2 * KVM Microsoft Hyper-V emulation
4 * derived from arch/x86/kvm/x86.c
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>
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>
19 * This work is licensed under the terms of the GNU GPL, version 2. See
20 * the COPYING file in the top-level directory.
29 #include <linux/kvm_host.h>
30 #include <linux/highmem.h>
31 #include <linux/nospec.h>
32 #include <asm/apicdef.h>
33 #include <trace/events/kvm.h>
37 static inline u64 synic_read_sint(struct kvm_vcpu_hv_synic *synic, int sint)
39 return atomic64_read(&synic->sint[sint]);
42 static inline int synic_get_sint_vector(u64 sint_value)
44 if (sint_value & HV_SYNIC_SINT_MASKED)
46 return sint_value & HV_SYNIC_SINT_VECTOR_MASK;
49 static bool synic_has_vector_connected(struct kvm_vcpu_hv_synic *synic,
54 for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
55 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
61 static bool synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic *synic,
67 for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
68 sint_value = synic_read_sint(synic, i);
69 if (synic_get_sint_vector(sint_value) == vector &&
70 sint_value & HV_SYNIC_SINT_AUTO_EOI)
76 static int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint,
81 vector = data & HV_SYNIC_SINT_VECTOR_MASK;
82 if (vector < 16 && !host)
85 * Guest may configure multiple SINTs to use the same vector, so
86 * we maintain a bitmap of vectors handled by synic, and a
87 * bitmap of vectors with auto-eoi behavior. The bitmaps are
88 * updated here, and atomically queried on fast paths.
91 atomic64_set(&synic->sint[sint], data);
93 if (synic_has_vector_connected(synic, vector))
94 __set_bit(vector, synic->vec_bitmap);
96 __clear_bit(vector, synic->vec_bitmap);
98 if (synic_has_vector_auto_eoi(synic, vector))
99 __set_bit(vector, synic->auto_eoi_bitmap);
101 __clear_bit(vector, synic->auto_eoi_bitmap);
103 /* Load SynIC vectors into EOI exit bitmap */
104 kvm_make_request(KVM_REQ_SCAN_IOAPIC, synic_to_vcpu(synic));
108 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vcpu_id)
110 struct kvm_vcpu *vcpu;
111 struct kvm_vcpu_hv_synic *synic;
113 if (vcpu_id >= atomic_read(&kvm->online_vcpus))
115 vcpu = kvm_get_vcpu(kvm, vcpu_id);
118 synic = vcpu_to_synic(vcpu);
119 return (synic->active) ? synic : NULL;
122 static void synic_clear_sint_msg_pending(struct kvm_vcpu_hv_synic *synic,
125 struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
128 struct hv_message *msg;
129 struct hv_message_page *msg_page;
131 gpa = synic->msg_page & PAGE_MASK;
132 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
133 if (is_error_page(page)) {
134 vcpu_err(vcpu, "Hyper-V SynIC can't get msg page, gpa 0x%llx\n",
138 msg_page = kmap_atomic(page);
140 msg = &msg_page->sint_message[sint];
141 msg->header.message_flags.msg_pending = 0;
143 kunmap_atomic(msg_page);
144 kvm_release_page_dirty(page);
145 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
148 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
150 struct kvm *kvm = vcpu->kvm;
151 struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
152 struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
153 struct kvm_vcpu_hv_stimer *stimer;
154 int gsi, idx, stimers_pending;
156 trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
158 if (synic->msg_page & HV_SYNIC_SIMP_ENABLE)
159 synic_clear_sint_msg_pending(synic, sint);
161 /* Try to deliver pending Hyper-V SynIC timers messages */
163 for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
164 stimer = &hv_vcpu->stimer[idx];
165 if (stimer->msg_pending &&
166 (stimer->config & HV_STIMER_ENABLE) &&
167 HV_STIMER_SINT(stimer->config) == sint) {
168 set_bit(stimer->index,
169 hv_vcpu->stimer_pending_bitmap);
174 kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
176 idx = srcu_read_lock(&kvm->irq_srcu);
177 gsi = atomic_read(&synic->sint_to_gsi[sint]);
179 kvm_notify_acked_gsi(kvm, gsi);
180 srcu_read_unlock(&kvm->irq_srcu, idx);
183 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
185 struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
186 struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
188 hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
189 hv_vcpu->exit.u.synic.msr = msr;
190 hv_vcpu->exit.u.synic.control = synic->control;
191 hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
192 hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
194 kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
197 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
198 u32 msr, u64 data, bool host)
200 struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
206 trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
210 case HV_X64_MSR_SCONTROL:
211 synic->control = data;
213 synic_exit(synic, msr);
215 case HV_X64_MSR_SVERSION:
220 synic->version = data;
222 case HV_X64_MSR_SIEFP:
223 if (data & HV_SYNIC_SIEFP_ENABLE)
224 if (kvm_clear_guest(vcpu->kvm,
225 data & PAGE_MASK, PAGE_SIZE)) {
229 synic->evt_page = data;
231 synic_exit(synic, msr);
233 case HV_X64_MSR_SIMP:
234 if (data & HV_SYNIC_SIMP_ENABLE)
235 if (kvm_clear_guest(vcpu->kvm,
236 data & PAGE_MASK, PAGE_SIZE)) {
240 synic->msg_page = data;
242 synic_exit(synic, msr);
244 case HV_X64_MSR_EOM: {
247 for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
248 kvm_hv_notify_acked_sint(vcpu, i);
251 case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
252 ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
261 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata)
270 case HV_X64_MSR_SCONTROL:
271 *pdata = synic->control;
273 case HV_X64_MSR_SVERSION:
274 *pdata = synic->version;
276 case HV_X64_MSR_SIEFP:
277 *pdata = synic->evt_page;
279 case HV_X64_MSR_SIMP:
280 *pdata = synic->msg_page;
285 case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
286 *pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
295 int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
297 struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
298 struct kvm_lapic_irq irq;
301 if (sint >= ARRAY_SIZE(synic->sint))
304 vector = synic_get_sint_vector(synic_read_sint(synic, sint));
308 memset(&irq, 0, sizeof(irq));
309 irq.dest_id = kvm_apic_id(vcpu->arch.apic);
310 irq.dest_mode = APIC_DEST_PHYSICAL;
311 irq.delivery_mode = APIC_DM_FIXED;
315 ret = kvm_irq_delivery_to_apic(vcpu->kvm, NULL, &irq, NULL);
316 trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
320 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vcpu_id, u32 sint)
322 struct kvm_vcpu_hv_synic *synic;
324 synic = synic_get(kvm, vcpu_id);
328 return synic_set_irq(synic, sint);
331 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
333 struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
336 trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
338 for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
339 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
340 kvm_hv_notify_acked_sint(vcpu, i);
343 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vcpu_id, u32 sint, int gsi)
345 struct kvm_vcpu_hv_synic *synic;
347 synic = synic_get(kvm, vcpu_id);
351 if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
354 atomic_set(&synic->sint_to_gsi[sint], gsi);
358 void kvm_hv_irq_routing_update(struct kvm *kvm)
360 struct kvm_irq_routing_table *irq_rt;
361 struct kvm_kernel_irq_routing_entry *e;
364 irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
365 lockdep_is_held(&kvm->irq_lock));
367 for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
368 hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
369 if (e->type == KVM_IRQ_ROUTING_HV_SINT)
370 kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
371 e->hv_sint.sint, gsi);
376 static void synic_init(struct kvm_vcpu_hv_synic *synic)
380 memset(synic, 0, sizeof(*synic));
381 synic->version = HV_SYNIC_VERSION_1;
382 for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
383 atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
384 atomic_set(&synic->sint_to_gsi[i], -1);
388 static u64 get_time_ref_counter(struct kvm *kvm)
390 struct kvm_hv *hv = &kvm->arch.hyperv;
391 struct kvm_vcpu *vcpu;
395 * The guest has not set up the TSC page or the clock isn't
396 * stable, fall back to get_kvmclock_ns.
398 if (!hv->tsc_ref.tsc_sequence)
399 return div_u64(get_kvmclock_ns(kvm), 100);
401 vcpu = kvm_get_vcpu(kvm, 0);
402 tsc = kvm_read_l1_tsc(vcpu, rdtsc());
403 return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64)
404 + hv->tsc_ref.tsc_offset;
407 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
410 struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
412 set_bit(stimer->index,
413 vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
414 kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
419 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
421 struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
423 trace_kvm_hv_stimer_cleanup(stimer_to_vcpu(stimer)->vcpu_id,
426 hrtimer_cancel(&stimer->timer);
427 clear_bit(stimer->index,
428 vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
429 stimer->msg_pending = false;
430 stimer->exp_time = 0;
433 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
435 struct kvm_vcpu_hv_stimer *stimer;
437 stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
438 trace_kvm_hv_stimer_callback(stimer_to_vcpu(stimer)->vcpu_id,
440 stimer_mark_pending(stimer, true);
442 return HRTIMER_NORESTART;
446 * stimer_start() assumptions:
447 * a) stimer->count is not equal to 0
448 * b) stimer->config has HV_STIMER_ENABLE flag
450 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
455 time_now = get_time_ref_counter(stimer_to_vcpu(stimer)->kvm);
456 ktime_now = ktime_get();
458 if (stimer->config & HV_STIMER_PERIODIC) {
459 if (stimer->exp_time) {
460 if (time_now >= stimer->exp_time) {
463 div64_u64_rem(time_now - stimer->exp_time,
464 stimer->count, &remainder);
466 time_now + (stimer->count - remainder);
469 stimer->exp_time = time_now + stimer->count;
471 trace_kvm_hv_stimer_start_periodic(
472 stimer_to_vcpu(stimer)->vcpu_id,
474 time_now, stimer->exp_time);
476 hrtimer_start(&stimer->timer,
477 ktime_add_ns(ktime_now,
478 100 * (stimer->exp_time - time_now)),
482 stimer->exp_time = stimer->count;
483 if (time_now >= stimer->count) {
485 * Expire timer according to Hypervisor Top-Level Functional
486 * specification v4(15.3.1):
487 * "If a one shot is enabled and the specified count is in
488 * the past, it will expire immediately."
490 stimer_mark_pending(stimer, false);
494 trace_kvm_hv_stimer_start_one_shot(stimer_to_vcpu(stimer)->vcpu_id,
496 time_now, stimer->count);
498 hrtimer_start(&stimer->timer,
499 ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)),
504 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
507 trace_kvm_hv_stimer_set_config(stimer_to_vcpu(stimer)->vcpu_id,
508 stimer->index, config, host);
510 stimer_cleanup(stimer);
511 if ((stimer->config & HV_STIMER_ENABLE) && HV_STIMER_SINT(config) == 0)
512 config &= ~HV_STIMER_ENABLE;
513 stimer->config = config;
514 stimer_mark_pending(stimer, false);
518 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
521 trace_kvm_hv_stimer_set_count(stimer_to_vcpu(stimer)->vcpu_id,
522 stimer->index, count, host);
524 stimer_cleanup(stimer);
525 stimer->count = count;
526 if (stimer->count == 0)
527 stimer->config &= ~HV_STIMER_ENABLE;
528 else if (stimer->config & HV_STIMER_AUTOENABLE)
529 stimer->config |= HV_STIMER_ENABLE;
530 stimer_mark_pending(stimer, false);
534 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
536 *pconfig = stimer->config;
540 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
542 *pcount = stimer->count;
546 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
547 struct hv_message *src_msg)
549 struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
552 struct hv_message *dst_msg;
554 struct hv_message_page *msg_page;
556 if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
559 gpa = synic->msg_page & PAGE_MASK;
560 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
561 if (is_error_page(page))
564 msg_page = kmap_atomic(page);
565 dst_msg = &msg_page->sint_message[sint];
566 if (sync_cmpxchg(&dst_msg->header.message_type, HVMSG_NONE,
567 src_msg->header.message_type) != HVMSG_NONE) {
568 dst_msg->header.message_flags.msg_pending = 1;
571 memcpy(&dst_msg->u.payload, &src_msg->u.payload,
572 src_msg->header.payload_size);
573 dst_msg->header.message_type = src_msg->header.message_type;
574 dst_msg->header.payload_size = src_msg->header.payload_size;
575 r = synic_set_irq(synic, sint);
581 kunmap_atomic(msg_page);
582 kvm_release_page_dirty(page);
583 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
587 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
589 struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
590 struct hv_message *msg = &stimer->msg;
591 struct hv_timer_message_payload *payload =
592 (struct hv_timer_message_payload *)&msg->u.payload;
594 payload->expiration_time = stimer->exp_time;
595 payload->delivery_time = get_time_ref_counter(vcpu->kvm);
596 return synic_deliver_msg(vcpu_to_synic(vcpu),
597 HV_STIMER_SINT(stimer->config), msg);
600 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
604 stimer->msg_pending = true;
605 r = stimer_send_msg(stimer);
606 trace_kvm_hv_stimer_expiration(stimer_to_vcpu(stimer)->vcpu_id,
609 stimer->msg_pending = false;
610 if (!(stimer->config & HV_STIMER_PERIODIC))
611 stimer->config &= ~HV_STIMER_ENABLE;
615 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
617 struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
618 struct kvm_vcpu_hv_stimer *stimer;
619 u64 time_now, exp_time;
622 for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
623 if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
624 stimer = &hv_vcpu->stimer[i];
625 if (stimer->config & HV_STIMER_ENABLE) {
626 exp_time = stimer->exp_time;
630 get_time_ref_counter(vcpu->kvm);
631 if (time_now >= exp_time)
632 stimer_expiration(stimer);
635 if ((stimer->config & HV_STIMER_ENABLE) &&
637 stimer_start(stimer);
639 stimer_cleanup(stimer);
644 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
646 struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
649 for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
650 stimer_cleanup(&hv_vcpu->stimer[i]);
653 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
655 struct hv_message *msg = &stimer->msg;
656 struct hv_timer_message_payload *payload =
657 (struct hv_timer_message_payload *)&msg->u.payload;
659 memset(&msg->header, 0, sizeof(msg->header));
660 msg->header.message_type = HVMSG_TIMER_EXPIRED;
661 msg->header.payload_size = sizeof(*payload);
663 payload->timer_index = stimer->index;
664 payload->expiration_time = 0;
665 payload->delivery_time = 0;
668 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
670 memset(stimer, 0, sizeof(*stimer));
671 stimer->index = timer_index;
672 hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
673 stimer->timer.function = stimer_timer_callback;
674 stimer_prepare_msg(stimer);
677 void kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
679 struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
682 synic_init(&hv_vcpu->synic);
684 bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
685 for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
686 stimer_init(&hv_vcpu->stimer[i], i);
689 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu)
692 * Hyper-V SynIC auto EOI SINT's are
693 * not compatible with APICV, so deactivate APICV
695 kvm_vcpu_deactivate_apicv(vcpu);
696 vcpu_to_synic(vcpu)->active = true;
700 static bool kvm_hv_msr_partition_wide(u32 msr)
705 case HV_X64_MSR_GUEST_OS_ID:
706 case HV_X64_MSR_HYPERCALL:
707 case HV_X64_MSR_REFERENCE_TSC:
708 case HV_X64_MSR_TIME_REF_COUNT:
709 case HV_X64_MSR_CRASH_CTL:
710 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
711 case HV_X64_MSR_RESET:
719 static int kvm_hv_msr_get_crash_data(struct kvm_vcpu *vcpu,
720 u32 index, u64 *pdata)
722 struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
723 size_t size = ARRAY_SIZE(hv->hv_crash_param);
725 if (WARN_ON_ONCE(index >= size))
728 *pdata = hv->hv_crash_param[array_index_nospec(index, size)];
732 static int kvm_hv_msr_get_crash_ctl(struct kvm_vcpu *vcpu, u64 *pdata)
734 struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
736 *pdata = hv->hv_crash_ctl;
740 static int kvm_hv_msr_set_crash_ctl(struct kvm_vcpu *vcpu, u64 data, bool host)
742 struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
745 hv->hv_crash_ctl = data & HV_X64_MSR_CRASH_CTL_NOTIFY;
747 if (!host && (data & HV_X64_MSR_CRASH_CTL_NOTIFY)) {
749 vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
750 hv->hv_crash_param[0],
751 hv->hv_crash_param[1],
752 hv->hv_crash_param[2],
753 hv->hv_crash_param[3],
754 hv->hv_crash_param[4]);
756 /* Send notification about crash to user space */
757 kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
763 static int kvm_hv_msr_set_crash_data(struct kvm_vcpu *vcpu,
766 struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
767 size_t size = ARRAY_SIZE(hv->hv_crash_param);
769 if (WARN_ON_ONCE(index >= size))
772 hv->hv_crash_param[array_index_nospec(index, size)] = data;
777 * The kvmclock and Hyper-V TSC page use similar formulas, and converting
778 * between them is possible:
781 * nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
785 * nsec/100 = ticks * scale / 2^64 + offset
787 * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
788 * By dividing the kvmclock formula by 100 and equating what's left we get:
789 * ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
790 * scale / 2^64 = tsc_to_system_mul * 2^(tsc_shift-32) / 100
791 * scale = tsc_to_system_mul * 2^(32+tsc_shift) / 100
793 * Now expand the kvmclock formula and divide by 100:
794 * nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
795 * - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
797 * nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
798 * - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
799 * + system_time / 100
801 * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
802 * nsec/100 = ticks * scale / 2^64
803 * - tsc_timestamp * scale / 2^64
804 * + system_time / 100
806 * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
807 * offset = system_time / 100 - tsc_timestamp * scale / 2^64
809 * These two equivalencies are implemented in this function.
811 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
812 HV_REFERENCE_TSC_PAGE *tsc_ref)
816 if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
820 * check if scale would overflow, if so we use the time ref counter
821 * tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
822 * tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
823 * tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
825 max_mul = 100ull << (32 - hv_clock->tsc_shift);
826 if (hv_clock->tsc_to_system_mul >= max_mul)
830 * Otherwise compute the scale and offset according to the formulas
834 mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
835 hv_clock->tsc_to_system_mul,
838 tsc_ref->tsc_offset = hv_clock->system_time;
839 do_div(tsc_ref->tsc_offset, 100);
840 tsc_ref->tsc_offset -=
841 mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
845 void kvm_hv_setup_tsc_page(struct kvm *kvm,
846 struct pvclock_vcpu_time_info *hv_clock)
848 struct kvm_hv *hv = &kvm->arch.hyperv;
852 BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
853 BUILD_BUG_ON(offsetof(HV_REFERENCE_TSC_PAGE, tsc_sequence) != 0);
855 if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
858 gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
860 * Because the TSC parameters only vary when there is a
861 * change in the master clock, do not bother with caching.
863 if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
864 &tsc_seq, sizeof(tsc_seq))))
868 * While we're computing and writing the parameters, force the
869 * guest to use the time reference count MSR.
871 hv->tsc_ref.tsc_sequence = 0;
872 if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
873 &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
876 if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
879 /* Ensure sequence is zero before writing the rest of the struct. */
881 if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
885 * Now switch to the TSC page mechanism by writing the sequence.
888 if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
891 /* Write the struct entirely before the non-zero sequence. */
894 hv->tsc_ref.tsc_sequence = tsc_seq;
895 kvm_write_guest(kvm, gfn_to_gpa(gfn),
896 &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence));
899 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
902 struct kvm *kvm = vcpu->kvm;
903 struct kvm_hv *hv = &kvm->arch.hyperv;
906 case HV_X64_MSR_GUEST_OS_ID:
907 hv->hv_guest_os_id = data;
908 /* setting guest os id to zero disables hypercall page */
909 if (!hv->hv_guest_os_id)
910 hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
912 case HV_X64_MSR_HYPERCALL: {
917 /* if guest os id is not set hypercall should remain disabled */
918 if (!hv->hv_guest_os_id)
920 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
921 hv->hv_hypercall = data;
924 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
925 addr = gfn_to_hva(kvm, gfn);
926 if (kvm_is_error_hva(addr))
928 kvm_x86_ops->patch_hypercall(vcpu, instructions);
929 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
930 if (__copy_to_user((void __user *)addr, instructions, 4))
932 hv->hv_hypercall = data;
933 mark_page_dirty(kvm, gfn);
936 case HV_X64_MSR_REFERENCE_TSC:
937 hv->hv_tsc_page = data;
938 if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE)
939 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
941 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
942 return kvm_hv_msr_set_crash_data(vcpu,
943 msr - HV_X64_MSR_CRASH_P0,
945 case HV_X64_MSR_CRASH_CTL:
946 return kvm_hv_msr_set_crash_ctl(vcpu, data, host);
947 case HV_X64_MSR_RESET:
949 vcpu_debug(vcpu, "hyper-v reset requested\n");
950 kvm_make_request(KVM_REQ_HV_RESET, vcpu);
954 vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
961 /* Calculate cpu time spent by current task in 100ns units */
962 static u64 current_task_runtime_100ns(void)
964 cputime_t utime, stime;
966 task_cputime_adjusted(current, &utime, &stime);
967 return div_u64(cputime_to_nsecs(utime + stime), 100);
970 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
972 struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
975 case HV_X64_MSR_APIC_ASSIST_PAGE: {
979 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
981 if (kvm_lapic_enable_pv_eoi(vcpu, 0))
985 gfn = data >> HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT;
986 addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
987 if (kvm_is_error_hva(addr))
989 if (__clear_user((void __user *)addr, PAGE_SIZE))
992 kvm_vcpu_mark_page_dirty(vcpu, gfn);
993 if (kvm_lapic_enable_pv_eoi(vcpu,
994 gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
999 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1000 case HV_X64_MSR_ICR:
1001 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1002 case HV_X64_MSR_TPR:
1003 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1004 case HV_X64_MSR_VP_RUNTIME:
1007 hv->runtime_offset = data - current_task_runtime_100ns();
1009 case HV_X64_MSR_SCONTROL:
1010 case HV_X64_MSR_SVERSION:
1011 case HV_X64_MSR_SIEFP:
1012 case HV_X64_MSR_SIMP:
1013 case HV_X64_MSR_EOM:
1014 case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1015 return synic_set_msr(vcpu_to_synic(vcpu), msr, data, host);
1016 case HV_X64_MSR_STIMER0_CONFIG:
1017 case HV_X64_MSR_STIMER1_CONFIG:
1018 case HV_X64_MSR_STIMER2_CONFIG:
1019 case HV_X64_MSR_STIMER3_CONFIG: {
1020 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1022 return stimer_set_config(vcpu_to_stimer(vcpu, timer_index),
1025 case HV_X64_MSR_STIMER0_COUNT:
1026 case HV_X64_MSR_STIMER1_COUNT:
1027 case HV_X64_MSR_STIMER2_COUNT:
1028 case HV_X64_MSR_STIMER3_COUNT: {
1029 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1031 return stimer_set_count(vcpu_to_stimer(vcpu, timer_index),
1035 vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
1043 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1046 struct kvm *kvm = vcpu->kvm;
1047 struct kvm_hv *hv = &kvm->arch.hyperv;
1050 case HV_X64_MSR_GUEST_OS_ID:
1051 data = hv->hv_guest_os_id;
1053 case HV_X64_MSR_HYPERCALL:
1054 data = hv->hv_hypercall;
1056 case HV_X64_MSR_TIME_REF_COUNT:
1057 data = get_time_ref_counter(kvm);
1059 case HV_X64_MSR_REFERENCE_TSC:
1060 data = hv->hv_tsc_page;
1062 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1063 return kvm_hv_msr_get_crash_data(vcpu,
1064 msr - HV_X64_MSR_CRASH_P0,
1066 case HV_X64_MSR_CRASH_CTL:
1067 return kvm_hv_msr_get_crash_ctl(vcpu, pdata);
1068 case HV_X64_MSR_RESET:
1072 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1080 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1083 struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
1086 case HV_X64_MSR_VP_INDEX: {
1090 kvm_for_each_vcpu(r, v, vcpu->kvm) {
1098 case HV_X64_MSR_EOI:
1099 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1100 case HV_X64_MSR_ICR:
1101 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1102 case HV_X64_MSR_TPR:
1103 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1104 case HV_X64_MSR_APIC_ASSIST_PAGE:
1105 data = hv->hv_vapic;
1107 case HV_X64_MSR_VP_RUNTIME:
1108 data = current_task_runtime_100ns() + hv->runtime_offset;
1110 case HV_X64_MSR_SCONTROL:
1111 case HV_X64_MSR_SVERSION:
1112 case HV_X64_MSR_SIEFP:
1113 case HV_X64_MSR_SIMP:
1114 case HV_X64_MSR_EOM:
1115 case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1116 return synic_get_msr(vcpu_to_synic(vcpu), msr, pdata);
1117 case HV_X64_MSR_STIMER0_CONFIG:
1118 case HV_X64_MSR_STIMER1_CONFIG:
1119 case HV_X64_MSR_STIMER2_CONFIG:
1120 case HV_X64_MSR_STIMER3_CONFIG: {
1121 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1123 return stimer_get_config(vcpu_to_stimer(vcpu, timer_index),
1126 case HV_X64_MSR_STIMER0_COUNT:
1127 case HV_X64_MSR_STIMER1_COUNT:
1128 case HV_X64_MSR_STIMER2_COUNT:
1129 case HV_X64_MSR_STIMER3_COUNT: {
1130 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1132 return stimer_get_count(vcpu_to_stimer(vcpu, timer_index),
1136 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1143 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1145 if (kvm_hv_msr_partition_wide(msr)) {
1148 mutex_lock(&vcpu->kvm->lock);
1149 r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1150 mutex_unlock(&vcpu->kvm->lock);
1153 return kvm_hv_set_msr(vcpu, msr, data, host);
1156 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1158 if (kvm_hv_msr_partition_wide(msr)) {
1161 mutex_lock(&vcpu->kvm->lock);
1162 r = kvm_hv_get_msr_pw(vcpu, msr, pdata);
1163 mutex_unlock(&vcpu->kvm->lock);
1166 return kvm_hv_get_msr(vcpu, msr, pdata);
1169 bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1171 return kvm->arch.hyperv.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1174 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
1178 longmode = is_64_bit_mode(vcpu);
1180 kvm_register_write(vcpu, VCPU_REGS_RAX, result);
1182 kvm_register_write(vcpu, VCPU_REGS_RDX, result >> 32);
1183 kvm_register_write(vcpu, VCPU_REGS_RAX, result & 0xffffffff);
1187 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
1189 struct kvm_run *run = vcpu->run;
1191 kvm_hv_hypercall_set_result(vcpu, run->hyperv.u.hcall.result);
1195 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
1197 u64 param, ingpa, outgpa, ret;
1198 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
1199 bool fast, longmode;
1202 * hypercall generates UD from non zero cpl and real mode
1205 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
1206 kvm_queue_exception(vcpu, UD_VECTOR);
1210 longmode = is_64_bit_mode(vcpu);
1213 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
1214 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
1215 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
1216 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
1217 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
1218 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
1220 #ifdef CONFIG_X86_64
1222 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
1223 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
1224 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
1228 code = param & 0xffff;
1229 fast = (param >> 16) & 0x1;
1230 rep_cnt = (param >> 32) & 0xfff;
1231 rep_idx = (param >> 48) & 0xfff;
1233 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
1235 /* Hypercall continuation is not supported yet */
1236 if (rep_cnt || rep_idx) {
1237 res = HV_STATUS_INVALID_HYPERCALL_CODE;
1242 case HVCALL_NOTIFY_LONG_SPIN_WAIT:
1243 kvm_vcpu_on_spin(vcpu);
1245 case HVCALL_POST_MESSAGE:
1246 case HVCALL_SIGNAL_EVENT:
1247 /* don't bother userspace if it has no way to handle it */
1248 if (!vcpu_to_synic(vcpu)->active) {
1249 res = HV_STATUS_INVALID_HYPERCALL_CODE;
1252 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
1253 vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
1254 vcpu->run->hyperv.u.hcall.input = param;
1255 vcpu->run->hyperv.u.hcall.params[0] = ingpa;
1256 vcpu->run->hyperv.u.hcall.params[1] = outgpa;
1257 vcpu->arch.complete_userspace_io =
1258 kvm_hv_hypercall_complete_userspace;
1261 res = HV_STATUS_INVALID_HYPERCALL_CODE;
1266 ret = res | (((u64)rep_done & 0xfff) << 32);
1267 kvm_hv_hypercall_set_result(vcpu, ret);