xref: /linux/arch/x86/kvm/hyperv.c (revision 8137a49e1567726eb10fcf55ad141ac19804ca6b)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * KVM Microsoft Hyper-V emulation
4  *
5  * derived from arch/x86/kvm/x86.c
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright (C) 2008 Qumranet, Inc.
9  * Copyright IBM Corporation, 2008
10  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11  * Copyright (C) 2015 Andrey Smetanin <asmetanin@virtuozzo.com>
12  *
13  * Authors:
14  *   Avi Kivity   <avi@qumranet.com>
15  *   Yaniv Kamay  <yaniv@qumranet.com>
16  *   Amit Shah    <amit.shah@qumranet.com>
17  *   Ben-Ami Yassour <benami@il.ibm.com>
18  *   Andrey Smetanin <asmetanin@virtuozzo.com>
19  */
20 
21 #include "x86.h"
22 #include "lapic.h"
23 #include "ioapic.h"
24 #include "cpuid.h"
25 #include "hyperv.h"
26 #include "xen.h"
27 
28 #include <linux/cpu.h>
29 #include <linux/kvm_host.h>
30 #include <linux/highmem.h>
31 #include <linux/sched/cputime.h>
32 #include <linux/eventfd.h>
33 
34 #include <asm/apicdef.h>
35 #include <trace/events/kvm.h>
36 
37 #include "trace.h"
38 #include "irq.h"
39 #include "fpu.h"
40 
41 /* "Hv#1" signature */
42 #define HYPERV_CPUID_SIGNATURE_EAX 0x31237648
43 
44 #define KVM_HV_MAX_SPARSE_VCPU_SET_BITS DIV_ROUND_UP(KVM_MAX_VCPUS, 64)
45 
46 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
47 				bool vcpu_kick);
48 
49 static inline u64 synic_read_sint(struct kvm_vcpu_hv_synic *synic, int sint)
50 {
51 	return atomic64_read(&synic->sint[sint]);
52 }
53 
54 static inline int synic_get_sint_vector(u64 sint_value)
55 {
56 	if (sint_value & HV_SYNIC_SINT_MASKED)
57 		return -1;
58 	return sint_value & HV_SYNIC_SINT_VECTOR_MASK;
59 }
60 
61 static bool synic_has_vector_connected(struct kvm_vcpu_hv_synic *synic,
62 				      int vector)
63 {
64 	int i;
65 
66 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
67 		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
68 			return true;
69 	}
70 	return false;
71 }
72 
73 static bool synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic *synic,
74 				     int vector)
75 {
76 	int i;
77 	u64 sint_value;
78 
79 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
80 		sint_value = synic_read_sint(synic, i);
81 		if (synic_get_sint_vector(sint_value) == vector &&
82 		    sint_value & HV_SYNIC_SINT_AUTO_EOI)
83 			return true;
84 	}
85 	return false;
86 }
87 
88 static void synic_update_vector(struct kvm_vcpu_hv_synic *synic,
89 				int vector)
90 {
91 	if (vector < HV_SYNIC_FIRST_VALID_VECTOR)
92 		return;
93 
94 	if (synic_has_vector_connected(synic, vector))
95 		__set_bit(vector, synic->vec_bitmap);
96 	else
97 		__clear_bit(vector, synic->vec_bitmap);
98 
99 	if (synic_has_vector_auto_eoi(synic, vector))
100 		__set_bit(vector, synic->auto_eoi_bitmap);
101 	else
102 		__clear_bit(vector, synic->auto_eoi_bitmap);
103 }
104 
105 static int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint,
106 			  u64 data, bool host)
107 {
108 	int vector, old_vector;
109 	bool masked;
110 
111 	vector = data & HV_SYNIC_SINT_VECTOR_MASK;
112 	masked = data & HV_SYNIC_SINT_MASKED;
113 
114 	/*
115 	 * Valid vectors are 16-255, however, nested Hyper-V attempts to write
116 	 * default '0x10000' value on boot and this should not #GP. We need to
117 	 * allow zero-initing the register from host as well.
118 	 */
119 	if (vector < HV_SYNIC_FIRST_VALID_VECTOR && !host && !masked)
120 		return 1;
121 	/*
122 	 * Guest may configure multiple SINTs to use the same vector, so
123 	 * we maintain a bitmap of vectors handled by synic, and a
124 	 * bitmap of vectors with auto-eoi behavior.  The bitmaps are
125 	 * updated here, and atomically queried on fast paths.
126 	 */
127 	old_vector = synic_read_sint(synic, sint) & HV_SYNIC_SINT_VECTOR_MASK;
128 
129 	atomic64_set(&synic->sint[sint], data);
130 
131 	synic_update_vector(synic, old_vector);
132 
133 	synic_update_vector(synic, vector);
134 
135 	/* Load SynIC vectors into EOI exit bitmap */
136 	kvm_make_request(KVM_REQ_SCAN_IOAPIC, hv_synic_to_vcpu(synic));
137 	return 0;
138 }
139 
140 static struct kvm_vcpu *get_vcpu_by_vpidx(struct kvm *kvm, u32 vpidx)
141 {
142 	struct kvm_vcpu *vcpu = NULL;
143 	int i;
144 
145 	if (vpidx >= KVM_MAX_VCPUS)
146 		return NULL;
147 
148 	vcpu = kvm_get_vcpu(kvm, vpidx);
149 	if (vcpu && kvm_hv_get_vpindex(vcpu) == vpidx)
150 		return vcpu;
151 	kvm_for_each_vcpu(i, vcpu, kvm)
152 		if (kvm_hv_get_vpindex(vcpu) == vpidx)
153 			return vcpu;
154 	return NULL;
155 }
156 
157 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vpidx)
158 {
159 	struct kvm_vcpu *vcpu;
160 	struct kvm_vcpu_hv_synic *synic;
161 
162 	vcpu = get_vcpu_by_vpidx(kvm, vpidx);
163 	if (!vcpu || !to_hv_vcpu(vcpu))
164 		return NULL;
165 	synic = to_hv_synic(vcpu);
166 	return (synic->active) ? synic : NULL;
167 }
168 
169 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
170 {
171 	struct kvm *kvm = vcpu->kvm;
172 	struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
173 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
174 	struct kvm_vcpu_hv_stimer *stimer;
175 	int gsi, idx;
176 
177 	trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
178 
179 	/* Try to deliver pending Hyper-V SynIC timers messages */
180 	for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
181 		stimer = &hv_vcpu->stimer[idx];
182 		if (stimer->msg_pending && stimer->config.enable &&
183 		    !stimer->config.direct_mode &&
184 		    stimer->config.sintx == sint)
185 			stimer_mark_pending(stimer, false);
186 	}
187 
188 	idx = srcu_read_lock(&kvm->irq_srcu);
189 	gsi = atomic_read(&synic->sint_to_gsi[sint]);
190 	if (gsi != -1)
191 		kvm_notify_acked_gsi(kvm, gsi);
192 	srcu_read_unlock(&kvm->irq_srcu, idx);
193 }
194 
195 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
196 {
197 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
198 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
199 
200 	hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
201 	hv_vcpu->exit.u.synic.msr = msr;
202 	hv_vcpu->exit.u.synic.control = synic->control;
203 	hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
204 	hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
205 
206 	kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
207 }
208 
209 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
210 			 u32 msr, u64 data, bool host)
211 {
212 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
213 	int ret;
214 
215 	if (!synic->active && !host)
216 		return 1;
217 
218 	trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
219 
220 	ret = 0;
221 	switch (msr) {
222 	case HV_X64_MSR_SCONTROL:
223 		synic->control = data;
224 		if (!host)
225 			synic_exit(synic, msr);
226 		break;
227 	case HV_X64_MSR_SVERSION:
228 		if (!host) {
229 			ret = 1;
230 			break;
231 		}
232 		synic->version = data;
233 		break;
234 	case HV_X64_MSR_SIEFP:
235 		if ((data & HV_SYNIC_SIEFP_ENABLE) && !host &&
236 		    !synic->dont_zero_synic_pages)
237 			if (kvm_clear_guest(vcpu->kvm,
238 					    data & PAGE_MASK, PAGE_SIZE)) {
239 				ret = 1;
240 				break;
241 			}
242 		synic->evt_page = data;
243 		if (!host)
244 			synic_exit(synic, msr);
245 		break;
246 	case HV_X64_MSR_SIMP:
247 		if ((data & HV_SYNIC_SIMP_ENABLE) && !host &&
248 		    !synic->dont_zero_synic_pages)
249 			if (kvm_clear_guest(vcpu->kvm,
250 					    data & PAGE_MASK, PAGE_SIZE)) {
251 				ret = 1;
252 				break;
253 			}
254 		synic->msg_page = data;
255 		if (!host)
256 			synic_exit(synic, msr);
257 		break;
258 	case HV_X64_MSR_EOM: {
259 		int i;
260 
261 		for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
262 			kvm_hv_notify_acked_sint(vcpu, i);
263 		break;
264 	}
265 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
266 		ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
267 		break;
268 	default:
269 		ret = 1;
270 		break;
271 	}
272 	return ret;
273 }
274 
275 static bool kvm_hv_is_syndbg_enabled(struct kvm_vcpu *vcpu)
276 {
277 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
278 
279 	return hv_vcpu->cpuid_cache.syndbg_cap_eax &
280 		HV_X64_SYNDBG_CAP_ALLOW_KERNEL_DEBUGGING;
281 }
282 
283 static int kvm_hv_syndbg_complete_userspace(struct kvm_vcpu *vcpu)
284 {
285 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
286 
287 	if (vcpu->run->hyperv.u.syndbg.msr == HV_X64_MSR_SYNDBG_CONTROL)
288 		hv->hv_syndbg.control.status =
289 			vcpu->run->hyperv.u.syndbg.status;
290 	return 1;
291 }
292 
293 static void syndbg_exit(struct kvm_vcpu *vcpu, u32 msr)
294 {
295 	struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
296 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
297 
298 	hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNDBG;
299 	hv_vcpu->exit.u.syndbg.msr = msr;
300 	hv_vcpu->exit.u.syndbg.control = syndbg->control.control;
301 	hv_vcpu->exit.u.syndbg.send_page = syndbg->control.send_page;
302 	hv_vcpu->exit.u.syndbg.recv_page = syndbg->control.recv_page;
303 	hv_vcpu->exit.u.syndbg.pending_page = syndbg->control.pending_page;
304 	vcpu->arch.complete_userspace_io =
305 			kvm_hv_syndbg_complete_userspace;
306 
307 	kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
308 }
309 
310 static int syndbg_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
311 {
312 	struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
313 
314 	if (!kvm_hv_is_syndbg_enabled(vcpu) && !host)
315 		return 1;
316 
317 	trace_kvm_hv_syndbg_set_msr(vcpu->vcpu_id,
318 				    to_hv_vcpu(vcpu)->vp_index, msr, data);
319 	switch (msr) {
320 	case HV_X64_MSR_SYNDBG_CONTROL:
321 		syndbg->control.control = data;
322 		if (!host)
323 			syndbg_exit(vcpu, msr);
324 		break;
325 	case HV_X64_MSR_SYNDBG_STATUS:
326 		syndbg->control.status = data;
327 		break;
328 	case HV_X64_MSR_SYNDBG_SEND_BUFFER:
329 		syndbg->control.send_page = data;
330 		break;
331 	case HV_X64_MSR_SYNDBG_RECV_BUFFER:
332 		syndbg->control.recv_page = data;
333 		break;
334 	case HV_X64_MSR_SYNDBG_PENDING_BUFFER:
335 		syndbg->control.pending_page = data;
336 		if (!host)
337 			syndbg_exit(vcpu, msr);
338 		break;
339 	case HV_X64_MSR_SYNDBG_OPTIONS:
340 		syndbg->options = data;
341 		break;
342 	default:
343 		break;
344 	}
345 
346 	return 0;
347 }
348 
349 static int syndbg_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
350 {
351 	struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
352 
353 	if (!kvm_hv_is_syndbg_enabled(vcpu) && !host)
354 		return 1;
355 
356 	switch (msr) {
357 	case HV_X64_MSR_SYNDBG_CONTROL:
358 		*pdata = syndbg->control.control;
359 		break;
360 	case HV_X64_MSR_SYNDBG_STATUS:
361 		*pdata = syndbg->control.status;
362 		break;
363 	case HV_X64_MSR_SYNDBG_SEND_BUFFER:
364 		*pdata = syndbg->control.send_page;
365 		break;
366 	case HV_X64_MSR_SYNDBG_RECV_BUFFER:
367 		*pdata = syndbg->control.recv_page;
368 		break;
369 	case HV_X64_MSR_SYNDBG_PENDING_BUFFER:
370 		*pdata = syndbg->control.pending_page;
371 		break;
372 	case HV_X64_MSR_SYNDBG_OPTIONS:
373 		*pdata = syndbg->options;
374 		break;
375 	default:
376 		break;
377 	}
378 
379 	trace_kvm_hv_syndbg_get_msr(vcpu->vcpu_id, kvm_hv_get_vpindex(vcpu), msr, *pdata);
380 
381 	return 0;
382 }
383 
384 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata,
385 			 bool host)
386 {
387 	int ret;
388 
389 	if (!synic->active && !host)
390 		return 1;
391 
392 	ret = 0;
393 	switch (msr) {
394 	case HV_X64_MSR_SCONTROL:
395 		*pdata = synic->control;
396 		break;
397 	case HV_X64_MSR_SVERSION:
398 		*pdata = synic->version;
399 		break;
400 	case HV_X64_MSR_SIEFP:
401 		*pdata = synic->evt_page;
402 		break;
403 	case HV_X64_MSR_SIMP:
404 		*pdata = synic->msg_page;
405 		break;
406 	case HV_X64_MSR_EOM:
407 		*pdata = 0;
408 		break;
409 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
410 		*pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
411 		break;
412 	default:
413 		ret = 1;
414 		break;
415 	}
416 	return ret;
417 }
418 
419 static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
420 {
421 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
422 	struct kvm_lapic_irq irq;
423 	int ret, vector;
424 
425 	if (sint >= ARRAY_SIZE(synic->sint))
426 		return -EINVAL;
427 
428 	vector = synic_get_sint_vector(synic_read_sint(synic, sint));
429 	if (vector < 0)
430 		return -ENOENT;
431 
432 	memset(&irq, 0, sizeof(irq));
433 	irq.shorthand = APIC_DEST_SELF;
434 	irq.dest_mode = APIC_DEST_PHYSICAL;
435 	irq.delivery_mode = APIC_DM_FIXED;
436 	irq.vector = vector;
437 	irq.level = 1;
438 
439 	ret = kvm_irq_delivery_to_apic(vcpu->kvm, vcpu->arch.apic, &irq, NULL);
440 	trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
441 	return ret;
442 }
443 
444 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vpidx, u32 sint)
445 {
446 	struct kvm_vcpu_hv_synic *synic;
447 
448 	synic = synic_get(kvm, vpidx);
449 	if (!synic)
450 		return -EINVAL;
451 
452 	return synic_set_irq(synic, sint);
453 }
454 
455 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
456 {
457 	struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
458 	int i;
459 
460 	trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
461 
462 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
463 		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
464 			kvm_hv_notify_acked_sint(vcpu, i);
465 }
466 
467 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi)
468 {
469 	struct kvm_vcpu_hv_synic *synic;
470 
471 	synic = synic_get(kvm, vpidx);
472 	if (!synic)
473 		return -EINVAL;
474 
475 	if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
476 		return -EINVAL;
477 
478 	atomic_set(&synic->sint_to_gsi[sint], gsi);
479 	return 0;
480 }
481 
482 void kvm_hv_irq_routing_update(struct kvm *kvm)
483 {
484 	struct kvm_irq_routing_table *irq_rt;
485 	struct kvm_kernel_irq_routing_entry *e;
486 	u32 gsi;
487 
488 	irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
489 					lockdep_is_held(&kvm->irq_lock));
490 
491 	for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
492 		hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
493 			if (e->type == KVM_IRQ_ROUTING_HV_SINT)
494 				kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
495 						    e->hv_sint.sint, gsi);
496 		}
497 	}
498 }
499 
500 static void synic_init(struct kvm_vcpu_hv_synic *synic)
501 {
502 	int i;
503 
504 	memset(synic, 0, sizeof(*synic));
505 	synic->version = HV_SYNIC_VERSION_1;
506 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
507 		atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
508 		atomic_set(&synic->sint_to_gsi[i], -1);
509 	}
510 }
511 
512 static u64 get_time_ref_counter(struct kvm *kvm)
513 {
514 	struct kvm_hv *hv = to_kvm_hv(kvm);
515 	struct kvm_vcpu *vcpu;
516 	u64 tsc;
517 
518 	/*
519 	 * Fall back to get_kvmclock_ns() when TSC page hasn't been set up,
520 	 * is broken, disabled or being updated.
521 	 */
522 	if (hv->hv_tsc_page_status != HV_TSC_PAGE_SET)
523 		return div_u64(get_kvmclock_ns(kvm), 100);
524 
525 	vcpu = kvm_get_vcpu(kvm, 0);
526 	tsc = kvm_read_l1_tsc(vcpu, rdtsc());
527 	return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64)
528 		+ hv->tsc_ref.tsc_offset;
529 }
530 
531 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
532 				bool vcpu_kick)
533 {
534 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
535 
536 	set_bit(stimer->index,
537 		to_hv_vcpu(vcpu)->stimer_pending_bitmap);
538 	kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
539 	if (vcpu_kick)
540 		kvm_vcpu_kick(vcpu);
541 }
542 
543 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
544 {
545 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
546 
547 	trace_kvm_hv_stimer_cleanup(hv_stimer_to_vcpu(stimer)->vcpu_id,
548 				    stimer->index);
549 
550 	hrtimer_cancel(&stimer->timer);
551 	clear_bit(stimer->index,
552 		  to_hv_vcpu(vcpu)->stimer_pending_bitmap);
553 	stimer->msg_pending = false;
554 	stimer->exp_time = 0;
555 }
556 
557 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
558 {
559 	struct kvm_vcpu_hv_stimer *stimer;
560 
561 	stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
562 	trace_kvm_hv_stimer_callback(hv_stimer_to_vcpu(stimer)->vcpu_id,
563 				     stimer->index);
564 	stimer_mark_pending(stimer, true);
565 
566 	return HRTIMER_NORESTART;
567 }
568 
569 /*
570  * stimer_start() assumptions:
571  * a) stimer->count is not equal to 0
572  * b) stimer->config has HV_STIMER_ENABLE flag
573  */
574 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
575 {
576 	u64 time_now;
577 	ktime_t ktime_now;
578 
579 	time_now = get_time_ref_counter(hv_stimer_to_vcpu(stimer)->kvm);
580 	ktime_now = ktime_get();
581 
582 	if (stimer->config.periodic) {
583 		if (stimer->exp_time) {
584 			if (time_now >= stimer->exp_time) {
585 				u64 remainder;
586 
587 				div64_u64_rem(time_now - stimer->exp_time,
588 					      stimer->count, &remainder);
589 				stimer->exp_time =
590 					time_now + (stimer->count - remainder);
591 			}
592 		} else
593 			stimer->exp_time = time_now + stimer->count;
594 
595 		trace_kvm_hv_stimer_start_periodic(
596 					hv_stimer_to_vcpu(stimer)->vcpu_id,
597 					stimer->index,
598 					time_now, stimer->exp_time);
599 
600 		hrtimer_start(&stimer->timer,
601 			      ktime_add_ns(ktime_now,
602 					   100 * (stimer->exp_time - time_now)),
603 			      HRTIMER_MODE_ABS);
604 		return 0;
605 	}
606 	stimer->exp_time = stimer->count;
607 	if (time_now >= stimer->count) {
608 		/*
609 		 * Expire timer according to Hypervisor Top-Level Functional
610 		 * specification v4(15.3.1):
611 		 * "If a one shot is enabled and the specified count is in
612 		 * the past, it will expire immediately."
613 		 */
614 		stimer_mark_pending(stimer, false);
615 		return 0;
616 	}
617 
618 	trace_kvm_hv_stimer_start_one_shot(hv_stimer_to_vcpu(stimer)->vcpu_id,
619 					   stimer->index,
620 					   time_now, stimer->count);
621 
622 	hrtimer_start(&stimer->timer,
623 		      ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)),
624 		      HRTIMER_MODE_ABS);
625 	return 0;
626 }
627 
628 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
629 			     bool host)
630 {
631 	union hv_stimer_config new_config = {.as_uint64 = config},
632 		old_config = {.as_uint64 = stimer->config.as_uint64};
633 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
634 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
635 	struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
636 
637 	if (!synic->active && !host)
638 		return 1;
639 
640 	if (unlikely(!host && hv_vcpu->enforce_cpuid && new_config.direct_mode &&
641 		     !(hv_vcpu->cpuid_cache.features_edx &
642 		       HV_STIMER_DIRECT_MODE_AVAILABLE)))
643 		return 1;
644 
645 	trace_kvm_hv_stimer_set_config(hv_stimer_to_vcpu(stimer)->vcpu_id,
646 				       stimer->index, config, host);
647 
648 	stimer_cleanup(stimer);
649 	if (old_config.enable &&
650 	    !new_config.direct_mode && new_config.sintx == 0)
651 		new_config.enable = 0;
652 	stimer->config.as_uint64 = new_config.as_uint64;
653 
654 	if (stimer->config.enable)
655 		stimer_mark_pending(stimer, false);
656 
657 	return 0;
658 }
659 
660 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
661 			    bool host)
662 {
663 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
664 	struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
665 
666 	if (!synic->active && !host)
667 		return 1;
668 
669 	trace_kvm_hv_stimer_set_count(hv_stimer_to_vcpu(stimer)->vcpu_id,
670 				      stimer->index, count, host);
671 
672 	stimer_cleanup(stimer);
673 	stimer->count = count;
674 	if (stimer->count == 0)
675 		stimer->config.enable = 0;
676 	else if (stimer->config.auto_enable)
677 		stimer->config.enable = 1;
678 
679 	if (stimer->config.enable)
680 		stimer_mark_pending(stimer, false);
681 
682 	return 0;
683 }
684 
685 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
686 {
687 	*pconfig = stimer->config.as_uint64;
688 	return 0;
689 }
690 
691 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
692 {
693 	*pcount = stimer->count;
694 	return 0;
695 }
696 
697 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
698 			     struct hv_message *src_msg, bool no_retry)
699 {
700 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
701 	int msg_off = offsetof(struct hv_message_page, sint_message[sint]);
702 	gfn_t msg_page_gfn;
703 	struct hv_message_header hv_hdr;
704 	int r;
705 
706 	if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
707 		return -ENOENT;
708 
709 	msg_page_gfn = synic->msg_page >> PAGE_SHIFT;
710 
711 	/*
712 	 * Strictly following the spec-mandated ordering would assume setting
713 	 * .msg_pending before checking .message_type.  However, this function
714 	 * is only called in vcpu context so the entire update is atomic from
715 	 * guest POV and thus the exact order here doesn't matter.
716 	 */
717 	r = kvm_vcpu_read_guest_page(vcpu, msg_page_gfn, &hv_hdr.message_type,
718 				     msg_off + offsetof(struct hv_message,
719 							header.message_type),
720 				     sizeof(hv_hdr.message_type));
721 	if (r < 0)
722 		return r;
723 
724 	if (hv_hdr.message_type != HVMSG_NONE) {
725 		if (no_retry)
726 			return 0;
727 
728 		hv_hdr.message_flags.msg_pending = 1;
729 		r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn,
730 					      &hv_hdr.message_flags,
731 					      msg_off +
732 					      offsetof(struct hv_message,
733 						       header.message_flags),
734 					      sizeof(hv_hdr.message_flags));
735 		if (r < 0)
736 			return r;
737 		return -EAGAIN;
738 	}
739 
740 	r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn, src_msg, msg_off,
741 				      sizeof(src_msg->header) +
742 				      src_msg->header.payload_size);
743 	if (r < 0)
744 		return r;
745 
746 	r = synic_set_irq(synic, sint);
747 	if (r < 0)
748 		return r;
749 	if (r == 0)
750 		return -EFAULT;
751 	return 0;
752 }
753 
754 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
755 {
756 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
757 	struct hv_message *msg = &stimer->msg;
758 	struct hv_timer_message_payload *payload =
759 			(struct hv_timer_message_payload *)&msg->u.payload;
760 
761 	/*
762 	 * To avoid piling up periodic ticks, don't retry message
763 	 * delivery for them (within "lazy" lost ticks policy).
764 	 */
765 	bool no_retry = stimer->config.periodic;
766 
767 	payload->expiration_time = stimer->exp_time;
768 	payload->delivery_time = get_time_ref_counter(vcpu->kvm);
769 	return synic_deliver_msg(to_hv_synic(vcpu),
770 				 stimer->config.sintx, msg,
771 				 no_retry);
772 }
773 
774 static int stimer_notify_direct(struct kvm_vcpu_hv_stimer *stimer)
775 {
776 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
777 	struct kvm_lapic_irq irq = {
778 		.delivery_mode = APIC_DM_FIXED,
779 		.vector = stimer->config.apic_vector
780 	};
781 
782 	if (lapic_in_kernel(vcpu))
783 		return !kvm_apic_set_irq(vcpu, &irq, NULL);
784 	return 0;
785 }
786 
787 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
788 {
789 	int r, direct = stimer->config.direct_mode;
790 
791 	stimer->msg_pending = true;
792 	if (!direct)
793 		r = stimer_send_msg(stimer);
794 	else
795 		r = stimer_notify_direct(stimer);
796 	trace_kvm_hv_stimer_expiration(hv_stimer_to_vcpu(stimer)->vcpu_id,
797 				       stimer->index, direct, r);
798 	if (!r) {
799 		stimer->msg_pending = false;
800 		if (!(stimer->config.periodic))
801 			stimer->config.enable = 0;
802 	}
803 }
804 
805 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
806 {
807 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
808 	struct kvm_vcpu_hv_stimer *stimer;
809 	u64 time_now, exp_time;
810 	int i;
811 
812 	if (!hv_vcpu)
813 		return;
814 
815 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
816 		if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
817 			stimer = &hv_vcpu->stimer[i];
818 			if (stimer->config.enable) {
819 				exp_time = stimer->exp_time;
820 
821 				if (exp_time) {
822 					time_now =
823 						get_time_ref_counter(vcpu->kvm);
824 					if (time_now >= exp_time)
825 						stimer_expiration(stimer);
826 				}
827 
828 				if ((stimer->config.enable) &&
829 				    stimer->count) {
830 					if (!stimer->msg_pending)
831 						stimer_start(stimer);
832 				} else
833 					stimer_cleanup(stimer);
834 			}
835 		}
836 }
837 
838 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
839 {
840 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
841 	int i;
842 
843 	if (!hv_vcpu)
844 		return;
845 
846 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
847 		stimer_cleanup(&hv_vcpu->stimer[i]);
848 
849 	kfree(hv_vcpu);
850 	vcpu->arch.hyperv = NULL;
851 }
852 
853 bool kvm_hv_assist_page_enabled(struct kvm_vcpu *vcpu)
854 {
855 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
856 
857 	if (!hv_vcpu)
858 		return false;
859 
860 	if (!(hv_vcpu->hv_vapic & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE))
861 		return false;
862 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
863 }
864 EXPORT_SYMBOL_GPL(kvm_hv_assist_page_enabled);
865 
866 bool kvm_hv_get_assist_page(struct kvm_vcpu *vcpu,
867 			    struct hv_vp_assist_page *assist_page)
868 {
869 	if (!kvm_hv_assist_page_enabled(vcpu))
870 		return false;
871 	return !kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data,
872 				      assist_page, sizeof(*assist_page));
873 }
874 EXPORT_SYMBOL_GPL(kvm_hv_get_assist_page);
875 
876 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
877 {
878 	struct hv_message *msg = &stimer->msg;
879 	struct hv_timer_message_payload *payload =
880 			(struct hv_timer_message_payload *)&msg->u.payload;
881 
882 	memset(&msg->header, 0, sizeof(msg->header));
883 	msg->header.message_type = HVMSG_TIMER_EXPIRED;
884 	msg->header.payload_size = sizeof(*payload);
885 
886 	payload->timer_index = stimer->index;
887 	payload->expiration_time = 0;
888 	payload->delivery_time = 0;
889 }
890 
891 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
892 {
893 	memset(stimer, 0, sizeof(*stimer));
894 	stimer->index = timer_index;
895 	hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
896 	stimer->timer.function = stimer_timer_callback;
897 	stimer_prepare_msg(stimer);
898 }
899 
900 static int kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
901 {
902 	struct kvm_vcpu_hv *hv_vcpu;
903 	int i;
904 
905 	hv_vcpu = kzalloc(sizeof(struct kvm_vcpu_hv), GFP_KERNEL_ACCOUNT);
906 	if (!hv_vcpu)
907 		return -ENOMEM;
908 
909 	vcpu->arch.hyperv = hv_vcpu;
910 	hv_vcpu->vcpu = vcpu;
911 
912 	synic_init(&hv_vcpu->synic);
913 
914 	bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
915 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
916 		stimer_init(&hv_vcpu->stimer[i], i);
917 
918 	hv_vcpu->vp_index = kvm_vcpu_get_idx(vcpu);
919 
920 	return 0;
921 }
922 
923 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
924 {
925 	struct kvm_vcpu_hv_synic *synic;
926 	int r;
927 
928 	if (!to_hv_vcpu(vcpu)) {
929 		r = kvm_hv_vcpu_init(vcpu);
930 		if (r)
931 			return r;
932 	}
933 
934 	synic = to_hv_synic(vcpu);
935 
936 	/*
937 	 * Hyper-V SynIC auto EOI SINT's are
938 	 * not compatible with APICV, so request
939 	 * to deactivate APICV permanently.
940 	 */
941 	kvm_request_apicv_update(vcpu->kvm, false, APICV_INHIBIT_REASON_HYPERV);
942 	synic->active = true;
943 	synic->dont_zero_synic_pages = dont_zero_synic_pages;
944 	synic->control = HV_SYNIC_CONTROL_ENABLE;
945 	return 0;
946 }
947 
948 static bool kvm_hv_msr_partition_wide(u32 msr)
949 {
950 	bool r = false;
951 
952 	switch (msr) {
953 	case HV_X64_MSR_GUEST_OS_ID:
954 	case HV_X64_MSR_HYPERCALL:
955 	case HV_X64_MSR_REFERENCE_TSC:
956 	case HV_X64_MSR_TIME_REF_COUNT:
957 	case HV_X64_MSR_CRASH_CTL:
958 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
959 	case HV_X64_MSR_RESET:
960 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
961 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
962 	case HV_X64_MSR_TSC_EMULATION_STATUS:
963 	case HV_X64_MSR_SYNDBG_OPTIONS:
964 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
965 		r = true;
966 		break;
967 	}
968 
969 	return r;
970 }
971 
972 static int kvm_hv_msr_get_crash_data(struct kvm *kvm, u32 index, u64 *pdata)
973 {
974 	struct kvm_hv *hv = to_kvm_hv(kvm);
975 	size_t size = ARRAY_SIZE(hv->hv_crash_param);
976 
977 	if (WARN_ON_ONCE(index >= size))
978 		return -EINVAL;
979 
980 	*pdata = hv->hv_crash_param[array_index_nospec(index, size)];
981 	return 0;
982 }
983 
984 static int kvm_hv_msr_get_crash_ctl(struct kvm *kvm, u64 *pdata)
985 {
986 	struct kvm_hv *hv = to_kvm_hv(kvm);
987 
988 	*pdata = hv->hv_crash_ctl;
989 	return 0;
990 }
991 
992 static int kvm_hv_msr_set_crash_ctl(struct kvm *kvm, u64 data)
993 {
994 	struct kvm_hv *hv = to_kvm_hv(kvm);
995 
996 	hv->hv_crash_ctl = data & HV_CRASH_CTL_CRASH_NOTIFY;
997 
998 	return 0;
999 }
1000 
1001 static int kvm_hv_msr_set_crash_data(struct kvm *kvm, u32 index, u64 data)
1002 {
1003 	struct kvm_hv *hv = to_kvm_hv(kvm);
1004 	size_t size = ARRAY_SIZE(hv->hv_crash_param);
1005 
1006 	if (WARN_ON_ONCE(index >= size))
1007 		return -EINVAL;
1008 
1009 	hv->hv_crash_param[array_index_nospec(index, size)] = data;
1010 	return 0;
1011 }
1012 
1013 /*
1014  * The kvmclock and Hyper-V TSC page use similar formulas, and converting
1015  * between them is possible:
1016  *
1017  * kvmclock formula:
1018  *    nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
1019  *           + system_time
1020  *
1021  * Hyper-V formula:
1022  *    nsec/100 = ticks * scale / 2^64 + offset
1023  *
1024  * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
1025  * By dividing the kvmclock formula by 100 and equating what's left we get:
1026  *    ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1027  *            scale / 2^64 =         tsc_to_system_mul * 2^(tsc_shift-32) / 100
1028  *            scale        =         tsc_to_system_mul * 2^(32+tsc_shift) / 100
1029  *
1030  * Now expand the kvmclock formula and divide by 100:
1031  *    nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
1032  *           - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
1033  *           + system_time
1034  *    nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1035  *               - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1036  *               + system_time / 100
1037  *
1038  * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
1039  *    nsec/100 = ticks * scale / 2^64
1040  *               - tsc_timestamp * scale / 2^64
1041  *               + system_time / 100
1042  *
1043  * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
1044  *    offset = system_time / 100 - tsc_timestamp * scale / 2^64
1045  *
1046  * These two equivalencies are implemented in this function.
1047  */
1048 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
1049 					struct ms_hyperv_tsc_page *tsc_ref)
1050 {
1051 	u64 max_mul;
1052 
1053 	if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
1054 		return false;
1055 
1056 	/*
1057 	 * check if scale would overflow, if so we use the time ref counter
1058 	 *    tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
1059 	 *    tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
1060 	 *    tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
1061 	 */
1062 	max_mul = 100ull << (32 - hv_clock->tsc_shift);
1063 	if (hv_clock->tsc_to_system_mul >= max_mul)
1064 		return false;
1065 
1066 	/*
1067 	 * Otherwise compute the scale and offset according to the formulas
1068 	 * derived above.
1069 	 */
1070 	tsc_ref->tsc_scale =
1071 		mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
1072 				hv_clock->tsc_to_system_mul,
1073 				100);
1074 
1075 	tsc_ref->tsc_offset = hv_clock->system_time;
1076 	do_div(tsc_ref->tsc_offset, 100);
1077 	tsc_ref->tsc_offset -=
1078 		mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
1079 	return true;
1080 }
1081 
1082 /*
1083  * Don't touch TSC page values if the guest has opted for TSC emulation after
1084  * migration. KVM doesn't fully support reenlightenment notifications and TSC
1085  * access emulation and Hyper-V is known to expect the values in TSC page to
1086  * stay constant before TSC access emulation is disabled from guest side
1087  * (HV_X64_MSR_TSC_EMULATION_STATUS). KVM userspace is expected to preserve TSC
1088  * frequency and guest visible TSC value across migration (and prevent it when
1089  * TSC scaling is unsupported).
1090  */
1091 static inline bool tsc_page_update_unsafe(struct kvm_hv *hv)
1092 {
1093 	return (hv->hv_tsc_page_status != HV_TSC_PAGE_GUEST_CHANGED) &&
1094 		hv->hv_tsc_emulation_control;
1095 }
1096 
1097 void kvm_hv_setup_tsc_page(struct kvm *kvm,
1098 			   struct pvclock_vcpu_time_info *hv_clock)
1099 {
1100 	struct kvm_hv *hv = to_kvm_hv(kvm);
1101 	u32 tsc_seq;
1102 	u64 gfn;
1103 
1104 	BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
1105 	BUILD_BUG_ON(offsetof(struct ms_hyperv_tsc_page, tsc_sequence) != 0);
1106 
1107 	if (hv->hv_tsc_page_status == HV_TSC_PAGE_BROKEN ||
1108 	    hv->hv_tsc_page_status == HV_TSC_PAGE_UNSET)
1109 		return;
1110 
1111 	mutex_lock(&hv->hv_lock);
1112 	if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1113 		goto out_unlock;
1114 
1115 	gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
1116 	/*
1117 	 * Because the TSC parameters only vary when there is a
1118 	 * change in the master clock, do not bother with caching.
1119 	 */
1120 	if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
1121 				    &tsc_seq, sizeof(tsc_seq))))
1122 		goto out_err;
1123 
1124 	if (tsc_seq && tsc_page_update_unsafe(hv)) {
1125 		if (kvm_read_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
1126 			goto out_err;
1127 
1128 		hv->hv_tsc_page_status = HV_TSC_PAGE_SET;
1129 		goto out_unlock;
1130 	}
1131 
1132 	/*
1133 	 * While we're computing and writing the parameters, force the
1134 	 * guest to use the time reference count MSR.
1135 	 */
1136 	hv->tsc_ref.tsc_sequence = 0;
1137 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
1138 			    &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
1139 		goto out_err;
1140 
1141 	if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
1142 		goto out_err;
1143 
1144 	/* Ensure sequence is zero before writing the rest of the struct.  */
1145 	smp_wmb();
1146 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
1147 		goto out_err;
1148 
1149 	/*
1150 	 * Now switch to the TSC page mechanism by writing the sequence.
1151 	 */
1152 	tsc_seq++;
1153 	if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
1154 		tsc_seq = 1;
1155 
1156 	/* Write the struct entirely before the non-zero sequence.  */
1157 	smp_wmb();
1158 
1159 	hv->tsc_ref.tsc_sequence = tsc_seq;
1160 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
1161 			    &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
1162 		goto out_err;
1163 
1164 	hv->hv_tsc_page_status = HV_TSC_PAGE_SET;
1165 	goto out_unlock;
1166 
1167 out_err:
1168 	hv->hv_tsc_page_status = HV_TSC_PAGE_BROKEN;
1169 out_unlock:
1170 	mutex_unlock(&hv->hv_lock);
1171 }
1172 
1173 void kvm_hv_invalidate_tsc_page(struct kvm *kvm)
1174 {
1175 	struct kvm_hv *hv = to_kvm_hv(kvm);
1176 	u64 gfn;
1177 	int idx;
1178 
1179 	if (hv->hv_tsc_page_status == HV_TSC_PAGE_BROKEN ||
1180 	    hv->hv_tsc_page_status == HV_TSC_PAGE_UNSET ||
1181 	    tsc_page_update_unsafe(hv))
1182 		return;
1183 
1184 	mutex_lock(&hv->hv_lock);
1185 
1186 	if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1187 		goto out_unlock;
1188 
1189 	/* Preserve HV_TSC_PAGE_GUEST_CHANGED/HV_TSC_PAGE_HOST_CHANGED states */
1190 	if (hv->hv_tsc_page_status == HV_TSC_PAGE_SET)
1191 		hv->hv_tsc_page_status = HV_TSC_PAGE_UPDATING;
1192 
1193 	gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
1194 
1195 	hv->tsc_ref.tsc_sequence = 0;
1196 
1197 	/*
1198 	 * Take the srcu lock as memslots will be accessed to check the gfn
1199 	 * cache generation against the memslots generation.
1200 	 */
1201 	idx = srcu_read_lock(&kvm->srcu);
1202 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
1203 			    &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
1204 		hv->hv_tsc_page_status = HV_TSC_PAGE_BROKEN;
1205 	srcu_read_unlock(&kvm->srcu, idx);
1206 
1207 out_unlock:
1208 	mutex_unlock(&hv->hv_lock);
1209 }
1210 
1211 
1212 static bool hv_check_msr_access(struct kvm_vcpu_hv *hv_vcpu, u32 msr)
1213 {
1214 	if (!hv_vcpu->enforce_cpuid)
1215 		return true;
1216 
1217 	switch (msr) {
1218 	case HV_X64_MSR_GUEST_OS_ID:
1219 	case HV_X64_MSR_HYPERCALL:
1220 		return hv_vcpu->cpuid_cache.features_eax &
1221 			HV_MSR_HYPERCALL_AVAILABLE;
1222 	case HV_X64_MSR_VP_RUNTIME:
1223 		return hv_vcpu->cpuid_cache.features_eax &
1224 			HV_MSR_VP_RUNTIME_AVAILABLE;
1225 	case HV_X64_MSR_TIME_REF_COUNT:
1226 		return hv_vcpu->cpuid_cache.features_eax &
1227 			HV_MSR_TIME_REF_COUNT_AVAILABLE;
1228 	case HV_X64_MSR_VP_INDEX:
1229 		return hv_vcpu->cpuid_cache.features_eax &
1230 			HV_MSR_VP_INDEX_AVAILABLE;
1231 	case HV_X64_MSR_RESET:
1232 		return hv_vcpu->cpuid_cache.features_eax &
1233 			HV_MSR_RESET_AVAILABLE;
1234 	case HV_X64_MSR_REFERENCE_TSC:
1235 		return hv_vcpu->cpuid_cache.features_eax &
1236 			HV_MSR_REFERENCE_TSC_AVAILABLE;
1237 	case HV_X64_MSR_SCONTROL:
1238 	case HV_X64_MSR_SVERSION:
1239 	case HV_X64_MSR_SIEFP:
1240 	case HV_X64_MSR_SIMP:
1241 	case HV_X64_MSR_EOM:
1242 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1243 		return hv_vcpu->cpuid_cache.features_eax &
1244 			HV_MSR_SYNIC_AVAILABLE;
1245 	case HV_X64_MSR_STIMER0_CONFIG:
1246 	case HV_X64_MSR_STIMER1_CONFIG:
1247 	case HV_X64_MSR_STIMER2_CONFIG:
1248 	case HV_X64_MSR_STIMER3_CONFIG:
1249 	case HV_X64_MSR_STIMER0_COUNT:
1250 	case HV_X64_MSR_STIMER1_COUNT:
1251 	case HV_X64_MSR_STIMER2_COUNT:
1252 	case HV_X64_MSR_STIMER3_COUNT:
1253 		return hv_vcpu->cpuid_cache.features_eax &
1254 			HV_MSR_SYNTIMER_AVAILABLE;
1255 	case HV_X64_MSR_EOI:
1256 	case HV_X64_MSR_ICR:
1257 	case HV_X64_MSR_TPR:
1258 	case HV_X64_MSR_VP_ASSIST_PAGE:
1259 		return hv_vcpu->cpuid_cache.features_eax &
1260 			HV_MSR_APIC_ACCESS_AVAILABLE;
1261 		break;
1262 	case HV_X64_MSR_TSC_FREQUENCY:
1263 	case HV_X64_MSR_APIC_FREQUENCY:
1264 		return hv_vcpu->cpuid_cache.features_eax &
1265 			HV_ACCESS_FREQUENCY_MSRS;
1266 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1267 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1268 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1269 		return hv_vcpu->cpuid_cache.features_eax &
1270 			HV_ACCESS_REENLIGHTENMENT;
1271 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1272 	case HV_X64_MSR_CRASH_CTL:
1273 		return hv_vcpu->cpuid_cache.features_edx &
1274 			HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
1275 	case HV_X64_MSR_SYNDBG_OPTIONS:
1276 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1277 		return hv_vcpu->cpuid_cache.features_edx &
1278 			HV_FEATURE_DEBUG_MSRS_AVAILABLE;
1279 	default:
1280 		break;
1281 	}
1282 
1283 	return false;
1284 }
1285 
1286 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
1287 			     bool host)
1288 {
1289 	struct kvm *kvm = vcpu->kvm;
1290 	struct kvm_hv *hv = to_kvm_hv(kvm);
1291 
1292 	if (unlikely(!host && !hv_check_msr_access(to_hv_vcpu(vcpu), msr)))
1293 		return 1;
1294 
1295 	switch (msr) {
1296 	case HV_X64_MSR_GUEST_OS_ID:
1297 		hv->hv_guest_os_id = data;
1298 		/* setting guest os id to zero disables hypercall page */
1299 		if (!hv->hv_guest_os_id)
1300 			hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1301 		break;
1302 	case HV_X64_MSR_HYPERCALL: {
1303 		u8 instructions[9];
1304 		int i = 0;
1305 		u64 addr;
1306 
1307 		/* if guest os id is not set hypercall should remain disabled */
1308 		if (!hv->hv_guest_os_id)
1309 			break;
1310 		if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1311 			hv->hv_hypercall = data;
1312 			break;
1313 		}
1314 
1315 		/*
1316 		 * If Xen and Hyper-V hypercalls are both enabled, disambiguate
1317 		 * the same way Xen itself does, by setting the bit 31 of EAX
1318 		 * which is RsvdZ in the 32-bit Hyper-V hypercall ABI and just
1319 		 * going to be clobbered on 64-bit.
1320 		 */
1321 		if (kvm_xen_hypercall_enabled(kvm)) {
1322 			/* orl $0x80000000, %eax */
1323 			instructions[i++] = 0x0d;
1324 			instructions[i++] = 0x00;
1325 			instructions[i++] = 0x00;
1326 			instructions[i++] = 0x00;
1327 			instructions[i++] = 0x80;
1328 		}
1329 
1330 		/* vmcall/vmmcall */
1331 		static_call(kvm_x86_patch_hypercall)(vcpu, instructions + i);
1332 		i += 3;
1333 
1334 		/* ret */
1335 		((unsigned char *)instructions)[i++] = 0xc3;
1336 
1337 		addr = data & HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_MASK;
1338 		if (kvm_vcpu_write_guest(vcpu, addr, instructions, i))
1339 			return 1;
1340 		hv->hv_hypercall = data;
1341 		break;
1342 	}
1343 	case HV_X64_MSR_REFERENCE_TSC:
1344 		hv->hv_tsc_page = data;
1345 		if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE) {
1346 			if (!host)
1347 				hv->hv_tsc_page_status = HV_TSC_PAGE_GUEST_CHANGED;
1348 			else
1349 				hv->hv_tsc_page_status = HV_TSC_PAGE_HOST_CHANGED;
1350 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1351 		} else {
1352 			hv->hv_tsc_page_status = HV_TSC_PAGE_UNSET;
1353 		}
1354 		break;
1355 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1356 		return kvm_hv_msr_set_crash_data(kvm,
1357 						 msr - HV_X64_MSR_CRASH_P0,
1358 						 data);
1359 	case HV_X64_MSR_CRASH_CTL:
1360 		if (host)
1361 			return kvm_hv_msr_set_crash_ctl(kvm, data);
1362 
1363 		if (data & HV_CRASH_CTL_CRASH_NOTIFY) {
1364 			vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
1365 				   hv->hv_crash_param[0],
1366 				   hv->hv_crash_param[1],
1367 				   hv->hv_crash_param[2],
1368 				   hv->hv_crash_param[3],
1369 				   hv->hv_crash_param[4]);
1370 
1371 			/* Send notification about crash to user space */
1372 			kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
1373 		}
1374 		break;
1375 	case HV_X64_MSR_RESET:
1376 		if (data == 1) {
1377 			vcpu_debug(vcpu, "hyper-v reset requested\n");
1378 			kvm_make_request(KVM_REQ_HV_RESET, vcpu);
1379 		}
1380 		break;
1381 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1382 		hv->hv_reenlightenment_control = data;
1383 		break;
1384 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1385 		hv->hv_tsc_emulation_control = data;
1386 		break;
1387 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1388 		if (data && !host)
1389 			return 1;
1390 
1391 		hv->hv_tsc_emulation_status = data;
1392 		break;
1393 	case HV_X64_MSR_TIME_REF_COUNT:
1394 		/* read-only, but still ignore it if host-initiated */
1395 		if (!host)
1396 			return 1;
1397 		break;
1398 	case HV_X64_MSR_SYNDBG_OPTIONS:
1399 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1400 		return syndbg_set_msr(vcpu, msr, data, host);
1401 	default:
1402 		vcpu_unimpl(vcpu, "Hyper-V unhandled wrmsr: 0x%x data 0x%llx\n",
1403 			    msr, data);
1404 		return 1;
1405 	}
1406 	return 0;
1407 }
1408 
1409 /* Calculate cpu time spent by current task in 100ns units */
1410 static u64 current_task_runtime_100ns(void)
1411 {
1412 	u64 utime, stime;
1413 
1414 	task_cputime_adjusted(current, &utime, &stime);
1415 
1416 	return div_u64(utime + stime, 100);
1417 }
1418 
1419 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1420 {
1421 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
1422 
1423 	if (unlikely(!host && !hv_check_msr_access(hv_vcpu, msr)))
1424 		return 1;
1425 
1426 	switch (msr) {
1427 	case HV_X64_MSR_VP_INDEX: {
1428 		struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1429 		int vcpu_idx = kvm_vcpu_get_idx(vcpu);
1430 		u32 new_vp_index = (u32)data;
1431 
1432 		if (!host || new_vp_index >= KVM_MAX_VCPUS)
1433 			return 1;
1434 
1435 		if (new_vp_index == hv_vcpu->vp_index)
1436 			return 0;
1437 
1438 		/*
1439 		 * The VP index is initialized to vcpu_index by
1440 		 * kvm_hv_vcpu_postcreate so they initially match.  Now the
1441 		 * VP index is changing, adjust num_mismatched_vp_indexes if
1442 		 * it now matches or no longer matches vcpu_idx.
1443 		 */
1444 		if (hv_vcpu->vp_index == vcpu_idx)
1445 			atomic_inc(&hv->num_mismatched_vp_indexes);
1446 		else if (new_vp_index == vcpu_idx)
1447 			atomic_dec(&hv->num_mismatched_vp_indexes);
1448 
1449 		hv_vcpu->vp_index = new_vp_index;
1450 		break;
1451 	}
1452 	case HV_X64_MSR_VP_ASSIST_PAGE: {
1453 		u64 gfn;
1454 		unsigned long addr;
1455 
1456 		if (!(data & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE)) {
1457 			hv_vcpu->hv_vapic = data;
1458 			if (kvm_lapic_enable_pv_eoi(vcpu, 0, 0))
1459 				return 1;
1460 			break;
1461 		}
1462 		gfn = data >> HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT;
1463 		addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1464 		if (kvm_is_error_hva(addr))
1465 			return 1;
1466 
1467 		/*
1468 		 * Clear apic_assist portion of struct hv_vp_assist_page
1469 		 * only, there can be valuable data in the rest which needs
1470 		 * to be preserved e.g. on migration.
1471 		 */
1472 		if (__put_user(0, (u32 __user *)addr))
1473 			return 1;
1474 		hv_vcpu->hv_vapic = data;
1475 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
1476 		if (kvm_lapic_enable_pv_eoi(vcpu,
1477 					    gfn_to_gpa(gfn) | KVM_MSR_ENABLED,
1478 					    sizeof(struct hv_vp_assist_page)))
1479 			return 1;
1480 		break;
1481 	}
1482 	case HV_X64_MSR_EOI:
1483 		return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1484 	case HV_X64_MSR_ICR:
1485 		return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1486 	case HV_X64_MSR_TPR:
1487 		return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1488 	case HV_X64_MSR_VP_RUNTIME:
1489 		if (!host)
1490 			return 1;
1491 		hv_vcpu->runtime_offset = data - current_task_runtime_100ns();
1492 		break;
1493 	case HV_X64_MSR_SCONTROL:
1494 	case HV_X64_MSR_SVERSION:
1495 	case HV_X64_MSR_SIEFP:
1496 	case HV_X64_MSR_SIMP:
1497 	case HV_X64_MSR_EOM:
1498 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1499 		return synic_set_msr(to_hv_synic(vcpu), msr, data, host);
1500 	case HV_X64_MSR_STIMER0_CONFIG:
1501 	case HV_X64_MSR_STIMER1_CONFIG:
1502 	case HV_X64_MSR_STIMER2_CONFIG:
1503 	case HV_X64_MSR_STIMER3_CONFIG: {
1504 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1505 
1506 		return stimer_set_config(to_hv_stimer(vcpu, timer_index),
1507 					 data, host);
1508 	}
1509 	case HV_X64_MSR_STIMER0_COUNT:
1510 	case HV_X64_MSR_STIMER1_COUNT:
1511 	case HV_X64_MSR_STIMER2_COUNT:
1512 	case HV_X64_MSR_STIMER3_COUNT: {
1513 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1514 
1515 		return stimer_set_count(to_hv_stimer(vcpu, timer_index),
1516 					data, host);
1517 	}
1518 	case HV_X64_MSR_TSC_FREQUENCY:
1519 	case HV_X64_MSR_APIC_FREQUENCY:
1520 		/* read-only, but still ignore it if host-initiated */
1521 		if (!host)
1522 			return 1;
1523 		break;
1524 	default:
1525 		vcpu_unimpl(vcpu, "Hyper-V unhandled wrmsr: 0x%x data 0x%llx\n",
1526 			    msr, data);
1527 		return 1;
1528 	}
1529 
1530 	return 0;
1531 }
1532 
1533 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1534 			     bool host)
1535 {
1536 	u64 data = 0;
1537 	struct kvm *kvm = vcpu->kvm;
1538 	struct kvm_hv *hv = to_kvm_hv(kvm);
1539 
1540 	if (unlikely(!host && !hv_check_msr_access(to_hv_vcpu(vcpu), msr)))
1541 		return 1;
1542 
1543 	switch (msr) {
1544 	case HV_X64_MSR_GUEST_OS_ID:
1545 		data = hv->hv_guest_os_id;
1546 		break;
1547 	case HV_X64_MSR_HYPERCALL:
1548 		data = hv->hv_hypercall;
1549 		break;
1550 	case HV_X64_MSR_TIME_REF_COUNT:
1551 		data = get_time_ref_counter(kvm);
1552 		break;
1553 	case HV_X64_MSR_REFERENCE_TSC:
1554 		data = hv->hv_tsc_page;
1555 		break;
1556 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1557 		return kvm_hv_msr_get_crash_data(kvm,
1558 						 msr - HV_X64_MSR_CRASH_P0,
1559 						 pdata);
1560 	case HV_X64_MSR_CRASH_CTL:
1561 		return kvm_hv_msr_get_crash_ctl(kvm, pdata);
1562 	case HV_X64_MSR_RESET:
1563 		data = 0;
1564 		break;
1565 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1566 		data = hv->hv_reenlightenment_control;
1567 		break;
1568 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1569 		data = hv->hv_tsc_emulation_control;
1570 		break;
1571 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1572 		data = hv->hv_tsc_emulation_status;
1573 		break;
1574 	case HV_X64_MSR_SYNDBG_OPTIONS:
1575 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1576 		return syndbg_get_msr(vcpu, msr, pdata, host);
1577 	default:
1578 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1579 		return 1;
1580 	}
1581 
1582 	*pdata = data;
1583 	return 0;
1584 }
1585 
1586 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1587 			  bool host)
1588 {
1589 	u64 data = 0;
1590 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
1591 
1592 	if (unlikely(!host && !hv_check_msr_access(hv_vcpu, msr)))
1593 		return 1;
1594 
1595 	switch (msr) {
1596 	case HV_X64_MSR_VP_INDEX:
1597 		data = hv_vcpu->vp_index;
1598 		break;
1599 	case HV_X64_MSR_EOI:
1600 		return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1601 	case HV_X64_MSR_ICR:
1602 		return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1603 	case HV_X64_MSR_TPR:
1604 		return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1605 	case HV_X64_MSR_VP_ASSIST_PAGE:
1606 		data = hv_vcpu->hv_vapic;
1607 		break;
1608 	case HV_X64_MSR_VP_RUNTIME:
1609 		data = current_task_runtime_100ns() + hv_vcpu->runtime_offset;
1610 		break;
1611 	case HV_X64_MSR_SCONTROL:
1612 	case HV_X64_MSR_SVERSION:
1613 	case HV_X64_MSR_SIEFP:
1614 	case HV_X64_MSR_SIMP:
1615 	case HV_X64_MSR_EOM:
1616 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1617 		return synic_get_msr(to_hv_synic(vcpu), msr, pdata, host);
1618 	case HV_X64_MSR_STIMER0_CONFIG:
1619 	case HV_X64_MSR_STIMER1_CONFIG:
1620 	case HV_X64_MSR_STIMER2_CONFIG:
1621 	case HV_X64_MSR_STIMER3_CONFIG: {
1622 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1623 
1624 		return stimer_get_config(to_hv_stimer(vcpu, timer_index),
1625 					 pdata);
1626 	}
1627 	case HV_X64_MSR_STIMER0_COUNT:
1628 	case HV_X64_MSR_STIMER1_COUNT:
1629 	case HV_X64_MSR_STIMER2_COUNT:
1630 	case HV_X64_MSR_STIMER3_COUNT: {
1631 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1632 
1633 		return stimer_get_count(to_hv_stimer(vcpu, timer_index),
1634 					pdata);
1635 	}
1636 	case HV_X64_MSR_TSC_FREQUENCY:
1637 		data = (u64)vcpu->arch.virtual_tsc_khz * 1000;
1638 		break;
1639 	case HV_X64_MSR_APIC_FREQUENCY:
1640 		data = APIC_BUS_FREQUENCY;
1641 		break;
1642 	default:
1643 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1644 		return 1;
1645 	}
1646 	*pdata = data;
1647 	return 0;
1648 }
1649 
1650 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1651 {
1652 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1653 
1654 	if (!host && !vcpu->arch.hyperv_enabled)
1655 		return 1;
1656 
1657 	if (!to_hv_vcpu(vcpu)) {
1658 		if (kvm_hv_vcpu_init(vcpu))
1659 			return 1;
1660 	}
1661 
1662 	if (kvm_hv_msr_partition_wide(msr)) {
1663 		int r;
1664 
1665 		mutex_lock(&hv->hv_lock);
1666 		r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1667 		mutex_unlock(&hv->hv_lock);
1668 		return r;
1669 	} else
1670 		return kvm_hv_set_msr(vcpu, msr, data, host);
1671 }
1672 
1673 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
1674 {
1675 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1676 
1677 	if (!host && !vcpu->arch.hyperv_enabled)
1678 		return 1;
1679 
1680 	if (!to_hv_vcpu(vcpu)) {
1681 		if (kvm_hv_vcpu_init(vcpu))
1682 			return 1;
1683 	}
1684 
1685 	if (kvm_hv_msr_partition_wide(msr)) {
1686 		int r;
1687 
1688 		mutex_lock(&hv->hv_lock);
1689 		r = kvm_hv_get_msr_pw(vcpu, msr, pdata, host);
1690 		mutex_unlock(&hv->hv_lock);
1691 		return r;
1692 	} else
1693 		return kvm_hv_get_msr(vcpu, msr, pdata, host);
1694 }
1695 
1696 static __always_inline unsigned long *sparse_set_to_vcpu_mask(
1697 	struct kvm *kvm, u64 *sparse_banks, u64 valid_bank_mask,
1698 	u64 *vp_bitmap, unsigned long *vcpu_bitmap)
1699 {
1700 	struct kvm_hv *hv = to_kvm_hv(kvm);
1701 	struct kvm_vcpu *vcpu;
1702 	int i, bank, sbank = 0;
1703 
1704 	memset(vp_bitmap, 0,
1705 	       KVM_HV_MAX_SPARSE_VCPU_SET_BITS * sizeof(*vp_bitmap));
1706 	for_each_set_bit(bank, (unsigned long *)&valid_bank_mask,
1707 			 KVM_HV_MAX_SPARSE_VCPU_SET_BITS)
1708 		vp_bitmap[bank] = sparse_banks[sbank++];
1709 
1710 	if (likely(!atomic_read(&hv->num_mismatched_vp_indexes))) {
1711 		/* for all vcpus vp_index == vcpu_idx */
1712 		return (unsigned long *)vp_bitmap;
1713 	}
1714 
1715 	bitmap_zero(vcpu_bitmap, KVM_MAX_VCPUS);
1716 	kvm_for_each_vcpu(i, vcpu, kvm) {
1717 		if (test_bit(kvm_hv_get_vpindex(vcpu), (unsigned long *)vp_bitmap))
1718 			__set_bit(i, vcpu_bitmap);
1719 	}
1720 	return vcpu_bitmap;
1721 }
1722 
1723 struct kvm_hv_hcall {
1724 	u64 param;
1725 	u64 ingpa;
1726 	u64 outgpa;
1727 	u16 code;
1728 	u16 rep_cnt;
1729 	u16 rep_idx;
1730 	bool fast;
1731 	bool rep;
1732 	sse128_t xmm[HV_HYPERCALL_MAX_XMM_REGISTERS];
1733 };
1734 
1735 static u64 kvm_hv_flush_tlb(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc, bool ex)
1736 {
1737 	int i;
1738 	gpa_t gpa;
1739 	struct kvm *kvm = vcpu->kvm;
1740 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
1741 	struct hv_tlb_flush_ex flush_ex;
1742 	struct hv_tlb_flush flush;
1743 	u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1744 	DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1745 	unsigned long *vcpu_mask;
1746 	u64 valid_bank_mask;
1747 	u64 sparse_banks[64];
1748 	int sparse_banks_len;
1749 	bool all_cpus;
1750 
1751 	if (!ex) {
1752 		if (hc->fast) {
1753 			flush.address_space = hc->ingpa;
1754 			flush.flags = hc->outgpa;
1755 			flush.processor_mask = sse128_lo(hc->xmm[0]);
1756 		} else {
1757 			if (unlikely(kvm_read_guest(kvm, hc->ingpa,
1758 						    &flush, sizeof(flush))))
1759 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
1760 		}
1761 
1762 		trace_kvm_hv_flush_tlb(flush.processor_mask,
1763 				       flush.address_space, flush.flags);
1764 
1765 		valid_bank_mask = BIT_ULL(0);
1766 		sparse_banks[0] = flush.processor_mask;
1767 
1768 		/*
1769 		 * Work around possible WS2012 bug: it sends hypercalls
1770 		 * with processor_mask = 0x0 and HV_FLUSH_ALL_PROCESSORS clear,
1771 		 * while also expecting us to flush something and crashing if
1772 		 * we don't. Let's treat processor_mask == 0 same as
1773 		 * HV_FLUSH_ALL_PROCESSORS.
1774 		 */
1775 		all_cpus = (flush.flags & HV_FLUSH_ALL_PROCESSORS) ||
1776 			flush.processor_mask == 0;
1777 	} else {
1778 		if (hc->fast) {
1779 			flush_ex.address_space = hc->ingpa;
1780 			flush_ex.flags = hc->outgpa;
1781 			memcpy(&flush_ex.hv_vp_set,
1782 			       &hc->xmm[0], sizeof(hc->xmm[0]));
1783 		} else {
1784 			if (unlikely(kvm_read_guest(kvm, hc->ingpa, &flush_ex,
1785 						    sizeof(flush_ex))))
1786 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
1787 		}
1788 
1789 		trace_kvm_hv_flush_tlb_ex(flush_ex.hv_vp_set.valid_bank_mask,
1790 					  flush_ex.hv_vp_set.format,
1791 					  flush_ex.address_space,
1792 					  flush_ex.flags);
1793 
1794 		valid_bank_mask = flush_ex.hv_vp_set.valid_bank_mask;
1795 		all_cpus = flush_ex.hv_vp_set.format !=
1796 			HV_GENERIC_SET_SPARSE_4K;
1797 
1798 		sparse_banks_len = bitmap_weight((unsigned long *)&valid_bank_mask, 64);
1799 
1800 		if (!sparse_banks_len && !all_cpus)
1801 			goto ret_success;
1802 
1803 		if (!all_cpus) {
1804 			if (hc->fast) {
1805 				if (sparse_banks_len > HV_HYPERCALL_MAX_XMM_REGISTERS - 1)
1806 					return HV_STATUS_INVALID_HYPERCALL_INPUT;
1807 				for (i = 0; i < sparse_banks_len; i += 2) {
1808 					sparse_banks[i] = sse128_lo(hc->xmm[i / 2 + 1]);
1809 					sparse_banks[i + 1] = sse128_hi(hc->xmm[i / 2 + 1]);
1810 				}
1811 			} else {
1812 				gpa = hc->ingpa + offsetof(struct hv_tlb_flush_ex,
1813 							   hv_vp_set.bank_contents);
1814 				if (unlikely(kvm_read_guest(kvm, gpa, sparse_banks,
1815 							    sparse_banks_len *
1816 							    sizeof(sparse_banks[0]))))
1817 					return HV_STATUS_INVALID_HYPERCALL_INPUT;
1818 			}
1819 		}
1820 	}
1821 
1822 	cpumask_clear(&hv_vcpu->tlb_flush);
1823 
1824 	vcpu_mask = all_cpus ? NULL :
1825 		sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1826 					vp_bitmap, vcpu_bitmap);
1827 
1828 	/*
1829 	 * vcpu->arch.cr3 may not be up-to-date for running vCPUs so we can't
1830 	 * analyze it here, flush TLB regardless of the specified address space.
1831 	 */
1832 	kvm_make_vcpus_request_mask(kvm, KVM_REQ_TLB_FLUSH_GUEST,
1833 				    NULL, vcpu_mask, &hv_vcpu->tlb_flush);
1834 
1835 ret_success:
1836 	/* We always do full TLB flush, set 'Reps completed' = 'Rep Count' */
1837 	return (u64)HV_STATUS_SUCCESS |
1838 		((u64)hc->rep_cnt << HV_HYPERCALL_REP_COMP_OFFSET);
1839 }
1840 
1841 static void kvm_send_ipi_to_many(struct kvm *kvm, u32 vector,
1842 				 unsigned long *vcpu_bitmap)
1843 {
1844 	struct kvm_lapic_irq irq = {
1845 		.delivery_mode = APIC_DM_FIXED,
1846 		.vector = vector
1847 	};
1848 	struct kvm_vcpu *vcpu;
1849 	int i;
1850 
1851 	kvm_for_each_vcpu(i, vcpu, kvm) {
1852 		if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
1853 			continue;
1854 
1855 		/* We fail only when APIC is disabled */
1856 		kvm_apic_set_irq(vcpu, &irq, NULL);
1857 	}
1858 }
1859 
1860 static u64 kvm_hv_send_ipi(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc, bool ex)
1861 {
1862 	struct kvm *kvm = vcpu->kvm;
1863 	struct hv_send_ipi_ex send_ipi_ex;
1864 	struct hv_send_ipi send_ipi;
1865 	u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1866 	DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1867 	unsigned long *vcpu_mask;
1868 	unsigned long valid_bank_mask;
1869 	u64 sparse_banks[64];
1870 	int sparse_banks_len;
1871 	u32 vector;
1872 	bool all_cpus;
1873 
1874 	if (!ex) {
1875 		if (!hc->fast) {
1876 			if (unlikely(kvm_read_guest(kvm, hc->ingpa, &send_ipi,
1877 						    sizeof(send_ipi))))
1878 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
1879 			sparse_banks[0] = send_ipi.cpu_mask;
1880 			vector = send_ipi.vector;
1881 		} else {
1882 			/* 'reserved' part of hv_send_ipi should be 0 */
1883 			if (unlikely(hc->ingpa >> 32 != 0))
1884 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
1885 			sparse_banks[0] = hc->outgpa;
1886 			vector = (u32)hc->ingpa;
1887 		}
1888 		all_cpus = false;
1889 		valid_bank_mask = BIT_ULL(0);
1890 
1891 		trace_kvm_hv_send_ipi(vector, sparse_banks[0]);
1892 	} else {
1893 		if (unlikely(kvm_read_guest(kvm, hc->ingpa, &send_ipi_ex,
1894 					    sizeof(send_ipi_ex))))
1895 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1896 
1897 		trace_kvm_hv_send_ipi_ex(send_ipi_ex.vector,
1898 					 send_ipi_ex.vp_set.format,
1899 					 send_ipi_ex.vp_set.valid_bank_mask);
1900 
1901 		vector = send_ipi_ex.vector;
1902 		valid_bank_mask = send_ipi_ex.vp_set.valid_bank_mask;
1903 		sparse_banks_len = bitmap_weight(&valid_bank_mask, 64) *
1904 			sizeof(sparse_banks[0]);
1905 
1906 		all_cpus = send_ipi_ex.vp_set.format == HV_GENERIC_SET_ALL;
1907 
1908 		if (!sparse_banks_len)
1909 			goto ret_success;
1910 
1911 		if (!all_cpus &&
1912 		    kvm_read_guest(kvm,
1913 				   hc->ingpa + offsetof(struct hv_send_ipi_ex,
1914 							vp_set.bank_contents),
1915 				   sparse_banks,
1916 				   sparse_banks_len))
1917 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1918 	}
1919 
1920 	if ((vector < HV_IPI_LOW_VECTOR) || (vector > HV_IPI_HIGH_VECTOR))
1921 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
1922 
1923 	vcpu_mask = all_cpus ? NULL :
1924 		sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1925 					vp_bitmap, vcpu_bitmap);
1926 
1927 	kvm_send_ipi_to_many(kvm, vector, vcpu_mask);
1928 
1929 ret_success:
1930 	return HV_STATUS_SUCCESS;
1931 }
1932 
1933 void kvm_hv_set_cpuid(struct kvm_vcpu *vcpu)
1934 {
1935 	struct kvm_cpuid_entry2 *entry;
1936 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
1937 
1938 	entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_INTERFACE, 0);
1939 	if (entry && entry->eax == HYPERV_CPUID_SIGNATURE_EAX) {
1940 		vcpu->arch.hyperv_enabled = true;
1941 	} else {
1942 		vcpu->arch.hyperv_enabled = false;
1943 		return;
1944 	}
1945 
1946 	if (!to_hv_vcpu(vcpu) && kvm_hv_vcpu_init(vcpu))
1947 		return;
1948 
1949 	hv_vcpu = to_hv_vcpu(vcpu);
1950 
1951 	entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_FEATURES, 0);
1952 	if (entry) {
1953 		hv_vcpu->cpuid_cache.features_eax = entry->eax;
1954 		hv_vcpu->cpuid_cache.features_ebx = entry->ebx;
1955 		hv_vcpu->cpuid_cache.features_edx = entry->edx;
1956 	} else {
1957 		hv_vcpu->cpuid_cache.features_eax = 0;
1958 		hv_vcpu->cpuid_cache.features_ebx = 0;
1959 		hv_vcpu->cpuid_cache.features_edx = 0;
1960 	}
1961 
1962 	entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_ENLIGHTMENT_INFO, 0);
1963 	if (entry) {
1964 		hv_vcpu->cpuid_cache.enlightenments_eax = entry->eax;
1965 		hv_vcpu->cpuid_cache.enlightenments_ebx = entry->ebx;
1966 	} else {
1967 		hv_vcpu->cpuid_cache.enlightenments_eax = 0;
1968 		hv_vcpu->cpuid_cache.enlightenments_ebx = 0;
1969 	}
1970 
1971 	entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES, 0);
1972 	if (entry)
1973 		hv_vcpu->cpuid_cache.syndbg_cap_eax = entry->eax;
1974 	else
1975 		hv_vcpu->cpuid_cache.syndbg_cap_eax = 0;
1976 }
1977 
1978 int kvm_hv_set_enforce_cpuid(struct kvm_vcpu *vcpu, bool enforce)
1979 {
1980 	struct kvm_vcpu_hv *hv_vcpu;
1981 	int ret = 0;
1982 
1983 	if (!to_hv_vcpu(vcpu)) {
1984 		if (enforce) {
1985 			ret = kvm_hv_vcpu_init(vcpu);
1986 			if (ret)
1987 				return ret;
1988 		} else {
1989 			return 0;
1990 		}
1991 	}
1992 
1993 	hv_vcpu = to_hv_vcpu(vcpu);
1994 	hv_vcpu->enforce_cpuid = enforce;
1995 
1996 	return ret;
1997 }
1998 
1999 bool kvm_hv_hypercall_enabled(struct kvm_vcpu *vcpu)
2000 {
2001 	return vcpu->arch.hyperv_enabled && to_kvm_hv(vcpu->kvm)->hv_guest_os_id;
2002 }
2003 
2004 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
2005 {
2006 	bool longmode;
2007 
2008 	longmode = is_64_bit_mode(vcpu);
2009 	if (longmode)
2010 		kvm_rax_write(vcpu, result);
2011 	else {
2012 		kvm_rdx_write(vcpu, result >> 32);
2013 		kvm_rax_write(vcpu, result & 0xffffffff);
2014 	}
2015 }
2016 
2017 static int kvm_hv_hypercall_complete(struct kvm_vcpu *vcpu, u64 result)
2018 {
2019 	kvm_hv_hypercall_set_result(vcpu, result);
2020 	++vcpu->stat.hypercalls;
2021 	return kvm_skip_emulated_instruction(vcpu);
2022 }
2023 
2024 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
2025 {
2026 	return kvm_hv_hypercall_complete(vcpu, vcpu->run->hyperv.u.hcall.result);
2027 }
2028 
2029 static u16 kvm_hvcall_signal_event(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc)
2030 {
2031 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
2032 	struct eventfd_ctx *eventfd;
2033 
2034 	if (unlikely(!hc->fast)) {
2035 		int ret;
2036 		gpa_t gpa = hc->ingpa;
2037 
2038 		if ((gpa & (__alignof__(hc->ingpa) - 1)) ||
2039 		    offset_in_page(gpa) + sizeof(hc->ingpa) > PAGE_SIZE)
2040 			return HV_STATUS_INVALID_ALIGNMENT;
2041 
2042 		ret = kvm_vcpu_read_guest(vcpu, gpa,
2043 					  &hc->ingpa, sizeof(hc->ingpa));
2044 		if (ret < 0)
2045 			return HV_STATUS_INVALID_ALIGNMENT;
2046 	}
2047 
2048 	/*
2049 	 * Per spec, bits 32-47 contain the extra "flag number".  However, we
2050 	 * have no use for it, and in all known usecases it is zero, so just
2051 	 * report lookup failure if it isn't.
2052 	 */
2053 	if (hc->ingpa & 0xffff00000000ULL)
2054 		return HV_STATUS_INVALID_PORT_ID;
2055 	/* remaining bits are reserved-zero */
2056 	if (hc->ingpa & ~KVM_HYPERV_CONN_ID_MASK)
2057 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
2058 
2059 	/* the eventfd is protected by vcpu->kvm->srcu, but conn_to_evt isn't */
2060 	rcu_read_lock();
2061 	eventfd = idr_find(&hv->conn_to_evt, hc->ingpa);
2062 	rcu_read_unlock();
2063 	if (!eventfd)
2064 		return HV_STATUS_INVALID_PORT_ID;
2065 
2066 	eventfd_signal(eventfd, 1);
2067 	return HV_STATUS_SUCCESS;
2068 }
2069 
2070 static bool is_xmm_fast_hypercall(struct kvm_hv_hcall *hc)
2071 {
2072 	switch (hc->code) {
2073 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
2074 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
2075 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
2076 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
2077 		return true;
2078 	}
2079 
2080 	return false;
2081 }
2082 
2083 static void kvm_hv_hypercall_read_xmm(struct kvm_hv_hcall *hc)
2084 {
2085 	int reg;
2086 
2087 	kvm_fpu_get();
2088 	for (reg = 0; reg < HV_HYPERCALL_MAX_XMM_REGISTERS; reg++)
2089 		_kvm_read_sse_reg(reg, &hc->xmm[reg]);
2090 	kvm_fpu_put();
2091 }
2092 
2093 static bool hv_check_hypercall_access(struct kvm_vcpu_hv *hv_vcpu, u16 code)
2094 {
2095 	if (!hv_vcpu->enforce_cpuid)
2096 		return true;
2097 
2098 	switch (code) {
2099 	case HVCALL_NOTIFY_LONG_SPIN_WAIT:
2100 		return hv_vcpu->cpuid_cache.enlightenments_ebx &&
2101 			hv_vcpu->cpuid_cache.enlightenments_ebx != U32_MAX;
2102 	case HVCALL_POST_MESSAGE:
2103 		return hv_vcpu->cpuid_cache.features_ebx & HV_POST_MESSAGES;
2104 	case HVCALL_SIGNAL_EVENT:
2105 		return hv_vcpu->cpuid_cache.features_ebx & HV_SIGNAL_EVENTS;
2106 	case HVCALL_POST_DEBUG_DATA:
2107 	case HVCALL_RETRIEVE_DEBUG_DATA:
2108 	case HVCALL_RESET_DEBUG_SESSION:
2109 		/*
2110 		 * Return 'true' when SynDBG is disabled so the resulting code
2111 		 * will be HV_STATUS_INVALID_HYPERCALL_CODE.
2112 		 */
2113 		return !kvm_hv_is_syndbg_enabled(hv_vcpu->vcpu) ||
2114 			hv_vcpu->cpuid_cache.features_ebx & HV_DEBUGGING;
2115 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
2116 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
2117 		if (!(hv_vcpu->cpuid_cache.enlightenments_eax &
2118 		      HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED))
2119 			return false;
2120 		fallthrough;
2121 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
2122 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
2123 		return hv_vcpu->cpuid_cache.enlightenments_eax &
2124 			HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
2125 	case HVCALL_SEND_IPI_EX:
2126 		if (!(hv_vcpu->cpuid_cache.enlightenments_eax &
2127 		      HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED))
2128 			return false;
2129 		fallthrough;
2130 	case HVCALL_SEND_IPI:
2131 		return hv_vcpu->cpuid_cache.enlightenments_eax &
2132 			HV_X64_CLUSTER_IPI_RECOMMENDED;
2133 	default:
2134 		break;
2135 	}
2136 
2137 	return true;
2138 }
2139 
2140 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
2141 {
2142 	struct kvm_hv_hcall hc;
2143 	u64 ret = HV_STATUS_SUCCESS;
2144 
2145 	/*
2146 	 * hypercall generates UD from non zero cpl and real mode
2147 	 * per HYPER-V spec
2148 	 */
2149 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0 || !is_protmode(vcpu)) {
2150 		kvm_queue_exception(vcpu, UD_VECTOR);
2151 		return 1;
2152 	}
2153 
2154 #ifdef CONFIG_X86_64
2155 	if (is_64_bit_mode(vcpu)) {
2156 		hc.param = kvm_rcx_read(vcpu);
2157 		hc.ingpa = kvm_rdx_read(vcpu);
2158 		hc.outgpa = kvm_r8_read(vcpu);
2159 	} else
2160 #endif
2161 	{
2162 		hc.param = ((u64)kvm_rdx_read(vcpu) << 32) |
2163 			    (kvm_rax_read(vcpu) & 0xffffffff);
2164 		hc.ingpa = ((u64)kvm_rbx_read(vcpu) << 32) |
2165 			    (kvm_rcx_read(vcpu) & 0xffffffff);
2166 		hc.outgpa = ((u64)kvm_rdi_read(vcpu) << 32) |
2167 			     (kvm_rsi_read(vcpu) & 0xffffffff);
2168 	}
2169 
2170 	hc.code = hc.param & 0xffff;
2171 	hc.fast = !!(hc.param & HV_HYPERCALL_FAST_BIT);
2172 	hc.rep_cnt = (hc.param >> HV_HYPERCALL_REP_COMP_OFFSET) & 0xfff;
2173 	hc.rep_idx = (hc.param >> HV_HYPERCALL_REP_START_OFFSET) & 0xfff;
2174 	hc.rep = !!(hc.rep_cnt || hc.rep_idx);
2175 
2176 	if (hc.fast && is_xmm_fast_hypercall(&hc))
2177 		kvm_hv_hypercall_read_xmm(&hc);
2178 
2179 	trace_kvm_hv_hypercall(hc.code, hc.fast, hc.rep_cnt, hc.rep_idx,
2180 			       hc.ingpa, hc.outgpa);
2181 
2182 	if (unlikely(!hv_check_hypercall_access(to_hv_vcpu(vcpu), hc.code))) {
2183 		ret = HV_STATUS_ACCESS_DENIED;
2184 		goto hypercall_complete;
2185 	}
2186 
2187 	switch (hc.code) {
2188 	case HVCALL_NOTIFY_LONG_SPIN_WAIT:
2189 		if (unlikely(hc.rep)) {
2190 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2191 			break;
2192 		}
2193 		kvm_vcpu_on_spin(vcpu, true);
2194 		break;
2195 	case HVCALL_SIGNAL_EVENT:
2196 		if (unlikely(hc.rep)) {
2197 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2198 			break;
2199 		}
2200 		ret = kvm_hvcall_signal_event(vcpu, &hc);
2201 		if (ret != HV_STATUS_INVALID_PORT_ID)
2202 			break;
2203 		fallthrough;	/* maybe userspace knows this conn_id */
2204 	case HVCALL_POST_MESSAGE:
2205 		/* don't bother userspace if it has no way to handle it */
2206 		if (unlikely(hc.rep || !to_hv_synic(vcpu)->active)) {
2207 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2208 			break;
2209 		}
2210 		vcpu->run->exit_reason = KVM_EXIT_HYPERV;
2211 		vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
2212 		vcpu->run->hyperv.u.hcall.input = hc.param;
2213 		vcpu->run->hyperv.u.hcall.params[0] = hc.ingpa;
2214 		vcpu->run->hyperv.u.hcall.params[1] = hc.outgpa;
2215 		vcpu->arch.complete_userspace_io =
2216 				kvm_hv_hypercall_complete_userspace;
2217 		return 0;
2218 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
2219 		if (unlikely(!hc.rep_cnt || hc.rep_idx)) {
2220 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2221 			break;
2222 		}
2223 		ret = kvm_hv_flush_tlb(vcpu, &hc, false);
2224 		break;
2225 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
2226 		if (unlikely(hc.rep)) {
2227 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2228 			break;
2229 		}
2230 		ret = kvm_hv_flush_tlb(vcpu, &hc, false);
2231 		break;
2232 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
2233 		if (unlikely(!hc.rep_cnt || hc.rep_idx)) {
2234 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2235 			break;
2236 		}
2237 		ret = kvm_hv_flush_tlb(vcpu, &hc, true);
2238 		break;
2239 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
2240 		if (unlikely(hc.rep)) {
2241 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2242 			break;
2243 		}
2244 		ret = kvm_hv_flush_tlb(vcpu, &hc, true);
2245 		break;
2246 	case HVCALL_SEND_IPI:
2247 		if (unlikely(hc.rep)) {
2248 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2249 			break;
2250 		}
2251 		ret = kvm_hv_send_ipi(vcpu, &hc, false);
2252 		break;
2253 	case HVCALL_SEND_IPI_EX:
2254 		if (unlikely(hc.fast || hc.rep)) {
2255 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2256 			break;
2257 		}
2258 		ret = kvm_hv_send_ipi(vcpu, &hc, true);
2259 		break;
2260 	case HVCALL_POST_DEBUG_DATA:
2261 	case HVCALL_RETRIEVE_DEBUG_DATA:
2262 		if (unlikely(hc.fast)) {
2263 			ret = HV_STATUS_INVALID_PARAMETER;
2264 			break;
2265 		}
2266 		fallthrough;
2267 	case HVCALL_RESET_DEBUG_SESSION: {
2268 		struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
2269 
2270 		if (!kvm_hv_is_syndbg_enabled(vcpu)) {
2271 			ret = HV_STATUS_INVALID_HYPERCALL_CODE;
2272 			break;
2273 		}
2274 
2275 		if (!(syndbg->options & HV_X64_SYNDBG_OPTION_USE_HCALLS)) {
2276 			ret = HV_STATUS_OPERATION_DENIED;
2277 			break;
2278 		}
2279 		vcpu->run->exit_reason = KVM_EXIT_HYPERV;
2280 		vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
2281 		vcpu->run->hyperv.u.hcall.input = hc.param;
2282 		vcpu->run->hyperv.u.hcall.params[0] = hc.ingpa;
2283 		vcpu->run->hyperv.u.hcall.params[1] = hc.outgpa;
2284 		vcpu->arch.complete_userspace_io =
2285 				kvm_hv_hypercall_complete_userspace;
2286 		return 0;
2287 	}
2288 	default:
2289 		ret = HV_STATUS_INVALID_HYPERCALL_CODE;
2290 		break;
2291 	}
2292 
2293 hypercall_complete:
2294 	return kvm_hv_hypercall_complete(vcpu, ret);
2295 }
2296 
2297 void kvm_hv_init_vm(struct kvm *kvm)
2298 {
2299 	struct kvm_hv *hv = to_kvm_hv(kvm);
2300 
2301 	mutex_init(&hv->hv_lock);
2302 	idr_init(&hv->conn_to_evt);
2303 }
2304 
2305 void kvm_hv_destroy_vm(struct kvm *kvm)
2306 {
2307 	struct kvm_hv *hv = to_kvm_hv(kvm);
2308 	struct eventfd_ctx *eventfd;
2309 	int i;
2310 
2311 	idr_for_each_entry(&hv->conn_to_evt, eventfd, i)
2312 		eventfd_ctx_put(eventfd);
2313 	idr_destroy(&hv->conn_to_evt);
2314 }
2315 
2316 static int kvm_hv_eventfd_assign(struct kvm *kvm, u32 conn_id, int fd)
2317 {
2318 	struct kvm_hv *hv = to_kvm_hv(kvm);
2319 	struct eventfd_ctx *eventfd;
2320 	int ret;
2321 
2322 	eventfd = eventfd_ctx_fdget(fd);
2323 	if (IS_ERR(eventfd))
2324 		return PTR_ERR(eventfd);
2325 
2326 	mutex_lock(&hv->hv_lock);
2327 	ret = idr_alloc(&hv->conn_to_evt, eventfd, conn_id, conn_id + 1,
2328 			GFP_KERNEL_ACCOUNT);
2329 	mutex_unlock(&hv->hv_lock);
2330 
2331 	if (ret >= 0)
2332 		return 0;
2333 
2334 	if (ret == -ENOSPC)
2335 		ret = -EEXIST;
2336 	eventfd_ctx_put(eventfd);
2337 	return ret;
2338 }
2339 
2340 static int kvm_hv_eventfd_deassign(struct kvm *kvm, u32 conn_id)
2341 {
2342 	struct kvm_hv *hv = to_kvm_hv(kvm);
2343 	struct eventfd_ctx *eventfd;
2344 
2345 	mutex_lock(&hv->hv_lock);
2346 	eventfd = idr_remove(&hv->conn_to_evt, conn_id);
2347 	mutex_unlock(&hv->hv_lock);
2348 
2349 	if (!eventfd)
2350 		return -ENOENT;
2351 
2352 	synchronize_srcu(&kvm->srcu);
2353 	eventfd_ctx_put(eventfd);
2354 	return 0;
2355 }
2356 
2357 int kvm_vm_ioctl_hv_eventfd(struct kvm *kvm, struct kvm_hyperv_eventfd *args)
2358 {
2359 	if ((args->flags & ~KVM_HYPERV_EVENTFD_DEASSIGN) ||
2360 	    (args->conn_id & ~KVM_HYPERV_CONN_ID_MASK))
2361 		return -EINVAL;
2362 
2363 	if (args->flags == KVM_HYPERV_EVENTFD_DEASSIGN)
2364 		return kvm_hv_eventfd_deassign(kvm, args->conn_id);
2365 	return kvm_hv_eventfd_assign(kvm, args->conn_id, args->fd);
2366 }
2367 
2368 int kvm_get_hv_cpuid(struct kvm_vcpu *vcpu, struct kvm_cpuid2 *cpuid,
2369 		     struct kvm_cpuid_entry2 __user *entries)
2370 {
2371 	uint16_t evmcs_ver = 0;
2372 	struct kvm_cpuid_entry2 cpuid_entries[] = {
2373 		{ .function = HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS },
2374 		{ .function = HYPERV_CPUID_INTERFACE },
2375 		{ .function = HYPERV_CPUID_VERSION },
2376 		{ .function = HYPERV_CPUID_FEATURES },
2377 		{ .function = HYPERV_CPUID_ENLIGHTMENT_INFO },
2378 		{ .function = HYPERV_CPUID_IMPLEMENT_LIMITS },
2379 		{ .function = HYPERV_CPUID_SYNDBG_VENDOR_AND_MAX_FUNCTIONS },
2380 		{ .function = HYPERV_CPUID_SYNDBG_INTERFACE },
2381 		{ .function = HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES	},
2382 		{ .function = HYPERV_CPUID_NESTED_FEATURES },
2383 	};
2384 	int i, nent = ARRAY_SIZE(cpuid_entries);
2385 
2386 	if (kvm_x86_ops.nested_ops->get_evmcs_version)
2387 		evmcs_ver = kvm_x86_ops.nested_ops->get_evmcs_version(vcpu);
2388 
2389 	/* Skip NESTED_FEATURES if eVMCS is not supported */
2390 	if (!evmcs_ver)
2391 		--nent;
2392 
2393 	if (cpuid->nent < nent)
2394 		return -E2BIG;
2395 
2396 	if (cpuid->nent > nent)
2397 		cpuid->nent = nent;
2398 
2399 	for (i = 0; i < nent; i++) {
2400 		struct kvm_cpuid_entry2 *ent = &cpuid_entries[i];
2401 		u32 signature[3];
2402 
2403 		switch (ent->function) {
2404 		case HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS:
2405 			memcpy(signature, "Linux KVM Hv", 12);
2406 
2407 			ent->eax = HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES;
2408 			ent->ebx = signature[0];
2409 			ent->ecx = signature[1];
2410 			ent->edx = signature[2];
2411 			break;
2412 
2413 		case HYPERV_CPUID_INTERFACE:
2414 			ent->eax = HYPERV_CPUID_SIGNATURE_EAX;
2415 			break;
2416 
2417 		case HYPERV_CPUID_VERSION:
2418 			/*
2419 			 * We implement some Hyper-V 2016 functions so let's use
2420 			 * this version.
2421 			 */
2422 			ent->eax = 0x00003839;
2423 			ent->ebx = 0x000A0000;
2424 			break;
2425 
2426 		case HYPERV_CPUID_FEATURES:
2427 			ent->eax |= HV_MSR_VP_RUNTIME_AVAILABLE;
2428 			ent->eax |= HV_MSR_TIME_REF_COUNT_AVAILABLE;
2429 			ent->eax |= HV_MSR_SYNIC_AVAILABLE;
2430 			ent->eax |= HV_MSR_SYNTIMER_AVAILABLE;
2431 			ent->eax |= HV_MSR_APIC_ACCESS_AVAILABLE;
2432 			ent->eax |= HV_MSR_HYPERCALL_AVAILABLE;
2433 			ent->eax |= HV_MSR_VP_INDEX_AVAILABLE;
2434 			ent->eax |= HV_MSR_RESET_AVAILABLE;
2435 			ent->eax |= HV_MSR_REFERENCE_TSC_AVAILABLE;
2436 			ent->eax |= HV_ACCESS_FREQUENCY_MSRS;
2437 			ent->eax |= HV_ACCESS_REENLIGHTENMENT;
2438 
2439 			ent->ebx |= HV_POST_MESSAGES;
2440 			ent->ebx |= HV_SIGNAL_EVENTS;
2441 
2442 			ent->edx |= HV_X64_HYPERCALL_XMM_INPUT_AVAILABLE;
2443 			ent->edx |= HV_FEATURE_FREQUENCY_MSRS_AVAILABLE;
2444 			ent->edx |= HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
2445 
2446 			ent->ebx |= HV_DEBUGGING;
2447 			ent->edx |= HV_X64_GUEST_DEBUGGING_AVAILABLE;
2448 			ent->edx |= HV_FEATURE_DEBUG_MSRS_AVAILABLE;
2449 
2450 			/*
2451 			 * Direct Synthetic timers only make sense with in-kernel
2452 			 * LAPIC
2453 			 */
2454 			if (!vcpu || lapic_in_kernel(vcpu))
2455 				ent->edx |= HV_STIMER_DIRECT_MODE_AVAILABLE;
2456 
2457 			break;
2458 
2459 		case HYPERV_CPUID_ENLIGHTMENT_INFO:
2460 			ent->eax |= HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
2461 			ent->eax |= HV_X64_APIC_ACCESS_RECOMMENDED;
2462 			ent->eax |= HV_X64_RELAXED_TIMING_RECOMMENDED;
2463 			ent->eax |= HV_X64_CLUSTER_IPI_RECOMMENDED;
2464 			ent->eax |= HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED;
2465 			if (evmcs_ver)
2466 				ent->eax |= HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
2467 			if (!cpu_smt_possible())
2468 				ent->eax |= HV_X64_NO_NONARCH_CORESHARING;
2469 			/*
2470 			 * Default number of spinlock retry attempts, matches
2471 			 * HyperV 2016.
2472 			 */
2473 			ent->ebx = 0x00000FFF;
2474 
2475 			break;
2476 
2477 		case HYPERV_CPUID_IMPLEMENT_LIMITS:
2478 			/* Maximum number of virtual processors */
2479 			ent->eax = KVM_MAX_VCPUS;
2480 			/*
2481 			 * Maximum number of logical processors, matches
2482 			 * HyperV 2016.
2483 			 */
2484 			ent->ebx = 64;
2485 
2486 			break;
2487 
2488 		case HYPERV_CPUID_NESTED_FEATURES:
2489 			ent->eax = evmcs_ver;
2490 
2491 			break;
2492 
2493 		case HYPERV_CPUID_SYNDBG_VENDOR_AND_MAX_FUNCTIONS:
2494 			memcpy(signature, "Linux KVM Hv", 12);
2495 
2496 			ent->eax = 0;
2497 			ent->ebx = signature[0];
2498 			ent->ecx = signature[1];
2499 			ent->edx = signature[2];
2500 			break;
2501 
2502 		case HYPERV_CPUID_SYNDBG_INTERFACE:
2503 			memcpy(signature, "VS#1\0\0\0\0\0\0\0\0", 12);
2504 			ent->eax = signature[0];
2505 			break;
2506 
2507 		case HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES:
2508 			ent->eax |= HV_X64_SYNDBG_CAP_ALLOW_KERNEL_DEBUGGING;
2509 			break;
2510 
2511 		default:
2512 			break;
2513 		}
2514 	}
2515 
2516 	if (copy_to_user(entries, cpuid_entries,
2517 			 nent * sizeof(struct kvm_cpuid_entry2)))
2518 		return -EFAULT;
2519 
2520 	return 0;
2521 }
2522