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
  40 /* "Hv#1" signature */
  41 #define HYPERV_CPUID_SIGNATURE_EAX 0x31237648
  42
  43 #define KVM_HV_MAX_SPARSE_VCPU_SET_BITS DIV_ROUND_UP(KVM_MAX_VCPUS, 64)
  44
  45 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
  46                                 bool vcpu_kick);
  47
  48 static inline u64 synic_read_sint(struct kvm_vcpu_hv_synic *synic, int sint)
  49 {
  50         return atomic64_read(&synic->sint[sint]);
  51 }
  52
  53 static inline int synic_get_sint_vector(u64 sint_value)
  54 {
  55         if (sint_value & HV_SYNIC_SINT_MASKED)
  56                 return -1;
  57         return sint_value & HV_SYNIC_SINT_VECTOR_MASK;
  58 }
  59
  60 static bool synic_has_vector_connected(struct kvm_vcpu_hv_synic *synic,
  61                                       int vector)
  62 {
  63         int i;
  64
  65         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
  66                 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
  67                         return true;
  68         }
  69         return false;
  70 }
  71
  72 static bool synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic *synic,
  73                                      int vector)
  74 {
  75         int i;
  76         u64 sint_value;
  77
  78         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
  79                 sint_value = synic_read_sint(synic, i);
  80                 if (synic_get_sint_vector(sint_value) == vector &&
  81                     sint_value & HV_SYNIC_SINT_AUTO_EOI)
  82                         return true;
  83         }
  84         return false;
  85 }
  86
  87 static void synic_update_vector(struct kvm_vcpu_hv_synic *synic,
  88                                 int vector)
  89 {
  90         if (vector < HV_SYNIC_FIRST_VALID_VECTOR)
  91                 return;
  92
  93         if (synic_has_vector_connected(synic, vector))
  94                 __set_bit(vector, synic->vec_bitmap);
  95         else
  96                 __clear_bit(vector, synic->vec_bitmap);
  97
  98         if (synic_has_vector_auto_eoi(synic, vector))
  99                 __set_bit(vector, synic->auto_eoi_bitmap);
 100         else
 101                 __clear_bit(vector, synic->auto_eoi_bitmap);
 102 }
 103
 104 static int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint,
 105                           u64 data, bool host)
 106 {
 107         int vector, old_vector;
 108         bool masked;
 109
 110         vector = data & HV_SYNIC_SINT_VECTOR_MASK;
 111         masked = data & HV_SYNIC_SINT_MASKED;
 112
 113         /*
 114          * Valid vectors are 16-255, however, nested Hyper-V attempts to write
 115          * default '0x10000' value on boot and this should not #GP. We need to
 116          * allow zero-initing the register from host as well.
 117          */
 118         if (vector < HV_SYNIC_FIRST_VALID_VECTOR && !host && !masked)
 119                 return 1;
 120         /*
 121          * Guest may configure multiple SINTs to use the same vector, so
 122          * we maintain a bitmap of vectors handled by synic, and a
 123          * bitmap of vectors with auto-eoi behavior.  The bitmaps are
 124          * updated here, and atomically queried on fast paths.
 125          */
 126         old_vector = synic_read_sint(synic, sint) & HV_SYNIC_SINT_VECTOR_MASK;
 127
 128         atomic64_set(&synic->sint[sint], data);
 129
 130         synic_update_vector(synic, old_vector);
 131
 132         synic_update_vector(synic, vector);
 133
 134         /* Load SynIC vectors into EOI exit bitmap */
 135         kvm_make_request(KVM_REQ_SCAN_IOAPIC, hv_synic_to_vcpu(synic));
 136         return 0;
 137 }
 138
 139 static struct kvm_vcpu *get_vcpu_by_vpidx(struct kvm *kvm, u32 vpidx)
 140 {
 141         struct kvm_vcpu *vcpu = NULL;
 142         int i;
 143
 144         if (vpidx >= KVM_MAX_VCPUS)
 145                 return NULL;
 146
 147         vcpu = kvm_get_vcpu(kvm, vpidx);
 148         if (vcpu && kvm_hv_get_vpindex(vcpu) == vpidx)
 149                 return vcpu;
 150         kvm_for_each_vcpu(i, vcpu, kvm)
 151                 if (kvm_hv_get_vpindex(vcpu) == vpidx)
 152                         return vcpu;
 153         return NULL;
 154 }
 155
 156 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vpidx)
 157 {
 158         struct kvm_vcpu *vcpu;
 159         struct kvm_vcpu_hv_synic *synic;
 160
 161         vcpu = get_vcpu_by_vpidx(kvm, vpidx);
 162         if (!vcpu || !to_hv_vcpu(vcpu))
 163                 return NULL;
 164         synic = to_hv_synic(vcpu);
 165         return (synic->active) ? synic : NULL;
 166 }
 167
 168 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
 169 {
 170         struct kvm *kvm = vcpu->kvm;
 171         struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
 172         struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
 173         struct kvm_vcpu_hv_stimer *stimer;
 174         int gsi, idx;
 175
 176         trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
 177
 178         /* Try to deliver pending Hyper-V SynIC timers messages */
 179         for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
 180                 stimer = &hv_vcpu->stimer[idx];
 181                 if (stimer->msg_pending && stimer->config.enable &&
 182                     !stimer->config.direct_mode &&
 183                     stimer->config.sintx == sint)
 184                         stimer_mark_pending(stimer, false);
 185         }
 186
 187         idx = srcu_read_lock(&kvm->irq_srcu);
 188         gsi = atomic_read(&synic->sint_to_gsi[sint]);
 189         if (gsi != -1)
 190                 kvm_notify_acked_gsi(kvm, gsi);
 191         srcu_read_unlock(&kvm->irq_srcu, idx);
 192 }
 193
 194 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
 195 {
 196         struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
 197         struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
 198
 199         hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
 200         hv_vcpu->exit.u.synic.msr = msr;
 201         hv_vcpu->exit.u.synic.control = synic->control;
 202         hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
 203         hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
 204
 205         kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
 206 }
 207
 208 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
 209                          u32 msr, u64 data, bool host)
 210 {
 211         struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
 212         int ret;
 213
 214         if (!synic->active && !host)
 215                 return 1;
 216
 217         trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
 218
 219         ret = 0;
 220         switch (msr) {
 221         case HV_X64_MSR_SCONTROL:
 222                 synic->control = data;
 223                 if (!host)
 224                         synic_exit(synic, msr);
 225                 break;
 226         case HV_X64_MSR_SVERSION:
 227                 if (!host) {
 228                         ret = 1;
 229                         break;
 230                 }
 231                 synic->version = data;
 232                 break;
 233         case HV_X64_MSR_SIEFP:
 234                 if ((data & HV_SYNIC_SIEFP_ENABLE) && !host &&
 235                     !synic->dont_zero_synic_pages)
 236                         if (kvm_clear_guest(vcpu->kvm,
 237                                             data & PAGE_MASK, PAGE_SIZE)) {
 238                                 ret = 1;
 239                                 break;
 240                         }
 241                 synic->evt_page = data;
 242                 if (!host)
 243                         synic_exit(synic, msr);
 244                 break;
 245         case HV_X64_MSR_SIMP:
 246                 if ((data & HV_SYNIC_SIMP_ENABLE) && !host &&
 247                     !synic->dont_zero_synic_pages)
 248                         if (kvm_clear_guest(vcpu->kvm,
 249                                             data & PAGE_MASK, PAGE_SIZE)) {
 250                                 ret = 1;
 251                                 break;
 252                         }
 253                 synic->msg_page = data;
 254                 if (!host)
 255                         synic_exit(synic, msr);
 256                 break;
 257         case HV_X64_MSR_EOM: {
 258                 int i;
 259
 260                 for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
 261                         kvm_hv_notify_acked_sint(vcpu, i);
 262                 break;
 263         }
 264         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
 265                 ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
 266                 break;
 267         default:
 268                 ret = 1;
 269                 break;
 270         }
 271         return ret;
 272 }
 273
 274 static bool kvm_hv_is_syndbg_enabled(struct kvm_vcpu *vcpu)
 275 {
 276         struct kvm_cpuid_entry2 *entry;
 277
 278         entry = kvm_find_cpuid_entry(vcpu,
 279                                      HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES,
 280                                      0);
 281         if (!entry)
 282                 return false;
 283
 284         return entry->eax & HV_X64_SYNDBG_CAP_ALLOW_KERNEL_DEBUGGING;
 285 }
 286
 287 static int kvm_hv_syndbg_complete_userspace(struct kvm_vcpu *vcpu)
 288 {
 289         struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
 290
 291         if (vcpu->run->hyperv.u.syndbg.msr == HV_X64_MSR_SYNDBG_CONTROL)
 292                 hv->hv_syndbg.control.status =
 293                         vcpu->run->hyperv.u.syndbg.status;
 294         return 1;
 295 }
 296
 297 static void syndbg_exit(struct kvm_vcpu *vcpu, u32 msr)
 298 {
 299         struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
 300         struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
 301
 302         hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNDBG;
 303         hv_vcpu->exit.u.syndbg.msr = msr;
 304         hv_vcpu->exit.u.syndbg.control = syndbg->control.control;
 305         hv_vcpu->exit.u.syndbg.send_page = syndbg->control.send_page;
 306         hv_vcpu->exit.u.syndbg.recv_page = syndbg->control.recv_page;
 307         hv_vcpu->exit.u.syndbg.pending_page = syndbg->control.pending_page;
 308         vcpu->arch.complete_userspace_io =
 309                         kvm_hv_syndbg_complete_userspace;
 310
 311         kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
 312 }
 313
 314 static int syndbg_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
 315 {
 316         struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
 317
 318         if (!kvm_hv_is_syndbg_enabled(vcpu) && !host)
 319                 return 1;
 320
 321         trace_kvm_hv_syndbg_set_msr(vcpu->vcpu_id,
 322                                     to_hv_vcpu(vcpu)->vp_index, msr, data);
 323         switch (msr) {
 324         case HV_X64_MSR_SYNDBG_CONTROL:
 325                 syndbg->control.control = data;
 326                 if (!host)
 327                         syndbg_exit(vcpu, msr);
 328                 break;
 329         case HV_X64_MSR_SYNDBG_STATUS:
 330                 syndbg->control.status = data;
 331                 break;
 332         case HV_X64_MSR_SYNDBG_SEND_BUFFER:
 333                 syndbg->control.send_page = data;
 334                 break;
 335         case HV_X64_MSR_SYNDBG_RECV_BUFFER:
 336                 syndbg->control.recv_page = data;
 337                 break;
 338         case HV_X64_MSR_SYNDBG_PENDING_BUFFER:
 339                 syndbg->control.pending_page = data;
 340                 if (!host)
 341                         syndbg_exit(vcpu, msr);
 342                 break;
 343         case HV_X64_MSR_SYNDBG_OPTIONS:
 344                 syndbg->options = data;
 345                 break;
 346         default:
 347                 break;
 348         }
 349
 350         return 0;
 351 }
 352
 353 static int syndbg_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
 354 {
 355         struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
 356
 357         if (!kvm_hv_is_syndbg_enabled(vcpu) && !host)
 358                 return 1;
 359
 360         switch (msr) {
 361         case HV_X64_MSR_SYNDBG_CONTROL:
 362                 *pdata = syndbg->control.control;
 363                 break;
 364         case HV_X64_MSR_SYNDBG_STATUS:
 365                 *pdata = syndbg->control.status;
 366                 break;
 367         case HV_X64_MSR_SYNDBG_SEND_BUFFER:
 368                 *pdata = syndbg->control.send_page;
 369                 break;
 370         case HV_X64_MSR_SYNDBG_RECV_BUFFER:
 371                 *pdata = syndbg->control.recv_page;
 372                 break;
 373         case HV_X64_MSR_SYNDBG_PENDING_BUFFER:
 374                 *pdata = syndbg->control.pending_page;
 375                 break;
 376         case HV_X64_MSR_SYNDBG_OPTIONS:
 377                 *pdata = syndbg->options;
 378                 break;
 379         default:
 380                 break;
 381         }
 382
 383         trace_kvm_hv_syndbg_get_msr(vcpu->vcpu_id, kvm_hv_get_vpindex(vcpu), msr, *pdata);
 384
 385         return 0;
 386 }
 387
 388 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata,
 389                          bool host)
 390 {
 391         int ret;
 392
 393         if (!synic->active && !host)
 394                 return 1;
 395
 396         ret = 0;
 397         switch (msr) {
 398         case HV_X64_MSR_SCONTROL:
 399                 *pdata = synic->control;
 400                 break;
 401         case HV_X64_MSR_SVERSION:
 402                 *pdata = synic->version;
 403                 break;
 404         case HV_X64_MSR_SIEFP:
 405                 *pdata = synic->evt_page;
 406                 break;
 407         case HV_X64_MSR_SIMP:
 408                 *pdata = synic->msg_page;
 409                 break;
 410         case HV_X64_MSR_EOM:
 411                 *pdata = 0;
 412                 break;
 413         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
 414                 *pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
 415                 break;
 416         default:
 417                 ret = 1;
 418                 break;
 419         }
 420         return ret;
 421 }
 422
 423 static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
 424 {
 425         struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
 426         struct kvm_lapic_irq irq;
 427         int ret, vector;
 428
 429         if (sint >= ARRAY_SIZE(synic->sint))
 430                 return -EINVAL;
 431
 432         vector = synic_get_sint_vector(synic_read_sint(synic, sint));
 433         if (vector < 0)
 434                 return -ENOENT;
 435
 436         memset(&irq, 0, sizeof(irq));
 437         irq.shorthand = APIC_DEST_SELF;
 438         irq.dest_mode = APIC_DEST_PHYSICAL;
 439         irq.delivery_mode = APIC_DM_FIXED;
 440         irq.vector = vector;
 441         irq.level = 1;
 442
 443         ret = kvm_irq_delivery_to_apic(vcpu->kvm, vcpu->arch.apic, &irq, NULL);
 444         trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
 445         return ret;
 446 }
 447
 448 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vpidx, u32 sint)
 449 {
 450         struct kvm_vcpu_hv_synic *synic;
 451
 452         synic = synic_get(kvm, vpidx);
 453         if (!synic)
 454                 return -EINVAL;
 455
 456         return synic_set_irq(synic, sint);
 457 }
 458
 459 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
 460 {
 461         struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
 462         int i;
 463
 464         trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
 465
 466         for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
 467                 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
 468                         kvm_hv_notify_acked_sint(vcpu, i);
 469 }
 470
 471 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi)
 472 {
 473         struct kvm_vcpu_hv_synic *synic;
 474
 475         synic = synic_get(kvm, vpidx);
 476         if (!synic)
 477                 return -EINVAL;
 478
 479         if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
 480                 return -EINVAL;
 481
 482         atomic_set(&synic->sint_to_gsi[sint], gsi);
 483         return 0;
 484 }
 485
 486 void kvm_hv_irq_routing_update(struct kvm *kvm)
 487 {
 488         struct kvm_irq_routing_table *irq_rt;
 489         struct kvm_kernel_irq_routing_entry *e;
 490         u32 gsi;
 491
 492         irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
 493                                         lockdep_is_held(&kvm->irq_lock));
 494
 495         for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
 496                 hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
 497                         if (e->type == KVM_IRQ_ROUTING_HV_SINT)
 498                                 kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
 499                                                     e->hv_sint.sint, gsi);
 500                 }
 501         }
 502 }
 503
 504 static void synic_init(struct kvm_vcpu_hv_synic *synic)
 505 {
 506         int i;
 507
 508         memset(synic, 0, sizeof(*synic));
 509         synic->version = HV_SYNIC_VERSION_1;
 510         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
 511                 atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
 512                 atomic_set(&synic->sint_to_gsi[i], -1);
 513         }
 514 }
 515
 516 static u64 get_time_ref_counter(struct kvm *kvm)
 517 {
 518         struct kvm_hv *hv = to_kvm_hv(kvm);
 519         struct kvm_vcpu *vcpu;
 520         u64 tsc;
 521
 522         /*
 523          * The guest has not set up the TSC page or the clock isn't
 524          * stable, fall back to get_kvmclock_ns.
 525          */
 526         if (!hv->tsc_ref.tsc_sequence)
 527                 return div_u64(get_kvmclock_ns(kvm), 100);
 528
 529         vcpu = kvm_get_vcpu(kvm, 0);
 530         tsc = kvm_read_l1_tsc(vcpu, rdtsc());
 531         return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64)
 532                 + hv->tsc_ref.tsc_offset;
 533 }
 534
 535 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
 536                                 bool vcpu_kick)
 537 {
 538         struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
 539
 540         set_bit(stimer->index,
 541                 to_hv_vcpu(vcpu)->stimer_pending_bitmap);
 542         kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
 543         if (vcpu_kick)
 544                 kvm_vcpu_kick(vcpu);
 545 }
 546
 547 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
 548 {
 549         struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
 550
 551         trace_kvm_hv_stimer_cleanup(hv_stimer_to_vcpu(stimer)->vcpu_id,
 552                                     stimer->index);
 553
 554         hrtimer_cancel(&stimer->timer);
 555         clear_bit(stimer->index,
 556                   to_hv_vcpu(vcpu)->stimer_pending_bitmap);
 557         stimer->msg_pending = false;
 558         stimer->exp_time = 0;
 559 }
 560
 561 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
 562 {
 563         struct kvm_vcpu_hv_stimer *stimer;
 564
 565         stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
 566         trace_kvm_hv_stimer_callback(hv_stimer_to_vcpu(stimer)->vcpu_id,
 567                                      stimer->index);
 568         stimer_mark_pending(stimer, true);
 569
 570         return HRTIMER_NORESTART;
 571 }
 572
 573 /*
 574  * stimer_start() assumptions:
 575  * a) stimer->count is not equal to 0
 576  * b) stimer->config has HV_STIMER_ENABLE flag
 577  */
 578 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
 579 {
 580         u64 time_now;
 581         ktime_t ktime_now;
 582
 583         time_now = get_time_ref_counter(hv_stimer_to_vcpu(stimer)->kvm);
 584         ktime_now = ktime_get();
 585
 586         if (stimer->config.periodic) {
 587                 if (stimer->exp_time) {
 588                         if (time_now >= stimer->exp_time) {
 589                                 u64 remainder;
 590
 591                                 div64_u64_rem(time_now - stimer->exp_time,
 592                                               stimer->count, &remainder);
 593                                 stimer->exp_time =
 594                                         time_now + (stimer->count - remainder);
 595                         }
 596                 } else
 597                         stimer->exp_time = time_now + stimer->count;
 598
 599                 trace_kvm_hv_stimer_start_periodic(
 600                                         hv_stimer_to_vcpu(stimer)->vcpu_id,
 601                                         stimer->index,
 602                                         time_now, stimer->exp_time);
 603
 604                 hrtimer_start(&stimer->timer,
 605                               ktime_add_ns(ktime_now,
 606                                            100 * (stimer->exp_time - time_now)),
 607                               HRTIMER_MODE_ABS);
 608                 return 0;
 609         }
 610         stimer->exp_time = stimer->count;
 611         if (time_now >= stimer->count) {
 612                 /*
 613                  * Expire timer according to Hypervisor Top-Level Functional
 614                  * specification v4(15.3.1):
 615                  * "If a one shot is enabled and the specified count is in
 616                  * the past, it will expire immediately."
 617                  */
 618                 stimer_mark_pending(stimer, false);
 619                 return 0;
 620         }
 621
 622         trace_kvm_hv_stimer_start_one_shot(hv_stimer_to_vcpu(stimer)->vcpu_id,
 623                                            stimer->index,
 624                                            time_now, stimer->count);
 625
 626         hrtimer_start(&stimer->timer,
 627                       ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)),
 628                       HRTIMER_MODE_ABS);
 629         return 0;
 630 }
 631
 632 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
 633                              bool host)
 634 {
 635         union hv_stimer_config new_config = {.as_uint64 = config},
 636                 old_config = {.as_uint64 = stimer->config.as_uint64};
 637         struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
 638         struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
 639
 640         if (!synic->active && !host)
 641                 return 1;
 642
 643         trace_kvm_hv_stimer_set_config(hv_stimer_to_vcpu(stimer)->vcpu_id,
 644                                        stimer->index, config, host);
 645
 646         stimer_cleanup(stimer);
 647         if (old_config.enable &&
 648             !new_config.direct_mode && new_config.sintx == 0)
 649                 new_config.enable = 0;
 650         stimer->config.as_uint64 = new_config.as_uint64;
 651
 652         if (stimer->config.enable)
 653                 stimer_mark_pending(stimer, false);
 654
 655         return 0;
 656 }
 657
 658 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
 659                             bool host)
 660 {
 661         struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
 662         struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
 663
 664         if (!synic->active && !host)
 665                 return 1;
 666
 667         trace_kvm_hv_stimer_set_count(hv_stimer_to_vcpu(stimer)->vcpu_id,
 668                                       stimer->index, count, host);
 669
 670         stimer_cleanup(stimer);
 671         stimer->count = count;
 672         if (stimer->count == 0)
 673                 stimer->config.enable = 0;
 674         else if (stimer->config.auto_enable)
 675                 stimer->config.enable = 1;
 676
 677         if (stimer->config.enable)
 678                 stimer_mark_pending(stimer, false);
 679
 680         return 0;
 681 }
 682
 683 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
 684 {
 685         *pconfig = stimer->config.as_uint64;
 686         return 0;
 687 }
 688
 689 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
 690 {
 691         *pcount = stimer->count;
 692         return 0;
 693 }
 694
 695 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
 696                              struct hv_message *src_msg, bool no_retry)
 697 {
 698         struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
 699         int msg_off = offsetof(struct hv_message_page, sint_message[sint]);
 700         gfn_t msg_page_gfn;
 701         struct hv_message_header hv_hdr;
 702         int r;
 703
 704         if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
 705                 return -ENOENT;
 706
 707         msg_page_gfn = synic->msg_page >> PAGE_SHIFT;
 708
 709         /*
 710          * Strictly following the spec-mandated ordering would assume setting
 711          * .msg_pending before checking .message_type.  However, this function
 712          * is only called in vcpu context so the entire update is atomic from
 713          * guest POV and thus the exact order here doesn't matter.
 714          */
 715         r = kvm_vcpu_read_guest_page(vcpu, msg_page_gfn, &hv_hdr.message_type,
 716                                      msg_off + offsetof(struct hv_message,
 717                                                         header.message_type),
 718                                      sizeof(hv_hdr.message_type));
 719         if (r < 0)
 720                 return r;
 721
 722         if (hv_hdr.message_type != HVMSG_NONE) {
 723                 if (no_retry)
 724                         return 0;
 725
 726                 hv_hdr.message_flags.msg_pending = 1;
 727                 r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn,
 728                                               &hv_hdr.message_flags,
 729                                               msg_off +
 730                                               offsetof(struct hv_message,
 731                                                        header.message_flags),
 732                                               sizeof(hv_hdr.message_flags));
 733                 if (r < 0)
 734                         return r;
 735                 return -EAGAIN;
 736         }
 737
 738         r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn, src_msg, msg_off,
 739                                       sizeof(src_msg->header) +
 740                                       src_msg->header.payload_size);
 741         if (r < 0)
 742                 return r;
 743
 744         r = synic_set_irq(synic, sint);
 745         if (r < 0)
 746                 return r;
 747         if (r == 0)
 748                 return -EFAULT;
 749         return 0;
 750 }
 751
 752 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
 753 {
 754         struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
 755         struct hv_message *msg = &stimer->msg;
 756         struct hv_timer_message_payload *payload =
 757                         (struct hv_timer_message_payload *)&msg->u.payload;
 758
 759         /*
 760          * To avoid piling up periodic ticks, don't retry message
 761          * delivery for them (within "lazy" lost ticks policy).
 762          */
 763         bool no_retry = stimer->config.periodic;
 764
 765         payload->expiration_time = stimer->exp_time;
 766         payload->delivery_time = get_time_ref_counter(vcpu->kvm);
 767         return synic_deliver_msg(to_hv_synic(vcpu),
 768                                  stimer->config.sintx, msg,
 769                                  no_retry);
 770 }
 771
 772 static int stimer_notify_direct(struct kvm_vcpu_hv_stimer *stimer)
 773 {
 774         struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
 775         struct kvm_lapic_irq irq = {
 776                 .delivery_mode = APIC_DM_FIXED,
 777                 .vector = stimer->config.apic_vector
 778         };
 779
 780         if (lapic_in_kernel(vcpu))
 781                 return !kvm_apic_set_irq(vcpu, &irq, NULL);
 782         return 0;
 783 }
 784
 785 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
 786 {
 787         int r, direct = stimer->config.direct_mode;
 788
 789         stimer->msg_pending = true;
 790         if (!direct)
 791                 r = stimer_send_msg(stimer);
 792         else
 793                 r = stimer_notify_direct(stimer);
 794         trace_kvm_hv_stimer_expiration(hv_stimer_to_vcpu(stimer)->vcpu_id,
 795                                        stimer->index, direct, r);
 796         if (!r) {
 797                 stimer->msg_pending = false;
 798                 if (!(stimer->config.periodic))
 799                         stimer->config.enable = 0;
 800         }
 801 }
 802
 803 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
 804 {
 805         struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
 806         struct kvm_vcpu_hv_stimer *stimer;
 807         u64 time_now, exp_time;
 808         int i;
 809
 810         if (!hv_vcpu)
 811                 return;
 812
 813         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 814                 if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
 815                         stimer = &hv_vcpu->stimer[i];
 816                         if (stimer->config.enable) {
 817                                 exp_time = stimer->exp_time;
 818
 819                                 if (exp_time) {
 820                                         time_now =
 821                                                 get_time_ref_counter(vcpu->kvm);
 822                                         if (time_now >= exp_time)
 823                                                 stimer_expiration(stimer);
 824                                 }
 825
 826                                 if ((stimer->config.enable) &&
 827                                     stimer->count) {
 828                                         if (!stimer->msg_pending)
 829                                                 stimer_start(stimer);
 830                                 } else
 831                                         stimer_cleanup(stimer);
 832                         }
 833                 }
 834 }
 835
 836 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
 837 {
 838         struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
 839         int i;
 840
 841         if (!hv_vcpu)
 842                 return;
 843
 844         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 845                 stimer_cleanup(&hv_vcpu->stimer[i]);
 846
 847         kfree(hv_vcpu);
 848         vcpu->arch.hyperv = NULL;
 849 }
 850
 851 bool kvm_hv_assist_page_enabled(struct kvm_vcpu *vcpu)
 852 {
 853         struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
 854
 855         if (!hv_vcpu)
 856                 return false;
 857
 858         if (!(hv_vcpu->hv_vapic & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE))
 859                 return false;
 860         return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
 861 }
 862 EXPORT_SYMBOL_GPL(kvm_hv_assist_page_enabled);
 863
 864 bool kvm_hv_get_assist_page(struct kvm_vcpu *vcpu,
 865                             struct hv_vp_assist_page *assist_page)
 866 {
 867         if (!kvm_hv_assist_page_enabled(vcpu))
 868                 return false;
 869         return !kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data,
 870                                       assist_page, sizeof(*assist_page));
 871 }
 872 EXPORT_SYMBOL_GPL(kvm_hv_get_assist_page);
 873
 874 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
 875 {
 876         struct hv_message *msg = &stimer->msg;
 877         struct hv_timer_message_payload *payload =
 878                         (struct hv_timer_message_payload *)&msg->u.payload;
 879
 880         memset(&msg->header, 0, sizeof(msg->header));
 881         msg->header.message_type = HVMSG_TIMER_EXPIRED;
 882         msg->header.payload_size = sizeof(*payload);
 883
 884         payload->timer_index = stimer->index;
 885         payload->expiration_time = 0;
 886         payload->delivery_time = 0;
 887 }
 888
 889 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
 890 {
 891         memset(stimer, 0, sizeof(*stimer));
 892         stimer->index = timer_index;
 893         hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
 894         stimer->timer.function = stimer_timer_callback;
 895         stimer_prepare_msg(stimer);
 896 }
 897
 898 static int kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
 899 {
 900         struct kvm_vcpu_hv *hv_vcpu;
 901         int i;
 902
 903         hv_vcpu = kzalloc(sizeof(struct kvm_vcpu_hv), GFP_KERNEL_ACCOUNT);
 904         if (!hv_vcpu)
 905                 return -ENOMEM;
 906
 907         vcpu->arch.hyperv = hv_vcpu;
 908         hv_vcpu->vcpu = vcpu;
 909
 910         synic_init(&hv_vcpu->synic);
 911
 912         bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
 913         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 914                 stimer_init(&hv_vcpu->stimer[i], i);
 915
 916         hv_vcpu->vp_index = kvm_vcpu_get_idx(vcpu);
 917
 918         return 0;
 919 }
 920
 921 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
 922 {
 923         struct kvm_vcpu_hv_synic *synic;
 924         int r;
 925
 926         if (!to_hv_vcpu(vcpu)) {
 927                 r = kvm_hv_vcpu_init(vcpu);
 928                 if (r)
 929                         return r;
 930         }
 931
 932         synic = to_hv_synic(vcpu);
 933
 934         /*
 935          * Hyper-V SynIC auto EOI SINT's are
 936          * not compatible with APICV, so request
 937          * to deactivate APICV permanently.
 938          */
 939         kvm_request_apicv_update(vcpu->kvm, false, APICV_INHIBIT_REASON_HYPERV);
 940         synic->active = true;
 941         synic->dont_zero_synic_pages = dont_zero_synic_pages;
 942         synic->control = HV_SYNIC_CONTROL_ENABLE;
 943         return 0;
 944 }
 945
 946 static bool kvm_hv_msr_partition_wide(u32 msr)
 947 {
 948         bool r = false;
 949
 950         switch (msr) {
 951         case HV_X64_MSR_GUEST_OS_ID:
 952         case HV_X64_MSR_HYPERCALL:
 953         case HV_X64_MSR_REFERENCE_TSC:
 954         case HV_X64_MSR_TIME_REF_COUNT:
 955         case HV_X64_MSR_CRASH_CTL:
 956         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
 957         case HV_X64_MSR_RESET:
 958         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
 959         case HV_X64_MSR_TSC_EMULATION_CONTROL:
 960         case HV_X64_MSR_TSC_EMULATION_STATUS:
 961         case HV_X64_MSR_SYNDBG_OPTIONS:
 962         case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
 963                 r = true;
 964                 break;
 965         }
 966
 967         return r;
 968 }
 969
 970 static int kvm_hv_msr_get_crash_data(struct kvm *kvm, u32 index, u64 *pdata)
 971 {
 972         struct kvm_hv *hv = to_kvm_hv(kvm);
 973         size_t size = ARRAY_SIZE(hv->hv_crash_param);
 974
 975         if (WARN_ON_ONCE(index >= size))
 976                 return -EINVAL;
 977
 978         *pdata = hv->hv_crash_param[array_index_nospec(index, size)];
 979         return 0;
 980 }
 981
 982 static int kvm_hv_msr_get_crash_ctl(struct kvm *kvm, u64 *pdata)
 983 {
 984         struct kvm_hv *hv = to_kvm_hv(kvm);
 985
 986         *pdata = hv->hv_crash_ctl;
 987         return 0;
 988 }
 989
 990 static int kvm_hv_msr_set_crash_ctl(struct kvm *kvm, u64 data)
 991 {
 992         struct kvm_hv *hv = to_kvm_hv(kvm);
 993
 994         hv->hv_crash_ctl = data & HV_CRASH_CTL_CRASH_NOTIFY;
 995
 996         return 0;
 997 }
 998
 999 static int kvm_hv_msr_set_crash_data(struct kvm *kvm, u32 index, u64 data)
1000 {
1001         struct kvm_hv *hv = to_kvm_hv(kvm);
1002         size_t size = ARRAY_SIZE(hv->hv_crash_param);
1003
1004         if (WARN_ON_ONCE(index >= size))
1005                 return -EINVAL;
1006
1007         hv->hv_crash_param[array_index_nospec(index, size)] = data;
1008         return 0;
1009 }
1010
1011 /*
1012  * The kvmclock and Hyper-V TSC page use similar formulas, and converting
1013  * between them is possible:
1014  *
1015  * kvmclock formula:
1016  *    nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
1017  *           + system_time
1018  *
1019  * Hyper-V formula:
1020  *    nsec/100 = ticks * scale / 2^64 + offset
1021  *
1022  * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
1023  * By dividing the kvmclock formula by 100 and equating what's left we get:
1024  *    ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1025  *            scale / 2^64 =         tsc_to_system_mul * 2^(tsc_shift-32) / 100
1026  *            scale        =         tsc_to_system_mul * 2^(32+tsc_shift) / 100
1027  *
1028  * Now expand the kvmclock formula and divide by 100:
1029  *    nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
1030  *           - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
1031  *           + system_time
1032  *    nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1033  *               - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1034  *               + system_time / 100
1035  *
1036  * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
1037  *    nsec/100 = ticks * scale / 2^64
1038  *               - tsc_timestamp * scale / 2^64
1039  *               + system_time / 100
1040  *
1041  * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
1042  *    offset = system_time / 100 - tsc_timestamp * scale / 2^64
1043  *
1044  * These two equivalencies are implemented in this function.
1045  */
1046 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
1047                                         struct ms_hyperv_tsc_page *tsc_ref)
1048 {
1049         u64 max_mul;
1050
1051         if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
1052                 return false;
1053
1054         /*
1055          * check if scale would overflow, if so we use the time ref counter
1056          *    tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
1057          *    tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
1058          *    tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
1059          */
1060         max_mul = 100ull << (32 - hv_clock->tsc_shift);
1061         if (hv_clock->tsc_to_system_mul >= max_mul)
1062                 return false;
1063
1064         /*
1065          * Otherwise compute the scale and offset according to the formulas
1066          * derived above.
1067          */
1068         tsc_ref->tsc_scale =
1069                 mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
1070                                 hv_clock->tsc_to_system_mul,
1071                                 100);
1072
1073         tsc_ref->tsc_offset = hv_clock->system_time;
1074         do_div(tsc_ref->tsc_offset, 100);
1075         tsc_ref->tsc_offset -=
1076                 mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
1077         return true;
1078 }
1079
1080 void kvm_hv_setup_tsc_page(struct kvm *kvm,
1081                            struct pvclock_vcpu_time_info *hv_clock)
1082 {
1083         struct kvm_hv *hv = to_kvm_hv(kvm);
1084         u32 tsc_seq;
1085         u64 gfn;
1086
1087         BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
1088         BUILD_BUG_ON(offsetof(struct ms_hyperv_tsc_page, tsc_sequence) != 0);
1089
1090         if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1091                 return;
1092
1093         mutex_lock(&hv->hv_lock);
1094         if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1095                 goto out_unlock;
1096
1097         gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
1098         /*
1099          * Because the TSC parameters only vary when there is a
1100          * change in the master clock, do not bother with caching.
1101          */
1102         if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
1103                                     &tsc_seq, sizeof(tsc_seq))))
1104                 goto out_unlock;
1105
1106         /*
1107          * While we're computing and writing the parameters, force the
1108          * guest to use the time reference count MSR.
1109          */
1110         hv->tsc_ref.tsc_sequence = 0;
1111         if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
1112                             &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
1113                 goto out_unlock;
1114
1115         if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
1116                 goto out_unlock;
1117
1118         /* Ensure sequence is zero before writing the rest of the struct.  */
1119         smp_wmb();
1120         if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
1121                 goto out_unlock;
1122
1123         /*
1124          * Now switch to the TSC page mechanism by writing the sequence.
1125          */
1126         tsc_seq++;
1127         if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
1128                 tsc_seq = 1;
1129
1130         /* Write the struct entirely before the non-zero sequence.  */
1131         smp_wmb();
1132
1133         hv->tsc_ref.tsc_sequence = tsc_seq;
1134         kvm_write_guest(kvm, gfn_to_gpa(gfn),
1135                         &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence));
1136 out_unlock:
1137         mutex_unlock(&hv->hv_lock);
1138 }
1139
1140 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
1141                              bool host)
1142 {
1143         struct kvm *kvm = vcpu->kvm;
1144         struct kvm_hv *hv = to_kvm_hv(kvm);
1145
1146         switch (msr) {
1147         case HV_X64_MSR_GUEST_OS_ID:
1148                 hv->hv_guest_os_id = data;
1149                 /* setting guest os id to zero disables hypercall page */
1150                 if (!hv->hv_guest_os_id)
1151                         hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1152                 break;
1153         case HV_X64_MSR_HYPERCALL: {
1154                 u8 instructions[9];
1155                 int i = 0;
1156                 u64 addr;
1157
1158                 /* if guest os id is not set hypercall should remain disabled */
1159                 if (!hv->hv_guest_os_id)
1160                         break;
1161                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1162                         hv->hv_hypercall = data;
1163                         break;
1164                 }
1165
1166                 /*
1167                  * If Xen and Hyper-V hypercalls are both enabled, disambiguate
1168                  * the same way Xen itself does, by setting the bit 31 of EAX
1169                  * which is RsvdZ in the 32-bit Hyper-V hypercall ABI and just
1170                  * going to be clobbered on 64-bit.
1171                  */
1172                 if (kvm_xen_hypercall_enabled(kvm)) {
1173                         /* orl $0x80000000, %eax */
1174                         instructions[i++] = 0x0d;
1175                         instructions[i++] = 0x00;
1176                         instructions[i++] = 0x00;
1177                         instructions[i++] = 0x00;
1178                         instructions[i++] = 0x80;
1179                 }
1180
1181                 /* vmcall/vmmcall */
1182                 static_call(kvm_x86_patch_hypercall)(vcpu, instructions + i);
1183                 i += 3;
1184
1185                 /* ret */
1186                 ((unsigned char *)instructions)[i++] = 0xc3;
1187
1188                 addr = data & HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_MASK;
1189                 if (kvm_vcpu_write_guest(vcpu, addr, instructions, i))
1190                         return 1;
1191                 hv->hv_hypercall = data;
1192                 break;
1193         }
1194         case HV_X64_MSR_REFERENCE_TSC:
1195                 hv->hv_tsc_page = data;
1196                 if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE)
1197                         kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1198                 break;
1199         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1200                 return kvm_hv_msr_set_crash_data(kvm,
1201                                                  msr - HV_X64_MSR_CRASH_P0,
1202                                                  data);
1203         case HV_X64_MSR_CRASH_CTL:
1204                 if (host)
1205                         return kvm_hv_msr_set_crash_ctl(kvm, data);
1206
1207                 if (data & HV_CRASH_CTL_CRASH_NOTIFY) {
1208                         vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
1209                                    hv->hv_crash_param[0],
1210                                    hv->hv_crash_param[1],
1211                                    hv->hv_crash_param[2],
1212                                    hv->hv_crash_param[3],
1213                                    hv->hv_crash_param[4]);
1214
1215                         /* Send notification about crash to user space */
1216                         kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
1217                 }
1218                 break;
1219         case HV_X64_MSR_RESET:
1220                 if (data == 1) {
1221                         vcpu_debug(vcpu, "hyper-v reset requested\n");
1222                         kvm_make_request(KVM_REQ_HV_RESET, vcpu);
1223                 }
1224                 break;
1225         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1226                 hv->hv_reenlightenment_control = data;
1227                 break;
1228         case HV_X64_MSR_TSC_EMULATION_CONTROL:
1229                 hv->hv_tsc_emulation_control = data;
1230                 break;
1231         case HV_X64_MSR_TSC_EMULATION_STATUS:
1232                 hv->hv_tsc_emulation_status = data;
1233                 break;
1234         case HV_X64_MSR_TIME_REF_COUNT:
1235                 /* read-only, but still ignore it if host-initiated */
1236                 if (!host)
1237                         return 1;
1238                 break;
1239         case HV_X64_MSR_SYNDBG_OPTIONS:
1240         case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1241                 return syndbg_set_msr(vcpu, msr, data, host);
1242         default:
1243                 vcpu_unimpl(vcpu, "Hyper-V unhandled wrmsr: 0x%x data 0x%llx\n",
1244                             msr, data);
1245                 return 1;
1246         }
1247         return 0;
1248 }
1249
1250 /* Calculate cpu time spent by current task in 100ns units */
1251 static u64 current_task_runtime_100ns(void)
1252 {
1253         u64 utime, stime;
1254
1255         task_cputime_adjusted(current, &utime, &stime);
1256
1257         return div_u64(utime + stime, 100);
1258 }
1259
1260 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1261 {
1262         struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
1263
1264         switch (msr) {
1265         case HV_X64_MSR_VP_INDEX: {
1266                 struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1267                 int vcpu_idx = kvm_vcpu_get_idx(vcpu);
1268                 u32 new_vp_index = (u32)data;
1269
1270                 if (!host || new_vp_index >= KVM_MAX_VCPUS)
1271                         return 1;
1272
1273                 if (new_vp_index == hv_vcpu->vp_index)
1274                         return 0;
1275
1276                 /*
1277                  * The VP index is initialized to vcpu_index by
1278                  * kvm_hv_vcpu_postcreate so they initially match.  Now the
1279                  * VP index is changing, adjust num_mismatched_vp_indexes if
1280                  * it now matches or no longer matches vcpu_idx.
1281                  */
1282                 if (hv_vcpu->vp_index == vcpu_idx)
1283                         atomic_inc(&hv->num_mismatched_vp_indexes);
1284                 else if (new_vp_index == vcpu_idx)
1285                         atomic_dec(&hv->num_mismatched_vp_indexes);
1286
1287                 hv_vcpu->vp_index = new_vp_index;
1288                 break;
1289         }
1290         case HV_X64_MSR_VP_ASSIST_PAGE: {
1291                 u64 gfn;
1292                 unsigned long addr;
1293
1294                 if (!(data & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE)) {
1295                         hv_vcpu->hv_vapic = data;
1296                         if (kvm_lapic_enable_pv_eoi(vcpu, 0, 0))
1297                                 return 1;
1298                         break;
1299                 }
1300                 gfn = data >> HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT;
1301                 addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1302                 if (kvm_is_error_hva(addr))
1303                         return 1;
1304
1305                 /*
1306                  * Clear apic_assist portion of struct hv_vp_assist_page
1307                  * only, there can be valuable data in the rest which needs
1308                  * to be preserved e.g. on migration.
1309                  */
1310                 if (__put_user(0, (u32 __user *)addr))
1311                         return 1;
1312                 hv_vcpu->hv_vapic = data;
1313                 kvm_vcpu_mark_page_dirty(vcpu, gfn);
1314                 if (kvm_lapic_enable_pv_eoi(vcpu,
1315                                             gfn_to_gpa(gfn) | KVM_MSR_ENABLED,
1316                                             sizeof(struct hv_vp_assist_page)))
1317                         return 1;
1318                 break;
1319         }
1320         case HV_X64_MSR_EOI:
1321                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1322         case HV_X64_MSR_ICR:
1323                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1324         case HV_X64_MSR_TPR:
1325                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1326         case HV_X64_MSR_VP_RUNTIME:
1327                 if (!host)
1328                         return 1;
1329                 hv_vcpu->runtime_offset = data - current_task_runtime_100ns();
1330                 break;
1331         case HV_X64_MSR_SCONTROL:
1332         case HV_X64_MSR_SVERSION:
1333         case HV_X64_MSR_SIEFP:
1334         case HV_X64_MSR_SIMP:
1335         case HV_X64_MSR_EOM:
1336         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1337                 return synic_set_msr(to_hv_synic(vcpu), msr, data, host);
1338         case HV_X64_MSR_STIMER0_CONFIG:
1339         case HV_X64_MSR_STIMER1_CONFIG:
1340         case HV_X64_MSR_STIMER2_CONFIG:
1341         case HV_X64_MSR_STIMER3_CONFIG: {
1342                 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1343
1344                 return stimer_set_config(to_hv_stimer(vcpu, timer_index),
1345                                          data, host);
1346         }
1347         case HV_X64_MSR_STIMER0_COUNT:
1348         case HV_X64_MSR_STIMER1_COUNT:
1349         case HV_X64_MSR_STIMER2_COUNT:
1350         case HV_X64_MSR_STIMER3_COUNT: {
1351                 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1352
1353                 return stimer_set_count(to_hv_stimer(vcpu, timer_index),
1354                                         data, host);
1355         }
1356         case HV_X64_MSR_TSC_FREQUENCY:
1357         case HV_X64_MSR_APIC_FREQUENCY:
1358                 /* read-only, but still ignore it if host-initiated */
1359                 if (!host)
1360                         return 1;
1361                 break;
1362         default:
1363                 vcpu_unimpl(vcpu, "Hyper-V unhandled wrmsr: 0x%x data 0x%llx\n",
1364                             msr, data);
1365                 return 1;
1366         }
1367
1368         return 0;
1369 }
1370
1371 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1372                              bool host)
1373 {
1374         u64 data = 0;
1375         struct kvm *kvm = vcpu->kvm;
1376         struct kvm_hv *hv = to_kvm_hv(kvm);
1377
1378         switch (msr) {
1379         case HV_X64_MSR_GUEST_OS_ID:
1380                 data = hv->hv_guest_os_id;
1381                 break;
1382         case HV_X64_MSR_HYPERCALL:
1383                 data = hv->hv_hypercall;
1384                 break;
1385         case HV_X64_MSR_TIME_REF_COUNT:
1386                 data = get_time_ref_counter(kvm);
1387                 break;
1388         case HV_X64_MSR_REFERENCE_TSC:
1389                 data = hv->hv_tsc_page;
1390                 break;
1391         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1392                 return kvm_hv_msr_get_crash_data(kvm,
1393                                                  msr - HV_X64_MSR_CRASH_P0,
1394                                                  pdata);
1395         case HV_X64_MSR_CRASH_CTL:
1396                 return kvm_hv_msr_get_crash_ctl(kvm, pdata);
1397         case HV_X64_MSR_RESET:
1398                 data = 0;
1399                 break;
1400         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1401                 data = hv->hv_reenlightenment_control;
1402                 break;
1403         case HV_X64_MSR_TSC_EMULATION_CONTROL:
1404                 data = hv->hv_tsc_emulation_control;
1405                 break;
1406         case HV_X64_MSR_TSC_EMULATION_STATUS:
1407                 data = hv->hv_tsc_emulation_status;
1408                 break;
1409         case HV_X64_MSR_SYNDBG_OPTIONS:
1410         case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1411                 return syndbg_get_msr(vcpu, msr, pdata, host);
1412         default:
1413                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1414                 return 1;
1415         }
1416
1417         *pdata = data;
1418         return 0;
1419 }
1420
1421 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1422                           bool host)
1423 {
1424         u64 data = 0;
1425         struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
1426
1427         switch (msr) {
1428         case HV_X64_MSR_VP_INDEX:
1429                 data = hv_vcpu->vp_index;
1430                 break;
1431         case HV_X64_MSR_EOI:
1432                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1433         case HV_X64_MSR_ICR:
1434                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1435         case HV_X64_MSR_TPR:
1436                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1437         case HV_X64_MSR_VP_ASSIST_PAGE:
1438                 data = hv_vcpu->hv_vapic;
1439                 break;
1440         case HV_X64_MSR_VP_RUNTIME:
1441                 data = current_task_runtime_100ns() + hv_vcpu->runtime_offset;
1442                 break;
1443         case HV_X64_MSR_SCONTROL:
1444         case HV_X64_MSR_SVERSION:
1445         case HV_X64_MSR_SIEFP:
1446         case HV_X64_MSR_SIMP:
1447         case HV_X64_MSR_EOM:
1448         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1449                 return synic_get_msr(to_hv_synic(vcpu), msr, pdata, host);
1450         case HV_X64_MSR_STIMER0_CONFIG:
1451         case HV_X64_MSR_STIMER1_CONFIG:
1452         case HV_X64_MSR_STIMER2_CONFIG:
1453         case HV_X64_MSR_STIMER3_CONFIG: {
1454                 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1455
1456                 return stimer_get_config(to_hv_stimer(vcpu, timer_index),
1457                                          pdata);
1458         }
1459         case HV_X64_MSR_STIMER0_COUNT:
1460         case HV_X64_MSR_STIMER1_COUNT:
1461         case HV_X64_MSR_STIMER2_COUNT:
1462         case HV_X64_MSR_STIMER3_COUNT: {
1463                 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1464
1465                 return stimer_get_count(to_hv_stimer(vcpu, timer_index),
1466                                         pdata);
1467         }
1468         case HV_X64_MSR_TSC_FREQUENCY:
1469                 data = (u64)vcpu->arch.virtual_tsc_khz * 1000;
1470                 break;
1471         case HV_X64_MSR_APIC_FREQUENCY:
1472                 data = APIC_BUS_FREQUENCY;
1473                 break;
1474         default:
1475                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1476                 return 1;
1477         }
1478         *pdata = data;
1479         return 0;
1480 }
1481
1482 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1483 {
1484         struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1485
1486         if (!host && !vcpu->arch.hyperv_enabled)
1487                 return 1;
1488
1489         if (!to_hv_vcpu(vcpu)) {
1490                 if (kvm_hv_vcpu_init(vcpu))
1491                         return 1;
1492         }
1493
1494         if (kvm_hv_msr_partition_wide(msr)) {
1495                 int r;
1496
1497                 mutex_lock(&hv->hv_lock);
1498                 r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1499                 mutex_unlock(&hv->hv_lock);
1500                 return r;
1501         } else
1502                 return kvm_hv_set_msr(vcpu, msr, data, host);
1503 }
1504
1505 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
1506 {
1507         struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1508
1509         if (!host && !vcpu->arch.hyperv_enabled)
1510                 return 1;
1511
1512         if (!to_hv_vcpu(vcpu)) {
1513                 if (kvm_hv_vcpu_init(vcpu))
1514                         return 1;
1515         }
1516
1517         if (kvm_hv_msr_partition_wide(msr)) {
1518                 int r;
1519
1520                 mutex_lock(&hv->hv_lock);
1521                 r = kvm_hv_get_msr_pw(vcpu, msr, pdata, host);
1522                 mutex_unlock(&hv->hv_lock);
1523                 return r;
1524         } else
1525                 return kvm_hv_get_msr(vcpu, msr, pdata, host);
1526 }
1527
1528 static __always_inline unsigned long *sparse_set_to_vcpu_mask(
1529         struct kvm *kvm, u64 *sparse_banks, u64 valid_bank_mask,
1530         u64 *vp_bitmap, unsigned long *vcpu_bitmap)
1531 {
1532         struct kvm_hv *hv = to_kvm_hv(kvm);
1533         struct kvm_vcpu *vcpu;
1534         int i, bank, sbank = 0;
1535
1536         memset(vp_bitmap, 0,
1537                KVM_HV_MAX_SPARSE_VCPU_SET_BITS * sizeof(*vp_bitmap));
1538         for_each_set_bit(bank, (unsigned long *)&valid_bank_mask,
1539                          KVM_HV_MAX_SPARSE_VCPU_SET_BITS)
1540                 vp_bitmap[bank] = sparse_banks[sbank++];
1541
1542         if (likely(!atomic_read(&hv->num_mismatched_vp_indexes))) {
1543                 /* for all vcpus vp_index == vcpu_idx */
1544                 return (unsigned long *)vp_bitmap;
1545         }
1546
1547         bitmap_zero(vcpu_bitmap, KVM_MAX_VCPUS);
1548         kvm_for_each_vcpu(i, vcpu, kvm) {
1549                 if (test_bit(kvm_hv_get_vpindex(vcpu), (unsigned long *)vp_bitmap))
1550                         __set_bit(i, vcpu_bitmap);
1551         }
1552         return vcpu_bitmap;
1553 }
1554
1555 static u64 kvm_hv_flush_tlb(struct kvm_vcpu *vcpu, u64 ingpa, u16 rep_cnt, bool ex)
1556 {
1557         struct kvm *kvm = vcpu->kvm;
1558         struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
1559         struct hv_tlb_flush_ex flush_ex;
1560         struct hv_tlb_flush flush;
1561         u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1562         DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1563         unsigned long *vcpu_mask;
1564         u64 valid_bank_mask;
1565         u64 sparse_banks[64];
1566         int sparse_banks_len;
1567         bool all_cpus;
1568
1569         if (!ex) {
1570                 if (unlikely(kvm_read_guest(kvm, ingpa, &flush, sizeof(flush))))
1571                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
1572
1573                 trace_kvm_hv_flush_tlb(flush.processor_mask,
1574                                        flush.address_space, flush.flags);
1575
1576                 valid_bank_mask = BIT_ULL(0);
1577                 sparse_banks[0] = flush.processor_mask;
1578
1579                 /*
1580                  * Work around possible WS2012 bug: it sends hypercalls
1581                  * with processor_mask = 0x0 and HV_FLUSH_ALL_PROCESSORS clear,
1582                  * while also expecting us to flush something and crashing if
1583                  * we don't. Let's treat processor_mask == 0 same as
1584                  * HV_FLUSH_ALL_PROCESSORS.
1585                  */
1586                 all_cpus = (flush.flags & HV_FLUSH_ALL_PROCESSORS) ||
1587                         flush.processor_mask == 0;
1588         } else {
1589                 if (unlikely(kvm_read_guest(kvm, ingpa, &flush_ex,
1590                                             sizeof(flush_ex))))
1591                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
1592
1593                 trace_kvm_hv_flush_tlb_ex(flush_ex.hv_vp_set.valid_bank_mask,
1594                                           flush_ex.hv_vp_set.format,
1595                                           flush_ex.address_space,
1596                                           flush_ex.flags);
1597
1598                 valid_bank_mask = flush_ex.hv_vp_set.valid_bank_mask;
1599                 all_cpus = flush_ex.hv_vp_set.format !=
1600                         HV_GENERIC_SET_SPARSE_4K;
1601
1602                 sparse_banks_len =
1603                         bitmap_weight((unsigned long *)&valid_bank_mask, 64) *
1604                         sizeof(sparse_banks[0]);
1605
1606                 if (!sparse_banks_len && !all_cpus)
1607                         goto ret_success;
1608
1609                 if (!all_cpus &&
1610                     kvm_read_guest(kvm,
1611                                    ingpa + offsetof(struct hv_tlb_flush_ex,
1612                                                     hv_vp_set.bank_contents),
1613                                    sparse_banks,
1614                                    sparse_banks_len))
1615                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
1616         }
1617
1618         cpumask_clear(&hv_vcpu->tlb_flush);
1619
1620         vcpu_mask = all_cpus ? NULL :
1621                 sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1622                                         vp_bitmap, vcpu_bitmap);
1623
1624         /*
1625          * vcpu->arch.cr3 may not be up-to-date for running vCPUs so we can't
1626          * analyze it here, flush TLB regardless of the specified address space.
1627          */
1628         kvm_make_vcpus_request_mask(kvm, KVM_REQ_HV_TLB_FLUSH,
1629                                     NULL, vcpu_mask, &hv_vcpu->tlb_flush);
1630
1631 ret_success:
1632         /* We always do full TLB flush, set rep_done = rep_cnt. */
1633         return (u64)HV_STATUS_SUCCESS |
1634                 ((u64)rep_cnt << HV_HYPERCALL_REP_COMP_OFFSET);
1635 }
1636
1637 static void kvm_send_ipi_to_many(struct kvm *kvm, u32 vector,
1638                                  unsigned long *vcpu_bitmap)
1639 {
1640         struct kvm_lapic_irq irq = {
1641                 .delivery_mode = APIC_DM_FIXED,
1642                 .vector = vector
1643         };
1644         struct kvm_vcpu *vcpu;
1645         int i;
1646
1647         kvm_for_each_vcpu(i, vcpu, kvm) {
1648                 if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
1649                         continue;
1650
1651                 /* We fail only when APIC is disabled */
1652                 kvm_apic_set_irq(vcpu, &irq, NULL);
1653         }
1654 }
1655
1656 static u64 kvm_hv_send_ipi(struct kvm_vcpu *vcpu, u64 ingpa, u64 outgpa,
1657                            bool ex, bool fast)
1658 {
1659         struct kvm *kvm = vcpu->kvm;
1660         struct hv_send_ipi_ex send_ipi_ex;
1661         struct hv_send_ipi send_ipi;
1662         u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1663         DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1664         unsigned long *vcpu_mask;
1665         unsigned long valid_bank_mask;
1666         u64 sparse_banks[64];
1667         int sparse_banks_len;
1668         u32 vector;
1669         bool all_cpus;
1670
1671         if (!ex) {
1672                 if (!fast) {
1673                         if (unlikely(kvm_read_guest(kvm, ingpa, &send_ipi,
1674                                                     sizeof(send_ipi))))
1675                                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1676                         sparse_banks[0] = send_ipi.cpu_mask;
1677                         vector = send_ipi.vector;
1678                 } else {
1679                         /* 'reserved' part of hv_send_ipi should be 0 */
1680                         if (unlikely(ingpa >> 32 != 0))
1681                                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1682                         sparse_banks[0] = outgpa;
1683                         vector = (u32)ingpa;
1684                 }
1685                 all_cpus = false;
1686                 valid_bank_mask = BIT_ULL(0);
1687
1688                 trace_kvm_hv_send_ipi(vector, sparse_banks[0]);
1689         } else {
1690                 if (unlikely(kvm_read_guest(kvm, ingpa, &send_ipi_ex,
1691                                             sizeof(send_ipi_ex))))
1692                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
1693
1694                 trace_kvm_hv_send_ipi_ex(send_ipi_ex.vector,
1695                                          send_ipi_ex.vp_set.format,
1696                                          send_ipi_ex.vp_set.valid_bank_mask);
1697
1698                 vector = send_ipi_ex.vector;
1699                 valid_bank_mask = send_ipi_ex.vp_set.valid_bank_mask;
1700                 sparse_banks_len = bitmap_weight(&valid_bank_mask, 64) *
1701                         sizeof(sparse_banks[0]);
1702
1703                 all_cpus = send_ipi_ex.vp_set.format == HV_GENERIC_SET_ALL;
1704
1705                 if (!sparse_banks_len)
1706                         goto ret_success;
1707
1708                 if (!all_cpus &&
1709                     kvm_read_guest(kvm,
1710                                    ingpa + offsetof(struct hv_send_ipi_ex,
1711                                                     vp_set.bank_contents),
1712                                    sparse_banks,
1713                                    sparse_banks_len))
1714                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
1715         }
1716
1717         if ((vector < HV_IPI_LOW_VECTOR) || (vector > HV_IPI_HIGH_VECTOR))
1718                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1719
1720         vcpu_mask = all_cpus ? NULL :
1721                 sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1722                                         vp_bitmap, vcpu_bitmap);
1723
1724         kvm_send_ipi_to_many(kvm, vector, vcpu_mask);
1725
1726 ret_success:
1727         return HV_STATUS_SUCCESS;
1728 }
1729
1730 void kvm_hv_set_cpuid(struct kvm_vcpu *vcpu)
1731 {
1732         struct kvm_cpuid_entry2 *entry;
1733
1734         entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_INTERFACE, 0);
1735         if (entry && entry->eax == HYPERV_CPUID_SIGNATURE_EAX)
1736                 vcpu->arch.hyperv_enabled = true;
1737         else
1738                 vcpu->arch.hyperv_enabled = false;
1739 }
1740
1741 bool kvm_hv_hypercall_enabled(struct kvm_vcpu *vcpu)
1742 {
1743         return vcpu->arch.hyperv_enabled && to_kvm_hv(vcpu->kvm)->hv_guest_os_id;
1744 }
1745
1746 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
1747 {
1748         bool longmode;
1749
1750         longmode = is_64_bit_mode(vcpu);
1751         if (longmode)
1752                 kvm_rax_write(vcpu, result);
1753         else {
1754                 kvm_rdx_write(vcpu, result >> 32);
1755                 kvm_rax_write(vcpu, result & 0xffffffff);
1756         }
1757 }
1758
1759 static int kvm_hv_hypercall_complete(struct kvm_vcpu *vcpu, u64 result)
1760 {
1761         kvm_hv_hypercall_set_result(vcpu, result);
1762         ++vcpu->stat.hypercalls;
1763         return kvm_skip_emulated_instruction(vcpu);
1764 }
1765
1766 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
1767 {
1768         return kvm_hv_hypercall_complete(vcpu, vcpu->run->hyperv.u.hcall.result);
1769 }
1770
1771 static u16 kvm_hvcall_signal_event(struct kvm_vcpu *vcpu, bool fast, u64 param)
1772 {
1773         struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1774         struct eventfd_ctx *eventfd;
1775
1776         if (unlikely(!fast)) {
1777                 int ret;
1778                 gpa_t gpa = param;
1779
1780                 if ((gpa & (__alignof__(param) - 1)) ||
1781                     offset_in_page(gpa) + sizeof(param) > PAGE_SIZE)
1782                         return HV_STATUS_INVALID_ALIGNMENT;
1783
1784                 ret = kvm_vcpu_read_guest(vcpu, gpa, &param, sizeof(param));
1785                 if (ret < 0)
1786                         return HV_STATUS_INVALID_ALIGNMENT;
1787         }
1788
1789         /*
1790          * Per spec, bits 32-47 contain the extra "flag number".  However, we
1791          * have no use for it, and in all known usecases it is zero, so just
1792          * report lookup failure if it isn't.
1793          */
1794         if (param & 0xffff00000000ULL)
1795                 return HV_STATUS_INVALID_PORT_ID;
1796         /* remaining bits are reserved-zero */
1797         if (param & ~KVM_HYPERV_CONN_ID_MASK)
1798                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1799
1800         /* the eventfd is protected by vcpu->kvm->srcu, but conn_to_evt isn't */
1801         rcu_read_lock();
1802         eventfd = idr_find(&hv->conn_to_evt, param);
1803         rcu_read_unlock();
1804         if (!eventfd)
1805                 return HV_STATUS_INVALID_PORT_ID;
1806
1807         eventfd_signal(eventfd, 1);
1808         return HV_STATUS_SUCCESS;
1809 }
1810
1811 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
1812 {
1813         u64 param, ingpa, outgpa, ret = HV_STATUS_SUCCESS;
1814         uint16_t code, rep_idx, rep_cnt;
1815         bool fast, rep;
1816
1817         /*
1818          * hypercall generates UD from non zero cpl and real mode
1819          * per HYPER-V spec
1820          */
1821         if (static_call(kvm_x86_get_cpl)(vcpu) != 0 || !is_protmode(vcpu)) {
1822                 kvm_queue_exception(vcpu, UD_VECTOR);
1823                 return 1;
1824         }
1825
1826 #ifdef CONFIG_X86_64
1827         if (is_64_bit_mode(vcpu)) {
1828                 param = kvm_rcx_read(vcpu);
1829                 ingpa = kvm_rdx_read(vcpu);
1830                 outgpa = kvm_r8_read(vcpu);
1831         } else
1832 #endif
1833         {
1834                 param = ((u64)kvm_rdx_read(vcpu) << 32) |
1835                         (kvm_rax_read(vcpu) & 0xffffffff);
1836                 ingpa = ((u64)kvm_rbx_read(vcpu) << 32) |
1837                         (kvm_rcx_read(vcpu) & 0xffffffff);
1838                 outgpa = ((u64)kvm_rdi_read(vcpu) << 32) |
1839                         (kvm_rsi_read(vcpu) & 0xffffffff);
1840         }
1841
1842         code = param & 0xffff;
1843         fast = !!(param & HV_HYPERCALL_FAST_BIT);
1844         rep_cnt = (param >> HV_HYPERCALL_REP_COMP_OFFSET) & 0xfff;
1845         rep_idx = (param >> HV_HYPERCALL_REP_START_OFFSET) & 0xfff;
1846         rep = !!(rep_cnt || rep_idx);
1847
1848         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
1849
1850         switch (code) {
1851         case HVCALL_NOTIFY_LONG_SPIN_WAIT:
1852                 if (unlikely(rep)) {
1853                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1854                         break;
1855                 }
1856                 kvm_vcpu_on_spin(vcpu, true);
1857                 break;
1858         case HVCALL_SIGNAL_EVENT:
1859                 if (unlikely(rep)) {
1860                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1861                         break;
1862                 }
1863                 ret = kvm_hvcall_signal_event(vcpu, fast, ingpa);
1864                 if (ret != HV_STATUS_INVALID_PORT_ID)
1865                         break;
1866                 fallthrough;    /* maybe userspace knows this conn_id */
1867         case HVCALL_POST_MESSAGE:
1868                 /* don't bother userspace if it has no way to handle it */
1869                 if (unlikely(rep || !to_hv_synic(vcpu)->active)) {
1870                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1871                         break;
1872                 }
1873                 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
1874                 vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
1875                 vcpu->run->hyperv.u.hcall.input = param;
1876                 vcpu->run->hyperv.u.hcall.params[0] = ingpa;
1877                 vcpu->run->hyperv.u.hcall.params[1] = outgpa;
1878                 vcpu->arch.complete_userspace_io =
1879                                 kvm_hv_hypercall_complete_userspace;
1880                 return 0;
1881         case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
1882                 if (unlikely(fast || !rep_cnt || rep_idx)) {
1883                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1884                         break;
1885                 }
1886                 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false);
1887                 break;
1888         case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
1889                 if (unlikely(fast || rep)) {
1890                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1891                         break;
1892                 }
1893                 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false);
1894                 break;
1895         case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
1896                 if (unlikely(fast || !rep_cnt || rep_idx)) {
1897                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1898                         break;
1899                 }
1900                 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
1901                 break;
1902         case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
1903                 if (unlikely(fast || rep)) {
1904                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1905                         break;
1906                 }
1907                 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
1908                 break;
1909         case HVCALL_SEND_IPI:
1910                 if (unlikely(rep)) {
1911                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1912                         break;
1913                 }
1914                 ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, false, fast);
1915                 break;
1916         case HVCALL_SEND_IPI_EX:
1917                 if (unlikely(fast || rep)) {
1918                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1919                         break;
1920                 }
1921                 ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, true, false);
1922                 break;
1923         case HVCALL_POST_DEBUG_DATA:
1924         case HVCALL_RETRIEVE_DEBUG_DATA:
1925                 if (unlikely(fast)) {
1926                         ret = HV_STATUS_INVALID_PARAMETER;
1927                         break;
1928                 }
1929                 fallthrough;
1930         case HVCALL_RESET_DEBUG_SESSION: {
1931                 struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
1932
1933                 if (!kvm_hv_is_syndbg_enabled(vcpu)) {
1934                         ret = HV_STATUS_INVALID_HYPERCALL_CODE;
1935                         break;
1936                 }
1937
1938                 if (!(syndbg->options & HV_X64_SYNDBG_OPTION_USE_HCALLS)) {
1939                         ret = HV_STATUS_OPERATION_DENIED;
1940                         break;
1941                 }
1942                 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
1943                 vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
1944                 vcpu->run->hyperv.u.hcall.input = param;
1945                 vcpu->run->hyperv.u.hcall.params[0] = ingpa;
1946                 vcpu->run->hyperv.u.hcall.params[1] = outgpa;
1947                 vcpu->arch.complete_userspace_io =
1948                                 kvm_hv_hypercall_complete_userspace;
1949                 return 0;
1950         }
1951         default:
1952                 ret = HV_STATUS_INVALID_HYPERCALL_CODE;
1953                 break;
1954         }
1955
1956         return kvm_hv_hypercall_complete(vcpu, ret);
1957 }
1958
1959 void kvm_hv_init_vm(struct kvm *kvm)
1960 {
1961         struct kvm_hv *hv = to_kvm_hv(kvm);
1962
1963         mutex_init(&hv->hv_lock);
1964         idr_init(&hv->conn_to_evt);
1965 }
1966
1967 void kvm_hv_destroy_vm(struct kvm *kvm)
1968 {
1969         struct kvm_hv *hv = to_kvm_hv(kvm);
1970         struct eventfd_ctx *eventfd;
1971         int i;
1972
1973         idr_for_each_entry(&hv->conn_to_evt, eventfd, i)
1974                 eventfd_ctx_put(eventfd);
1975         idr_destroy(&hv->conn_to_evt);
1976 }
1977
1978 static int kvm_hv_eventfd_assign(struct kvm *kvm, u32 conn_id, int fd)
1979 {
1980         struct kvm_hv *hv = to_kvm_hv(kvm);
1981         struct eventfd_ctx *eventfd;
1982         int ret;
1983
1984         eventfd = eventfd_ctx_fdget(fd);
1985         if (IS_ERR(eventfd))
1986                 return PTR_ERR(eventfd);
1987
1988         mutex_lock(&hv->hv_lock);
1989         ret = idr_alloc(&hv->conn_to_evt, eventfd, conn_id, conn_id + 1,
1990                         GFP_KERNEL_ACCOUNT);
1991         mutex_unlock(&hv->hv_lock);
1992
1993         if (ret >= 0)
1994                 return 0;
1995
1996         if (ret == -ENOSPC)
1997                 ret = -EEXIST;
1998         eventfd_ctx_put(eventfd);
1999         return ret;
2000 }
2001
2002 static int kvm_hv_eventfd_deassign(struct kvm *kvm, u32 conn_id)
2003 {
2004         struct kvm_hv *hv = to_kvm_hv(kvm);
2005         struct eventfd_ctx *eventfd;
2006
2007         mutex_lock(&hv->hv_lock);
2008         eventfd = idr_remove(&hv->conn_to_evt, conn_id);
2009         mutex_unlock(&hv->hv_lock);
2010
2011         if (!eventfd)
2012                 return -ENOENT;
2013
2014         synchronize_srcu(&kvm->srcu);
2015         eventfd_ctx_put(eventfd);
2016         return 0;
2017 }
2018
2019 int kvm_vm_ioctl_hv_eventfd(struct kvm *kvm, struct kvm_hyperv_eventfd *args)
2020 {
2021         if ((args->flags & ~KVM_HYPERV_EVENTFD_DEASSIGN) ||
2022             (args->conn_id & ~KVM_HYPERV_CONN_ID_MASK))
2023                 return -EINVAL;
2024
2025         if (args->flags == KVM_HYPERV_EVENTFD_DEASSIGN)
2026                 return kvm_hv_eventfd_deassign(kvm, args->conn_id);
2027         return kvm_hv_eventfd_assign(kvm, args->conn_id, args->fd);
2028 }
2029
2030 int kvm_get_hv_cpuid(struct kvm_vcpu *vcpu, struct kvm_cpuid2 *cpuid,
2031                      struct kvm_cpuid_entry2 __user *entries)
2032 {
2033         uint16_t evmcs_ver = 0;
2034         struct kvm_cpuid_entry2 cpuid_entries[] = {
2035                 { .function = HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS },
2036                 { .function = HYPERV_CPUID_INTERFACE },
2037                 { .function = HYPERV_CPUID_VERSION },
2038                 { .function = HYPERV_CPUID_FEATURES },
2039                 { .function = HYPERV_CPUID_ENLIGHTMENT_INFO },
2040                 { .function = HYPERV_CPUID_IMPLEMENT_LIMITS },
2041                 { .function = HYPERV_CPUID_SYNDBG_VENDOR_AND_MAX_FUNCTIONS },
2042                 { .function = HYPERV_CPUID_SYNDBG_INTERFACE },
2043                 { .function = HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES },
2044                 { .function = HYPERV_CPUID_NESTED_FEATURES },
2045         };
2046         int i, nent = ARRAY_SIZE(cpuid_entries);
2047
2048         if (kvm_x86_ops.nested_ops->get_evmcs_version)
2049                 evmcs_ver = kvm_x86_ops.nested_ops->get_evmcs_version(vcpu);
2050
2051         /* Skip NESTED_FEATURES if eVMCS is not supported */
2052         if (!evmcs_ver)
2053                 --nent;
2054
2055         if (cpuid->nent < nent)
2056                 return -E2BIG;
2057
2058         if (cpuid->nent > nent)
2059                 cpuid->nent = nent;
2060
2061         for (i = 0; i < nent; i++) {
2062                 struct kvm_cpuid_entry2 *ent = &cpuid_entries[i];
2063                 u32 signature[3];
2064
2065                 switch (ent->function) {
2066                 case HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS:
2067                         memcpy(signature, "Linux KVM Hv", 12);
2068
2069                         ent->eax = HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES;
2070                         ent->ebx = signature[0];
2071                         ent->ecx = signature[1];
2072                         ent->edx = signature[2];
2073                         break;
2074
2075                 case HYPERV_CPUID_INTERFACE:
2076                         ent->eax = HYPERV_CPUID_SIGNATURE_EAX;
2077                         break;
2078
2079                 case HYPERV_CPUID_VERSION:
2080                         /*
2081                          * We implement some Hyper-V 2016 functions so let's use
2082                          * this version.
2083                          */
2084                         ent->eax = 0x00003839;
2085                         ent->ebx = 0x000A0000;
2086                         break;
2087
2088                 case HYPERV_CPUID_FEATURES:
2089                         ent->eax |= HV_MSR_VP_RUNTIME_AVAILABLE;
2090                         ent->eax |= HV_MSR_TIME_REF_COUNT_AVAILABLE;
2091                         ent->eax |= HV_MSR_SYNIC_AVAILABLE;
2092                         ent->eax |= HV_MSR_SYNTIMER_AVAILABLE;
2093                         ent->eax |= HV_MSR_APIC_ACCESS_AVAILABLE;
2094                         ent->eax |= HV_MSR_HYPERCALL_AVAILABLE;
2095                         ent->eax |= HV_MSR_VP_INDEX_AVAILABLE;
2096                         ent->eax |= HV_MSR_RESET_AVAILABLE;
2097                         ent->eax |= HV_MSR_REFERENCE_TSC_AVAILABLE;
2098                         ent->eax |= HV_ACCESS_FREQUENCY_MSRS;
2099                         ent->eax |= HV_ACCESS_REENLIGHTENMENT;
2100
2101                         ent->ebx |= HV_POST_MESSAGES;
2102                         ent->ebx |= HV_SIGNAL_EVENTS;
2103
2104                         ent->edx |= HV_FEATURE_FREQUENCY_MSRS_AVAILABLE;
2105                         ent->edx |= HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
2106
2107                         ent->ebx |= HV_DEBUGGING;
2108                         ent->edx |= HV_X64_GUEST_DEBUGGING_AVAILABLE;
2109                         ent->edx |= HV_FEATURE_DEBUG_MSRS_AVAILABLE;
2110
2111                         /*
2112                          * Direct Synthetic timers only make sense with in-kernel
2113                          * LAPIC
2114                          */
2115                         if (!vcpu || lapic_in_kernel(vcpu))
2116                                 ent->edx |= HV_STIMER_DIRECT_MODE_AVAILABLE;
2117
2118                         break;
2119
2120                 case HYPERV_CPUID_ENLIGHTMENT_INFO:
2121                         ent->eax |= HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
2122                         ent->eax |= HV_X64_APIC_ACCESS_RECOMMENDED;
2123                         ent->eax |= HV_X64_RELAXED_TIMING_RECOMMENDED;
2124                         ent->eax |= HV_X64_CLUSTER_IPI_RECOMMENDED;
2125                         ent->eax |= HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED;
2126                         if (evmcs_ver)
2127                                 ent->eax |= HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
2128                         if (!cpu_smt_possible())
2129                                 ent->eax |= HV_X64_NO_NONARCH_CORESHARING;
2130                         /*
2131                          * Default number of spinlock retry attempts, matches
2132                          * HyperV 2016.
2133                          */
2134                         ent->ebx = 0x00000FFF;
2135
2136                         break;
2137
2138                 case HYPERV_CPUID_IMPLEMENT_LIMITS:
2139                         /* Maximum number of virtual processors */
2140                         ent->eax = KVM_MAX_VCPUS;
2141                         /*
2142                          * Maximum number of logical processors, matches
2143                          * HyperV 2016.
2144                          */
2145                         ent->ebx = 64;
2146
2147                         break;
2148
2149                 case HYPERV_CPUID_NESTED_FEATURES:
2150                         ent->eax = evmcs_ver;
2151
2152                         break;
2153
2154                 case HYPERV_CPUID_SYNDBG_VENDOR_AND_MAX_FUNCTIONS:
2155                         memcpy(signature, "Linux KVM Hv", 12);
2156
2157                         ent->eax = 0;
2158                         ent->ebx = signature[0];
2159                         ent->ecx = signature[1];
2160                         ent->edx = signature[2];
2161                         break;
2162
2163                 case HYPERV_CPUID_SYNDBG_INTERFACE:
2164                         memcpy(signature, "VS#1\0\0\0\0\0\0\0\0", 12);
2165                         ent->eax = signature[0];
2166                         break;
2167
2168                 case HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES:
2169                         ent->eax |= HV_X64_SYNDBG_CAP_ALLOW_KERNEL_DEBUGGING;
2170                         break;
2171
2172                 default:
2173                         break;
2174                 }
2175         }
2176
2177         if (copy_to_user(entries, cpuid_entries,
2178                          nent * sizeof(struct kvm_cpuid_entry2)))
2179                 return -EFAULT;
2180
2181         return 0;
2182 }