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 <asm/apicdef.h>
32 #include <trace/events/kvm.h>
36 static inline u64 synic_read_sint(struct kvm_vcpu_hv_synic *synic, int sint)
38 return atomic64_read(&synic->sint[sint]);
41 static inline int synic_get_sint_vector(u64 sint_value)
43 if (sint_value & HV_SYNIC_SINT_MASKED)
45 return sint_value & HV_SYNIC_SINT_VECTOR_MASK;
48 static bool synic_has_vector_connected(struct kvm_vcpu_hv_synic *synic,
53 for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
54 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
60 static bool synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic *synic,
66 for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
67 sint_value = synic_read_sint(synic, i);
68 if (synic_get_sint_vector(sint_value) == vector &&
69 sint_value & HV_SYNIC_SINT_AUTO_EOI)
75 static int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint,
80 vector = data & HV_SYNIC_SINT_VECTOR_MASK;
81 if (vector < 16 && !host)
84 * Guest may configure multiple SINTs to use the same vector, so
85 * we maintain a bitmap of vectors handled by synic, and a
86 * bitmap of vectors with auto-eoi behavior. The bitmaps are
87 * updated here, and atomically queried on fast paths.
90 atomic64_set(&synic->sint[sint], data);
92 if (synic_has_vector_connected(synic, vector))
93 __set_bit(vector, synic->vec_bitmap);
95 __clear_bit(vector, synic->vec_bitmap);
97 if (synic_has_vector_auto_eoi(synic, vector))
98 __set_bit(vector, synic->auto_eoi_bitmap);
100 __clear_bit(vector, synic->auto_eoi_bitmap);
102 /* Load SynIC vectors into EOI exit bitmap */
103 kvm_make_request(KVM_REQ_SCAN_IOAPIC, synic_to_vcpu(synic));
107 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vcpu_id)
109 struct kvm_vcpu *vcpu;
110 struct kvm_vcpu_hv_synic *synic;
112 if (vcpu_id >= atomic_read(&kvm->online_vcpus))
114 vcpu = kvm_get_vcpu(kvm, vcpu_id);
117 synic = vcpu_to_synic(vcpu);
118 return (synic->active) ? synic : NULL;
121 static void synic_clear_sint_msg_pending(struct kvm_vcpu_hv_synic *synic,
124 struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
127 struct hv_message *msg;
128 struct hv_message_page *msg_page;
130 gpa = synic->msg_page & PAGE_MASK;
131 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
132 if (is_error_page(page)) {
133 vcpu_err(vcpu, "Hyper-V SynIC can't get msg page, gpa 0x%llx\n",
137 msg_page = kmap_atomic(page);
139 msg = &msg_page->sint_message[sint];
140 msg->header.message_flags.msg_pending = 0;
142 kunmap_atomic(msg_page);
143 kvm_release_page_dirty(page);
144 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
147 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
149 struct kvm *kvm = vcpu->kvm;
150 struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
151 struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
152 struct kvm_vcpu_hv_stimer *stimer;
153 int gsi, idx, stimers_pending;
155 trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
157 if (synic->msg_page & HV_SYNIC_SIMP_ENABLE)
158 synic_clear_sint_msg_pending(synic, sint);
160 /* Try to deliver pending Hyper-V SynIC timers messages */
162 for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
163 stimer = &hv_vcpu->stimer[idx];
164 if (stimer->msg_pending &&
165 (stimer->config & HV_STIMER_ENABLE) &&
166 HV_STIMER_SINT(stimer->config) == sint) {
167 set_bit(stimer->index,
168 hv_vcpu->stimer_pending_bitmap);
173 kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
175 idx = srcu_read_lock(&kvm->irq_srcu);
176 gsi = atomic_read(&synic->sint_to_gsi[sint]);
178 kvm_notify_acked_gsi(kvm, gsi);
179 srcu_read_unlock(&kvm->irq_srcu, idx);
182 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
184 struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
185 struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
187 hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
188 hv_vcpu->exit.u.synic.msr = msr;
189 hv_vcpu->exit.u.synic.control = synic->control;
190 hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
191 hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
193 kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
196 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
197 u32 msr, u64 data, bool host)
199 struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
205 trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
209 case HV_X64_MSR_SCONTROL:
210 synic->control = data;
212 synic_exit(synic, msr);
214 case HV_X64_MSR_SVERSION:
219 synic->version = data;
221 case HV_X64_MSR_SIEFP:
222 if (data & HV_SYNIC_SIEFP_ENABLE)
223 if (kvm_clear_guest(vcpu->kvm,
224 data & PAGE_MASK, PAGE_SIZE)) {
228 synic->evt_page = data;
230 synic_exit(synic, msr);
232 case HV_X64_MSR_SIMP:
233 if (data & HV_SYNIC_SIMP_ENABLE)
234 if (kvm_clear_guest(vcpu->kvm,
235 data & PAGE_MASK, PAGE_SIZE)) {
239 synic->msg_page = data;
241 synic_exit(synic, msr);
243 case HV_X64_MSR_EOM: {
246 for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
247 kvm_hv_notify_acked_sint(vcpu, i);
250 case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
251 ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
260 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata)
269 case HV_X64_MSR_SCONTROL:
270 *pdata = synic->control;
272 case HV_X64_MSR_SVERSION:
273 *pdata = synic->version;
275 case HV_X64_MSR_SIEFP:
276 *pdata = synic->evt_page;
278 case HV_X64_MSR_SIMP:
279 *pdata = synic->msg_page;
284 case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
285 *pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
294 static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
296 struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
297 struct kvm_lapic_irq irq;
300 if (sint >= ARRAY_SIZE(synic->sint))
303 vector = synic_get_sint_vector(synic_read_sint(synic, sint));
307 memset(&irq, 0, sizeof(irq));
308 irq.dest_id = kvm_apic_id(vcpu->arch.apic);
309 irq.dest_mode = APIC_DEST_PHYSICAL;
310 irq.delivery_mode = APIC_DM_FIXED;
314 ret = kvm_irq_delivery_to_apic(vcpu->kvm, NULL, &irq, NULL);
315 trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
319 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vcpu_id, u32 sint)
321 struct kvm_vcpu_hv_synic *synic;
323 synic = synic_get(kvm, vcpu_id);
327 return synic_set_irq(synic, sint);
330 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
332 struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
335 trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
337 for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
338 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
339 kvm_hv_notify_acked_sint(vcpu, i);
342 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vcpu_id, u32 sint, int gsi)
344 struct kvm_vcpu_hv_synic *synic;
346 synic = synic_get(kvm, vcpu_id);
350 if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
353 atomic_set(&synic->sint_to_gsi[sint], gsi);
357 void kvm_hv_irq_routing_update(struct kvm *kvm)
359 struct kvm_irq_routing_table *irq_rt;
360 struct kvm_kernel_irq_routing_entry *e;
363 irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
364 lockdep_is_held(&kvm->irq_lock));
366 for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
367 hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
368 if (e->type == KVM_IRQ_ROUTING_HV_SINT)
369 kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
370 e->hv_sint.sint, gsi);
375 static void synic_init(struct kvm_vcpu_hv_synic *synic)
379 memset(synic, 0, sizeof(*synic));
380 synic->version = HV_SYNIC_VERSION_1;
381 for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
382 atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
383 atomic_set(&synic->sint_to_gsi[i], -1);
387 static u64 get_time_ref_counter(struct kvm *kvm)
389 struct kvm_hv *hv = &kvm->arch.hyperv;
390 struct kvm_vcpu *vcpu;
394 * The guest has not set up the TSC page or the clock isn't
395 * stable, fall back to get_kvmclock_ns.
397 if (!hv->tsc_ref.tsc_sequence)
398 return div_u64(get_kvmclock_ns(kvm), 100);
400 vcpu = kvm_get_vcpu(kvm, 0);
401 tsc = kvm_read_l1_tsc(vcpu, rdtsc());
402 return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64)
403 + hv->tsc_ref.tsc_offset;
406 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
409 struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
411 set_bit(stimer->index,
412 vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
413 kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
418 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
420 struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
422 trace_kvm_hv_stimer_cleanup(stimer_to_vcpu(stimer)->vcpu_id,
425 hrtimer_cancel(&stimer->timer);
426 clear_bit(stimer->index,
427 vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
428 stimer->msg_pending = false;
429 stimer->exp_time = 0;
432 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
434 struct kvm_vcpu_hv_stimer *stimer;
436 stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
437 trace_kvm_hv_stimer_callback(stimer_to_vcpu(stimer)->vcpu_id,
439 stimer_mark_pending(stimer, true);
441 return HRTIMER_NORESTART;
445 * stimer_start() assumptions:
446 * a) stimer->count is not equal to 0
447 * b) stimer->config has HV_STIMER_ENABLE flag
449 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
454 time_now = get_time_ref_counter(stimer_to_vcpu(stimer)->kvm);
455 ktime_now = ktime_get();
457 if (stimer->config & HV_STIMER_PERIODIC) {
458 if (stimer->exp_time) {
459 if (time_now >= stimer->exp_time) {
462 div64_u64_rem(time_now - stimer->exp_time,
463 stimer->count, &remainder);
465 time_now + (stimer->count - remainder);
468 stimer->exp_time = time_now + stimer->count;
470 trace_kvm_hv_stimer_start_periodic(
471 stimer_to_vcpu(stimer)->vcpu_id,
473 time_now, stimer->exp_time);
475 hrtimer_start(&stimer->timer,
476 ktime_add_ns(ktime_now,
477 100 * (stimer->exp_time - time_now)),
481 stimer->exp_time = stimer->count;
482 if (time_now >= stimer->count) {
484 * Expire timer according to Hypervisor Top-Level Functional
485 * specification v4(15.3.1):
486 * "If a one shot is enabled and the specified count is in
487 * the past, it will expire immediately."
489 stimer_mark_pending(stimer, false);
493 trace_kvm_hv_stimer_start_one_shot(stimer_to_vcpu(stimer)->vcpu_id,
495 time_now, stimer->count);
497 hrtimer_start(&stimer->timer,
498 ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)),
503 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
506 trace_kvm_hv_stimer_set_config(stimer_to_vcpu(stimer)->vcpu_id,
507 stimer->index, config, host);
509 stimer_cleanup(stimer);
510 if ((stimer->config & HV_STIMER_ENABLE) && HV_STIMER_SINT(config) == 0)
511 config &= ~HV_STIMER_ENABLE;
512 stimer->config = config;
513 stimer_mark_pending(stimer, false);
517 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
520 trace_kvm_hv_stimer_set_count(stimer_to_vcpu(stimer)->vcpu_id,
521 stimer->index, count, host);
523 stimer_cleanup(stimer);
524 stimer->count = count;
525 if (stimer->count == 0)
526 stimer->config &= ~HV_STIMER_ENABLE;
527 else if (stimer->config & HV_STIMER_AUTOENABLE)
528 stimer->config |= HV_STIMER_ENABLE;
529 stimer_mark_pending(stimer, false);
533 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
535 *pconfig = stimer->config;
539 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
541 *pcount = stimer->count;
545 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
546 struct hv_message *src_msg)
548 struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
551 struct hv_message *dst_msg;
553 struct hv_message_page *msg_page;
555 if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
558 gpa = synic->msg_page & PAGE_MASK;
559 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
560 if (is_error_page(page))
563 msg_page = kmap_atomic(page);
564 dst_msg = &msg_page->sint_message[sint];
565 if (sync_cmpxchg(&dst_msg->header.message_type, HVMSG_NONE,
566 src_msg->header.message_type) != HVMSG_NONE) {
567 dst_msg->header.message_flags.msg_pending = 1;
570 memcpy(&dst_msg->u.payload, &src_msg->u.payload,
571 src_msg->header.payload_size);
572 dst_msg->header.message_type = src_msg->header.message_type;
573 dst_msg->header.payload_size = src_msg->header.payload_size;
574 r = synic_set_irq(synic, sint);
580 kunmap_atomic(msg_page);
581 kvm_release_page_dirty(page);
582 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
586 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
588 struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
589 struct hv_message *msg = &stimer->msg;
590 struct hv_timer_message_payload *payload =
591 (struct hv_timer_message_payload *)&msg->u.payload;
593 payload->expiration_time = stimer->exp_time;
594 payload->delivery_time = get_time_ref_counter(vcpu->kvm);
595 return synic_deliver_msg(vcpu_to_synic(vcpu),
596 HV_STIMER_SINT(stimer->config), msg);
599 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
603 stimer->msg_pending = true;
604 r = stimer_send_msg(stimer);
605 trace_kvm_hv_stimer_expiration(stimer_to_vcpu(stimer)->vcpu_id,
608 stimer->msg_pending = false;
609 if (!(stimer->config & HV_STIMER_PERIODIC))
610 stimer->config &= ~HV_STIMER_ENABLE;
614 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
616 struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
617 struct kvm_vcpu_hv_stimer *stimer;
618 u64 time_now, exp_time;
621 for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
622 if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
623 stimer = &hv_vcpu->stimer[i];
624 if (stimer->config & HV_STIMER_ENABLE) {
625 exp_time = stimer->exp_time;
629 get_time_ref_counter(vcpu->kvm);
630 if (time_now >= exp_time)
631 stimer_expiration(stimer);
634 if ((stimer->config & HV_STIMER_ENABLE) &&
636 stimer_start(stimer);
638 stimer_cleanup(stimer);
643 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
645 struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
648 for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
649 stimer_cleanup(&hv_vcpu->stimer[i]);
652 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
654 struct hv_message *msg = &stimer->msg;
655 struct hv_timer_message_payload *payload =
656 (struct hv_timer_message_payload *)&msg->u.payload;
658 memset(&msg->header, 0, sizeof(msg->header));
659 msg->header.message_type = HVMSG_TIMER_EXPIRED;
660 msg->header.payload_size = sizeof(*payload);
662 payload->timer_index = stimer->index;
663 payload->expiration_time = 0;
664 payload->delivery_time = 0;
667 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
669 memset(stimer, 0, sizeof(*stimer));
670 stimer->index = timer_index;
671 hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
672 stimer->timer.function = stimer_timer_callback;
673 stimer_prepare_msg(stimer);
676 void kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
678 struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
681 synic_init(&hv_vcpu->synic);
683 bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
684 for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
685 stimer_init(&hv_vcpu->stimer[i], i);
688 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu)
691 * Hyper-V SynIC auto EOI SINT's are
692 * not compatible with APICV, so deactivate APICV
694 kvm_vcpu_deactivate_apicv(vcpu);
695 vcpu_to_synic(vcpu)->active = true;
699 static bool kvm_hv_msr_partition_wide(u32 msr)
704 case HV_X64_MSR_GUEST_OS_ID:
705 case HV_X64_MSR_HYPERCALL:
706 case HV_X64_MSR_REFERENCE_TSC:
707 case HV_X64_MSR_TIME_REF_COUNT:
708 case HV_X64_MSR_CRASH_CTL:
709 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
710 case HV_X64_MSR_RESET:
718 static int kvm_hv_msr_get_crash_data(struct kvm_vcpu *vcpu,
719 u32 index, u64 *pdata)
721 struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
723 if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param)))
726 *pdata = hv->hv_crash_param[index];
730 static int kvm_hv_msr_get_crash_ctl(struct kvm_vcpu *vcpu, u64 *pdata)
732 struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
734 *pdata = hv->hv_crash_ctl;
738 static int kvm_hv_msr_set_crash_ctl(struct kvm_vcpu *vcpu, u64 data, bool host)
740 struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
743 hv->hv_crash_ctl = data & HV_X64_MSR_CRASH_CTL_NOTIFY;
745 if (!host && (data & HV_X64_MSR_CRASH_CTL_NOTIFY)) {
747 vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
748 hv->hv_crash_param[0],
749 hv->hv_crash_param[1],
750 hv->hv_crash_param[2],
751 hv->hv_crash_param[3],
752 hv->hv_crash_param[4]);
754 /* Send notification about crash to user space */
755 kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
761 static int kvm_hv_msr_set_crash_data(struct kvm_vcpu *vcpu,
764 struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
766 if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param)))
769 hv->hv_crash_param[index] = data;
774 * The kvmclock and Hyper-V TSC page use similar formulas, and converting
775 * between them is possible:
778 * nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
782 * nsec/100 = ticks * scale / 2^64 + offset
784 * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
785 * By dividing the kvmclock formula by 100 and equating what's left we get:
786 * ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
787 * scale / 2^64 = tsc_to_system_mul * 2^(tsc_shift-32) / 100
788 * scale = tsc_to_system_mul * 2^(32+tsc_shift) / 100
790 * Now expand the kvmclock formula and divide by 100:
791 * nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
792 * - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
794 * nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
795 * - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
796 * + system_time / 100
798 * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
799 * nsec/100 = ticks * scale / 2^64
800 * - tsc_timestamp * scale / 2^64
801 * + system_time / 100
803 * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
804 * offset = system_time / 100 - tsc_timestamp * scale / 2^64
806 * These two equivalencies are implemented in this function.
808 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
809 HV_REFERENCE_TSC_PAGE *tsc_ref)
813 if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
817 * check if scale would overflow, if so we use the time ref counter
818 * tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
819 * tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
820 * tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
822 max_mul = 100ull << (32 - hv_clock->tsc_shift);
823 if (hv_clock->tsc_to_system_mul >= max_mul)
827 * Otherwise compute the scale and offset according to the formulas
831 mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
832 hv_clock->tsc_to_system_mul,
835 tsc_ref->tsc_offset = hv_clock->system_time;
836 do_div(tsc_ref->tsc_offset, 100);
837 tsc_ref->tsc_offset -=
838 mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
842 void kvm_hv_setup_tsc_page(struct kvm *kvm,
843 struct pvclock_vcpu_time_info *hv_clock)
845 struct kvm_hv *hv = &kvm->arch.hyperv;
849 BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
850 BUILD_BUG_ON(offsetof(HV_REFERENCE_TSC_PAGE, tsc_sequence) != 0);
852 if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
855 mutex_lock(&kvm->arch.hyperv.hv_lock);
856 if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
859 gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
861 * Because the TSC parameters only vary when there is a
862 * change in the master clock, do not bother with caching.
864 if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
865 &tsc_seq, sizeof(tsc_seq))))
869 * While we're computing and writing the parameters, force the
870 * guest to use the time reference count MSR.
872 hv->tsc_ref.tsc_sequence = 0;
873 if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
874 &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
877 if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
880 /* Ensure sequence is zero before writing the rest of the struct. */
882 if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
886 * Now switch to the TSC page mechanism by writing the sequence.
889 if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
892 /* Write the struct entirely before the non-zero sequence. */
895 hv->tsc_ref.tsc_sequence = tsc_seq;
896 kvm_write_guest(kvm, gfn_to_gpa(gfn),
897 &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence));
899 mutex_unlock(&kvm->arch.hyperv.hv_lock);
902 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
905 struct kvm *kvm = vcpu->kvm;
906 struct kvm_hv *hv = &kvm->arch.hyperv;
909 case HV_X64_MSR_GUEST_OS_ID:
910 hv->hv_guest_os_id = data;
911 /* setting guest os id to zero disables hypercall page */
912 if (!hv->hv_guest_os_id)
913 hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
915 case HV_X64_MSR_HYPERCALL: {
920 /* if guest os id is not set hypercall should remain disabled */
921 if (!hv->hv_guest_os_id)
923 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
924 hv->hv_hypercall = data;
927 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
928 addr = gfn_to_hva(kvm, gfn);
929 if (kvm_is_error_hva(addr))
931 kvm_x86_ops->patch_hypercall(vcpu, instructions);
932 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
933 if (__copy_to_user((void __user *)addr, instructions, 4))
935 hv->hv_hypercall = data;
936 mark_page_dirty(kvm, gfn);
939 case HV_X64_MSR_REFERENCE_TSC:
940 hv->hv_tsc_page = data;
941 if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE)
942 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
944 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
945 return kvm_hv_msr_set_crash_data(vcpu,
946 msr - HV_X64_MSR_CRASH_P0,
948 case HV_X64_MSR_CRASH_CTL:
949 return kvm_hv_msr_set_crash_ctl(vcpu, data, host);
950 case HV_X64_MSR_RESET:
952 vcpu_debug(vcpu, "hyper-v reset requested\n");
953 kvm_make_request(KVM_REQ_HV_RESET, vcpu);
957 vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
964 /* Calculate cpu time spent by current task in 100ns units */
965 static u64 current_task_runtime_100ns(void)
967 cputime_t utime, stime;
969 task_cputime_adjusted(current, &utime, &stime);
970 return div_u64(cputime_to_nsecs(utime + stime), 100);
973 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
975 struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
978 case HV_X64_MSR_APIC_ASSIST_PAGE: {
982 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
984 if (kvm_lapic_enable_pv_eoi(vcpu, 0))
988 gfn = data >> HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT;
989 addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
990 if (kvm_is_error_hva(addr))
992 if (__clear_user((void __user *)addr, PAGE_SIZE))
995 kvm_vcpu_mark_page_dirty(vcpu, gfn);
996 if (kvm_lapic_enable_pv_eoi(vcpu,
997 gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
1001 case HV_X64_MSR_EOI:
1002 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1003 case HV_X64_MSR_ICR:
1004 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1005 case HV_X64_MSR_TPR:
1006 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1007 case HV_X64_MSR_VP_RUNTIME:
1010 hv->runtime_offset = data - current_task_runtime_100ns();
1012 case HV_X64_MSR_SCONTROL:
1013 case HV_X64_MSR_SVERSION:
1014 case HV_X64_MSR_SIEFP:
1015 case HV_X64_MSR_SIMP:
1016 case HV_X64_MSR_EOM:
1017 case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1018 return synic_set_msr(vcpu_to_synic(vcpu), msr, data, host);
1019 case HV_X64_MSR_STIMER0_CONFIG:
1020 case HV_X64_MSR_STIMER1_CONFIG:
1021 case HV_X64_MSR_STIMER2_CONFIG:
1022 case HV_X64_MSR_STIMER3_CONFIG: {
1023 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1025 return stimer_set_config(vcpu_to_stimer(vcpu, timer_index),
1028 case HV_X64_MSR_STIMER0_COUNT:
1029 case HV_X64_MSR_STIMER1_COUNT:
1030 case HV_X64_MSR_STIMER2_COUNT:
1031 case HV_X64_MSR_STIMER3_COUNT: {
1032 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1034 return stimer_set_count(vcpu_to_stimer(vcpu, timer_index),
1038 vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
1046 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1049 struct kvm *kvm = vcpu->kvm;
1050 struct kvm_hv *hv = &kvm->arch.hyperv;
1053 case HV_X64_MSR_GUEST_OS_ID:
1054 data = hv->hv_guest_os_id;
1056 case HV_X64_MSR_HYPERCALL:
1057 data = hv->hv_hypercall;
1059 case HV_X64_MSR_TIME_REF_COUNT:
1060 data = get_time_ref_counter(kvm);
1062 case HV_X64_MSR_REFERENCE_TSC:
1063 data = hv->hv_tsc_page;
1065 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1066 return kvm_hv_msr_get_crash_data(vcpu,
1067 msr - HV_X64_MSR_CRASH_P0,
1069 case HV_X64_MSR_CRASH_CTL:
1070 return kvm_hv_msr_get_crash_ctl(vcpu, pdata);
1071 case HV_X64_MSR_RESET:
1075 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1083 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1086 struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
1089 case HV_X64_MSR_VP_INDEX: {
1093 kvm_for_each_vcpu(r, v, vcpu->kvm) {
1101 case HV_X64_MSR_EOI:
1102 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1103 case HV_X64_MSR_ICR:
1104 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1105 case HV_X64_MSR_TPR:
1106 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1107 case HV_X64_MSR_APIC_ASSIST_PAGE:
1108 data = hv->hv_vapic;
1110 case HV_X64_MSR_VP_RUNTIME:
1111 data = current_task_runtime_100ns() + hv->runtime_offset;
1113 case HV_X64_MSR_SCONTROL:
1114 case HV_X64_MSR_SVERSION:
1115 case HV_X64_MSR_SIEFP:
1116 case HV_X64_MSR_SIMP:
1117 case HV_X64_MSR_EOM:
1118 case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1119 return synic_get_msr(vcpu_to_synic(vcpu), msr, pdata);
1120 case HV_X64_MSR_STIMER0_CONFIG:
1121 case HV_X64_MSR_STIMER1_CONFIG:
1122 case HV_X64_MSR_STIMER2_CONFIG:
1123 case HV_X64_MSR_STIMER3_CONFIG: {
1124 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1126 return stimer_get_config(vcpu_to_stimer(vcpu, timer_index),
1129 case HV_X64_MSR_STIMER0_COUNT:
1130 case HV_X64_MSR_STIMER1_COUNT:
1131 case HV_X64_MSR_STIMER2_COUNT:
1132 case HV_X64_MSR_STIMER3_COUNT: {
1133 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1135 return stimer_get_count(vcpu_to_stimer(vcpu, timer_index),
1139 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1146 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1148 if (kvm_hv_msr_partition_wide(msr)) {
1151 mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1152 r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1153 mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1156 return kvm_hv_set_msr(vcpu, msr, data, host);
1159 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1161 if (kvm_hv_msr_partition_wide(msr)) {
1164 mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1165 r = kvm_hv_get_msr_pw(vcpu, msr, pdata);
1166 mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1169 return kvm_hv_get_msr(vcpu, msr, pdata);
1172 bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1174 return READ_ONCE(kvm->arch.hyperv.hv_hypercall) & HV_X64_MSR_HYPERCALL_ENABLE;
1177 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
1181 longmode = is_64_bit_mode(vcpu);
1183 kvm_register_write(vcpu, VCPU_REGS_RAX, result);
1185 kvm_register_write(vcpu, VCPU_REGS_RDX, result >> 32);
1186 kvm_register_write(vcpu, VCPU_REGS_RAX, result & 0xffffffff);
1190 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
1192 struct kvm_run *run = vcpu->run;
1194 kvm_hv_hypercall_set_result(vcpu, run->hyperv.u.hcall.result);
1198 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
1200 u64 param, ingpa, outgpa, ret;
1201 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
1202 bool fast, longmode;
1205 * hypercall generates UD from non zero cpl and real mode
1208 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
1209 kvm_queue_exception(vcpu, UD_VECTOR);
1213 longmode = is_64_bit_mode(vcpu);
1216 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
1217 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
1218 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
1219 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
1220 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
1221 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
1223 #ifdef CONFIG_X86_64
1225 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
1226 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
1227 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
1231 code = param & 0xffff;
1232 fast = (param >> 16) & 0x1;
1233 rep_cnt = (param >> 32) & 0xfff;
1234 rep_idx = (param >> 48) & 0xfff;
1236 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
1238 /* Hypercall continuation is not supported yet */
1239 if (rep_cnt || rep_idx) {
1240 res = HV_STATUS_INVALID_HYPERCALL_CODE;
1245 case HVCALL_NOTIFY_LONG_SPIN_WAIT:
1246 kvm_vcpu_on_spin(vcpu);
1248 case HVCALL_POST_MESSAGE:
1249 case HVCALL_SIGNAL_EVENT:
1250 /* don't bother userspace if it has no way to handle it */
1251 if (!vcpu_to_synic(vcpu)->active) {
1252 res = HV_STATUS_INVALID_HYPERCALL_CODE;
1255 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
1256 vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
1257 vcpu->run->hyperv.u.hcall.input = param;
1258 vcpu->run->hyperv.u.hcall.params[0] = ingpa;
1259 vcpu->run->hyperv.u.hcall.params[1] = outgpa;
1260 vcpu->arch.complete_userspace_io =
1261 kvm_hv_hypercall_complete_userspace;
1264 res = HV_STATUS_INVALID_HYPERCALL_CODE;
1269 ret = res | (((u64)rep_done & 0xfff) << 32);
1270 kvm_hv_hypercall_set_result(vcpu, ret);