_page_dirty(vcpu, gfn);
1493                 if (kvm_lapic_set_pv_eoi(vcpu,
1494                                             gfn_to_gpa(gfn) | KVM_MSR_ENABLED,
1495                                             sizeof(struct hv_vp_assist_page)))
1496                         return 1;
1497                 break;
1498         }
1499         case HV_X64_MSR_EOI:
1500                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1501         case HV_X64_MSR_ICR:
1502                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1503         case HV_X64_MSR_TPR:
1504                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1505         case HV_X64_MSR_VP_RUNTIME:
1506                 if (!host)
1507                         return 1;
1508                 hv_vcpu->runtime_offset = data - current_task_runtime_100ns();
1509                 break;
1510         case HV_X64_MSR_SCONTROL:
1511         case HV_X64_MSR_SVERSION:
1512         case HV_X64_MSR_SIEFP:
1513         case HV_X64_MSR_SIMP:
1514         case HV_X64_MSR_EOM:
1515         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1516                 return synic_set_msr(to_hv_synic(vcpu), msr, data, host);
1517         case HV_X64_MSR_STIMER0_CONFIG:
1518         case HV_X64_MSR_STIMER1_CONFIG:
1519         case HV_X64_MSR_STIMER2_CONFIG:
1520         case HV_X64_MSR_STIMER3_CONFIG: {
1521                 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1522
1523                 return stimer_set_config(to_hv_stimer(vcpu, timer_index),
1524                                          data, host);
1525         }
1526         case HV_X64_MSR_STIMER0_COUNT:
1527         case HV_X64_MSR_STIMER1_COUNT:
1528         case HV_X64_MSR_STIMER2_COUNT:
1529         case HV_X64_MSR_STIMER3_COUNT: {
1530                 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1531
1532                 return stimer_set_count(to_hv_stimer(vcpu, timer_index),
1533                                         data, host);
1534         }
1535         case HV_X64_MSR_TSC_FREQUENCY:
1536         case HV_X64_MSR_APIC_FREQUENCY:
1537                 /* read-only, but still ignore it if host-initiated */
1538                 if (!host)
1539                         return 1;
1540                 break;
1541         default:
1542                 vcpu_unimpl(vcpu, "Hyper-V unhandled wrmsr: 0x%x data 0x%llx\n",
1543                             msr, data);
1544                 return 1;
1545         }
1546
1547         return 0;
1548 }
1549
1550 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1551                              bool host)
1552 {
1553         u64 data = 0;
1554         struct kvm *kvm = vcpu->kvm;
1555         struct kvm_hv *hv = to_kvm_hv(kvm);
1556
1557         if (unlikely(!host && !hv_check_msr_access(to_hv_vcpu(vcpu), msr)))
1558                 return 1;
1559
1560         switch (msr) {
1561         case HV_X64_MSR_GUEST_OS_ID:
1562                 data = hv->hv_guest_os_id;
1563                 break;
1564         case HV_X64_MSR_HYPERCALL:
1565                 data = hv->hv_hypercall;
1566                 break;
1567         case HV_X64_MSR_TIME_REF_COUNT:
1568                 data = get_time_ref_counter(kvm);
1569                 break;
1570         case HV_X64_MSR_REFERENCE_TSC:
1571                 data = hv->hv_tsc_page;
1572                 break;
1573         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1574                 return kvm_hv_msr_get_crash_data(kvm,
1575                                                  msr - HV_X64_MSR_CRASH_P0,
1576                                                  pdata);
1577         case HV_X64_MSR_CRASH_CTL:
1578                 return kvm_hv_msr_get_crash_ctl(kvm, pdata);
1579         case HV_X64_MSR_RESET:
1580                 data = 0;
1581                 break;
1582         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1583                 data = hv->hv_reenlightenment_control;
1584                 break;
1585         case HV_X64_MSR_TSC_EMULATION_CONTROL:
1586                 data = hv->hv_tsc_emulation_control;
1587                 break;
1588         case HV_X64_MSR_TSC_EMULATION_STATUS:
1589                 data = hv->hv_tsc_emulation_status;
1590                 break;
1591         case HV_X64_MSR_SYNDBG_OPTIONS:
1592         case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1593                 return syndbg_get_msr(vcpu, msr, pdata, host);
1594         default:
1595                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1596                 return 1;
1597         }
1598
1599         *pdata = data;
1600         return 0;
1601 }
1602
1603 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1604                           bool host)
1605 {
1606         u64 data = 0;
1607         struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
1608
1609         if (unlikely(!host && !hv_check_msr_access(hv_vcpu, msr)))
1610                 return 1;
1611
1612         switch (msr) {
1613         case HV_X64_MSR_VP_INDEX:
1614                 data = hv_vcpu->vp_index;
1615                 break;
1616         case HV_X64_MSR_EOI:
1617                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1618         case HV_X64_MSR_ICR:
1619                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1620         case HV_X64_MSR_TPR:
1621                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1622         case HV_X64_MSR_VP_ASSIST_PAGE:
1623                 data = hv_vcpu->hv_vapic;
1624                 break;
1625         case HV_X64_MSR_VP_RUNTIME:
1626                 data = current_task_runtime_100ns() + hv_vcpu->runtime_offset;
1627                 break;
1628         case HV_X64_MSR_SCONTROL:
1629         case HV_X64_MSR_SVERSION:
1630         case HV_X64_MSR_SIEFP:
1631         case HV_X64_MSR_SIMP:
1632         case HV_X64_MSR_EOM:
1633         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1634                 return synic_get_msr(to_hv_synic(vcpu), msr, pdata, host);
1635         case HV_X64_MSR_STIMER0_CONFIG:
1636         case HV_X64_MSR_STIMER1_CONFIG:
1637         case HV_X64_MSR_STIMER2_CONFIG:
1638         case HV_X64_MSR_STIMER3_CONFIG: {
1639                 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1640
1641                 return stimer_get_config(to_hv_stimer(vcpu, timer_index),
1642                                          pdata);
1643         }
1644         case HV_X64_MSR_STIMER0_COUNT:
1645         case HV_X64_MSR_STIMER1_COUNT:
1646         case HV_X64_MSR_STIMER2_COUNT:
1647         case HV_X64_MSR_STIMER3_COUNT: {
1648                 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1649
1650                 return stimer_get_count(to_hv_stimer(vcpu, timer_index),
1651                                         pdata);
1652         }
1653         case HV_X64_MSR_TSC_FREQUENCY:
1654                 data = (u64)vcpu->arch.virtual_tsc_khz * 1000;
1655                 break;
1656         case HV_X64_MSR_APIC_FREQUENCY:
1657                 data = APIC_BUS_FREQUENCY;
1658                 break;
1659         default:
1660                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1661                 return 1;
1662         }
1663         *pdata = data;
1664         return 0;
1665 }
1666
1667 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1668 {
1669         struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1670
1671         if (!host && !vcpu->arch.hyperv_enabled)
1672                 return 1;
1673
1674         if (!to_hv_vcpu(vcpu)) {
1675                 if (kvm_hv_vcpu_init(vcpu))
1676                         return 1;
1677         }
1678
1679         if (kvm_hv_msr_partition_wide(msr)) {
1680                 int r;
1681
1682                 mutex_lock(&hv->hv_lock);
1683                 r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1684                 mutex_unlock(&hv->hv_lock);
1685                 return r;
1686         } else
1687                 return kvm_hv_set_msr(vcpu, msr, data, host);
1688 }
1689
1690 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
1691 {
1692         struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1693
1694         if (!host && !vcpu->arch.hyperv_enabled)
1695                 return 1;
1696
1697         if (!to_hv_vcpu(vcpu)) {
1698                 if (kvm_hv_vcpu_init(vcpu))
1699                         return 1;
1700         }
1701
1702         if (kvm_hv_msr_partition_wide(msr)) {
1703                 int r;
1704
1705                 mutex_lock(&hv->hv_lock);
1706                 r = kvm_hv_get_msr_pw(vcpu, msr, pdata, host);
1707                 mutex_unlock(&hv->hv_lock);
1708                 return r;
1709         } else
1710                 return kvm_hv_get_msr(vcpu, msr, pdata, host);
1711 }
1712
1713 static __always_inline unsigned long *sparse_set_to_vcpu_mask(
1714         struct kvm *kvm, u64 *sparse_banks, u64 valid_bank_mask,
1715         u64 *vp_bitmap, unsigned long *vcpu_bitmap)
1716 {
1717         struct kvm_hv *hv = to_kvm_hv(kvm);
1718         struct kvm_vcpu *vcpu;
1719         int bank, sbank = 0;
1720         unsigned long i;
1721
1722         memset(vp_bitmap, 0,
1723                KVM_HV_MAX_SPARSE_VCPU_SET_BITS * sizeof(*vp_bitmap));
1724         for_each_set_bit(bank, (unsigned long *)&valid_bank_mask,
1725                          KVM_HV_MAX_SPARSE_VCPU_SET_BITS)
1726                 vp_bitmap[bank] = sparse_banks[sbank++];
1727
1728         if (likely(!atomic_read(&hv->num_mismatched_vp_indexes))) {
1729                 /* for all vcpus vp_index == vcpu_idx */
1730                 return (unsigned long *)vp_bitmap;
1731         }
1732
1733         bitmap_zero(vcpu_bitmap, KVM_MAX_VCPUS);
1734         kvm_for_each_vcpu(i, vcpu, kvm) {
1735                 if (test_bit(kvm_hv_get_vpindex(vcpu), (unsigned long *)vp_bitmap))
1736                         __set_bit(i, vcpu_bitmap);
1737         }
1738         return vcpu_bitmap;
1739 }
1740
1741 struct kvm_hv_hcall {
1742         u64 param;
1743         u64 ingpa;
1744         u64 outgpa;
1745         u16 code;
1746         u16 rep_cnt;
1747         u16 rep_idx;
1748         bool fast;
1749         bool rep;
1750         sse128_t xmm[HV_HYPERCALL_MAX_XMM_REGISTERS];
1751 };
1752
1753 static u64 kvm_hv_flush_tlb(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc, bool ex)
1754 {
1755         int i;
1756         gpa_t gpa;
1757         struct kvm *kvm = vcpu->kvm;
1758         struct hv_tlb_flush_ex flush_ex;
1759         struct hv_tlb_flush flush;
1760         u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1761         DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1762         unsigned long *vcpu_mask;
1763         u64 valid_bank_mask;
1764         u64 sparse_banks[64];
1765         int sparse_banks_len;
1766         bool all_cpus;
1767
1768         if (!ex) {
1769                 if (hc->fast) {
1770                         flush.address_space = hc->ingpa;
1771                         flush.flags = hc->outgpa;
1772                         flush.processor_mask = sse128_lo(hc->xmm[0]);
1773                 } else {
1774                         if (unlikely(kvm_read_guest(kvm, hc->ingpa,
1775                                                     &flush, sizeof(flush))))
1776                                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1777                 }
1778
1779                 trace_kvm_hv_flush_tlb(flush.processor_mask,
1780                                        flush.address_space, flush.flags);
1781
1782                 valid_bank_mask = BIT_ULL(0);
1783                 sparse_banks[0] = flush.processor_mask;
1784
1785                 /*
1786                  * Work around possible WS2012 bug: it sends hypercalls
1787                  * with processor_mask = 0x0 and HV_FLUSH_ALL_PROCESSORS clear,
1788                  * while also expecting us to flush something and crashing if
1789                  * we don't. Let's treat processor_mask == 0 same as
1790                  * HV_FLUSH_ALL_PROCESSORS.
1791                  */
1792                 all_cpus = (flush.flags & HV_FLUSH_ALL_PROCESSORS) ||
1793                         flush.processor_mask == 0;
1794         } else {
1795                 if (hc->fast) {
1796                         flush_ex.address_space = hc->ingpa;
1797                         flush_ex.flags = hc->outgpa;
1798                         memcpy(&flush_ex.hv_vp_set,
1799                                &hc->xmm[0], sizeof(hc->xmm[0]));
1800                 } else {
1801                         if (unlikely(kvm_read_guest(kvm, hc->ingpa, &flush_ex,
1802                                                     sizeof(flush_ex))))
1803                                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1804                 }
1805
1806                 trace_kvm_hv_flush_tlb_ex(flush_ex.hv_vp_set.valid_bank_mask,
1807                                           flush_ex.hv_vp_set.format,
1808                                           flush_ex.address_space,
1809                                           flush_ex.flags);
1810
1811                 valid_bank_mask = flush_ex.hv_vp_set.valid_bank_mask;
1812                 all_cpus = flush_ex.hv_vp_set.format !=
1813                         HV_GENERIC_SET_SPARSE_4K;
1814
1815                 sparse_banks_len = bitmap_weight((unsigned long *)&valid_bank_mask, 64);
1816
1817                 if (!sparse_banks_len && !all_cpus)
1818                         goto ret_success;
1819
1820                 if (!all_cpus) {
1821                         if (hc->fast) {
1822                                 if (sparse_banks_len > HV_HYPERCALL_MAX_XMM_REGISTERS - 1)
1823                                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
1824                                 for (i = 0; i < sparse_banks_len; i += 2) {
1825                                         sparse_banks[i] = sse128_lo(hc->xmm[i / 2 + 1]);
1826                                         sparse_banks[i + 1] = sse128_hi(hc->xmm[i / 2 + 1]);
1827                                 }
1828                         } else {
1829                                 gpa = hc->ingpa + offsetof(struct hv_tlb_flush_ex,
1830                                                            hv_vp_set.bank_contents);
1831                                 if (unlikely(kvm_read_guest(kvm, gpa, sparse_banks,
1832                                                             sparse_banks_len *
1833                                                             sizeof(sparse_banks[0]))))
1834                                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
1835                         }
1836                 }
1837         }
1838
1839         /*
1840          * vcpu->arch.cr3 may not be up-to-date for running vCPUs so we can't
1841          * analyze it here, flush TLB regardless of the specified address space.
1842          */
1843         if (all_cpus) {
1844                 kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH_GUEST);
1845         } else {
1846                 vcpu_mask = sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1847                                                     vp_bitmap, vcpu_bitmap);
1848
1849                 kvm_make_vcpus_request_mask(kvm, KVM_REQ_TLB_FLUSH_GUEST,
1850                                             vcpu_mask);
1851         }
1852
1853 ret_success:
1854         /* We always do full TLB flush, set 'Reps completed' = 'Rep Count' */
1855         return (u64)HV_STATUS_SUCCESS |
1856                 ((u64)hc->rep_cnt << HV_HYPERCALL_REP_COMP_OFFSET);
1857 }
1858
1859 static void kvm_send_ipi_to_many(struct kvm *kvm, u32 vector,
1860                                  unsigned long *vcpu_bitmap)
1861 {
1862         struct kvm_lapic_irq irq = {
1863                 .delivery_mode = APIC_DM_FIXED,
1864                 .vector = vector
1865         };
1866         struct kvm_vcpu *vcpu;
1867         unsigned long i;
1868
1869         kvm_for_each_vcpu(i, vcpu, kvm) {
1870                 if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
1871                         continue;
1872
1873                 /* We fail only when APIC is disabled */
1874                 kvm_apic_set_irq(vcpu, &irq, NULL);
1875         }
1876 }
1877
1878 static u64 kvm_hv_send_ipi(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc, bool ex)
1879 {
1880         struct kvm *kvm = vcpu->kvm;
1881         struct hv_send_ipi_ex send_ipi_ex;
1882         struct hv_send_ipi send_ipi;
1883         u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1884         DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1885         unsigned long *vcpu_mask;
1886         unsigned long valid_bank_mask;
1887         u64 sparse_banks[64];
1888         int sparse_banks_len;
1889         u32 vector;
1890         bool all_cpus;
1891
1892         if (!ex) {
1893                 if (!hc->fast) {
1894                         if (unlikely(kvm_read_guest(kvm, hc->ingpa, &send_ipi,
1895                                                     sizeof(send_ipi))))
1896                                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1897                         sparse_banks[0] = send_ipi.cpu_mask;
1898                         vector = send_ipi.vector;
1899                 } else {
1900                         /* 'reserved' part of hv_send_ipi should be 0 */
1901                         if (unlikely(hc->ingpa >> 32 != 0))
1902                                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1903                         sparse_banks[0] = hc->outgpa;
1904                         vector = (u32)hc->ingpa;
1905                 }
1906                 all_cpus = false;
1907                 valid_bank_mask = BIT_ULL(0);
1908
1909                 trace_kvm_hv_send_ipi(vector, sparse_banks[0]);
1910         } else {
1911                 if (unlikely(kvm_read_guest(kvm, hc->ingpa, &send_ipi_ex,
1912                                             sizeof(send_ipi_ex))))
1913                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
1914
1915                 trace_kvm_hv_send_ipi_ex(send_ipi_ex.vector,
1916                                          send_ipi_ex.vp_set.format,
1917                                          send_ipi_ex.vp_set.valid_bank_mask);
1918
1919                 vector = send_ipi_ex.vector;
1920                 valid_bank_mask = send_ipi_ex.vp_set.valid_bank_mask;
1921                 sparse_banks_len = bitmap_weight(&valid_bank_mask, 64) *
1922                         sizeof(sparse_banks[0]);
1923
1924                 all_cpus = send_ipi_ex.vp_set.format == HV_GENERIC_SET_ALL;
1925
1926                 if (all_cpus)
1927                         goto check_and_send_ipi;
1928
1929                 if (!sparse_banks_len)
1930                         goto ret_success;
1931
1932                 if (kvm_read_guest(kvm,
1933                                    hc->ingpa + offsetof(struct hv_send_ipi_ex,
1934                                                         vp_set.bank_contents),
1935                                    sparse_banks,
1936                                    sparse_banks_len))
1937                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
1938         }
1939
1940 check_and_send_ipi:
1941         if ((vector < HV_IPI_LOW_VECTOR) || (vector > HV_IPI_HIGH_VECTOR))
1942                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1943
1944         vcpu_mask = all_cpus ? NULL :
1945                 sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1946                                         vp_bitmap, vcpu_bitmap);
1947
1948         kvm_send_ipi_to_many(kvm, vector, vcpu_mask);
1949
1950 ret_success:
1951         return HV_STATUS_SUCCESS;
1952 }
1953
1954 void kvm_hv_set_cpuid(struct kvm_vcpu *vcpu)
1955 {
1956         struct kvm_cpuid_entry2 *entry;
1957         struct kvm_vcpu_hv *hv_vcpu;
1958
1959         entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_INTERFACE, 0);
1960         if (entry && entry->eax == HYPERV_CPUID_SIGNATURE_EAX) {
1961                 vcpu->arch.hyperv_enabled = true;
1962         } else {
1963                 vcpu->arch.hyperv_enabled = false;
1964                 return;
1965         }
1966
1967         if (!to_hv_vcpu(vcpu) && kvm_hv_vcpu_init(vcpu))
1968                 return;
1969
1970         hv_vcpu = to_hv_vcpu(vcpu);
1971
1972         entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_FEATURES, 0);
1973         if (entry) {
1974                 hv_vcpu->cpuid_cache.features_eax = entry->eax;
1975                 hv_vcpu->cpuid_cache.features_ebx = entry->ebx;
1976                 hv_vcpu->cpuid_cache.features_edx = entry->edx;
1977         } else {
1978                 hv_vcpu->cpuid_cache.features_eax = 0;
1979                 hv_vcpu->cpuid_cache.features_ebx = 0;
1980                 hv_vcpu->cpuid_cache.features_edx = 0;
1981         }
1982
1983         entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_ENLIGHTMENT_INFO, 0);
1984         if (entry) {
1985                 hv_vcpu->cpuid_cache.enlightenments_eax = entry->eax;
1986                 hv_vcpu->cpuid_cache.enlightenments_ebx = entry->ebx;
1987         } else {
1988                 hv_vcpu->cpuid_cache.enlightenments_eax = 0;
1989                 hv_vcpu->cpuid_cache.enlightenments_ebx = 0;
1990         }
1991
1992         entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES, 0);
1993         if (entry)
1994                 hv_vcpu->cpuid_cache.syndbg_cap_eax = entry->eax;
1995         else
1996                 hv_vcpu->cpuid_cache.syndbg_cap_eax = 0;
1997 }
1998
1999 int kvm_hv_set_enforce_cpuid(struct kvm_vcpu *vcpu, bool enforce)
2000 {
2001         struct kvm_vcpu_hv *hv_vcpu;
2002         int ret = 0;
2003
2004         if (!to_hv_vcpu(vcpu)) {
2005                 if (enforce) {
2006                         ret = kvm_hv_vcpu_init(vcpu);
2007                         if (ret)
2008                                 return ret;
2009                 } else {
2010                         return 0;
2011                 }
2012         }
2013
2014         hv_vcpu = to_hv_vcpu(vcpu);
2015         hv_vcpu->enforce_cpuid = enforce;
2016
2017         return ret;
2018 }
2019
2020 bool kvm_hv_hypercall_enabled(struct kvm_vcpu *vcpu)
2021 {
2022         return vcpu->arch.hyperv_enabled && to_kvm_hv(vcpu->kvm)->hv_guest_os_id;
2023 }
2024
2025 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
2026 {
2027         bool longmode;
2028
2029         longmode = is_64_bit_hypercall(vcpu);
2030         if (longmode)
2031                 kvm_rax_write(vcpu, result);
2032         else {
2033                 kvm_rdx_write(vcpu, result >> 32);
2034                 kvm_rax_write(vcpu, result & 0xffffffff);
2035         }
2036 }
2037
2038 static int kvm_hv_hypercall_complete(struct kvm_vcpu *vcpu, u64 result)
2039 {
2040         trace_kvm_hv_hypercall_done(result);
2041         kvm_hv_hypercall_set_result(vcpu, result);
2042         ++vcpu->stat.hypercalls;
2043         return kvm_skip_emulated_instruction(vcpu);
2044 }
2045
2046 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
2047 {
2048         return kvm_hv_hypercall_complete(vcpu, vcpu->run->hyperv.u.hcall.result);
2049 }
2050
2051 static u16 kvm_hvcall_signal_event(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc)
2052 {
2053         struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
2054         struct eventfd_ctx *eventfd;
2055
2056         if (unlikely(!hc->fast)) {
2057                 int ret;
2058                 gpa_t gpa = hc->ingpa;
2059
2060                 if ((gpa & (__alignof__(hc->ingpa) - 1)) ||
2061                     offset_in_page(gpa) + sizeof(hc->ingpa) > PAGE_SIZE)
2062                         return HV_STATUS_INVALID_ALIGNMENT;
2063
2064                 ret = kvm_vcpu_read_guest(vcpu, gpa,
2065                                           &hc->ingpa, sizeof(hc->ingpa));
2066                 if (ret < 0)
2067                         return HV_STATUS_INVALID_ALIGNMENT;
2068         }
2069
2070         /*
2071          * Per spec, bits 32-47 contain the extra "flag number".  However, we
2072          * have no use for it, and in all known usecases it is zero, so just
2073          * report lookup failure if it isn't.
2074          */
2075         if (hc->ingpa & 0xffff00000000ULL)
2076                 return HV_STATUS_INVALID_PORT_ID;
2077         /* remaining bits are reserved-zero */
2078         if (hc->ingpa & ~KVM_HYPERV_CONN_ID_MASK)
2079                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
2080
2081         /* the eventfd is protected by vcpu->kvm->srcu, but conn_to_evt isn't */
2082         rcu_read_lock();
2083         eventfd = idr_find(&hv->conn_to_evt, hc->ingpa);
2084         rcu_read_unlock();
2085         if (!eventfd)
2086                 return HV_STATUS_INVALID_PORT_ID;
2087
2088         eventfd_signal(eventfd, 1);
2089         return HV_STATUS_SUCCESS;
2090 }
2091
2092 static bool is_xmm_fast_hypercall(struct kvm_hv_hcall *hc)
2093 {
2094         switch (hc->code) {
2095         case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
2096         case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
2097         case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
2098         case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
2099                 return true;
2100         }
2101
2102         return false;
2103 }
2104
2105 static void kvm_hv_hypercall_read_xmm(struct kvm_hv_hcall *hc)
2106 {
2107         int reg;
2108
2109         kvm_fpu_get();
2110         for (reg = 0; reg < HV_HYPERCALL_MAX_XMM_REGISTERS; reg++)
2111                 _kvm_read_sse_reg(reg, &hc->xmm[reg]);
2112         kvm_fpu_put();
2113 }
2114
2115 static bool hv_check_hypercall_access(struct kvm_vcpu_hv *hv_vcpu, u16 code)
2116 {
2117         if (!hv_vcpu->enforce_cpuid)
2118                 return true;
2119
2120         switch (code) {
2121         case HVCALL_NOTIFY_LONG_SPIN_WAIT:
2122                 return hv_vcpu->cpuid_cache.enlightenments_ebx &&
2123                         hv_vcpu->cpuid_cache.enlightenments_ebx != U32_MAX;
2124         case HVCALL_POST_MESSAGE:
2125                 return hv_vcpu->cpuid_cache.features_ebx & HV_POST_MESSAGES;
2126         case HVCALL_SIGNAL_EVENT:
2127                 return hv_vcpu->cpuid_cache.features_ebx & HV_SIGNAL_EVENTS;
2128         case HVCALL_POST_DEBUG_DATA:
2129         case HVCALL_RETRIEVE_DEBUG_DATA:
2130         case HVCALL_RESET_DEBUG_SESSION:
2131                 /*
2132                  * Return 'true' when SynDBG is disabled so the resulting code
2133                  * will be HV_STATUS_INVALID_HYPERCALL_CODE.
2134                  */
2135                 return !kvm_hv_is_syndbg_enabled(hv_vcpu->vcpu) ||
2136                         hv_vcpu->cpuid_cache.features_ebx & HV_DEBUGGING;
2137         case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
2138         case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
2139                 if (!(hv_vcpu->cpuid_cache.enlightenments_eax &
2140                       HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED))
2141                         return false;
2142                 fallthrough;
2143         case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
2144         case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
2145                 return hv_vcpu->cpuid_cache.enlightenments_eax &
2146                         HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
2147         case HVCALL_SEND_IPI_EX:
2148                 if (!(hv_vcpu->cpuid_cache.enlightenments_eax &
2149                       HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED))
2150                         return false;
2151                 fallthrough;
2152         case HVCALL_SEND_IPI:
2153                 return hv_vcpu->cpuid_cache.enlightenments_eax &
2154                         HV_X64_CLUSTER_IPI_RECOMMENDED;
2155         default:
2156                 break;
2157         }
2158
2159         return true;
2160 }
2161
2162 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
2163 {
2164         struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
2165         struct kvm_hv_hcall hc;
2166         u64 ret = HV_STATUS_SUCCESS;
2167
2168         /*
2169          * hypercall generates UD from non zero cpl and real mode
2170          * per HYPER-V spec
2171          */
2172         if (static_call(kvm_x86_get_cpl)(vcpu) != 0 || !is_protmode(vcpu)) {
2173                 kvm_queue_exception(vcpu, UD_VECTOR);
2174                 return 1;
2175         }
2176
2177 #ifdef CONFIG_X86_64
2178         if (is_64_bit_hypercall(vcpu)) {
2179                 hc.param = kvm_rcx_read(vcpu);
2180                 hc.ingpa = kvm_rdx_read(vcpu);
2181                 hc.outgpa = kvm_r8_read(vcpu);
2182         } else
2183 #endif
2184         {
2185                 hc.param = ((u64)kvm_rdx_read(vcpu) << 32) |
2186                             (kvm_rax_read(vcpu) & 0xffffffff);
2187                 hc.ingpa = ((u64)kvm_rbx_read(vcpu) << 32) |
2188                             (kvm_rcx_read(vcpu) & 0xffffffff);
2189                 hc.outgpa = ((u64)kvm_rdi_read(vcpu) << 32) |
2190                              (kvm_rsi_read(vcpu) & 0xffffffff);
2191         }
2192
2193         hc.code = hc.param & 0xffff;
2194         hc.fast = !!(hc.param & HV_HYPERCALL_FAST_BIT);
2195         hc.rep_cnt = (hc.param >> HV_HYPERCALL_REP_COMP_OFFSET) & 0xfff;
2196         hc.rep_idx = (hc.param >> HV_HYPERCALL_REP_START_OFFSET) & 0xfff;
2197         hc.rep = !!(hc.rep_cnt || hc.rep_idx);
2198
2199         trace_kvm_hv_hypercall(hc.code, hc.fast, hc.rep_cnt, hc.rep_idx,
2200                                hc.ingpa, hc.outgpa);
2201
2202         if (unlikely(!hv_check_hypercall_access(hv_vcpu, hc.code))) {
2203                 ret = HV_STATUS_ACCESS_DENIED;
2204                 goto hypercall_complete;
2205         }
2206
2207         if (hc.fast && is_xmm_fast_hypercall(&hc)) {
2208                 if (unlikely(hv_vcpu->enforce_cpuid &&
2209                              !(hv_vcpu->cpuid_cache.features_edx &
2210                                HV_X64_HYPERCALL_XMM_INPUT_AVAILABLE))) {
2211                         kvm_queue_exception(vcpu, UD_VECTOR);
2212                         return 1;
2213                 }
2214
2215                 kvm_hv_hypercall_read_xmm(&hc);
2216         }
2217
2218         switch (hc.code) {
2219         case HVCALL_NOTIFY_LONG_SPIN_WAIT:
2220                 if (unlikely(hc.rep)) {
2221                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2222                         break;
2223                 }
2224                 kvm_vcpu_on_spin(vcpu, true);
2225                 break;
2226         case HVCALL_SIGNAL_EVENT:
2227                 if (unlikely(hc.rep)) {
2228                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2229                         break;
2230                 }
2231                 ret = kvm_hvcall_signal_event(vcpu, &hc);
2232                 if (ret != HV_STATUS_INVALID_PORT_ID)
2233                         break;
2234                 fallthrough;    /* maybe userspace knows this conn_id */
2235         case HVCALL_POST_MESSAGE:
2236                 /* don't bother userspace if it has no way to handle it */
2237                 if (unlikely(hc.rep || !to_hv_synic(vcpu)->active)) {
2238                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2239                         break;
2240                 }
2241                 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
2242                 vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
2243                 vcpu->run->hyperv.u.hcall.input = hc.param;
2244                 vcpu->run->hyperv.u.hcall.params[0] = hc.ingpa;
2245                 vcpu->run->hyperv.u.hcall.params[1] = hc.outgpa;
2246                 vcpu->arch.complete_userspace_io =
2247                                 kvm_hv_hypercall_complete_userspace;
2248                 return 0;
2249         case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
2250                 if (unlikely(!hc.rep_cnt || hc.rep_idx)) {
2251                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2252                         break;
2253                 }
2254                 ret = kvm_hv_flush_tlb(vcpu, &hc, false);
2255                 break;
2256         case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
2257                 if (unlikely(hc.rep)) {
2258                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2259                         break;
2260                 }
2261                 ret = kvm_hv_flush_tlb(vcpu, &hc, false);
2262                 break;
2263         case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
2264                 if (unlikely(!hc.rep_cnt || hc.rep_idx)) {
2265                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2266                         break;
2267                 }
2268                 ret = kvm_hv_flush_tlb(vcpu, &hc, true);
2269                 break;
2270         case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
2271                 if (unlikely(hc.rep)) {
2272                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2273                         break;
2274                 }
2275                 ret = kvm_hv_flush_tlb(vcpu, &hc, true);
2276                 break;
2277         case HVCALL_SEND_IPI:
2278                 if (unlikely(hc.rep)) {
2279                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2280                         break;
2281                 }
2282                 ret = kvm_hv_send_ipi(vcpu, &hc, false);
2283                 break;
2284         case HVCALL_SEND_IPI_EX:
2285                 if (unlikely(hc.fast || hc.rep)) {
2286                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2287                         break;
2288                 }
2289                 ret = kvm_hv_send_ipi(vcpu, &hc, true);
2290                 break;
2291         case HVCALL_POST_DEBUG_DATA:
2292         case HVCALL_RETRIEVE_DEBUG_DATA:
2293                 if (unlikely(hc.fast)) {
2294                         ret = HV_STATUS_INVALID_PARAMETER;
2295                         break;
2296                 }
2297                 fallthrough;
2298         case HVCALL_RESET_DEBUG_SESSION: {
2299                 struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
2300
2301                 if (!kvm_hv_is_syndbg_enabled(vcpu)) {
2302                         ret = HV_STATUS_INVALID_HYPERCALL_CODE;
2303                         break;
2304                 }
2305
2306                 if (!(syndbg->options & HV_X64_SYNDBG_OPTION_USE_HCALLS)) {
2307                         ret = HV_STATUS_OPERATION_DENIED;
2308                         break;
2309                 }
2310                 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
2311                 vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
2312                 vcpu->run->hyperv.u.hcall.input = hc.param;
2313                 vcpu->run->hyperv.u.hcall.params[0] = hc.ingpa;
2314                 vcpu->run->hyperv.u.hcall.params[1] = hc.outgpa;
2315                 vcpu->arch.complete_userspace_io =
2316                                 kvm_hv_hypercall_complete_userspace;
2317                 return 0;
2318         }
2319         default:
2320                 ret = HV_STATUS_INVALID_HYPERCALL_CODE;
2321                 break;
2322         }
2323
2324 hypercall_complete:
2325         return kvm_hv_hypercall_complete(vcpu, ret);
2326 }
2327
2328 void kvm_hv_init_vm(struct kvm *kvm)
2329 {
2330         struct kvm_hv *hv = to_kvm_hv(kvm);
2331
2332         mutex_init(&hv->hv_lock);
2333         idr_init(&hv->conn_to_evt);
2334 }
2335
2336 void kvm_hv_destroy_vm(struct kvm *kvm)
2337 {
2338         struct kvm_hv *hv = to_kvm_hv(kvm);
2339         struct eventfd_ctx *eventfd;
2340         int i;
2341
2342         idr_for_each_entry(&hv->conn_to_evt, eventfd, i)
2343                 eventfd_ctx_put(eventfd);
2344         idr_destroy(&hv->conn_to_evt);
2345 }
2346
2347 static int kvm_hv_eventfd_assign(struct kvm *kvm, u32 conn_id, int fd)
2348 {
2349         struct kvm_hv *hv = to_kvm_hv(kvm);
2350         struct eventfd_ctx *eventfd;
2351         int ret;
2352
2353         eventfd = eventfd_ctx_fdget(fd);
2354         if (IS_ERR(eventfd))
2355                 return PTR_ERR(eventfd);
2356
2357         mutex_lock(&hv->hv_lock);
2358         ret = idr_alloc(&hv->conn_to_evt, eventfd, conn_id, conn_id + 1,
2359                         GFP_KERNEL_ACCOUNT);
2360         mutex_unlock(&hv->hv_lock);
2361
2362         if (ret >= 0)
2363                 return 0;
2364
2365         if (ret == -ENOSPC)
2366                 ret = -EEXIST;
2367         eventfd_ctx_put(eventfd);
2368         return ret;
2369 }
2370
2371 static int kvm_hv_eventfd_deassign(struct kvm *kvm, u32 conn_id)
2372 {
2373         struct kvm_hv *hv = to_kvm_hv(kvm);
2374         struct eventfd_ctx *eventfd;
2375
2376         mutex_lock(&hv->hv_lock);
2377         eventfd = idr_remove(&hv->conn_to_evt, conn_id);
2378         mutex_unlock(&hv->hv_lock);
2379
2380         if (!eventfd)
2381                 return -ENOENT;
2382
2383         synchronize_srcu(&kvm->srcu);
2384         eventfd_ctx_put(eventfd);
2385         return 0;
2386 }
2387
2388 int kvm_vm_ioctl_hv_eventfd(struct kvm *kvm, struct kvm_hyperv_eventfd *args)
2389 {
2390         if ((args->flags & ~KVM_HYPERV_EVENTFD_DEASSIGN) ||
2391             (args->conn_id & ~KVM_HYPERV_CONN_ID_MASK))
2392                 return -EINVAL;
2393
2394         if (args->flags == KVM_HYPERV_EVENTFD_DEASSIGN)
2395                 return kvm_hv_eventfd_deassign(kvm, args->conn_id);
2396         return kvm_hv_eventfd_assign(kvm, args->conn_id, args->fd);
2397 }
2398
2399 int kvm_get_hv_cpuid(struct kvm_vcpu *vcpu, struct kvm_cpuid2 *cpuid,
2400                      struct kvm_cpuid_entry2 __user *entries)
2401 {
2402         uint16_t evmcs_ver = 0;
2403         struct kvm_cpuid_entry2 cpuid_entries[] = {
2404                 { .function = HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS },
2405                 { .function = HYPERV_CPUID_INTERFACE },
2406                 { .function = HYPERV_CPUID_VERSION },
2407                 { .function = HYPERV_CPUID_FEATURES },
2408                 { .function = HYPERV_CPUID_ENLIGHTMENT_INFO },
2409                 { .function = HYPERV_CPUID_IMPLEMENT_LIMITS },
2410                 { .function = HYPERV_CPUID_SYNDBG_VENDOR_AND_MAX_FUNCTIONS },
2411                 { .function = HYPERV_CPUID_SYNDBG_INTERFACE },
2412                 { .function = HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES },
2413                 { .function = HYPERV_CPUID_NESTED_FEATURES },
2414         };
2415         int i, nent = ARRAY_SIZE(cpuid_entries);
2416
2417         if (kvm_x86_ops.nested_ops->get_evmcs_version)
2418                 evmcs_ver = kvm_x86_ops.nested_ops->get_evmcs_version(vcpu);
2419
2420         /* Skip NESTED_FEATURES if eVMCS is not supported */
2421         if (!evmcs_ver)
2422                 --nent;
2423
2424         if (cpuid->nent < nent)
2425                 return -E2BIG;
2426
2427         if (cpuid->nent > nent)
2428                 cpuid->nent = nent;
2429
2430         for (i = 0; i < nent; i++) {
2431                 struct kvm_cpuid_entry2 *ent = &cpuid_entries[i];
2432                 u32 signature[3];
2433
2434                 switch (ent->function) {
2435                 case HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS:
2436                         memcpy(signature, "Linux KVM Hv", 12);
2437
2438                         ent->eax = HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES;
2439                         ent->ebx = signature[0];
2440                         ent->ecx = signature[1];
2441                         ent->edx = signature[2];
2442                         break;
2443
2444                 case HYPERV_CPUID_INTERFACE:
2445                         ent->eax = HYPERV_CPUID_SIGNATURE_EAX;
2446                         break;
2447
2448                 case HYPERV_CPUID_VERSION:
2449                         /*
2450                          * We implement some Hyper-V 2016 functions so let's use
2451                          * this version.
2452                          */
2453                         ent->eax = 0x00003839;
2454                         ent->ebx = 0x000A0000;
2455                         break;
2456
2457                 case HYPERV_CPUID_FEATURES:
2458                         ent->eax |= HV_MSR_VP_RUNTIME_AVAILABLE;
2459                         ent->eax |= HV_MSR_TIME_REF_COUNT_AVAILABLE;
2460                         ent->eax |= HV_MSR_SYNIC_AVAILABLE;
2461                         ent->eax |= HV_MSR_SYNTIMER_AVAILABLE;
2462                         ent->eax |= HV_MSR_APIC_ACCESS_AVAILABLE;
2463                         ent->eax |= HV_MSR_HYPERCALL_AVAILABLE;
2464                         ent->eax |= HV_MSR_VP_INDEX_AVAILABLE;
2465                         ent->eax |= HV_MSR_RESET_AVAILABLE;
2466                         ent->eax |= HV_MSR_REFERENCE_TSC_AVAILABLE;
2467                         ent->eax |= HV_ACCESS_FREQUENCY_MSRS;
2468                         ent->eax |= HV_ACCESS_REENLIGHTENMENT;
2469
2470                         ent->ebx |= HV_POST_MESSAGES;
2471                         ent->ebx |= HV_SIGNAL_EVENTS;
2472
2473                         ent->edx |= HV_X64_HYPERCALL_XMM_INPUT_AVAILABLE;
2474                         ent->edx |= HV_FEATURE_FREQUENCY_MSRS_AVAILABLE;
2475                         ent->edx |= HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
2476
2477                         ent->ebx |= HV_DEBUGGING;
2478                         ent->edx |= HV_X64_GUEST_DEBUGGING_AVAILABLE;
2479                         ent->edx |= HV_FEATURE_DEBUG_MSRS_AVAILABLE;
2480
2481                         /*
2482                          * Direct Synthetic timers only make sense with in-kernel
2483                          * LAPIC
2484                          */
2485                         if (!vcpu || lapic_in_kernel(vcpu))
2486                                 ent->edx |= HV_STIMER_DIRECT_MODE_AVAILABLE;
2487
2488                         break;
2489
2490                 case HYPERV_CPUID_ENLIGHTMENT_INFO:
2491                         ent->eax |= HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
2492                         ent->eax |= HV_X64_APIC_ACCESS_RECOMMENDED;
2493                         ent->eax |= HV_X64_RELAXED_TIMING_RECOMMENDED;
2494                         ent->eax |= HV_X64_CLUSTER_IPI_RECOMMENDED;
2495                         ent->eax |= HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED;
2496                         if (evmcs_ver)
2497                                 ent->eax |= HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
2498                         if (!cpu_smt_possible())
2499                                 ent->eax |= HV_X64_NO_NONARCH_CORESHARING;
2500
2501                         ent->eax |= HV_DEPRECATING_AEOI_RECOMMENDED;
2502                         /*
2503                          * Default number of spinlock retry attempts, matches
2504                          * HyperV 2016.
2505                          */
2506                         ent->ebx = 0x00000FFF;
2507
2508                         break;
2509
2510                 case HYPERV_CPUID_IMPLEMENT_LIMITS:
2511                         /* Maximum number of virtual processors */
2512                         ent->eax = KVM_MAX_VCPUS;
2513                         /*
2514                          * Maximum number of logical processors, matches
2515                          * HyperV 2016.
2516                          */
2517                         ent->ebx = 64;
2518
2519                         break;
2520
2521                 case HYPERV_CPUID_NESTED_FEATURES:
2522                         ent->eax = evmcs_ver;
2523                         if (evmcs_ver)
2524                                 ent->eax |= HV_X64_NESTED_MSR_BITMAP;
2525
2526                         break;
2527
2528                 case HYPERV_CPUID_SYNDBG_VENDOR_AND_MAX_FUNCTIONS:
2529                         memcpy(signature, "Linux KVM Hv", 12);
2530
2531                         ent->eax = 0;
2532                         ent->ebx = signature[0];
2533                         ent->ecx = signature[1];
2534                         ent->edx = signature[2];
2535                         break;
2536
2537                 case HYPERV_CPUID_SYNDBG_INTERFACE:
2538                         memcpy(signature, "VS#1\0\0\0\0\0\0\0\0", 12);
2539                         ent->eax = signature[0];
2540                         break;
2541
2542                 case HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES:
2543                         ent->eax |= HV_X64_SYNDBG_CAP_ALLOW_KERNEL_DEBUGGING;
2544                         break;
2545
2546                 default:
2547                         break;
2548                 }
2549         }
2550
2551         if (copy_to_user(entries, cpuid_entries,
2552                          nent * sizeof(struct kvm_cpuid_entry2)))
2553                 return -EFAULT;
2554
2555         return 0;
2556 }