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 int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint,
  79                           u64 data, bool host)
  80 {
  81         int vector;
  82
  83         vector = data & HV_SYNIC_SINT_VECTOR_MASK;
  84         if (vector < 16 && !host)
  85                 return 1;
  86         /*
  87          * Guest may configure multiple SINTs to use the same vector, so
  88          * we maintain a bitmap of vectors handled by synic, and a
  89          * bitmap of vectors with auto-eoi behavior.  The bitmaps are
  90          * updated here, and atomically queried on fast paths.
  91          */
  92
  93         atomic64_set(&synic->sint[sint], data);
  94
  95         if (synic_has_vector_connected(synic, vector))
  96                 __set_bit(vector, synic->vec_bitmap);
  97         else
  98                 __clear_bit(vector, synic->vec_bitmap);
  99
 100         if (synic_has_vector_auto_eoi(synic, vector))
 101                 __set_bit(vector, synic->auto_eoi_bitmap);
 102         else
 103                 __clear_bit(vector, synic->auto_eoi_bitmap);
 104
 105         /* Load SynIC vectors into EOI exit bitmap */
 106         kvm_make_request(KVM_REQ_SCAN_IOAPIC, synic_to_vcpu(synic));
 107         return 0;
 108 }
 109
 110 static struct kvm_vcpu *get_vcpu_by_vpidx(struct kvm *kvm, u32 vpidx)
 111 {
 112         struct kvm_vcpu *vcpu = NULL;
 113         int i;
 114
 115         if (vpidx < KVM_MAX_VCPUS)
 116                 vcpu = kvm_get_vcpu(kvm, vpidx);
 117         if (vcpu && vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
 118                 return vcpu;
 119         kvm_for_each_vcpu(i, vcpu, kvm)
 120                 if (vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
 121                         return vcpu;
 122         return NULL;
 123 }
 124
 125 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vpidx)
 126 {
 127         struct kvm_vcpu *vcpu;
 128         struct kvm_vcpu_hv_synic *synic;
 129
 130         vcpu = get_vcpu_by_vpidx(kvm, vpidx);
 131         if (!vcpu)
 132                 return NULL;
 133         synic = vcpu_to_synic(vcpu);
 134         return (synic->active) ? synic : NULL;
 135 }
 136
 137 static void synic_clear_sint_msg_pending(struct kvm_vcpu_hv_synic *synic,
 138                                         u32 sint)
 139 {
 140         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 141         struct page *page;
 142         gpa_t gpa;
 143         struct hv_message *msg;
 144         struct hv_message_page *msg_page;
 145
 146         gpa = synic->msg_page & PAGE_MASK;
 147         page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
 148         if (is_error_page(page)) {
 149                 vcpu_err(vcpu, "Hyper-V SynIC can't get msg page, gpa 0x%llx\n",
 150                          gpa);
 151                 return;
 152         }
 153         msg_page = kmap_atomic(page);
 154
 155         msg = &msg_page->sint_message[sint];
 156         msg->header.message_flags.msg_pending = 0;
 157
 158         kunmap_atomic(msg_page);
 159         kvm_release_page_dirty(page);
 160         kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
 161 }
 162
 163 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
 164 {
 165         struct kvm *kvm = vcpu->kvm;
 166         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
 167         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 168         struct kvm_vcpu_hv_stimer *stimer;
 169         int gsi, idx, stimers_pending;
 170
 171         trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
 172
 173         if (synic->msg_page & HV_SYNIC_SIMP_ENABLE)
 174                 synic_clear_sint_msg_pending(synic, sint);
 175
 176         /* Try to deliver pending Hyper-V SynIC timers messages */
 177         stimers_pending = 0;
 178         for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
 179                 stimer = &hv_vcpu->stimer[idx];
 180                 if (stimer->msg_pending &&
 181                     (stimer->config & HV_STIMER_ENABLE) &&
 182                     HV_STIMER_SINT(stimer->config) == sint) {
 183                         set_bit(stimer->index,
 184                                 hv_vcpu->stimer_pending_bitmap);
 185                         stimers_pending++;
 186                 }
 187         }
 188         if (stimers_pending)
 189                 kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
 190
 191         idx = srcu_read_lock(&kvm->irq_srcu);
 192         gsi = atomic_read(&synic->sint_to_gsi[sint]);
 193         if (gsi != -1)
 194                 kvm_notify_acked_gsi(kvm, gsi);
 195         srcu_read_unlock(&kvm->irq_srcu, idx);
 196 }
 197
 198 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
 199 {
 200         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 201         struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
 202
 203         hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
 204         hv_vcpu->exit.u.synic.msr = msr;
 205         hv_vcpu->exit.u.synic.control = synic->control;
 206         hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
 207         hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
 208
 209         kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
 210 }
 211
 212 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
 213                          u32 msr, u64 data, bool host)
 214 {
 215         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 216         int ret;
 217
 218         if (!synic->active)
 219                 return 1;
 220
 221         trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
 222
 223         ret = 0;
 224         switch (msr) {
 225         case HV_X64_MSR_SCONTROL:
 226                 synic->control = data;
 227                 if (!host)
 228                         synic_exit(synic, msr);
 229                 break;
 230         case HV_X64_MSR_SVERSION:
 231                 if (!host) {
 232                         ret = 1;
 233                         break;
 234                 }
 235                 synic->version = data;
 236                 break;
 237         case HV_X64_MSR_SIEFP:
 238                 if ((data & HV_SYNIC_SIEFP_ENABLE) && !host &&
 239                     !synic->dont_zero_synic_pages)
 240                         if (kvm_clear_guest(vcpu->kvm,
 241                                             data & PAGE_MASK, PAGE_SIZE)) {
 242                                 ret = 1;
 243                                 break;
 244                         }
 245                 synic->evt_page = data;
 246                 if (!host)
 247                         synic_exit(synic, msr);
 248                 break;
 249         case HV_X64_MSR_SIMP:
 250                 if ((data & HV_SYNIC_SIMP_ENABLE) && !host &&
 251                     !synic->dont_zero_synic_pages)
 252                         if (kvm_clear_guest(vcpu->kvm,
 253                                             data & PAGE_MASK, PAGE_SIZE)) {
 254                                 ret = 1;
 255                                 break;
 256                         }
 257                 synic->msg_page = data;
 258                 if (!host)
 259                         synic_exit(synic, msr);
 260                 break;
 261         case HV_X64_MSR_EOM: {
 262                 int i;
 263
 264                 for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
 265                         kvm_hv_notify_acked_sint(vcpu, i);
 266                 break;
 267         }
 268         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
 269                 ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
 270                 break;
 271         default:
 272                 ret = 1;
 273                 break;
 274         }
 275         return ret;
 276 }
 277
 278 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata)
 279 {
 280         int ret;
 281
 282         if (!synic->active)
 283                 return 1;
 284
 285         ret = 0;
 286         switch (msr) {
 287         case HV_X64_MSR_SCONTROL:
 288                 *pdata = synic->control;
 289                 break;
 290         case HV_X64_MSR_SVERSION:
 291                 *pdata = synic->version;
 292                 break;
 293         case HV_X64_MSR_SIEFP:
 294                 *pdata = synic->evt_page;
 295                 break;
 296         case HV_X64_MSR_SIMP:
 297                 *pdata = synic->msg_page;
 298                 break;
 299         case HV_X64_MSR_EOM:
 300                 *pdata = 0;
 301                 break;
 302         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
 303                 *pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
 304                 break;
 305         default:
 306                 ret = 1;
 307                 break;
 308         }
 309         return ret;
 310 }
 311
 312 static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
 313 {
 314         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 315         struct kvm_lapic_irq irq;
 316         int ret, vector;
 317
 318         if (sint >= ARRAY_SIZE(synic->sint))
 319                 return -EINVAL;
 320
 321         vector = synic_get_sint_vector(synic_read_sint(synic, sint));
 322         if (vector < 0)
 323                 return -ENOENT;
 324
 325         memset(&irq, 0, sizeof(irq));
 326         irq.shorthand = APIC_DEST_SELF;
 327         irq.dest_mode = APIC_DEST_PHYSICAL;
 328         irq.delivery_mode = APIC_DM_FIXED;
 329         irq.vector = vector;
 330         irq.level = 1;
 331
 332         ret = kvm_irq_delivery_to_apic(vcpu->kvm, vcpu->arch.apic, &irq, NULL);
 333         trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
 334         return ret;
 335 }
 336
 337 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vpidx, u32 sint)
 338 {
 339         struct kvm_vcpu_hv_synic *synic;
 340
 341         synic = synic_get(kvm, vpidx);
 342         if (!synic)
 343                 return -EINVAL;
 344
 345         return synic_set_irq(synic, sint);
 346 }
 347
 348 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
 349 {
 350         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
 351         int i;
 352
 353         trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
 354
 355         for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
 356                 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
 357                         kvm_hv_notify_acked_sint(vcpu, i);
 358 }
 359
 360 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi)
 361 {
 362         struct kvm_vcpu_hv_synic *synic;
 363
 364         synic = synic_get(kvm, vpidx);
 365         if (!synic)
 366                 return -EINVAL;
 367
 368         if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
 369                 return -EINVAL;
 370
 371         atomic_set(&synic->sint_to_gsi[sint], gsi);
 372         return 0;
 373 }
 374
 375 void kvm_hv_irq_routing_update(struct kvm *kvm)
 376 {
 377         struct kvm_irq_routing_table *irq_rt;
 378         struct kvm_kernel_irq_routing_entry *e;
 379         u32 gsi;
 380
 381         irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
 382                                         lockdep_is_held(&kvm->irq_lock));
 383
 384         for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
 385                 hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
 386                         if (e->type == KVM_IRQ_ROUTING_HV_SINT)
 387                                 kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
 388                                                     e->hv_sint.sint, gsi);
 389                 }
 390         }
 391 }
 392
 393 static void synic_init(struct kvm_vcpu_hv_synic *synic)
 394 {
 395         int i;
 396
 397         memset(synic, 0, sizeof(*synic));
 398         synic->version = HV_SYNIC_VERSION_1;
 399         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
 400                 atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
 401                 atomic_set(&synic->sint_to_gsi[i], -1);
 402         }
 403 }
 404
 405 static u64 get_time_ref_counter(struct kvm *kvm)
 406 {
 407         struct kvm_hv *hv = &kvm->arch.hyperv;
 408         struct kvm_vcpu *vcpu;
 409         u64 tsc;
 410
 411         /*
 412          * The guest has not set up the TSC page or the clock isn't
 413          * stable, fall back to get_kvmclock_ns.
 414          */
 415         if (!hv->tsc_ref.tsc_sequence)
 416                 return div_u64(get_kvmclock_ns(kvm), 100);
 417
 418         vcpu = kvm_get_vcpu(kvm, 0);
 419         tsc = kvm_read_l1_tsc(vcpu, rdtsc());
 420         return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64)
 421                 + hv->tsc_ref.tsc_offset;
 422 }
 423
 424 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
 425                                 bool vcpu_kick)
 426 {
 427         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
 428
 429         set_bit(stimer->index,
 430                 vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
 431         kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
 432         if (vcpu_kick)
 433                 kvm_vcpu_kick(vcpu);
 434 }
 435
 436 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
 437 {
 438         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
 439
 440         trace_kvm_hv_stimer_cleanup(stimer_to_vcpu(stimer)->vcpu_id,
 441                                     stimer->index);
 442
 443         hrtimer_cancel(&stimer->timer);
 444         clear_bit(stimer->index,
 445                   vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
 446         stimer->msg_pending = false;
 447         stimer->exp_time = 0;
 448 }
 449
 450 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
 451 {
 452         struct kvm_vcpu_hv_stimer *stimer;
 453
 454         stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
 455         trace_kvm_hv_stimer_callback(stimer_to_vcpu(stimer)->vcpu_id,
 456                                      stimer->index);
 457         stimer_mark_pending(stimer, true);
 458
 459         return HRTIMER_NORESTART;
 460 }
 461
 462 /*
 463  * stimer_start() assumptions:
 464  * a) stimer->count is not equal to 0
 465  * b) stimer->config has HV_STIMER_ENABLE flag
 466  */
 467 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
 468 {
 469         u64 time_now;
 470         ktime_t ktime_now;
 471
 472         time_now = get_time_ref_counter(stimer_to_vcpu(stimer)->kvm);
 473         ktime_now = ktime_get();
 474
 475         if (stimer->config & HV_STIMER_PERIODIC) {
 476                 if (stimer->exp_time) {
 477                         if (time_now >= stimer->exp_time) {
 478                                 u64 remainder;
 479
 480                                 div64_u64_rem(time_now - stimer->exp_time,
 481                                               stimer->count, &remainder);
 482                                 stimer->exp_time =
 483                                         time_now + (stimer->count - remainder);
 484                         }
 485                 } else
 486                         stimer->exp_time = time_now + stimer->count;
 487
 488                 trace_kvm_hv_stimer_start_periodic(
 489                                         stimer_to_vcpu(stimer)->vcpu_id,
 490                                         stimer->index,
 491                                         time_now, stimer->exp_time);
 492
 493                 hrtimer_start(&stimer->timer,
 494                               ktime_add_ns(ktime_now,
 495                                            100 * (stimer->exp_time - time_now)),
 496                               HRTIMER_MODE_ABS);
 497                 return 0;
 498         }
 499         stimer->exp_time = stimer->count;
 500         if (time_now >= stimer->count) {
 501                 /*
 502                  * Expire timer according to Hypervisor Top-Level Functional
 503                  * specification v4(15.3.1):
 504                  * "If a one shot is enabled and the specified count is in
 505                  * the past, it will expire immediately."
 506                  */
 507                 stimer_mark_pending(stimer, false);
 508                 return 0;
 509         }
 510
 511         trace_kvm_hv_stimer_start_one_shot(stimer_to_vcpu(stimer)->vcpu_id,
 512                                            stimer->index,
 513                                            time_now, stimer->count);
 514
 515         hrtimer_start(&stimer->timer,
 516                       ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)),
 517                       HRTIMER_MODE_ABS);
 518         return 0;
 519 }
 520
 521 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
 522                              bool host)
 523 {
 524         trace_kvm_hv_stimer_set_config(stimer_to_vcpu(stimer)->vcpu_id,
 525                                        stimer->index, config, host);
 526
 527         stimer_cleanup(stimer);
 528         if ((stimer->config & HV_STIMER_ENABLE) && HV_STIMER_SINT(config) == 0)
 529                 config &= ~HV_STIMER_ENABLE;
 530         stimer->config = config;
 531         stimer_mark_pending(stimer, false);
 532         return 0;
 533 }
 534
 535 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
 536                             bool host)
 537 {
 538         trace_kvm_hv_stimer_set_count(stimer_to_vcpu(stimer)->vcpu_id,
 539                                       stimer->index, count, host);
 540
 541         stimer_cleanup(stimer);
 542         stimer->count = count;
 543         if (stimer->count == 0)
 544                 stimer->config &= ~HV_STIMER_ENABLE;
 545         else if (stimer->config & HV_STIMER_AUTOENABLE)
 546                 stimer->config |= HV_STIMER_ENABLE;
 547         stimer_mark_pending(stimer, false);
 548         return 0;
 549 }
 550
 551 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
 552 {
 553         *pconfig = stimer->config;
 554         return 0;
 555 }
 556
 557 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
 558 {
 559         *pcount = stimer->count;
 560         return 0;
 561 }
 562
 563 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
 564                              struct hv_message *src_msg)
 565 {
 566         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
 567         struct page *page;
 568         gpa_t gpa;
 569         struct hv_message *dst_msg;
 570         int r;
 571         struct hv_message_page *msg_page;
 572
 573         if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
 574                 return -ENOENT;
 575
 576         gpa = synic->msg_page & PAGE_MASK;
 577         page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
 578         if (is_error_page(page))
 579                 return -EFAULT;
 580
 581         msg_page = kmap_atomic(page);
 582         dst_msg = &msg_page->sint_message[sint];
 583         if (sync_cmpxchg(&dst_msg->header.message_type, HVMSG_NONE,
 584                          src_msg->header.message_type) != HVMSG_NONE) {
 585                 dst_msg->header.message_flags.msg_pending = 1;
 586                 r = -EAGAIN;
 587         } else {
 588                 memcpy(&dst_msg->u.payload, &src_msg->u.payload,
 589                        src_msg->header.payload_size);
 590                 dst_msg->header.message_type = src_msg->header.message_type;
 591                 dst_msg->header.payload_size = src_msg->header.payload_size;
 592                 r = synic_set_irq(synic, sint);
 593                 if (r >= 1)
 594                         r = 0;
 595                 else if (r == 0)
 596                         r = -EFAULT;
 597         }
 598         kunmap_atomic(msg_page);
 599         kvm_release_page_dirty(page);
 600         kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
 601         return r;
 602 }
 603
 604 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
 605 {
 606         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
 607         struct hv_message *msg = &stimer->msg;
 608         struct hv_timer_message_payload *payload =
 609                         (struct hv_timer_message_payload *)&msg->u.payload;
 610
 611         payload->expiration_time = stimer->exp_time;
 612         payload->delivery_time = get_time_ref_counter(vcpu->kvm);
 613         return synic_deliver_msg(vcpu_to_synic(vcpu),
 614                                  HV_STIMER_SINT(stimer->config), msg);
 615 }
 616
 617 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
 618 {
 619         int r;
 620
 621         stimer->msg_pending = true;
 622         r = stimer_send_msg(stimer);
 623         trace_kvm_hv_stimer_expiration(stimer_to_vcpu(stimer)->vcpu_id,
 624                                        stimer->index, r);
 625         if (!r) {
 626                 stimer->msg_pending = false;
 627                 if (!(stimer->config & HV_STIMER_PERIODIC))
 628                         stimer->config &= ~HV_STIMER_ENABLE;
 629         }
 630 }
 631
 632 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
 633 {
 634         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 635         struct kvm_vcpu_hv_stimer *stimer;
 636         u64 time_now, exp_time;
 637         int i;
 638
 639         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 640                 if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
 641                         stimer = &hv_vcpu->stimer[i];
 642                         if (stimer->config & HV_STIMER_ENABLE) {
 643                                 exp_time = stimer->exp_time;
 644
 645                                 if (exp_time) {
 646                                         time_now =
 647                                                 get_time_ref_counter(vcpu->kvm);
 648                                         if (time_now >= exp_time)
 649                                                 stimer_expiration(stimer);
 650                                 }
 651
 652                                 if ((stimer->config & HV_STIMER_ENABLE) &&
 653                                     stimer->count) {
 654                                         if (!stimer->msg_pending)
 655                                                 stimer_start(stimer);
 656                                 } else
 657                                         stimer_cleanup(stimer);
 658                         }
 659                 }
 660 }
 661
 662 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
 663 {
 664         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 665         int i;
 666
 667         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 668                 stimer_cleanup(&hv_vcpu->stimer[i]);
 669 }
 670
 671 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
 672 {
 673         struct hv_message *msg = &stimer->msg;
 674         struct hv_timer_message_payload *payload =
 675                         (struct hv_timer_message_payload *)&msg->u.payload;
 676
 677         memset(&msg->header, 0, sizeof(msg->header));
 678         msg->header.message_type = HVMSG_TIMER_EXPIRED;
 679         msg->header.payload_size = sizeof(*payload);
 680
 681         payload->timer_index = stimer->index;
 682         payload->expiration_time = 0;
 683         payload->delivery_time = 0;
 684 }
 685
 686 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
 687 {
 688         memset(stimer, 0, sizeof(*stimer));
 689         stimer->index = timer_index;
 690         hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
 691         stimer->timer.function = stimer_timer_callback;
 692         stimer_prepare_msg(stimer);
 693 }
 694
 695 void kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
 696 {
 697         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 698         int i;
 699
 700         synic_init(&hv_vcpu->synic);
 701
 702         bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
 703         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
 704                 stimer_init(&hv_vcpu->stimer[i], i);
 705 }
 706
 707 void kvm_hv_vcpu_postcreate(struct kvm_vcpu *vcpu)
 708 {
 709         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
 710
 711         hv_vcpu->vp_index = kvm_vcpu_get_idx(vcpu);
 712 }
 713
 714 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
 715 {
 716         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
 717
 718         /*
 719          * Hyper-V SynIC auto EOI SINT's are
 720          * not compatible with APICV, so deactivate APICV
 721          */
 722         kvm_vcpu_deactivate_apicv(vcpu);
 723         synic->active = true;
 724         synic->dont_zero_synic_pages = dont_zero_synic_pages;
 725         return 0;
 726 }
 727
 728 static bool kvm_hv_msr_partition_wide(u32 msr)
 729 {
 730         bool r = false;
 731
 732         switch (msr) {
 733         case HV_X64_MSR_GUEST_OS_ID:
 734         case HV_X64_MSR_HYPERCALL:
 735         case HV_X64_MSR_REFERENCE_TSC:
 736         case HV_X64_MSR_TIME_REF_COUNT:
 737         case HV_X64_MSR_CRASH_CTL:
 738         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
 739         case HV_X64_MSR_RESET:
 740         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
 741         case HV_X64_MSR_TSC_EMULATION_CONTROL:
 742         case HV_X64_MSR_TSC_EMULATION_STATUS:
 743                 r = true;
 744                 break;
 745         }
 746
 747         return r;
 748 }
 749
 750 static int kvm_hv_msr_get_crash_data(struct kvm_vcpu *vcpu,
 751                                      u32 index, u64 *pdata)
 752 {
 753         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 754
 755         if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param)))
 756                 return -EINVAL;
 757
 758         *pdata = hv->hv_crash_param[index];
 759         return 0;
 760 }
 761
 762 static int kvm_hv_msr_get_crash_ctl(struct kvm_vcpu *vcpu, u64 *pdata)
 763 {
 764         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 765
 766         *pdata = hv->hv_crash_ctl;
 767         return 0;
 768 }
 769
 770 static int kvm_hv_msr_set_crash_ctl(struct kvm_vcpu *vcpu, u64 data, bool host)
 771 {
 772         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 773
 774         if (host)
 775                 hv->hv_crash_ctl = data & HV_X64_MSR_CRASH_CTL_NOTIFY;
 776
 777         if (!host && (data & HV_X64_MSR_CRASH_CTL_NOTIFY)) {
 778
 779                 vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
 780                           hv->hv_crash_param[0],
 781                           hv->hv_crash_param[1],
 782                           hv->hv_crash_param[2],
 783                           hv->hv_crash_param[3],
 784                           hv->hv_crash_param[4]);
 785
 786                 /* Send notification about crash to user space */
 787                 kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
 788         }
 789
 790         return 0;
 791 }
 792
 793 static int kvm_hv_msr_set_crash_data(struct kvm_vcpu *vcpu,
 794                                      u32 index, u64 data)
 795 {
 796         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
 797
 798         if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param)))
 799                 return -EINVAL;
 800
 801         hv->hv_crash_param[index] = data;
 802         return 0;
 803 }
 804
 805 /*
 806  * The kvmclock and Hyper-V TSC page use similar formulas, and converting
 807  * between them is possible:
 808  *
 809  * kvmclock formula:
 810  *    nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
 811  *           + system_time
 812  *
 813  * Hyper-V formula:
 814  *    nsec/100 = ticks * scale / 2^64 + offset
 815  *
 816  * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
 817  * By dividing the kvmclock formula by 100 and equating what's left we get:
 818  *    ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
 819  *            scale / 2^64 =         tsc_to_system_mul * 2^(tsc_shift-32) / 100
 820  *            scale        =         tsc_to_system_mul * 2^(32+tsc_shift) / 100
 821  *
 822  * Now expand the kvmclock formula and divide by 100:
 823  *    nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
 824  *           - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
 825  *           + system_time
 826  *    nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
 827  *               - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
 828  *               + system_time / 100
 829  *
 830  * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
 831  *    nsec/100 = ticks * scale / 2^64
 832  *               - tsc_timestamp * scale / 2^64
 833  *               + system_time / 100
 834  *
 835  * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
 836  *    offset = system_time / 100 - tsc_timestamp * scale / 2^64
 837  *
 838  * These two equivalencies are implemented in this function.
 839  */
 840 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
 841                                         HV_REFERENCE_TSC_PAGE *tsc_ref)
 842 {
 843         u64 max_mul;
 844
 845         if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
 846                 return false;
 847
 848         /*
 849          * check if scale would overflow, if so we use the time ref counter
 850          *    tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
 851          *    tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
 852          *    tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
 853          */
 854         max_mul = 100ull << (32 - hv_clock->tsc_shift);
 855         if (hv_clock->tsc_to_system_mul >= max_mul)
 856                 return false;
 857
 858         /*
 859          * Otherwise compute the scale and offset according to the formulas
 860          * derived above.
 861          */
 862         tsc_ref->tsc_scale =
 863                 mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
 864                                 hv_clock->tsc_to_system_mul,
 865                                 100);
 866
 867         tsc_ref->tsc_offset = hv_clock->system_time;
 868         do_div(tsc_ref->tsc_offset, 100);
 869         tsc_ref->tsc_offset -=
 870                 mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
 871         return true;
 872 }
 873
 874 void kvm_hv_setup_tsc_page(struct kvm *kvm,
 875                            struct pvclock_vcpu_time_info *hv_clock)
 876 {
 877         struct kvm_hv *hv = &kvm->arch.hyperv;
 878         u32 tsc_seq;
 879         u64 gfn;
 880
 881         BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
 882         BUILD_BUG_ON(offsetof(HV_REFERENCE_TSC_PAGE, tsc_sequence) != 0);
 883
 884         if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
 885                 return;
 886
 887         mutex_lock(&kvm->arch.hyperv.hv_lock);
 888         if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
 889                 goto out_unlock;
 890
 891         gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
 892         /*
 893          * Because the TSC parameters only vary when there is a
 894          * change in the master clock, do not bother with caching.
 895          */
 896         if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
 897                                     &tsc_seq, sizeof(tsc_seq))))
 898                 goto out_unlock;
 899
 900         /*
 901          * While we're computing and writing the parameters, force the
 902          * guest to use the time reference count MSR.
 903          */
 904         hv->tsc_ref.tsc_sequence = 0;
 905         if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
 906                             &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
 907                 goto out_unlock;
 908
 909         if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
 910                 goto out_unlock;
 911
 912         /* Ensure sequence is zero before writing the rest of the struct.  */
 913         smp_wmb();
 914         if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
 915                 goto out_unlock;
 916
 917         /*
 918          * Now switch to the TSC page mechanism by writing the sequence.
 919          */
 920         tsc_seq++;
 921         if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
 922                 tsc_seq = 1;
 923
 924         /* Write the struct entirely before the non-zero sequence.  */
 925         smp_wmb();
 926
 927         hv->tsc_ref.tsc_sequence = tsc_seq;
 928         kvm_write_guest(kvm, gfn_to_gpa(gfn),
 929                         &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence));
 930 out_unlock:
 931         mutex_unlock(&kvm->arch.hyperv.hv_lock);
 932 }
 933
 934 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
 935                              bool host)
 936 {
 937         struct kvm *kvm = vcpu->kvm;
 938         struct kvm_hv *hv = &kvm->arch.hyperv;
 939
 940         switch (msr) {
 941         case HV_X64_MSR_GUEST_OS_ID:
 942                 hv->hv_guest_os_id = data;
 943                 /* setting guest os id to zero disables hypercall page */
 944                 if (!hv->hv_guest_os_id)
 945                         hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
 946                 break;
 947         case HV_X64_MSR_HYPERCALL: {
 948                 u64 gfn;
 949                 unsigned long addr;
 950                 u8 instructions[4];
 951
 952                 /* if guest os id is not set hypercall should remain disabled */
 953                 if (!hv->hv_guest_os_id)
 954                         break;
 955                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
 956                         hv->hv_hypercall = data;
 957                         break;
 958                 }
 959                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
 960                 addr = gfn_to_hva(kvm, gfn);
 961                 if (kvm_is_error_hva(addr))
 962                         return 1;
 963                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
 964                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
 965                 if (__copy_to_user((void __user *)addr, instructions, 4))
 966                         return 1;
 967                 hv->hv_hypercall = data;
 968                 mark_page_dirty(kvm, gfn);
 969                 break;
 970         }
 971         case HV_X64_MSR_REFERENCE_TSC:
 972                 hv->hv_tsc_page = data;
 973                 if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE)
 974                         kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
 975                 break;
 976         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
 977                 return kvm_hv_msr_set_crash_data(vcpu,
 978                                                  msr - HV_X64_MSR_CRASH_P0,
 979                                                  data);
 980         case HV_X64_MSR_CRASH_CTL:
 981                 return kvm_hv_msr_set_crash_ctl(vcpu, data, host);
 982         case HV_X64_MSR_RESET:
 983                 if (data == 1) {
 984                         vcpu_debug(vcpu, "hyper-v reset requested\n");
 985                         kvm_make_request(KVM_REQ_HV_RESET, vcpu);
 986                 }
 987                 break;
 988         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
 989                 hv->hv_reenlightenment_control = data;
 990                 break;
 991         case HV_X64_MSR_TSC_EMULATION_CONTROL:
 992                 hv->hv_tsc_emulation_control = data;
 993                 break;
 994         case HV_X64_MSR_TSC_EMULATION_STATUS:
 995                 hv->hv_tsc_emulation_status = data;
 996                 break;
 997         default:
 998                 vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
 999                             msr, data);
1000                 return 1;
1001         }
1002         return 0;
1003 }
1004
1005 /* Calculate cpu time spent by current task in 100ns units */
1006 static u64 current_task_runtime_100ns(void)
1007 {
1008         u64 utime, stime;
1009
1010         task_cputime_adjusted(current, &utime, &stime);
1011
1012         return div_u64(utime + stime, 100);
1013 }
1014
1015 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1016 {
1017         struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
1018
1019         switch (msr) {
1020         case HV_X64_MSR_VP_INDEX:
1021                 if (!host)
1022                         return 1;
1023                 hv->vp_index = (u32)data;
1024                 break;
1025         case HV_X64_MSR_APIC_ASSIST_PAGE: {
1026                 u64 gfn;
1027                 unsigned long addr;
1028
1029                 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1030                         hv->hv_vapic = data;
1031                         if (kvm_lapic_enable_pv_eoi(vcpu, 0))
1032                                 return 1;
1033                         break;
1034                 }
1035                 gfn = data >> HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT;
1036                 addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1037                 if (kvm_is_error_hva(addr))
1038                         return 1;
1039                 if (__clear_user((void __user *)addr, PAGE_SIZE))
1040                         return 1;
1041                 hv->hv_vapic = data;
1042                 kvm_vcpu_mark_page_dirty(vcpu, gfn);
1043                 if (kvm_lapic_enable_pv_eoi(vcpu,
1044                                             gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
1045                         return 1;
1046                 break;
1047         }
1048         case HV_X64_MSR_EOI:
1049                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1050         case HV_X64_MSR_ICR:
1051                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1052         case HV_X64_MSR_TPR:
1053                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1054         case HV_X64_MSR_VP_RUNTIME:
1055                 if (!host)
1056                         return 1;
1057                 hv->runtime_offset = data - current_task_runtime_100ns();
1058                 break;
1059         case HV_X64_MSR_SCONTROL:
1060         case HV_X64_MSR_SVERSION:
1061         case HV_X64_MSR_SIEFP:
1062         case HV_X64_MSR_SIMP:
1063         case HV_X64_MSR_EOM:
1064         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1065                 return synic_set_msr(vcpu_to_synic(vcpu), msr, data, host);
1066         case HV_X64_MSR_STIMER0_CONFIG:
1067         case HV_X64_MSR_STIMER1_CONFIG:
1068         case HV_X64_MSR_STIMER2_CONFIG:
1069         case HV_X64_MSR_STIMER3_CONFIG: {
1070                 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1071
1072                 return stimer_set_config(vcpu_to_stimer(vcpu, timer_index),
1073                                          data, host);
1074         }
1075         case HV_X64_MSR_STIMER0_COUNT:
1076         case HV_X64_MSR_STIMER1_COUNT:
1077         case HV_X64_MSR_STIMER2_COUNT:
1078         case HV_X64_MSR_STIMER3_COUNT: {
1079                 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1080
1081                 return stimer_set_count(vcpu_to_stimer(vcpu, timer_index),
1082                                         data, host);
1083         }
1084         default:
1085                 vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
1086                             msr, data);
1087                 return 1;
1088         }
1089
1090         return 0;
1091 }
1092
1093 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1094 {
1095         u64 data = 0;
1096         struct kvm *kvm = vcpu->kvm;
1097         struct kvm_hv *hv = &kvm->arch.hyperv;
1098
1099         switch (msr) {
1100         case HV_X64_MSR_GUEST_OS_ID:
1101                 data = hv->hv_guest_os_id;
1102                 break;
1103         case HV_X64_MSR_HYPERCALL:
1104                 data = hv->hv_hypercall;
1105                 break;
1106         case HV_X64_MSR_TIME_REF_COUNT:
1107                 data = get_time_ref_counter(kvm);
1108                 break;
1109         case HV_X64_MSR_REFERENCE_TSC:
1110                 data = hv->hv_tsc_page;
1111                 break;
1112         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1113                 return kvm_hv_msr_get_crash_data(vcpu,
1114                                                  msr - HV_X64_MSR_CRASH_P0,
1115                                                  pdata);
1116         case HV_X64_MSR_CRASH_CTL:
1117                 return kvm_hv_msr_get_crash_ctl(vcpu, pdata);
1118         case HV_X64_MSR_RESET:
1119                 data = 0;
1120                 break;
1121         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1122                 data = hv->hv_reenlightenment_control;
1123                 break;
1124         case HV_X64_MSR_TSC_EMULATION_CONTROL:
1125                 data = hv->hv_tsc_emulation_control;
1126                 break;
1127         case HV_X64_MSR_TSC_EMULATION_STATUS:
1128                 data = hv->hv_tsc_emulation_status;
1129                 break;
1130         default:
1131                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1132                 return 1;
1133         }
1134
1135         *pdata = data;
1136         return 0;
1137 }
1138
1139 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1140 {
1141         u64 data = 0;
1142         struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
1143
1144         switch (msr) {
1145         case HV_X64_MSR_VP_INDEX:
1146                 data = hv->vp_index;
1147                 break;
1148         case HV_X64_MSR_EOI:
1149                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1150         case HV_X64_MSR_ICR:
1151                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1152         case HV_X64_MSR_TPR:
1153                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1154         case HV_X64_MSR_APIC_ASSIST_PAGE:
1155                 data = hv->hv_vapic;
1156                 break;
1157         case HV_X64_MSR_VP_RUNTIME:
1158                 data = current_task_runtime_100ns() + hv->runtime_offset;
1159                 break;
1160         case HV_X64_MSR_SCONTROL:
1161         case HV_X64_MSR_SVERSION:
1162         case HV_X64_MSR_SIEFP:
1163         case HV_X64_MSR_SIMP:
1164         case HV_X64_MSR_EOM:
1165         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1166                 return synic_get_msr(vcpu_to_synic(vcpu), msr, pdata);
1167         case HV_X64_MSR_STIMER0_CONFIG:
1168         case HV_X64_MSR_STIMER1_CONFIG:
1169         case HV_X64_MSR_STIMER2_CONFIG:
1170         case HV_X64_MSR_STIMER3_CONFIG: {
1171                 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1172
1173                 return stimer_get_config(vcpu_to_stimer(vcpu, timer_index),
1174                                          pdata);
1175         }
1176         case HV_X64_MSR_STIMER0_COUNT:
1177         case HV_X64_MSR_STIMER1_COUNT:
1178         case HV_X64_MSR_STIMER2_COUNT:
1179         case HV_X64_MSR_STIMER3_COUNT: {
1180                 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1181
1182                 return stimer_get_count(vcpu_to_stimer(vcpu, timer_index),
1183                                         pdata);
1184         }
1185         case HV_X64_MSR_TSC_FREQUENCY:
1186                 data = (u64)vcpu->arch.virtual_tsc_khz * 1000;
1187                 break;
1188         case HV_X64_MSR_APIC_FREQUENCY:
1189                 data = APIC_BUS_FREQUENCY;
1190                 break;
1191         default:
1192                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1193                 return 1;
1194         }
1195         *pdata = data;
1196         return 0;
1197 }
1198
1199 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1200 {
1201         if (kvm_hv_msr_partition_wide(msr)) {
1202                 int r;
1203
1204                 mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1205                 r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1206                 mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1207                 return r;
1208         } else
1209                 return kvm_hv_set_msr(vcpu, msr, data, host);
1210 }
1211
1212 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1213 {
1214         if (kvm_hv_msr_partition_wide(msr)) {
1215                 int r;
1216
1217                 mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1218                 r = kvm_hv_get_msr_pw(vcpu, msr, pdata);
1219                 mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1220                 return r;
1221         } else
1222                 return kvm_hv_get_msr(vcpu, msr, pdata);
1223 }
1224
1225 bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1226 {
1227         return READ_ONCE(kvm->arch.hyperv.hv_hypercall) & HV_X64_MSR_HYPERCALL_ENABLE;
1228 }
1229
1230 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
1231 {
1232         bool longmode;
1233
1234         longmode = is_64_bit_mode(vcpu);
1235         if (longmode)
1236                 kvm_register_write(vcpu, VCPU_REGS_RAX, result);
1237         else {
1238                 kvm_register_write(vcpu, VCPU_REGS_RDX, result >> 32);
1239                 kvm_register_write(vcpu, VCPU_REGS_RAX, result & 0xffffffff);
1240         }
1241 }
1242
1243 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
1244 {
1245         struct kvm_run *run = vcpu->run;
1246
1247         kvm_hv_hypercall_set_result(vcpu, run->hyperv.u.hcall.result);
1248         return 1;
1249 }
1250
1251 static u16 kvm_hvcall_signal_event(struct kvm_vcpu *vcpu, bool fast, u64 param)
1252 {
1253         struct eventfd_ctx *eventfd;
1254
1255         if (unlikely(!fast)) {
1256                 int ret;
1257                 gpa_t gpa = param;
1258
1259                 if ((gpa & (__alignof__(param) - 1)) ||
1260                     offset_in_page(gpa) + sizeof(param) > PAGE_SIZE)
1261                         return HV_STATUS_INVALID_ALIGNMENT;
1262
1263                 ret = kvm_vcpu_read_guest(vcpu, gpa, &param, sizeof(param));
1264                 if (ret < 0)
1265                         return HV_STATUS_INVALID_ALIGNMENT;
1266         }
1267
1268         /*
1269          * Per spec, bits 32-47 contain the extra "flag number".  However, we
1270          * have no use for it, and in all known usecases it is zero, so just
1271          * report lookup failure if it isn't.
1272          */
1273         if (param & 0xffff00000000ULL)
1274                 return HV_STATUS_INVALID_PORT_ID;
1275         /* remaining bits are reserved-zero */
1276         if (param & ~KVM_HYPERV_CONN_ID_MASK)
1277                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1278
1279         /* conn_to_evt is protected by vcpu->kvm->srcu */
1280         eventfd = idr_find(&vcpu->kvm->arch.hyperv.conn_to_evt, param);
1281         if (!eventfd)
1282                 return HV_STATUS_INVALID_PORT_ID;
1283
1284         eventfd_signal(eventfd, 1);
1285         return HV_STATUS_SUCCESS;
1286 }
1287
1288 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
1289 {
1290         u64 param, ingpa, outgpa, ret;
1291         uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
1292         bool fast, longmode;
1293
1294         /*
1295          * hypercall generates UD from non zero cpl and real mode
1296          * per HYPER-V spec
1297          */
1298         if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
1299                 kvm_queue_exception(vcpu, UD_VECTOR);
1300                 return 1;
1301         }
1302
1303         longmode = is_64_bit_mode(vcpu);
1304
1305         if (!longmode) {
1306                 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
1307                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
1308                 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
1309                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
1310                 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
1311                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
1312         }
1313 #ifdef CONFIG_X86_64
1314         else {
1315                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
1316                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
1317                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
1318         }
1319 #endif
1320
1321         code = param & 0xffff;
1322         fast = (param >> 16) & 0x1;
1323         rep_cnt = (param >> 32) & 0xfff;
1324         rep_idx = (param >> 48) & 0xfff;
1325
1326         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
1327
1328         /* Hypercall continuation is not supported yet */
1329         if (rep_cnt || rep_idx) {
1330                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
1331                 goto set_result;
1332         }
1333
1334         switch (code) {
1335         case HVCALL_NOTIFY_LONG_SPIN_WAIT:
1336                 kvm_vcpu_on_spin(vcpu, true);
1337                 break;
1338         case HVCALL_SIGNAL_EVENT:
1339                 res = kvm_hvcall_signal_event(vcpu, fast, ingpa);
1340                 if (res != HV_STATUS_INVALID_PORT_ID)
1341                         break;
1342                 /* maybe userspace knows this conn_id: fall through */
1343         case HVCALL_POST_MESSAGE:
1344                 /* don't bother userspace if it has no way to handle it */
1345                 if (!vcpu_to_synic(vcpu)->active) {
1346                         res = HV_STATUS_INVALID_HYPERCALL_CODE;
1347                         break;
1348                 }
1349                 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
1350                 vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
1351                 vcpu->run->hyperv.u.hcall.input = param;
1352                 vcpu->run->hyperv.u.hcall.params[0] = ingpa;
1353                 vcpu->run->hyperv.u.hcall.params[1] = outgpa;
1354                 vcpu->arch.complete_userspace_io =
1355                                 kvm_hv_hypercall_complete_userspace;
1356                 return 0;
1357         default:
1358                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
1359                 break;
1360         }
1361
1362 set_result:
1363         ret = res | (((u64)rep_done & 0xfff) << 32);
1364         kvm_hv_hypercall_set_result(vcpu, ret);
1365         return 1;
1366 }
1367
1368 void kvm_hv_init_vm(struct kvm *kvm)
1369 {
1370         mutex_init(&kvm->arch.hyperv.hv_lock);
1371         idr_init(&kvm->arch.hyperv.conn_to_evt);
1372 }
1373
1374 void kvm_hv_destroy_vm(struct kvm *kvm)
1375 {
1376         struct eventfd_ctx *eventfd;
1377         int i;
1378
1379         idr_for_each_entry(&kvm->arch.hyperv.conn_to_evt, eventfd, i)
1380                 eventfd_ctx_put(eventfd);
1381         idr_destroy(&kvm->arch.hyperv.conn_to_evt);
1382 }
1383
1384 static int kvm_hv_eventfd_assign(struct kvm *kvm, u32 conn_id, int fd)
1385 {
1386         struct kvm_hv *hv = &kvm->arch.hyperv;
1387         struct eventfd_ctx *eventfd;
1388         int ret;
1389
1390         eventfd = eventfd_ctx_fdget(fd);
1391         if (IS_ERR(eventfd))
1392                 return PTR_ERR(eventfd);
1393
1394         mutex_lock(&hv->hv_lock);
1395         ret = idr_alloc(&hv->conn_to_evt, eventfd, conn_id, conn_id + 1,
1396                         GFP_KERNEL);
1397         mutex_unlock(&hv->hv_lock);
1398
1399         if (ret >= 0)
1400                 return 0;
1401
1402         if (ret == -ENOSPC)
1403                 ret = -EEXIST;
1404         eventfd_ctx_put(eventfd);
1405         return ret;
1406 }
1407
1408 static int kvm_hv_eventfd_deassign(struct kvm *kvm, u32 conn_id)
1409 {
1410         struct kvm_hv *hv = &kvm->arch.hyperv;
1411         struct eventfd_ctx *eventfd;
1412
1413         mutex_lock(&hv->hv_lock);
1414         eventfd = idr_remove(&hv->conn_to_evt, conn_id);
1415         mutex_unlock(&hv->hv_lock);
1416
1417         if (!eventfd)
1418                 return -ENOENT;
1419
1420         synchronize_srcu(&kvm->srcu);
1421         eventfd_ctx_put(eventfd);
1422         return 0;
1423 }
1424
1425 int kvm_vm_ioctl_hv_eventfd(struct kvm *kvm, struct kvm_hyperv_eventfd *args)
1426 {
1427         if ((args->flags & ~KVM_HYPERV_EVENTFD_DEASSIGN) ||
1428             (args->conn_id & ~KVM_HYPERV_CONN_ID_MASK))
1429                 return -EINVAL;
1430
1431         if (args->flags == KVM_HYPERV_EVENTFD_DEASSIGN)
1432                 return kvm_hv_eventfd_deassign(kvm, args->conn_id);
1433         return kvm_hv_eventfd_assign(kvm, args->conn_id, args->fd);
1434 }