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 "hyperv.h"
  25
  26 #include <linux/kvm_host.h>
  27 #include <linux/highmem.h>
  28 #include <linux/sched/cputime.h>
  29 #include <linux/eventfd.h>
  30
  31 #include <asm/apicdef.h>
  32 #include <trace/events/kvm.h>
  33
  34 #include "trace.h"
  35
  36 #define KVM_HV_MAX_SPARSE_VCPU_SET_BITS DIV_ROUND_UP(KVM_MAX_VCPUS, 64)
  37
  38 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
  39                                 bool vcpu_kick);
  40
  41 static inline u64 synic_read_sint(struct kvm_vcpu_hv_synic *synic, int sint)
  42 {
  43         return atomic64_read(&synic->sint[sint]);
  44 }
  45
  46 static inline int synic_get_sint_vector(u64 sint_value)
  47 {
  48         if (sint_value & HV_SYNIC_SINT_MASKED)
  49                 return -1;
  50         return sint_value & HV_SYNIC_SINT_VECTOR_MASK;
  51 }
  52
  53 static bool synic_has_vector_connected(struct kvm_vcpu_hv_synic *synic,
  54                                       int vector)
  55 {
  56         int i;
  57
  58         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
  59                 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
  60                         return true;
  61         }
  62         return false;
  63 }
  64
  65 static bool synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic *synic,
  66                                      int vector)
  67 {
  68         int i;
  69         u64 sint_value;
  70
  71         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
  72                 sint_value = synic_read_sint(synic, i);
  73                 if (synic_get_sint_vector(sint_value) == vector &&
  74                     sint_value & HV_SYNIC_SINT_AUTO_EOI)
  75                         return true;
  76         }
  77         return false;
  78 }
  79
  80 static void synic_update_vector(struct kvm_vcpu_hv_synic *synic,
  81                                 int vector)
  82 {
  83         if (vector < HV_SYNIC_FIRST_VALID_VECTOR)
  84                 return;
  85
  86         if (synic_has_vector_connected(synic, vector))
  87                 __set_bit(vector, synic->vec_bitmap);
  88         else
  89                 __clear_bit(vector, synic->vec_bitmap);
  90
  91         if (synic_has_vector_auto_eoi(synic, vector))
  92                 __set_bit(vector, synic->auto_eoi_bitmap);
  93         else
  94                 __clear_bit(vector, synic->auto_eoi_bitmap);
  95 }
  96
  97 static int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint,
  98                           u64 data, bool host)
  99 {
 100         int vector, old_vector;
 101         bool masked;
 102
 103         vector = data & HV_SYNIC_SINT_VECTOR_MASK;
 104         masked = data & HV_SYNIC_SINT_MASKED;
 105
 106         /*
 107          * Valid vectors are 16-255, however, nested Hyper-V attempts to write
 108          * default '0x10000' value on boot and this should not #GP. We need to
 109          * allow zero-initing the register from host as well.
 110          */
 111         if (vector < HV_SYNIC_FIRST_VALID_VECTOR && !host && !masked)
 112                 return 1;
 113         /*
 114          * Guest may configure multiple SINTs to use the same vector, so
 115          * we maintain a bitmap of vectors handled by synic, and a
 116          * bitmap of vectors with auto-eoi behavior.  The bitmaps are
 117          * updated here, and atomically queried on fast paths.
 118          */
 119         old_vector = synic_read_sint(synic, sint) & HV_SYNIC_SINT_VECTOR_MASK;
 120
 121         atomic64_set(&synic->sint[sint], data);
 122
 123         synic_update_vector(synic, old_vector);
 124
 125         synic_update_vector(synic, vector);
 126
 127         /* Load SynIC vectors into EOI exit bitmap */
 128         kvm_make_request(KVM_REQ_SCAN_IOAPIC, synic_to_vcpu(synic));
 129         return 0;
 130 }
 131
 132 static struct kvm_vcpu *get_vcpu_by_vpidx(struct kvm *kvm, u32 vpidx)
 133 {
 134         struct kvm_vcpu *vcpu = NULL;
 135         int i;
 136
 137         if (vpidx >= KVM_MAX_VCPUS)
 138                 return NULL;
 139
 140         vcpu = kvm_get_vcpu(kvm, vpidx);
 141         if (vcpu && vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
 142                 return vcpu;
 143         kvm_for_each_vcpu(i, vcpu, kvm)
 144                 if (vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
 145                         return vcpu;
 146         return NULL;
 147 }
 148
 149 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vpidx)
 150 {
 151         struct kvm_vcpu *vcpu;
 152         struct kvm_vcpu_hv_synic *synic;
 153
 154         vcpu = get_vcpu_by_vpidx(kvm, vpidx);
 155         if (!vcpu)
 156                 return NULL;
 157         synic = vcpu_to_synic(vcpu);
 158         return (synic->active) ? synic : NULL;
 159 }
 160
 161 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
 162 {
 163         struct kvm *kvm = vcpu->kvm;
 164         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
 165         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 166         struct kvm_vcpu_hv_stimer *stimer;
 167         int gsi, idx;
 168
 169         trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
 170
 171         /* Try to deliver pending Hyper-V SynIC timers messages */
 172         for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
 173                 stimer = &hv_vcpu->stimer[idx];
 174                 if (stimer->msg_pending && stimer->config.enable &&
 175                     !stimer->config.direct_mode &&
 176                     stimer->config.sintx == sint)
 177                         stimer_mark_pending(stimer, false);
 178         }
 179
 180         idx = srcu_read_lock(&kvm->irq_srcu);
 181         gsi = atomic_read(&synic->sint_to_gsi[sint]);
 182         if (gsi != -1)
 183                 kvm_notify_acked_gsi(kvm, gsi);
 184         srcu_read_unlock(&kvm->irq_srcu, idx);
 185 }
 186
 187 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
 188 {
 189         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 190         struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
 191
 192         hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
 193         hv_vcpu->exit.u.synic.msr = msr;
 194         hv_vcpu->exit.u.synic.control = synic->control;
 195         hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
 196         hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
 197
 198         kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
 199 }
 200
 201 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
 202                          u32 msr, u64 data, bool host)
 203 {
 204         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 205         int ret;
 206
 207         if (!synic->active && !host)
 208                 return 1;
 209
 210         trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
 211
 212         ret = 0;
 213         switch (msr) {
 214         case HV_X64_MSR_SCONTROL:
 215                 synic->control = data;
 216                 if (!host)
 217                         synic_exit(synic, msr);
 218                 break;
 219         case HV_X64_MSR_SVERSION:
 220                 if (!host) {
 221                         ret = 1;
 222                         break;
 223                 }
 224                 synic->version = data;
 225                 break;
 226         case HV_X64_MSR_SIEFP:
 227                 if ((data & HV_SYNIC_SIEFP_ENABLE) && !host &&
 228                     !synic->dont_zero_synic_pages)
 229                         if (kvm_clear_guest(vcpu->kvm,
 230                                             data & PAGE_MASK, PAGE_SIZE)) {
 231                                 ret = 1;
 232                                 break;
 233                         }
 234                 synic->evt_page = data;
 235                 if (!host)
 236                         synic_exit(synic, msr);
 237                 break;
 238         case HV_X64_MSR_SIMP:
 239                 if ((data & HV_SYNIC_SIMP_ENABLE) && !host &&
 240                     !synic->dont_zero_synic_pages)
 241                         if (kvm_clear_guest(vcpu->kvm,
 242                                             data & PAGE_MASK, PAGE_SIZE)) {
 243                                 ret = 1;
 244                                 break;
 245                         }
 246                 synic->msg_page = data;
 247                 if (!host)
 248                         synic_exit(synic, msr);
 249                 break;
 250         case HV_X64_MSR_EOM: {
 251                 int i;
 252
 253                 for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
 254                         kvm_hv_notify_acked_sint(vcpu, i);
 255                 break;
 256         }
 257         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
 258                 ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
 259                 break;
 260         default:
 261                 ret = 1;
 262                 break;
 263         }
 264         return ret;
 265 }
 266
 267 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata,
 268                          bool host)
 269 {
 270         int ret;
 271
 272         if (!synic->active && !host)
 273                 return 1;
 274
 275         ret = 0;
 276         switch (msr) {
 277         case HV_X64_MSR_SCONTROL:
 278                 *pdata = synic->control;
 279                 break;
 280         case HV_X64_MSR_SVERSION:
 281                 *pdata = synic->version;
 282                 break;
 283         case HV_X64_MSR_SIEFP:
 284                 *pdata = synic->evt_page;
 285                 break;
 286         case HV_X64_MSR_SIMP:
 287                 *pdata = synic->msg_page;
 288                 break;
 289         case HV_X64_MSR_EOM:
 290                 *pdata = 0;
 291                 break;
 292         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
 293                 *pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
 294                 break;
 295         default:
 296                 ret = 1;
 297                 break;
 298         }
 299         return ret;
 300 }
 301
 302 static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
 303 {
 304         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 305         struct kvm_lapic_irq irq;
 306         int ret, vector;
 307
 308         if (sint >= ARRAY_SIZE(synic->sint))
 309                 return -EINVAL;
 310
 311         vector = synic_get_sint_vector(synic_read_sint(synic, sint));
 312         if (vector < 0)
 313                 return -ENOENT;
 314
 315         memset(&irq, 0, sizeof(irq));
 316         irq.shorthand = APIC_DEST_SELF;
 317         irq.dest_mode = APIC_DEST_PHYSICAL;
 318         irq.delivery_mode = APIC_DM_FIXED;
 319         irq.vector = vector;
 320         irq.level = 1;
 321
 322         ret = kvm_irq_delivery_to_apic(vcpu->kvm, vcpu->arch.apic, &irq, NULL);
 323         trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
 324         return ret;
 325 }
 326
 327 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vpidx, u32 sint)
 328 {
 329         struct kvm_vcpu_hv_synic *synic;
 330
 331         synic = synic_get(kvm, vpidx);
 332         if (!synic)
 333                 return -EINVAL;
 334
 335         return synic_set_irq(synic, sint);
 336 }
 337
 338 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
 339 {
 340         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
 341         int i;
 342
 343         trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
 344
 345         for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
 346                 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
 347                         kvm_hv_notify_acked_sint(vcpu, i);
 348 }
 349
 350 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi)
 351 {
 352         struct kvm_vcpu_hv_synic *synic;
 353
 354         synic = synic_get(kvm, vpidx);
 355         if (!synic)
 356                 return -EINVAL;
 357
 358         if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
 359                 return -EINVAL;
 360
 361         atomic_set(&synic->sint_to_gsi[sint], gsi);
 362         return 0;
 363 }
 364
 365 void kvm_hv_irq_routing_update(struct kvm *kvm)
 366 {
 367         struct kvm_irq_routing_table *irq_rt;
 368         struct kvm_kernel_irq_routing_entry *e;
 369         u32 gsi;
 370
 371         irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
 372                                         lockdep_is_held(&kvm->irq_lock));
 373
 374         for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
 375                 hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
 376                         if (e->type == KVM_IRQ_ROUTING_HV_SINT)
 377                                 kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
 378                                                     e->hv_sint.sint, gsi);
 379                 }
 380         }
 381 }
 382
 383 static void synic_init(struct kvm_vcpu_hv_synic *synic)
 384 {
 385         int i;
 386
 387         memset(synic, 0, sizeof(*synic));
 388         synic->version = HV_SYNIC_VERSION_1;
 389         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
 390                 atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
 391                 atomic_set(&synic->sint_to_gsi[i], -1);
 392         }
 393 }
 394
 395 static u64 get_time_ref_counter(struct kvm *kvm)
 396 {
 397         struct kvm_hv *hv = &kvm->arch.hyperv;
 398         struct kvm_vcpu *vcpu;
 399         u64 tsc;
 400
 401         /*
 402          * The guest has not set up the TSC page or the clock isn't
 403          * stable, fall back to get_kvmclock_ns.
 404          */
 405         if (!hv->tsc_ref.tsc_sequence)
 406                 return div_u64(get_kvmclock_ns(kvm), 100);
 407
 408         vcpu = kvm_get_vcpu(kvm, 0);
 409         tsc = kvm_read_l1_tsc(vcpu, rdtsc());
 410         return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64)
 411                 + hv->tsc_ref.tsc_offset;
 412 }
 413
 414 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
 415                                 bool vcpu_kick)
 416 {
 417         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
 418
 419         set_bit(stimer->index,
 420                 vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
 421         kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
 422         if (vcpu_kick)
 423                 kvm_vcpu_kick(vcpu);
 424 }
 425
 426 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
 427 {
 428         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
 429
 430         trace_kvm_hv_stimer_cleanup(stimer_to_vcpu(stimer)->vcpu_id,
 431                                     stimer->index);
 432
 433         hrtimer_cancel(&stimer->timer);
 434         clear_bit(stimer->index,
 435                   vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
 436         stimer->msg_pending = false;
 437         stimer->exp_time = 0;
 438 }
 439
 440 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
 441 {
 442         struct kvm_vcpu_hv_stimer *stimer;
 443
 444         stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
 445         trace_kvm_hv_stimer_callback(stimer_to_vcpu(stimer)->vcpu_id,
 446                                      stimer->index);
 447         stimer_mark_pending(stimer, true);
 448
 449         return HRTIMER_NORESTART;
 450 }
 451
 452 /*
 453  * stimer_start() assumptions:
 454  * a) stimer->count is not equal to 0
 455  * b) stimer->config has HV_STIMER_ENABLE flag
 456  */
 457 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
 458 {
 459         u64 time_now;
 460         ktime_t ktime_now;
 461
 462         time_now = get_time_ref_counter(stimer_to_vcpu(stimer)->kvm);
 463         ktime_now = ktime_get();
 464
 465         if (stimer->config.periodic) {
 466                 if (stimer->exp_time) {
 467                         if (time_now >= stimer->exp_time) {
 468                                 u64 remainder;
 469
 470                                 div64_u64_rem(time_now - stimer->exp_time,
 471                                               stimer->count, &remainder);
 472                                 stimer->exp_time =
 473                                         time_now + (stimer->count - remainder);
 474                         }
 475                 } else
 476                         stimer->exp_time = time_now + stimer->count;
 477
 478                 trace_kvm_hv_stimer_start_periodic(
 479                                         stimer_to_vcpu(stimer)->vcpu_id,
 480                                         stimer->index,
 481                                         time_now, stimer->exp_time);
 482
 483                 hrtimer_start(&stimer->timer,
 484                               ktime_add_ns(ktime_now,
 485                                            100 * (stimer->exp_time - time_now)),
 486                               HRTIMER_MODE_ABS);
 487                 return 0;
 488         }
 489         stimer->exp_time = stimer->count;
 490         if (time_now >= stimer->count) {
 491                 /*
 492                  * Expire timer according to Hypervisor Top-Level Functional
 493                  * specification v4(15.3.1):
 494                  * "If a one shot is enabled and the specified count is in
 495                  * the past, it will expire immediately."
 496                  */
 497                 stimer_mark_pending(stimer, false);
 498                 return 0;
 499         }
 500
 501         trace_kvm_hv_stimer_start_one_shot(stimer_to_vcpu(stimer)->vcpu_id,
 502                                            stimer->index,
 503                                            time_now, stimer->count);
 504
 505         hrtimer_start(&stimer->timer,
 506                       ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)),
 507                       HRTIMER_MODE_ABS);
 508         return 0;
 509 }
 510
 511 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
 512                              bool host)
 513 {
 514         union hv_stimer_config new_config = {.as_uint64 = config},
 515                 old_config = {.as_uint64 = stimer->config.as_uint64};
 516
 517         trace_kvm_hv_stimer_set_config(stimer_to_vcpu(stimer)->vcpu_id,
 518                                        stimer->index, config, host);
 519
 520         stimer_cleanup(stimer);
 521         if (old_config.enable &&
 522             !new_config.direct_mode && new_config.sintx == 0)
 523                 new_config.enable = 0;
 524         stimer->config.as_uint64 = new_config.as_uint64;
 525
 526         if (stimer->config.enable)
 527                 stimer_mark_pending(stimer, false);
 528
 529         return 0;
 530 }
 531
 532 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
 533                             bool host)
 534 {
 535         trace_kvm_hv_stimer_set_count(stimer_to_vcpu(stimer)->vcpu_id,
 536                                       stimer->index, count, host);
 537
 538         stimer_cleanup(stimer);
 539         stimer->count = count;
 540         if (stimer->count == 0)
 541                 stimer->config.enable = 0;
 542         else if (stimer->config.auto_enable)
 543                 stimer->config.enable = 1;
 544
 545         if (stimer->config.enable)
 546                 stimer_mark_pending(stimer, false);
 547
 548         return 0;
 549 }
 550
 551 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
 552 {
 553         *pconfig = stimer->config.as_uint64;
 554         return 0;
 555 }
 556
 557 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
 558 {
 559         *pcount = stimer->count;
 560         return 0;
 561 }
 562
 563 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
 564                              struct hv_message *src_msg, bool no_retry)
 565 {
 566         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 567         int msg_off = offsetof(struct hv_message_page, sint_message[sint]);
 568         gfn_t msg_page_gfn;
 569         struct hv_message_header hv_hdr;
 570         int r;
 571
 572         if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
 573                 return -ENOENT;
 574
 575         msg_page_gfn = synic->msg_page >> PAGE_SHIFT;
 576
 577         /*
 578          * Strictly following the spec-mandated ordering would assume setting
 579          * .msg_pending before checking .message_type.  However, this function
 580          * is only called in vcpu context so the entire update is atomic from
 581          * guest POV and thus the exact order here doesn't matter.
 582          */
 583         r = kvm_vcpu_read_guest_page(vcpu, msg_page_gfn, &hv_hdr.message_type,
 584                                      msg_off + offsetof(struct hv_message,
 585                                                         header.message_type),
 586                                      sizeof(hv_hdr.message_type));
 587         if (r < 0)
 588                 return r;
 589
 590         if (hv_hdr.message_type != HVMSG_NONE) {
 591                 if (no_retry)
 592                         return 0;
 593
 594                 hv_hdr.message_flags.msg_pending = 1;
 595                 r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn,
 596                                               &hv_hdr.message_flags,
 597                                               msg_off +
 598                                               offsetof(struct hv_message,
 599                                                        header.message_flags),
 600                                               sizeof(hv_hdr.message_flags));
 601                 if (r < 0)
 602                         return r;
 603                 return -EAGAIN;
 604         }
 605
 606         r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn, src_msg, msg_off,
 607                                       sizeof(src_msg->header) +
 608                                       src_msg->header.payload_size);
 609         if (r < 0)
 610                 return r;
 611
 612         r = synic_set_irq(synic, sint);
 613         if (r < 0)
 614                 return r;
 615         if (r == 0)
 616                 return -EFAULT;
 617         return 0;
 618 }
 619
 620 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
 621 {
 622         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
 623         struct hv_message *msg = &stimer->msg;
 624         struct hv_timer_message_payload *payload =
 625                         (struct hv_timer_message_payload *)&msg->u.payload;
 626
 627         /*
 628          * To avoid piling up periodic ticks, don't retry message
 629          * delivery for them (within "lazy" lost ticks policy).
 630          */
 631         bool no_retry = stimer->config.periodic;
 632
 633         payload->expiration_time = stimer->exp_time;
 634         payload->delivery_time = get_time_ref_counter(vcpu->kvm);
 635         return synic_deliver_msg(vcpu_to_synic(vcpu),
 636                                  stimer->config.sintx, msg,
 637                                  no_retry);
 638 }
 639
 640 static int stimer_notify_direct(struct kvm_vcpu_hv_stimer *stimer)
 641 {
 642         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
 643         struct kvm_lapic_irq irq = {
 644                 .delivery_mode = APIC_DM_FIXED,
 645                 .vector = stimer->config.apic_vector
 646         };
 647
 648         if (lapic_in_kernel(vcpu))
 649                 return !kvm_apic_set_irq(vcpu, &irq, NULL);
 650         return 0;
 651 }
 652
 653 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
 654 {
 655         int r, direct = stimer->config.direct_mode;
 656
 657         stimer->msg_pending = true;
 658         if (!direct)
 659                 r = stimer_send_msg(stimer);
 660         else
 661                 r = stimer_notify_direct(stimer);
 662         trace_kvm_hv_stimer_expiration(stimer_to_vcpu(stimer)->vcpu_id,
 663                                        stimer->index, direct, r);
 664         if (!r) {
 665                 stimer->msg_pending = false;
 666                 if (!(stimer->config.periodic))
 667                         stimer->config.enable = 0;
 668         }
 669 }
 670
 671 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
 672 {
 673         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 674         struct kvm_vcpu_hv_stimer *stimer;
 675         u64 time_now, exp_time;
 676         int i;
 677
 678         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 679                 if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
 680                         stimer = &hv_vcpu->stimer[i];
 681                         if (stimer->config.enable) {
 682                                 exp_time = stimer->exp_time;
 683
 684                                 if (exp_time) {
 685                                         time_now =
 686                                                 get_time_ref_counter(vcpu->kvm);
 687                                         if (time_now >= exp_time)
 688                                                 stimer_expiration(stimer);
 689                                 }
 690
 691                                 if ((stimer->config.enable) &&
 692                                     stimer->count) {
 693                                         if (!stimer->msg_pending)
 694                                                 stimer_start(stimer);
 695                                 } else
 696                                         stimer_cleanup(stimer);
 697                         }
 698                 }
 699 }
 700
 701 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
 702 {
 703         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 704         int i;
 705
 706         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 707                 stimer_cleanup(&hv_vcpu->stimer[i]);
 708 }
 709
 710 bool kvm_hv_assist_page_enabled(struct kvm_vcpu *vcpu)
 711 {
 712         if (!(vcpu->arch.hyperv.hv_vapic & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE))
 713                 return false;
 714         return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
 715 }
 716 EXPORT_SYMBOL_GPL(kvm_hv_assist_page_enabled);
 717
 718 bool kvm_hv_get_assist_page(struct kvm_vcpu *vcpu,
 719                             struct hv_vp_assist_page *assist_page)
 720 {
 721         if (!kvm_hv_assist_page_enabled(vcpu))
 722                 return false;
 723         return !kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data,
 724                                       assist_page, sizeof(*assist_page));
 725 }
 726 EXPORT_SYMBOL_GPL(kvm_hv_get_assist_page);
 727
 728 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
 729 {
 730         struct hv_message *msg = &stimer->msg;
 731         struct hv_timer_message_payload *payload =
 732                         (struct hv_timer_message_payload *)&msg->u.payload;
 733
 734         memset(&msg->header, 0, sizeof(msg->header));
 735         msg->header.message_type = HVMSG_TIMER_EXPIRED;
 736         msg->header.payload_size = sizeof(*payload);
 737
 738         payload->timer_index = stimer->index;
 739         payload->expiration_time = 0;
 740         payload->delivery_time = 0;
 741 }
 742
 743 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
 744 {
 745         memset(stimer, 0, sizeof(*stimer));
 746         stimer->index = timer_index;
 747         hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
 748         stimer->timer.function = stimer_timer_callback;
 749         stimer_prepare_msg(stimer);
 750 }
 751
 752 void kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
 753 {
 754         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 755         int i;
 756
 757         synic_init(&hv_vcpu->synic);
 758
 759         bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
 760         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 761                 stimer_init(&hv_vcpu->stimer[i], i);
 762 }
 763
 764 void kvm_hv_vcpu_postcreate(struct kvm_vcpu *vcpu)
 765 {
 766         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 767
 768         hv_vcpu->vp_index = kvm_vcpu_get_idx(vcpu);
 769 }
 770
 771 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
 772 {
 773         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
 774
 775         /*
 776          * Hyper-V SynIC auto EOI SINT's are
 777          * not compatible with APICV, so deactivate APICV
 778          */
 779         kvm_vcpu_deactivate_apicv(vcpu);
 780         synic->active = true;
 781         synic->dont_zero_synic_pages = dont_zero_synic_pages;
 782         return 0;
 783 }
 784
 785 static bool kvm_hv_msr_partition_wide(u32 msr)
 786 {
 787         bool r = false;
 788
 789         switch (msr) {
 790         case HV_X64_MSR_GUEST_OS_ID:
 791         case HV_X64_MSR_HYPERCALL:
 792         case HV_X64_MSR_REFERENCE_TSC:
 793         case HV_X64_MSR_TIME_REF_COUNT:
 794         case HV_X64_MSR_CRASH_CTL:
 795         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
 796         case HV_X64_MSR_RESET:
 797         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
 798         case HV_X64_MSR_TSC_EMULATION_CONTROL:
 799         case HV_X64_MSR_TSC_EMULATION_STATUS:
 800                 r = true;
 801                 break;
 802         }
 803
 804         return r;
 805 }
 806
 807 static int kvm_hv_msr_get_crash_data(struct kvm_vcpu *vcpu,
 808                                      u32 index, u64 *pdata)
 809 {
 810         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 811
 812         if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param)))
 813                 return -EINVAL;
 814
 815         *pdata = hv->hv_crash_param[index];
 816         return 0;
 817 }
 818
 819 static int kvm_hv_msr_get_crash_ctl(struct kvm_vcpu *vcpu, u64 *pdata)
 820 {
 821         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 822
 823         *pdata = hv->hv_crash_ctl;
 824         return 0;
 825 }
 826
 827 static int kvm_hv_msr_set_crash_ctl(struct kvm_vcpu *vcpu, u64 data, bool host)
 828 {
 829         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 830
 831         if (host)
 832                 hv->hv_crash_ctl = data & HV_CRASH_CTL_CRASH_NOTIFY;
 833
 834         if (!host && (data & HV_CRASH_CTL_CRASH_NOTIFY)) {
 835
 836                 vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
 837                           hv->hv_crash_param[0],
 838                           hv->hv_crash_param[1],
 839                           hv->hv_crash_param[2],
 840                           hv->hv_crash_param[3],
 841                           hv->hv_crash_param[4]);
 842
 843                 /* Send notification about crash to user space */
 844                 kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
 845         }
 846
 847         return 0;
 848 }
 849
 850 static int kvm_hv_msr_set_crash_data(struct kvm_vcpu *vcpu,
 851                                      u32 index, u64 data)
 852 {
 853         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 854
 855         if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param)))
 856                 return -EINVAL;
 857
 858         hv->hv_crash_param[index] = data;
 859         return 0;
 860 }
 861
 862 /*
 863  * The kvmclock and Hyper-V TSC page use similar formulas, and converting
 864  * between them is possible:
 865  *
 866  * kvmclock formula:
 867  *    nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
 868  *           + system_time
 869  *
 870  * Hyper-V formula:
 871  *    nsec/100 = ticks * scale / 2^64 + offset
 872  *
 873  * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
 874  * By dividing the kvmclock formula by 100 and equating what's left we get:
 875  *    ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
 876  *            scale / 2^64 =         tsc_to_system_mul * 2^(tsc_shift-32) / 100
 877  *            scale        =         tsc_to_system_mul * 2^(32+tsc_shift) / 100
 878  *
 879  * Now expand the kvmclock formula and divide by 100:
 880  *    nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
 881  *           - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
 882  *           + system_time
 883  *    nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
 884  *               - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
 885  *               + system_time / 100
 886  *
 887  * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
 888  *    nsec/100 = ticks * scale / 2^64
 889  *               - tsc_timestamp * scale / 2^64
 890  *               + system_time / 100
 891  *
 892  * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
 893  *    offset = system_time / 100 - tsc_timestamp * scale / 2^64
 894  *
 895  * These two equivalencies are implemented in this function.
 896  */
 897 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
 898                                         HV_REFERENCE_TSC_PAGE *tsc_ref)
 899 {
 900         u64 max_mul;
 901
 902         if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
 903                 return false;
 904
 905         /*
 906          * check if scale would overflow, if so we use the time ref counter
 907          *    tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
 908          *    tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
 909          *    tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
 910          */
 911         max_mul = 100ull << (32 - hv_clock->tsc_shift);
 912         if (hv_clock->tsc_to_system_mul >= max_mul)
 913                 return false;
 914
 915         /*
 916          * Otherwise compute the scale and offset according to the formulas
 917          * derived above.
 918          */
 919         tsc_ref->tsc_scale =
 920                 mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
 921                                 hv_clock->tsc_to_system_mul,
 922                                 100);
 923
 924         tsc_ref->tsc_offset = hv_clock->system_time;
 925         do_div(tsc_ref->tsc_offset, 100);
 926         tsc_ref->tsc_offset -=
 927                 mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
 928         return true;
 929 }
 930
 931 void kvm_hv_setup_tsc_page(struct kvm *kvm,
 932                            struct pvclock_vcpu_time_info *hv_clock)
 933 {
 934         struct kvm_hv *hv = &kvm->arch.hyperv;
 935         u32 tsc_seq;
 936         u64 gfn;
 937
 938         BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
 939         BUILD_BUG_ON(offsetof(HV_REFERENCE_TSC_PAGE, tsc_sequence) != 0);
 940
 941         if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
 942                 return;
 943
 944         mutex_lock(&kvm->arch.hyperv.hv_lock);
 945         if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
 946                 goto out_unlock;
 947
 948         gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
 949         /*
 950          * Because the TSC parameters only vary when there is a
 951          * change in the master clock, do not bother with caching.
 952          */
 953         if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
 954                                     &tsc_seq, sizeof(tsc_seq))))
 955                 goto out_unlock;
 956
 957         /*
 958          * While we're computing and writing the parameters, force the
 959          * guest to use the time reference count MSR.
 960          */
 961         hv->tsc_ref.tsc_sequence = 0;
 962         if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
 963                             &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
 964                 goto out_unlock;
 965
 966         if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
 967                 goto out_unlock;
 968
 969         /* Ensure sequence is zero before writing the rest of the struct.  */
 970         smp_wmb();
 971         if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
 972                 goto out_unlock;
 973
 974         /*
 975          * Now switch to the TSC page mechanism by writing the sequence.
 976          */
 977         tsc_seq++;
 978         if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
 979                 tsc_seq = 1;
 980
 981         /* Write the struct entirely before the non-zero sequence.  */
 982         smp_wmb();
 983
 984         hv->tsc_ref.tsc_sequence = tsc_seq;
 985         kvm_write_guest(kvm, gfn_to_gpa(gfn),
 986                         &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence));
 987 out_unlock:
 988         mutex_unlock(&kvm->arch.hyperv.hv_lock);
 989 }
 990
 991 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
 992                              bool host)
 993 {
 994         struct kvm *kvm = vcpu->kvm;
 995         struct kvm_hv *hv = &kvm->arch.hyperv;
 996
 997         switch (msr) {
 998         case HV_X64_MSR_GUEST_OS_ID:
 999                 hv->hv_guest_os_id = data;
1000                 /* setting guest os id to zero disables hypercall page */
1001                 if (!hv->hv_guest_os_id)
1002                         hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1003                 break;
1004         case HV_X64_MSR_HYPERCALL: {
1005                 u64 gfn;
1006                 unsigned long addr;
1007                 u8 instructions[4];
1008
1009                 /* if guest os id is not set hypercall should remain disabled */
1010                 if (!hv->hv_guest_os_id)
1011                         break;
1012                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1013                         hv->hv_hypercall = data;
1014                         break;
1015                 }
1016                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1017                 addr = gfn_to_hva(kvm, gfn);
1018                 if (kvm_is_error_hva(addr))
1019                         return 1;
1020                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1021                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1022                 if (__copy_to_user((void __user *)addr, instructions, 4))
1023                         return 1;
1024                 hv->hv_hypercall = data;
1025                 mark_page_dirty(kvm, gfn);
1026                 break;
1027         }
1028         case HV_X64_MSR_REFERENCE_TSC:
1029                 hv->hv_tsc_page = data;
1030                 if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE)
1031                         kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1032                 break;
1033         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1034                 return kvm_hv_msr_set_crash_data(vcpu,
1035                                                  msr - HV_X64_MSR_CRASH_P0,
1036                                                  data);
1037         case HV_X64_MSR_CRASH_CTL:
1038                 return kvm_hv_msr_set_crash_ctl(vcpu, data, host);
1039         case HV_X64_MSR_RESET:
1040                 if (data == 1) {
1041                         vcpu_debug(vcpu, "hyper-v reset requested\n");
1042                         kvm_make_request(KVM_REQ_HV_RESET, vcpu);
1043                 }
1044                 break;
1045         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1046                 hv->hv_reenlightenment_control = data;
1047                 break;
1048         case HV_X64_MSR_TSC_EMULATION_CONTROL:
1049                 hv->hv_tsc_emulation_control = data;
1050                 break;
1051         case HV_X64_MSR_TSC_EMULATION_STATUS:
1052                 hv->hv_tsc_emulation_status = data;
1053                 break;
1054         case HV_X64_MSR_TIME_REF_COUNT:
1055                 /* read-only, but still ignore it if host-initiated */
1056                 if (!host)
1057                         return 1;
1058                 break;
1059         default:
1060                 vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
1061                             msr, data);
1062                 return 1;
1063         }
1064         return 0;
1065 }
1066
1067 /* Calculate cpu time spent by current task in 100ns units */
1068 static u64 current_task_runtime_100ns(void)
1069 {
1070         u64 utime, stime;
1071
1072         task_cputime_adjusted(current, &utime, &stime);
1073
1074         return div_u64(utime + stime, 100);
1075 }
1076
1077 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1078 {
1079         struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
1080
1081         switch (msr) {
1082         case HV_X64_MSR_VP_INDEX: {
1083                 struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
1084                 int vcpu_idx = kvm_vcpu_get_idx(vcpu);
1085                 u32 new_vp_index = (u32)data;
1086
1087                 if (!host || new_vp_index >= KVM_MAX_VCPUS)
1088                         return 1;
1089
1090                 if (new_vp_index == hv_vcpu->vp_index)
1091                         return 0;
1092
1093                 /*
1094                  * The VP index is initialized to vcpu_index by
1095                  * kvm_hv_vcpu_postcreate so they initially match.  Now the
1096                  * VP index is changing, adjust num_mismatched_vp_indexes if
1097                  * it now matches or no longer matches vcpu_idx.
1098                  */
1099                 if (hv_vcpu->vp_index == vcpu_idx)
1100                         atomic_inc(&hv->num_mismatched_vp_indexes);
1101                 else if (new_vp_index == vcpu_idx)
1102                         atomic_dec(&hv->num_mismatched_vp_indexes);
1103
1104                 hv_vcpu->vp_index = new_vp_index;
1105                 break;
1106         }
1107         case HV_X64_MSR_VP_ASSIST_PAGE: {
1108                 u64 gfn;
1109                 unsigned long addr;
1110
1111                 if (!(data & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE)) {
1112                         hv_vcpu->hv_vapic = data;
1113                         if (kvm_lapic_enable_pv_eoi(vcpu, 0, 0))
1114                                 return 1;
1115                         break;
1116                 }
1117                 gfn = data >> HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT;
1118                 addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1119                 if (kvm_is_error_hva(addr))
1120                         return 1;
1121
1122                 /*
1123                  * Clear apic_assist portion of f(struct hv_vp_assist_page
1124                  * only, there can be valuable data in the rest which needs
1125                  * to be preserved e.g. on migration.
1126                  */
1127                 if (__clear_user((void __user *)addr, sizeof(u32)))
1128                         return 1;
1129                 hv_vcpu->hv_vapic = data;
1130                 kvm_vcpu_mark_page_dirty(vcpu, gfn);
1131                 if (kvm_lapic_enable_pv_eoi(vcpu,
1132                                             gfn_to_gpa(gfn) | KVM_MSR_ENABLED,
1133                                             sizeof(struct hv_vp_assist_page)))
1134                         return 1;
1135                 break;
1136         }
1137         case HV_X64_MSR_EOI:
1138                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1139         case HV_X64_MSR_ICR:
1140                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1141         case HV_X64_MSR_TPR:
1142                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1143         case HV_X64_MSR_VP_RUNTIME:
1144                 if (!host)
1145                         return 1;
1146                 hv_vcpu->runtime_offset = data - current_task_runtime_100ns();
1147                 break;
1148         case HV_X64_MSR_SCONTROL:
1149         case HV_X64_MSR_SVERSION:
1150         case HV_X64_MSR_SIEFP:
1151         case HV_X64_MSR_SIMP:
1152         case HV_X64_MSR_EOM:
1153         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1154                 return synic_set_msr(vcpu_to_synic(vcpu), msr, data, host);
1155         case HV_X64_MSR_STIMER0_CONFIG:
1156         case HV_X64_MSR_STIMER1_CONFIG:
1157         case HV_X64_MSR_STIMER2_CONFIG:
1158         case HV_X64_MSR_STIMER3_CONFIG: {
1159                 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1160
1161                 return stimer_set_config(vcpu_to_stimer(vcpu, timer_index),
1162                                          data, host);
1163         }
1164         case HV_X64_MSR_STIMER0_COUNT:
1165         case HV_X64_MSR_STIMER1_COUNT:
1166         case HV_X64_MSR_STIMER2_COUNT:
1167         case HV_X64_MSR_STIMER3_COUNT: {
1168                 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1169
1170                 return stimer_set_count(vcpu_to_stimer(vcpu, timer_index),
1171                                         data, host);
1172         }
1173         case HV_X64_MSR_TSC_FREQUENCY:
1174         case HV_X64_MSR_APIC_FREQUENCY:
1175                 /* read-only, but still ignore it if host-initiated */
1176                 if (!host)
1177                         return 1;
1178                 break;
1179         default:
1180                 vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
1181                             msr, data);
1182                 return 1;
1183         }
1184
1185         return 0;
1186 }
1187
1188 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1189 {
1190         u64 data = 0;
1191         struct kvm *kvm = vcpu->kvm;
1192         struct kvm_hv *hv = &kvm->arch.hyperv;
1193
1194         switch (msr) {
1195         case HV_X64_MSR_GUEST_OS_ID:
1196                 data = hv->hv_guest_os_id;
1197                 break;
1198         case HV_X64_MSR_HYPERCALL:
1199                 data = hv->hv_hypercall;
1200                 break;
1201         case HV_X64_MSR_TIME_REF_COUNT:
1202                 data = get_time_ref_counter(kvm);
1203                 break;
1204         case HV_X64_MSR_REFERENCE_TSC:
1205                 data = hv->hv_tsc_page;
1206                 break;
1207         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1208                 return kvm_hv_msr_get_crash_data(vcpu,
1209                                                  msr - HV_X64_MSR_CRASH_P0,
1210                                                  pdata);
1211         case HV_X64_MSR_CRASH_CTL:
1212                 return kvm_hv_msr_get_crash_ctl(vcpu, pdata);
1213         case HV_X64_MSR_RESET:
1214                 data = 0;
1215                 break;
1216         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1217                 data = hv->hv_reenlightenment_control;
1218                 break;
1219         case HV_X64_MSR_TSC_EMULATION_CONTROL:
1220                 data = hv->hv_tsc_emulation_control;
1221                 break;
1222         case HV_X64_MSR_TSC_EMULATION_STATUS:
1223                 data = hv->hv_tsc_emulation_status;
1224                 break;
1225         default:
1226                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1227                 return 1;
1228         }
1229
1230         *pdata = data;
1231         return 0;
1232 }
1233
1234 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1235                           bool host)
1236 {
1237         u64 data = 0;
1238         struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
1239
1240         switch (msr) {
1241         case HV_X64_MSR_VP_INDEX:
1242                 data = hv_vcpu->vp_index;
1243                 break;
1244         case HV_X64_MSR_EOI:
1245                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1246         case HV_X64_MSR_ICR:
1247                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1248         case HV_X64_MSR_TPR:
1249                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1250         case HV_X64_MSR_VP_ASSIST_PAGE:
1251                 data = hv_vcpu->hv_vapic;
1252                 break;
1253         case HV_X64_MSR_VP_RUNTIME:
1254                 data = current_task_runtime_100ns() + hv_vcpu->runtime_offset;
1255                 break;
1256         case HV_X64_MSR_SCONTROL:
1257         case HV_X64_MSR_SVERSION:
1258         case HV_X64_MSR_SIEFP:
1259         case HV_X64_MSR_SIMP:
1260         case HV_X64_MSR_EOM:
1261         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1262                 return synic_get_msr(vcpu_to_synic(vcpu), msr, pdata, host);
1263         case HV_X64_MSR_STIMER0_CONFIG:
1264         case HV_X64_MSR_STIMER1_CONFIG:
1265         case HV_X64_MSR_STIMER2_CONFIG:
1266         case HV_X64_MSR_STIMER3_CONFIG: {
1267                 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1268
1269                 return stimer_get_config(vcpu_to_stimer(vcpu, timer_index),
1270                                          pdata);
1271         }
1272         case HV_X64_MSR_STIMER0_COUNT:
1273         case HV_X64_MSR_STIMER1_COUNT:
1274         case HV_X64_MSR_STIMER2_COUNT:
1275         case HV_X64_MSR_STIMER3_COUNT: {
1276                 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1277
1278                 return stimer_get_count(vcpu_to_stimer(vcpu, timer_index),
1279                                         pdata);
1280         }
1281         case HV_X64_MSR_TSC_FREQUENCY:
1282                 data = (u64)vcpu->arch.virtual_tsc_khz * 1000;
1283                 break;
1284         case HV_X64_MSR_APIC_FREQUENCY:
1285                 data = APIC_BUS_FREQUENCY;
1286                 break;
1287         default:
1288                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1289                 return 1;
1290         }
1291         *pdata = data;
1292         return 0;
1293 }
1294
1295 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1296 {
1297         if (kvm_hv_msr_partition_wide(msr)) {
1298                 int r;
1299
1300                 mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1301                 r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1302                 mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1303                 return r;
1304         } else
1305                 return kvm_hv_set_msr(vcpu, msr, data, host);
1306 }
1307
1308 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
1309 {
1310         if (kvm_hv_msr_partition_wide(msr)) {
1311                 int r;
1312
1313                 mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1314                 r = kvm_hv_get_msr_pw(vcpu, msr, pdata);
1315                 mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1316                 return r;
1317         } else
1318                 return kvm_hv_get_msr(vcpu, msr, pdata, host);
1319 }
1320
1321 static __always_inline unsigned long *sparse_set_to_vcpu_mask(
1322         struct kvm *kvm, u64 *sparse_banks, u64 valid_bank_mask,
1323         u64 *vp_bitmap, unsigned long *vcpu_bitmap)
1324 {
1325         struct kvm_hv *hv = &kvm->arch.hyperv;
1326         struct kvm_vcpu *vcpu;
1327         int i, bank, sbank = 0;
1328
1329         memset(vp_bitmap, 0,
1330                KVM_HV_MAX_SPARSE_VCPU_SET_BITS * sizeof(*vp_bitmap));
1331         for_each_set_bit(bank, (unsigned long *)&valid_bank_mask,
1332                          KVM_HV_MAX_SPARSE_VCPU_SET_BITS)
1333                 vp_bitmap[bank] = sparse_banks[sbank++];
1334
1335         if (likely(!atomic_read(&hv->num_mismatched_vp_indexes))) {
1336                 /* for all vcpus vp_index == vcpu_idx */
1337                 return (unsigned long *)vp_bitmap;
1338         }
1339
1340         bitmap_zero(vcpu_bitmap, KVM_MAX_VCPUS);
1341         kvm_for_each_vcpu(i, vcpu, kvm) {
1342                 if (test_bit(vcpu_to_hv_vcpu(vcpu)->vp_index,
1343                              (unsigned long *)vp_bitmap))
1344                         __set_bit(i, vcpu_bitmap);
1345         }
1346         return vcpu_bitmap;
1347 }
1348
1349 static u64 kvm_hv_flush_tlb(struct kvm_vcpu *current_vcpu, u64 ingpa,
1350                             u16 rep_cnt, bool ex)
1351 {
1352         struct kvm *kvm = current_vcpu->kvm;
1353         struct kvm_vcpu_hv *hv_vcpu = &current_vcpu->arch.hyperv;
1354         struct hv_tlb_flush_ex flush_ex;
1355         struct hv_tlb_flush flush;
1356         u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1357         DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1358         unsigned long *vcpu_mask;
1359         u64 valid_bank_mask;
1360         u64 sparse_banks[64];
1361         int sparse_banks_len;
1362         bool all_cpus;
1363
1364         if (!ex) {
1365                 if (unlikely(kvm_read_guest(kvm, ingpa, &flush, sizeof(flush))))
1366                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
1367
1368                 trace_kvm_hv_flush_tlb(flush.processor_mask,
1369                                        flush.address_space, flush.flags);
1370
1371                 valid_bank_mask = BIT_ULL(0);
1372                 sparse_banks[0] = flush.processor_mask;
1373
1374                 /*
1375                  * Work around possible WS2012 bug: it sends hypercalls
1376                  * with processor_mask = 0x0 and HV_FLUSH_ALL_PROCESSORS clear,
1377                  * while also expecting us to flush something and crashing if
1378                  * we don't. Let's treat processor_mask == 0 same as
1379                  * HV_FLUSH_ALL_PROCESSORS.
1380                  */
1381                 all_cpus = (flush.flags & HV_FLUSH_ALL_PROCESSORS) ||
1382                         flush.processor_mask == 0;
1383         } else {
1384                 if (unlikely(kvm_read_guest(kvm, ingpa, &flush_ex,
1385                                             sizeof(flush_ex))))
1386                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
1387
1388                 trace_kvm_hv_flush_tlb_ex(flush_ex.hv_vp_set.valid_bank_mask,
1389                                           flush_ex.hv_vp_set.format,
1390                                           flush_ex.address_space,
1391                                           flush_ex.flags);
1392
1393                 valid_bank_mask = flush_ex.hv_vp_set.valid_bank_mask;
1394                 all_cpus = flush_ex.hv_vp_set.format !=
1395                         HV_GENERIC_SET_SPARSE_4K;
1396
1397                 sparse_banks_len =
1398                         bitmap_weight((unsigned long *)&valid_bank_mask, 64) *
1399                         sizeof(sparse_banks[0]);
1400
1401                 if (!sparse_banks_len && !all_cpus)
1402                         goto ret_success;
1403
1404                 if (!all_cpus &&
1405                     kvm_read_guest(kvm,
1406                                    ingpa + offsetof(struct hv_tlb_flush_ex,
1407                                                     hv_vp_set.bank_contents),
1408                                    sparse_banks,
1409                                    sparse_banks_len))
1410                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
1411         }
1412
1413         cpumask_clear(&hv_vcpu->tlb_flush);
1414
1415         vcpu_mask = all_cpus ? NULL :
1416                 sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1417                                         vp_bitmap, vcpu_bitmap);
1418
1419         /*
1420          * vcpu->arch.cr3 may not be up-to-date for running vCPUs so we can't
1421          * analyze it here, flush TLB regardless of the specified address space.
1422          */
1423         kvm_make_vcpus_request_mask(kvm,
1424                                     KVM_REQ_TLB_FLUSH | KVM_REQUEST_NO_WAKEUP,
1425                                     vcpu_mask, &hv_vcpu->tlb_flush);
1426
1427 ret_success:
1428         /* We always do full TLB flush, set rep_done = rep_cnt. */
1429         return (u64)HV_STATUS_SUCCESS |
1430                 ((u64)rep_cnt << HV_HYPERCALL_REP_COMP_OFFSET);
1431 }
1432
1433 static void kvm_send_ipi_to_many(struct kvm *kvm, u32 vector,
1434                                  unsigned long *vcpu_bitmap)
1435 {
1436         struct kvm_lapic_irq irq = {
1437                 .delivery_mode = APIC_DM_FIXED,
1438                 .vector = vector
1439         };
1440         struct kvm_vcpu *vcpu;
1441         int i;
1442
1443         kvm_for_each_vcpu(i, vcpu, kvm) {
1444                 if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
1445                         continue;
1446
1447                 /* We fail only when APIC is disabled */
1448                 kvm_apic_set_irq(vcpu, &irq, NULL);
1449         }
1450 }
1451
1452 static u64 kvm_hv_send_ipi(struct kvm_vcpu *current_vcpu, u64 ingpa, u64 outgpa,
1453                            bool ex, bool fast)
1454 {
1455         struct kvm *kvm = current_vcpu->kvm;
1456         struct hv_send_ipi_ex send_ipi_ex;
1457         struct hv_send_ipi send_ipi;
1458         u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1459         DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1460         unsigned long *vcpu_mask;
1461         unsigned long valid_bank_mask;
1462         u64 sparse_banks[64];
1463         int sparse_banks_len;
1464         u32 vector;
1465         bool all_cpus;
1466
1467         if (!ex) {
1468                 if (!fast) {
1469                         if (unlikely(kvm_read_guest(kvm, ingpa, &send_ipi,
1470                                                     sizeof(send_ipi))))
1471                                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1472                         sparse_banks[0] = send_ipi.cpu_mask;
1473                         vector = send_ipi.vector;
1474                 } else {
1475                         /* 'reserved' part of hv_send_ipi should be 0 */
1476                         if (unlikely(ingpa >> 32 != 0))
1477                                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1478                         sparse_banks[0] = outgpa;
1479                         vector = (u32)ingpa;
1480                 }
1481                 all_cpus = false;
1482                 valid_bank_mask = BIT_ULL(0);
1483
1484                 trace_kvm_hv_send_ipi(vector, sparse_banks[0]);
1485         } else {
1486                 if (unlikely(kvm_read_guest(kvm, ingpa, &send_ipi_ex,
1487                                             sizeof(send_ipi_ex))))
1488                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
1489
1490                 trace_kvm_hv_send_ipi_ex(send_ipi_ex.vector,
1491                                          send_ipi_ex.vp_set.format,
1492                                          send_ipi_ex.vp_set.valid_bank_mask);
1493
1494                 vector = send_ipi_ex.vector;
1495                 valid_bank_mask = send_ipi_ex.vp_set.valid_bank_mask;
1496                 sparse_banks_len = bitmap_weight(&valid_bank_mask, 64) *
1497                         sizeof(sparse_banks[0]);
1498
1499                 all_cpus = send_ipi_ex.vp_set.format == HV_GENERIC_SET_ALL;
1500
1501                 if (!sparse_banks_len)
1502                         goto ret_success;
1503
1504                 if (!all_cpus &&
1505                     kvm_read_guest(kvm,
1506                                    ingpa + offsetof(struct hv_send_ipi_ex,
1507                                                     vp_set.bank_contents),
1508                                    sparse_banks,
1509                                    sparse_banks_len))
1510                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
1511         }
1512
1513         if ((vector < HV_IPI_LOW_VECTOR) || (vector > HV_IPI_HIGH_VECTOR))
1514                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1515
1516         vcpu_mask = all_cpus ? NULL :
1517                 sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1518                                         vp_bitmap, vcpu_bitmap);
1519
1520         kvm_send_ipi_to_many(kvm, vector, vcpu_mask);
1521
1522 ret_success:
1523         return HV_STATUS_SUCCESS;
1524 }
1525
1526 bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1527 {
1528         return READ_ONCE(kvm->arch.hyperv.hv_hypercall) & HV_X64_MSR_HYPERCALL_ENABLE;
1529 }
1530
1531 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
1532 {
1533         bool longmode;
1534
1535         longmode = is_64_bit_mode(vcpu);
1536         if (longmode)
1537                 kvm_rax_write(vcpu, result);
1538         else {
1539                 kvm_rdx_write(vcpu, result >> 32);
1540                 kvm_rax_write(vcpu, result & 0xffffffff);
1541         }
1542 }
1543
1544 static int kvm_hv_hypercall_complete(struct kvm_vcpu *vcpu, u64 result)
1545 {
1546         kvm_hv_hypercall_set_result(vcpu, result);
1547         ++vcpu->stat.hypercalls;
1548         return kvm_skip_emulated_instruction(vcpu);
1549 }
1550
1551 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
1552 {
1553         return kvm_hv_hypercall_complete(vcpu, vcpu->run->hyperv.u.hcall.result);
1554 }
1555
1556 static u16 kvm_hvcall_signal_event(struct kvm_vcpu *vcpu, bool fast, u64 param)
1557 {
1558         struct eventfd_ctx *eventfd;
1559
1560         if (unlikely(!fast)) {
1561                 int ret;
1562                 gpa_t gpa = param;
1563
1564                 if ((gpa & (__alignof__(param) - 1)) ||
1565                     offset_in_page(gpa) + sizeof(param) > PAGE_SIZE)
1566                         return HV_STATUS_INVALID_ALIGNMENT;
1567
1568                 ret = kvm_vcpu_read_guest(vcpu, gpa, &param, sizeof(param));
1569                 if (ret < 0)
1570                         return HV_STATUS_INVALID_ALIGNMENT;
1571         }
1572
1573         /*
1574          * Per spec, bits 32-47 contain the extra "flag number".  However, we
1575          * have no use for it, and in all known usecases it is zero, so just
1576          * report lookup failure if it isn't.
1577          */
1578         if (param & 0xffff00000000ULL)
1579                 return HV_STATUS_INVALID_PORT_ID;
1580         /* remaining bits are reserved-zero */
1581         if (param & ~KVM_HYPERV_CONN_ID_MASK)
1582                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1583
1584         /* the eventfd is protected by vcpu->kvm->srcu, but conn_to_evt isn't */
1585         rcu_read_lock();
1586         eventfd = idr_find(&vcpu->kvm->arch.hyperv.conn_to_evt, param);
1587         rcu_read_unlock();
1588         if (!eventfd)
1589                 return HV_STATUS_INVALID_PORT_ID;
1590
1591         eventfd_signal(eventfd, 1);
1592         return HV_STATUS_SUCCESS;
1593 }
1594
1595 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
1596 {
1597         u64 param, ingpa, outgpa, ret = HV_STATUS_SUCCESS;
1598         uint16_t code, rep_idx, rep_cnt;
1599         bool fast, rep;
1600
1601         /*
1602          * hypercall generates UD from non zero cpl and real mode
1603          * per HYPER-V spec
1604          */
1605         if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
1606                 kvm_queue_exception(vcpu, UD_VECTOR);
1607                 return 1;
1608         }
1609
1610 #ifdef CONFIG_X86_64
1611         if (is_64_bit_mode(vcpu)) {
1612                 param = kvm_rcx_read(vcpu);
1613                 ingpa = kvm_rdx_read(vcpu);
1614                 outgpa = kvm_r8_read(vcpu);
1615         } else
1616 #endif
1617         {
1618                 param = ((u64)kvm_rdx_read(vcpu) << 32) |
1619                         (kvm_rax_read(vcpu) & 0xffffffff);
1620                 ingpa = ((u64)kvm_rbx_read(vcpu) << 32) |
1621                         (kvm_rcx_read(vcpu) & 0xffffffff);
1622                 outgpa = ((u64)kvm_rdi_read(vcpu) << 32) |
1623                         (kvm_rsi_read(vcpu) & 0xffffffff);
1624         }
1625
1626         code = param & 0xffff;
1627         fast = !!(param & HV_HYPERCALL_FAST_BIT);
1628         rep_cnt = (param >> HV_HYPERCALL_REP_COMP_OFFSET) & 0xfff;
1629         rep_idx = (param >> HV_HYPERCALL_REP_START_OFFSET) & 0xfff;
1630         rep = !!(rep_cnt || rep_idx);
1631
1632         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
1633
1634         switch (code) {
1635         case HVCALL_NOTIFY_LONG_SPIN_WAIT:
1636                 if (unlikely(rep)) {
1637                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1638                         break;
1639                 }
1640                 kvm_vcpu_on_spin(vcpu, true);
1641                 break;
1642         case HVCALL_SIGNAL_EVENT:
1643                 if (unlikely(rep)) {
1644                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1645                         break;
1646                 }
1647                 ret = kvm_hvcall_signal_event(vcpu, fast, ingpa);
1648                 if (ret != HV_STATUS_INVALID_PORT_ID)
1649                         break;
1650                 /* fall through - maybe userspace knows this conn_id. */
1651         case HVCALL_POST_MESSAGE:
1652                 /* don't bother userspace if it has no way to handle it */
1653                 if (unlikely(rep || !vcpu_to_synic(vcpu)->active)) {
1654                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1655                         break;
1656                 }
1657                 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
1658                 vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
1659                 vcpu->run->hyperv.u.hcall.input = param;
1660                 vcpu->run->hyperv.u.hcall.params[0] = ingpa;
1661                 vcpu->run->hyperv.u.hcall.params[1] = outgpa;
1662                 vcpu->arch.complete_userspace_io =
1663                                 kvm_hv_hypercall_complete_userspace;
1664                 return 0;
1665         case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
1666                 if (unlikely(fast || !rep_cnt || rep_idx)) {
1667                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1668                         break;
1669                 }
1670                 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false);
1671                 break;
1672         case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
1673                 if (unlikely(fast || rep)) {
1674                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1675                         break;
1676                 }
1677                 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false);
1678                 break;
1679         case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
1680                 if (unlikely(fast || !rep_cnt || rep_idx)) {
1681                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1682                         break;
1683                 }
1684                 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
1685                 break;
1686         case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
1687                 if (unlikely(fast || rep)) {
1688                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1689                         break;
1690                 }
1691                 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
1692                 break;
1693         case HVCALL_SEND_IPI:
1694                 if (unlikely(rep)) {
1695                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1696                         break;
1697                 }
1698                 ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, false, fast);
1699                 break;
1700         case HVCALL_SEND_IPI_EX:
1701                 if (unlikely(fast || rep)) {
1702                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1703                         break;
1704                 }
1705                 ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, true, false);
1706                 break;
1707         default:
1708                 ret = HV_STATUS_INVALID_HYPERCALL_CODE;
1709                 break;
1710         }
1711
1712         return kvm_hv_hypercall_complete(vcpu, ret);
1713 }
1714
1715 void kvm_hv_init_vm(struct kvm *kvm)
1716 {
1717         mutex_init(&kvm->arch.hyperv.hv_lock);
1718         idr_init(&kvm->arch.hyperv.conn_to_evt);
1719 }
1720
1721 void kvm_hv_destroy_vm(struct kvm *kvm)
1722 {
1723         struct eventfd_ctx *eventfd;
1724         int i;
1725
1726         idr_for_each_entry(&kvm->arch.hyperv.conn_to_evt, eventfd, i)
1727                 eventfd_ctx_put(eventfd);
1728         idr_destroy(&kvm->arch.hyperv.conn_to_evt);
1729 }
1730
1731 static int kvm_hv_eventfd_assign(struct kvm *kvm, u32 conn_id, int fd)
1732 {
1733         struct kvm_hv *hv = &kvm->arch.hyperv;
1734         struct eventfd_ctx *eventfd;
1735         int ret;
1736
1737         eventfd = eventfd_ctx_fdget(fd);
1738         if (IS_ERR(eventfd))
1739                 return PTR_ERR(eventfd);
1740
1741         mutex_lock(&hv->hv_lock);
1742         ret = idr_alloc(&hv->conn_to_evt, eventfd, conn_id, conn_id + 1,
1743                         GFP_KERNEL_ACCOUNT);
1744         mutex_unlock(&hv->hv_lock);
1745
1746         if (ret >= 0)
1747                 return 0;
1748
1749         if (ret == -ENOSPC)
1750                 ret = -EEXIST;
1751         eventfd_ctx_put(eventfd);
1752         return ret;
1753 }
1754
1755 static int kvm_hv_eventfd_deassign(struct kvm *kvm, u32 conn_id)
1756 {
1757         struct kvm_hv *hv = &kvm->arch.hyperv;
1758         struct eventfd_ctx *eventfd;
1759
1760         mutex_lock(&hv->hv_lock);
1761         eventfd = idr_remove(&hv->conn_to_evt, conn_id);
1762         mutex_unlock(&hv->hv_lock);
1763
1764         if (!eventfd)
1765                 return -ENOENT;
1766
1767         synchronize_srcu(&kvm->srcu);
1768         eventfd_ctx_put(eventfd);
1769         return 0;
1770 }
1771
1772 int kvm_vm_ioctl_hv_eventfd(struct kvm *kvm, struct kvm_hyperv_eventfd *args)
1773 {
1774         if ((args->flags & ~KVM_HYPERV_EVENTFD_DEASSIGN) ||
1775             (args->conn_id & ~KVM_HYPERV_CONN_ID_MASK))
1776                 return -EINVAL;
1777
1778         if (args->flags == KVM_HYPERV_EVENTFD_DEASSIGN)
1779                 return kvm_hv_eventfd_deassign(kvm, args->conn_id);
1780         return kvm_hv_eventfd_assign(kvm, args->conn_id, args->fd);
1781 }
1782
1783 int kvm_vcpu_ioctl_get_hv_cpuid(struct kvm_vcpu *vcpu, struct kvm_cpuid2 *cpuid,
1784                                 struct kvm_cpuid_entry2 __user *entries)
1785 {
1786         uint16_t evmcs_ver = 0;
1787         struct kvm_cpuid_entry2 cpuid_entries[] = {
1788                 { .function = HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS },
1789                 { .function = HYPERV_CPUID_INTERFACE },
1790                 { .function = HYPERV_CPUID_VERSION },
1791                 { .function = HYPERV_CPUID_FEATURES },
1792                 { .function = HYPERV_CPUID_ENLIGHTMENT_INFO },
1793                 { .function = HYPERV_CPUID_IMPLEMENT_LIMITS },
1794                 { .function = HYPERV_CPUID_NESTED_FEATURES },
1795         };
1796         int i, nent = ARRAY_SIZE(cpuid_entries);
1797
1798         if (kvm_x86_ops->nested_get_evmcs_version)
1799                 evmcs_ver = kvm_x86_ops->nested_get_evmcs_version(vcpu);
1800
1801         /* Skip NESTED_FEATURES if eVMCS is not supported */
1802         if (!evmcs_ver)
1803                 --nent;
1804
1805         if (cpuid->nent < nent)
1806                 return -E2BIG;
1807
1808         if (cpuid->nent > nent)
1809                 cpuid->nent = nent;
1810
1811         for (i = 0; i < nent; i++) {
1812                 struct kvm_cpuid_entry2 *ent = &cpuid_entries[i];
1813                 u32 signature[3];
1814
1815                 switch (ent->function) {
1816                 case HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS:
1817                         memcpy(signature, "Linux KVM Hv", 12);
1818
1819                         ent->eax = HYPERV_CPUID_NESTED_FEATURES;
1820                         ent->ebx = signature[0];
1821                         ent->ecx = signature[1];
1822                         ent->edx = signature[2];
1823                         break;
1824
1825                 case HYPERV_CPUID_INTERFACE:
1826                         memcpy(signature, "Hv#1\0\0\0\0\0\0\0\0", 12);
1827                         ent->eax = signature[0];
1828                         break;
1829
1830                 case HYPERV_CPUID_VERSION:
1831                         /*
1832                          * We implement some Hyper-V 2016 functions so let's use
1833                          * this version.
1834                          */
1835                         ent->eax = 0x00003839;
1836                         ent->ebx = 0x000A0000;
1837                         break;
1838
1839                 case HYPERV_CPUID_FEATURES:
1840                         ent->eax |= HV_X64_MSR_VP_RUNTIME_AVAILABLE;
1841                         ent->eax |= HV_MSR_TIME_REF_COUNT_AVAILABLE;
1842                         ent->eax |= HV_X64_MSR_SYNIC_AVAILABLE;
1843                         ent->eax |= HV_MSR_SYNTIMER_AVAILABLE;
1844                         ent->eax |= HV_X64_MSR_APIC_ACCESS_AVAILABLE;
1845                         ent->eax |= HV_X64_MSR_HYPERCALL_AVAILABLE;
1846                         ent->eax |= HV_X64_MSR_VP_INDEX_AVAILABLE;
1847                         ent->eax |= HV_X64_MSR_RESET_AVAILABLE;
1848                         ent->eax |= HV_MSR_REFERENCE_TSC_AVAILABLE;
1849                         ent->eax |= HV_X64_ACCESS_FREQUENCY_MSRS;
1850                         ent->eax |= HV_X64_ACCESS_REENLIGHTENMENT;
1851
1852                         ent->ebx |= HV_X64_POST_MESSAGES;
1853                         ent->ebx |= HV_X64_SIGNAL_EVENTS;
1854
1855                         ent->edx |= HV_FEATURE_FREQUENCY_MSRS_AVAILABLE;
1856                         ent->edx |= HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
1857
1858                         /*
1859                          * Direct Synthetic timers only make sense with in-kernel
1860                          * LAPIC
1861                          */
1862                         if (lapic_in_kernel(vcpu))
1863                                 ent->edx |= HV_STIMER_DIRECT_MODE_AVAILABLE;
1864
1865                         break;
1866
1867                 case HYPERV_CPUID_ENLIGHTMENT_INFO:
1868                         ent->eax |= HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
1869                         ent->eax |= HV_X64_APIC_ACCESS_RECOMMENDED;
1870                         ent->eax |= HV_X64_RELAXED_TIMING_RECOMMENDED;
1871                         ent->eax |= HV_X64_CLUSTER_IPI_RECOMMENDED;
1872                         ent->eax |= HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED;
1873                         if (evmcs_ver)
1874                                 ent->eax |= HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
1875
1876                         /*
1877                          * Default number of spinlock retry attempts, matches
1878                          * HyperV 2016.
1879                          */
1880                         ent->ebx = 0x00000FFF;
1881
1882                         break;
1883
1884                 case HYPERV_CPUID_IMPLEMENT_LIMITS:
1885                         /* Maximum number of virtual processors */
1886                         ent->eax = KVM_MAX_VCPUS;
1887                         /*
1888                          * Maximum number of logical processors, matches
1889                          * HyperV 2016.
1890                          */
1891                         ent->ebx = 64;
1892
1893                         break;
1894
1895                 case HYPERV_CPUID_NESTED_FEATURES:
1896                         ent->eax = evmcs_ver;
1897
1898                         break;
1899
1900                 default:
1901                         break;
1902                 }
1903         }
1904
1905         if (copy_to_user(entries, cpuid_entries,
1906                          nent * sizeof(struct kvm_cpuid_entry2)))
1907                 return -EFAULT;
1908
1909         return 0;
1910 }