1 // SPDX-License-Identifier: GPL-2.0-only
3 * KVM Microsoft Hyper-V emulation
5 * derived from arch/x86/kvm/x86.c
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>
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>
26 #include <linux/kvm_host.h>
27 #include <linux/highmem.h>
28 #include <linux/sched/cputime.h>
29 #include <linux/eventfd.h>
31 #include <asm/apicdef.h>
32 #include <trace/events/kvm.h>
36 #define KVM_HV_MAX_SPARSE_VCPU_SET_BITS DIV_ROUND_UP(KVM_MAX_VCPUS, 64)
38 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
41 static inline u64 synic_read_sint(struct kvm_vcpu_hv_synic *synic, int sint)
43 return atomic64_read(&synic->sint[sint]);
46 static inline int synic_get_sint_vector(u64 sint_value)
48 if (sint_value & HV_SYNIC_SINT_MASKED)
50 return sint_value & HV_SYNIC_SINT_VECTOR_MASK;
53 static bool synic_has_vector_connected(struct kvm_vcpu_hv_synic *synic,
58 for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
59 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
65 static bool synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic *synic,
71 for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
72 sint_value = synic_read_sint(synic, i);
73 if (synic_get_sint_vector(sint_value) == vector &&
74 sint_value & HV_SYNIC_SINT_AUTO_EOI)
80 static void synic_update_vector(struct kvm_vcpu_hv_synic *synic,
83 if (vector < HV_SYNIC_FIRST_VALID_VECTOR)
86 if (synic_has_vector_connected(synic, vector))
87 __set_bit(vector, synic->vec_bitmap);
89 __clear_bit(vector, synic->vec_bitmap);
91 if (synic_has_vector_auto_eoi(synic, vector))
92 __set_bit(vector, synic->auto_eoi_bitmap);
94 __clear_bit(vector, synic->auto_eoi_bitmap);
97 static int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint,
100 int vector, old_vector;
103 vector = data & HV_SYNIC_SINT_VECTOR_MASK;
104 masked = data & HV_SYNIC_SINT_MASKED;
107 * Valid vectors are 16-255, however, nested Hyper-V attempts to write
108 * default '0x10000' value on boot and this should not #GP. We need to
109 * allow zero-initing the register from host as well.
111 if (vector < HV_SYNIC_FIRST_VALID_VECTOR && !host && !masked)
114 * Guest may configure multiple SINTs to use the same vector, so
115 * we maintain a bitmap of vectors handled by synic, and a
116 * bitmap of vectors with auto-eoi behavior. The bitmaps are
117 * updated here, and atomically queried on fast paths.
119 old_vector = synic_read_sint(synic, sint) & HV_SYNIC_SINT_VECTOR_MASK;
121 atomic64_set(&synic->sint[sint], data);
123 synic_update_vector(synic, old_vector);
125 synic_update_vector(synic, vector);
127 /* Load SynIC vectors into EOI exit bitmap */
128 kvm_make_request(KVM_REQ_SCAN_IOAPIC, synic_to_vcpu(synic));
132 static struct kvm_vcpu *get_vcpu_by_vpidx(struct kvm *kvm, u32 vpidx)
134 struct kvm_vcpu *vcpu = NULL;
137 if (vpidx >= KVM_MAX_VCPUS)
140 vcpu = kvm_get_vcpu(kvm, vpidx);
141 if (vcpu && vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
143 kvm_for_each_vcpu(i, vcpu, kvm)
144 if (vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
149 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vpidx)
151 struct kvm_vcpu *vcpu;
152 struct kvm_vcpu_hv_synic *synic;
154 vcpu = get_vcpu_by_vpidx(kvm, vpidx);
157 synic = vcpu_to_synic(vcpu);
158 return (synic->active) ? synic : NULL;
161 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
163 struct kvm *kvm = vcpu->kvm;
164 struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
165 struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
166 struct kvm_vcpu_hv_stimer *stimer;
169 trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
171 /* Try to deliver pending Hyper-V SynIC timers messages */
172 for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
173 stimer = &hv_vcpu->stimer[idx];
174 if (stimer->msg_pending && stimer->config.enable &&
175 !stimer->config.direct_mode &&
176 stimer->config.sintx == sint)
177 stimer_mark_pending(stimer, false);
180 idx = srcu_read_lock(&kvm->irq_srcu);
181 gsi = atomic_read(&synic->sint_to_gsi[sint]);
183 kvm_notify_acked_gsi(kvm, gsi);
184 srcu_read_unlock(&kvm->irq_srcu, idx);
187 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
189 struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
190 struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
192 hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
193 hv_vcpu->exit.u.synic.msr = msr;
194 hv_vcpu->exit.u.synic.control = synic->control;
195 hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
196 hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
198 kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
201 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
202 u32 msr, u64 data, bool host)
204 struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
207 if (!synic->active && !host)
210 trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
214 case HV_X64_MSR_SCONTROL:
215 synic->control = data;
217 synic_exit(synic, msr);
219 case HV_X64_MSR_SVERSION:
224 synic->version = data;
226 case HV_X64_MSR_SIEFP:
227 if ((data & HV_SYNIC_SIEFP_ENABLE) && !host &&
228 !synic->dont_zero_synic_pages)
229 if (kvm_clear_guest(vcpu->kvm,
230 data & PAGE_MASK, PAGE_SIZE)) {
234 synic->evt_page = data;
236 synic_exit(synic, msr);
238 case HV_X64_MSR_SIMP:
239 if ((data & HV_SYNIC_SIMP_ENABLE) && !host &&
240 !synic->dont_zero_synic_pages)
241 if (kvm_clear_guest(vcpu->kvm,
242 data & PAGE_MASK, PAGE_SIZE)) {
246 synic->msg_page = data;
248 synic_exit(synic, msr);
250 case HV_X64_MSR_EOM: {
253 for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
254 kvm_hv_notify_acked_sint(vcpu, i);
257 case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
258 ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
267 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata,
272 if (!synic->active && !host)
277 case HV_X64_MSR_SCONTROL:
278 *pdata = synic->control;
280 case HV_X64_MSR_SVERSION:
281 *pdata = synic->version;
283 case HV_X64_MSR_SIEFP:
284 *pdata = synic->evt_page;
286 case HV_X64_MSR_SIMP:
287 *pdata = synic->msg_page;
292 case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
293 *pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
302 static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
304 struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
305 struct kvm_lapic_irq irq;
308 if (sint >= ARRAY_SIZE(synic->sint))
311 vector = synic_get_sint_vector(synic_read_sint(synic, sint));
315 memset(&irq, 0, sizeof(irq));
316 irq.shorthand = APIC_DEST_SELF;
317 irq.dest_mode = APIC_DEST_PHYSICAL;
318 irq.delivery_mode = APIC_DM_FIXED;
322 ret = kvm_irq_delivery_to_apic(vcpu->kvm, vcpu->arch.apic, &irq, NULL);
323 trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
327 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vpidx, u32 sint)
329 struct kvm_vcpu_hv_synic *synic;
331 synic = synic_get(kvm, vpidx);
335 return synic_set_irq(synic, sint);
338 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
340 struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
343 trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
345 for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
346 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
347 kvm_hv_notify_acked_sint(vcpu, i);
350 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi)
352 struct kvm_vcpu_hv_synic *synic;
354 synic = synic_get(kvm, vpidx);
358 if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
361 atomic_set(&synic->sint_to_gsi[sint], gsi);
365 void kvm_hv_irq_routing_update(struct kvm *kvm)
367 struct kvm_irq_routing_table *irq_rt;
368 struct kvm_kernel_irq_routing_entry *e;
371 irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
372 lockdep_is_held(&kvm->irq_lock));
374 for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
375 hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
376 if (e->type == KVM_IRQ_ROUTING_HV_SINT)
377 kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
378 e->hv_sint.sint, gsi);
383 static void synic_init(struct kvm_vcpu_hv_synic *synic)
387 memset(synic, 0, sizeof(*synic));
388 synic->version = HV_SYNIC_VERSION_1;
389 for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
390 atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
391 atomic_set(&synic->sint_to_gsi[i], -1);
395 static u64 get_time_ref_counter(struct kvm *kvm)
397 struct kvm_hv *hv = &kvm->arch.hyperv;
398 struct kvm_vcpu *vcpu;
402 * The guest has not set up the TSC page or the clock isn't
403 * stable, fall back to get_kvmclock_ns.
405 if (!hv->tsc_ref.tsc_sequence)
406 return div_u64(get_kvmclock_ns(kvm), 100);
408 vcpu = kvm_get_vcpu(kvm, 0);
409 tsc = kvm_read_l1_tsc(vcpu, rdtsc());
410 return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64)
411 + hv->tsc_ref.tsc_offset;
414 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
417 struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
419 set_bit(stimer->index,
420 vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
421 kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
426 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
428 struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
430 trace_kvm_hv_stimer_cleanup(stimer_to_vcpu(stimer)->vcpu_id,
433 hrtimer_cancel(&stimer->timer);
434 clear_bit(stimer->index,
435 vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
436 stimer->msg_pending = false;
437 stimer->exp_time = 0;
440 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
442 struct kvm_vcpu_hv_stimer *stimer;
444 stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
445 trace_kvm_hv_stimer_callback(stimer_to_vcpu(stimer)->vcpu_id,
447 stimer_mark_pending(stimer, true);
449 return HRTIMER_NORESTART;
453 * stimer_start() assumptions:
454 * a) stimer->count is not equal to 0
455 * b) stimer->config has HV_STIMER_ENABLE flag
457 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
462 time_now = get_time_ref_counter(stimer_to_vcpu(stimer)->kvm);
463 ktime_now = ktime_get();
465 if (stimer->config.periodic) {
466 if (stimer->exp_time) {
467 if (time_now >= stimer->exp_time) {
470 div64_u64_rem(time_now - stimer->exp_time,
471 stimer->count, &remainder);
473 time_now + (stimer->count - remainder);
476 stimer->exp_time = time_now + stimer->count;
478 trace_kvm_hv_stimer_start_periodic(
479 stimer_to_vcpu(stimer)->vcpu_id,
481 time_now, stimer->exp_time);
483 hrtimer_start(&stimer->timer,
484 ktime_add_ns(ktime_now,
485 100 * (stimer->exp_time - time_now)),
489 stimer->exp_time = stimer->count;
490 if (time_now >= stimer->count) {
492 * Expire timer according to Hypervisor Top-Level Functional
493 * specification v4(15.3.1):
494 * "If a one shot is enabled and the specified count is in
495 * the past, it will expire immediately."
497 stimer_mark_pending(stimer, false);
501 trace_kvm_hv_stimer_start_one_shot(stimer_to_vcpu(stimer)->vcpu_id,
503 time_now, stimer->count);
505 hrtimer_start(&stimer->timer,
506 ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)),
511 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
514 union hv_stimer_config new_config = {.as_uint64 = config},
515 old_config = {.as_uint64 = stimer->config.as_uint64};
517 trace_kvm_hv_stimer_set_config(stimer_to_vcpu(stimer)->vcpu_id,
518 stimer->index, config, host);
520 stimer_cleanup(stimer);
521 if (old_config.enable &&
522 !new_config.direct_mode && new_config.sintx == 0)
523 new_config.enable = 0;
524 stimer->config.as_uint64 = new_config.as_uint64;
526 if (stimer->config.enable)
527 stimer_mark_pending(stimer, false);
532 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
535 trace_kvm_hv_stimer_set_count(stimer_to_vcpu(stimer)->vcpu_id,
536 stimer->index, count, host);
538 stimer_cleanup(stimer);
539 stimer->count = count;
540 if (stimer->count == 0)
541 stimer->config.enable = 0;
542 else if (stimer->config.auto_enable)
543 stimer->config.enable = 1;
545 if (stimer->config.enable)
546 stimer_mark_pending(stimer, false);
551 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
553 *pconfig = stimer->config.as_uint64;
557 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
559 *pcount = stimer->count;
563 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
564 struct hv_message *src_msg, bool no_retry)
566 struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
567 int msg_off = offsetof(struct hv_message_page, sint_message[sint]);
569 struct hv_message_header hv_hdr;
572 if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
575 msg_page_gfn = synic->msg_page >> PAGE_SHIFT;
578 * Strictly following the spec-mandated ordering would assume setting
579 * .msg_pending before checking .message_type. However, this function
580 * is only called in vcpu context so the entire update is atomic from
581 * guest POV and thus the exact order here doesn't matter.
583 r = kvm_vcpu_read_guest_page(vcpu, msg_page_gfn, &hv_hdr.message_type,
584 msg_off + offsetof(struct hv_message,
585 header.message_type),
586 sizeof(hv_hdr.message_type));
590 if (hv_hdr.message_type != HVMSG_NONE) {
594 hv_hdr.message_flags.msg_pending = 1;
595 r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn,
596 &hv_hdr.message_flags,
598 offsetof(struct hv_message,
599 header.message_flags),
600 sizeof(hv_hdr.message_flags));
606 r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn, src_msg, msg_off,
607 sizeof(src_msg->header) +
608 src_msg->header.payload_size);
612 r = synic_set_irq(synic, sint);
620 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
622 struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
623 struct hv_message *msg = &stimer->msg;
624 struct hv_timer_message_payload *payload =
625 (struct hv_timer_message_payload *)&msg->u.payload;
628 * To avoid piling up periodic ticks, don't retry message
629 * delivery for them (within "lazy" lost ticks policy).
631 bool no_retry = stimer->config.periodic;
633 payload->expiration_time = stimer->exp_time;
634 payload->delivery_time = get_time_ref_counter(vcpu->kvm);
635 return synic_deliver_msg(vcpu_to_synic(vcpu),
636 stimer->config.sintx, msg,
640 static int stimer_notify_direct(struct kvm_vcpu_hv_stimer *stimer)
642 struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
643 struct kvm_lapic_irq irq = {
644 .delivery_mode = APIC_DM_FIXED,
645 .vector = stimer->config.apic_vector
648 return !kvm_apic_set_irq(vcpu, &irq, NULL);
651 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
653 int r, direct = stimer->config.direct_mode;
655 stimer->msg_pending = true;
657 r = stimer_send_msg(stimer);
659 r = stimer_notify_direct(stimer);
660 trace_kvm_hv_stimer_expiration(stimer_to_vcpu(stimer)->vcpu_id,
661 stimer->index, direct, r);
663 stimer->msg_pending = false;
664 if (!(stimer->config.periodic))
665 stimer->config.enable = 0;
669 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
671 struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
672 struct kvm_vcpu_hv_stimer *stimer;
673 u64 time_now, exp_time;
676 for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
677 if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
678 stimer = &hv_vcpu->stimer[i];
679 if (stimer->config.enable) {
680 exp_time = stimer->exp_time;
684 get_time_ref_counter(vcpu->kvm);
685 if (time_now >= exp_time)
686 stimer_expiration(stimer);
689 if ((stimer->config.enable) &&
691 if (!stimer->msg_pending)
692 stimer_start(stimer);
694 stimer_cleanup(stimer);
699 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
701 struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
704 for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
705 stimer_cleanup(&hv_vcpu->stimer[i]);
708 bool kvm_hv_assist_page_enabled(struct kvm_vcpu *vcpu)
710 if (!(vcpu->arch.hyperv.hv_vapic & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE))
712 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
714 EXPORT_SYMBOL_GPL(kvm_hv_assist_page_enabled);
716 bool kvm_hv_get_assist_page(struct kvm_vcpu *vcpu,
717 struct hv_vp_assist_page *assist_page)
719 if (!kvm_hv_assist_page_enabled(vcpu))
721 return !kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data,
722 assist_page, sizeof(*assist_page));
724 EXPORT_SYMBOL_GPL(kvm_hv_get_assist_page);
726 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
728 struct hv_message *msg = &stimer->msg;
729 struct hv_timer_message_payload *payload =
730 (struct hv_timer_message_payload *)&msg->u.payload;
732 memset(&msg->header, 0, sizeof(msg->header));
733 msg->header.message_type = HVMSG_TIMER_EXPIRED;
734 msg->header.payload_size = sizeof(*payload);
736 payload->timer_index = stimer->index;
737 payload->expiration_time = 0;
738 payload->delivery_time = 0;
741 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
743 memset(stimer, 0, sizeof(*stimer));
744 stimer->index = timer_index;
745 hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
746 stimer->timer.function = stimer_timer_callback;
747 stimer_prepare_msg(stimer);
750 void kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
752 struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
755 synic_init(&hv_vcpu->synic);
757 bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
758 for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
759 stimer_init(&hv_vcpu->stimer[i], i);
762 void kvm_hv_vcpu_postcreate(struct kvm_vcpu *vcpu)
764 struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
766 hv_vcpu->vp_index = kvm_vcpu_get_idx(vcpu);
769 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
771 struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
774 * Hyper-V SynIC auto EOI SINT's are
775 * not compatible with APICV, so deactivate APICV
777 kvm_vcpu_deactivate_apicv(vcpu);
778 synic->active = true;
779 synic->dont_zero_synic_pages = dont_zero_synic_pages;
783 static bool kvm_hv_msr_partition_wide(u32 msr)
788 case HV_X64_MSR_GUEST_OS_ID:
789 case HV_X64_MSR_HYPERCALL:
790 case HV_X64_MSR_REFERENCE_TSC:
791 case HV_X64_MSR_TIME_REF_COUNT:
792 case HV_X64_MSR_CRASH_CTL:
793 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
794 case HV_X64_MSR_RESET:
795 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
796 case HV_X64_MSR_TSC_EMULATION_CONTROL:
797 case HV_X64_MSR_TSC_EMULATION_STATUS:
805 static int kvm_hv_msr_get_crash_data(struct kvm_vcpu *vcpu,
806 u32 index, u64 *pdata)
808 struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
810 if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param)))
813 *pdata = hv->hv_crash_param[index];
817 static int kvm_hv_msr_get_crash_ctl(struct kvm_vcpu *vcpu, u64 *pdata)
819 struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
821 *pdata = hv->hv_crash_ctl;
825 static int kvm_hv_msr_set_crash_ctl(struct kvm_vcpu *vcpu, u64 data, bool host)
827 struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
830 hv->hv_crash_ctl = data & HV_CRASH_CTL_CRASH_NOTIFY;
832 if (!host && (data & HV_CRASH_CTL_CRASH_NOTIFY)) {
834 vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
835 hv->hv_crash_param[0],
836 hv->hv_crash_param[1],
837 hv->hv_crash_param[2],
838 hv->hv_crash_param[3],
839 hv->hv_crash_param[4]);
841 /* Send notification about crash to user space */
842 kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
848 static int kvm_hv_msr_set_crash_data(struct kvm_vcpu *vcpu,
851 struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
853 if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param)))
856 hv->hv_crash_param[index] = data;
861 * The kvmclock and Hyper-V TSC page use similar formulas, and converting
862 * between them is possible:
865 * nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
869 * nsec/100 = ticks * scale / 2^64 + offset
871 * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
872 * By dividing the kvmclock formula by 100 and equating what's left we get:
873 * ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
874 * scale / 2^64 = tsc_to_system_mul * 2^(tsc_shift-32) / 100
875 * scale = tsc_to_system_mul * 2^(32+tsc_shift) / 100
877 * Now expand the kvmclock formula and divide by 100:
878 * nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
879 * - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
881 * nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
882 * - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
883 * + system_time / 100
885 * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
886 * nsec/100 = ticks * scale / 2^64
887 * - tsc_timestamp * scale / 2^64
888 * + system_time / 100
890 * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
891 * offset = system_time / 100 - tsc_timestamp * scale / 2^64
893 * These two equivalencies are implemented in this function.
895 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
896 HV_REFERENCE_TSC_PAGE *tsc_ref)
900 if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
904 * check if scale would overflow, if so we use the time ref counter
905 * tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
906 * tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
907 * tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
909 max_mul = 100ull << (32 - hv_clock->tsc_shift);
910 if (hv_clock->tsc_to_system_mul >= max_mul)
914 * Otherwise compute the scale and offset according to the formulas
918 mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
919 hv_clock->tsc_to_system_mul,
922 tsc_ref->tsc_offset = hv_clock->system_time;
923 do_div(tsc_ref->tsc_offset, 100);
924 tsc_ref->tsc_offset -=
925 mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
929 void kvm_hv_setup_tsc_page(struct kvm *kvm,
930 struct pvclock_vcpu_time_info *hv_clock)
932 struct kvm_hv *hv = &kvm->arch.hyperv;
936 BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
937 BUILD_BUG_ON(offsetof(HV_REFERENCE_TSC_PAGE, tsc_sequence) != 0);
939 if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
942 mutex_lock(&kvm->arch.hyperv.hv_lock);
943 if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
946 gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
948 * Because the TSC parameters only vary when there is a
949 * change in the master clock, do not bother with caching.
951 if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
952 &tsc_seq, sizeof(tsc_seq))))
956 * While we're computing and writing the parameters, force the
957 * guest to use the time reference count MSR.
959 hv->tsc_ref.tsc_sequence = 0;
960 if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
961 &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
964 if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
967 /* Ensure sequence is zero before writing the rest of the struct. */
969 if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
973 * Now switch to the TSC page mechanism by writing the sequence.
976 if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
979 /* Write the struct entirely before the non-zero sequence. */
982 hv->tsc_ref.tsc_sequence = tsc_seq;
983 kvm_write_guest(kvm, gfn_to_gpa(gfn),
984 &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence));
986 mutex_unlock(&kvm->arch.hyperv.hv_lock);
989 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
992 struct kvm *kvm = vcpu->kvm;
993 struct kvm_hv *hv = &kvm->arch.hyperv;
996 case HV_X64_MSR_GUEST_OS_ID:
997 hv->hv_guest_os_id = data;
998 /* setting guest os id to zero disables hypercall page */
999 if (!hv->hv_guest_os_id)
1000 hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1002 case HV_X64_MSR_HYPERCALL: {
1007 /* if guest os id is not set hypercall should remain disabled */
1008 if (!hv->hv_guest_os_id)
1010 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1011 hv->hv_hypercall = data;
1014 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1015 addr = gfn_to_hva(kvm, gfn);
1016 if (kvm_is_error_hva(addr))
1018 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1019 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1020 if (__copy_to_user((void __user *)addr, instructions, 4))
1022 hv->hv_hypercall = data;
1023 mark_page_dirty(kvm, gfn);
1026 case HV_X64_MSR_REFERENCE_TSC:
1027 hv->hv_tsc_page = data;
1028 if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE)
1029 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1031 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1032 return kvm_hv_msr_set_crash_data(vcpu,
1033 msr - HV_X64_MSR_CRASH_P0,
1035 case HV_X64_MSR_CRASH_CTL:
1036 return kvm_hv_msr_set_crash_ctl(vcpu, data, host);
1037 case HV_X64_MSR_RESET:
1039 vcpu_debug(vcpu, "hyper-v reset requested\n");
1040 kvm_make_request(KVM_REQ_HV_RESET, vcpu);
1043 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1044 hv->hv_reenlightenment_control = data;
1046 case HV_X64_MSR_TSC_EMULATION_CONTROL:
1047 hv->hv_tsc_emulation_control = data;
1049 case HV_X64_MSR_TSC_EMULATION_STATUS:
1050 hv->hv_tsc_emulation_status = data;
1052 case HV_X64_MSR_TIME_REF_COUNT:
1053 /* read-only, but still ignore it if host-initiated */
1058 vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
1065 /* Calculate cpu time spent by current task in 100ns units */
1066 static u64 current_task_runtime_100ns(void)
1070 task_cputime_adjusted(current, &utime, &stime);
1072 return div_u64(utime + stime, 100);
1075 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1077 struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
1080 case HV_X64_MSR_VP_INDEX: {
1081 struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
1082 int vcpu_idx = kvm_vcpu_get_idx(vcpu);
1083 u32 new_vp_index = (u32)data;
1085 if (!host || new_vp_index >= KVM_MAX_VCPUS)
1088 if (new_vp_index == hv_vcpu->vp_index)
1092 * The VP index is initialized to vcpu_index by
1093 * kvm_hv_vcpu_postcreate so they initially match. Now the
1094 * VP index is changing, adjust num_mismatched_vp_indexes if
1095 * it now matches or no longer matches vcpu_idx.
1097 if (hv_vcpu->vp_index == vcpu_idx)
1098 atomic_inc(&hv->num_mismatched_vp_indexes);
1099 else if (new_vp_index == vcpu_idx)
1100 atomic_dec(&hv->num_mismatched_vp_indexes);
1102 hv_vcpu->vp_index = new_vp_index;
1105 case HV_X64_MSR_VP_ASSIST_PAGE: {
1109 if (!(data & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE)) {
1110 hv_vcpu->hv_vapic = data;
1111 if (kvm_lapic_enable_pv_eoi(vcpu, 0, 0))
1115 gfn = data >> HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT;
1116 addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1117 if (kvm_is_error_hva(addr))
1121 * Clear apic_assist portion of f(struct hv_vp_assist_page
1122 * only, there can be valuable data in the rest which needs
1123 * to be preserved e.g. on migration.
1125 if (__clear_user((void __user *)addr, sizeof(u32)))
1127 hv_vcpu->hv_vapic = data;
1128 kvm_vcpu_mark_page_dirty(vcpu, gfn);
1129 if (kvm_lapic_enable_pv_eoi(vcpu,
1130 gfn_to_gpa(gfn) | KVM_MSR_ENABLED,
1131 sizeof(struct hv_vp_assist_page)))
1135 case HV_X64_MSR_EOI:
1136 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1137 case HV_X64_MSR_ICR:
1138 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1139 case HV_X64_MSR_TPR:
1140 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1141 case HV_X64_MSR_VP_RUNTIME:
1144 hv_vcpu->runtime_offset = data - current_task_runtime_100ns();
1146 case HV_X64_MSR_SCONTROL:
1147 case HV_X64_MSR_SVERSION:
1148 case HV_X64_MSR_SIEFP:
1149 case HV_X64_MSR_SIMP:
1150 case HV_X64_MSR_EOM:
1151 case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1152 return synic_set_msr(vcpu_to_synic(vcpu), msr, data, host);
1153 case HV_X64_MSR_STIMER0_CONFIG:
1154 case HV_X64_MSR_STIMER1_CONFIG:
1155 case HV_X64_MSR_STIMER2_CONFIG:
1156 case HV_X64_MSR_STIMER3_CONFIG: {
1157 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1159 return stimer_set_config(vcpu_to_stimer(vcpu, timer_index),
1162 case HV_X64_MSR_STIMER0_COUNT:
1163 case HV_X64_MSR_STIMER1_COUNT:
1164 case HV_X64_MSR_STIMER2_COUNT:
1165 case HV_X64_MSR_STIMER3_COUNT: {
1166 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1168 return stimer_set_count(vcpu_to_stimer(vcpu, timer_index),
1171 case HV_X64_MSR_TSC_FREQUENCY:
1172 case HV_X64_MSR_APIC_FREQUENCY:
1173 /* read-only, but still ignore it if host-initiated */
1178 vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
1186 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1189 struct kvm *kvm = vcpu->kvm;
1190 struct kvm_hv *hv = &kvm->arch.hyperv;
1193 case HV_X64_MSR_GUEST_OS_ID:
1194 data = hv->hv_guest_os_id;
1196 case HV_X64_MSR_HYPERCALL:
1197 data = hv->hv_hypercall;
1199 case HV_X64_MSR_TIME_REF_COUNT:
1200 data = get_time_ref_counter(kvm);
1202 case HV_X64_MSR_REFERENCE_TSC:
1203 data = hv->hv_tsc_page;
1205 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1206 return kvm_hv_msr_get_crash_data(vcpu,
1207 msr - HV_X64_MSR_CRASH_P0,
1209 case HV_X64_MSR_CRASH_CTL:
1210 return kvm_hv_msr_get_crash_ctl(vcpu, pdata);
1211 case HV_X64_MSR_RESET:
1214 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1215 data = hv->hv_reenlightenment_control;
1217 case HV_X64_MSR_TSC_EMULATION_CONTROL:
1218 data = hv->hv_tsc_emulation_control;
1220 case HV_X64_MSR_TSC_EMULATION_STATUS:
1221 data = hv->hv_tsc_emulation_status;
1224 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1232 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1236 struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
1239 case HV_X64_MSR_VP_INDEX:
1240 data = hv_vcpu->vp_index;
1242 case HV_X64_MSR_EOI:
1243 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1244 case HV_X64_MSR_ICR:
1245 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1246 case HV_X64_MSR_TPR:
1247 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1248 case HV_X64_MSR_VP_ASSIST_PAGE:
1249 data = hv_vcpu->hv_vapic;
1251 case HV_X64_MSR_VP_RUNTIME:
1252 data = current_task_runtime_100ns() + hv_vcpu->runtime_offset;
1254 case HV_X64_MSR_SCONTROL:
1255 case HV_X64_MSR_SVERSION:
1256 case HV_X64_MSR_SIEFP:
1257 case HV_X64_MSR_SIMP:
1258 case HV_X64_MSR_EOM:
1259 case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1260 return synic_get_msr(vcpu_to_synic(vcpu), msr, pdata, host);
1261 case HV_X64_MSR_STIMER0_CONFIG:
1262 case HV_X64_MSR_STIMER1_CONFIG:
1263 case HV_X64_MSR_STIMER2_CONFIG:
1264 case HV_X64_MSR_STIMER3_CONFIG: {
1265 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1267 return stimer_get_config(vcpu_to_stimer(vcpu, timer_index),
1270 case HV_X64_MSR_STIMER0_COUNT:
1271 case HV_X64_MSR_STIMER1_COUNT:
1272 case HV_X64_MSR_STIMER2_COUNT:
1273 case HV_X64_MSR_STIMER3_COUNT: {
1274 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1276 return stimer_get_count(vcpu_to_stimer(vcpu, timer_index),
1279 case HV_X64_MSR_TSC_FREQUENCY:
1280 data = (u64)vcpu->arch.virtual_tsc_khz * 1000;
1282 case HV_X64_MSR_APIC_FREQUENCY:
1283 data = APIC_BUS_FREQUENCY;
1286 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1293 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1295 if (kvm_hv_msr_partition_wide(msr)) {
1298 mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1299 r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1300 mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1303 return kvm_hv_set_msr(vcpu, msr, data, host);
1306 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
1308 if (kvm_hv_msr_partition_wide(msr)) {
1311 mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1312 r = kvm_hv_get_msr_pw(vcpu, msr, pdata);
1313 mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1316 return kvm_hv_get_msr(vcpu, msr, pdata, host);
1319 static __always_inline unsigned long *sparse_set_to_vcpu_mask(
1320 struct kvm *kvm, u64 *sparse_banks, u64 valid_bank_mask,
1321 u64 *vp_bitmap, unsigned long *vcpu_bitmap)
1323 struct kvm_hv *hv = &kvm->arch.hyperv;
1324 struct kvm_vcpu *vcpu;
1325 int i, bank, sbank = 0;
1327 memset(vp_bitmap, 0,
1328 KVM_HV_MAX_SPARSE_VCPU_SET_BITS * sizeof(*vp_bitmap));
1329 for_each_set_bit(bank, (unsigned long *)&valid_bank_mask,
1330 KVM_HV_MAX_SPARSE_VCPU_SET_BITS)
1331 vp_bitmap[bank] = sparse_banks[sbank++];
1333 if (likely(!atomic_read(&hv->num_mismatched_vp_indexes))) {
1334 /* for all vcpus vp_index == vcpu_idx */
1335 return (unsigned long *)vp_bitmap;
1338 bitmap_zero(vcpu_bitmap, KVM_MAX_VCPUS);
1339 kvm_for_each_vcpu(i, vcpu, kvm) {
1340 if (test_bit(vcpu_to_hv_vcpu(vcpu)->vp_index,
1341 (unsigned long *)vp_bitmap))
1342 __set_bit(i, vcpu_bitmap);
1347 static u64 kvm_hv_flush_tlb(struct kvm_vcpu *current_vcpu, u64 ingpa,
1348 u16 rep_cnt, bool ex)
1350 struct kvm *kvm = current_vcpu->kvm;
1351 struct kvm_vcpu_hv *hv_vcpu = ¤t_vcpu->arch.hyperv;
1352 struct hv_tlb_flush_ex flush_ex;
1353 struct hv_tlb_flush flush;
1354 u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1355 DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1356 unsigned long *vcpu_mask;
1357 u64 valid_bank_mask;
1358 u64 sparse_banks[64];
1359 int sparse_banks_len;
1363 if (unlikely(kvm_read_guest(kvm, ingpa, &flush, sizeof(flush))))
1364 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1366 trace_kvm_hv_flush_tlb(flush.processor_mask,
1367 flush.address_space, flush.flags);
1369 valid_bank_mask = BIT_ULL(0);
1370 sparse_banks[0] = flush.processor_mask;
1373 * Work around possible WS2012 bug: it sends hypercalls
1374 * with processor_mask = 0x0 and HV_FLUSH_ALL_PROCESSORS clear,
1375 * while also expecting us to flush something and crashing if
1376 * we don't. Let's treat processor_mask == 0 same as
1377 * HV_FLUSH_ALL_PROCESSORS.
1379 all_cpus = (flush.flags & HV_FLUSH_ALL_PROCESSORS) ||
1380 flush.processor_mask == 0;
1382 if (unlikely(kvm_read_guest(kvm, ingpa, &flush_ex,
1384 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1386 trace_kvm_hv_flush_tlb_ex(flush_ex.hv_vp_set.valid_bank_mask,
1387 flush_ex.hv_vp_set.format,
1388 flush_ex.address_space,
1391 valid_bank_mask = flush_ex.hv_vp_set.valid_bank_mask;
1392 all_cpus = flush_ex.hv_vp_set.format !=
1393 HV_GENERIC_SET_SPARSE_4K;
1396 bitmap_weight((unsigned long *)&valid_bank_mask, 64) *
1397 sizeof(sparse_banks[0]);
1399 if (!sparse_banks_len && !all_cpus)
1404 ingpa + offsetof(struct hv_tlb_flush_ex,
1405 hv_vp_set.bank_contents),
1408 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1411 cpumask_clear(&hv_vcpu->tlb_flush);
1413 vcpu_mask = all_cpus ? NULL :
1414 sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1415 vp_bitmap, vcpu_bitmap);
1418 * vcpu->arch.cr3 may not be up-to-date for running vCPUs so we can't
1419 * analyze it here, flush TLB regardless of the specified address space.
1421 kvm_make_vcpus_request_mask(kvm,
1422 KVM_REQ_TLB_FLUSH | KVM_REQUEST_NO_WAKEUP,
1423 vcpu_mask, &hv_vcpu->tlb_flush);
1426 /* We always do full TLB flush, set rep_done = rep_cnt. */
1427 return (u64)HV_STATUS_SUCCESS |
1428 ((u64)rep_cnt << HV_HYPERCALL_REP_COMP_OFFSET);
1431 static void kvm_send_ipi_to_many(struct kvm *kvm, u32 vector,
1432 unsigned long *vcpu_bitmap)
1434 struct kvm_lapic_irq irq = {
1435 .delivery_mode = APIC_DM_FIXED,
1438 struct kvm_vcpu *vcpu;
1441 kvm_for_each_vcpu(i, vcpu, kvm) {
1442 if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
1445 /* We fail only when APIC is disabled */
1446 kvm_apic_set_irq(vcpu, &irq, NULL);
1450 static u64 kvm_hv_send_ipi(struct kvm_vcpu *current_vcpu, u64 ingpa, u64 outgpa,
1453 struct kvm *kvm = current_vcpu->kvm;
1454 struct hv_send_ipi_ex send_ipi_ex;
1455 struct hv_send_ipi send_ipi;
1456 u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1457 DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1458 unsigned long *vcpu_mask;
1459 unsigned long valid_bank_mask;
1460 u64 sparse_banks[64];
1461 int sparse_banks_len;
1467 if (unlikely(kvm_read_guest(kvm, ingpa, &send_ipi,
1469 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1470 sparse_banks[0] = send_ipi.cpu_mask;
1471 vector = send_ipi.vector;
1473 /* 'reserved' part of hv_send_ipi should be 0 */
1474 if (unlikely(ingpa >> 32 != 0))
1475 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1476 sparse_banks[0] = outgpa;
1477 vector = (u32)ingpa;
1480 valid_bank_mask = BIT_ULL(0);
1482 trace_kvm_hv_send_ipi(vector, sparse_banks[0]);
1484 if (unlikely(kvm_read_guest(kvm, ingpa, &send_ipi_ex,
1485 sizeof(send_ipi_ex))))
1486 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1488 trace_kvm_hv_send_ipi_ex(send_ipi_ex.vector,
1489 send_ipi_ex.vp_set.format,
1490 send_ipi_ex.vp_set.valid_bank_mask);
1492 vector = send_ipi_ex.vector;
1493 valid_bank_mask = send_ipi_ex.vp_set.valid_bank_mask;
1494 sparse_banks_len = bitmap_weight(&valid_bank_mask, 64) *
1495 sizeof(sparse_banks[0]);
1497 all_cpus = send_ipi_ex.vp_set.format == HV_GENERIC_SET_ALL;
1499 if (!sparse_banks_len)
1504 ingpa + offsetof(struct hv_send_ipi_ex,
1505 vp_set.bank_contents),
1508 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1511 if ((vector < HV_IPI_LOW_VECTOR) || (vector > HV_IPI_HIGH_VECTOR))
1512 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1514 vcpu_mask = all_cpus ? NULL :
1515 sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1516 vp_bitmap, vcpu_bitmap);
1518 kvm_send_ipi_to_many(kvm, vector, vcpu_mask);
1521 return HV_STATUS_SUCCESS;
1524 bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1526 return READ_ONCE(kvm->arch.hyperv.hv_hypercall) & HV_X64_MSR_HYPERCALL_ENABLE;
1529 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
1533 longmode = is_64_bit_mode(vcpu);
1535 kvm_rax_write(vcpu, result);
1537 kvm_rdx_write(vcpu, result >> 32);
1538 kvm_rax_write(vcpu, result & 0xffffffff);
1542 static int kvm_hv_hypercall_complete(struct kvm_vcpu *vcpu, u64 result)
1544 kvm_hv_hypercall_set_result(vcpu, result);
1545 ++vcpu->stat.hypercalls;
1546 return kvm_skip_emulated_instruction(vcpu);
1549 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
1551 return kvm_hv_hypercall_complete(vcpu, vcpu->run->hyperv.u.hcall.result);
1554 static u16 kvm_hvcall_signal_event(struct kvm_vcpu *vcpu, bool fast, u64 param)
1556 struct eventfd_ctx *eventfd;
1558 if (unlikely(!fast)) {
1562 if ((gpa & (__alignof__(param) - 1)) ||
1563 offset_in_page(gpa) + sizeof(param) > PAGE_SIZE)
1564 return HV_STATUS_INVALID_ALIGNMENT;
1566 ret = kvm_vcpu_read_guest(vcpu, gpa, ¶m, sizeof(param));
1568 return HV_STATUS_INVALID_ALIGNMENT;
1572 * Per spec, bits 32-47 contain the extra "flag number". However, we
1573 * have no use for it, and in all known usecases it is zero, so just
1574 * report lookup failure if it isn't.
1576 if (param & 0xffff00000000ULL)
1577 return HV_STATUS_INVALID_PORT_ID;
1578 /* remaining bits are reserved-zero */
1579 if (param & ~KVM_HYPERV_CONN_ID_MASK)
1580 return HV_STATUS_INVALID_HYPERCALL_INPUT;
1582 /* the eventfd is protected by vcpu->kvm->srcu, but conn_to_evt isn't */
1584 eventfd = idr_find(&vcpu->kvm->arch.hyperv.conn_to_evt, param);
1587 return HV_STATUS_INVALID_PORT_ID;
1589 eventfd_signal(eventfd, 1);
1590 return HV_STATUS_SUCCESS;
1593 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
1595 u64 param, ingpa, outgpa, ret = HV_STATUS_SUCCESS;
1596 uint16_t code, rep_idx, rep_cnt;
1600 * hypercall generates UD from non zero cpl and real mode
1603 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
1604 kvm_queue_exception(vcpu, UD_VECTOR);
1608 #ifdef CONFIG_X86_64
1609 if (is_64_bit_mode(vcpu)) {
1610 param = kvm_rcx_read(vcpu);
1611 ingpa = kvm_rdx_read(vcpu);
1612 outgpa = kvm_r8_read(vcpu);
1616 param = ((u64)kvm_rdx_read(vcpu) << 32) |
1617 (kvm_rax_read(vcpu) & 0xffffffff);
1618 ingpa = ((u64)kvm_rbx_read(vcpu) << 32) |
1619 (kvm_rcx_read(vcpu) & 0xffffffff);
1620 outgpa = ((u64)kvm_rdi_read(vcpu) << 32) |
1621 (kvm_rsi_read(vcpu) & 0xffffffff);
1624 code = param & 0xffff;
1625 fast = !!(param & HV_HYPERCALL_FAST_BIT);
1626 rep_cnt = (param >> HV_HYPERCALL_REP_COMP_OFFSET) & 0xfff;
1627 rep_idx = (param >> HV_HYPERCALL_REP_START_OFFSET) & 0xfff;
1628 rep = !!(rep_cnt || rep_idx);
1630 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
1633 case HVCALL_NOTIFY_LONG_SPIN_WAIT:
1634 if (unlikely(rep)) {
1635 ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1638 kvm_vcpu_on_spin(vcpu, true);
1640 case HVCALL_SIGNAL_EVENT:
1641 if (unlikely(rep)) {
1642 ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1645 ret = kvm_hvcall_signal_event(vcpu, fast, ingpa);
1646 if (ret != HV_STATUS_INVALID_PORT_ID)
1648 /* fall through - maybe userspace knows this conn_id. */
1649 case HVCALL_POST_MESSAGE:
1650 /* don't bother userspace if it has no way to handle it */
1651 if (unlikely(rep || !vcpu_to_synic(vcpu)->active)) {
1652 ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1655 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
1656 vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
1657 vcpu->run->hyperv.u.hcall.input = param;
1658 vcpu->run->hyperv.u.hcall.params[0] = ingpa;
1659 vcpu->run->hyperv.u.hcall.params[1] = outgpa;
1660 vcpu->arch.complete_userspace_io =
1661 kvm_hv_hypercall_complete_userspace;
1663 case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
1664 if (unlikely(fast || !rep_cnt || rep_idx)) {
1665 ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1668 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false);
1670 case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
1671 if (unlikely(fast || rep)) {
1672 ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1675 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false);
1677 case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
1678 if (unlikely(fast || !rep_cnt || rep_idx)) {
1679 ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1682 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
1684 case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
1685 if (unlikely(fast || rep)) {
1686 ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1689 ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
1691 case HVCALL_SEND_IPI:
1692 if (unlikely(rep)) {
1693 ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1696 ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, false, fast);
1698 case HVCALL_SEND_IPI_EX:
1699 if (unlikely(fast || rep)) {
1700 ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1703 ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, true, false);
1706 ret = HV_STATUS_INVALID_HYPERCALL_CODE;
1710 return kvm_hv_hypercall_complete(vcpu, ret);
1713 void kvm_hv_init_vm(struct kvm *kvm)
1715 mutex_init(&kvm->arch.hyperv.hv_lock);
1716 idr_init(&kvm->arch.hyperv.conn_to_evt);
1719 void kvm_hv_destroy_vm(struct kvm *kvm)
1721 struct eventfd_ctx *eventfd;
1724 idr_for_each_entry(&kvm->arch.hyperv.conn_to_evt, eventfd, i)
1725 eventfd_ctx_put(eventfd);
1726 idr_destroy(&kvm->arch.hyperv.conn_to_evt);
1729 static int kvm_hv_eventfd_assign(struct kvm *kvm, u32 conn_id, int fd)
1731 struct kvm_hv *hv = &kvm->arch.hyperv;
1732 struct eventfd_ctx *eventfd;
1735 eventfd = eventfd_ctx_fdget(fd);
1736 if (IS_ERR(eventfd))
1737 return PTR_ERR(eventfd);
1739 mutex_lock(&hv->hv_lock);
1740 ret = idr_alloc(&hv->conn_to_evt, eventfd, conn_id, conn_id + 1,
1741 GFP_KERNEL_ACCOUNT);
1742 mutex_unlock(&hv->hv_lock);
1749 eventfd_ctx_put(eventfd);
1753 static int kvm_hv_eventfd_deassign(struct kvm *kvm, u32 conn_id)
1755 struct kvm_hv *hv = &kvm->arch.hyperv;
1756 struct eventfd_ctx *eventfd;
1758 mutex_lock(&hv->hv_lock);
1759 eventfd = idr_remove(&hv->conn_to_evt, conn_id);
1760 mutex_unlock(&hv->hv_lock);
1765 synchronize_srcu(&kvm->srcu);
1766 eventfd_ctx_put(eventfd);
1770 int kvm_vm_ioctl_hv_eventfd(struct kvm *kvm, struct kvm_hyperv_eventfd *args)
1772 if ((args->flags & ~KVM_HYPERV_EVENTFD_DEASSIGN) ||
1773 (args->conn_id & ~KVM_HYPERV_CONN_ID_MASK))
1776 if (args->flags == KVM_HYPERV_EVENTFD_DEASSIGN)
1777 return kvm_hv_eventfd_deassign(kvm, args->conn_id);
1778 return kvm_hv_eventfd_assign(kvm, args->conn_id, args->fd);
1781 int kvm_vcpu_ioctl_get_hv_cpuid(struct kvm_vcpu *vcpu, struct kvm_cpuid2 *cpuid,
1782 struct kvm_cpuid_entry2 __user *entries)
1784 uint16_t evmcs_ver = 0;
1785 struct kvm_cpuid_entry2 cpuid_entries[] = {
1786 { .function = HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS },
1787 { .function = HYPERV_CPUID_INTERFACE },
1788 { .function = HYPERV_CPUID_VERSION },
1789 { .function = HYPERV_CPUID_FEATURES },
1790 { .function = HYPERV_CPUID_ENLIGHTMENT_INFO },
1791 { .function = HYPERV_CPUID_IMPLEMENT_LIMITS },
1792 { .function = HYPERV_CPUID_NESTED_FEATURES },
1794 int i, nent = ARRAY_SIZE(cpuid_entries);
1796 if (kvm_x86_ops->nested_get_evmcs_version)
1797 evmcs_ver = kvm_x86_ops->nested_get_evmcs_version(vcpu);
1799 /* Skip NESTED_FEATURES if eVMCS is not supported */
1803 if (cpuid->nent < nent)
1806 if (cpuid->nent > nent)
1809 for (i = 0; i < nent; i++) {
1810 struct kvm_cpuid_entry2 *ent = &cpuid_entries[i];
1813 switch (ent->function) {
1814 case HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS:
1815 memcpy(signature, "Linux KVM Hv", 12);
1817 ent->eax = HYPERV_CPUID_NESTED_FEATURES;
1818 ent->ebx = signature[0];
1819 ent->ecx = signature[1];
1820 ent->edx = signature[2];
1823 case HYPERV_CPUID_INTERFACE:
1824 memcpy(signature, "Hv#1\0\0\0\0\0\0\0\0", 12);
1825 ent->eax = signature[0];
1828 case HYPERV_CPUID_VERSION:
1830 * We implement some Hyper-V 2016 functions so let's use
1833 ent->eax = 0x00003839;
1834 ent->ebx = 0x000A0000;
1837 case HYPERV_CPUID_FEATURES:
1838 ent->eax |= HV_X64_MSR_VP_RUNTIME_AVAILABLE;
1839 ent->eax |= HV_MSR_TIME_REF_COUNT_AVAILABLE;
1840 ent->eax |= HV_X64_MSR_SYNIC_AVAILABLE;
1841 ent->eax |= HV_MSR_SYNTIMER_AVAILABLE;
1842 ent->eax |= HV_X64_MSR_APIC_ACCESS_AVAILABLE;
1843 ent->eax |= HV_X64_MSR_HYPERCALL_AVAILABLE;
1844 ent->eax |= HV_X64_MSR_VP_INDEX_AVAILABLE;
1845 ent->eax |= HV_X64_MSR_RESET_AVAILABLE;
1846 ent->eax |= HV_MSR_REFERENCE_TSC_AVAILABLE;
1847 ent->eax |= HV_X64_ACCESS_FREQUENCY_MSRS;
1848 ent->eax |= HV_X64_ACCESS_REENLIGHTENMENT;
1850 ent->ebx |= HV_X64_POST_MESSAGES;
1851 ent->ebx |= HV_X64_SIGNAL_EVENTS;
1853 ent->edx |= HV_FEATURE_FREQUENCY_MSRS_AVAILABLE;
1854 ent->edx |= HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
1855 ent->edx |= HV_STIMER_DIRECT_MODE_AVAILABLE;
1859 case HYPERV_CPUID_ENLIGHTMENT_INFO:
1860 ent->eax |= HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
1861 ent->eax |= HV_X64_APIC_ACCESS_RECOMMENDED;
1862 ent->eax |= HV_X64_RELAXED_TIMING_RECOMMENDED;
1863 ent->eax |= HV_X64_CLUSTER_IPI_RECOMMENDED;
1864 ent->eax |= HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED;
1866 ent->eax |= HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
1869 * Default number of spinlock retry attempts, matches
1872 ent->ebx = 0x00000FFF;
1876 case HYPERV_CPUID_IMPLEMENT_LIMITS:
1877 /* Maximum number of virtual processors */
1878 ent->eax = KVM_MAX_VCPUS;
1880 * Maximum number of logical processors, matches
1887 case HYPERV_CPUID_NESTED_FEATURES:
1888 ent->eax = evmcs_ver;
1897 if (copy_to_user(entries, cpuid_entries,
1898 nent * sizeof(struct kvm_cpuid_entry2)))