1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * KVM Microsoft Hyper-V emulation 4 * 5 * derived from arch/x86/kvm/x86.c 6 * 7 * Copyright (C) 2006 Qumranet, Inc. 8 * Copyright (C) 2008 Qumranet, Inc. 9 * Copyright IBM Corporation, 2008 10 * Copyright 2010 Red Hat, Inc. and/or its affiliates. 11 * Copyright (C) 2015 Andrey Smetanin <asmetanin@virtuozzo.com> 12 * 13 * Authors: 14 * Avi Kivity <avi@qumranet.com> 15 * Yaniv Kamay <yaniv@qumranet.com> 16 * Amit Shah <amit.shah@qumranet.com> 17 * Ben-Ami Yassour <benami@il.ibm.com> 18 * Andrey Smetanin <asmetanin@virtuozzo.com> 19 */ 20 21 #include "x86.h" 22 #include "lapic.h" 23 #include "ioapic.h" 24 #include "cpuid.h" 25 #include "hyperv.h" 26 #include "xen.h" 27 28 #include <linux/cpu.h> 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 #include "irq.h" 39 40 /* "Hv#1" signature */ 41 #define HYPERV_CPUID_SIGNATURE_EAX 0x31237648 42 43 #define KVM_HV_MAX_SPARSE_VCPU_SET_BITS DIV_ROUND_UP(KVM_MAX_VCPUS, 64) 44 45 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer, 46 bool vcpu_kick); 47 48 static inline u64 synic_read_sint(struct kvm_vcpu_hv_synic *synic, int sint) 49 { 50 return atomic64_read(&synic->sint[sint]); 51 } 52 53 static inline int synic_get_sint_vector(u64 sint_value) 54 { 55 if (sint_value & HV_SYNIC_SINT_MASKED) 56 return -1; 57 return sint_value & HV_SYNIC_SINT_VECTOR_MASK; 58 } 59 60 static bool synic_has_vector_connected(struct kvm_vcpu_hv_synic *synic, 61 int vector) 62 { 63 int i; 64 65 for (i = 0; i < ARRAY_SIZE(synic->sint); i++) { 66 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector) 67 return true; 68 } 69 return false; 70 } 71 72 static bool synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic *synic, 73 int vector) 74 { 75 int i; 76 u64 sint_value; 77 78 for (i = 0; i < ARRAY_SIZE(synic->sint); i++) { 79 sint_value = synic_read_sint(synic, i); 80 if (synic_get_sint_vector(sint_value) == vector && 81 sint_value & HV_SYNIC_SINT_AUTO_EOI) 82 return true; 83 } 84 return false; 85 } 86 87 static void synic_update_vector(struct kvm_vcpu_hv_synic *synic, 88 int vector) 89 { 90 if (vector < HV_SYNIC_FIRST_VALID_VECTOR) 91 return; 92 93 if (synic_has_vector_connected(synic, vector)) 94 __set_bit(vector, synic->vec_bitmap); 95 else 96 __clear_bit(vector, synic->vec_bitmap); 97 98 if (synic_has_vector_auto_eoi(synic, vector)) 99 __set_bit(vector, synic->auto_eoi_bitmap); 100 else 101 __clear_bit(vector, synic->auto_eoi_bitmap); 102 } 103 104 static int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint, 105 u64 data, bool host) 106 { 107 int vector, old_vector; 108 bool masked; 109 110 vector = data & HV_SYNIC_SINT_VECTOR_MASK; 111 masked = data & HV_SYNIC_SINT_MASKED; 112 113 /* 114 * Valid vectors are 16-255, however, nested Hyper-V attempts to write 115 * default '0x10000' value on boot and this should not #GP. We need to 116 * allow zero-initing the register from host as well. 117 */ 118 if (vector < HV_SYNIC_FIRST_VALID_VECTOR && !host && !masked) 119 return 1; 120 /* 121 * Guest may configure multiple SINTs to use the same vector, so 122 * we maintain a bitmap of vectors handled by synic, and a 123 * bitmap of vectors with auto-eoi behavior. The bitmaps are 124 * updated here, and atomically queried on fast paths. 125 */ 126 old_vector = synic_read_sint(synic, sint) & HV_SYNIC_SINT_VECTOR_MASK; 127 128 atomic64_set(&synic->sint[sint], data); 129 130 synic_update_vector(synic, old_vector); 131 132 synic_update_vector(synic, vector); 133 134 /* Load SynIC vectors into EOI exit bitmap */ 135 kvm_make_request(KVM_REQ_SCAN_IOAPIC, hv_synic_to_vcpu(synic)); 136 return 0; 137 } 138 139 static struct kvm_vcpu *get_vcpu_by_vpidx(struct kvm *kvm, u32 vpidx) 140 { 141 struct kvm_vcpu *vcpu = NULL; 142 int i; 143 144 if (vpidx >= KVM_MAX_VCPUS) 145 return NULL; 146 147 vcpu = kvm_get_vcpu(kvm, vpidx); 148 if (vcpu && kvm_hv_get_vpindex(vcpu) == vpidx) 149 return vcpu; 150 kvm_for_each_vcpu(i, vcpu, kvm) 151 if (kvm_hv_get_vpindex(vcpu) == vpidx) 152 return vcpu; 153 return NULL; 154 } 155 156 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vpidx) 157 { 158 struct kvm_vcpu *vcpu; 159 struct kvm_vcpu_hv_synic *synic; 160 161 vcpu = get_vcpu_by_vpidx(kvm, vpidx); 162 if (!vcpu || !to_hv_vcpu(vcpu)) 163 return NULL; 164 synic = to_hv_synic(vcpu); 165 return (synic->active) ? synic : NULL; 166 } 167 168 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint) 169 { 170 struct kvm *kvm = vcpu->kvm; 171 struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu); 172 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu); 173 struct kvm_vcpu_hv_stimer *stimer; 174 int gsi, idx; 175 176 trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint); 177 178 /* Try to deliver pending Hyper-V SynIC timers messages */ 179 for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) { 180 stimer = &hv_vcpu->stimer[idx]; 181 if (stimer->msg_pending && stimer->config.enable && 182 !stimer->config.direct_mode && 183 stimer->config.sintx == sint) 184 stimer_mark_pending(stimer, false); 185 } 186 187 idx = srcu_read_lock(&kvm->irq_srcu); 188 gsi = atomic_read(&synic->sint_to_gsi[sint]); 189 if (gsi != -1) 190 kvm_notify_acked_gsi(kvm, gsi); 191 srcu_read_unlock(&kvm->irq_srcu, idx); 192 } 193 194 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr) 195 { 196 struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic); 197 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu); 198 199 hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC; 200 hv_vcpu->exit.u.synic.msr = msr; 201 hv_vcpu->exit.u.synic.control = synic->control; 202 hv_vcpu->exit.u.synic.evt_page = synic->evt_page; 203 hv_vcpu->exit.u.synic.msg_page = synic->msg_page; 204 205 kvm_make_request(KVM_REQ_HV_EXIT, vcpu); 206 } 207 208 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic, 209 u32 msr, u64 data, bool host) 210 { 211 struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic); 212 int ret; 213 214 if (!synic->active && !host) 215 return 1; 216 217 trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host); 218 219 ret = 0; 220 switch (msr) { 221 case HV_X64_MSR_SCONTROL: 222 synic->control = data; 223 if (!host) 224 synic_exit(synic, msr); 225 break; 226 case HV_X64_MSR_SVERSION: 227 if (!host) { 228 ret = 1; 229 break; 230 } 231 synic->version = data; 232 break; 233 case HV_X64_MSR_SIEFP: 234 if ((data & HV_SYNIC_SIEFP_ENABLE) && !host && 235 !synic->dont_zero_synic_pages) 236 if (kvm_clear_guest(vcpu->kvm, 237 data & PAGE_MASK, PAGE_SIZE)) { 238 ret = 1; 239 break; 240 } 241 synic->evt_page = data; 242 if (!host) 243 synic_exit(synic, msr); 244 break; 245 case HV_X64_MSR_SIMP: 246 if ((data & HV_SYNIC_SIMP_ENABLE) && !host && 247 !synic->dont_zero_synic_pages) 248 if (kvm_clear_guest(vcpu->kvm, 249 data & PAGE_MASK, PAGE_SIZE)) { 250 ret = 1; 251 break; 252 } 253 synic->msg_page = data; 254 if (!host) 255 synic_exit(synic, msr); 256 break; 257 case HV_X64_MSR_EOM: { 258 int i; 259 260 for (i = 0; i < ARRAY_SIZE(synic->sint); i++) 261 kvm_hv_notify_acked_sint(vcpu, i); 262 break; 263 } 264 case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15: 265 ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host); 266 break; 267 default: 268 ret = 1; 269 break; 270 } 271 return ret; 272 } 273 274 static bool kvm_hv_is_syndbg_enabled(struct kvm_vcpu *vcpu) 275 { 276 struct kvm_cpuid_entry2 *entry; 277 278 entry = kvm_find_cpuid_entry(vcpu, 279 HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES, 280 0); 281 if (!entry) 282 return false; 283 284 return entry->eax & HV_X64_SYNDBG_CAP_ALLOW_KERNEL_DEBUGGING; 285 } 286 287 static int kvm_hv_syndbg_complete_userspace(struct kvm_vcpu *vcpu) 288 { 289 struct kvm_hv *hv = to_kvm_hv(vcpu->kvm); 290 291 if (vcpu->run->hyperv.u.syndbg.msr == HV_X64_MSR_SYNDBG_CONTROL) 292 hv->hv_syndbg.control.status = 293 vcpu->run->hyperv.u.syndbg.status; 294 return 1; 295 } 296 297 static void syndbg_exit(struct kvm_vcpu *vcpu, u32 msr) 298 { 299 struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu); 300 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu); 301 302 hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNDBG; 303 hv_vcpu->exit.u.syndbg.msr = msr; 304 hv_vcpu->exit.u.syndbg.control = syndbg->control.control; 305 hv_vcpu->exit.u.syndbg.send_page = syndbg->control.send_page; 306 hv_vcpu->exit.u.syndbg.recv_page = syndbg->control.recv_page; 307 hv_vcpu->exit.u.syndbg.pending_page = syndbg->control.pending_page; 308 vcpu->arch.complete_userspace_io = 309 kvm_hv_syndbg_complete_userspace; 310 311 kvm_make_request(KVM_REQ_HV_EXIT, vcpu); 312 } 313 314 static int syndbg_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host) 315 { 316 struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu); 317 318 if (!kvm_hv_is_syndbg_enabled(vcpu) && !host) 319 return 1; 320 321 trace_kvm_hv_syndbg_set_msr(vcpu->vcpu_id, 322 to_hv_vcpu(vcpu)->vp_index, msr, data); 323 switch (msr) { 324 case HV_X64_MSR_SYNDBG_CONTROL: 325 syndbg->control.control = data; 326 if (!host) 327 syndbg_exit(vcpu, msr); 328 break; 329 case HV_X64_MSR_SYNDBG_STATUS: 330 syndbg->control.status = data; 331 break; 332 case HV_X64_MSR_SYNDBG_SEND_BUFFER: 333 syndbg->control.send_page = data; 334 break; 335 case HV_X64_MSR_SYNDBG_RECV_BUFFER: 336 syndbg->control.recv_page = data; 337 break; 338 case HV_X64_MSR_SYNDBG_PENDING_BUFFER: 339 syndbg->control.pending_page = data; 340 if (!host) 341 syndbg_exit(vcpu, msr); 342 break; 343 case HV_X64_MSR_SYNDBG_OPTIONS: 344 syndbg->options = data; 345 break; 346 default: 347 break; 348 } 349 350 return 0; 351 } 352 353 static int syndbg_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host) 354 { 355 struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu); 356 357 if (!kvm_hv_is_syndbg_enabled(vcpu) && !host) 358 return 1; 359 360 switch (msr) { 361 case HV_X64_MSR_SYNDBG_CONTROL: 362 *pdata = syndbg->control.control; 363 break; 364 case HV_X64_MSR_SYNDBG_STATUS: 365 *pdata = syndbg->control.status; 366 break; 367 case HV_X64_MSR_SYNDBG_SEND_BUFFER: 368 *pdata = syndbg->control.send_page; 369 break; 370 case HV_X64_MSR_SYNDBG_RECV_BUFFER: 371 *pdata = syndbg->control.recv_page; 372 break; 373 case HV_X64_MSR_SYNDBG_PENDING_BUFFER: 374 *pdata = syndbg->control.pending_page; 375 break; 376 case HV_X64_MSR_SYNDBG_OPTIONS: 377 *pdata = syndbg->options; 378 break; 379 default: 380 break; 381 } 382 383 trace_kvm_hv_syndbg_get_msr(vcpu->vcpu_id, kvm_hv_get_vpindex(vcpu), msr, *pdata); 384 385 return 0; 386 } 387 388 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata, 389 bool host) 390 { 391 int ret; 392 393 if (!synic->active && !host) 394 return 1; 395 396 ret = 0; 397 switch (msr) { 398 case HV_X64_MSR_SCONTROL: 399 *pdata = synic->control; 400 break; 401 case HV_X64_MSR_SVERSION: 402 *pdata = synic->version; 403 break; 404 case HV_X64_MSR_SIEFP: 405 *pdata = synic->evt_page; 406 break; 407 case HV_X64_MSR_SIMP: 408 *pdata = synic->msg_page; 409 break; 410 case HV_X64_MSR_EOM: 411 *pdata = 0; 412 break; 413 case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15: 414 *pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]); 415 break; 416 default: 417 ret = 1; 418 break; 419 } 420 return ret; 421 } 422 423 static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint) 424 { 425 struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic); 426 struct kvm_lapic_irq irq; 427 int ret, vector; 428 429 if (sint >= ARRAY_SIZE(synic->sint)) 430 return -EINVAL; 431 432 vector = synic_get_sint_vector(synic_read_sint(synic, sint)); 433 if (vector < 0) 434 return -ENOENT; 435 436 memset(&irq, 0, sizeof(irq)); 437 irq.shorthand = APIC_DEST_SELF; 438 irq.dest_mode = APIC_DEST_PHYSICAL; 439 irq.delivery_mode = APIC_DM_FIXED; 440 irq.vector = vector; 441 irq.level = 1; 442 443 ret = kvm_irq_delivery_to_apic(vcpu->kvm, vcpu->arch.apic, &irq, NULL); 444 trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret); 445 return ret; 446 } 447 448 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vpidx, u32 sint) 449 { 450 struct kvm_vcpu_hv_synic *synic; 451 452 synic = synic_get(kvm, vpidx); 453 if (!synic) 454 return -EINVAL; 455 456 return synic_set_irq(synic, sint); 457 } 458 459 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector) 460 { 461 struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu); 462 int i; 463 464 trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector); 465 466 for (i = 0; i < ARRAY_SIZE(synic->sint); i++) 467 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector) 468 kvm_hv_notify_acked_sint(vcpu, i); 469 } 470 471 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi) 472 { 473 struct kvm_vcpu_hv_synic *synic; 474 475 synic = synic_get(kvm, vpidx); 476 if (!synic) 477 return -EINVAL; 478 479 if (sint >= ARRAY_SIZE(synic->sint_to_gsi)) 480 return -EINVAL; 481 482 atomic_set(&synic->sint_to_gsi[sint], gsi); 483 return 0; 484 } 485 486 void kvm_hv_irq_routing_update(struct kvm *kvm) 487 { 488 struct kvm_irq_routing_table *irq_rt; 489 struct kvm_kernel_irq_routing_entry *e; 490 u32 gsi; 491 492 irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu, 493 lockdep_is_held(&kvm->irq_lock)); 494 495 for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) { 496 hlist_for_each_entry(e, &irq_rt->map[gsi], link) { 497 if (e->type == KVM_IRQ_ROUTING_HV_SINT) 498 kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu, 499 e->hv_sint.sint, gsi); 500 } 501 } 502 } 503 504 static void synic_init(struct kvm_vcpu_hv_synic *synic) 505 { 506 int i; 507 508 memset(synic, 0, sizeof(*synic)); 509 synic->version = HV_SYNIC_VERSION_1; 510 for (i = 0; i < ARRAY_SIZE(synic->sint); i++) { 511 atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED); 512 atomic_set(&synic->sint_to_gsi[i], -1); 513 } 514 } 515 516 static u64 get_time_ref_counter(struct kvm *kvm) 517 { 518 struct kvm_hv *hv = to_kvm_hv(kvm); 519 struct kvm_vcpu *vcpu; 520 u64 tsc; 521 522 /* 523 * Fall back to get_kvmclock_ns() when TSC page hasn't been set up, 524 * is broken, disabled or being updated. 525 */ 526 if (hv->hv_tsc_page_status != HV_TSC_PAGE_SET) 527 return div_u64(get_kvmclock_ns(kvm), 100); 528 529 vcpu = kvm_get_vcpu(kvm, 0); 530 tsc = kvm_read_l1_tsc(vcpu, rdtsc()); 531 return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64) 532 + hv->tsc_ref.tsc_offset; 533 } 534 535 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer, 536 bool vcpu_kick) 537 { 538 struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer); 539 540 set_bit(stimer->index, 541 to_hv_vcpu(vcpu)->stimer_pending_bitmap); 542 kvm_make_request(KVM_REQ_HV_STIMER, vcpu); 543 if (vcpu_kick) 544 kvm_vcpu_kick(vcpu); 545 } 546 547 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer) 548 { 549 struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer); 550 551 trace_kvm_hv_stimer_cleanup(hv_stimer_to_vcpu(stimer)->vcpu_id, 552 stimer->index); 553 554 hrtimer_cancel(&stimer->timer); 555 clear_bit(stimer->index, 556 to_hv_vcpu(vcpu)->stimer_pending_bitmap); 557 stimer->msg_pending = false; 558 stimer->exp_time = 0; 559 } 560 561 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer) 562 { 563 struct kvm_vcpu_hv_stimer *stimer; 564 565 stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer); 566 trace_kvm_hv_stimer_callback(hv_stimer_to_vcpu(stimer)->vcpu_id, 567 stimer->index); 568 stimer_mark_pending(stimer, true); 569 570 return HRTIMER_NORESTART; 571 } 572 573 /* 574 * stimer_start() assumptions: 575 * a) stimer->count is not equal to 0 576 * b) stimer->config has HV_STIMER_ENABLE flag 577 */ 578 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer) 579 { 580 u64 time_now; 581 ktime_t ktime_now; 582 583 time_now = get_time_ref_counter(hv_stimer_to_vcpu(stimer)->kvm); 584 ktime_now = ktime_get(); 585 586 if (stimer->config.periodic) { 587 if (stimer->exp_time) { 588 if (time_now >= stimer->exp_time) { 589 u64 remainder; 590 591 div64_u64_rem(time_now - stimer->exp_time, 592 stimer->count, &remainder); 593 stimer->exp_time = 594 time_now + (stimer->count - remainder); 595 } 596 } else 597 stimer->exp_time = time_now + stimer->count; 598 599 trace_kvm_hv_stimer_start_periodic( 600 hv_stimer_to_vcpu(stimer)->vcpu_id, 601 stimer->index, 602 time_now, stimer->exp_time); 603 604 hrtimer_start(&stimer->timer, 605 ktime_add_ns(ktime_now, 606 100 * (stimer->exp_time - time_now)), 607 HRTIMER_MODE_ABS); 608 return 0; 609 } 610 stimer->exp_time = stimer->count; 611 if (time_now >= stimer->count) { 612 /* 613 * Expire timer according to Hypervisor Top-Level Functional 614 * specification v4(15.3.1): 615 * "If a one shot is enabled and the specified count is in 616 * the past, it will expire immediately." 617 */ 618 stimer_mark_pending(stimer, false); 619 return 0; 620 } 621 622 trace_kvm_hv_stimer_start_one_shot(hv_stimer_to_vcpu(stimer)->vcpu_id, 623 stimer->index, 624 time_now, stimer->count); 625 626 hrtimer_start(&stimer->timer, 627 ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)), 628 HRTIMER_MODE_ABS); 629 return 0; 630 } 631 632 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config, 633 bool host) 634 { 635 union hv_stimer_config new_config = {.as_uint64 = config}, 636 old_config = {.as_uint64 = stimer->config.as_uint64}; 637 struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer); 638 struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu); 639 640 if (!synic->active && !host) 641 return 1; 642 643 trace_kvm_hv_stimer_set_config(hv_stimer_to_vcpu(stimer)->vcpu_id, 644 stimer->index, config, host); 645 646 stimer_cleanup(stimer); 647 if (old_config.enable && 648 !new_config.direct_mode && new_config.sintx == 0) 649 new_config.enable = 0; 650 stimer->config.as_uint64 = new_config.as_uint64; 651 652 if (stimer->config.enable) 653 stimer_mark_pending(stimer, false); 654 655 return 0; 656 } 657 658 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count, 659 bool host) 660 { 661 struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer); 662 struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu); 663 664 if (!synic->active && !host) 665 return 1; 666 667 trace_kvm_hv_stimer_set_count(hv_stimer_to_vcpu(stimer)->vcpu_id, 668 stimer->index, count, host); 669 670 stimer_cleanup(stimer); 671 stimer->count = count; 672 if (stimer->count == 0) 673 stimer->config.enable = 0; 674 else if (stimer->config.auto_enable) 675 stimer->config.enable = 1; 676 677 if (stimer->config.enable) 678 stimer_mark_pending(stimer, false); 679 680 return 0; 681 } 682 683 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig) 684 { 685 *pconfig = stimer->config.as_uint64; 686 return 0; 687 } 688 689 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount) 690 { 691 *pcount = stimer->count; 692 return 0; 693 } 694 695 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint, 696 struct hv_message *src_msg, bool no_retry) 697 { 698 struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic); 699 int msg_off = offsetof(struct hv_message_page, sint_message[sint]); 700 gfn_t msg_page_gfn; 701 struct hv_message_header hv_hdr; 702 int r; 703 704 if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE)) 705 return -ENOENT; 706 707 msg_page_gfn = synic->msg_page >> PAGE_SHIFT; 708 709 /* 710 * Strictly following the spec-mandated ordering would assume setting 711 * .msg_pending before checking .message_type. However, this function 712 * is only called in vcpu context so the entire update is atomic from 713 * guest POV and thus the exact order here doesn't matter. 714 */ 715 r = kvm_vcpu_read_guest_page(vcpu, msg_page_gfn, &hv_hdr.message_type, 716 msg_off + offsetof(struct hv_message, 717 header.message_type), 718 sizeof(hv_hdr.message_type)); 719 if (r < 0) 720 return r; 721 722 if (hv_hdr.message_type != HVMSG_NONE) { 723 if (no_retry) 724 return 0; 725 726 hv_hdr.message_flags.msg_pending = 1; 727 r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn, 728 &hv_hdr.message_flags, 729 msg_off + 730 offsetof(struct hv_message, 731 header.message_flags), 732 sizeof(hv_hdr.message_flags)); 733 if (r < 0) 734 return r; 735 return -EAGAIN; 736 } 737 738 r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn, src_msg, msg_off, 739 sizeof(src_msg->header) + 740 src_msg->header.payload_size); 741 if (r < 0) 742 return r; 743 744 r = synic_set_irq(synic, sint); 745 if (r < 0) 746 return r; 747 if (r == 0) 748 return -EFAULT; 749 return 0; 750 } 751 752 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer) 753 { 754 struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer); 755 struct hv_message *msg = &stimer->msg; 756 struct hv_timer_message_payload *payload = 757 (struct hv_timer_message_payload *)&msg->u.payload; 758 759 /* 760 * To avoid piling up periodic ticks, don't retry message 761 * delivery for them (within "lazy" lost ticks policy). 762 */ 763 bool no_retry = stimer->config.periodic; 764 765 payload->expiration_time = stimer->exp_time; 766 payload->delivery_time = get_time_ref_counter(vcpu->kvm); 767 return synic_deliver_msg(to_hv_synic(vcpu), 768 stimer->config.sintx, msg, 769 no_retry); 770 } 771 772 static int stimer_notify_direct(struct kvm_vcpu_hv_stimer *stimer) 773 { 774 struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer); 775 struct kvm_lapic_irq irq = { 776 .delivery_mode = APIC_DM_FIXED, 777 .vector = stimer->config.apic_vector 778 }; 779 780 if (lapic_in_kernel(vcpu)) 781 return !kvm_apic_set_irq(vcpu, &irq, NULL); 782 return 0; 783 } 784 785 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer) 786 { 787 int r, direct = stimer->config.direct_mode; 788 789 stimer->msg_pending = true; 790 if (!direct) 791 r = stimer_send_msg(stimer); 792 else 793 r = stimer_notify_direct(stimer); 794 trace_kvm_hv_stimer_expiration(hv_stimer_to_vcpu(stimer)->vcpu_id, 795 stimer->index, direct, r); 796 if (!r) { 797 stimer->msg_pending = false; 798 if (!(stimer->config.periodic)) 799 stimer->config.enable = 0; 800 } 801 } 802 803 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu) 804 { 805 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu); 806 struct kvm_vcpu_hv_stimer *stimer; 807 u64 time_now, exp_time; 808 int i; 809 810 if (!hv_vcpu) 811 return; 812 813 for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++) 814 if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) { 815 stimer = &hv_vcpu->stimer[i]; 816 if (stimer->config.enable) { 817 exp_time = stimer->exp_time; 818 819 if (exp_time) { 820 time_now = 821 get_time_ref_counter(vcpu->kvm); 822 if (time_now >= exp_time) 823 stimer_expiration(stimer); 824 } 825 826 if ((stimer->config.enable) && 827 stimer->count) { 828 if (!stimer->msg_pending) 829 stimer_start(stimer); 830 } else 831 stimer_cleanup(stimer); 832 } 833 } 834 } 835 836 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu) 837 { 838 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu); 839 int i; 840 841 if (!hv_vcpu) 842 return; 843 844 for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++) 845 stimer_cleanup(&hv_vcpu->stimer[i]); 846 847 kfree(hv_vcpu); 848 vcpu->arch.hyperv = NULL; 849 } 850 851 bool kvm_hv_assist_page_enabled(struct kvm_vcpu *vcpu) 852 { 853 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu); 854 855 if (!hv_vcpu) 856 return false; 857 858 if (!(hv_vcpu->hv_vapic & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE)) 859 return false; 860 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED; 861 } 862 EXPORT_SYMBOL_GPL(kvm_hv_assist_page_enabled); 863 864 bool kvm_hv_get_assist_page(struct kvm_vcpu *vcpu, 865 struct hv_vp_assist_page *assist_page) 866 { 867 if (!kvm_hv_assist_page_enabled(vcpu)) 868 return false; 869 return !kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, 870 assist_page, sizeof(*assist_page)); 871 } 872 EXPORT_SYMBOL_GPL(kvm_hv_get_assist_page); 873 874 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer) 875 { 876 struct hv_message *msg = &stimer->msg; 877 struct hv_timer_message_payload *payload = 878 (struct hv_timer_message_payload *)&msg->u.payload; 879 880 memset(&msg->header, 0, sizeof(msg->header)); 881 msg->header.message_type = HVMSG_TIMER_EXPIRED; 882 msg->header.payload_size = sizeof(*payload); 883 884 payload->timer_index = stimer->index; 885 payload->expiration_time = 0; 886 payload->delivery_time = 0; 887 } 888 889 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index) 890 { 891 memset(stimer, 0, sizeof(*stimer)); 892 stimer->index = timer_index; 893 hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); 894 stimer->timer.function = stimer_timer_callback; 895 stimer_prepare_msg(stimer); 896 } 897 898 static int kvm_hv_vcpu_init(struct kvm_vcpu *vcpu) 899 { 900 struct kvm_vcpu_hv *hv_vcpu; 901 int i; 902 903 hv_vcpu = kzalloc(sizeof(struct kvm_vcpu_hv), GFP_KERNEL_ACCOUNT); 904 if (!hv_vcpu) 905 return -ENOMEM; 906 907 vcpu->arch.hyperv = hv_vcpu; 908 hv_vcpu->vcpu = vcpu; 909 910 synic_init(&hv_vcpu->synic); 911 912 bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT); 913 for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++) 914 stimer_init(&hv_vcpu->stimer[i], i); 915 916 hv_vcpu->vp_index = kvm_vcpu_get_idx(vcpu); 917 918 return 0; 919 } 920 921 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages) 922 { 923 struct kvm_vcpu_hv_synic *synic; 924 int r; 925 926 if (!to_hv_vcpu(vcpu)) { 927 r = kvm_hv_vcpu_init(vcpu); 928 if (r) 929 return r; 930 } 931 932 synic = to_hv_synic(vcpu); 933 934 /* 935 * Hyper-V SynIC auto EOI SINT's are 936 * not compatible with APICV, so request 937 * to deactivate APICV permanently. 938 */ 939 kvm_request_apicv_update(vcpu->kvm, false, APICV_INHIBIT_REASON_HYPERV); 940 synic->active = true; 941 synic->dont_zero_synic_pages = dont_zero_synic_pages; 942 synic->control = HV_SYNIC_CONTROL_ENABLE; 943 return 0; 944 } 945 946 static bool kvm_hv_msr_partition_wide(u32 msr) 947 { 948 bool r = false; 949 950 switch (msr) { 951 case HV_X64_MSR_GUEST_OS_ID: 952 case HV_X64_MSR_HYPERCALL: 953 case HV_X64_MSR_REFERENCE_TSC: 954 case HV_X64_MSR_TIME_REF_COUNT: 955 case HV_X64_MSR_CRASH_CTL: 956 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4: 957 case HV_X64_MSR_RESET: 958 case HV_X64_MSR_REENLIGHTENMENT_CONTROL: 959 case HV_X64_MSR_TSC_EMULATION_CONTROL: 960 case HV_X64_MSR_TSC_EMULATION_STATUS: 961 case HV_X64_MSR_SYNDBG_OPTIONS: 962 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER: 963 r = true; 964 break; 965 } 966 967 return r; 968 } 969 970 static int kvm_hv_msr_get_crash_data(struct kvm *kvm, u32 index, u64 *pdata) 971 { 972 struct kvm_hv *hv = to_kvm_hv(kvm); 973 size_t size = ARRAY_SIZE(hv->hv_crash_param); 974 975 if (WARN_ON_ONCE(index >= size)) 976 return -EINVAL; 977 978 *pdata = hv->hv_crash_param[array_index_nospec(index, size)]; 979 return 0; 980 } 981 982 static int kvm_hv_msr_get_crash_ctl(struct kvm *kvm, u64 *pdata) 983 { 984 struct kvm_hv *hv = to_kvm_hv(kvm); 985 986 *pdata = hv->hv_crash_ctl; 987 return 0; 988 } 989 990 static int kvm_hv_msr_set_crash_ctl(struct kvm *kvm, u64 data) 991 { 992 struct kvm_hv *hv = to_kvm_hv(kvm); 993 994 hv->hv_crash_ctl = data & HV_CRASH_CTL_CRASH_NOTIFY; 995 996 return 0; 997 } 998 999 static int kvm_hv_msr_set_crash_data(struct kvm *kvm, u32 index, u64 data) 1000 { 1001 struct kvm_hv *hv = to_kvm_hv(kvm); 1002 size_t size = ARRAY_SIZE(hv->hv_crash_param); 1003 1004 if (WARN_ON_ONCE(index >= size)) 1005 return -EINVAL; 1006 1007 hv->hv_crash_param[array_index_nospec(index, size)] = data; 1008 return 0; 1009 } 1010 1011 /* 1012 * The kvmclock and Hyper-V TSC page use similar formulas, and converting 1013 * between them is possible: 1014 * 1015 * kvmclock formula: 1016 * nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32) 1017 * + system_time 1018 * 1019 * Hyper-V formula: 1020 * nsec/100 = ticks * scale / 2^64 + offset 1021 * 1022 * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula. 1023 * By dividing the kvmclock formula by 100 and equating what's left we get: 1024 * ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100 1025 * scale / 2^64 = tsc_to_system_mul * 2^(tsc_shift-32) / 100 1026 * scale = tsc_to_system_mul * 2^(32+tsc_shift) / 100 1027 * 1028 * Now expand the kvmclock formula and divide by 100: 1029 * nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32) 1030 * - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) 1031 * + system_time 1032 * nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100 1033 * - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100 1034 * + system_time / 100 1035 * 1036 * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64: 1037 * nsec/100 = ticks * scale / 2^64 1038 * - tsc_timestamp * scale / 2^64 1039 * + system_time / 100 1040 * 1041 * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out: 1042 * offset = system_time / 100 - tsc_timestamp * scale / 2^64 1043 * 1044 * These two equivalencies are implemented in this function. 1045 */ 1046 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock, 1047 struct ms_hyperv_tsc_page *tsc_ref) 1048 { 1049 u64 max_mul; 1050 1051 if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT)) 1052 return false; 1053 1054 /* 1055 * check if scale would overflow, if so we use the time ref counter 1056 * tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64 1057 * tsc_to_system_mul / 100 >= 2^(32-tsc_shift) 1058 * tsc_to_system_mul >= 100 * 2^(32-tsc_shift) 1059 */ 1060 max_mul = 100ull << (32 - hv_clock->tsc_shift); 1061 if (hv_clock->tsc_to_system_mul >= max_mul) 1062 return false; 1063 1064 /* 1065 * Otherwise compute the scale and offset according to the formulas 1066 * derived above. 1067 */ 1068 tsc_ref->tsc_scale = 1069 mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift), 1070 hv_clock->tsc_to_system_mul, 1071 100); 1072 1073 tsc_ref->tsc_offset = hv_clock->system_time; 1074 do_div(tsc_ref->tsc_offset, 100); 1075 tsc_ref->tsc_offset -= 1076 mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64); 1077 return true; 1078 } 1079 1080 /* 1081 * Don't touch TSC page values if the guest has opted for TSC emulation after 1082 * migration. KVM doesn't fully support reenlightenment notifications and TSC 1083 * access emulation and Hyper-V is known to expect the values in TSC page to 1084 * stay constant before TSC access emulation is disabled from guest side 1085 * (HV_X64_MSR_TSC_EMULATION_STATUS). KVM userspace is expected to preserve TSC 1086 * frequency and guest visible TSC value across migration (and prevent it when 1087 * TSC scaling is unsupported). 1088 */ 1089 static inline bool tsc_page_update_unsafe(struct kvm_hv *hv) 1090 { 1091 return (hv->hv_tsc_page_status != HV_TSC_PAGE_GUEST_CHANGED) && 1092 hv->hv_tsc_emulation_control; 1093 } 1094 1095 void kvm_hv_setup_tsc_page(struct kvm *kvm, 1096 struct pvclock_vcpu_time_info *hv_clock) 1097 { 1098 struct kvm_hv *hv = to_kvm_hv(kvm); 1099 u32 tsc_seq; 1100 u64 gfn; 1101 1102 BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence)); 1103 BUILD_BUG_ON(offsetof(struct ms_hyperv_tsc_page, tsc_sequence) != 0); 1104 1105 if (hv->hv_tsc_page_status == HV_TSC_PAGE_BROKEN || 1106 hv->hv_tsc_page_status == HV_TSC_PAGE_UNSET) 1107 return; 1108 1109 mutex_lock(&hv->hv_lock); 1110 if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE)) 1111 goto out_unlock; 1112 1113 gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT; 1114 /* 1115 * Because the TSC parameters only vary when there is a 1116 * change in the master clock, do not bother with caching. 1117 */ 1118 if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn), 1119 &tsc_seq, sizeof(tsc_seq)))) 1120 goto out_err; 1121 1122 if (tsc_seq && tsc_page_update_unsafe(hv)) { 1123 if (kvm_read_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref))) 1124 goto out_err; 1125 1126 hv->hv_tsc_page_status = HV_TSC_PAGE_SET; 1127 goto out_unlock; 1128 } 1129 1130 /* 1131 * While we're computing and writing the parameters, force the 1132 * guest to use the time reference count MSR. 1133 */ 1134 hv->tsc_ref.tsc_sequence = 0; 1135 if (kvm_write_guest(kvm, gfn_to_gpa(gfn), 1136 &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence))) 1137 goto out_err; 1138 1139 if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref)) 1140 goto out_err; 1141 1142 /* Ensure sequence is zero before writing the rest of the struct. */ 1143 smp_wmb(); 1144 if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref))) 1145 goto out_err; 1146 1147 /* 1148 * Now switch to the TSC page mechanism by writing the sequence. 1149 */ 1150 tsc_seq++; 1151 if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0) 1152 tsc_seq = 1; 1153 1154 /* Write the struct entirely before the non-zero sequence. */ 1155 smp_wmb(); 1156 1157 hv->tsc_ref.tsc_sequence = tsc_seq; 1158 if (kvm_write_guest(kvm, gfn_to_gpa(gfn), 1159 &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence))) 1160 goto out_err; 1161 1162 hv->hv_tsc_page_status = HV_TSC_PAGE_SET; 1163 goto out_unlock; 1164 1165 out_err: 1166 hv->hv_tsc_page_status = HV_TSC_PAGE_BROKEN; 1167 out_unlock: 1168 mutex_unlock(&hv->hv_lock); 1169 } 1170 1171 void kvm_hv_invalidate_tsc_page(struct kvm *kvm) 1172 { 1173 struct kvm_hv *hv = to_kvm_hv(kvm); 1174 u64 gfn; 1175 1176 if (hv->hv_tsc_page_status == HV_TSC_PAGE_BROKEN || 1177 hv->hv_tsc_page_status == HV_TSC_PAGE_UNSET || 1178 tsc_page_update_unsafe(hv)) 1179 return; 1180 1181 mutex_lock(&hv->hv_lock); 1182 1183 if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE)) 1184 goto out_unlock; 1185 1186 /* Preserve HV_TSC_PAGE_GUEST_CHANGED/HV_TSC_PAGE_HOST_CHANGED states */ 1187 if (hv->hv_tsc_page_status == HV_TSC_PAGE_SET) 1188 hv->hv_tsc_page_status = HV_TSC_PAGE_UPDATING; 1189 1190 gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT; 1191 1192 hv->tsc_ref.tsc_sequence = 0; 1193 if (kvm_write_guest(kvm, gfn_to_gpa(gfn), 1194 &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence))) 1195 hv->hv_tsc_page_status = HV_TSC_PAGE_BROKEN; 1196 1197 out_unlock: 1198 mutex_unlock(&hv->hv_lock); 1199 } 1200 1201 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data, 1202 bool host) 1203 { 1204 struct kvm *kvm = vcpu->kvm; 1205 struct kvm_hv *hv = to_kvm_hv(kvm); 1206 1207 switch (msr) { 1208 case HV_X64_MSR_GUEST_OS_ID: 1209 hv->hv_guest_os_id = data; 1210 /* setting guest os id to zero disables hypercall page */ 1211 if (!hv->hv_guest_os_id) 1212 hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE; 1213 break; 1214 case HV_X64_MSR_HYPERCALL: { 1215 u8 instructions[9]; 1216 int i = 0; 1217 u64 addr; 1218 1219 /* if guest os id is not set hypercall should remain disabled */ 1220 if (!hv->hv_guest_os_id) 1221 break; 1222 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) { 1223 hv->hv_hypercall = data; 1224 break; 1225 } 1226 1227 /* 1228 * If Xen and Hyper-V hypercalls are both enabled, disambiguate 1229 * the same way Xen itself does, by setting the bit 31 of EAX 1230 * which is RsvdZ in the 32-bit Hyper-V hypercall ABI and just 1231 * going to be clobbered on 64-bit. 1232 */ 1233 if (kvm_xen_hypercall_enabled(kvm)) { 1234 /* orl $0x80000000, %eax */ 1235 instructions[i++] = 0x0d; 1236 instructions[i++] = 0x00; 1237 instructions[i++] = 0x00; 1238 instructions[i++] = 0x00; 1239 instructions[i++] = 0x80; 1240 } 1241 1242 /* vmcall/vmmcall */ 1243 static_call(kvm_x86_patch_hypercall)(vcpu, instructions + i); 1244 i += 3; 1245 1246 /* ret */ 1247 ((unsigned char *)instructions)[i++] = 0xc3; 1248 1249 addr = data & HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_MASK; 1250 if (kvm_vcpu_write_guest(vcpu, addr, instructions, i)) 1251 return 1; 1252 hv->hv_hypercall = data; 1253 break; 1254 } 1255 case HV_X64_MSR_REFERENCE_TSC: 1256 hv->hv_tsc_page = data; 1257 if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE) { 1258 if (!host) 1259 hv->hv_tsc_page_status = HV_TSC_PAGE_GUEST_CHANGED; 1260 else 1261 hv->hv_tsc_page_status = HV_TSC_PAGE_HOST_CHANGED; 1262 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); 1263 } else { 1264 hv->hv_tsc_page_status = HV_TSC_PAGE_UNSET; 1265 } 1266 break; 1267 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4: 1268 return kvm_hv_msr_set_crash_data(kvm, 1269 msr - HV_X64_MSR_CRASH_P0, 1270 data); 1271 case HV_X64_MSR_CRASH_CTL: 1272 if (host) 1273 return kvm_hv_msr_set_crash_ctl(kvm, data); 1274 1275 if (data & HV_CRASH_CTL_CRASH_NOTIFY) { 1276 vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n", 1277 hv->hv_crash_param[0], 1278 hv->hv_crash_param[1], 1279 hv->hv_crash_param[2], 1280 hv->hv_crash_param[3], 1281 hv->hv_crash_param[4]); 1282 1283 /* Send notification about crash to user space */ 1284 kvm_make_request(KVM_REQ_HV_CRASH, vcpu); 1285 } 1286 break; 1287 case HV_X64_MSR_RESET: 1288 if (data == 1) { 1289 vcpu_debug(vcpu, "hyper-v reset requested\n"); 1290 kvm_make_request(KVM_REQ_HV_RESET, vcpu); 1291 } 1292 break; 1293 case HV_X64_MSR_REENLIGHTENMENT_CONTROL: 1294 hv->hv_reenlightenment_control = data; 1295 break; 1296 case HV_X64_MSR_TSC_EMULATION_CONTROL: 1297 hv->hv_tsc_emulation_control = data; 1298 break; 1299 case HV_X64_MSR_TSC_EMULATION_STATUS: 1300 if (data && !host) 1301 return 1; 1302 1303 hv->hv_tsc_emulation_status = data; 1304 break; 1305 case HV_X64_MSR_TIME_REF_COUNT: 1306 /* read-only, but still ignore it if host-initiated */ 1307 if (!host) 1308 return 1; 1309 break; 1310 case HV_X64_MSR_SYNDBG_OPTIONS: 1311 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER: 1312 return syndbg_set_msr(vcpu, msr, data, host); 1313 default: 1314 vcpu_unimpl(vcpu, "Hyper-V unhandled wrmsr: 0x%x data 0x%llx\n", 1315 msr, data); 1316 return 1; 1317 } 1318 return 0; 1319 } 1320 1321 /* Calculate cpu time spent by current task in 100ns units */ 1322 static u64 current_task_runtime_100ns(void) 1323 { 1324 u64 utime, stime; 1325 1326 task_cputime_adjusted(current, &utime, &stime); 1327 1328 return div_u64(utime + stime, 100); 1329 } 1330 1331 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host) 1332 { 1333 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu); 1334 1335 switch (msr) { 1336 case HV_X64_MSR_VP_INDEX: { 1337 struct kvm_hv *hv = to_kvm_hv(vcpu->kvm); 1338 int vcpu_idx = kvm_vcpu_get_idx(vcpu); 1339 u32 new_vp_index = (u32)data; 1340 1341 if (!host || new_vp_index >= KVM_MAX_VCPUS) 1342 return 1; 1343 1344 if (new_vp_index == hv_vcpu->vp_index) 1345 return 0; 1346 1347 /* 1348 * The VP index is initialized to vcpu_index by 1349 * kvm_hv_vcpu_postcreate so they initially match. Now the 1350 * VP index is changing, adjust num_mismatched_vp_indexes if 1351 * it now matches or no longer matches vcpu_idx. 1352 */ 1353 if (hv_vcpu->vp_index == vcpu_idx) 1354 atomic_inc(&hv->num_mismatched_vp_indexes); 1355 else if (new_vp_index == vcpu_idx) 1356 atomic_dec(&hv->num_mismatched_vp_indexes); 1357 1358 hv_vcpu->vp_index = new_vp_index; 1359 break; 1360 } 1361 case HV_X64_MSR_VP_ASSIST_PAGE: { 1362 u64 gfn; 1363 unsigned long addr; 1364 1365 if (!(data & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE)) { 1366 hv_vcpu->hv_vapic = data; 1367 if (kvm_lapic_enable_pv_eoi(vcpu, 0, 0)) 1368 return 1; 1369 break; 1370 } 1371 gfn = data >> HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT; 1372 addr = kvm_vcpu_gfn_to_hva(vcpu, gfn); 1373 if (kvm_is_error_hva(addr)) 1374 return 1; 1375 1376 /* 1377 * Clear apic_assist portion of struct hv_vp_assist_page 1378 * only, there can be valuable data in the rest which needs 1379 * to be preserved e.g. on migration. 1380 */ 1381 if (__put_user(0, (u32 __user *)addr)) 1382 return 1; 1383 hv_vcpu->hv_vapic = data; 1384 kvm_vcpu_mark_page_dirty(vcpu, gfn); 1385 if (kvm_lapic_enable_pv_eoi(vcpu, 1386 gfn_to_gpa(gfn) | KVM_MSR_ENABLED, 1387 sizeof(struct hv_vp_assist_page))) 1388 return 1; 1389 break; 1390 } 1391 case HV_X64_MSR_EOI: 1392 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data); 1393 case HV_X64_MSR_ICR: 1394 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data); 1395 case HV_X64_MSR_TPR: 1396 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data); 1397 case HV_X64_MSR_VP_RUNTIME: 1398 if (!host) 1399 return 1; 1400 hv_vcpu->runtime_offset = data - current_task_runtime_100ns(); 1401 break; 1402 case HV_X64_MSR_SCONTROL: 1403 case HV_X64_MSR_SVERSION: 1404 case HV_X64_MSR_SIEFP: 1405 case HV_X64_MSR_SIMP: 1406 case HV_X64_MSR_EOM: 1407 case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15: 1408 return synic_set_msr(to_hv_synic(vcpu), msr, data, host); 1409 case HV_X64_MSR_STIMER0_CONFIG: 1410 case HV_X64_MSR_STIMER1_CONFIG: 1411 case HV_X64_MSR_STIMER2_CONFIG: 1412 case HV_X64_MSR_STIMER3_CONFIG: { 1413 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2; 1414 1415 return stimer_set_config(to_hv_stimer(vcpu, timer_index), 1416 data, host); 1417 } 1418 case HV_X64_MSR_STIMER0_COUNT: 1419 case HV_X64_MSR_STIMER1_COUNT: 1420 case HV_X64_MSR_STIMER2_COUNT: 1421 case HV_X64_MSR_STIMER3_COUNT: { 1422 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2; 1423 1424 return stimer_set_count(to_hv_stimer(vcpu, timer_index), 1425 data, host); 1426 } 1427 case HV_X64_MSR_TSC_FREQUENCY: 1428 case HV_X64_MSR_APIC_FREQUENCY: 1429 /* read-only, but still ignore it if host-initiated */ 1430 if (!host) 1431 return 1; 1432 break; 1433 default: 1434 vcpu_unimpl(vcpu, "Hyper-V unhandled wrmsr: 0x%x data 0x%llx\n", 1435 msr, data); 1436 return 1; 1437 } 1438 1439 return 0; 1440 } 1441 1442 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, 1443 bool host) 1444 { 1445 u64 data = 0; 1446 struct kvm *kvm = vcpu->kvm; 1447 struct kvm_hv *hv = to_kvm_hv(kvm); 1448 1449 switch (msr) { 1450 case HV_X64_MSR_GUEST_OS_ID: 1451 data = hv->hv_guest_os_id; 1452 break; 1453 case HV_X64_MSR_HYPERCALL: 1454 data = hv->hv_hypercall; 1455 break; 1456 case HV_X64_MSR_TIME_REF_COUNT: 1457 data = get_time_ref_counter(kvm); 1458 break; 1459 case HV_X64_MSR_REFERENCE_TSC: 1460 data = hv->hv_tsc_page; 1461 break; 1462 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4: 1463 return kvm_hv_msr_get_crash_data(kvm, 1464 msr - HV_X64_MSR_CRASH_P0, 1465 pdata); 1466 case HV_X64_MSR_CRASH_CTL: 1467 return kvm_hv_msr_get_crash_ctl(kvm, pdata); 1468 case HV_X64_MSR_RESET: 1469 data = 0; 1470 break; 1471 case HV_X64_MSR_REENLIGHTENMENT_CONTROL: 1472 data = hv->hv_reenlightenment_control; 1473 break; 1474 case HV_X64_MSR_TSC_EMULATION_CONTROL: 1475 data = hv->hv_tsc_emulation_control; 1476 break; 1477 case HV_X64_MSR_TSC_EMULATION_STATUS: 1478 data = hv->hv_tsc_emulation_status; 1479 break; 1480 case HV_X64_MSR_SYNDBG_OPTIONS: 1481 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER: 1482 return syndbg_get_msr(vcpu, msr, pdata, host); 1483 default: 1484 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr); 1485 return 1; 1486 } 1487 1488 *pdata = data; 1489 return 0; 1490 } 1491 1492 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, 1493 bool host) 1494 { 1495 u64 data = 0; 1496 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu); 1497 1498 switch (msr) { 1499 case HV_X64_MSR_VP_INDEX: 1500 data = hv_vcpu->vp_index; 1501 break; 1502 case HV_X64_MSR_EOI: 1503 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata); 1504 case HV_X64_MSR_ICR: 1505 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata); 1506 case HV_X64_MSR_TPR: 1507 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata); 1508 case HV_X64_MSR_VP_ASSIST_PAGE: 1509 data = hv_vcpu->hv_vapic; 1510 break; 1511 case HV_X64_MSR_VP_RUNTIME: 1512 data = current_task_runtime_100ns() + hv_vcpu->runtime_offset; 1513 break; 1514 case HV_X64_MSR_SCONTROL: 1515 case HV_X64_MSR_SVERSION: 1516 case HV_X64_MSR_SIEFP: 1517 case HV_X64_MSR_SIMP: 1518 case HV_X64_MSR_EOM: 1519 case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15: 1520 return synic_get_msr(to_hv_synic(vcpu), msr, pdata, host); 1521 case HV_X64_MSR_STIMER0_CONFIG: 1522 case HV_X64_MSR_STIMER1_CONFIG: 1523 case HV_X64_MSR_STIMER2_CONFIG: 1524 case HV_X64_MSR_STIMER3_CONFIG: { 1525 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2; 1526 1527 return stimer_get_config(to_hv_stimer(vcpu, timer_index), 1528 pdata); 1529 } 1530 case HV_X64_MSR_STIMER0_COUNT: 1531 case HV_X64_MSR_STIMER1_COUNT: 1532 case HV_X64_MSR_STIMER2_COUNT: 1533 case HV_X64_MSR_STIMER3_COUNT: { 1534 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2; 1535 1536 return stimer_get_count(to_hv_stimer(vcpu, timer_index), 1537 pdata); 1538 } 1539 case HV_X64_MSR_TSC_FREQUENCY: 1540 data = (u64)vcpu->arch.virtual_tsc_khz * 1000; 1541 break; 1542 case HV_X64_MSR_APIC_FREQUENCY: 1543 data = APIC_BUS_FREQUENCY; 1544 break; 1545 default: 1546 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr); 1547 return 1; 1548 } 1549 *pdata = data; 1550 return 0; 1551 } 1552 1553 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host) 1554 { 1555 struct kvm_hv *hv = to_kvm_hv(vcpu->kvm); 1556 1557 if (!host && !vcpu->arch.hyperv_enabled) 1558 return 1; 1559 1560 if (!to_hv_vcpu(vcpu)) { 1561 if (kvm_hv_vcpu_init(vcpu)) 1562 return 1; 1563 } 1564 1565 if (kvm_hv_msr_partition_wide(msr)) { 1566 int r; 1567 1568 mutex_lock(&hv->hv_lock); 1569 r = kvm_hv_set_msr_pw(vcpu, msr, data, host); 1570 mutex_unlock(&hv->hv_lock); 1571 return r; 1572 } else 1573 return kvm_hv_set_msr(vcpu, msr, data, host); 1574 } 1575 1576 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host) 1577 { 1578 struct kvm_hv *hv = to_kvm_hv(vcpu->kvm); 1579 1580 if (!host && !vcpu->arch.hyperv_enabled) 1581 return 1; 1582 1583 if (!to_hv_vcpu(vcpu)) { 1584 if (kvm_hv_vcpu_init(vcpu)) 1585 return 1; 1586 } 1587 1588 if (kvm_hv_msr_partition_wide(msr)) { 1589 int r; 1590 1591 mutex_lock(&hv->hv_lock); 1592 r = kvm_hv_get_msr_pw(vcpu, msr, pdata, host); 1593 mutex_unlock(&hv->hv_lock); 1594 return r; 1595 } else 1596 return kvm_hv_get_msr(vcpu, msr, pdata, host); 1597 } 1598 1599 static __always_inline unsigned long *sparse_set_to_vcpu_mask( 1600 struct kvm *kvm, u64 *sparse_banks, u64 valid_bank_mask, 1601 u64 *vp_bitmap, unsigned long *vcpu_bitmap) 1602 { 1603 struct kvm_hv *hv = to_kvm_hv(kvm); 1604 struct kvm_vcpu *vcpu; 1605 int i, bank, sbank = 0; 1606 1607 memset(vp_bitmap, 0, 1608 KVM_HV_MAX_SPARSE_VCPU_SET_BITS * sizeof(*vp_bitmap)); 1609 for_each_set_bit(bank, (unsigned long *)&valid_bank_mask, 1610 KVM_HV_MAX_SPARSE_VCPU_SET_BITS) 1611 vp_bitmap[bank] = sparse_banks[sbank++]; 1612 1613 if (likely(!atomic_read(&hv->num_mismatched_vp_indexes))) { 1614 /* for all vcpus vp_index == vcpu_idx */ 1615 return (unsigned long *)vp_bitmap; 1616 } 1617 1618 bitmap_zero(vcpu_bitmap, KVM_MAX_VCPUS); 1619 kvm_for_each_vcpu(i, vcpu, kvm) { 1620 if (test_bit(kvm_hv_get_vpindex(vcpu), (unsigned long *)vp_bitmap)) 1621 __set_bit(i, vcpu_bitmap); 1622 } 1623 return vcpu_bitmap; 1624 } 1625 1626 static u64 kvm_hv_flush_tlb(struct kvm_vcpu *vcpu, u64 ingpa, u16 rep_cnt, bool ex) 1627 { 1628 struct kvm *kvm = vcpu->kvm; 1629 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu); 1630 struct hv_tlb_flush_ex flush_ex; 1631 struct hv_tlb_flush flush; 1632 u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS]; 1633 DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS); 1634 unsigned long *vcpu_mask; 1635 u64 valid_bank_mask; 1636 u64 sparse_banks[64]; 1637 int sparse_banks_len; 1638 bool all_cpus; 1639 1640 if (!ex) { 1641 if (unlikely(kvm_read_guest(kvm, ingpa, &flush, sizeof(flush)))) 1642 return HV_STATUS_INVALID_HYPERCALL_INPUT; 1643 1644 trace_kvm_hv_flush_tlb(flush.processor_mask, 1645 flush.address_space, flush.flags); 1646 1647 valid_bank_mask = BIT_ULL(0); 1648 sparse_banks[0] = flush.processor_mask; 1649 1650 /* 1651 * Work around possible WS2012 bug: it sends hypercalls 1652 * with processor_mask = 0x0 and HV_FLUSH_ALL_PROCESSORS clear, 1653 * while also expecting us to flush something and crashing if 1654 * we don't. Let's treat processor_mask == 0 same as 1655 * HV_FLUSH_ALL_PROCESSORS. 1656 */ 1657 all_cpus = (flush.flags & HV_FLUSH_ALL_PROCESSORS) || 1658 flush.processor_mask == 0; 1659 } else { 1660 if (unlikely(kvm_read_guest(kvm, ingpa, &flush_ex, 1661 sizeof(flush_ex)))) 1662 return HV_STATUS_INVALID_HYPERCALL_INPUT; 1663 1664 trace_kvm_hv_flush_tlb_ex(flush_ex.hv_vp_set.valid_bank_mask, 1665 flush_ex.hv_vp_set.format, 1666 flush_ex.address_space, 1667 flush_ex.flags); 1668 1669 valid_bank_mask = flush_ex.hv_vp_set.valid_bank_mask; 1670 all_cpus = flush_ex.hv_vp_set.format != 1671 HV_GENERIC_SET_SPARSE_4K; 1672 1673 sparse_banks_len = 1674 bitmap_weight((unsigned long *)&valid_bank_mask, 64) * 1675 sizeof(sparse_banks[0]); 1676 1677 if (!sparse_banks_len && !all_cpus) 1678 goto ret_success; 1679 1680 if (!all_cpus && 1681 kvm_read_guest(kvm, 1682 ingpa + offsetof(struct hv_tlb_flush_ex, 1683 hv_vp_set.bank_contents), 1684 sparse_banks, 1685 sparse_banks_len)) 1686 return HV_STATUS_INVALID_HYPERCALL_INPUT; 1687 } 1688 1689 cpumask_clear(&hv_vcpu->tlb_flush); 1690 1691 vcpu_mask = all_cpus ? NULL : 1692 sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask, 1693 vp_bitmap, vcpu_bitmap); 1694 1695 /* 1696 * vcpu->arch.cr3 may not be up-to-date for running vCPUs so we can't 1697 * analyze it here, flush TLB regardless of the specified address space. 1698 */ 1699 kvm_make_vcpus_request_mask(kvm, KVM_REQ_HV_TLB_FLUSH, 1700 NULL, vcpu_mask, &hv_vcpu->tlb_flush); 1701 1702 ret_success: 1703 /* We always do full TLB flush, set rep_done = rep_cnt. */ 1704 return (u64)HV_STATUS_SUCCESS | 1705 ((u64)rep_cnt << HV_HYPERCALL_REP_COMP_OFFSET); 1706 } 1707 1708 static void kvm_send_ipi_to_many(struct kvm *kvm, u32 vector, 1709 unsigned long *vcpu_bitmap) 1710 { 1711 struct kvm_lapic_irq irq = { 1712 .delivery_mode = APIC_DM_FIXED, 1713 .vector = vector 1714 }; 1715 struct kvm_vcpu *vcpu; 1716 int i; 1717 1718 kvm_for_each_vcpu(i, vcpu, kvm) { 1719 if (vcpu_bitmap && !test_bit(i, vcpu_bitmap)) 1720 continue; 1721 1722 /* We fail only when APIC is disabled */ 1723 kvm_apic_set_irq(vcpu, &irq, NULL); 1724 } 1725 } 1726 1727 static u64 kvm_hv_send_ipi(struct kvm_vcpu *vcpu, u64 ingpa, u64 outgpa, 1728 bool ex, bool fast) 1729 { 1730 struct kvm *kvm = vcpu->kvm; 1731 struct hv_send_ipi_ex send_ipi_ex; 1732 struct hv_send_ipi send_ipi; 1733 u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS]; 1734 DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS); 1735 unsigned long *vcpu_mask; 1736 unsigned long valid_bank_mask; 1737 u64 sparse_banks[64]; 1738 int sparse_banks_len; 1739 u32 vector; 1740 bool all_cpus; 1741 1742 if (!ex) { 1743 if (!fast) { 1744 if (unlikely(kvm_read_guest(kvm, ingpa, &send_ipi, 1745 sizeof(send_ipi)))) 1746 return HV_STATUS_INVALID_HYPERCALL_INPUT; 1747 sparse_banks[0] = send_ipi.cpu_mask; 1748 vector = send_ipi.vector; 1749 } else { 1750 /* 'reserved' part of hv_send_ipi should be 0 */ 1751 if (unlikely(ingpa >> 32 != 0)) 1752 return HV_STATUS_INVALID_HYPERCALL_INPUT; 1753 sparse_banks[0] = outgpa; 1754 vector = (u32)ingpa; 1755 } 1756 all_cpus = false; 1757 valid_bank_mask = BIT_ULL(0); 1758 1759 trace_kvm_hv_send_ipi(vector, sparse_banks[0]); 1760 } else { 1761 if (unlikely(kvm_read_guest(kvm, ingpa, &send_ipi_ex, 1762 sizeof(send_ipi_ex)))) 1763 return HV_STATUS_INVALID_HYPERCALL_INPUT; 1764 1765 trace_kvm_hv_send_ipi_ex(send_ipi_ex.vector, 1766 send_ipi_ex.vp_set.format, 1767 send_ipi_ex.vp_set.valid_bank_mask); 1768 1769 vector = send_ipi_ex.vector; 1770 valid_bank_mask = send_ipi_ex.vp_set.valid_bank_mask; 1771 sparse_banks_len = bitmap_weight(&valid_bank_mask, 64) * 1772 sizeof(sparse_banks[0]); 1773 1774 all_cpus = send_ipi_ex.vp_set.format == HV_GENERIC_SET_ALL; 1775 1776 if (!sparse_banks_len) 1777 goto ret_success; 1778 1779 if (!all_cpus && 1780 kvm_read_guest(kvm, 1781 ingpa + offsetof(struct hv_send_ipi_ex, 1782 vp_set.bank_contents), 1783 sparse_banks, 1784 sparse_banks_len)) 1785 return HV_STATUS_INVALID_HYPERCALL_INPUT; 1786 } 1787 1788 if ((vector < HV_IPI_LOW_VECTOR) || (vector > HV_IPI_HIGH_VECTOR)) 1789 return HV_STATUS_INVALID_HYPERCALL_INPUT; 1790 1791 vcpu_mask = all_cpus ? NULL : 1792 sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask, 1793 vp_bitmap, vcpu_bitmap); 1794 1795 kvm_send_ipi_to_many(kvm, vector, vcpu_mask); 1796 1797 ret_success: 1798 return HV_STATUS_SUCCESS; 1799 } 1800 1801 void kvm_hv_set_cpuid(struct kvm_vcpu *vcpu) 1802 { 1803 struct kvm_cpuid_entry2 *entry; 1804 1805 entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_INTERFACE, 0); 1806 if (entry && entry->eax == HYPERV_CPUID_SIGNATURE_EAX) 1807 vcpu->arch.hyperv_enabled = true; 1808 else 1809 vcpu->arch.hyperv_enabled = false; 1810 } 1811 1812 bool kvm_hv_hypercall_enabled(struct kvm_vcpu *vcpu) 1813 { 1814 return vcpu->arch.hyperv_enabled && to_kvm_hv(vcpu->kvm)->hv_guest_os_id; 1815 } 1816 1817 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result) 1818 { 1819 bool longmode; 1820 1821 longmode = is_64_bit_mode(vcpu); 1822 if (longmode) 1823 kvm_rax_write(vcpu, result); 1824 else { 1825 kvm_rdx_write(vcpu, result >> 32); 1826 kvm_rax_write(vcpu, result & 0xffffffff); 1827 } 1828 } 1829 1830 static int kvm_hv_hypercall_complete(struct kvm_vcpu *vcpu, u64 result) 1831 { 1832 kvm_hv_hypercall_set_result(vcpu, result); 1833 ++vcpu->stat.hypercalls; 1834 return kvm_skip_emulated_instruction(vcpu); 1835 } 1836 1837 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu) 1838 { 1839 return kvm_hv_hypercall_complete(vcpu, vcpu->run->hyperv.u.hcall.result); 1840 } 1841 1842 static u16 kvm_hvcall_signal_event(struct kvm_vcpu *vcpu, bool fast, u64 param) 1843 { 1844 struct kvm_hv *hv = to_kvm_hv(vcpu->kvm); 1845 struct eventfd_ctx *eventfd; 1846 1847 if (unlikely(!fast)) { 1848 int ret; 1849 gpa_t gpa = param; 1850 1851 if ((gpa & (__alignof__(param) - 1)) || 1852 offset_in_page(gpa) + sizeof(param) > PAGE_SIZE) 1853 return HV_STATUS_INVALID_ALIGNMENT; 1854 1855 ret = kvm_vcpu_read_guest(vcpu, gpa, ¶m, sizeof(param)); 1856 if (ret < 0) 1857 return HV_STATUS_INVALID_ALIGNMENT; 1858 } 1859 1860 /* 1861 * Per spec, bits 32-47 contain the extra "flag number". However, we 1862 * have no use for it, and in all known usecases it is zero, so just 1863 * report lookup failure if it isn't. 1864 */ 1865 if (param & 0xffff00000000ULL) 1866 return HV_STATUS_INVALID_PORT_ID; 1867 /* remaining bits are reserved-zero */ 1868 if (param & ~KVM_HYPERV_CONN_ID_MASK) 1869 return HV_STATUS_INVALID_HYPERCALL_INPUT; 1870 1871 /* the eventfd is protected by vcpu->kvm->srcu, but conn_to_evt isn't */ 1872 rcu_read_lock(); 1873 eventfd = idr_find(&hv->conn_to_evt, param); 1874 rcu_read_unlock(); 1875 if (!eventfd) 1876 return HV_STATUS_INVALID_PORT_ID; 1877 1878 eventfd_signal(eventfd, 1); 1879 return HV_STATUS_SUCCESS; 1880 } 1881 1882 int kvm_hv_hypercall(struct kvm_vcpu *vcpu) 1883 { 1884 u64 param, ingpa, outgpa, ret = HV_STATUS_SUCCESS; 1885 uint16_t code, rep_idx, rep_cnt; 1886 bool fast, rep; 1887 1888 /* 1889 * hypercall generates UD from non zero cpl and real mode 1890 * per HYPER-V spec 1891 */ 1892 if (static_call(kvm_x86_get_cpl)(vcpu) != 0 || !is_protmode(vcpu)) { 1893 kvm_queue_exception(vcpu, UD_VECTOR); 1894 return 1; 1895 } 1896 1897 #ifdef CONFIG_X86_64 1898 if (is_64_bit_mode(vcpu)) { 1899 param = kvm_rcx_read(vcpu); 1900 ingpa = kvm_rdx_read(vcpu); 1901 outgpa = kvm_r8_read(vcpu); 1902 } else 1903 #endif 1904 { 1905 param = ((u64)kvm_rdx_read(vcpu) << 32) | 1906 (kvm_rax_read(vcpu) & 0xffffffff); 1907 ingpa = ((u64)kvm_rbx_read(vcpu) << 32) | 1908 (kvm_rcx_read(vcpu) & 0xffffffff); 1909 outgpa = ((u64)kvm_rdi_read(vcpu) << 32) | 1910 (kvm_rsi_read(vcpu) & 0xffffffff); 1911 } 1912 1913 code = param & 0xffff; 1914 fast = !!(param & HV_HYPERCALL_FAST_BIT); 1915 rep_cnt = (param >> HV_HYPERCALL_REP_COMP_OFFSET) & 0xfff; 1916 rep_idx = (param >> HV_HYPERCALL_REP_START_OFFSET) & 0xfff; 1917 rep = !!(rep_cnt || rep_idx); 1918 1919 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa); 1920 1921 switch (code) { 1922 case HVCALL_NOTIFY_LONG_SPIN_WAIT: 1923 if (unlikely(rep)) { 1924 ret = HV_STATUS_INVALID_HYPERCALL_INPUT; 1925 break; 1926 } 1927 kvm_vcpu_on_spin(vcpu, true); 1928 break; 1929 case HVCALL_SIGNAL_EVENT: 1930 if (unlikely(rep)) { 1931 ret = HV_STATUS_INVALID_HYPERCALL_INPUT; 1932 break; 1933 } 1934 ret = kvm_hvcall_signal_event(vcpu, fast, ingpa); 1935 if (ret != HV_STATUS_INVALID_PORT_ID) 1936 break; 1937 fallthrough; /* maybe userspace knows this conn_id */ 1938 case HVCALL_POST_MESSAGE: 1939 /* don't bother userspace if it has no way to handle it */ 1940 if (unlikely(rep || !to_hv_synic(vcpu)->active)) { 1941 ret = HV_STATUS_INVALID_HYPERCALL_INPUT; 1942 break; 1943 } 1944 vcpu->run->exit_reason = KVM_EXIT_HYPERV; 1945 vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL; 1946 vcpu->run->hyperv.u.hcall.input = param; 1947 vcpu->run->hyperv.u.hcall.params[0] = ingpa; 1948 vcpu->run->hyperv.u.hcall.params[1] = outgpa; 1949 vcpu->arch.complete_userspace_io = 1950 kvm_hv_hypercall_complete_userspace; 1951 return 0; 1952 case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST: 1953 if (unlikely(fast || !rep_cnt || rep_idx)) { 1954 ret = HV_STATUS_INVALID_HYPERCALL_INPUT; 1955 break; 1956 } 1957 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false); 1958 break; 1959 case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE: 1960 if (unlikely(fast || rep)) { 1961 ret = HV_STATUS_INVALID_HYPERCALL_INPUT; 1962 break; 1963 } 1964 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false); 1965 break; 1966 case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX: 1967 if (unlikely(fast || !rep_cnt || rep_idx)) { 1968 ret = HV_STATUS_INVALID_HYPERCALL_INPUT; 1969 break; 1970 } 1971 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true); 1972 break; 1973 case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX: 1974 if (unlikely(fast || rep)) { 1975 ret = HV_STATUS_INVALID_HYPERCALL_INPUT; 1976 break; 1977 } 1978 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true); 1979 break; 1980 case HVCALL_SEND_IPI: 1981 if (unlikely(rep)) { 1982 ret = HV_STATUS_INVALID_HYPERCALL_INPUT; 1983 break; 1984 } 1985 ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, false, fast); 1986 break; 1987 case HVCALL_SEND_IPI_EX: 1988 if (unlikely(fast || rep)) { 1989 ret = HV_STATUS_INVALID_HYPERCALL_INPUT; 1990 break; 1991 } 1992 ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, true, false); 1993 break; 1994 case HVCALL_POST_DEBUG_DATA: 1995 case HVCALL_RETRIEVE_DEBUG_DATA: 1996 if (unlikely(fast)) { 1997 ret = HV_STATUS_INVALID_PARAMETER; 1998 break; 1999 } 2000 fallthrough; 2001 case HVCALL_RESET_DEBUG_SESSION: { 2002 struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu); 2003 2004 if (!kvm_hv_is_syndbg_enabled(vcpu)) { 2005 ret = HV_STATUS_INVALID_HYPERCALL_CODE; 2006 break; 2007 } 2008 2009 if (!(syndbg->options & HV_X64_SYNDBG_OPTION_USE_HCALLS)) { 2010 ret = HV_STATUS_OPERATION_DENIED; 2011 break; 2012 } 2013 vcpu->run->exit_reason = KVM_EXIT_HYPERV; 2014 vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL; 2015 vcpu->run->hyperv.u.hcall.input = param; 2016 vcpu->run->hyperv.u.hcall.params[0] = ingpa; 2017 vcpu->run->hyperv.u.hcall.params[1] = outgpa; 2018 vcpu->arch.complete_userspace_io = 2019 kvm_hv_hypercall_complete_userspace; 2020 return 0; 2021 } 2022 default: 2023 ret = HV_STATUS_INVALID_HYPERCALL_CODE; 2024 break; 2025 } 2026 2027 return kvm_hv_hypercall_complete(vcpu, ret); 2028 } 2029 2030 void kvm_hv_init_vm(struct kvm *kvm) 2031 { 2032 struct kvm_hv *hv = to_kvm_hv(kvm); 2033 2034 mutex_init(&hv->hv_lock); 2035 idr_init(&hv->conn_to_evt); 2036 } 2037 2038 void kvm_hv_destroy_vm(struct kvm *kvm) 2039 { 2040 struct kvm_hv *hv = to_kvm_hv(kvm); 2041 struct eventfd_ctx *eventfd; 2042 int i; 2043 2044 idr_for_each_entry(&hv->conn_to_evt, eventfd, i) 2045 eventfd_ctx_put(eventfd); 2046 idr_destroy(&hv->conn_to_evt); 2047 } 2048 2049 static int kvm_hv_eventfd_assign(struct kvm *kvm, u32 conn_id, int fd) 2050 { 2051 struct kvm_hv *hv = to_kvm_hv(kvm); 2052 struct eventfd_ctx *eventfd; 2053 int ret; 2054 2055 eventfd = eventfd_ctx_fdget(fd); 2056 if (IS_ERR(eventfd)) 2057 return PTR_ERR(eventfd); 2058 2059 mutex_lock(&hv->hv_lock); 2060 ret = idr_alloc(&hv->conn_to_evt, eventfd, conn_id, conn_id + 1, 2061 GFP_KERNEL_ACCOUNT); 2062 mutex_unlock(&hv->hv_lock); 2063 2064 if (ret >= 0) 2065 return 0; 2066 2067 if (ret == -ENOSPC) 2068 ret = -EEXIST; 2069 eventfd_ctx_put(eventfd); 2070 return ret; 2071 } 2072 2073 static int kvm_hv_eventfd_deassign(struct kvm *kvm, u32 conn_id) 2074 { 2075 struct kvm_hv *hv = to_kvm_hv(kvm); 2076 struct eventfd_ctx *eventfd; 2077 2078 mutex_lock(&hv->hv_lock); 2079 eventfd = idr_remove(&hv->conn_to_evt, conn_id); 2080 mutex_unlock(&hv->hv_lock); 2081 2082 if (!eventfd) 2083 return -ENOENT; 2084 2085 synchronize_srcu(&kvm->srcu); 2086 eventfd_ctx_put(eventfd); 2087 return 0; 2088 } 2089 2090 int kvm_vm_ioctl_hv_eventfd(struct kvm *kvm, struct kvm_hyperv_eventfd *args) 2091 { 2092 if ((args->flags & ~KVM_HYPERV_EVENTFD_DEASSIGN) || 2093 (args->conn_id & ~KVM_HYPERV_CONN_ID_MASK)) 2094 return -EINVAL; 2095 2096 if (args->flags == KVM_HYPERV_EVENTFD_DEASSIGN) 2097 return kvm_hv_eventfd_deassign(kvm, args->conn_id); 2098 return kvm_hv_eventfd_assign(kvm, args->conn_id, args->fd); 2099 } 2100 2101 int kvm_get_hv_cpuid(struct kvm_vcpu *vcpu, struct kvm_cpuid2 *cpuid, 2102 struct kvm_cpuid_entry2 __user *entries) 2103 { 2104 uint16_t evmcs_ver = 0; 2105 struct kvm_cpuid_entry2 cpuid_entries[] = { 2106 { .function = HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS }, 2107 { .function = HYPERV_CPUID_INTERFACE }, 2108 { .function = HYPERV_CPUID_VERSION }, 2109 { .function = HYPERV_CPUID_FEATURES }, 2110 { .function = HYPERV_CPUID_ENLIGHTMENT_INFO }, 2111 { .function = HYPERV_CPUID_IMPLEMENT_LIMITS }, 2112 { .function = HYPERV_CPUID_SYNDBG_VENDOR_AND_MAX_FUNCTIONS }, 2113 { .function = HYPERV_CPUID_SYNDBG_INTERFACE }, 2114 { .function = HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES }, 2115 { .function = HYPERV_CPUID_NESTED_FEATURES }, 2116 }; 2117 int i, nent = ARRAY_SIZE(cpuid_entries); 2118 2119 if (kvm_x86_ops.nested_ops->get_evmcs_version) 2120 evmcs_ver = kvm_x86_ops.nested_ops->get_evmcs_version(vcpu); 2121 2122 /* Skip NESTED_FEATURES if eVMCS is not supported */ 2123 if (!evmcs_ver) 2124 --nent; 2125 2126 if (cpuid->nent < nent) 2127 return -E2BIG; 2128 2129 if (cpuid->nent > nent) 2130 cpuid->nent = nent; 2131 2132 for (i = 0; i < nent; i++) { 2133 struct kvm_cpuid_entry2 *ent = &cpuid_entries[i]; 2134 u32 signature[3]; 2135 2136 switch (ent->function) { 2137 case HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS: 2138 memcpy(signature, "Linux KVM Hv", 12); 2139 2140 ent->eax = HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES; 2141 ent->ebx = signature[0]; 2142 ent->ecx = signature[1]; 2143 ent->edx = signature[2]; 2144 break; 2145 2146 case HYPERV_CPUID_INTERFACE: 2147 ent->eax = HYPERV_CPUID_SIGNATURE_EAX; 2148 break; 2149 2150 case HYPERV_CPUID_VERSION: 2151 /* 2152 * We implement some Hyper-V 2016 functions so let's use 2153 * this version. 2154 */ 2155 ent->eax = 0x00003839; 2156 ent->ebx = 0x000A0000; 2157 break; 2158 2159 case HYPERV_CPUID_FEATURES: 2160 ent->eax |= HV_MSR_VP_RUNTIME_AVAILABLE; 2161 ent->eax |= HV_MSR_TIME_REF_COUNT_AVAILABLE; 2162 ent->eax |= HV_MSR_SYNIC_AVAILABLE; 2163 ent->eax |= HV_MSR_SYNTIMER_AVAILABLE; 2164 ent->eax |= HV_MSR_APIC_ACCESS_AVAILABLE; 2165 ent->eax |= HV_MSR_HYPERCALL_AVAILABLE; 2166 ent->eax |= HV_MSR_VP_INDEX_AVAILABLE; 2167 ent->eax |= HV_MSR_RESET_AVAILABLE; 2168 ent->eax |= HV_MSR_REFERENCE_TSC_AVAILABLE; 2169 ent->eax |= HV_ACCESS_FREQUENCY_MSRS; 2170 ent->eax |= HV_ACCESS_REENLIGHTENMENT; 2171 2172 ent->ebx |= HV_POST_MESSAGES; 2173 ent->ebx |= HV_SIGNAL_EVENTS; 2174 2175 ent->edx |= HV_FEATURE_FREQUENCY_MSRS_AVAILABLE; 2176 ent->edx |= HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE; 2177 2178 ent->ebx |= HV_DEBUGGING; 2179 ent->edx |= HV_X64_GUEST_DEBUGGING_AVAILABLE; 2180 ent->edx |= HV_FEATURE_DEBUG_MSRS_AVAILABLE; 2181 2182 /* 2183 * Direct Synthetic timers only make sense with in-kernel 2184 * LAPIC 2185 */ 2186 if (!vcpu || lapic_in_kernel(vcpu)) 2187 ent->edx |= HV_STIMER_DIRECT_MODE_AVAILABLE; 2188 2189 break; 2190 2191 case HYPERV_CPUID_ENLIGHTMENT_INFO: 2192 ent->eax |= HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED; 2193 ent->eax |= HV_X64_APIC_ACCESS_RECOMMENDED; 2194 ent->eax |= HV_X64_RELAXED_TIMING_RECOMMENDED; 2195 ent->eax |= HV_X64_CLUSTER_IPI_RECOMMENDED; 2196 ent->eax |= HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED; 2197 if (evmcs_ver) 2198 ent->eax |= HV_X64_ENLIGHTENED_VMCS_RECOMMENDED; 2199 if (!cpu_smt_possible()) 2200 ent->eax |= HV_X64_NO_NONARCH_CORESHARING; 2201 /* 2202 * Default number of spinlock retry attempts, matches 2203 * HyperV 2016. 2204 */ 2205 ent->ebx = 0x00000FFF; 2206 2207 break; 2208 2209 case HYPERV_CPUID_IMPLEMENT_LIMITS: 2210 /* Maximum number of virtual processors */ 2211 ent->eax = KVM_MAX_VCPUS; 2212 /* 2213 * Maximum number of logical processors, matches 2214 * HyperV 2016. 2215 */ 2216 ent->ebx = 64; 2217 2218 break; 2219 2220 case HYPERV_CPUID_NESTED_FEATURES: 2221 ent->eax = evmcs_ver; 2222 2223 break; 2224 2225 case HYPERV_CPUID_SYNDBG_VENDOR_AND_MAX_FUNCTIONS: 2226 memcpy(signature, "Linux KVM Hv", 12); 2227 2228 ent->eax = 0; 2229 ent->ebx = signature[0]; 2230 ent->ecx = signature[1]; 2231 ent->edx = signature[2]; 2232 break; 2233 2234 case HYPERV_CPUID_SYNDBG_INTERFACE: 2235 memcpy(signature, "VS#1\0\0\0\0\0\0\0\0", 12); 2236 ent->eax = signature[0]; 2237 break; 2238 2239 case HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES: 2240 ent->eax |= HV_X64_SYNDBG_CAP_ALLOW_KERNEL_DEBUGGING; 2241 break; 2242 2243 default: 2244 break; 2245 } 2246 } 2247 2248 if (copy_to_user(entries, cpuid_entries, 2249 nent * sizeof(struct kvm_cpuid_entry2))) 2250 return -EFAULT; 2251 2252 return 0; 2253 } 2254