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
  27 #include <linux/cpu.h>
  28 #include <linux/kvm_host.h>
  29 #include <linux/highmem.h>
  30 #include <linux/sched/cputime.h>
  31 #include <linux/eventfd.h>
  32
  33 #include <asm/apicdef.h>
  34 #include <trace/events/kvm.h>
  35
  36 #include "trace.h"
  37 #include "irq.h"
  38
  39 #define KVM_HV_MAX_SPARSE_VCPU_SET_BITS DIV_ROUND_UP(KVM_MAX_VCPUS, 64)
  40
  41 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
  42                                 bool vcpu_kick);
  43
  44 static inline u64 synic_read_sint(struct kvm_vcpu_hv_synic *synic, int sint)
  45 {
  46         return atomic64_read(&synic->sint[sint]);
  47 }
  48
  49 static inline int synic_get_sint_vector(u64 sint_value)
  50 {
  51         if (sint_value & HV_SYNIC_SINT_MASKED)
  52                 return -1;
  53         return sint_value & HV_SYNIC_SINT_VECTOR_MASK;
  54 }
  55
  56 static bool synic_has_vector_connected(struct kvm_vcpu_hv_synic *synic,
  57                                       int vector)
  58 {
  59         int i;
  60
  61         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
  62                 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
  63                         return true;
  64         }
  65         return false;
  66 }
  67
  68 static bool synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic *synic,
  69                                      int vector)
  70 {
  71         int i;
  72         u64 sint_value;
  73
  74         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
  75                 sint_value = synic_read_sint(synic, i);
  76                 if (synic_get_sint_vector(sint_value) == vector &&
  77                     sint_value & HV_SYNIC_SINT_AUTO_EOI)
  78                         return true;
  79         }
  80         return false;
  81 }
  82
  83 static void synic_update_vector(struct kvm_vcpu_hv_synic *synic,
  84                                 int vector)
  85 {
  86         if (vector < HV_SYNIC_FIRST_VALID_VECTOR)
  87                 return;
  88
  89         if (synic_has_vector_connected(synic, vector))
  90                 __set_bit(vector, synic->vec_bitmap);
  91         else
  92                 __clear_bit(vector, synic->vec_bitmap);
  93
  94         if (synic_has_vector_auto_eoi(synic, vector))
  95                 __set_bit(vector, synic->auto_eoi_bitmap);
  96         else
  97                 __clear_bit(vector, synic->auto_eoi_bitmap);
  98 }
  99
 100 static int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint,
 101                           u64 data, bool host)
 102 {
 103         int vector, old_vector;
 104         bool masked;
 105
 106         vector = data & HV_SYNIC_SINT_VECTOR_MASK;
 107         masked = data & HV_SYNIC_SINT_MASKED;
 108
 109         /*
 110          * Valid vectors are 16-255, however, nested Hyper-V attempts to write
 111          * default '0x10000' value on boot and this should not #GP. We need to
 112          * allow zero-initing the register from host as well.
 113          */
 114         if (vector < HV_SYNIC_FIRST_VALID_VECTOR && !host && !masked)
 115                 return 1;
 116         /*
 117          * Guest may configure multiple SINTs to use the same vector, so
 118          * we maintain a bitmap of vectors handled by synic, and a
 119          * bitmap of vectors with auto-eoi behavior.  The bitmaps are
 120          * updated here, and atomically queried on fast paths.
 121          */
 122         old_vector = synic_read_sint(synic, sint) & HV_SYNIC_SINT_VECTOR_MASK;
 123
 124         atomic64_set(&synic->sint[sint], data);
 125
 126         synic_update_vector(synic, old_vector);
 127
 128         synic_update_vector(synic, vector);
 129
 130         /* Load SynIC vectors into EOI exit bitmap */
 131         kvm_make_request(KVM_REQ_SCAN_IOAPIC, synic_to_vcpu(synic));
 132         return 0;
 133 }
 134
 135 static struct kvm_vcpu *get_vcpu_by_vpidx(struct kvm *kvm, u32 vpidx)
 136 {
 137         struct kvm_vcpu *vcpu = NULL;
 138         int i;
 139
 140         if (vpidx >= KVM_MAX_VCPUS)
 141                 return NULL;
 142
 143         vcpu = kvm_get_vcpu(kvm, vpidx);
 144         if (vcpu && vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
 145                 return vcpu;
 146         kvm_for_each_vcpu(i, vcpu, kvm)
 147                 if (vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
 148                         return vcpu;
 149         return NULL;
 150 }
 151
 152 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vpidx)
 153 {
 154         struct kvm_vcpu *vcpu;
 155         struct kvm_vcpu_hv_synic *synic;
 156
 157         vcpu = get_vcpu_by_vpidx(kvm, vpidx);
 158         if (!vcpu)
 159                 return NULL;
 160         synic = vcpu_to_synic(vcpu);
 161         return (synic->active) ? synic : NULL;
 162 }
 163
 164 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
 165 {
 166         struct kvm *kvm = vcpu->kvm;
 167         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
 168         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 169         struct kvm_vcpu_hv_stimer *stimer;
 170         int gsi, idx;
 171
 172         trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
 173
 174         /* Try to deliver pending Hyper-V SynIC timers messages */
 175         for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
 176                 stimer = &hv_vcpu->stimer[idx];
 177                 if (stimer->msg_pending && stimer->config.enable &&
 178                     !stimer->config.direct_mode &&
 179                     stimer->config.sintx == sint)
 180                         stimer_mark_pending(stimer, false);
 181         }
 182
 183         idx = srcu_read_lock(&kvm->irq_srcu);
 184         gsi = atomic_read(&synic->sint_to_gsi[sint]);
 185         if (gsi != -1)
 186                 kvm_notify_acked_gsi(kvm, gsi);
 187         srcu_read_unlock(&kvm->irq_srcu, idx);
 188 }
 189
 190 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
 191 {
 192         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 193         struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
 194
 195         hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
 196         hv_vcpu->exit.u.synic.msr = msr;
 197         hv_vcpu->exit.u.synic.control = synic->control;
 198         hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
 199         hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
 200
 201         kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
 202 }
 203
 204 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
 205                          u32 msr, u64 data, bool host)
 206 {
 207         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 208         int ret;
 209
 210         if (!synic->active && !host)
 211                 return 1;
 212
 213         trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
 214
 215         ret = 0;
 216         switch (msr) {
 217         case HV_X64_MSR_SCONTROL:
 218                 synic->control = data;
 219                 if (!host)
 220                         synic_exit(synic, msr);
 221                 break;
 222         case HV_X64_MSR_SVERSION:
 223                 if (!host) {
 224                         ret = 1;
 225                         break;
 226                 }
 227                 synic->version = data;
 228                 break;
 229         case HV_X64_MSR_SIEFP:
 230                 if ((data & HV_SYNIC_SIEFP_ENABLE) && !host &&
 231                     !synic->dont_zero_synic_pages)
 232                         if (kvm_clear_guest(vcpu->kvm,
 233                                             data & PAGE_MASK, PAGE_SIZE)) {
 234                                 ret = 1;
 235                                 break;
 236                         }
 237                 synic->evt_page = data;
 238                 if (!host)
 239                         synic_exit(synic, msr);
 240                 break;
 241         case HV_X64_MSR_SIMP:
 242                 if ((data & HV_SYNIC_SIMP_ENABLE) && !host &&
 243                     !synic->dont_zero_synic_pages)
 244                         if (kvm_clear_guest(vcpu->kvm,
 245                                             data & PAGE_MASK, PAGE_SIZE)) {
 246                                 ret = 1;
 247                                 break;
 248                         }
 249                 synic->msg_page = data;
 250                 if (!host)
 251                         synic_exit(synic, msr);
 252                 break;
 253         case HV_X64_MSR_EOM: {
 254                 int i;
 255
 256                 for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
 257                         kvm_hv_notify_acked_sint(vcpu, i);
 258                 break;
 259         }
 260         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
 261                 ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
 262                 break;
 263         default:
 264                 ret = 1;
 265                 break;
 266         }
 267         return ret;
 268 }
 269
 270 static bool kvm_hv_is_syndbg_enabled(struct kvm_vcpu *vcpu)
 271 {
 272         struct kvm_cpuid_entry2 *entry;
 273
 274         entry = kvm_find_cpuid_entry(vcpu,
 275                                      HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES,
 276                                      0);
 277         if (!entry)
 278                 return false;
 279
 280         return entry->eax & HV_X64_SYNDBG_CAP_ALLOW_KERNEL_DEBUGGING;
 281 }
 282
 283 static int kvm_hv_syndbg_complete_userspace(struct kvm_vcpu *vcpu)
 284 {
 285         struct kvm *kvm = vcpu->kvm;
 286         struct kvm_hv *hv = &kvm->arch.hyperv;
 287
 288         if (vcpu->run->hyperv.u.syndbg.msr == HV_X64_MSR_SYNDBG_CONTROL)
 289                 hv->hv_syndbg.control.status =
 290                         vcpu->run->hyperv.u.syndbg.status;
 291         return 1;
 292 }
 293
 294 static void syndbg_exit(struct kvm_vcpu *vcpu, u32 msr)
 295 {
 296         struct kvm_hv_syndbg *syndbg = vcpu_to_hv_syndbg(vcpu);
 297         struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
 298
 299         hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNDBG;
 300         hv_vcpu->exit.u.syndbg.msr = msr;
 301         hv_vcpu->exit.u.syndbg.control = syndbg->control.control;
 302         hv_vcpu->exit.u.syndbg.send_page = syndbg->control.send_page;
 303         hv_vcpu->exit.u.syndbg.recv_page = syndbg->control.recv_page;
 304         hv_vcpu->exit.u.syndbg.pending_page = syndbg->control.pending_page;
 305         vcpu->arch.complete_userspace_io =
 306                         kvm_hv_syndbg_complete_userspace;
 307
 308         kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
 309 }
 310
 311 static int syndbg_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
 312 {
 313         struct kvm_hv_syndbg *syndbg = vcpu_to_hv_syndbg(vcpu);
 314
 315         if (!kvm_hv_is_syndbg_enabled(vcpu) && !host)
 316                 return 1;
 317
 318         trace_kvm_hv_syndbg_set_msr(vcpu->vcpu_id,
 319                                     vcpu_to_hv_vcpu(vcpu)->vp_index, msr, data);
 320         switch (msr) {
 321         case HV_X64_MSR_SYNDBG_CONTROL:
 322                 syndbg->control.control = data;
 323                 if (!host)
 324                         syndbg_exit(vcpu, msr);
 325                 break;
 326         case HV_X64_MSR_SYNDBG_STATUS:
 327                 syndbg->control.status = data;
 328                 break;
 329         case HV_X64_MSR_SYNDBG_SEND_BUFFER:
 330                 syndbg->control.send_page = data;
 331                 break;
 332         case HV_X64_MSR_SYNDBG_RECV_BUFFER:
 333                 syndbg->control.recv_page = data;
 334                 break;
 335         case HV_X64_MSR_SYNDBG_PENDING_BUFFER:
 336                 syndbg->control.pending_page = data;
 337                 if (!host)
 338                         syndbg_exit(vcpu, msr);
 339                 break;
 340         case HV_X64_MSR_SYNDBG_OPTIONS:
 341                 syndbg->options = data;
 342                 break;
 343         default:
 344                 break;
 345         }
 346
 347         return 0;
 348 }
 349
 350 static int syndbg_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
 351 {
 352         struct kvm_hv_syndbg *syndbg = vcpu_to_hv_syndbg(vcpu);
 353
 354         if (!kvm_hv_is_syndbg_enabled(vcpu) && !host)
 355                 return 1;
 356
 357         switch (msr) {
 358         case HV_X64_MSR_SYNDBG_CONTROL:
 359                 *pdata = syndbg->control.control;
 360                 break;
 361         case HV_X64_MSR_SYNDBG_STATUS:
 362                 *pdata = syndbg->control.status;
 363                 break;
 364         case HV_X64_MSR_SYNDBG_SEND_BUFFER:
 365                 *pdata = syndbg->control.send_page;
 366                 break;
 367         case HV_X64_MSR_SYNDBG_RECV_BUFFER:
 368                 *pdata = syndbg->control.recv_page;
 369                 break;
 370         case HV_X64_MSR_SYNDBG_PENDING_BUFFER:
 371                 *pdata = syndbg->control.pending_page;
 372                 break;
 373         case HV_X64_MSR_SYNDBG_OPTIONS:
 374                 *pdata = syndbg->options;
 375                 break;
 376         default:
 377                 break;
 378         }
 379
 380         trace_kvm_hv_syndbg_get_msr(vcpu->vcpu_id,
 381                                     vcpu_to_hv_vcpu(vcpu)->vp_index, msr,
 382                                     *pdata);
 383
 384         return 0;
 385 }
 386
 387 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata,
 388                          bool host)
 389 {
 390         int ret;
 391
 392         if (!synic->active && !host)
 393                 return 1;
 394
 395         ret = 0;
 396         switch (msr) {
 397         case HV_X64_MSR_SCONTROL:
 398                 *pdata = synic->control;
 399                 break;
 400         case HV_X64_MSR_SVERSION:
 401                 *pdata = synic->version;
 402                 break;
 403         case HV_X64_MSR_SIEFP:
 404                 *pdata = synic->evt_page;
 405                 break;
 406         case HV_X64_MSR_SIMP:
 407                 *pdata = synic->msg_page;
 408                 break;
 409         case HV_X64_MSR_EOM:
 410                 *pdata = 0;
 411                 break;
 412         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
 413                 *pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
 414                 break;
 415         default:
 416                 ret = 1;
 417                 break;
 418         }
 419         return ret;
 420 }
 421
 422 static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
 423 {
 424         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 425         struct kvm_lapic_irq irq;
 426         int ret, vector;
 427
 428         if (sint >= ARRAY_SIZE(synic->sint))
 429                 return -EINVAL;
 430
 431         vector = synic_get_sint_vector(synic_read_sint(synic, sint));
 432         if (vector < 0)
 433                 return -ENOENT;
 434
 435         memset(&irq, 0, sizeof(irq));
 436         irq.shorthand = APIC_DEST_SELF;
 437         irq.dest_mode = APIC_DEST_PHYSICAL;
 438         irq.delivery_mode = APIC_DM_FIXED;
 439         irq.vector = vector;
 440         irq.level = 1;
 441
 442         ret = kvm_irq_delivery_to_apic(vcpu->kvm, vcpu->arch.apic, &irq, NULL);
 443         trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
 444         return ret;
 445 }
 446
 447 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vpidx, u32 sint)
 448 {
 449         struct kvm_vcpu_hv_synic *synic;
 450
 451         synic = synic_get(kvm, vpidx);
 452         if (!synic)
 453                 return -EINVAL;
 454
 455         return synic_set_irq(synic, sint);
 456 }
 457
 458 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
 459 {
 460         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
 461         int i;
 462
 463         trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
 464
 465         for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
 466                 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
 467                         kvm_hv_notify_acked_sint(vcpu, i);
 468 }
 469
 470 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi)
 471 {
 472         struct kvm_vcpu_hv_synic *synic;
 473
 474         synic = synic_get(kvm, vpidx);
 475         if (!synic)
 476                 return -EINVAL;
 477
 478         if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
 479                 return -EINVAL;
 480
 481         atomic_set(&synic->sint_to_gsi[sint], gsi);
 482         return 0;
 483 }
 484
 485 void kvm_hv_irq_routing_update(struct kvm *kvm)
 486 {
 487         struct kvm_irq_routing_table *irq_rt;
 488         struct kvm_kernel_irq_routing_entry *e;
 489         u32 gsi;
 490
 491         irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
 492                                         lockdep_is_held(&kvm->irq_lock));
 493
 494         for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
 495                 hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
 496                         if (e->type == KVM_IRQ_ROUTING_HV_SINT)
 497                                 kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
 498                                                     e->hv_sint.sint, gsi);
 499                 }
 500         }
 501 }
 502
 503 static void synic_init(struct kvm_vcpu_hv_synic *synic)
 504 {
 505         int i;
 506
 507         memset(synic, 0, sizeof(*synic));
 508         synic->version = HV_SYNIC_VERSION_1;
 509         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
 510                 atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
 511                 atomic_set(&synic->sint_to_gsi[i], -1);
 512         }
 513 }
 514
 515 static u64 get_time_ref_counter(struct kvm *kvm)
 516 {
 517         struct kvm_hv *hv = &kvm->arch.hyperv;
 518         struct kvm_vcpu *vcpu;
 519         u64 tsc;
 520
 521         /*
 522          * The guest has not set up the TSC page or the clock isn't
 523          * stable, fall back to get_kvmclock_ns.
 524          */
 525         if (!hv->tsc_ref.tsc_sequence)
 526                 return div_u64(get_kvmclock_ns(kvm), 100);
 527
 528         vcpu = kvm_get_vcpu(kvm, 0);
 529         tsc = kvm_read_l1_tsc(vcpu, rdtsc());
 530         return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64)
 531                 + hv->tsc_ref.tsc_offset;
 532 }
 533
 534 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
 535                                 bool vcpu_kick)
 536 {
 537         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
 538
 539         set_bit(stimer->index,
 540                 vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
 541         kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
 542         if (vcpu_kick)
 543                 kvm_vcpu_kick(vcpu);
 544 }
 545
 546 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
 547 {
 548         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
 549
 550         trace_kvm_hv_stimer_cleanup(stimer_to_vcpu(stimer)->vcpu_id,
 551                                     stimer->index);
 552
 553         hrtimer_cancel(&stimer->timer);
 554         clear_bit(stimer->index,
 555                   vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
 556         stimer->msg_pending = false;
 557         stimer->exp_time = 0;
 558 }
 559
 560 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
 561 {
 562         struct kvm_vcpu_hv_stimer *stimer;
 563
 564         stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
 565         trace_kvm_hv_stimer_callback(stimer_to_vcpu(stimer)->vcpu_id,
 566                                      stimer->index);
 567         stimer_mark_pending(stimer, true);
 568
 569         return HRTIMER_NORESTART;
 570 }
 571
 572 /*
 573  * stimer_start() assumptions:
 574  * a) stimer->count is not equal to 0
 575  * b) stimer->config has HV_STIMER_ENABLE flag
 576  */
 577 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
 578 {
 579         u64 time_now;
 580         ktime_t ktime_now;
 581
 582         time_now = get_time_ref_counter(stimer_to_vcpu(stimer)->kvm);
 583         ktime_now = ktime_get();
 584
 585         if (stimer->config.periodic) {
 586                 if (stimer->exp_time) {
 587                         if (time_now >= stimer->exp_time) {
 588                                 u64 remainder;
 589
 590                                 div64_u64_rem(time_now - stimer->exp_time,
 591                                               stimer->count, &remainder);
 592                                 stimer->exp_time =
 593                                         time_now + (stimer->count - remainder);
 594                         }
 595                 } else
 596                         stimer->exp_time = time_now + stimer->count;
 597
 598                 trace_kvm_hv_stimer_start_periodic(
 599                                         stimer_to_vcpu(stimer)->vcpu_id,
 600                                         stimer->index,
 601                                         time_now, stimer->exp_time);
 602
 603                 hrtimer_start(&stimer->timer,
 604                               ktime_add_ns(ktime_now,
 605                                            100 * (stimer->exp_time - time_now)),
 606                               HRTIMER_MODE_ABS);
 607                 return 0;
 608         }
 609         stimer->exp_time = stimer->count;
 610         if (time_now >= stimer->count) {
 611                 /*
 612                  * Expire timer according to Hypervisor Top-Level Functional
 613                  * specification v4(15.3.1):
 614                  * "If a one shot is enabled and the specified count is in
 615                  * the past, it will expire immediately."
 616                  */
 617                 stimer_mark_pending(stimer, false);
 618                 return 0;
 619         }
 620
 621         trace_kvm_hv_stimer_start_one_shot(stimer_to_vcpu(stimer)->vcpu_id,
 622                                            stimer->index,
 623                                            time_now, stimer->count);
 624
 625         hrtimer_start(&stimer->timer,
 626                       ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)),
 627                       HRTIMER_MODE_ABS);
 628         return 0;
 629 }
 630
 631 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
 632                              bool host)
 633 {
 634         union hv_stimer_config new_config = {.as_uint64 = config},
 635                 old_config = {.as_uint64 = stimer->config.as_uint64};
 636
 637         trace_kvm_hv_stimer_set_config(stimer_to_vcpu(stimer)->vcpu_id,
 638                                        stimer->index, config, host);
 639
 640         stimer_cleanup(stimer);
 641         if (old_config.enable &&
 642             !new_config.direct_mode && new_config.sintx == 0)
 643                 new_config.enable = 0;
 644         stimer->config.as_uint64 = new_config.as_uint64;
 645
 646         if (stimer->config.enable)
 647                 stimer_mark_pending(stimer, false);
 648
 649         return 0;
 650 }
 651
 652 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
 653                             bool host)
 654 {
 655         trace_kvm_hv_stimer_set_count(stimer_to_vcpu(stimer)->vcpu_id,
 656                                       stimer->index, count, host);
 657
 658         stimer_cleanup(stimer);
 659         stimer->count = count;
 660         if (stimer->count == 0)
 661                 stimer->config.enable = 0;
 662         else if (stimer->config.auto_enable)
 663                 stimer->config.enable = 1;
 664
 665         if (stimer->config.enable)
 666                 stimer_mark_pending(stimer, false);
 667
 668         return 0;
 669 }
 670
 671 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
 672 {
 673         *pconfig = stimer->config.as_uint64;
 674         return 0;
 675 }
 676
 677 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
 678 {
 679         *pcount = stimer->count;
 680         return 0;
 681 }
 682
 683 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
 684                              struct hv_message *src_msg, bool no_retry)
 685 {
 686         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 687         int msg_off = offsetof(struct hv_message_page, sint_message[sint]);
 688         gfn_t msg_page_gfn;
 689         struct hv_message_header hv_hdr;
 690         int r;
 691
 692         if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
 693                 return -ENOENT;
 694
 695         msg_page_gfn = synic->msg_page >> PAGE_SHIFT;
 696
 697         /*
 698          * Strictly following the spec-mandated ordering would assume setting
 699          * .msg_pending before checking .message_type.  However, this function
 700          * is only called in vcpu context so the entire update is atomic from
 701          * guest POV and thus the exact order here doesn't matter.
 702          */
 703         r = kvm_vcpu_read_guest_page(vcpu, msg_page_gfn, &hv_hdr.message_type,
 704                                      msg_off + offsetof(struct hv_message,
 705                                                         header.message_type),
 706                                      sizeof(hv_hdr.message_type));
 707         if (r < 0)
 708                 return r;
 709
 710         if (hv_hdr.message_type != HVMSG_NONE) {
 711                 if (no_retry)
 712                         return 0;
 713
 714                 hv_hdr.message_flags.msg_pending = 1;
 715                 r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn,
 716                                               &hv_hdr.message_flags,
 717                                               msg_off +
 718                                               offsetof(struct hv_message,
 719                                                        header.message_flags),
 720                                               sizeof(hv_hdr.message_flags));
 721                 if (r < 0)
 722                         return r;
 723                 return -EAGAIN;
 724         }
 725
 726         r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn, src_msg, msg_off,
 727                                       sizeof(src_msg->header) +
 728                                       src_msg->header.payload_size);
 729         if (r < 0)
 730                 return r;
 731
 732         r = synic_set_irq(synic, sint);
 733         if (r < 0)
 734                 return r;
 735         if (r == 0)
 736                 return -EFAULT;
 737         return 0;
 738 }
 739
 740 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
 741 {
 742         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
 743         struct hv_message *msg = &stimer->msg;
 744         struct hv_timer_message_payload *payload =
 745                         (struct hv_timer_message_payload *)&msg->u.payload;
 746
 747         /*
 748          * To avoid piling up periodic ticks, don't retry message
 749          * delivery for them (within "lazy" lost ticks policy).
 750          */
 751         bool no_retry = stimer->config.periodic;
 752
 753         payload->expiration_time = stimer->exp_time;
 754         payload->delivery_time = get_time_ref_counter(vcpu->kvm);
 755         return synic_deliver_msg(vcpu_to_synic(vcpu),
 756                                  stimer->config.sintx, msg,
 757                                  no_retry);
 758 }
 759
 760 static int stimer_notify_direct(struct kvm_vcpu_hv_stimer *stimer)
 761 {
 762         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
 763         struct kvm_lapic_irq irq = {
 764                 .delivery_mode = APIC_DM_FIXED,
 765                 .vector = stimer->config.apic_vector
 766         };
 767
 768         if (lapic_in_kernel(vcpu))
 769                 return !kvm_apic_set_irq(vcpu, &irq, NULL);
 770         return 0;
 771 }
 772
 773 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
 774 {
 775         int r, direct = stimer->config.direct_mode;
 776
 777         stimer->msg_pending = true;
 778         if (!direct)
 779                 r = stimer_send_msg(stimer);
 780         else
 781                 r = stimer_notify_direct(stimer);
 782         trace_kvm_hv_stimer_expiration(stimer_to_vcpu(stimer)->vcpu_id,
 783                                        stimer->index, direct, r);
 784         if (!r) {
 785                 stimer->msg_pending = false;
 786                 if (!(stimer->config.periodic))
 787                         stimer->config.enable = 0;
 788         }
 789 }
 790
 791 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
 792 {
 793         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 794         struct kvm_vcpu_hv_stimer *stimer;
 795         u64 time_now, exp_time;
 796         int i;
 797
 798         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 799                 if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
 800                         stimer = &hv_vcpu->stimer[i];
 801                         if (stimer->config.enable) {
 802                                 exp_time = stimer->exp_time;
 803
 804                                 if (exp_time) {
 805                                         time_now =
 806                                                 get_time_ref_counter(vcpu->kvm);
 807                                         if (time_now >= exp_time)
 808                                                 stimer_expiration(stimer);
 809                                 }
 810
 811                                 if ((stimer->config.enable) &&
 812                                     stimer->count) {
 813                                         if (!stimer->msg_pending)
 814                                                 stimer_start(stimer);
 815                                 } else
 816                                         stimer_cleanup(stimer);
 817                         }
 818                 }
 819 }
 820
 821 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
 822 {
 823         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 824         int i;
 825
 826         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 827                 stimer_cleanup(&hv_vcpu->stimer[i]);
 828 }
 829
 830 bool kvm_hv_assist_page_enabled(struct kvm_vcpu *vcpu)
 831 {
 832         if (!(vcpu->arch.hyperv.hv_vapic & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE))
 833                 return false;
 834         return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
 835 }
 836 EXPORT_SYMBOL_GPL(kvm_hv_assist_page_enabled);
 837
 838 bool kvm_hv_get_assist_page(struct kvm_vcpu *vcpu,
 839                             struct hv_vp_assist_page *assist_page)
 840 {
 841         if (!kvm_hv_assist_page_enabled(vcpu))
 842                 return false;
 843         return !kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data,
 844                                       assist_page, sizeof(*assist_page));
 845 }
 846 EXPORT_SYMBOL_GPL(kvm_hv_get_assist_page);
 847
 848 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
 849 {
 850         struct hv_message *msg = &stimer->msg;
 851         struct hv_timer_message_payload *payload =
 852                         (struct hv_timer_message_payload *)&msg->u.payload;
 853
 854         memset(&msg->header, 0, sizeof(msg->header));
 855         msg->header.message_type = HVMSG_TIMER_EXPIRED;
 856         msg->header.payload_size = sizeof(*payload);
 857
 858         payload->timer_index = stimer->index;
 859         payload->expiration_time = 0;
 860         payload->delivery_time = 0;
 861 }
 862
 863 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
 864 {
 865         memset(stimer, 0, sizeof(*stimer));
 866         stimer->index = timer_index;
 867         hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
 868         stimer->timer.function = stimer_timer_callback;
 869         stimer_prepare_msg(stimer);
 870 }
 871
 872 void kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
 873 {
 874         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 875         int i;
 876
 877         synic_init(&hv_vcpu->synic);
 878
 879         bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
 880         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 881                 stimer_init(&hv_vcpu->stimer[i], i);
 882 }
 883
 884 void kvm_hv_vcpu_postcreate(struct kvm_vcpu *vcpu)
 885 {
 886         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 887
 888         hv_vcpu->vp_index = kvm_vcpu_get_idx(vcpu);
 889 }
 890
 891 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
 892 {
 893         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
 894
 895         /*
 896          * Hyper-V SynIC auto EOI SINT's are
 897          * not compatible with APICV, so request
 898          * to deactivate APICV permanently.
 899          */
 900         kvm_request_apicv_update(vcpu->kvm, false, APICV_INHIBIT_REASON_HYPERV);
 901         synic->active = true;
 902         synic->dont_zero_synic_pages = dont_zero_synic_pages;
 903         return 0;
 904 }
 905
 906 static bool kvm_hv_msr_partition_wide(u32 msr)
 907 {
 908         bool r = false;
 909
 910         switch (msr) {
 911         case HV_X64_MSR_GUEST_OS_ID:
 912         case HV_X64_MSR_HYPERCALL:
 913         case HV_X64_MSR_REFERENCE_TSC:
 914         case HV_X64_MSR_TIME_REF_COUNT:
 915         case HV_X64_MSR_CRASH_CTL:
 916         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
 917         case HV_X64_MSR_RESET:
 918         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
 919         case HV_X64_MSR_TSC_EMULATION_CONTROL:
 920         case HV_X64_MSR_TSC_EMULATION_STATUS:
 921         case HV_X64_MSR_SYNDBG_OPTIONS:
 922         case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
 923                 r = true;
 924                 break;
 925         }
 926
 927         return r;
 928 }
 929
 930 static int kvm_hv_msr_get_crash_data(struct kvm_vcpu *vcpu,
 931                                      u32 index, u64 *pdata)
 932 {
 933         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 934         size_t size = ARRAY_SIZE(hv->hv_crash_param);
 935
 936         if (WARN_ON_ONCE(index >= size))
 937                 return -EINVAL;
 938
 939         *pdata = hv->hv_crash_param[array_index_nospec(index, size)];
 940         return 0;
 941 }
 942
 943 static int kvm_hv_msr_get_crash_ctl(struct kvm_vcpu *vcpu, u64 *pdata)
 944 {
 945         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 946
 947         *pdata = hv->hv_crash_ctl;
 948         return 0;
 949 }
 950
 951 static int kvm_hv_msr_set_crash_ctl(struct kvm_vcpu *vcpu, u64 data, bool host)
 952 {
 953         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 954
 955         if (host)
 956                 hv->hv_crash_ctl = data & HV_CRASH_CTL_CRASH_NOTIFY;
 957
 958         if (!host && (data & HV_CRASH_CTL_CRASH_NOTIFY)) {
 959
 960                 vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
 961                           hv->hv_crash_param[0],
 962                           hv->hv_crash_param[1],
 963                           hv->hv_crash_param[2],
 964                           hv->hv_crash_param[3],
 965                           hv->hv_crash_param[4]);
 966
 967                 /* Send notification about crash to user space */
 968                 kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
 969         }
 970
 971         return 0;
 972 }
 973
 974 static int kvm_hv_msr_set_crash_data(struct kvm_vcpu *vcpu,
 975                                      u32 index, u64 data)
 976 {
 977         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 978         size_t size = ARRAY_SIZE(hv->hv_crash_param);
 979
 980         if (WARN_ON_ONCE(index >= size))
 981                 return -EINVAL;
 982
 983         hv->hv_crash_param[array_index_nospec(index, size)] = data;
 984         return 0;
 985 }
 986
 987 /*
 988  * The kvmclock and Hyper-V TSC page use similar formulas, and converting
 989  * between them is possible:
 990  *
 991  * kvmclock formula:
 992  *    nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
 993  *           + system_time
 994  *
 995  * Hyper-V formula:
 996  *    nsec/100 = ticks * scale / 2^64 + offset
 997  *
 998  * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
 999  * By dividing the kvmclock formula by 100 and equating what's left we get:
1000  *    ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1001  *            scale / 2^64 =         tsc_to_system_mul * 2^(tsc_shift-32) / 100
1002  *            scale        =         tsc_to_system_mul * 2^(32+tsc_shift) / 100
1003  *
1004  * Now expand the kvmclock formula and divide by 100:
1005  *    nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
1006  *           - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
1007  *           + system_time
1008  *    nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1009  *               - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1010  *               + system_time / 100
1011  *
1012  * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
1013  *    nsec/100 = ticks * scale / 2^64
1014  *               - tsc_timestamp * scale / 2^64
1015  *               + system_time / 100
1016  *
1017  * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
1018  *    offset = system_time / 100 - tsc_timestamp * scale / 2^64
1019  *
1020  * These two equivalencies are implemented in this function.
1021  */
1022 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
1023                                         HV_REFERENCE_TSC_PAGE *tsc_ref)
1024 {
1025         u64 max_mul;
1026
1027         if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
1028                 return false;
1029
1030         /*
1031          * check if scale would overflow, if so we use the time ref counter
1032          *    tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
1033          *    tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
1034          *    tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
1035          */
1036         max_mul = 100ull << (32 - hv_clock->tsc_shift);
1037         if (hv_clock->tsc_to_system_mul >= max_mul)
1038                 return false;
1039
1040         /*
1041          * Otherwise compute the scale and offset according to the formulas
1042          * derived above.
1043          */
1044         tsc_ref->tsc_scale =
1045                 mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
1046                                 hv_clock->tsc_to_system_mul,
1047                                 100);
1048
1049         tsc_ref->tsc_offset = hv_clock->system_time;
1050         do_div(tsc_ref->tsc_offset, 100);
1051         tsc_ref->tsc_offset -=
1052                 mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
1053         return true;
1054 }
1055
1056 void kvm_hv_setup_tsc_page(struct kvm *kvm,
1057                            struct pvclock_vcpu_time_info *hv_clock)
1058 {
1059         struct kvm_hv *hv = &kvm->arch.hyperv;
1060         u32 tsc_seq;
1061         u64 gfn;
1062
1063         BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
1064         BUILD_BUG_ON(offsetof(HV_REFERENCE_TSC_PAGE, tsc_sequence) != 0);
1065
1066         if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1067                 return;
1068
1069         mutex_lock(&kvm->arch.hyperv.hv_lock);
1070         if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1071                 goto out_unlock;
1072
1073         gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
1074         /*
1075          * Because the TSC parameters only vary when there is a
1076          * change in the master clock, do not bother with caching.
1077          */
1078         if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
1079                                     &tsc_seq, sizeof(tsc_seq))))
1080                 goto out_unlock;
1081
1082         /*
1083          * While we're computing and writing the parameters, force the
1084          * guest to use the time reference count MSR.
1085          */
1086         hv->tsc_ref.tsc_sequence = 0;
1087         if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
1088                             &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
1089                 goto out_unlock;
1090
1091         if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
1092                 goto out_unlock;
1093
1094         /* Ensure sequence is zero before writing the rest of the struct.  */
1095         smp_wmb();
1096         if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
1097                 goto out_unlock;
1098
1099         /*
1100          * Now switch to the TSC page mechanism by writing the sequence.
1101          */
1102         tsc_seq++;
1103         if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
1104                 tsc_seq = 1;
1105
1106         /* Write the struct entirely before the non-zero sequence.  */
1107         smp_wmb();
1108
1109         hv->tsc_ref.tsc_sequence = tsc_seq;
1110         kvm_write_guest(kvm, gfn_to_gpa(gfn),
1111                         &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence));
1112 out_unlock:
1113         mutex_unlock(&kvm->arch.hyperv.hv_lock);
1114 }
1115
1116 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
1117                              bool host)
1118 {
1119         struct kvm *kvm = vcpu->kvm;
1120         struct kvm_hv *hv = &kvm->arch.hyperv;
1121
1122         switch (msr) {
1123         case HV_X64_MSR_GUEST_OS_ID:
1124                 hv->hv_guest_os_id = data;
1125                 /* setting guest os id to zero disables hypercall page */
1126                 if (!hv->hv_guest_os_id)
1127                         hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1128                 break;
1129         case HV_X64_MSR_HYPERCALL: {
1130                 u64 gfn;
1131                 unsigned long addr;
1132                 u8 instructions[4];
1133
1134                 /* if guest os id is not set hypercall should remain disabled */
1135                 if (!hv->hv_guest_os_id)
1136                         break;
1137                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1138                         hv->hv_hypercall = data;
1139                         break;
1140                 }
1141                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1142                 addr = gfn_to_hva(kvm, gfn);
1143                 if (kvm_is_error_hva(addr))
1144                         return 1;
1145                 kvm_x86_ops.patch_hypercall(vcpu, instructions);
1146                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1147                 if (__copy_to_user((void __user *)addr, instructions, 4))
1148                         return 1;
1149                 hv->hv_hypercall = data;
1150                 mark_page_dirty(kvm, gfn);
1151                 break;
1152         }
1153         case HV_X64_MSR_REFERENCE_TSC:
1154                 hv->hv_tsc_page = data;
1155                 if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE)
1156                         kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1157                 break;
1158         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1159                 return kvm_hv_msr_set_crash_data(vcpu,
1160                                                  msr - HV_X64_MSR_CRASH_P0,
1161                                                  data);
1162         case HV_X64_MSR_CRASH_CTL:
1163                 return kvm_hv_msr_set_crash_ctl(vcpu, data, host);
1164         case HV_X64_MSR_RESET:
1165                 if (data == 1) {
1166                         vcpu_debug(vcpu, "hyper-v reset requested\n");
1167                         kvm_make_request(KVM_REQ_HV_RESET, vcpu);
1168                 }
1169                 break;
1170         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1171                 hv->hv_reenlightenment_control = data;
1172                 break;
1173         case HV_X64_MSR_TSC_EMULATION_CONTROL:
1174                 hv->hv_tsc_emulation_control = data;
1175                 break;
1176         case HV_X64_MSR_TSC_EMULATION_STATUS:
1177                 hv->hv_tsc_emulation_status = data;
1178                 break;
1179         case HV_X64_MSR_TIME_REF_COUNT:
1180                 /* read-only, but still ignore it if host-initiated */
1181                 if (!host)
1182                         return 1;
1183                 break;
1184         case HV_X64_MSR_SYNDBG_OPTIONS:
1185         case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1186                 return syndbg_set_msr(vcpu, msr, data, host);
1187         default:
1188                 vcpu_unimpl(vcpu, "Hyper-V unhandled wrmsr: 0x%x data 0x%llx\n",
1189                             msr, data);
1190                 return 1;
1191         }
1192         return 0;
1193 }
1194
1195 /* Calculate cpu time spent by current task in 100ns units */
1196 static u64 current_task_runtime_100ns(void)
1197 {
1198         u64 utime, stime;
1199
1200         task_cputime_adjusted(current, &utime, &stime);
1201
1202         return div_u64(utime + stime, 100);
1203 }
1204
1205 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1206 {
1207         struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
1208
1209         switch (msr) {
1210         case HV_X64_MSR_VP_INDEX: {
1211                 struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
1212                 int vcpu_idx = kvm_vcpu_get_idx(vcpu);
1213                 u32 new_vp_index = (u32)data;
1214
1215                 if (!host || new_vp_index >= KVM_MAX_VCPUS)
1216                         return 1;
1217
1218                 if (new_vp_index == hv_vcpu->vp_index)
1219                         return 0;
1220
1221                 /*
1222                  * The VP index is initialized to vcpu_index by
1223                  * kvm_hv_vcpu_postcreate so they initially match.  Now the
1224                  * VP index is changing, adjust num_mismatched_vp_indexes if
1225                  * it now matches or no longer matches vcpu_idx.
1226                  */
1227                 if (hv_vcpu->vp_index == vcpu_idx)
1228                         atomic_inc(&hv->num_mismatched_vp_indexes);
1229                 else if (new_vp_index == vcpu_idx)
1230                         atomic_dec(&hv->num_mismatched_vp_indexes);
1231
1232                 hv_vcpu->vp_index = new_vp_index;
1233                 break;
1234         }
1235         case HV_X64_MSR_VP_ASSIST_PAGE: {
1236                 u64 gfn;
1237                 unsigned long addr;
1238
1239                 if (!(data & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE)) {
1240                         hv_vcpu->hv_vapic = data;
1241                         if (kvm_lapic_enable_pv_eoi(vcpu, 0, 0))
1242                                 return 1;
1243                         break;
1244                 }
1245                 gfn = data >> HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT;
1246                 addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1247                 if (kvm_is_error_hva(addr))
1248                         return 1;
1249
1250                 /*
1251                  * Clear apic_assist portion of struct hv_vp_assist_page
1252                  * only, there can be valuable data in the rest which needs
1253                  * to be preserved e.g. on migration.
1254                  */
1255                 if (__clear_user((void __user *)addr, sizeof(u32)))
1256                         return 1;
1257                 hv_vcpu->hv_vapic = data;
1258                 kvm_vcpu_mark_page_dirty(vcpu, gfn);
1259                 if (kvm_lapic_enable_pv_eoi(vcpu,
1260                                             gfn_to_gpa(gfn) | KVM_MSR_ENABLED,
1261                                             sizeof(struct hv_vp_assist_page)))
1262                         return 1;
1263                 break;
1264         }
1265         case HV_X64_MSR_EOI:
1266                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1267         case HV_X64_MSR_ICR:
1268                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1269         case HV_X64_MSR_TPR:
1270                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1271         case HV_X64_MSR_VP_RUNTIME:
1272                 if (!host)
1273                         return 1;
1274                 hv_vcpu->runtime_offset = data - current_task_runtime_100ns();
1275                 break;
1276         case HV_X64_MSR_SCONTROL:
1277         case HV_X64_MSR_SVERSION:
1278         case HV_X64_MSR_SIEFP:
1279         case HV_X64_MSR_SIMP:
1280         case HV_X64_MSR_EOM:
1281         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1282                 return synic_set_msr(vcpu_to_synic(vcpu), msr, data, host);
1283         case HV_X64_MSR_STIMER0_CONFIG:
1284         case HV_X64_MSR_STIMER1_CONFIG:
1285         case HV_X64_MSR_STIMER2_CONFIG:
1286         case HV_X64_MSR_STIMER3_CONFIG: {
1287                 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1288
1289                 return stimer_set_config(vcpu_to_stimer(vcpu, timer_index),
1290                                          data, host);
1291         }
1292         case HV_X64_MSR_STIMER0_COUNT:
1293         case HV_X64_MSR_STIMER1_COUNT:
1294         case HV_X64_MSR_STIMER2_COUNT:
1295         case HV_X64_MSR_STIMER3_COUNT: {
1296                 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1297
1298                 return stimer_set_count(vcpu_to_stimer(vcpu, timer_index),
1299                                         data, host);
1300         }
1301         case HV_X64_MSR_TSC_FREQUENCY:
1302         case HV_X64_MSR_APIC_FREQUENCY:
1303                 /* read-only, but still ignore it if host-initiated */
1304                 if (!host)
1305                         return 1;
1306                 break;
1307         default:
1308                 vcpu_unimpl(vcpu, "Hyper-V unhandled wrmsr: 0x%x data 0x%llx\n",
1309                             msr, data);
1310                 return 1;
1311         }
1312
1313         return 0;
1314 }
1315
1316 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1317                              bool host)
1318 {
1319         u64 data = 0;
1320         struct kvm *kvm = vcpu->kvm;
1321         struct kvm_hv *hv = &kvm->arch.hyperv;
1322
1323         switch (msr) {
1324         case HV_X64_MSR_GUEST_OS_ID:
1325                 data = hv->hv_guest_os_id;
1326                 break;
1327         case HV_X64_MSR_HYPERCALL:
1328                 data = hv->hv_hypercall;
1329                 break;
1330         case HV_X64_MSR_TIME_REF_COUNT:
1331                 data = get_time_ref_counter(kvm);
1332                 break;
1333         case HV_X64_MSR_REFERENCE_TSC:
1334                 data = hv->hv_tsc_page;
1335                 break;
1336         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1337                 return kvm_hv_msr_get_crash_data(vcpu,
1338                                                  msr - HV_X64_MSR_CRASH_P0,
1339                                                  pdata);
1340         case HV_X64_MSR_CRASH_CTL:
1341                 return kvm_hv_msr_get_crash_ctl(vcpu, pdata);
1342         case HV_X64_MSR_RESET:
1343                 data = 0;
1344                 break;
1345         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1346                 data = hv->hv_reenlightenment_control;
1347                 break;
1348         case HV_X64_MSR_TSC_EMULATION_CONTROL:
1349                 data = hv->hv_tsc_emulation_control;
1350                 break;
1351         case HV_X64_MSR_TSC_EMULATION_STATUS:
1352                 data = hv->hv_tsc_emulation_status;
1353                 break;
1354         case HV_X64_MSR_SYNDBG_OPTIONS:
1355         case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1356                 return syndbg_get_msr(vcpu, msr, pdata, host);
1357         default:
1358                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1359                 return 1;
1360         }
1361
1362         *pdata = data;
1363         return 0;
1364 }
1365
1366 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1367                           bool host)
1368 {
1369         u64 data = 0;
1370         struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
1371
1372         switch (msr) {
1373         case HV_X64_MSR_VP_INDEX:
1374                 data = hv_vcpu->vp_index;
1375                 break;
1376         case HV_X64_MSR_EOI:
1377                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1378         case HV_X64_MSR_ICR:
1379                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1380         case HV_X64_MSR_TPR:
1381                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1382         case HV_X64_MSR_VP_ASSIST_PAGE:
1383                 data = hv_vcpu->hv_vapic;
1384                 break;
1385         case HV_X64_MSR_VP_RUNTIME:
1386                 data = current_task_runtime_100ns() + hv_vcpu->runtime_offset;
1387                 break;
1388         case HV_X64_MSR_SCONTROL:
1389         case HV_X64_MSR_SVERSION:
1390         case HV_X64_MSR_SIEFP:
1391         case HV_X64_MSR_SIMP:
1392         case HV_X64_MSR_EOM:
1393         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1394                 return synic_get_msr(vcpu_to_synic(vcpu), msr, pdata, host);
1395         case HV_X64_MSR_STIMER0_CONFIG:
1396         case HV_X64_MSR_STIMER1_CONFIG:
1397         case HV_X64_MSR_STIMER2_CONFIG:
1398         case HV_X64_MSR_STIMER3_CONFIG: {
1399                 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1400
1401                 return stimer_get_config(vcpu_to_stimer(vcpu, timer_index),
1402                                          pdata);
1403         }
1404         case HV_X64_MSR_STIMER0_COUNT:
1405         case HV_X64_MSR_STIMER1_COUNT:
1406         case HV_X64_MSR_STIMER2_COUNT:
1407         case HV_X64_MSR_STIMER3_COUNT: {
1408                 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1409
1410                 return stimer_get_count(vcpu_to_stimer(vcpu, timer_index),
1411                                         pdata);
1412         }
1413         case HV_X64_MSR_TSC_FREQUENCY:
1414                 data = (u64)vcpu->arch.virtual_tsc_khz * 1000;
1415                 break;
1416         case HV_X64_MSR_APIC_FREQUENCY:
1417                 data = APIC_BUS_FREQUENCY;
1418                 break;
1419         default:
1420                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1421                 return 1;
1422         }
1423         *pdata = data;
1424         return 0;
1425 }
1426
1427 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1428 {
1429         if (kvm_hv_msr_partition_wide(msr)) {
1430                 int r;
1431
1432                 mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1433                 r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1434                 mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1435                 return r;
1436         } else
1437                 return kvm_hv_set_msr(vcpu, msr, data, host);
1438 }
1439
1440 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
1441 {
1442         if (kvm_hv_msr_partition_wide(msr)) {
1443                 int r;
1444
1445                 mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1446                 r = kvm_hv_get_msr_pw(vcpu, msr, pdata, host);
1447                 mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1448                 return r;
1449         } else
1450                 return kvm_hv_get_msr(vcpu, msr, pdata, host);
1451 }
1452
1453 static __always_inline unsigned long *sparse_set_to_vcpu_mask(
1454         struct kvm *kvm, u64 *sparse_banks, u64 valid_bank_mask,
1455         u64 *vp_bitmap, unsigned long *vcpu_bitmap)
1456 {
1457         struct kvm_hv *hv = &kvm->arch.hyperv;
1458         struct kvm_vcpu *vcpu;
1459         int i, bank, sbank = 0;
1460
1461         memset(vp_bitmap, 0,
1462                KVM_HV_MAX_SPARSE_VCPU_SET_BITS * sizeof(*vp_bitmap));
1463         for_each_set_bit(bank, (unsigned long *)&valid_bank_mask,
1464                          KVM_HV_MAX_SPARSE_VCPU_SET_BITS)
1465                 vp_bitmap[bank] = sparse_banks[sbank++];
1466
1467         if (likely(!atomic_read(&hv->num_mismatched_vp_indexes))) {
1468                 /* for all vcpus vp_index == vcpu_idx */
1469                 return (unsigned long *)vp_bitmap;
1470         }
1471
1472         bitmap_zero(vcpu_bitmap, KVM_MAX_VCPUS);
1473         kvm_for_each_vcpu(i, vcpu, kvm) {
1474                 if (test_bit(vcpu_to_hv_vcpu(vcpu)->vp_index,
1475                              (unsigned long *)vp_bitmap))
1476                         __set_bit(i, vcpu_bitmap);
1477         }
1478         return vcpu_bitmap;
1479 }
1480
1481 static u64 kvm_hv_flush_tlb(struct kvm_vcpu *current_vcpu, u64 ingpa,
1482                             u16 rep_cnt, bool ex)
1483 {
1484         struct kvm *kvm = current_vcpu->kvm;
1485         struct kvm_vcpu_hv *hv_vcpu = &current_vcpu->arch.hyperv;
1486         struct hv_tlb_flush_ex flush_ex;
1487         struct hv_tlb_flush flush;
1488         u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1489         DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1490         unsigned long *vcpu_mask;
1491         u64 valid_bank_mask;
1492         u64 sparse_banks[64];
1493         int sparse_banks_len;
1494         bool all_cpus;
1495
1496         if (!ex) {
1497                 if (unlikely(kvm_read_guest(kvm, ingpa, &flush, sizeof(flush))))
1498                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
1499
1500                 trace_kvm_hv_flush_tlb(flush.processor_mask,
1501                                        flush.address_space, flush.flags);
1502
1503                 valid_bank_mask = BIT_ULL(0);
1504                 sparse_banks[0] = flush.processor_mask;
1505
1506                 /*
1507                  * Work around possible WS2012 bug: it sends hypercalls
1508                  * with processor_mask = 0x0 and HV_FLUSH_ALL_PROCESSORS clear,
1509                  * while also expecting us to flush something and crashing if
1510                  * we don't. Let's treat processor_mask == 0 same as
1511                  * HV_FLUSH_ALL_PROCESSORS.
1512                  */
1513                 all_cpus = (flush.flags & HV_FLUSH_ALL_PROCESSORS) ||
1514                         flush.processor_mask == 0;
1515         } else {
1516                 if (unlikely(kvm_read_guest(kvm, ingpa, &flush_ex,
1517                                             sizeof(flush_ex))))
1518                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
1519
1520                 trace_kvm_hv_flush_tlb_ex(flush_ex.hv_vp_set.valid_bank_mask,
1521                                           flush_ex.hv_vp_set.format,
1522                                           flush_ex.address_space,
1523                                           flush_ex.flags);
1524
1525                 valid_bank_mask = flush_ex.hv_vp_set.valid_bank_mask;
1526                 all_cpus = flush_ex.hv_vp_set.format !=
1527                         HV_GENERIC_SET_SPARSE_4K;
1528
1529                 sparse_banks_len =
1530                         bitmap_weight((unsigned long *)&valid_bank_mask, 64) *
1531                         sizeof(sparse_banks[0]);
1532
1533                 if (!sparse_banks_len && !all_cpus)
1534                         goto ret_success;
1535
1536                 if (!all_cpus &&
1537                     kvm_read_guest(kvm,
1538                                    ingpa + offsetof(struct hv_tlb_flush_ex,
1539                                                     hv_vp_set.bank_contents),
1540                                    sparse_banks,
1541                                    sparse_banks_len))
1542                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
1543         }
1544
1545         cpumask_clear(&hv_vcpu->tlb_flush);
1546
1547         vcpu_mask = all_cpus ? NULL :
1548                 sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1549                                         vp_bitmap, vcpu_bitmap);
1550
1551         /*
1552          * vcpu->arch.cr3 may not be up-to-date for running vCPUs so we can't
1553          * analyze it here, flush TLB regardless of the specified address space.
1554          */
1555         kvm_make_vcpus_request_mask(kvm, KVM_REQ_HV_TLB_FLUSH,
1556                                     NULL, vcpu_mask, &hv_vcpu->tlb_flush);
1557
1558 ret_success:
1559         /* We always do full TLB flush, set rep_done = rep_cnt. */
1560         return (u64)HV_STATUS_SUCCESS |
1561                 ((u64)rep_cnt << HV_HYPERCALL_REP_COMP_OFFSET);
1562 }
1563
1564 static void kvm_send_ipi_to_many(struct kvm *kvm, u32 vector,
1565                                  unsigned long *vcpu_bitmap)
1566 {
1567         struct kvm_lapic_irq irq = {
1568                 .delivery_mode = APIC_DM_FIXED,
1569                 .vector = vector
1570         };
1571         struct kvm_vcpu *vcpu;
1572         int i;
1573
1574         kvm_for_each_vcpu(i, vcpu, kvm) {
1575                 if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
1576                         continue;
1577
1578                 /* We fail only when APIC is disabled */
1579                 kvm_apic_set_irq(vcpu, &irq, NULL);
1580         }
1581 }
1582
1583 static u64 kvm_hv_send_ipi(struct kvm_vcpu *current_vcpu, u64 ingpa, u64 outgpa,
1584                            bool ex, bool fast)
1585 {
1586         struct kvm *kvm = current_vcpu->kvm;
1587         struct hv_send_ipi_ex send_ipi_ex;
1588         struct hv_send_ipi send_ipi;
1589         u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1590         DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1591         unsigned long *vcpu_mask;
1592         unsigned long valid_bank_mask;
1593         u64 sparse_banks[64];
1594         int sparse_banks_len;
1595         u32 vector;
1596         bool all_cpus;
1597
1598         if (!ex) {
1599                 if (!fast) {
1600                         if (unlikely(kvm_read_guest(kvm, ingpa, &send_ipi,
1601                                                     sizeof(send_ipi))))
1602                                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1603                         sparse_banks[0] = send_ipi.cpu_mask;
1604                         vector = send_ipi.vector;
1605                 } else {
1606                         /* 'reserved' part of hv_send_ipi should be 0 */
1607                         if (unlikely(ingpa >> 32 != 0))
1608                                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1609                         sparse_banks[0] = outgpa;
1610                         vector = (u32)ingpa;
1611                 }
1612                 all_cpus = false;
1613                 valid_bank_mask = BIT_ULL(0);
1614
1615                 trace_kvm_hv_send_ipi(vector, sparse_banks[0]);
1616         } else {
1617                 if (unlikely(kvm_read_guest(kvm, ingpa, &send_ipi_ex,
1618                                             sizeof(send_ipi_ex))))
1619                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
1620
1621                 trace_kvm_hv_send_ipi_ex(send_ipi_ex.vector,
1622                                          send_ipi_ex.vp_set.format,
1623                                          send_ipi_ex.vp_set.valid_bank_mask);
1624
1625                 vector = send_ipi_ex.vector;
1626                 valid_bank_mask = send_ipi_ex.vp_set.valid_bank_mask;
1627                 sparse_banks_len = bitmap_weight(&valid_bank_mask, 64) *
1628                         sizeof(sparse_banks[0]);
1629
1630                 all_cpus = send_ipi_ex.vp_set.format == HV_GENERIC_SET_ALL;
1631
1632                 if (!sparse_banks_len)
1633                         goto ret_success;
1634
1635                 if (!all_cpus &&
1636                     kvm_read_guest(kvm,
1637                                    ingpa + offsetof(struct hv_send_ipi_ex,
1638                                                     vp_set.bank_contents),
1639                                    sparse_banks,
1640                                    sparse_banks_len))
1641                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
1642         }
1643
1644         if ((vector < HV_IPI_LOW_VECTOR) || (vector > HV_IPI_HIGH_VECTOR))
1645                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1646
1647         vcpu_mask = all_cpus ? NULL :
1648                 sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1649                                         vp_bitmap, vcpu_bitmap);
1650
1651         kvm_send_ipi_to_many(kvm, vector, vcpu_mask);
1652
1653 ret_success:
1654         return HV_STATUS_SUCCESS;
1655 }
1656
1657 bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1658 {
1659         return READ_ONCE(kvm->arch.hyperv.hv_hypercall) & HV_X64_MSR_HYPERCALL_ENABLE;
1660 }
1661
1662 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
1663 {
1664         bool longmode;
1665
1666         longmode = is_64_bit_mode(vcpu);
1667         if (longmode)
1668                 kvm_rax_write(vcpu, result);
1669         else {
1670                 kvm_rdx_write(vcpu, result >> 32);
1671                 kvm_rax_write(vcpu, result & 0xffffffff);
1672         }
1673 }
1674
1675 static int kvm_hv_hypercall_complete(struct kvm_vcpu *vcpu, u64 result)
1676 {
1677         kvm_hv_hypercall_set_result(vcpu, result);
1678         ++vcpu->stat.hypercalls;
1679         return kvm_skip_emulated_instruction(vcpu);
1680 }
1681
1682 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
1683 {
1684         return kvm_hv_hypercall_complete(vcpu, vcpu->run->hyperv.u.hcall.result);
1685 }
1686
1687 static u16 kvm_hvcall_signal_event(struct kvm_vcpu *vcpu, bool fast, u64 param)
1688 {
1689         struct eventfd_ctx *eventfd;
1690
1691         if (unlikely(!fast)) {
1692                 int ret;
1693                 gpa_t gpa = param;
1694
1695                 if ((gpa & (__alignof__(param) - 1)) ||
1696                     offset_in_page(gpa) + sizeof(param) > PAGE_SIZE)
1697                         return HV_STATUS_INVALID_ALIGNMENT;
1698
1699                 ret = kvm_vcpu_read_guest(vcpu, gpa, &param, sizeof(param));
1700                 if (ret < 0)
1701                         return HV_STATUS_INVALID_ALIGNMENT;
1702         }
1703
1704         /*
1705          * Per spec, bits 32-47 contain the extra "flag number".  However, we
1706          * have no use for it, and in all known usecases it is zero, so just
1707          * report lookup failure if it isn't.
1708          */
1709         if (param & 0xffff00000000ULL)
1710                 return HV_STATUS_INVALID_PORT_ID;
1711         /* remaining bits are reserved-zero */
1712         if (param & ~KVM_HYPERV_CONN_ID_MASK)
1713                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1714
1715         /* the eventfd is protected by vcpu->kvm->srcu, but conn_to_evt isn't */
1716         rcu_read_lock();
1717         eventfd = idr_find(&vcpu->kvm->arch.hyperv.conn_to_evt, param);
1718         rcu_read_unlock();
1719         if (!eventfd)
1720                 return HV_STATUS_INVALID_PORT_ID;
1721
1722         eventfd_signal(eventfd, 1);
1723         return HV_STATUS_SUCCESS;
1724 }
1725
1726 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
1727 {
1728         u64 param, ingpa, outgpa, ret = HV_STATUS_SUCCESS;
1729         uint16_t code, rep_idx, rep_cnt;
1730         bool fast, rep;
1731
1732         /*
1733          * hypercall generates UD from non zero cpl and real mode
1734          * per HYPER-V spec
1735          */
1736         if (kvm_x86_ops.get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
1737                 kvm_queue_exception(vcpu, UD_VECTOR);
1738                 return 1;
1739         }
1740
1741 #ifdef CONFIG_X86_64
1742         if (is_64_bit_mode(vcpu)) {
1743                 param = kvm_rcx_read(vcpu);
1744                 ingpa = kvm_rdx_read(vcpu);
1745                 outgpa = kvm_r8_read(vcpu);
1746         } else
1747 #endif
1748         {
1749                 param = ((u64)kvm_rdx_read(vcpu) << 32) |
1750                         (kvm_rax_read(vcpu) & 0xffffffff);
1751                 ingpa = ((u64)kvm_rbx_read(vcpu) << 32) |
1752                         (kvm_rcx_read(vcpu) & 0xffffffff);
1753                 outgpa = ((u64)kvm_rdi_read(vcpu) << 32) |
1754                         (kvm_rsi_read(vcpu) & 0xffffffff);
1755         }
1756
1757         code = param & 0xffff;
1758         fast = !!(param & HV_HYPERCALL_FAST_BIT);
1759         rep_cnt = (param >> HV_HYPERCALL_REP_COMP_OFFSET) & 0xfff;
1760         rep_idx = (param >> HV_HYPERCALL_REP_START_OFFSET) & 0xfff;
1761         rep = !!(rep_cnt || rep_idx);
1762
1763         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
1764
1765         switch (code) {
1766         case HVCALL_NOTIFY_LONG_SPIN_WAIT:
1767                 if (unlikely(rep)) {
1768                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1769                         break;
1770                 }
1771                 kvm_vcpu_on_spin(vcpu, true);
1772                 break;
1773         case HVCALL_SIGNAL_EVENT:
1774                 if (unlikely(rep)) {
1775                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1776                         break;
1777                 }
1778                 ret = kvm_hvcall_signal_event(vcpu, fast, ingpa);
1779                 if (ret != HV_STATUS_INVALID_PORT_ID)
1780                         break;
1781                 /* fall through - maybe userspace knows this conn_id. */
1782         case HVCALL_POST_MESSAGE:
1783                 /* don't bother userspace if it has no way to handle it */
1784                 if (unlikely(rep || !vcpu_to_synic(vcpu)->active)) {
1785                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1786                         break;
1787                 }
1788                 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
1789                 vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
1790                 vcpu->run->hyperv.u.hcall.input = param;
1791                 vcpu->run->hyperv.u.hcall.params[0] = ingpa;
1792                 vcpu->run->hyperv.u.hcall.params[1] = outgpa;
1793                 vcpu->arch.complete_userspace_io =
1794                                 kvm_hv_hypercall_complete_userspace;
1795                 return 0;
1796         case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
1797                 if (unlikely(fast || !rep_cnt || rep_idx)) {
1798                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1799                         break;
1800                 }
1801                 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false);
1802                 break;
1803         case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
1804                 if (unlikely(fast || rep)) {
1805                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1806                         break;
1807                 }
1808                 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false);
1809                 break;
1810         case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
1811                 if (unlikely(fast || !rep_cnt || rep_idx)) {
1812                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1813                         break;
1814                 }
1815                 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
1816                 break;
1817         case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
1818                 if (unlikely(fast || rep)) {
1819                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1820                         break;
1821                 }
1822                 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
1823                 break;
1824         case HVCALL_SEND_IPI:
1825                 if (unlikely(rep)) {
1826                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1827                         break;
1828                 }
1829                 ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, false, fast);
1830                 break;
1831         case HVCALL_SEND_IPI_EX:
1832                 if (unlikely(fast || rep)) {
1833                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1834                         break;
1835                 }
1836                 ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, true, false);
1837                 break;
1838         default:
1839                 ret = HV_STATUS_INVALID_HYPERCALL_CODE;
1840                 break;
1841         }
1842
1843         return kvm_hv_hypercall_complete(vcpu, ret);
1844 }
1845
1846 void kvm_hv_init_vm(struct kvm *kvm)
1847 {
1848         mutex_init(&kvm->arch.hyperv.hv_lock);
1849         idr_init(&kvm->arch.hyperv.conn_to_evt);
1850 }
1851
1852 void kvm_hv_destroy_vm(struct kvm *kvm)
1853 {
1854         struct eventfd_ctx *eventfd;
1855         int i;
1856
1857         idr_for_each_entry(&kvm->arch.hyperv.conn_to_evt, eventfd, i)
1858                 eventfd_ctx_put(eventfd);
1859         idr_destroy(&kvm->arch.hyperv.conn_to_evt);
1860 }
1861
1862 static int kvm_hv_eventfd_assign(struct kvm *kvm, u32 conn_id, int fd)
1863 {
1864         struct kvm_hv *hv = &kvm->arch.hyperv;
1865         struct eventfd_ctx *eventfd;
1866         int ret;
1867
1868         eventfd = eventfd_ctx_fdget(fd);
1869         if (IS_ERR(eventfd))
1870                 return PTR_ERR(eventfd);
1871
1872         mutex_lock(&hv->hv_lock);
1873         ret = idr_alloc(&hv->conn_to_evt, eventfd, conn_id, conn_id + 1,
1874                         GFP_KERNEL_ACCOUNT);
1875         mutex_unlock(&hv->hv_lock);
1876
1877         if (ret >= 0)
1878                 return 0;
1879
1880         if (ret == -ENOSPC)
1881                 ret = -EEXIST;
1882         eventfd_ctx_put(eventfd);
1883         return ret;
1884 }
1885
1886 static int kvm_hv_eventfd_deassign(struct kvm *kvm, u32 conn_id)
1887 {
1888         struct kvm_hv *hv = &kvm->arch.hyperv;
1889         struct eventfd_ctx *eventfd;
1890
1891         mutex_lock(&hv->hv_lock);
1892         eventfd = idr_remove(&hv->conn_to_evt, conn_id);
1893         mutex_unlock(&hv->hv_lock);
1894
1895         if (!eventfd)
1896                 return -ENOENT;
1897
1898         synchronize_srcu(&kvm->srcu);
1899         eventfd_ctx_put(eventfd);
1900         return 0;
1901 }
1902
1903 int kvm_vm_ioctl_hv_eventfd(struct kvm *kvm, struct kvm_hyperv_eventfd *args)
1904 {
1905         if ((args->flags & ~KVM_HYPERV_EVENTFD_DEASSIGN) ||
1906             (args->conn_id & ~KVM_HYPERV_CONN_ID_MASK))
1907                 return -EINVAL;
1908
1909         if (args->flags == KVM_HYPERV_EVENTFD_DEASSIGN)
1910                 return kvm_hv_eventfd_deassign(kvm, args->conn_id);
1911         return kvm_hv_eventfd_assign(kvm, args->conn_id, args->fd);
1912 }
1913
1914 int kvm_vcpu_ioctl_get_hv_cpuid(struct kvm_vcpu *vcpu, struct kvm_cpuid2 *cpuid,
1915                                 struct kvm_cpuid_entry2 __user *entries)
1916 {
1917         uint16_t evmcs_ver = 0;
1918         struct kvm_cpuid_entry2 cpuid_entries[] = {
1919                 { .function = HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS },
1920                 { .function = HYPERV_CPUID_INTERFACE },
1921                 { .function = HYPERV_CPUID_VERSION },
1922                 { .function = HYPERV_CPUID_FEATURES },
1923                 { .function = HYPERV_CPUID_ENLIGHTMENT_INFO },
1924                 { .function = HYPERV_CPUID_IMPLEMENT_LIMITS },
1925                 { .function = HYPERV_CPUID_SYNDBG_VENDOR_AND_MAX_FUNCTIONS },
1926                 { .function = HYPERV_CPUID_SYNDBG_INTERFACE },
1927                 { .function = HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES },
1928                 { .function = HYPERV_CPUID_NESTED_FEATURES },
1929         };
1930         int i, nent = ARRAY_SIZE(cpuid_entries);
1931
1932         if (kvm_x86_ops.nested_ops->get_evmcs_version)
1933                 evmcs_ver = kvm_x86_ops.nested_ops->get_evmcs_version(vcpu);
1934
1935         /* Skip NESTED_FEATURES if eVMCS is not supported */
1936         if (!evmcs_ver)
1937                 --nent;
1938
1939         if (cpuid->nent < nent)
1940                 return -E2BIG;
1941
1942         if (cpuid->nent > nent)
1943                 cpuid->nent = nent;
1944
1945         for (i = 0; i < nent; i++) {
1946                 struct kvm_cpuid_entry2 *ent = &cpuid_entries[i];
1947                 u32 signature[3];
1948
1949                 switch (ent->function) {
1950                 case HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS:
1951                         memcpy(signature, "Linux KVM Hv", 12);
1952
1953                         ent->eax = HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES;
1954                         ent->ebx = signature[0];
1955                         ent->ecx = signature[1];
1956                         ent->edx = signature[2];
1957                         break;
1958
1959                 case HYPERV_CPUID_INTERFACE:
1960                         memcpy(signature, "Hv#1\0\0\0\0\0\0\0\0", 12);
1961                         ent->eax = signature[0];
1962                         break;
1963
1964                 case HYPERV_CPUID_VERSION:
1965                         /*
1966                          * We implement some Hyper-V 2016 functions so let's use
1967                          * this version.
1968                          */
1969                         ent->eax = 0x00003839;
1970                         ent->ebx = 0x000A0000;
1971                         break;
1972
1973                 case HYPERV_CPUID_FEATURES:
1974                         ent->eax |= HV_X64_MSR_VP_RUNTIME_AVAILABLE;
1975                         ent->eax |= HV_MSR_TIME_REF_COUNT_AVAILABLE;
1976                         ent->eax |= HV_X64_MSR_SYNIC_AVAILABLE;
1977                         ent->eax |= HV_MSR_SYNTIMER_AVAILABLE;
1978                         ent->eax |= HV_X64_MSR_APIC_ACCESS_AVAILABLE;
1979                         ent->eax |= HV_X64_MSR_HYPERCALL_AVAILABLE;
1980                         ent->eax |= HV_X64_MSR_VP_INDEX_AVAILABLE;
1981                         ent->eax |= HV_X64_MSR_RESET_AVAILABLE;
1982                         ent->eax |= HV_MSR_REFERENCE_TSC_AVAILABLE;
1983                         ent->eax |= HV_X64_ACCESS_FREQUENCY_MSRS;
1984                         ent->eax |= HV_X64_ACCESS_REENLIGHTENMENT;
1985
1986                         ent->ebx |= HV_X64_POST_MESSAGES;
1987                         ent->ebx |= HV_X64_SIGNAL_EVENTS;
1988
1989                         ent->edx |= HV_FEATURE_FREQUENCY_MSRS_AVAILABLE;
1990                         ent->edx |= HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
1991
1992                         ent->ebx |= HV_X64_DEBUGGING;
1993                         ent->edx |= HV_X64_GUEST_DEBUGGING_AVAILABLE;
1994                         ent->edx |= HV_FEATURE_DEBUG_MSRS_AVAILABLE;
1995
1996                         /*
1997                          * Direct Synthetic timers only make sense with in-kernel
1998                          * LAPIC
1999                          */
2000                         if (lapic_in_kernel(vcpu))
2001                                 ent->edx |= HV_STIMER_DIRECT_MODE_AVAILABLE;
2002
2003                         break;
2004
2005                 case HYPERV_CPUID_ENLIGHTMENT_INFO:
2006                         ent->eax |= HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
2007                         ent->eax |= HV_X64_APIC_ACCESS_RECOMMENDED;
2008                         ent->eax |= HV_X64_RELAXED_TIMING_RECOMMENDED;
2009                         ent->eax |= HV_X64_CLUSTER_IPI_RECOMMENDED;
2010                         ent->eax |= HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED;
2011                         if (evmcs_ver)
2012                                 ent->eax |= HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
2013                         if (!cpu_smt_possible())
2014                                 ent->eax |= HV_X64_NO_NONARCH_CORESHARING;
2015                         /*
2016                          * Default number of spinlock retry attempts, matches
2017                          * HyperV 2016.
2018                          */
2019                         ent->ebx = 0x00000FFF;
2020
2021                         break;
2022
2023                 case HYPERV_CPUID_IMPLEMENT_LIMITS:
2024                         /* Maximum number of virtual processors */
2025                         ent->eax = KVM_MAX_VCPUS;
2026                         /*
2027                          * Maximum number of logical processors, matches
2028                          * HyperV 2016.
2029                          */
2030                         ent->ebx = 64;
2031
2032                         break;
2033
2034                 case HYPERV_CPUID_NESTED_FEATURES:
2035                         ent->eax = evmcs_ver;
2036
2037                         break;
2038
2039                 case HYPERV_CPUID_SYNDBG_VENDOR_AND_MAX_FUNCTIONS:
2040                         memcpy(signature, "Linux KVM Hv", 12);
2041
2042                         ent->eax = 0;
2043                         ent->ebx = signature[0];
2044                         ent->ecx = signature[1];
2045                         ent->edx = signature[2];
2046                         break;
2047
2048                 case HYPERV_CPUID_SYNDBG_INTERFACE:
2049                         memcpy(signature, "VS#1\0\0\0\0\0\0\0\0", 12);
2050                         ent->eax = signature[0];
2051                         break;
2052
2053                 case HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES:
2054                         ent->eax |= HV_X64_SYNDBG_CAP_ALLOW_KERNEL_DEBUGGING;
2055                         break;
2056
2057                 default:
2058                         break;
2059                 }
2060         }
2061
2062         if (copy_to_user(entries, cpuid_entries,
2063                          nent * sizeof(struct kvm_cpuid_entry2)))
2064                 return -EFAULT;
2065
2066         return 0;
2067 }