arch/x86/kvm/hyperv.c

   1 /*
   2  * KVM Microsoft Hyper-V emulation
   3  *
   4  * derived from arch/x86/kvm/x86.c
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  * Copyright (C) 2008 Qumranet, Inc.
   8  * Copyright IBM Corporation, 2008
   9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
  10  * Copyright (C) 2015 Andrey Smetanin <asmetanin@virtuozzo.com>
  11  *
  12  * Authors:
  13  *   Avi Kivity   <avi@qumranet.com>
  14  *   Yaniv Kamay  <yaniv@qumranet.com>
  15  *   Amit Shah    <amit.shah@qumranet.com>
  16  *   Ben-Ami Yassour <benami@il.ibm.com>
  17  *   Andrey Smetanin <asmetanin@virtuozzo.com>
  18  *
  19  * This work is licensed under the terms of the GNU GPL, version 2.  See
  20  * the COPYING file in the top-level directory.
  21  *
  22  */
  23
  24 #include "x86.h"
  25 #include "lapic.h"
  26 #include "ioapic.h"
  27 #include "hyperv.h"
  28
  29 #include <linux/kvm_host.h>
  30 #include <linux/highmem.h>
  31 #include <linux/sched/cputime.h>
  32 #include <linux/eventfd.h>
  33
  34 #include <asm/apicdef.h>
  35 #include <trace/events/kvm.h>
  36
  37 #include "trace.h"
  38
  39 static inline u64 synic_read_sint(struct kvm_vcpu_hv_synic *synic, int sint)
  40 {
  41         return atomic64_read(&synic->sint[sint]);
  42 }
  43
  44 static inline int synic_get_sint_vector(u64 sint_value)
  45 {
  46         if (sint_value & HV_SYNIC_SINT_MASKED)
  47                 return -1;
  48         return sint_value & HV_SYNIC_SINT_VECTOR_MASK;
  49 }
  50
  51 static bool synic_has_vector_connected(struct kvm_vcpu_hv_synic *synic,
  52                                       int vector)
  53 {
  54         int i;
  55
  56         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
  57                 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
  58                         return true;
  59         }
  60         return false;
  61 }
  62
  63 static bool synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic *synic,
  64                                      int vector)
  65 {
  66         int i;
  67         u64 sint_value;
  68
  69         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
  70                 sint_value = synic_read_sint(synic, i);
  71                 if (synic_get_sint_vector(sint_value) == vector &&
  72                     sint_value & HV_SYNIC_SINT_AUTO_EOI)
  73                         return true;
  74         }
  75         return false;
  76 }
  77
  78 static void synic_update_vector(struct kvm_vcpu_hv_synic *synic,
  79                                 int vector)
  80 {
  81         if (vector < HV_SYNIC_FIRST_VALID_VECTOR)
  82                 return;
  83
  84         if (synic_has_vector_connected(synic, vector))
  85                 __set_bit(vector, synic->vec_bitmap);
  86         else
  87                 __clear_bit(vector, synic->vec_bitmap);
  88
  89         if (synic_has_vector_auto_eoi(synic, vector))
  90                 __set_bit(vector, synic->auto_eoi_bitmap);
  91         else
  92                 __clear_bit(vector, synic->auto_eoi_bitmap);
  93 }
  94
  95 static int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint,
  96                           u64 data, bool host)
  97 {
  98         int vector, old_vector;
  99         bool masked;
 100
 101         vector = data & HV_SYNIC_SINT_VECTOR_MASK;
 102         masked = data & HV_SYNIC_SINT_MASKED;
 103
 104         /*
 105          * Valid vectors are 16-255, however, nested Hyper-V attempts to write
 106          * default '0x10000' value on boot and this should not #GP. We need to
 107          * allow zero-initing the register from host as well.
 108          */
 109         if (vector < HV_SYNIC_FIRST_VALID_VECTOR && !host && !masked)
 110                 return 1;
 111         /*
 112          * Guest may configure multiple SINTs to use the same vector, so
 113          * we maintain a bitmap of vectors handled by synic, and a
 114          * bitmap of vectors with auto-eoi behavior.  The bitmaps are
 115          * updated here, and atomically queried on fast paths.
 116          */
 117         old_vector = synic_read_sint(synic, sint) & HV_SYNIC_SINT_VECTOR_MASK;
 118
 119         atomic64_set(&synic->sint[sint], data);
 120
 121         synic_update_vector(synic, old_vector);
 122
 123         synic_update_vector(synic, vector);
 124
 125         /* Load SynIC vectors into EOI exit bitmap */
 126         kvm_make_request(KVM_REQ_SCAN_IOAPIC, synic_to_vcpu(synic));
 127         return 0;
 128 }
 129
 130 static struct kvm_vcpu *get_vcpu_by_vpidx(struct kvm *kvm, u32 vpidx)
 131 {
 132         struct kvm_vcpu *vcpu = NULL;
 133         int i;
 134
 135         if (vpidx >= KVM_MAX_VCPUS)
 136                 return NULL;
 137
 138         vcpu = kvm_get_vcpu(kvm, vpidx);
 139         if (vcpu && vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
 140                 return vcpu;
 141         kvm_for_each_vcpu(i, vcpu, kvm)
 142                 if (vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
 143                         return vcpu;
 144         return NULL;
 145 }
 146
 147 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vpidx)
 148 {
 149         struct kvm_vcpu *vcpu;
 150         struct kvm_vcpu_hv_synic *synic;
 151
 152         vcpu = get_vcpu_by_vpidx(kvm, vpidx);
 153         if (!vcpu)
 154                 return NULL;
 155         synic = vcpu_to_synic(vcpu);
 156         return (synic->active) ? synic : NULL;
 157 }
 158
 159 static void synic_clear_sint_msg_pending(struct kvm_vcpu_hv_synic *synic,
 160                                         u32 sint)
 161 {
 162         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 163         struct page *page;
 164         gpa_t gpa;
 165         struct hv_message *msg;
 166         struct hv_message_page *msg_page;
 167
 168         gpa = synic->msg_page & PAGE_MASK;
 169         page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
 170         if (is_error_page(page)) {
 171                 vcpu_err(vcpu, "Hyper-V SynIC can't get msg page, gpa 0x%llx\n",
 172                          gpa);
 173                 return;
 174         }
 175         msg_page = kmap_atomic(page);
 176
 177         msg = &msg_page->sint_message[sint];
 178         msg->header.message_flags.msg_pending = 0;
 179
 180         kunmap_atomic(msg_page);
 181         kvm_release_page_dirty(page);
 182         kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
 183 }
 184
 185 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
 186 {
 187         struct kvm *kvm = vcpu->kvm;
 188         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
 189         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 190         struct kvm_vcpu_hv_stimer *stimer;
 191         int gsi, idx, stimers_pending;
 192
 193         trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
 194
 195         if (synic->msg_page & HV_SYNIC_SIMP_ENABLE)
 196                 synic_clear_sint_msg_pending(synic, sint);
 197
 198         /* Try to deliver pending Hyper-V SynIC timers messages */
 199         stimers_pending = 0;
 200         for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
 201                 stimer = &hv_vcpu->stimer[idx];
 202                 if (stimer->msg_pending &&
 203                     (stimer->config & HV_STIMER_ENABLE) &&
 204                     HV_STIMER_SINT(stimer->config) == sint) {
 205                         set_bit(stimer->index,
 206                                 hv_vcpu->stimer_pending_bitmap);
 207                         stimers_pending++;
 208                 }
 209         }
 210         if (stimers_pending)
 211                 kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
 212
 213         idx = srcu_read_lock(&kvm->irq_srcu);
 214         gsi = atomic_read(&synic->sint_to_gsi[sint]);
 215         if (gsi != -1)
 216                 kvm_notify_acked_gsi(kvm, gsi);
 217         srcu_read_unlock(&kvm->irq_srcu, idx);
 218 }
 219
 220 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
 221 {
 222         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 223         struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
 224
 225         hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
 226         hv_vcpu->exit.u.synic.msr = msr;
 227         hv_vcpu->exit.u.synic.control = synic->control;
 228         hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
 229         hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
 230
 231         kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
 232 }
 233
 234 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
 235                          u32 msr, u64 data, bool host)
 236 {
 237         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 238         int ret;
 239
 240         if (!synic->active && !host)
 241                 return 1;
 242
 243         trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
 244
 245         ret = 0;
 246         switch (msr) {
 247         case HV_X64_MSR_SCONTROL:
 248                 synic->control = data;
 249                 if (!host)
 250                         synic_exit(synic, msr);
 251                 break;
 252         case HV_X64_MSR_SVERSION:
 253                 if (!host) {
 254                         ret = 1;
 255                         break;
 256                 }
 257                 synic->version = data;
 258                 break;
 259         case HV_X64_MSR_SIEFP:
 260                 if ((data & HV_SYNIC_SIEFP_ENABLE) && !host &&
 261                     !synic->dont_zero_synic_pages)
 262                         if (kvm_clear_guest(vcpu->kvm,
 263                                             data & PAGE_MASK, PAGE_SIZE)) {
 264                                 ret = 1;
 265                                 break;
 266                         }
 267                 synic->evt_page = data;
 268                 if (!host)
 269                         synic_exit(synic, msr);
 270                 break;
 271         case HV_X64_MSR_SIMP:
 272                 if ((data & HV_SYNIC_SIMP_ENABLE) && !host &&
 273                     !synic->dont_zero_synic_pages)
 274                         if (kvm_clear_guest(vcpu->kvm,
 275                                             data & PAGE_MASK, PAGE_SIZE)) {
 276                                 ret = 1;
 277                                 break;
 278                         }
 279                 synic->msg_page = data;
 280                 if (!host)
 281                         synic_exit(synic, msr);
 282                 break;
 283         case HV_X64_MSR_EOM: {
 284                 int i;
 285
 286                 for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
 287                         kvm_hv_notify_acked_sint(vcpu, i);
 288                 break;
 289         }
 290         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
 291                 ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
 292                 break;
 293         default:
 294                 ret = 1;
 295                 break;
 296         }
 297         return ret;
 298 }
 299
 300 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata,
 301                          bool host)
 302 {
 303         int ret;
 304
 305         if (!synic->active && !host)
 306                 return 1;
 307
 308         ret = 0;
 309         switch (msr) {
 310         case HV_X64_MSR_SCONTROL:
 311                 *pdata = synic->control;
 312                 break;
 313         case HV_X64_MSR_SVERSION:
 314                 *pdata = synic->version;
 315                 break;
 316         case HV_X64_MSR_SIEFP:
 317                 *pdata = synic->evt_page;
 318                 break;
 319         case HV_X64_MSR_SIMP:
 320                 *pdata = synic->msg_page;
 321                 break;
 322         case HV_X64_MSR_EOM:
 323                 *pdata = 0;
 324                 break;
 325         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
 326                 *pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
 327                 break;
 328         default:
 329                 ret = 1;
 330                 break;
 331         }
 332         return ret;
 333 }
 334
 335 static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
 336 {
 337         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 338         struct kvm_lapic_irq irq;
 339         int ret, vector;
 340
 341         if (sint >= ARRAY_SIZE(synic->sint))
 342                 return -EINVAL;
 343
 344         vector = synic_get_sint_vector(synic_read_sint(synic, sint));
 345         if (vector < 0)
 346                 return -ENOENT;
 347
 348         memset(&irq, 0, sizeof(irq));
 349         irq.shorthand = APIC_DEST_SELF;
 350         irq.dest_mode = APIC_DEST_PHYSICAL;
 351         irq.delivery_mode = APIC_DM_FIXED;
 352         irq.vector = vector;
 353         irq.level = 1;
 354
 355         ret = kvm_irq_delivery_to_apic(vcpu->kvm, vcpu->arch.apic, &irq, NULL);
 356         trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
 357         return ret;
 358 }
 359
 360 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vpidx, u32 sint)
 361 {
 362         struct kvm_vcpu_hv_synic *synic;
 363
 364         synic = synic_get(kvm, vpidx);
 365         if (!synic)
 366                 return -EINVAL;
 367
 368         return synic_set_irq(synic, sint);
 369 }
 370
 371 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
 372 {
 373         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
 374         int i;
 375
 376         trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
 377
 378         for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
 379                 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
 380                         kvm_hv_notify_acked_sint(vcpu, i);
 381 }
 382
 383 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi)
 384 {
 385         struct kvm_vcpu_hv_synic *synic;
 386
 387         synic = synic_get(kvm, vpidx);
 388         if (!synic)
 389                 return -EINVAL;
 390
 391         if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
 392                 return -EINVAL;
 393
 394         atomic_set(&synic->sint_to_gsi[sint], gsi);
 395         return 0;
 396 }
 397
 398 void kvm_hv_irq_routing_update(struct kvm *kvm)
 399 {
 400         struct kvm_irq_routing_table *irq_rt;
 401         struct kvm_kernel_irq_routing_entry *e;
 402         u32 gsi;
 403
 404         irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
 405                                         lockdep_is_held(&kvm->irq_lock));
 406
 407         for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
 408                 hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
 409                         if (e->type == KVM_IRQ_ROUTING_HV_SINT)
 410                                 kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
 411                                                     e->hv_sint.sint, gsi);
 412                 }
 413         }
 414 }
 415
 416 static void synic_init(struct kvm_vcpu_hv_synic *synic)
 417 {
 418         int i;
 419
 420         memset(synic, 0, sizeof(*synic));
 421         synic->version = HV_SYNIC_VERSION_1;
 422         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
 423                 atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
 424                 atomic_set(&synic->sint_to_gsi[i], -1);
 425         }
 426 }
 427
 428 static u64 get_time_ref_counter(struct kvm *kvm)
 429 {
 430         struct kvm_hv *hv = &kvm->arch.hyperv;
 431         struct kvm_vcpu *vcpu;
 432         u64 tsc;
 433
 434         /*
 435          * The guest has not set up the TSC page or the clock isn't
 436          * stable, fall back to get_kvmclock_ns.
 437          */
 438         if (!hv->tsc_ref.tsc_sequence)
 439                 return div_u64(get_kvmclock_ns(kvm), 100);
 440
 441         vcpu = kvm_get_vcpu(kvm, 0);
 442         tsc = kvm_read_l1_tsc(vcpu, rdtsc());
 443         return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64)
 444                 + hv->tsc_ref.tsc_offset;
 445 }
 446
 447 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
 448                                 bool vcpu_kick)
 449 {
 450         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
 451
 452         set_bit(stimer->index,
 453                 vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
 454         kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
 455         if (vcpu_kick)
 456                 kvm_vcpu_kick(vcpu);
 457 }
 458
 459 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
 460 {
 461         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
 462
 463         trace_kvm_hv_stimer_cleanup(stimer_to_vcpu(stimer)->vcpu_id,
 464                                     stimer->index);
 465
 466         hrtimer_cancel(&stimer->timer);
 467         clear_bit(stimer->index,
 468                   vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
 469         stimer->msg_pending = false;
 470         stimer->exp_time = 0;
 471 }
 472
 473 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
 474 {
 475         struct kvm_vcpu_hv_stimer *stimer;
 476
 477         stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
 478         trace_kvm_hv_stimer_callback(stimer_to_vcpu(stimer)->vcpu_id,
 479                                      stimer->index);
 480         stimer_mark_pending(stimer, true);
 481
 482         return HRTIMER_NORESTART;
 483 }
 484
 485 /*
 486  * stimer_start() assumptions:
 487  * a) stimer->count is not equal to 0
 488  * b) stimer->config has HV_STIMER_ENABLE flag
 489  */
 490 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
 491 {
 492         u64 time_now;
 493         ktime_t ktime_now;
 494
 495         time_now = get_time_ref_counter(stimer_to_vcpu(stimer)->kvm);
 496         ktime_now = ktime_get();
 497
 498         if (stimer->config & HV_STIMER_PERIODIC) {
 499                 if (stimer->exp_time) {
 500                         if (time_now >= stimer->exp_time) {
 501                                 u64 remainder;
 502
 503                                 div64_u64_rem(time_now - stimer->exp_time,
 504                                               stimer->count, &remainder);
 505                                 stimer->exp_time =
 506                                         time_now + (stimer->count - remainder);
 507                         }
 508                 } else
 509                         stimer->exp_time = time_now + stimer->count;
 510
 511                 trace_kvm_hv_stimer_start_periodic(
 512                                         stimer_to_vcpu(stimer)->vcpu_id,
 513                                         stimer->index,
 514                                         time_now, stimer->exp_time);
 515
 516                 hrtimer_start(&stimer->timer,
 517                               ktime_add_ns(ktime_now,
 518                                            100 * (stimer->exp_time - time_now)),
 519                               HRTIMER_MODE_ABS);
 520                 return 0;
 521         }
 522         stimer->exp_time = stimer->count;
 523         if (time_now >= stimer->count) {
 524                 /*
 525                  * Expire timer according to Hypervisor Top-Level Functional
 526                  * specification v4(15.3.1):
 527                  * "If a one shot is enabled and the specified count is in
 528                  * the past, it will expire immediately."
 529                  */
 530                 stimer_mark_pending(stimer, false);
 531                 return 0;
 532         }
 533
 534         trace_kvm_hv_stimer_start_one_shot(stimer_to_vcpu(stimer)->vcpu_id,
 535                                            stimer->index,
 536                                            time_now, stimer->count);
 537
 538         hrtimer_start(&stimer->timer,
 539                       ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)),
 540                       HRTIMER_MODE_ABS);
 541         return 0;
 542 }
 543
 544 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
 545                              bool host)
 546 {
 547         trace_kvm_hv_stimer_set_config(stimer_to_vcpu(stimer)->vcpu_id,
 548                                        stimer->index, config, host);
 549
 550         stimer_cleanup(stimer);
 551         if ((stimer->config & HV_STIMER_ENABLE) && HV_STIMER_SINT(config) == 0)
 552                 config &= ~HV_STIMER_ENABLE;
 553         stimer->config = config;
 554         stimer_mark_pending(stimer, false);
 555         return 0;
 556 }
 557
 558 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
 559                             bool host)
 560 {
 561         trace_kvm_hv_stimer_set_count(stimer_to_vcpu(stimer)->vcpu_id,
 562                                       stimer->index, count, host);
 563
 564         stimer_cleanup(stimer);
 565         stimer->count = count;
 566         if (stimer->count == 0)
 567                 stimer->config &= ~HV_STIMER_ENABLE;
 568         else if (stimer->config & HV_STIMER_AUTOENABLE)
 569                 stimer->config |= HV_STIMER_ENABLE;
 570         stimer_mark_pending(stimer, false);
 571         return 0;
 572 }
 573
 574 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
 575 {
 576         *pconfig = stimer->config;
 577         return 0;
 578 }
 579
 580 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
 581 {
 582         *pcount = stimer->count;
 583         return 0;
 584 }
 585
 586 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
 587                              struct hv_message *src_msg)
 588 {
 589         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 590         struct page *page;
 591         gpa_t gpa;
 592         struct hv_message *dst_msg;
 593         int r;
 594         struct hv_message_page *msg_page;
 595
 596         if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
 597                 return -ENOENT;
 598
 599         gpa = synic->msg_page & PAGE_MASK;
 600         page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
 601         if (is_error_page(page))
 602                 return -EFAULT;
 603
 604         msg_page = kmap_atomic(page);
 605         dst_msg = &msg_page->sint_message[sint];
 606         if (sync_cmpxchg(&dst_msg->header.message_type, HVMSG_NONE,
 607                          src_msg->header.message_type) != HVMSG_NONE) {
 608                 dst_msg->header.message_flags.msg_pending = 1;
 609                 r = -EAGAIN;
 610         } else {
 611                 memcpy(&dst_msg->u.payload, &src_msg->u.payload,
 612                        src_msg->header.payload_size);
 613                 dst_msg->header.message_type = src_msg->header.message_type;
 614                 dst_msg->header.payload_size = src_msg->header.payload_size;
 615                 r = synic_set_irq(synic, sint);
 616                 if (r >= 1)
 617                         r = 0;
 618                 else if (r == 0)
 619                         r = -EFAULT;
 620         }
 621         kunmap_atomic(msg_page);
 622         kvm_release_page_dirty(page);
 623         kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
 624         return r;
 625 }
 626
 627 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
 628 {
 629         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
 630         struct hv_message *msg = &stimer->msg;
 631         struct hv_timer_message_payload *payload =
 632                         (struct hv_timer_message_payload *)&msg->u.payload;
 633
 634         payload->expiration_time = stimer->exp_time;
 635         payload->delivery_time = get_time_ref_counter(vcpu->kvm);
 636         return synic_deliver_msg(vcpu_to_synic(vcpu),
 637                                  HV_STIMER_SINT(stimer->config), msg);
 638 }
 639
 640 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
 641 {
 642         int r;
 643
 644         stimer->msg_pending = true;
 645         r = stimer_send_msg(stimer);
 646         trace_kvm_hv_stimer_expiration(stimer_to_vcpu(stimer)->vcpu_id,
 647                                        stimer->index, r);
 648         if (!r) {
 649                 stimer->msg_pending = false;
 650                 if (!(stimer->config & HV_STIMER_PERIODIC))
 651                         stimer->config &= ~HV_STIMER_ENABLE;
 652         }
 653 }
 654
 655 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
 656 {
 657         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 658         struct kvm_vcpu_hv_stimer *stimer;
 659         u64 time_now, exp_time;
 660         int i;
 661
 662         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 663                 if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
 664                         stimer = &hv_vcpu->stimer[i];
 665                         if (stimer->config & HV_STIMER_ENABLE) {
 666                                 exp_time = stimer->exp_time;
 667
 668                                 if (exp_time) {
 669                                         time_now =
 670                                                 get_time_ref_counter(vcpu->kvm);
 671                                         if (time_now >= exp_time)
 672                                                 stimer_expiration(stimer);
 673                                 }
 674
 675                                 if ((stimer->config & HV_STIMER_ENABLE) &&
 676                                     stimer->count) {
 677                                         if (!stimer->msg_pending)
 678                                                 stimer_start(stimer);
 679                                 } else
 680                                         stimer_cleanup(stimer);
 681                         }
 682                 }
 683 }
 684
 685 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
 686 {
 687         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 688         int i;
 689
 690         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 691                 stimer_cleanup(&hv_vcpu->stimer[i]);
 692 }
 693
 694 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
 695 {
 696         struct hv_message *msg = &stimer->msg;
 697         struct hv_timer_message_payload *payload =
 698                         (struct hv_timer_message_payload *)&msg->u.payload;
 699
 700         memset(&msg->header, 0, sizeof(msg->header));
 701         msg->header.message_type = HVMSG_TIMER_EXPIRED;
 702         msg->header.payload_size = sizeof(*payload);
 703
 704         payload->timer_index = stimer->index;
 705         payload->expiration_time = 0;
 706         payload->delivery_time = 0;
 707 }
 708
 709 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
 710 {
 711         memset(stimer, 0, sizeof(*stimer));
 712         stimer->index = timer_index;
 713         hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
 714         stimer->timer.function = stimer_timer_callback;
 715         stimer_prepare_msg(stimer);
 716 }
 717
 718 void kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
 719 {
 720         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 721         int i;
 722
 723         synic_init(&hv_vcpu->synic);
 724
 725         bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
 726         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 727                 stimer_init(&hv_vcpu->stimer[i], i);
 728 }
 729
 730 void kvm_hv_vcpu_postcreate(struct kvm_vcpu *vcpu)
 731 {
 732         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 733
 734         hv_vcpu->vp_index = kvm_vcpu_get_idx(vcpu);
 735 }
 736
 737 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
 738 {
 739         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
 740
 741         /*
 742          * Hyper-V SynIC auto EOI SINT's are
 743          * not compatible with APICV, so deactivate APICV
 744          */
 745         kvm_vcpu_deactivate_apicv(vcpu);
 746         synic->active = true;
 747         synic->dont_zero_synic_pages = dont_zero_synic_pages;
 748         return 0;
 749 }
 750
 751 static bool kvm_hv_msr_partition_wide(u32 msr)
 752 {
 753         bool r = false;
 754
 755         switch (msr) {
 756         case HV_X64_MSR_GUEST_OS_ID:
 757         case HV_X64_MSR_HYPERCALL:
 758         case HV_X64_MSR_REFERENCE_TSC:
 759         case HV_X64_MSR_TIME_REF_COUNT:
 760         case HV_X64_MSR_CRASH_CTL:
 761         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
 762         case HV_X64_MSR_RESET:
 763         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
 764         case HV_X64_MSR_TSC_EMULATION_CONTROL:
 765         case HV_X64_MSR_TSC_EMULATION_STATUS:
 766                 r = true;
 767                 break;
 768         }
 769
 770         return r;
 771 }
 772
 773 static int kvm_hv_msr_get_crash_data(struct kvm_vcpu *vcpu,
 774                                      u32 index, u64 *pdata)
 775 {
 776         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 777
 778         if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param)))
 779                 return -EINVAL;
 780
 781         *pdata = hv->hv_crash_param[index];
 782         return 0;
 783 }
 784
 785 static int kvm_hv_msr_get_crash_ctl(struct kvm_vcpu *vcpu, u64 *pdata)
 786 {
 787         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 788
 789         *pdata = hv->hv_crash_ctl;
 790         return 0;
 791 }
 792
 793 static int kvm_hv_msr_set_crash_ctl(struct kvm_vcpu *vcpu, u64 data, bool host)
 794 {
 795         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 796
 797         if (host)
 798                 hv->hv_crash_ctl = data & HV_X64_MSR_CRASH_CTL_NOTIFY;
 799
 800         if (!host && (data & HV_X64_MSR_CRASH_CTL_NOTIFY)) {
 801
 802                 vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
 803                           hv->hv_crash_param[0],
 804                           hv->hv_crash_param[1],
 805                           hv->hv_crash_param[2],
 806                           hv->hv_crash_param[3],
 807                           hv->hv_crash_param[4]);
 808
 809                 /* Send notification about crash to user space */
 810                 kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
 811         }
 812
 813         return 0;
 814 }
 815
 816 static int kvm_hv_msr_set_crash_data(struct kvm_vcpu *vcpu,
 817                                      u32 index, u64 data)
 818 {
 819         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 820
 821         if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param)))
 822                 return -EINVAL;
 823
 824         hv->hv_crash_param[index] = data;
 825         return 0;
 826 }
 827
 828 /*
 829  * The kvmclock and Hyper-V TSC page use similar formulas, and converting
 830  * between them is possible:
 831  *
 832  * kvmclock formula:
 833  *    nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
 834  *           + system_time
 835  *
 836  * Hyper-V formula:
 837  *    nsec/100 = ticks * scale / 2^64 + offset
 838  *
 839  * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
 840  * By dividing the kvmclock formula by 100 and equating what's left we get:
 841  *    ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
 842  *            scale / 2^64 =         tsc_to_system_mul * 2^(tsc_shift-32) / 100
 843  *            scale        =         tsc_to_system_mul * 2^(32+tsc_shift) / 100
 844  *
 845  * Now expand the kvmclock formula and divide by 100:
 846  *    nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
 847  *           - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
 848  *           + system_time
 849  *    nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
 850  *               - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
 851  *               + system_time / 100
 852  *
 853  * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
 854  *    nsec/100 = ticks * scale / 2^64
 855  *               - tsc_timestamp * scale / 2^64
 856  *               + system_time / 100
 857  *
 858  * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
 859  *    offset = system_time / 100 - tsc_timestamp * scale / 2^64
 860  *
 861  * These two equivalencies are implemented in this function.
 862  */
 863 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
 864                                         HV_REFERENCE_TSC_PAGE *tsc_ref)
 865 {
 866         u64 max_mul;
 867
 868         if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
 869                 return false;
 870
 871         /*
 872          * check if scale would overflow, if so we use the time ref counter
 873          *    tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
 874          *    tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
 875          *    tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
 876          */
 877         max_mul = 100ull << (32 - hv_clock->tsc_shift);
 878         if (hv_clock->tsc_to_system_mul >= max_mul)
 879                 return false;
 880
 881         /*
 882          * Otherwise compute the scale and offset according to the formulas
 883          * derived above.
 884          */
 885         tsc_ref->tsc_scale =
 886                 mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
 887                                 hv_clock->tsc_to_system_mul,
 888                                 100);
 889
 890         tsc_ref->tsc_offset = hv_clock->system_time;
 891         do_div(tsc_ref->tsc_offset, 100);
 892         tsc_ref->tsc_offset -=
 893                 mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
 894         return true;
 895 }
 896
 897 void kvm_hv_setup_tsc_page(struct kvm *kvm,
 898                            struct pvclock_vcpu_time_info *hv_clock)
 899 {
 900         struct kvm_hv *hv = &kvm->arch.hyperv;
 901         u32 tsc_seq;
 902         u64 gfn;
 903
 904         BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
 905         BUILD_BUG_ON(offsetof(HV_REFERENCE_TSC_PAGE, tsc_sequence) != 0);
 906
 907         if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
 908                 return;
 909
 910         mutex_lock(&kvm->arch.hyperv.hv_lock);
 911         if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
 912                 goto out_unlock;
 913
 914         gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
 915         /*
 916          * Because the TSC parameters only vary when there is a
 917          * change in the master clock, do not bother with caching.
 918          */
 919         if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
 920                                     &tsc_seq, sizeof(tsc_seq))))
 921                 goto out_unlock;
 922
 923         /*
 924          * While we're computing and writing the parameters, force the
 925          * guest to use the time reference count MSR.
 926          */
 927         hv->tsc_ref.tsc_sequence = 0;
 928         if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
 929                             &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
 930                 goto out_unlock;
 931
 932         if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
 933                 goto out_unlock;
 934
 935         /* Ensure sequence is zero before writing the rest of the struct.  */
 936         smp_wmb();
 937         if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
 938                 goto out_unlock;
 939
 940         /*
 941          * Now switch to the TSC page mechanism by writing the sequence.
 942          */
 943         tsc_seq++;
 944         if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
 945                 tsc_seq = 1;
 946
 947         /* Write the struct entirely before the non-zero sequence.  */
 948         smp_wmb();
 949
 950         hv->tsc_ref.tsc_sequence = tsc_seq;
 951         kvm_write_guest(kvm, gfn_to_gpa(gfn),
 952                         &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence));
 953 out_unlock:
 954         mutex_unlock(&kvm->arch.hyperv.hv_lock);
 955 }
 956
 957 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
 958                              bool host)
 959 {
 960         struct kvm *kvm = vcpu->kvm;
 961         struct kvm_hv *hv = &kvm->arch.hyperv;
 962
 963         switch (msr) {
 964         case HV_X64_MSR_GUEST_OS_ID:
 965                 hv->hv_guest_os_id = data;
 966                 /* setting guest os id to zero disables hypercall page */
 967                 if (!hv->hv_guest_os_id)
 968                         hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
 969                 break;
 970         case HV_X64_MSR_HYPERCALL: {
 971                 u64 gfn;
 972                 unsigned long addr;
 973                 u8 instructions[4];
 974
 975                 /* if guest os id is not set hypercall should remain disabled */
 976                 if (!hv->hv_guest_os_id)
 977                         break;
 978                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
 979                         hv->hv_hypercall = data;
 980                         break;
 981                 }
 982                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
 983                 addr = gfn_to_hva(kvm, gfn);
 984                 if (kvm_is_error_hva(addr))
 985                         return 1;
 986                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
 987                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
 988                 if (__copy_to_user((void __user *)addr, instructions, 4))
 989                         return 1;
 990                 hv->hv_hypercall = data;
 991                 mark_page_dirty(kvm, gfn);
 992                 break;
 993         }
 994         case HV_X64_MSR_REFERENCE_TSC:
 995                 hv->hv_tsc_page = data;
 996                 if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE)
 997                         kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
 998                 break;
 999         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1000                 return kvm_hv_msr_set_crash_data(vcpu,
1001                                                  msr - HV_X64_MSR_CRASH_P0,
1002                                                  data);
1003         case HV_X64_MSR_CRASH_CTL:
1004                 return kvm_hv_msr_set_crash_ctl(vcpu, data, host);
1005         case HV_X64_MSR_RESET:
1006                 if (data == 1) {
1007                         vcpu_debug(vcpu, "hyper-v reset requested\n");
1008                         kvm_make_request(KVM_REQ_HV_RESET, vcpu);
1009                 }
1010                 break;
1011         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1012                 hv->hv_reenlightenment_control = data;
1013                 break;
1014         case HV_X64_MSR_TSC_EMULATION_CONTROL:
1015                 hv->hv_tsc_emulation_control = data;
1016                 break;
1017         case HV_X64_MSR_TSC_EMULATION_STATUS:
1018                 hv->hv_tsc_emulation_status = data;
1019                 break;
1020         case HV_X64_MSR_TIME_REF_COUNT:
1021                 /* read-only, but still ignore it if host-initiated */
1022                 if (!host)
1023                         return 1;
1024                 break;
1025         default:
1026                 vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
1027                             msr, data);
1028                 return 1;
1029         }
1030         return 0;
1031 }
1032
1033 /* Calculate cpu time spent by current task in 100ns units */
1034 static u64 current_task_runtime_100ns(void)
1035 {
1036         u64 utime, stime;
1037
1038         task_cputime_adjusted(current, &utime, &stime);
1039
1040         return div_u64(utime + stime, 100);
1041 }
1042
1043 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1044 {
1045         struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
1046
1047         switch (msr) {
1048         case HV_X64_MSR_VP_INDEX:
1049                 if (!host || (u32)data >= KVM_MAX_VCPUS)
1050                         return 1;
1051                 hv->vp_index = (u32)data;
1052                 break;
1053         case HV_X64_MSR_VP_ASSIST_PAGE: {
1054                 u64 gfn;
1055                 unsigned long addr;
1056
1057                 if (!(data & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE)) {
1058                         hv->hv_vapic = data;
1059                         if (kvm_lapic_enable_pv_eoi(vcpu, 0))
1060                                 return 1;
1061                         break;
1062                 }
1063                 gfn = data >> HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT;
1064                 addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1065                 if (kvm_is_error_hva(addr))
1066                         return 1;
1067                 if (__clear_user((void __user *)addr, PAGE_SIZE))
1068                         return 1;
1069                 hv->hv_vapic = data;
1070                 kvm_vcpu_mark_page_dirty(vcpu, gfn);
1071                 if (kvm_lapic_enable_pv_eoi(vcpu,
1072                                             gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
1073                         return 1;
1074                 break;
1075         }
1076         case HV_X64_MSR_EOI:
1077                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1078         case HV_X64_MSR_ICR:
1079                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1080         case HV_X64_MSR_TPR:
1081                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1082         case HV_X64_MSR_VP_RUNTIME:
1083                 if (!host)
1084                         return 1;
1085                 hv->runtime_offset = data - current_task_runtime_100ns();
1086                 break;
1087         case HV_X64_MSR_SCONTROL:
1088         case HV_X64_MSR_SVERSION:
1089         case HV_X64_MSR_SIEFP:
1090         case HV_X64_MSR_SIMP:
1091         case HV_X64_MSR_EOM:
1092         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1093                 return synic_set_msr(vcpu_to_synic(vcpu), msr, data, host);
1094         case HV_X64_MSR_STIMER0_CONFIG:
1095         case HV_X64_MSR_STIMER1_CONFIG:
1096         case HV_X64_MSR_STIMER2_CONFIG:
1097         case HV_X64_MSR_STIMER3_CONFIG: {
1098                 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1099
1100                 return stimer_set_config(vcpu_to_stimer(vcpu, timer_index),
1101                                          data, host);
1102         }
1103         case HV_X64_MSR_STIMER0_COUNT:
1104         case HV_X64_MSR_STIMER1_COUNT:
1105         case HV_X64_MSR_STIMER2_COUNT:
1106         case HV_X64_MSR_STIMER3_COUNT: {
1107                 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1108
1109                 return stimer_set_count(vcpu_to_stimer(vcpu, timer_index),
1110                                         data, host);
1111         }
1112         case HV_X64_MSR_TSC_FREQUENCY:
1113         case HV_X64_MSR_APIC_FREQUENCY:
1114                 /* read-only, but still ignore it if host-initiated */
1115                 if (!host)
1116                         return 1;
1117                 break;
1118         default:
1119                 vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
1120                             msr, data);
1121                 return 1;
1122         }
1123
1124         return 0;
1125 }
1126
1127 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1128 {
1129         u64 data = 0;
1130         struct kvm *kvm = vcpu->kvm;
1131         struct kvm_hv *hv = &kvm->arch.hyperv;
1132
1133         switch (msr) {
1134         case HV_X64_MSR_GUEST_OS_ID:
1135                 data = hv->hv_guest_os_id;
1136                 break;
1137         case HV_X64_MSR_HYPERCALL:
1138                 data = hv->hv_hypercall;
1139                 break;
1140         case HV_X64_MSR_TIME_REF_COUNT:
1141                 data = get_time_ref_counter(kvm);
1142                 break;
1143         case HV_X64_MSR_REFERENCE_TSC:
1144                 data = hv->hv_tsc_page;
1145                 break;
1146         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1147                 return kvm_hv_msr_get_crash_data(vcpu,
1148                                                  msr - HV_X64_MSR_CRASH_P0,
1149                                                  pdata);
1150         case HV_X64_MSR_CRASH_CTL:
1151                 return kvm_hv_msr_get_crash_ctl(vcpu, pdata);
1152         case HV_X64_MSR_RESET:
1153                 data = 0;
1154                 break;
1155         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1156                 data = hv->hv_reenlightenment_control;
1157                 break;
1158         case HV_X64_MSR_TSC_EMULATION_CONTROL:
1159                 data = hv->hv_tsc_emulation_control;
1160                 break;
1161         case HV_X64_MSR_TSC_EMULATION_STATUS:
1162                 data = hv->hv_tsc_emulation_status;
1163                 break;
1164         default:
1165                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1166                 return 1;
1167         }
1168
1169         *pdata = data;
1170         return 0;
1171 }
1172
1173 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1174                           bool host)
1175 {
1176         u64 data = 0;
1177         struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
1178
1179         switch (msr) {
1180         case HV_X64_MSR_VP_INDEX:
1181                 data = hv->vp_index;
1182                 break;
1183         case HV_X64_MSR_EOI:
1184                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1185         case HV_X64_MSR_ICR:
1186                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1187         case HV_X64_MSR_TPR:
1188                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1189         case HV_X64_MSR_VP_ASSIST_PAGE:
1190                 data = hv->hv_vapic;
1191                 break;
1192         case HV_X64_MSR_VP_RUNTIME:
1193                 data = current_task_runtime_100ns() + hv->runtime_offset;
1194                 break;
1195         case HV_X64_MSR_SCONTROL:
1196         case HV_X64_MSR_SVERSION:
1197         case HV_X64_MSR_SIEFP:
1198         case HV_X64_MSR_SIMP:
1199         case HV_X64_MSR_EOM:
1200         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1201                 return synic_get_msr(vcpu_to_synic(vcpu), msr, pdata, host);
1202         case HV_X64_MSR_STIMER0_CONFIG:
1203         case HV_X64_MSR_STIMER1_CONFIG:
1204         case HV_X64_MSR_STIMER2_CONFIG:
1205         case HV_X64_MSR_STIMER3_CONFIG: {
1206                 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1207
1208                 return stimer_get_config(vcpu_to_stimer(vcpu, timer_index),
1209                                          pdata);
1210         }
1211         case HV_X64_MSR_STIMER0_COUNT:
1212         case HV_X64_MSR_STIMER1_COUNT:
1213         case HV_X64_MSR_STIMER2_COUNT:
1214         case HV_X64_MSR_STIMER3_COUNT: {
1215                 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1216
1217                 return stimer_get_count(vcpu_to_stimer(vcpu, timer_index),
1218                                         pdata);
1219         }
1220         case HV_X64_MSR_TSC_FREQUENCY:
1221                 data = (u64)vcpu->arch.virtual_tsc_khz * 1000;
1222                 break;
1223         case HV_X64_MSR_APIC_FREQUENCY:
1224                 data = APIC_BUS_FREQUENCY;
1225                 break;
1226         default:
1227                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1228                 return 1;
1229         }
1230         *pdata = data;
1231         return 0;
1232 }
1233
1234 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1235 {
1236         if (kvm_hv_msr_partition_wide(msr)) {
1237                 int r;
1238
1239                 mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1240                 r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1241                 mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1242                 return r;
1243         } else
1244                 return kvm_hv_set_msr(vcpu, msr, data, host);
1245 }
1246
1247 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
1248 {
1249         if (kvm_hv_msr_partition_wide(msr)) {
1250                 int r;
1251
1252                 mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1253                 r = kvm_hv_get_msr_pw(vcpu, msr, pdata);
1254                 mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1255                 return r;
1256         } else
1257                 return kvm_hv_get_msr(vcpu, msr, pdata, host);
1258 }
1259
1260 static __always_inline int get_sparse_bank_no(u64 valid_bank_mask, int bank_no)
1261 {
1262         int i = 0, j;
1263
1264         if (!(valid_bank_mask & BIT_ULL(bank_no)))
1265                 return -1;
1266
1267         for (j = 0; j < bank_no; j++)
1268                 if (valid_bank_mask & BIT_ULL(j))
1269                         i++;
1270
1271         return i;
1272 }
1273
1274 static u64 kvm_hv_flush_tlb(struct kvm_vcpu *current_vcpu, u64 ingpa,
1275                             u16 rep_cnt, bool ex)
1276 {
1277         struct kvm *kvm = current_vcpu->kvm;
1278         struct kvm_vcpu_hv *hv_current = &current_vcpu->arch.hyperv;
1279         struct hv_tlb_flush_ex flush_ex;
1280         struct hv_tlb_flush flush;
1281         struct kvm_vcpu *vcpu;
1282         unsigned long vcpu_bitmap[BITS_TO_LONGS(KVM_MAX_VCPUS)] = {0};
1283         unsigned long valid_bank_mask = 0;
1284         u64 sparse_banks[64];
1285         int sparse_banks_len, i;
1286         bool all_cpus;
1287
1288         if (!ex) {
1289                 if (unlikely(kvm_read_guest(kvm, ingpa, &flush, sizeof(flush))))
1290                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
1291
1292                 trace_kvm_hv_flush_tlb(flush.processor_mask,
1293                                        flush.address_space, flush.flags);
1294
1295                 sparse_banks[0] = flush.processor_mask;
1296                 all_cpus = flush.flags & HV_FLUSH_ALL_PROCESSORS;
1297         } else {
1298                 if (unlikely(kvm_read_guest(kvm, ingpa, &flush_ex,
1299                                             sizeof(flush_ex))))
1300                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
1301
1302                 trace_kvm_hv_flush_tlb_ex(flush_ex.hv_vp_set.valid_bank_mask,
1303                                           flush_ex.hv_vp_set.format,
1304                                           flush_ex.address_space,
1305                                           flush_ex.flags);
1306
1307                 valid_bank_mask = flush_ex.hv_vp_set.valid_bank_mask;
1308                 all_cpus = flush_ex.hv_vp_set.format !=
1309                         HV_GENERIC_SET_SPARSE_4K;
1310
1311                 sparse_banks_len = bitmap_weight(&valid_bank_mask, 64) *
1312                         sizeof(sparse_banks[0]);
1313
1314                 if (!sparse_banks_len && !all_cpus)
1315                         goto ret_success;
1316
1317                 if (!all_cpus &&
1318                     kvm_read_guest(kvm,
1319                                    ingpa + offsetof(struct hv_tlb_flush_ex,
1320                                                     hv_vp_set.bank_contents),
1321                                    sparse_banks,
1322                                    sparse_banks_len))
1323                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
1324         }
1325
1326         cpumask_clear(&hv_current->tlb_lush);
1327
1328         kvm_for_each_vcpu(i, vcpu, kvm) {
1329                 struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
1330                 int bank = hv->vp_index / 64, sbank = 0;
1331
1332                 if (!all_cpus) {
1333                         /* Banks >64 can't be represented */
1334                         if (bank >= 64)
1335                                 continue;
1336
1337                         /* Non-ex hypercalls can only address first 64 vCPUs */
1338                         if (!ex && bank)
1339                                 continue;
1340
1341                         if (ex) {
1342                                 /*
1343                                  * Check is the bank of this vCPU is in sparse
1344                                  * set and get the sparse bank number.
1345                                  */
1346                                 sbank = get_sparse_bank_no(valid_bank_mask,
1347                                                            bank);
1348
1349                                 if (sbank < 0)
1350                                         continue;
1351                         }
1352
1353                         if (!(sparse_banks[sbank] & BIT_ULL(hv->vp_index % 64)))
1354                                 continue;
1355                 }
1356
1357                 /*
1358                  * vcpu->arch.cr3 may not be up-to-date for running vCPUs so we
1359                  * can't analyze it here, flush TLB regardless of the specified
1360                  * address space.
1361                  */
1362                 __set_bit(i, vcpu_bitmap);
1363         }
1364
1365         kvm_make_vcpus_request_mask(kvm,
1366                                     KVM_REQ_TLB_FLUSH | KVM_REQUEST_NO_WAKEUP,
1367                                     vcpu_bitmap, &hv_current->tlb_lush);
1368
1369 ret_success:
1370         /* We always do full TLB flush, set rep_done = rep_cnt. */
1371         return (u64)HV_STATUS_SUCCESS |
1372                 ((u64)rep_cnt << HV_HYPERCALL_REP_COMP_OFFSET);
1373 }
1374
1375 bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1376 {
1377         return READ_ONCE(kvm->arch.hyperv.hv_hypercall) & HV_X64_MSR_HYPERCALL_ENABLE;
1378 }
1379
1380 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
1381 {
1382         bool longmode;
1383
1384         longmode = is_64_bit_mode(vcpu);
1385         if (longmode)
1386                 kvm_register_write(vcpu, VCPU_REGS_RAX, result);
1387         else {
1388                 kvm_register_write(vcpu, VCPU_REGS_RDX, result >> 32);
1389                 kvm_register_write(vcpu, VCPU_REGS_RAX, result & 0xffffffff);
1390         }
1391 }
1392
1393 static int kvm_hv_hypercall_complete(struct kvm_vcpu *vcpu, u64 result)
1394 {
1395         kvm_hv_hypercall_set_result(vcpu, result);
1396         ++vcpu->stat.hypercalls;
1397         return kvm_skip_emulated_instruction(vcpu);
1398 }
1399
1400 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
1401 {
1402         return kvm_hv_hypercall_complete(vcpu, vcpu->run->hyperv.u.hcall.result);
1403 }
1404
1405 static u16 kvm_hvcall_signal_event(struct kvm_vcpu *vcpu, bool fast, u64 param)
1406 {
1407         struct eventfd_ctx *eventfd;
1408
1409         if (unlikely(!fast)) {
1410                 int ret;
1411                 gpa_t gpa = param;
1412
1413                 if ((gpa & (__alignof__(param) - 1)) ||
1414                     offset_in_page(gpa) + sizeof(param) > PAGE_SIZE)
1415                         return HV_STATUS_INVALID_ALIGNMENT;
1416
1417                 ret = kvm_vcpu_read_guest(vcpu, gpa, &param, sizeof(param));
1418                 if (ret < 0)
1419                         return HV_STATUS_INVALID_ALIGNMENT;
1420         }
1421
1422         /*
1423          * Per spec, bits 32-47 contain the extra "flag number".  However, we
1424          * have no use for it, and in all known usecases it is zero, so just
1425          * report lookup failure if it isn't.
1426          */
1427         if (param & 0xffff00000000ULL)
1428                 return HV_STATUS_INVALID_PORT_ID;
1429         /* remaining bits are reserved-zero */
1430         if (param & ~KVM_HYPERV_CONN_ID_MASK)
1431                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1432
1433         /* the eventfd is protected by vcpu->kvm->srcu, but conn_to_evt isn't */
1434         rcu_read_lock();
1435         eventfd = idr_find(&vcpu->kvm->arch.hyperv.conn_to_evt, param);
1436         rcu_read_unlock();
1437         if (!eventfd)
1438                 return HV_STATUS_INVALID_PORT_ID;
1439
1440         eventfd_signal(eventfd, 1);
1441         return HV_STATUS_SUCCESS;
1442 }
1443
1444 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
1445 {
1446         u64 param, ingpa, outgpa, ret = HV_STATUS_SUCCESS;
1447         uint16_t code, rep_idx, rep_cnt;
1448         bool fast, longmode, rep;
1449
1450         /*
1451          * hypercall generates UD from non zero cpl and real mode
1452          * per HYPER-V spec
1453          */
1454         if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
1455                 kvm_queue_exception(vcpu, UD_VECTOR);
1456                 return 1;
1457         }
1458
1459         longmode = is_64_bit_mode(vcpu);
1460
1461         if (!longmode) {
1462                 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
1463                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
1464                 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
1465                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
1466                 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
1467                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
1468         }
1469 #ifdef CONFIG_X86_64
1470         else {
1471                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
1472                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
1473                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
1474         }
1475 #endif
1476
1477         code = param & 0xffff;
1478         fast = !!(param & HV_HYPERCALL_FAST_BIT);
1479         rep_cnt = (param >> HV_HYPERCALL_REP_COMP_OFFSET) & 0xfff;
1480         rep_idx = (param >> HV_HYPERCALL_REP_START_OFFSET) & 0xfff;
1481         rep = !!(rep_cnt || rep_idx);
1482
1483         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
1484
1485         switch (code) {
1486         case HVCALL_NOTIFY_LONG_SPIN_WAIT:
1487                 if (unlikely(rep)) {
1488                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1489                         break;
1490                 }
1491                 kvm_vcpu_on_spin(vcpu, true);
1492                 break;
1493         case HVCALL_SIGNAL_EVENT:
1494                 if (unlikely(rep)) {
1495                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1496                         break;
1497                 }
1498                 ret = kvm_hvcall_signal_event(vcpu, fast, ingpa);
1499                 if (ret != HV_STATUS_INVALID_PORT_ID)
1500                         break;
1501                 /* maybe userspace knows this conn_id: fall through */
1502         case HVCALL_POST_MESSAGE:
1503                 /* don't bother userspace if it has no way to handle it */
1504                 if (unlikely(rep || !vcpu_to_synic(vcpu)->active)) {
1505                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1506                         break;
1507                 }
1508                 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
1509                 vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
1510                 vcpu->run->hyperv.u.hcall.input = param;
1511                 vcpu->run->hyperv.u.hcall.params[0] = ingpa;
1512                 vcpu->run->hyperv.u.hcall.params[1] = outgpa;
1513                 vcpu->arch.complete_userspace_io =
1514                                 kvm_hv_hypercall_complete_userspace;
1515                 return 0;
1516         case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
1517                 if (unlikely(fast || !rep_cnt || rep_idx)) {
1518                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1519                         break;
1520                 }
1521                 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false);
1522                 break;
1523         case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
1524                 if (unlikely(fast || rep)) {
1525                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1526                         break;
1527                 }
1528                 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false);
1529                 break;
1530         case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
1531                 if (unlikely(fast || !rep_cnt || rep_idx)) {
1532                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1533                         break;
1534                 }
1535                 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
1536                 break;
1537         case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
1538                 if (unlikely(fast || rep)) {
1539                         ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1540                         break;
1541                 }
1542                 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
1543                 break;
1544         default:
1545                 ret = HV_STATUS_INVALID_HYPERCALL_CODE;
1546                 break;
1547         }
1548
1549         return kvm_hv_hypercall_complete(vcpu, ret);
1550 }
1551
1552 void kvm_hv_init_vm(struct kvm *kvm)
1553 {
1554         mutex_init(&kvm->arch.hyperv.hv_lock);
1555         idr_init(&kvm->arch.hyperv.conn_to_evt);
1556 }
1557
1558 void kvm_hv_destroy_vm(struct kvm *kvm)
1559 {
1560         struct eventfd_ctx *eventfd;
1561         int i;
1562
1563         idr_for_each_entry(&kvm->arch.hyperv.conn_to_evt, eventfd, i)
1564                 eventfd_ctx_put(eventfd);
1565         idr_destroy(&kvm->arch.hyperv.conn_to_evt);
1566 }
1567
1568 static int kvm_hv_eventfd_assign(struct kvm *kvm, u32 conn_id, int fd)
1569 {
1570         struct kvm_hv *hv = &kvm->arch.hyperv;
1571         struct eventfd_ctx *eventfd;
1572         int ret;
1573
1574         eventfd = eventfd_ctx_fdget(fd);
1575         if (IS_ERR(eventfd))
1576                 return PTR_ERR(eventfd);
1577
1578         mutex_lock(&hv->hv_lock);
1579         ret = idr_alloc(&hv->conn_to_evt, eventfd, conn_id, conn_id + 1,
1580                         GFP_KERNEL);
1581         mutex_unlock(&hv->hv_lock);
1582
1583         if (ret >= 0)
1584                 return 0;
1585
1586         if (ret == -ENOSPC)
1587                 ret = -EEXIST;
1588         eventfd_ctx_put(eventfd);
1589         return ret;
1590 }
1591
1592 static int kvm_hv_eventfd_deassign(struct kvm *kvm, u32 conn_id)
1593 {
1594         struct kvm_hv *hv = &kvm->arch.hyperv;
1595         struct eventfd_ctx *eventfd;
1596
1597         mutex_lock(&hv->hv_lock);
1598         eventfd = idr_remove(&hv->conn_to_evt, conn_id);
1599         mutex_unlock(&hv->hv_lock);
1600
1601         if (!eventfd)
1602                 return -ENOENT;
1603
1604         synchronize_srcu(&kvm->srcu);
1605         eventfd_ctx_put(eventfd);
1606         return 0;
1607 }
1608
1609 int kvm_vm_ioctl_hv_eventfd(struct kvm *kvm, struct kvm_hyperv_eventfd *args)
1610 {
1611         if ((args->flags & ~KVM_HYPERV_EVENTFD_DEASSIGN) ||
1612             (args->conn_id & ~KVM_HYPERV_CONN_ID_MASK))
1613                 return -EINVAL;
1614
1615         if (args->flags == KVM_HYPERV_EVENTFD_DEASSIGN)
1616                 return kvm_hv_eventfd_deassign(kvm, args->conn_id);
1617         return kvm_hv_eventfd_assign(kvm, args->conn_id, args->fd);
1618 }