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 <[email protected]>
  11  *
  12  * Authors:
  13  *   Avi Kivity   <[email protected]>
  14  *   Yaniv Kamay  <[email protected]>
  15  *   Amit Shah    <[email protected]>
  16  *   Ben-Ami Yassour <[email protected]>
  17  *   Andrey Smetanin <[email protected]>
  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                 vcpu = kvm_get_vcpu(kvm, vpidx);
 137         if (vcpu && vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
 138                 return vcpu;
 139         kvm_for_each_vcpu(i, vcpu, kvm)
 140                 if (vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
 141                         return vcpu;
 142         return NULL;
 143 }
 144
 145 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vpidx)
 146 {
 147         struct kvm_vcpu *vcpu;
 148         struct kvm_vcpu_hv_synic *synic;
 149
 150         vcpu = get_vcpu_by_vpidx(kvm, vpidx);
 151         if (!vcpu)
 152                 return NULL;
 153         synic = vcpu_to_synic(vcpu);
 154         return (synic->active) ? synic : NULL;
 155 }
 156
 157 static void synic_clear_sint_msg_pending(struct kvm_vcpu_hv_synic *synic,
 158                                         u32 sint)
 159 {
 160         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 161         struct page *page;
 162         gpa_t gpa;
 163         struct hv_message *msg;
 164         struct hv_message_page *msg_page;
 165
 166         gpa = synic->msg_page & PAGE_MASK;
 167         page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
 168         if (is_error_page(page)) {
 169                 vcpu_err(vcpu, "Hyper-V SynIC can't get msg page, gpa 0x%llx\n",
 170                          gpa);
 171                 return;
 172         }
 173         msg_page = kmap_atomic(page);
 174
 175         msg = &msg_page->sint_message[sint];
 176         msg->header.message_flags.msg_pending = 0;
 177
 178         kunmap_atomic(msg_page);
 179         kvm_release_page_dirty(page);
 180         kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
 181 }
 182
 183 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
 184 {
 185         struct kvm *kvm = vcpu->kvm;
 186         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
 187         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 188         struct kvm_vcpu_hv_stimer *stimer;
 189         int gsi, idx, stimers_pending;
 190
 191         trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
 192
 193         if (synic->msg_page & HV_SYNIC_SIMP_ENABLE)
 194                 synic_clear_sint_msg_pending(synic, sint);
 195
 196         /* Try to deliver pending Hyper-V SynIC timers messages */
 197         stimers_pending = 0;
 198         for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
 199                 stimer = &hv_vcpu->stimer[idx];
 200                 if (stimer->msg_pending &&
 201                     (stimer->config & HV_STIMER_ENABLE) &&
 202                     HV_STIMER_SINT(stimer->config) == sint) {
 203                         set_bit(stimer->index,
 204                                 hv_vcpu->stimer_pending_bitmap);
 205                         stimers_pending++;
 206                 }
 207         }
 208         if (stimers_pending)
 209                 kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
 210
 211         idx = srcu_read_lock(&kvm->irq_srcu);
 212         gsi = atomic_read(&synic->sint_to_gsi[sint]);
 213         if (gsi != -1)
 214                 kvm_notify_acked_gsi(kvm, gsi);
 215         srcu_read_unlock(&kvm->irq_srcu, idx);
 216 }
 217
 218 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
 219 {
 220         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 221         struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
 222
 223         hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
 224         hv_vcpu->exit.u.synic.msr = msr;
 225         hv_vcpu->exit.u.synic.control = synic->control;
 226         hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
 227         hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
 228
 229         kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
 230 }
 231
 232 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
 233                          u32 msr, u64 data, bool host)
 234 {
 235         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 236         int ret;
 237
 238         if (!synic->active)
 239                 return 1;
 240
 241         trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
 242
 243         ret = 0;
 244         switch (msr) {
 245         case HV_X64_MSR_SCONTROL:
 246                 synic->control = data;
 247                 if (!host)
 248                         synic_exit(synic, msr);
 249                 break;
 250         case HV_X64_MSR_SVERSION:
 251                 if (!host) {
 252                         ret = 1;
 253                         break;
 254                 }
 255                 synic->version = data;
 256                 break;
 257         case HV_X64_MSR_SIEFP:
 258                 if ((data & HV_SYNIC_SIEFP_ENABLE) && !host &&
 259                     !synic->dont_zero_synic_pages)
 260                         if (kvm_clear_guest(vcpu->kvm,
 261                                             data & PAGE_MASK, PAGE_SIZE)) {
 262                                 ret = 1;
 263                                 break;
 264                         }
 265                 synic->evt_page = data;
 266                 if (!host)
 267                         synic_exit(synic, msr);
 268                 break;
 269         case HV_X64_MSR_SIMP:
 270                 if ((data & HV_SYNIC_SIMP_ENABLE) && !host &&
 271                     !synic->dont_zero_synic_pages)
 272                         if (kvm_clear_guest(vcpu->kvm,
 273                                             data & PAGE_MASK, PAGE_SIZE)) {
 274                                 ret = 1;
 275                                 break;
 276                         }
 277                 synic->msg_page = data;
 278                 if (!host)
 279                         synic_exit(synic, msr);
 280                 break;
 281         case HV_X64_MSR_EOM: {
 282                 int i;
 283
 284                 for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
 285                         kvm_hv_notify_acked_sint(vcpu, i);
 286                 break;
 287         }
 288         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
 289                 ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
 290                 break;
 291         default:
 292                 ret = 1;
 293                 break;
 294         }
 295         return ret;
 296 }
 297
 298 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata)
 299 {
 300         int ret;
 301
 302         if (!synic->active)
 303                 return 1;
 304
 305         ret = 0;
 306         switch (msr) {
 307         case HV_X64_MSR_SCONTROL:
 308                 *pdata = synic->control;
 309                 break;
 310         case HV_X64_MSR_SVERSION:
 311                 *pdata = synic->version;
 312                 break;
 313         case HV_X64_MSR_SIEFP:
 314                 *pdata = synic->evt_page;
 315                 break;
 316         case HV_X64_MSR_SIMP:
 317                 *pdata = synic->msg_page;
 318                 break;
 319         case HV_X64_MSR_EOM:
 320                 *pdata = 0;
 321                 break;
 322         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
 323                 *pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
 324                 break;
 325         default:
 326                 ret = 1;
 327                 break;
 328         }
 329         return ret;
 330 }
 331
 332 static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
 333 {
 334         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 335         struct kvm_lapic_irq irq;
 336         int ret, vector;
 337
 338         if (sint >= ARRAY_SIZE(synic->sint))
 339                 return -EINVAL;
 340
 341         vector = synic_get_sint_vector(synic_read_sint(synic, sint));
 342         if (vector < 0)
 343                 return -ENOENT;
 344
 345         memset(&irq, 0, sizeof(irq));
 346         irq.shorthand = APIC_DEST_SELF;
 347         irq.dest_mode = APIC_DEST_PHYSICAL;
 348         irq.delivery_mode = APIC_DM_FIXED;
 349         irq.vector = vector;
 350         irq.level = 1;
 351
 352         ret = kvm_irq_delivery_to_apic(vcpu->kvm, vcpu->arch.apic, &irq, NULL);
 353         trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
 354         return ret;
 355 }
 356
 357 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vpidx, u32 sint)
 358 {
 359         struct kvm_vcpu_hv_synic *synic;
 360
 361         synic = synic_get(kvm, vpidx);
 362         if (!synic)
 363                 return -EINVAL;
 364
 365         return synic_set_irq(synic, sint);
 366 }
 367
 368 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
 369 {
 370         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
 371         int i;
 372
 373         trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
 374
 375         for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
 376                 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
 377                         kvm_hv_notify_acked_sint(vcpu, i);
 378 }
 379
 380 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi)
 381 {
 382         struct kvm_vcpu_hv_synic *synic;
 383
 384         synic = synic_get(kvm, vpidx);
 385         if (!synic)
 386                 return -EINVAL;
 387
 388         if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
 389                 return -EINVAL;
 390
 391         atomic_set(&synic->sint_to_gsi[sint], gsi);
 392         return 0;
 393 }
 394
 395 void kvm_hv_irq_routing_update(struct kvm *kvm)
 396 {
 397         struct kvm_irq_routing_table *irq_rt;
 398         struct kvm_kernel_irq_routing_entry *e;
 399         u32 gsi;
 400
 401         irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
 402                                         lockdep_is_held(&kvm->irq_lock));
 403
 404         for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
 405                 hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
 406                         if (e->type == KVM_IRQ_ROUTING_HV_SINT)
 407                                 kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
 408                                                     e->hv_sint.sint, gsi);
 409                 }
 410         }
 411 }
 412
 413 static void synic_init(struct kvm_vcpu_hv_synic *synic)
 414 {
 415         int i;
 416
 417         memset(synic, 0, sizeof(*synic));
 418         synic->version = HV_SYNIC_VERSION_1;
 419         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
 420                 atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
 421                 atomic_set(&synic->sint_to_gsi[i], -1);
 422         }
 423 }
 424
 425 static u64 get_time_ref_counter(struct kvm *kvm)
 426 {
 427         struct kvm_hv *hv = &kvm->arch.hyperv;
 428         struct kvm_vcpu *vcpu;
 429         u64 tsc;
 430
 431         /*
 432          * The guest has not set up the TSC page or the clock isn't
 433          * stable, fall back to get_kvmclock_ns.
 434          */
 435         if (!hv->tsc_ref.tsc_sequence)
 436                 return div_u64(get_kvmclock_ns(kvm), 100);
 437
 438         vcpu = kvm_get_vcpu(kvm, 0);
 439         tsc = kvm_read_l1_tsc(vcpu, rdtsc());
 440         return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64)
 441                 + hv->tsc_ref.tsc_offset;
 442 }
 443
 444 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
 445                                 bool vcpu_kick)
 446 {
 447         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
 448
 449         set_bit(stimer->index,
 450                 vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
 451         kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
 452         if (vcpu_kick)
 453                 kvm_vcpu_kick(vcpu);
 454 }
 455
 456 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
 457 {
 458         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
 459
 460         trace_kvm_hv_stimer_cleanup(stimer_to_vcpu(stimer)->vcpu_id,
 461                                     stimer->index);
 462
 463         hrtimer_cancel(&stimer->timer);
 464         clear_bit(stimer->index,
 465                   vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
 466         stimer->msg_pending = false;
 467         stimer->exp_time = 0;
 468 }
 469
 470 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
 471 {
 472         struct kvm_vcpu_hv_stimer *stimer;
 473
 474         stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
 475         trace_kvm_hv_stimer_callback(stimer_to_vcpu(stimer)->vcpu_id,
 476                                      stimer->index);
 477         stimer_mark_pending(stimer, true);
 478
 479         return HRTIMER_NORESTART;
 480 }
 481
 482 /*
 483  * stimer_start() assumptions:
 484  * a) stimer->count is not equal to 0
 485  * b) stimer->config has HV_STIMER_ENABLE flag
 486  */
 487 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
 488 {
 489         u64 time_now;
 490         ktime_t ktime_now;
 491
 492         time_now = get_time_ref_counter(stimer_to_vcpu(stimer)->kvm);
 493         ktime_now = ktime_get();
 494
 495         if (stimer->config & HV_STIMER_PERIODIC) {
 496                 if (stimer->exp_time) {
 497                         if (time_now >= stimer->exp_time) {
 498                                 u64 remainder;
 499
 500                                 div64_u64_rem(time_now - stimer->exp_time,
 501                                               stimer->count, &remainder);
 502                                 stimer->exp_time =
 503                                         time_now + (stimer->count - remainder);
 504                         }
 505                 } else
 506                         stimer->exp_time = time_now + stimer->count;
 507
 508                 trace_kvm_hv_stimer_start_periodic(
 509                                         stimer_to_vcpu(stimer)->vcpu_id,
 510                                         stimer->index,
 511                                         time_now, stimer->exp_time);
 512
 513                 hrtimer_start(&stimer->timer,
 514                               ktime_add_ns(ktime_now,
 515                                            100 * (stimer->exp_time - time_now)),
 516                               HRTIMER_MODE_ABS);
 517                 return 0;
 518         }
 519         stimer->exp_time = stimer->count;
 520         if (time_now >= stimer->count) {
 521                 /*
 522                  * Expire timer according to Hypervisor Top-Level Functional
 523                  * specification v4(15.3.1):
 524                  * "If a one shot is enabled and the specified count is in
 525                  * the past, it will expire immediately."
 526                  */
 527                 stimer_mark_pending(stimer, false);
 528                 return 0;
 529         }
 530
 531         trace_kvm_hv_stimer_start_one_shot(stimer_to_vcpu(stimer)->vcpu_id,
 532                                            stimer->index,
 533                                            time_now, stimer->count);
 534
 535         hrtimer_start(&stimer->timer,
 536                       ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)),
 537                       HRTIMER_MODE_ABS);
 538         return 0;
 539 }
 540
 541 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
 542                              bool host)
 543 {
 544         trace_kvm_hv_stimer_set_config(stimer_to_vcpu(stimer)->vcpu_id,
 545                                        stimer->index, config, host);
 546
 547         stimer_cleanup(stimer);
 548         if ((stimer->config & HV_STIMER_ENABLE) && HV_STIMER_SINT(config) == 0)
 549                 config &= ~HV_STIMER_ENABLE;
 550         stimer->config = config;
 551         stimer_mark_pending(stimer, false);
 552         return 0;
 553 }
 554
 555 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
 556                             bool host)
 557 {
 558         trace_kvm_hv_stimer_set_count(stimer_to_vcpu(stimer)->vcpu_id,
 559                                       stimer->index, count, host);
 560
 561         stimer_cleanup(stimer);
 562         stimer->count = count;
 563         if (stimer->count == 0)
 564                 stimer->config &= ~HV_STIMER_ENABLE;
 565         else if (stimer->config & HV_STIMER_AUTOENABLE)
 566                 stimer->config |= HV_STIMER_ENABLE;
 567         stimer_mark_pending(stimer, false);
 568         return 0;
 569 }
 570
 571 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
 572 {
 573         *pconfig = stimer->config;
 574         return 0;
 575 }
 576
 577 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
 578 {
 579         *pcount = stimer->count;
 580         return 0;
 581 }
 582
 583 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
 584                              struct hv_message *src_msg)
 585 {
 586         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 587         struct page *page;
 588         gpa_t gpa;
 589         struct hv_message *dst_msg;
 590         int r;
 591         struct hv_message_page *msg_page;
 592
 593         if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
 594                 return -ENOENT;
 595
 596         gpa = synic->msg_page & PAGE_MASK;
 597         page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
 598         if (is_error_page(page))
 599                 return -EFAULT;
 600
 601         msg_page = kmap_atomic(page);
 602         dst_msg = &msg_page->sint_message[sint];
 603         if (sync_cmpxchg(&dst_msg->header.message_type, HVMSG_NONE,
 604                          src_msg->header.message_type) != HVMSG_NONE) {
 605                 dst_msg->header.message_flags.msg_pending = 1;
 606                 r = -EAGAIN;
 607         } else {
 608                 memcpy(&dst_msg->u.payload, &src_msg->u.payload,
 609                        src_msg->header.payload_size);
 610                 dst_msg->header.message_type = src_msg->header.message_type;
 611                 dst_msg->header.payload_size = src_msg->header.payload_size;
 612                 r = synic_set_irq(synic, sint);
 613                 if (r >= 1)
 614                         r = 0;
 615                 else if (r == 0)
 616                         r = -EFAULT;
 617         }
 618         kunmap_atomic(msg_page);
 619         kvm_release_page_dirty(page);
 620         kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
 621         return r;
 622 }
 623
 624 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
 625 {
 626         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
 627         struct hv_message *msg = &stimer->msg;
 628         struct hv_timer_message_payload *payload =
 629                         (struct hv_timer_message_payload *)&msg->u.payload;
 630
 631         payload->expiration_time = stimer->exp_time;
 632         payload->delivery_time = get_time_ref_counter(vcpu->kvm);
 633         return synic_deliver_msg(vcpu_to_synic(vcpu),
 634                                  HV_STIMER_SINT(stimer->config), msg);
 635 }
 636
 637 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
 638 {
 639         int r;
 640
 641         stimer->msg_pending = true;
 642         r = stimer_send_msg(stimer);
 643         trace_kvm_hv_stimer_expiration(stimer_to_vcpu(stimer)->vcpu_id,
 644                                        stimer->index, r);
 645         if (!r) {
 646                 stimer->msg_pending = false;
 647                 if (!(stimer->config & HV_STIMER_PERIODIC))
 648                         stimer->config &= ~HV_STIMER_ENABLE;
 649         }
 650 }
 651
 652 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
 653 {
 654         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 655         struct kvm_vcpu_hv_stimer *stimer;
 656         u64 time_now, exp_time;
 657         int i;
 658
 659         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 660                 if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
 661                         stimer = &hv_vcpu->stimer[i];
 662                         if (stimer->config & HV_STIMER_ENABLE) {
 663                                 exp_time = stimer->exp_time;
 664
 665                                 if (exp_time) {
 666                                         time_now =
 667                                                 get_time_ref_counter(vcpu->kvm);
 668                                         if (time_now >= exp_time)
 669                                                 stimer_expiration(stimer);
 670                                 }
 671
 672                                 if ((stimer->config & HV_STIMER_ENABLE) &&
 673                                     stimer->count) {
 674                                         if (!stimer->msg_pending)
 675                                                 stimer_start(stimer);
 676                                 } else
 677                                         stimer_cleanup(stimer);
 678                         }
 679                 }
 680 }
 681
 682 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
 683 {
 684         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 685         int i;
 686
 687         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 688                 stimer_cleanup(&hv_vcpu->stimer[i]);
 689 }
 690
 691 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
 692 {
 693         struct hv_message *msg = &stimer->msg;
 694         struct hv_timer_message_payload *payload =
 695                         (struct hv_timer_message_payload *)&msg->u.payload;
 696
 697         memset(&msg->header, 0, sizeof(msg->header));
 698         msg->header.message_type = HVMSG_TIMER_EXPIRED;
 699         msg->header.payload_size = sizeof(*payload);
 700
 701         payload->timer_index = stimer->index;
 702         payload->expiration_time = 0;
 703         payload->delivery_time = 0;
 704 }
 705
 706 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
 707 {
 708         memset(stimer, 0, sizeof(*stimer));
 709         stimer->index = timer_index;
 710         hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
 711         stimer->timer.function = stimer_timer_callback;
 712         stimer_prepare_msg(stimer);
 713 }
 714
 715 void kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
 716 {
 717         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 718         int i;
 719
 720         synic_init(&hv_vcpu->synic);
 721
 722         bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
 723         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 724                 stimer_init(&hv_vcpu->stimer[i], i);
 725 }
 726
 727 void kvm_hv_vcpu_postcreate(struct kvm_vcpu *vcpu)
 728 {
 729         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 730
 731         hv_vcpu->vp_index = kvm_vcpu_get_idx(vcpu);
 732 }
 733
 734 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
 735 {
 736         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
 737
 738         /*
 739          * Hyper-V SynIC auto EOI SINT's are
 740          * not compatible with APICV, so deactivate APICV
 741          */
 742         kvm_vcpu_deactivate_apicv(vcpu);
 743         synic->active = true;
 744         synic->dont_zero_synic_pages = dont_zero_synic_pages;
 745         return 0;
 746 }
 747
 748 static bool kvm_hv_msr_partition_wide(u32 msr)
 749 {
 750         bool r = false;
 751
 752         switch (msr) {
 753         case HV_X64_MSR_GUEST_OS_ID:
 754         case HV_X64_MSR_HYPERCALL:
 755         case HV_X64_MSR_REFERENCE_TSC:
 756         case HV_X64_MSR_TIME_REF_COUNT:
 757         case HV_X64_MSR_CRASH_CTL:
 758         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
 759         case HV_X64_MSR_RESET:
 760         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
 761         case HV_X64_MSR_TSC_EMULATION_CONTROL:
 762         case HV_X64_MSR_TSC_EMULATION_STATUS:
 763                 r = true;
 764                 break;
 765         }
 766
 767         return r;
 768 }
 769
 770 static int kvm_hv_msr_get_crash_data(struct kvm_vcpu *vcpu,
 771                                      u32 index, u64 *pdata)
 772 {
 773         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 774
 775         if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param)))
 776                 return -EINVAL;
 777
 778         *pdata = hv->hv_crash_param[index];
 779         return 0;
 780 }
 781
 782 static int kvm_hv_msr_get_crash_ctl(struct kvm_vcpu *vcpu, u64 *pdata)
 783 {
 784         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 785
 786         *pdata = hv->hv_crash_ctl;
 787         return 0;
 788 }
 789
 790 static int kvm_hv_msr_set_crash_ctl(struct kvm_vcpu *vcpu, u64 data, bool host)
 791 {
 792         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 793
 794         if (host)
 795                 hv->hv_crash_ctl = data & HV_X64_MSR_CRASH_CTL_NOTIFY;
 796
 797         if (!host && (data & HV_X64_MSR_CRASH_CTL_NOTIFY)) {
 798
 799                 vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
 800                           hv->hv_crash_param[0],
 801                           hv->hv_crash_param[1],
 802                           hv->hv_crash_param[2],
 803                           hv->hv_crash_param[3],
 804                           hv->hv_crash_param[4]);
 805
 806                 /* Send notification about crash to user space */
 807                 kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
 808         }
 809
 810         return 0;
 811 }
 812
 813 static int kvm_hv_msr_set_crash_data(struct kvm_vcpu *vcpu,
 814                                      u32 index, u64 data)
 815 {
 816         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 817
 818         if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param)))
 819                 return -EINVAL;
 820
 821         hv->hv_crash_param[index] = data;
 822         return 0;
 823 }
 824
 825 /*
 826  * The kvmclock and Hyper-V TSC page use similar formulas, and converting
 827  * between them is possible:
 828  *
 829  * kvmclock formula:
 830  *    nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
 831  *           + system_time
 832  *
 833  * Hyper-V formula:
 834  *    nsec/100 = ticks * scale / 2^64 + offset
 835  *
 836  * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
 837  * By dividing the kvmclock formula by 100 and equating what's left we get:
 838  *    ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
 839  *            scale / 2^64 =         tsc_to_system_mul * 2^(tsc_shift-32) / 100
 840  *            scale        =         tsc_to_system_mul * 2^(32+tsc_shift) / 100
 841  *
 842  * Now expand the kvmclock formula and divide by 100:
 843  *    nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
 844  *           - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
 845  *           + system_time
 846  *    nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
 847  *               - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
 848  *               + system_time / 100
 849  *
 850  * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
 851  *    nsec/100 = ticks * scale / 2^64
 852  *               - tsc_timestamp * scale / 2^64
 853  *               + system_time / 100
 854  *
 855  * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
 856  *    offset = system_time / 100 - tsc_timestamp * scale / 2^64
 857  *
 858  * These two equivalencies are implemented in this function.
 859  */
 860 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
 861                                         HV_REFERENCE_TSC_PAGE *tsc_ref)
 862 {
 863         u64 max_mul;
 864
 865         if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
 866                 return false;
 867
 868         /*
 869          * check if scale would overflow, if so we use the time ref counter
 870          *    tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
 871          *    tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
 872          *    tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
 873          */
 874         max_mul = 100ull << (32 - hv_clock->tsc_shift);
 875         if (hv_clock->tsc_to_system_mul >= max_mul)
 876                 return false;
 877
 878         /*
 879          * Otherwise compute the scale and offset according to the formulas
 880          * derived above.
 881          */
 882         tsc_ref->tsc_scale =
 883                 mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
 884                                 hv_clock->tsc_to_system_mul,
 885                                 100);
 886
 887         tsc_ref->tsc_offset = hv_clock->system_time;
 888         do_div(tsc_ref->tsc_offset, 100);
 889         tsc_ref->tsc_offset -=
 890                 mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
 891         return true;
 892 }
 893
 894 void kvm_hv_setup_tsc_page(struct kvm *kvm,
 895                            struct pvclock_vcpu_time_info *hv_clock)
 896 {
 897         struct kvm_hv *hv = &kvm->arch.hyperv;
 898         u32 tsc_seq;
 899         u64 gfn;
 900
 901         BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
 902         BUILD_BUG_ON(offsetof(HV_REFERENCE_TSC_PAGE, tsc_sequence) != 0);
 903
 904         if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
 905                 return;
 906
 907         mutex_lock(&kvm->arch.hyperv.hv_lock);
 908         if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
 909                 goto out_unlock;
 910
 911         gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
 912         /*
 913          * Because the TSC parameters only vary when there is a
 914          * change in the master clock, do not bother with caching.
 915          */
 916         if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
 917                                     &tsc_seq, sizeof(tsc_seq))))
 918                 goto out_unlock;
 919
 920         /*
 921          * While we're computing and writing the parameters, force the
 922          * guest to use the time reference count MSR.
 923          */
 924         hv->tsc_ref.tsc_sequence = 0;
 925         if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
 926                             &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
 927                 goto out_unlock;
 928
 929         if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
 930                 goto out_unlock;
 931
 932         /* Ensure sequence is zero before writing the rest of the struct.  */
 933         smp_wmb();
 934         if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
 935                 goto out_unlock;
 936
 937         /*
 938          * Now switch to the TSC page mechanism by writing the sequence.
 939          */
 940         tsc_seq++;
 941         if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
 942                 tsc_seq = 1;
 943
 944         /* Write the struct entirely before the non-zero sequence.  */
 945         smp_wmb();
 946
 947         hv->tsc_ref.tsc_sequence = tsc_seq;
 948         kvm_write_guest(kvm, gfn_to_gpa(gfn),
 949                         &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence));
 950 out_unlock:
 951         mutex_unlock(&kvm->arch.hyperv.hv_lock);
 952 }
 953
 954 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
 955                              bool host)
 956 {
 957         struct kvm *kvm = vcpu->kvm;
 958         struct kvm_hv *hv = &kvm->arch.hyperv;
 959
 960         switch (msr) {
 961         case HV_X64_MSR_GUEST_OS_ID:
 962                 hv->hv_guest_os_id = data;
 963                 /* setting guest os id to zero disables hypercall page */
 964                 if (!hv->hv_guest_os_id)
 965                         hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
 966                 break;
 967         case HV_X64_MSR_HYPERCALL: {
 968                 u64 gfn;
 969                 unsigned long addr;
 970                 u8 instructions[4];
 971
 972                 /* if guest os id is not set hypercall should remain disabled */
 973                 if (!hv->hv_guest_os_id)
 974                         break;
 975                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
 976                         hv->hv_hypercall = data;
 977                         break;
 978                 }
 979                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
 980                 addr = gfn_to_hva(kvm, gfn);
 981                 if (kvm_is_error_hva(addr))
 982                         return 1;
 983                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
 984                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
 985                 if (__copy_to_user((void __user *)addr, instructions, 4))
 986                         return 1;
 987                 hv->hv_hypercall = data;
 988                 mark_page_dirty(kvm, gfn);
 989                 break;
 990         }
 991         case HV_X64_MSR_REFERENCE_TSC:
 992                 hv->hv_tsc_page = data;
 993                 if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE)
 994                         kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
 995                 break;
 996         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
 997                 return kvm_hv_msr_set_crash_data(vcpu,
 998                                                  msr - HV_X64_MSR_CRASH_P0,
 999                                                  data);
1000         case HV_X64_MSR_CRASH_CTL:
1001                 return kvm_hv_msr_set_crash_ctl(vcpu, data, host);
1002         case HV_X64_MSR_RESET:
1003                 if (data == 1) {
1004                         vcpu_debug(vcpu, "hyper-v reset requested\n");
1005                         kvm_make_request(KVM_REQ_HV_RESET, vcpu);
1006                 }
1007                 break;
1008         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1009                 hv->hv_reenlightenment_control = data;
1010                 break;
1011         case HV_X64_MSR_TSC_EMULATION_CONTROL:
1012                 hv->hv_tsc_emulation_control = data;
1013                 break;
1014         case HV_X64_MSR_TSC_EMULATION_STATUS:
1015                 hv->hv_tsc_emulation_status = data;
1016                 break;
1017         default:
1018                 vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
1019                             msr, data);
1020                 return 1;
1021         }
1022         return 0;
1023 }
1024
1025 /* Calculate cpu time spent by current task in 100ns units */
1026 static u64 current_task_runtime_100ns(void)
1027 {
1028         u64 utime, stime;
1029
1030         task_cputime_adjusted(current, &utime, &stime);
1031
1032         return div_u64(utime + stime, 100);
1033 }
1034
1035 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1036 {
1037         struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
1038
1039         switch (msr) {
1040         case HV_X64_MSR_VP_INDEX:
1041                 if (!host)
1042                         return 1;
1043                 hv->vp_index = (u32)data;
1044                 break;
1045         case HV_X64_MSR_VP_ASSIST_PAGE: {
1046                 u64 gfn;
1047                 unsigned long addr;
1048
1049                 if (!(data & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE)) {
1050                         hv->hv_vapic = data;
1051                         if (kvm_lapic_enable_pv_eoi(vcpu, 0))
1052                                 return 1;
1053                         break;
1054                 }
1055                 gfn = data >> HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT;
1056                 addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1057                 if (kvm_is_error_hva(addr))
1058                         return 1;
1059                 if (__clear_user((void __user *)addr, PAGE_SIZE))
1060                         return 1;
1061                 hv->hv_vapic = data;
1062                 kvm_vcpu_mark_page_dirty(vcpu, gfn);
1063                 if (kvm_lapic_enable_pv_eoi(vcpu,
1064                                             gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
1065                         return 1;
1066                 break;
1067         }
1068         case HV_X64_MSR_EOI:
1069                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1070         case HV_X64_MSR_ICR:
1071                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1072         case HV_X64_MSR_TPR:
1073                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1074         case HV_X64_MSR_VP_RUNTIME:
1075                 if (!host)
1076                         return 1;
1077                 hv->runtime_offset = data - current_task_runtime_100ns();
1078                 break;
1079         case HV_X64_MSR_SCONTROL:
1080         case HV_X64_MSR_SVERSION:
1081         case HV_X64_MSR_SIEFP:
1082         case HV_X64_MSR_SIMP:
1083         case HV_X64_MSR_EOM:
1084         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1085                 return synic_set_msr(vcpu_to_synic(vcpu), msr, data, host);
1086         case HV_X64_MSR_STIMER0_CONFIG:
1087         case HV_X64_MSR_STIMER1_CONFIG:
1088         case HV_X64_MSR_STIMER2_CONFIG:
1089         case HV_X64_MSR_STIMER3_CONFIG: {
1090                 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1091
1092                 return stimer_set_config(vcpu_to_stimer(vcpu, timer_index),
1093                                          data, host);
1094         }
1095         case HV_X64_MSR_STIMER0_COUNT:
1096         case HV_X64_MSR_STIMER1_COUNT:
1097         case HV_X64_MSR_STIMER2_COUNT:
1098         case HV_X64_MSR_STIMER3_COUNT: {
1099                 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1100
1101                 return stimer_set_count(vcpu_to_stimer(vcpu, timer_index),
1102                                         data, host);
1103         }
1104         default:
1105                 vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
1106                             msr, data);
1107                 return 1;
1108         }
1109
1110         return 0;
1111 }
1112
1113 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1114 {
1115         u64 data = 0;
1116         struct kvm *kvm = vcpu->kvm;
1117         struct kvm_hv *hv = &kvm->arch.hyperv;
1118
1119         switch (msr) {
1120         case HV_X64_MSR_GUEST_OS_ID:
1121                 data = hv->hv_guest_os_id;
1122                 break;
1123         case HV_X64_MSR_HYPERCALL:
1124                 data = hv->hv_hypercall;
1125                 break;
1126         case HV_X64_MSR_TIME_REF_COUNT:
1127                 data = get_time_ref_counter(kvm);
1128                 break;
1129         case HV_X64_MSR_REFERENCE_TSC:
1130                 data = hv->hv_tsc_page;
1131                 break;
1132         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1133                 return kvm_hv_msr_get_crash_data(vcpu,
1134                                                  msr - HV_X64_MSR_CRASH_P0,
1135                                                  pdata);
1136         case HV_X64_MSR_CRASH_CTL:
1137                 return kvm_hv_msr_get_crash_ctl(vcpu, pdata);
1138         case HV_X64_MSR_RESET:
1139                 data = 0;
1140                 break;
1141         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1142                 data = hv->hv_reenlightenment_control;
1143                 break;
1144         case HV_X64_MSR_TSC_EMULATION_CONTROL:
1145                 data = hv->hv_tsc_emulation_control;
1146                 break;
1147         case HV_X64_MSR_TSC_EMULATION_STATUS:
1148                 data = hv->hv_tsc_emulation_status;
1149                 break;
1150         default:
1151                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1152                 return 1;
1153         }
1154
1155         *pdata = data;
1156         return 0;
1157 }
1158
1159 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1160 {
1161         u64 data = 0;
1162         struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
1163
1164         switch (msr) {
1165         case HV_X64_MSR_VP_INDEX:
1166                 data = hv->vp_index;
1167                 break;
1168         case HV_X64_MSR_EOI:
1169                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1170         case HV_X64_MSR_ICR:
1171                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1172         case HV_X64_MSR_TPR:
1173                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1174         case HV_X64_MSR_VP_ASSIST_PAGE:
1175                 data = hv->hv_vapic;
1176                 break;
1177         case HV_X64_MSR_VP_RUNTIME:
1178                 data = current_task_runtime_100ns() + hv->runtime_offset;
1179                 break;
1180         case HV_X64_MSR_SCONTROL:
1181         case HV_X64_MSR_SVERSION:
1182         case HV_X64_MSR_SIEFP:
1183         case HV_X64_MSR_SIMP:
1184         case HV_X64_MSR_EOM:
1185         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1186                 return synic_get_msr(vcpu_to_synic(vcpu), msr, pdata);
1187         case HV_X64_MSR_STIMER0_CONFIG:
1188         case HV_X64_MSR_STIMER1_CONFIG:
1189         case HV_X64_MSR_STIMER2_CONFIG:
1190         case HV_X64_MSR_STIMER3_CONFIG: {
1191                 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1192
1193                 return stimer_get_config(vcpu_to_stimer(vcpu, timer_index),
1194                                          pdata);
1195         }
1196         case HV_X64_MSR_STIMER0_COUNT:
1197         case HV_X64_MSR_STIMER1_COUNT:
1198         case HV_X64_MSR_STIMER2_COUNT:
1199         case HV_X64_MSR_STIMER3_COUNT: {
1200                 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1201
1202                 return stimer_get_count(vcpu_to_stimer(vcpu, timer_index),
1203                                         pdata);
1204         }
1205         case HV_X64_MSR_TSC_FREQUENCY:
1206                 data = (u64)vcpu->arch.virtual_tsc_khz * 1000;
1207                 break;
1208         case HV_X64_MSR_APIC_FREQUENCY:
1209                 data = APIC_BUS_FREQUENCY;
1210                 break;
1211         default:
1212                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1213                 return 1;
1214         }
1215         *pdata = data;
1216         return 0;
1217 }
1218
1219 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1220 {
1221         if (kvm_hv_msr_partition_wide(msr)) {
1222                 int r;
1223
1224                 mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1225                 r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1226                 mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1227                 return r;
1228         } else
1229                 return kvm_hv_set_msr(vcpu, msr, data, host);
1230 }
1231
1232 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1233 {
1234         if (kvm_hv_msr_partition_wide(msr)) {
1235                 int r;
1236
1237                 mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1238                 r = kvm_hv_get_msr_pw(vcpu, msr, pdata);
1239                 mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1240                 return r;
1241         } else
1242                 return kvm_hv_get_msr(vcpu, msr, pdata);
1243 }
1244
1245 bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1246 {
1247         return READ_ONCE(kvm->arch.hyperv.hv_hypercall) & HV_X64_MSR_HYPERCALL_ENABLE;
1248 }
1249
1250 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
1251 {
1252         bool longmode;
1253
1254         longmode = is_64_bit_mode(vcpu);
1255         if (longmode)
1256                 kvm_register_write(vcpu, VCPU_REGS_RAX, result);
1257         else {
1258                 kvm_register_write(vcpu, VCPU_REGS_RDX, result >> 32);
1259                 kvm_register_write(vcpu, VCPU_REGS_RAX, result & 0xffffffff);
1260         }
1261 }
1262
1263 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
1264 {
1265         struct kvm_run *run = vcpu->run;
1266
1267         kvm_hv_hypercall_set_result(vcpu, run->hyperv.u.hcall.result);
1268         return kvm_skip_emulated_instruction(vcpu);
1269 }
1270
1271 static u16 kvm_hvcall_signal_event(struct kvm_vcpu *vcpu, bool fast, u64 param)
1272 {
1273         struct eventfd_ctx *eventfd;
1274
1275         if (unlikely(!fast)) {
1276                 int ret;
1277                 gpa_t gpa = param;
1278
1279                 if ((gpa & (__alignof__(param) - 1)) ||
1280                     offset_in_page(gpa) + sizeof(param) > PAGE_SIZE)
1281                         return HV_STATUS_INVALID_ALIGNMENT;
1282
1283                 ret = kvm_vcpu_read_guest(vcpu, gpa, &param, sizeof(param));
1284                 if (ret < 0)
1285                         return HV_STATUS_INVALID_ALIGNMENT;
1286         }
1287
1288         /*
1289          * Per spec, bits 32-47 contain the extra "flag number".  However, we
1290          * have no use for it, and in all known usecases it is zero, so just
1291          * report lookup failure if it isn't.
1292          */
1293         if (param & 0xffff00000000ULL)
1294                 return HV_STATUS_INVALID_PORT_ID;
1295         /* remaining bits are reserved-zero */
1296         if (param & ~KVM_HYPERV_CONN_ID_MASK)
1297                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1298
1299         /* the eventfd is protected by vcpu->kvm->srcu, but conn_to_evt isn't */
1300         rcu_read_lock();
1301         eventfd = idr_find(&vcpu->kvm->arch.hyperv.conn_to_evt, param);
1302         rcu_read_unlock();
1303         if (!eventfd)
1304                 return HV_STATUS_INVALID_PORT_ID;
1305
1306         eventfd_signal(eventfd, 1);
1307         return HV_STATUS_SUCCESS;
1308 }
1309
1310 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
1311 {
1312         u64 param, ingpa, outgpa, ret = HV_STATUS_SUCCESS;
1313         uint16_t code, rep_idx, rep_cnt;
1314         bool fast, longmode;
1315
1316         /*
1317          * hypercall generates UD from non zero cpl and real mode
1318          * per HYPER-V spec
1319          */
1320         if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
1321                 kvm_queue_exception(vcpu, UD_VECTOR);
1322                 return 1;
1323         }
1324
1325         longmode = is_64_bit_mode(vcpu);
1326
1327         if (!longmode) {
1328                 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
1329                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
1330                 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
1331                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
1332                 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
1333                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
1334         }
1335 #ifdef CONFIG_X86_64
1336         else {
1337                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
1338                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
1339                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
1340         }
1341 #endif
1342
1343         code = param & 0xffff;
1344         fast = (param >> 16) & 0x1;
1345         rep_cnt = (param >> 32) & 0xfff;
1346         rep_idx = (param >> 48) & 0xfff;
1347
1348         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
1349
1350         /* Hypercall continuation is not supported yet */
1351         if (rep_cnt || rep_idx) {
1352                 ret = HV_STATUS_INVALID_HYPERCALL_CODE;
1353                 goto set_result;
1354         }
1355
1356         switch (code) {
1357         case HVCALL_NOTIFY_LONG_SPIN_WAIT:
1358                 kvm_vcpu_on_spin(vcpu, true);
1359                 break;
1360         case HVCALL_SIGNAL_EVENT:
1361                 ret = kvm_hvcall_signal_event(vcpu, fast, ingpa);
1362                 if (ret != HV_STATUS_INVALID_PORT_ID)
1363                         break;
1364                 /* maybe userspace knows this conn_id: fall through */
1365         case HVCALL_POST_MESSAGE:
1366                 /* don't bother userspace if it has no way to handle it */
1367                 if (!vcpu_to_synic(vcpu)->active) {
1368                         ret = HV_STATUS_INVALID_HYPERCALL_CODE;
1369                         break;
1370                 }
1371                 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
1372                 vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
1373                 vcpu->run->hyperv.u.hcall.input = param;
1374                 vcpu->run->hyperv.u.hcall.params[0] = ingpa;
1375                 vcpu->run->hyperv.u.hcall.params[1] = outgpa;
1376                 vcpu->arch.complete_userspace_io =
1377                                 kvm_hv_hypercall_complete_userspace;
1378                 return 0;
1379         default:
1380                 ret = HV_STATUS_INVALID_HYPERCALL_CODE;
1381                 break;
1382         }
1383
1384 set_result:
1385         kvm_hv_hypercall_set_result(vcpu, ret);
1386         return 1;
1387 }
1388
1389 void kvm_hv_init_vm(struct kvm *kvm)
1390 {
1391         mutex_init(&kvm->arch.hyperv.hv_lock);
1392         idr_init(&kvm->arch.hyperv.conn_to_evt);
1393 }
1394
1395 void kvm_hv_destroy_vm(struct kvm *kvm)
1396 {
1397         struct eventfd_ctx *eventfd;
1398         int i;
1399
1400         idr_for_each_entry(&kvm->arch.hyperv.conn_to_evt, eventfd, i)
1401                 eventfd_ctx_put(eventfd);
1402         idr_destroy(&kvm->arch.hyperv.conn_to_evt);
1403 }
1404
1405 static int kvm_hv_eventfd_assign(struct kvm *kvm, u32 conn_id, int fd)
1406 {
1407         struct kvm_hv *hv = &kvm->arch.hyperv;
1408         struct eventfd_ctx *eventfd;
1409         int ret;
1410
1411         eventfd = eventfd_ctx_fdget(fd);
1412         if (IS_ERR(eventfd))
1413                 return PTR_ERR(eventfd);
1414
1415         mutex_lock(&hv->hv_lock);
1416         ret = idr_alloc(&hv->conn_to_evt, eventfd, conn_id, conn_id + 1,
1417                         GFP_KERNEL);
1418         mutex_unlock(&hv->hv_lock);
1419
1420         if (ret >= 0)
1421                 return 0;
1422
1423         if (ret == -ENOSPC)
1424                 ret = -EEXIST;
1425         eventfd_ctx_put(eventfd);
1426         return ret;
1427 }
1428
1429 static int kvm_hv_eventfd_deassign(struct kvm *kvm, u32 conn_id)
1430 {
1431         struct kvm_hv *hv = &kvm->arch.hyperv;
1432         struct eventfd_ctx *eventfd;
1433
1434         mutex_lock(&hv->hv_lock);
1435         eventfd = idr_remove(&hv->conn_to_evt, conn_id);
1436         mutex_unlock(&hv->hv_lock);
1437
1438         if (!eventfd)
1439                 return -ENOENT;
1440
1441         synchronize_srcu(&kvm->srcu);
1442         eventfd_ctx_put(eventfd);
1443         return 0;
1444 }
1445
1446 int kvm_vm_ioctl_hv_eventfd(struct kvm *kvm, struct kvm_hyperv_eventfd *args)
1447 {
1448         if ((args->flags & ~KVM_HYPERV_EVENTFD_DEASSIGN) ||
1449             (args->conn_id & ~KVM_HYPERV_CONN_ID_MASK))
1450                 return -EINVAL;
1451
1452         if (args->flags == KVM_HYPERV_EVENTFD_DEASSIGN)
1453                 return kvm_hv_eventfd_deassign(kvm, args->conn_id);
1454         return kvm_hv_eventfd_assign(kvm, args->conn_id, args->fd);
1455 }