xref: /openbmc/linux/arch/x86/kvm/hyperv.c (revision 6d99a79c)
1 /*
2  * KVM Microsoft Hyper-V emulation
3  *
4  * derived from arch/x86/kvm/x86.c
5  *
6  * Copyright (C) 2006 Qumranet, Inc.
7  * Copyright (C) 2008 Qumranet, Inc.
8  * Copyright IBM Corporation, 2008
9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10  * Copyright (C) 2015 Andrey Smetanin <asmetanin@virtuozzo.com>
11  *
12  * Authors:
13  *   Avi Kivity   <avi@qumranet.com>
14  *   Yaniv Kamay  <yaniv@qumranet.com>
15  *   Amit Shah    <amit.shah@qumranet.com>
16  *   Ben-Ami Yassour <benami@il.ibm.com>
17  *   Andrey Smetanin <asmetanin@virtuozzo.com>
18  *
19  * This work is licensed under the terms of the GNU GPL, version 2.  See
20  * the COPYING file in the top-level directory.
21  *
22  */
23 
24 #include "x86.h"
25 #include "lapic.h"
26 #include "ioapic.h"
27 #include "hyperv.h"
28 
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 
39 #define KVM_HV_MAX_SPARSE_VCPU_SET_BITS DIV_ROUND_UP(KVM_MAX_VCPUS, 64)
40 
41 static inline u64 synic_read_sint(struct kvm_vcpu_hv_synic *synic, int sint)
42 {
43 	return atomic64_read(&synic->sint[sint]);
44 }
45 
46 static inline int synic_get_sint_vector(u64 sint_value)
47 {
48 	if (sint_value & HV_SYNIC_SINT_MASKED)
49 		return -1;
50 	return sint_value & HV_SYNIC_SINT_VECTOR_MASK;
51 }
52 
53 static bool synic_has_vector_connected(struct kvm_vcpu_hv_synic *synic,
54 				      int vector)
55 {
56 	int i;
57 
58 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
59 		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
60 			return true;
61 	}
62 	return false;
63 }
64 
65 static bool synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic *synic,
66 				     int vector)
67 {
68 	int i;
69 	u64 sint_value;
70 
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)
75 			return true;
76 	}
77 	return false;
78 }
79 
80 static void synic_update_vector(struct kvm_vcpu_hv_synic *synic,
81 				int vector)
82 {
83 	if (vector < HV_SYNIC_FIRST_VALID_VECTOR)
84 		return;
85 
86 	if (synic_has_vector_connected(synic, vector))
87 		__set_bit(vector, synic->vec_bitmap);
88 	else
89 		__clear_bit(vector, synic->vec_bitmap);
90 
91 	if (synic_has_vector_auto_eoi(synic, vector))
92 		__set_bit(vector, synic->auto_eoi_bitmap);
93 	else
94 		__clear_bit(vector, synic->auto_eoi_bitmap);
95 }
96 
97 static int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint,
98 			  u64 data, bool host)
99 {
100 	int vector, old_vector;
101 	bool masked;
102 
103 	vector = data & HV_SYNIC_SINT_VECTOR_MASK;
104 	masked = data & HV_SYNIC_SINT_MASKED;
105 
106 	/*
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.
110 	 */
111 	if (vector < HV_SYNIC_FIRST_VALID_VECTOR && !host && !masked)
112 		return 1;
113 	/*
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.
118 	 */
119 	old_vector = synic_read_sint(synic, sint) & HV_SYNIC_SINT_VECTOR_MASK;
120 
121 	atomic64_set(&synic->sint[sint], data);
122 
123 	synic_update_vector(synic, old_vector);
124 
125 	synic_update_vector(synic, vector);
126 
127 	/* Load SynIC vectors into EOI exit bitmap */
128 	kvm_make_request(KVM_REQ_SCAN_IOAPIC, synic_to_vcpu(synic));
129 	return 0;
130 }
131 
132 static struct kvm_vcpu *get_vcpu_by_vpidx(struct kvm *kvm, u32 vpidx)
133 {
134 	struct kvm_vcpu *vcpu = NULL;
135 	int i;
136 
137 	if (vpidx >= KVM_MAX_VCPUS)
138 		return NULL;
139 
140 	vcpu = kvm_get_vcpu(kvm, vpidx);
141 	if (vcpu && vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
142 		return vcpu;
143 	kvm_for_each_vcpu(i, vcpu, kvm)
144 		if (vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
145 			return vcpu;
146 	return NULL;
147 }
148 
149 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vpidx)
150 {
151 	struct kvm_vcpu *vcpu;
152 	struct kvm_vcpu_hv_synic *synic;
153 
154 	vcpu = get_vcpu_by_vpidx(kvm, vpidx);
155 	if (!vcpu)
156 		return NULL;
157 	synic = vcpu_to_synic(vcpu);
158 	return (synic->active) ? synic : NULL;
159 }
160 
161 static void synic_clear_sint_msg_pending(struct kvm_vcpu_hv_synic *synic,
162 					u32 sint)
163 {
164 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
165 	struct page *page;
166 	gpa_t gpa;
167 	struct hv_message *msg;
168 	struct hv_message_page *msg_page;
169 
170 	gpa = synic->msg_page & PAGE_MASK;
171 	page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
172 	if (is_error_page(page)) {
173 		vcpu_err(vcpu, "Hyper-V SynIC can't get msg page, gpa 0x%llx\n",
174 			 gpa);
175 		return;
176 	}
177 	msg_page = kmap_atomic(page);
178 
179 	msg = &msg_page->sint_message[sint];
180 	msg->header.message_flags.msg_pending = 0;
181 
182 	kunmap_atomic(msg_page);
183 	kvm_release_page_dirty(page);
184 	kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
185 }
186 
187 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
188 {
189 	struct kvm *kvm = vcpu->kvm;
190 	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
191 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
192 	struct kvm_vcpu_hv_stimer *stimer;
193 	int gsi, idx, stimers_pending;
194 
195 	trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
196 
197 	if (synic->msg_page & HV_SYNIC_SIMP_ENABLE)
198 		synic_clear_sint_msg_pending(synic, sint);
199 
200 	/* Try to deliver pending Hyper-V SynIC timers messages */
201 	stimers_pending = 0;
202 	for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
203 		stimer = &hv_vcpu->stimer[idx];
204 		if (stimer->msg_pending &&
205 		    (stimer->config & HV_STIMER_ENABLE) &&
206 		    HV_STIMER_SINT(stimer->config) == sint) {
207 			set_bit(stimer->index,
208 				hv_vcpu->stimer_pending_bitmap);
209 			stimers_pending++;
210 		}
211 	}
212 	if (stimers_pending)
213 		kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
214 
215 	idx = srcu_read_lock(&kvm->irq_srcu);
216 	gsi = atomic_read(&synic->sint_to_gsi[sint]);
217 	if (gsi != -1)
218 		kvm_notify_acked_gsi(kvm, gsi);
219 	srcu_read_unlock(&kvm->irq_srcu, idx);
220 }
221 
222 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
223 {
224 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
225 	struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
226 
227 	hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
228 	hv_vcpu->exit.u.synic.msr = msr;
229 	hv_vcpu->exit.u.synic.control = synic->control;
230 	hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
231 	hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
232 
233 	kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
234 }
235 
236 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
237 			 u32 msr, u64 data, bool host)
238 {
239 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
240 	int ret;
241 
242 	if (!synic->active && !host)
243 		return 1;
244 
245 	trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
246 
247 	ret = 0;
248 	switch (msr) {
249 	case HV_X64_MSR_SCONTROL:
250 		synic->control = data;
251 		if (!host)
252 			synic_exit(synic, msr);
253 		break;
254 	case HV_X64_MSR_SVERSION:
255 		if (!host) {
256 			ret = 1;
257 			break;
258 		}
259 		synic->version = data;
260 		break;
261 	case HV_X64_MSR_SIEFP:
262 		if ((data & HV_SYNIC_SIEFP_ENABLE) && !host &&
263 		    !synic->dont_zero_synic_pages)
264 			if (kvm_clear_guest(vcpu->kvm,
265 					    data & PAGE_MASK, PAGE_SIZE)) {
266 				ret = 1;
267 				break;
268 			}
269 		synic->evt_page = data;
270 		if (!host)
271 			synic_exit(synic, msr);
272 		break;
273 	case HV_X64_MSR_SIMP:
274 		if ((data & HV_SYNIC_SIMP_ENABLE) && !host &&
275 		    !synic->dont_zero_synic_pages)
276 			if (kvm_clear_guest(vcpu->kvm,
277 					    data & PAGE_MASK, PAGE_SIZE)) {
278 				ret = 1;
279 				break;
280 			}
281 		synic->msg_page = data;
282 		if (!host)
283 			synic_exit(synic, msr);
284 		break;
285 	case HV_X64_MSR_EOM: {
286 		int i;
287 
288 		for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
289 			kvm_hv_notify_acked_sint(vcpu, i);
290 		break;
291 	}
292 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
293 		ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
294 		break;
295 	default:
296 		ret = 1;
297 		break;
298 	}
299 	return ret;
300 }
301 
302 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata,
303 			 bool host)
304 {
305 	int ret;
306 
307 	if (!synic->active && !host)
308 		return 1;
309 
310 	ret = 0;
311 	switch (msr) {
312 	case HV_X64_MSR_SCONTROL:
313 		*pdata = synic->control;
314 		break;
315 	case HV_X64_MSR_SVERSION:
316 		*pdata = synic->version;
317 		break;
318 	case HV_X64_MSR_SIEFP:
319 		*pdata = synic->evt_page;
320 		break;
321 	case HV_X64_MSR_SIMP:
322 		*pdata = synic->msg_page;
323 		break;
324 	case HV_X64_MSR_EOM:
325 		*pdata = 0;
326 		break;
327 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
328 		*pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
329 		break;
330 	default:
331 		ret = 1;
332 		break;
333 	}
334 	return ret;
335 }
336 
337 static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
338 {
339 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
340 	struct kvm_lapic_irq irq;
341 	int ret, vector;
342 
343 	if (sint >= ARRAY_SIZE(synic->sint))
344 		return -EINVAL;
345 
346 	vector = synic_get_sint_vector(synic_read_sint(synic, sint));
347 	if (vector < 0)
348 		return -ENOENT;
349 
350 	memset(&irq, 0, sizeof(irq));
351 	irq.shorthand = APIC_DEST_SELF;
352 	irq.dest_mode = APIC_DEST_PHYSICAL;
353 	irq.delivery_mode = APIC_DM_FIXED;
354 	irq.vector = vector;
355 	irq.level = 1;
356 
357 	ret = kvm_irq_delivery_to_apic(vcpu->kvm, vcpu->arch.apic, &irq, NULL);
358 	trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
359 	return ret;
360 }
361 
362 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vpidx, u32 sint)
363 {
364 	struct kvm_vcpu_hv_synic *synic;
365 
366 	synic = synic_get(kvm, vpidx);
367 	if (!synic)
368 		return -EINVAL;
369 
370 	return synic_set_irq(synic, sint);
371 }
372 
373 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
374 {
375 	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
376 	int i;
377 
378 	trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
379 
380 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
381 		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
382 			kvm_hv_notify_acked_sint(vcpu, i);
383 }
384 
385 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi)
386 {
387 	struct kvm_vcpu_hv_synic *synic;
388 
389 	synic = synic_get(kvm, vpidx);
390 	if (!synic)
391 		return -EINVAL;
392 
393 	if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
394 		return -EINVAL;
395 
396 	atomic_set(&synic->sint_to_gsi[sint], gsi);
397 	return 0;
398 }
399 
400 void kvm_hv_irq_routing_update(struct kvm *kvm)
401 {
402 	struct kvm_irq_routing_table *irq_rt;
403 	struct kvm_kernel_irq_routing_entry *e;
404 	u32 gsi;
405 
406 	irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
407 					lockdep_is_held(&kvm->irq_lock));
408 
409 	for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
410 		hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
411 			if (e->type == KVM_IRQ_ROUTING_HV_SINT)
412 				kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
413 						    e->hv_sint.sint, gsi);
414 		}
415 	}
416 }
417 
418 static void synic_init(struct kvm_vcpu_hv_synic *synic)
419 {
420 	int i;
421 
422 	memset(synic, 0, sizeof(*synic));
423 	synic->version = HV_SYNIC_VERSION_1;
424 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
425 		atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
426 		atomic_set(&synic->sint_to_gsi[i], -1);
427 	}
428 }
429 
430 static u64 get_time_ref_counter(struct kvm *kvm)
431 {
432 	struct kvm_hv *hv = &kvm->arch.hyperv;
433 	struct kvm_vcpu *vcpu;
434 	u64 tsc;
435 
436 	/*
437 	 * The guest has not set up the TSC page or the clock isn't
438 	 * stable, fall back to get_kvmclock_ns.
439 	 */
440 	if (!hv->tsc_ref.tsc_sequence)
441 		return div_u64(get_kvmclock_ns(kvm), 100);
442 
443 	vcpu = kvm_get_vcpu(kvm, 0);
444 	tsc = kvm_read_l1_tsc(vcpu, rdtsc());
445 	return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64)
446 		+ hv->tsc_ref.tsc_offset;
447 }
448 
449 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
450 				bool vcpu_kick)
451 {
452 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
453 
454 	set_bit(stimer->index,
455 		vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
456 	kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
457 	if (vcpu_kick)
458 		kvm_vcpu_kick(vcpu);
459 }
460 
461 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
462 {
463 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
464 
465 	trace_kvm_hv_stimer_cleanup(stimer_to_vcpu(stimer)->vcpu_id,
466 				    stimer->index);
467 
468 	hrtimer_cancel(&stimer->timer);
469 	clear_bit(stimer->index,
470 		  vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
471 	stimer->msg_pending = false;
472 	stimer->exp_time = 0;
473 }
474 
475 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
476 {
477 	struct kvm_vcpu_hv_stimer *stimer;
478 
479 	stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
480 	trace_kvm_hv_stimer_callback(stimer_to_vcpu(stimer)->vcpu_id,
481 				     stimer->index);
482 	stimer_mark_pending(stimer, true);
483 
484 	return HRTIMER_NORESTART;
485 }
486 
487 /*
488  * stimer_start() assumptions:
489  * a) stimer->count is not equal to 0
490  * b) stimer->config has HV_STIMER_ENABLE flag
491  */
492 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
493 {
494 	u64 time_now;
495 	ktime_t ktime_now;
496 
497 	time_now = get_time_ref_counter(stimer_to_vcpu(stimer)->kvm);
498 	ktime_now = ktime_get();
499 
500 	if (stimer->config & HV_STIMER_PERIODIC) {
501 		if (stimer->exp_time) {
502 			if (time_now >= stimer->exp_time) {
503 				u64 remainder;
504 
505 				div64_u64_rem(time_now - stimer->exp_time,
506 					      stimer->count, &remainder);
507 				stimer->exp_time =
508 					time_now + (stimer->count - remainder);
509 			}
510 		} else
511 			stimer->exp_time = time_now + stimer->count;
512 
513 		trace_kvm_hv_stimer_start_periodic(
514 					stimer_to_vcpu(stimer)->vcpu_id,
515 					stimer->index,
516 					time_now, stimer->exp_time);
517 
518 		hrtimer_start(&stimer->timer,
519 			      ktime_add_ns(ktime_now,
520 					   100 * (stimer->exp_time - time_now)),
521 			      HRTIMER_MODE_ABS);
522 		return 0;
523 	}
524 	stimer->exp_time = stimer->count;
525 	if (time_now >= stimer->count) {
526 		/*
527 		 * Expire timer according to Hypervisor Top-Level Functional
528 		 * specification v4(15.3.1):
529 		 * "If a one shot is enabled and the specified count is in
530 		 * the past, it will expire immediately."
531 		 */
532 		stimer_mark_pending(stimer, false);
533 		return 0;
534 	}
535 
536 	trace_kvm_hv_stimer_start_one_shot(stimer_to_vcpu(stimer)->vcpu_id,
537 					   stimer->index,
538 					   time_now, stimer->count);
539 
540 	hrtimer_start(&stimer->timer,
541 		      ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)),
542 		      HRTIMER_MODE_ABS);
543 	return 0;
544 }
545 
546 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
547 			     bool host)
548 {
549 	trace_kvm_hv_stimer_set_config(stimer_to_vcpu(stimer)->vcpu_id,
550 				       stimer->index, config, host);
551 
552 	stimer_cleanup(stimer);
553 	if ((stimer->config & HV_STIMER_ENABLE) && HV_STIMER_SINT(config) == 0)
554 		config &= ~HV_STIMER_ENABLE;
555 	stimer->config = config;
556 	stimer_mark_pending(stimer, false);
557 	return 0;
558 }
559 
560 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
561 			    bool host)
562 {
563 	trace_kvm_hv_stimer_set_count(stimer_to_vcpu(stimer)->vcpu_id,
564 				      stimer->index, count, host);
565 
566 	stimer_cleanup(stimer);
567 	stimer->count = count;
568 	if (stimer->count == 0)
569 		stimer->config &= ~HV_STIMER_ENABLE;
570 	else if (stimer->config & HV_STIMER_AUTOENABLE)
571 		stimer->config |= HV_STIMER_ENABLE;
572 	stimer_mark_pending(stimer, false);
573 	return 0;
574 }
575 
576 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
577 {
578 	*pconfig = stimer->config;
579 	return 0;
580 }
581 
582 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
583 {
584 	*pcount = stimer->count;
585 	return 0;
586 }
587 
588 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
589 			     struct hv_message *src_msg)
590 {
591 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
592 	struct page *page;
593 	gpa_t gpa;
594 	struct hv_message *dst_msg;
595 	int r;
596 	struct hv_message_page *msg_page;
597 
598 	if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
599 		return -ENOENT;
600 
601 	gpa = synic->msg_page & PAGE_MASK;
602 	page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
603 	if (is_error_page(page))
604 		return -EFAULT;
605 
606 	msg_page = kmap_atomic(page);
607 	dst_msg = &msg_page->sint_message[sint];
608 	if (sync_cmpxchg(&dst_msg->header.message_type, HVMSG_NONE,
609 			 src_msg->header.message_type) != HVMSG_NONE) {
610 		dst_msg->header.message_flags.msg_pending = 1;
611 		r = -EAGAIN;
612 	} else {
613 		memcpy(&dst_msg->u.payload, &src_msg->u.payload,
614 		       src_msg->header.payload_size);
615 		dst_msg->header.message_type = src_msg->header.message_type;
616 		dst_msg->header.payload_size = src_msg->header.payload_size;
617 		r = synic_set_irq(synic, sint);
618 		if (r >= 1)
619 			r = 0;
620 		else if (r == 0)
621 			r = -EFAULT;
622 	}
623 	kunmap_atomic(msg_page);
624 	kvm_release_page_dirty(page);
625 	kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
626 	return r;
627 }
628 
629 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
630 {
631 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
632 	struct hv_message *msg = &stimer->msg;
633 	struct hv_timer_message_payload *payload =
634 			(struct hv_timer_message_payload *)&msg->u.payload;
635 
636 	payload->expiration_time = stimer->exp_time;
637 	payload->delivery_time = get_time_ref_counter(vcpu->kvm);
638 	return synic_deliver_msg(vcpu_to_synic(vcpu),
639 				 HV_STIMER_SINT(stimer->config), msg);
640 }
641 
642 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
643 {
644 	int r;
645 
646 	stimer->msg_pending = true;
647 	r = stimer_send_msg(stimer);
648 	trace_kvm_hv_stimer_expiration(stimer_to_vcpu(stimer)->vcpu_id,
649 				       stimer->index, r);
650 	if (!r) {
651 		stimer->msg_pending = false;
652 		if (!(stimer->config & HV_STIMER_PERIODIC))
653 			stimer->config &= ~HV_STIMER_ENABLE;
654 	}
655 }
656 
657 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
658 {
659 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
660 	struct kvm_vcpu_hv_stimer *stimer;
661 	u64 time_now, exp_time;
662 	int i;
663 
664 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
665 		if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
666 			stimer = &hv_vcpu->stimer[i];
667 			if (stimer->config & HV_STIMER_ENABLE) {
668 				exp_time = stimer->exp_time;
669 
670 				if (exp_time) {
671 					time_now =
672 						get_time_ref_counter(vcpu->kvm);
673 					if (time_now >= exp_time)
674 						stimer_expiration(stimer);
675 				}
676 
677 				if ((stimer->config & HV_STIMER_ENABLE) &&
678 				    stimer->count) {
679 					if (!stimer->msg_pending)
680 						stimer_start(stimer);
681 				} else
682 					stimer_cleanup(stimer);
683 			}
684 		}
685 }
686 
687 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
688 {
689 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
690 	int i;
691 
692 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
693 		stimer_cleanup(&hv_vcpu->stimer[i]);
694 }
695 
696 bool kvm_hv_assist_page_enabled(struct kvm_vcpu *vcpu)
697 {
698 	if (!(vcpu->arch.hyperv.hv_vapic & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE))
699 		return false;
700 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
701 }
702 EXPORT_SYMBOL_GPL(kvm_hv_assist_page_enabled);
703 
704 bool kvm_hv_get_assist_page(struct kvm_vcpu *vcpu,
705 			    struct hv_vp_assist_page *assist_page)
706 {
707 	if (!kvm_hv_assist_page_enabled(vcpu))
708 		return false;
709 	return !kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data,
710 				      assist_page, sizeof(*assist_page));
711 }
712 EXPORT_SYMBOL_GPL(kvm_hv_get_assist_page);
713 
714 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
715 {
716 	struct hv_message *msg = &stimer->msg;
717 	struct hv_timer_message_payload *payload =
718 			(struct hv_timer_message_payload *)&msg->u.payload;
719 
720 	memset(&msg->header, 0, sizeof(msg->header));
721 	msg->header.message_type = HVMSG_TIMER_EXPIRED;
722 	msg->header.payload_size = sizeof(*payload);
723 
724 	payload->timer_index = stimer->index;
725 	payload->expiration_time = 0;
726 	payload->delivery_time = 0;
727 }
728 
729 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
730 {
731 	memset(stimer, 0, sizeof(*stimer));
732 	stimer->index = timer_index;
733 	hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
734 	stimer->timer.function = stimer_timer_callback;
735 	stimer_prepare_msg(stimer);
736 }
737 
738 void kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
739 {
740 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
741 	int i;
742 
743 	synic_init(&hv_vcpu->synic);
744 
745 	bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
746 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
747 		stimer_init(&hv_vcpu->stimer[i], i);
748 }
749 
750 void kvm_hv_vcpu_postcreate(struct kvm_vcpu *vcpu)
751 {
752 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
753 
754 	hv_vcpu->vp_index = kvm_vcpu_get_idx(vcpu);
755 }
756 
757 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
758 {
759 	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
760 
761 	/*
762 	 * Hyper-V SynIC auto EOI SINT's are
763 	 * not compatible with APICV, so deactivate APICV
764 	 */
765 	kvm_vcpu_deactivate_apicv(vcpu);
766 	synic->active = true;
767 	synic->dont_zero_synic_pages = dont_zero_synic_pages;
768 	return 0;
769 }
770 
771 static bool kvm_hv_msr_partition_wide(u32 msr)
772 {
773 	bool r = false;
774 
775 	switch (msr) {
776 	case HV_X64_MSR_GUEST_OS_ID:
777 	case HV_X64_MSR_HYPERCALL:
778 	case HV_X64_MSR_REFERENCE_TSC:
779 	case HV_X64_MSR_TIME_REF_COUNT:
780 	case HV_X64_MSR_CRASH_CTL:
781 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
782 	case HV_X64_MSR_RESET:
783 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
784 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
785 	case HV_X64_MSR_TSC_EMULATION_STATUS:
786 		r = true;
787 		break;
788 	}
789 
790 	return r;
791 }
792 
793 static int kvm_hv_msr_get_crash_data(struct kvm_vcpu *vcpu,
794 				     u32 index, u64 *pdata)
795 {
796 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
797 
798 	if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param)))
799 		return -EINVAL;
800 
801 	*pdata = hv->hv_crash_param[index];
802 	return 0;
803 }
804 
805 static int kvm_hv_msr_get_crash_ctl(struct kvm_vcpu *vcpu, u64 *pdata)
806 {
807 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
808 
809 	*pdata = hv->hv_crash_ctl;
810 	return 0;
811 }
812 
813 static int kvm_hv_msr_set_crash_ctl(struct kvm_vcpu *vcpu, u64 data, bool host)
814 {
815 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
816 
817 	if (host)
818 		hv->hv_crash_ctl = data & HV_X64_MSR_CRASH_CTL_NOTIFY;
819 
820 	if (!host && (data & HV_X64_MSR_CRASH_CTL_NOTIFY)) {
821 
822 		vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
823 			  hv->hv_crash_param[0],
824 			  hv->hv_crash_param[1],
825 			  hv->hv_crash_param[2],
826 			  hv->hv_crash_param[3],
827 			  hv->hv_crash_param[4]);
828 
829 		/* Send notification about crash to user space */
830 		kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
831 	}
832 
833 	return 0;
834 }
835 
836 static int kvm_hv_msr_set_crash_data(struct kvm_vcpu *vcpu,
837 				     u32 index, u64 data)
838 {
839 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
840 
841 	if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param)))
842 		return -EINVAL;
843 
844 	hv->hv_crash_param[index] = data;
845 	return 0;
846 }
847 
848 /*
849  * The kvmclock and Hyper-V TSC page use similar formulas, and converting
850  * between them is possible:
851  *
852  * kvmclock formula:
853  *    nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
854  *           + system_time
855  *
856  * Hyper-V formula:
857  *    nsec/100 = ticks * scale / 2^64 + offset
858  *
859  * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
860  * By dividing the kvmclock formula by 100 and equating what's left we get:
861  *    ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
862  *            scale / 2^64 =         tsc_to_system_mul * 2^(tsc_shift-32) / 100
863  *            scale        =         tsc_to_system_mul * 2^(32+tsc_shift) / 100
864  *
865  * Now expand the kvmclock formula and divide by 100:
866  *    nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
867  *           - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
868  *           + system_time
869  *    nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
870  *               - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
871  *               + system_time / 100
872  *
873  * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
874  *    nsec/100 = ticks * scale / 2^64
875  *               - tsc_timestamp * scale / 2^64
876  *               + system_time / 100
877  *
878  * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
879  *    offset = system_time / 100 - tsc_timestamp * scale / 2^64
880  *
881  * These two equivalencies are implemented in this function.
882  */
883 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
884 					HV_REFERENCE_TSC_PAGE *tsc_ref)
885 {
886 	u64 max_mul;
887 
888 	if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
889 		return false;
890 
891 	/*
892 	 * check if scale would overflow, if so we use the time ref counter
893 	 *    tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
894 	 *    tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
895 	 *    tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
896 	 */
897 	max_mul = 100ull << (32 - hv_clock->tsc_shift);
898 	if (hv_clock->tsc_to_system_mul >= max_mul)
899 		return false;
900 
901 	/*
902 	 * Otherwise compute the scale and offset according to the formulas
903 	 * derived above.
904 	 */
905 	tsc_ref->tsc_scale =
906 		mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
907 				hv_clock->tsc_to_system_mul,
908 				100);
909 
910 	tsc_ref->tsc_offset = hv_clock->system_time;
911 	do_div(tsc_ref->tsc_offset, 100);
912 	tsc_ref->tsc_offset -=
913 		mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
914 	return true;
915 }
916 
917 void kvm_hv_setup_tsc_page(struct kvm *kvm,
918 			   struct pvclock_vcpu_time_info *hv_clock)
919 {
920 	struct kvm_hv *hv = &kvm->arch.hyperv;
921 	u32 tsc_seq;
922 	u64 gfn;
923 
924 	BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
925 	BUILD_BUG_ON(offsetof(HV_REFERENCE_TSC_PAGE, tsc_sequence) != 0);
926 
927 	if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
928 		return;
929 
930 	mutex_lock(&kvm->arch.hyperv.hv_lock);
931 	if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
932 		goto out_unlock;
933 
934 	gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
935 	/*
936 	 * Because the TSC parameters only vary when there is a
937 	 * change in the master clock, do not bother with caching.
938 	 */
939 	if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
940 				    &tsc_seq, sizeof(tsc_seq))))
941 		goto out_unlock;
942 
943 	/*
944 	 * While we're computing and writing the parameters, force the
945 	 * guest to use the time reference count MSR.
946 	 */
947 	hv->tsc_ref.tsc_sequence = 0;
948 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
949 			    &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
950 		goto out_unlock;
951 
952 	if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
953 		goto out_unlock;
954 
955 	/* Ensure sequence is zero before writing the rest of the struct.  */
956 	smp_wmb();
957 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
958 		goto out_unlock;
959 
960 	/*
961 	 * Now switch to the TSC page mechanism by writing the sequence.
962 	 */
963 	tsc_seq++;
964 	if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
965 		tsc_seq = 1;
966 
967 	/* Write the struct entirely before the non-zero sequence.  */
968 	smp_wmb();
969 
970 	hv->tsc_ref.tsc_sequence = tsc_seq;
971 	kvm_write_guest(kvm, gfn_to_gpa(gfn),
972 			&hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence));
973 out_unlock:
974 	mutex_unlock(&kvm->arch.hyperv.hv_lock);
975 }
976 
977 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
978 			     bool host)
979 {
980 	struct kvm *kvm = vcpu->kvm;
981 	struct kvm_hv *hv = &kvm->arch.hyperv;
982 
983 	switch (msr) {
984 	case HV_X64_MSR_GUEST_OS_ID:
985 		hv->hv_guest_os_id = data;
986 		/* setting guest os id to zero disables hypercall page */
987 		if (!hv->hv_guest_os_id)
988 			hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
989 		break;
990 	case HV_X64_MSR_HYPERCALL: {
991 		u64 gfn;
992 		unsigned long addr;
993 		u8 instructions[4];
994 
995 		/* if guest os id is not set hypercall should remain disabled */
996 		if (!hv->hv_guest_os_id)
997 			break;
998 		if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
999 			hv->hv_hypercall = data;
1000 			break;
1001 		}
1002 		gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1003 		addr = gfn_to_hva(kvm, gfn);
1004 		if (kvm_is_error_hva(addr))
1005 			return 1;
1006 		kvm_x86_ops->patch_hypercall(vcpu, instructions);
1007 		((unsigned char *)instructions)[3] = 0xc3; /* ret */
1008 		if (__copy_to_user((void __user *)addr, instructions, 4))
1009 			return 1;
1010 		hv->hv_hypercall = data;
1011 		mark_page_dirty(kvm, gfn);
1012 		break;
1013 	}
1014 	case HV_X64_MSR_REFERENCE_TSC:
1015 		hv->hv_tsc_page = data;
1016 		if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE)
1017 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1018 		break;
1019 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1020 		return kvm_hv_msr_set_crash_data(vcpu,
1021 						 msr - HV_X64_MSR_CRASH_P0,
1022 						 data);
1023 	case HV_X64_MSR_CRASH_CTL:
1024 		return kvm_hv_msr_set_crash_ctl(vcpu, data, host);
1025 	case HV_X64_MSR_RESET:
1026 		if (data == 1) {
1027 			vcpu_debug(vcpu, "hyper-v reset requested\n");
1028 			kvm_make_request(KVM_REQ_HV_RESET, vcpu);
1029 		}
1030 		break;
1031 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1032 		hv->hv_reenlightenment_control = data;
1033 		break;
1034 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1035 		hv->hv_tsc_emulation_control = data;
1036 		break;
1037 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1038 		hv->hv_tsc_emulation_status = data;
1039 		break;
1040 	case HV_X64_MSR_TIME_REF_COUNT:
1041 		/* read-only, but still ignore it if host-initiated */
1042 		if (!host)
1043 			return 1;
1044 		break;
1045 	default:
1046 		vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
1047 			    msr, data);
1048 		return 1;
1049 	}
1050 	return 0;
1051 }
1052 
1053 /* Calculate cpu time spent by current task in 100ns units */
1054 static u64 current_task_runtime_100ns(void)
1055 {
1056 	u64 utime, stime;
1057 
1058 	task_cputime_adjusted(current, &utime, &stime);
1059 
1060 	return div_u64(utime + stime, 100);
1061 }
1062 
1063 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1064 {
1065 	struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
1066 
1067 	switch (msr) {
1068 	case HV_X64_MSR_VP_INDEX: {
1069 		struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
1070 		int vcpu_idx = kvm_vcpu_get_idx(vcpu);
1071 		u32 new_vp_index = (u32)data;
1072 
1073 		if (!host || new_vp_index >= KVM_MAX_VCPUS)
1074 			return 1;
1075 
1076 		if (new_vp_index == hv_vcpu->vp_index)
1077 			return 0;
1078 
1079 		/*
1080 		 * The VP index is initialized to vcpu_index by
1081 		 * kvm_hv_vcpu_postcreate so they initially match.  Now the
1082 		 * VP index is changing, adjust num_mismatched_vp_indexes if
1083 		 * it now matches or no longer matches vcpu_idx.
1084 		 */
1085 		if (hv_vcpu->vp_index == vcpu_idx)
1086 			atomic_inc(&hv->num_mismatched_vp_indexes);
1087 		else if (new_vp_index == vcpu_idx)
1088 			atomic_dec(&hv->num_mismatched_vp_indexes);
1089 
1090 		hv_vcpu->vp_index = new_vp_index;
1091 		break;
1092 	}
1093 	case HV_X64_MSR_VP_ASSIST_PAGE: {
1094 		u64 gfn;
1095 		unsigned long addr;
1096 
1097 		if (!(data & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE)) {
1098 			hv_vcpu->hv_vapic = data;
1099 			if (kvm_lapic_enable_pv_eoi(vcpu, 0, 0))
1100 				return 1;
1101 			break;
1102 		}
1103 		gfn = data >> HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT;
1104 		addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1105 		if (kvm_is_error_hva(addr))
1106 			return 1;
1107 
1108 		/*
1109 		 * Clear apic_assist portion of f(struct hv_vp_assist_page
1110 		 * only, there can be valuable data in the rest which needs
1111 		 * to be preserved e.g. on migration.
1112 		 */
1113 		if (__clear_user((void __user *)addr, sizeof(u32)))
1114 			return 1;
1115 		hv_vcpu->hv_vapic = data;
1116 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
1117 		if (kvm_lapic_enable_pv_eoi(vcpu,
1118 					    gfn_to_gpa(gfn) | KVM_MSR_ENABLED,
1119 					    sizeof(struct hv_vp_assist_page)))
1120 			return 1;
1121 		break;
1122 	}
1123 	case HV_X64_MSR_EOI:
1124 		return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1125 	case HV_X64_MSR_ICR:
1126 		return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1127 	case HV_X64_MSR_TPR:
1128 		return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1129 	case HV_X64_MSR_VP_RUNTIME:
1130 		if (!host)
1131 			return 1;
1132 		hv_vcpu->runtime_offset = data - current_task_runtime_100ns();
1133 		break;
1134 	case HV_X64_MSR_SCONTROL:
1135 	case HV_X64_MSR_SVERSION:
1136 	case HV_X64_MSR_SIEFP:
1137 	case HV_X64_MSR_SIMP:
1138 	case HV_X64_MSR_EOM:
1139 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1140 		return synic_set_msr(vcpu_to_synic(vcpu), msr, data, host);
1141 	case HV_X64_MSR_STIMER0_CONFIG:
1142 	case HV_X64_MSR_STIMER1_CONFIG:
1143 	case HV_X64_MSR_STIMER2_CONFIG:
1144 	case HV_X64_MSR_STIMER3_CONFIG: {
1145 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1146 
1147 		return stimer_set_config(vcpu_to_stimer(vcpu, timer_index),
1148 					 data, host);
1149 	}
1150 	case HV_X64_MSR_STIMER0_COUNT:
1151 	case HV_X64_MSR_STIMER1_COUNT:
1152 	case HV_X64_MSR_STIMER2_COUNT:
1153 	case HV_X64_MSR_STIMER3_COUNT: {
1154 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1155 
1156 		return stimer_set_count(vcpu_to_stimer(vcpu, timer_index),
1157 					data, host);
1158 	}
1159 	case HV_X64_MSR_TSC_FREQUENCY:
1160 	case HV_X64_MSR_APIC_FREQUENCY:
1161 		/* read-only, but still ignore it if host-initiated */
1162 		if (!host)
1163 			return 1;
1164 		break;
1165 	default:
1166 		vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
1167 			    msr, data);
1168 		return 1;
1169 	}
1170 
1171 	return 0;
1172 }
1173 
1174 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1175 {
1176 	u64 data = 0;
1177 	struct kvm *kvm = vcpu->kvm;
1178 	struct kvm_hv *hv = &kvm->arch.hyperv;
1179 
1180 	switch (msr) {
1181 	case HV_X64_MSR_GUEST_OS_ID:
1182 		data = hv->hv_guest_os_id;
1183 		break;
1184 	case HV_X64_MSR_HYPERCALL:
1185 		data = hv->hv_hypercall;
1186 		break;
1187 	case HV_X64_MSR_TIME_REF_COUNT:
1188 		data = get_time_ref_counter(kvm);
1189 		break;
1190 	case HV_X64_MSR_REFERENCE_TSC:
1191 		data = hv->hv_tsc_page;
1192 		break;
1193 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1194 		return kvm_hv_msr_get_crash_data(vcpu,
1195 						 msr - HV_X64_MSR_CRASH_P0,
1196 						 pdata);
1197 	case HV_X64_MSR_CRASH_CTL:
1198 		return kvm_hv_msr_get_crash_ctl(vcpu, pdata);
1199 	case HV_X64_MSR_RESET:
1200 		data = 0;
1201 		break;
1202 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1203 		data = hv->hv_reenlightenment_control;
1204 		break;
1205 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1206 		data = hv->hv_tsc_emulation_control;
1207 		break;
1208 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1209 		data = hv->hv_tsc_emulation_status;
1210 		break;
1211 	default:
1212 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1213 		return 1;
1214 	}
1215 
1216 	*pdata = data;
1217 	return 0;
1218 }
1219 
1220 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1221 			  bool host)
1222 {
1223 	u64 data = 0;
1224 	struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
1225 
1226 	switch (msr) {
1227 	case HV_X64_MSR_VP_INDEX:
1228 		data = hv_vcpu->vp_index;
1229 		break;
1230 	case HV_X64_MSR_EOI:
1231 		return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1232 	case HV_X64_MSR_ICR:
1233 		return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1234 	case HV_X64_MSR_TPR:
1235 		return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1236 	case HV_X64_MSR_VP_ASSIST_PAGE:
1237 		data = hv_vcpu->hv_vapic;
1238 		break;
1239 	case HV_X64_MSR_VP_RUNTIME:
1240 		data = current_task_runtime_100ns() + hv_vcpu->runtime_offset;
1241 		break;
1242 	case HV_X64_MSR_SCONTROL:
1243 	case HV_X64_MSR_SVERSION:
1244 	case HV_X64_MSR_SIEFP:
1245 	case HV_X64_MSR_SIMP:
1246 	case HV_X64_MSR_EOM:
1247 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1248 		return synic_get_msr(vcpu_to_synic(vcpu), msr, pdata, host);
1249 	case HV_X64_MSR_STIMER0_CONFIG:
1250 	case HV_X64_MSR_STIMER1_CONFIG:
1251 	case HV_X64_MSR_STIMER2_CONFIG:
1252 	case HV_X64_MSR_STIMER3_CONFIG: {
1253 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1254 
1255 		return stimer_get_config(vcpu_to_stimer(vcpu, timer_index),
1256 					 pdata);
1257 	}
1258 	case HV_X64_MSR_STIMER0_COUNT:
1259 	case HV_X64_MSR_STIMER1_COUNT:
1260 	case HV_X64_MSR_STIMER2_COUNT:
1261 	case HV_X64_MSR_STIMER3_COUNT: {
1262 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1263 
1264 		return stimer_get_count(vcpu_to_stimer(vcpu, timer_index),
1265 					pdata);
1266 	}
1267 	case HV_X64_MSR_TSC_FREQUENCY:
1268 		data = (u64)vcpu->arch.virtual_tsc_khz * 1000;
1269 		break;
1270 	case HV_X64_MSR_APIC_FREQUENCY:
1271 		data = APIC_BUS_FREQUENCY;
1272 		break;
1273 	default:
1274 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1275 		return 1;
1276 	}
1277 	*pdata = data;
1278 	return 0;
1279 }
1280 
1281 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1282 {
1283 	if (kvm_hv_msr_partition_wide(msr)) {
1284 		int r;
1285 
1286 		mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1287 		r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1288 		mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1289 		return r;
1290 	} else
1291 		return kvm_hv_set_msr(vcpu, msr, data, host);
1292 }
1293 
1294 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
1295 {
1296 	if (kvm_hv_msr_partition_wide(msr)) {
1297 		int r;
1298 
1299 		mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1300 		r = kvm_hv_get_msr_pw(vcpu, msr, pdata);
1301 		mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1302 		return r;
1303 	} else
1304 		return kvm_hv_get_msr(vcpu, msr, pdata, host);
1305 }
1306 
1307 static __always_inline unsigned long *sparse_set_to_vcpu_mask(
1308 	struct kvm *kvm, u64 *sparse_banks, u64 valid_bank_mask,
1309 	u64 *vp_bitmap, unsigned long *vcpu_bitmap)
1310 {
1311 	struct kvm_hv *hv = &kvm->arch.hyperv;
1312 	struct kvm_vcpu *vcpu;
1313 	int i, bank, sbank = 0;
1314 
1315 	memset(vp_bitmap, 0,
1316 	       KVM_HV_MAX_SPARSE_VCPU_SET_BITS * sizeof(*vp_bitmap));
1317 	for_each_set_bit(bank, (unsigned long *)&valid_bank_mask,
1318 			 KVM_HV_MAX_SPARSE_VCPU_SET_BITS)
1319 		vp_bitmap[bank] = sparse_banks[sbank++];
1320 
1321 	if (likely(!atomic_read(&hv->num_mismatched_vp_indexes))) {
1322 		/* for all vcpus vp_index == vcpu_idx */
1323 		return (unsigned long *)vp_bitmap;
1324 	}
1325 
1326 	bitmap_zero(vcpu_bitmap, KVM_MAX_VCPUS);
1327 	kvm_for_each_vcpu(i, vcpu, kvm) {
1328 		if (test_bit(vcpu_to_hv_vcpu(vcpu)->vp_index,
1329 			     (unsigned long *)vp_bitmap))
1330 			__set_bit(i, vcpu_bitmap);
1331 	}
1332 	return vcpu_bitmap;
1333 }
1334 
1335 static u64 kvm_hv_flush_tlb(struct kvm_vcpu *current_vcpu, u64 ingpa,
1336 			    u16 rep_cnt, bool ex)
1337 {
1338 	struct kvm *kvm = current_vcpu->kvm;
1339 	struct kvm_vcpu_hv *hv_vcpu = &current_vcpu->arch.hyperv;
1340 	struct hv_tlb_flush_ex flush_ex;
1341 	struct hv_tlb_flush flush;
1342 	u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1343 	DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1344 	unsigned long *vcpu_mask;
1345 	u64 valid_bank_mask;
1346 	u64 sparse_banks[64];
1347 	int sparse_banks_len;
1348 	bool all_cpus;
1349 
1350 	if (!ex) {
1351 		if (unlikely(kvm_read_guest(kvm, ingpa, &flush, sizeof(flush))))
1352 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1353 
1354 		trace_kvm_hv_flush_tlb(flush.processor_mask,
1355 				       flush.address_space, flush.flags);
1356 
1357 		valid_bank_mask = BIT_ULL(0);
1358 		sparse_banks[0] = flush.processor_mask;
1359 		all_cpus = flush.flags & HV_FLUSH_ALL_PROCESSORS;
1360 	} else {
1361 		if (unlikely(kvm_read_guest(kvm, ingpa, &flush_ex,
1362 					    sizeof(flush_ex))))
1363 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1364 
1365 		trace_kvm_hv_flush_tlb_ex(flush_ex.hv_vp_set.valid_bank_mask,
1366 					  flush_ex.hv_vp_set.format,
1367 					  flush_ex.address_space,
1368 					  flush_ex.flags);
1369 
1370 		valid_bank_mask = flush_ex.hv_vp_set.valid_bank_mask;
1371 		all_cpus = flush_ex.hv_vp_set.format !=
1372 			HV_GENERIC_SET_SPARSE_4K;
1373 
1374 		sparse_banks_len =
1375 			bitmap_weight((unsigned long *)&valid_bank_mask, 64) *
1376 			sizeof(sparse_banks[0]);
1377 
1378 		if (!sparse_banks_len && !all_cpus)
1379 			goto ret_success;
1380 
1381 		if (!all_cpus &&
1382 		    kvm_read_guest(kvm,
1383 				   ingpa + offsetof(struct hv_tlb_flush_ex,
1384 						    hv_vp_set.bank_contents),
1385 				   sparse_banks,
1386 				   sparse_banks_len))
1387 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1388 	}
1389 
1390 	cpumask_clear(&hv_vcpu->tlb_flush);
1391 
1392 	vcpu_mask = all_cpus ? NULL :
1393 		sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1394 					vp_bitmap, vcpu_bitmap);
1395 
1396 	/*
1397 	 * vcpu->arch.cr3 may not be up-to-date for running vCPUs so we can't
1398 	 * analyze it here, flush TLB regardless of the specified address space.
1399 	 */
1400 	kvm_make_vcpus_request_mask(kvm,
1401 				    KVM_REQ_TLB_FLUSH | KVM_REQUEST_NO_WAKEUP,
1402 				    vcpu_mask, &hv_vcpu->tlb_flush);
1403 
1404 ret_success:
1405 	/* We always do full TLB flush, set rep_done = rep_cnt. */
1406 	return (u64)HV_STATUS_SUCCESS |
1407 		((u64)rep_cnt << HV_HYPERCALL_REP_COMP_OFFSET);
1408 }
1409 
1410 static void kvm_send_ipi_to_many(struct kvm *kvm, u32 vector,
1411 				 unsigned long *vcpu_bitmap)
1412 {
1413 	struct kvm_lapic_irq irq = {
1414 		.delivery_mode = APIC_DM_FIXED,
1415 		.vector = vector
1416 	};
1417 	struct kvm_vcpu *vcpu;
1418 	int i;
1419 
1420 	kvm_for_each_vcpu(i, vcpu, kvm) {
1421 		if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
1422 			continue;
1423 
1424 		/* We fail only when APIC is disabled */
1425 		kvm_apic_set_irq(vcpu, &irq, NULL);
1426 	}
1427 }
1428 
1429 static u64 kvm_hv_send_ipi(struct kvm_vcpu *current_vcpu, u64 ingpa, u64 outgpa,
1430 			   bool ex, bool fast)
1431 {
1432 	struct kvm *kvm = current_vcpu->kvm;
1433 	struct hv_send_ipi_ex send_ipi_ex;
1434 	struct hv_send_ipi send_ipi;
1435 	u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1436 	DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1437 	unsigned long *vcpu_mask;
1438 	unsigned long valid_bank_mask;
1439 	u64 sparse_banks[64];
1440 	int sparse_banks_len;
1441 	u32 vector;
1442 	bool all_cpus;
1443 
1444 	if (!ex) {
1445 		if (!fast) {
1446 			if (unlikely(kvm_read_guest(kvm, ingpa, &send_ipi,
1447 						    sizeof(send_ipi))))
1448 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
1449 			sparse_banks[0] = send_ipi.cpu_mask;
1450 			vector = send_ipi.vector;
1451 		} else {
1452 			/* 'reserved' part of hv_send_ipi should be 0 */
1453 			if (unlikely(ingpa >> 32 != 0))
1454 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
1455 			sparse_banks[0] = outgpa;
1456 			vector = (u32)ingpa;
1457 		}
1458 		all_cpus = false;
1459 		valid_bank_mask = BIT_ULL(0);
1460 
1461 		trace_kvm_hv_send_ipi(vector, sparse_banks[0]);
1462 	} else {
1463 		if (unlikely(kvm_read_guest(kvm, ingpa, &send_ipi_ex,
1464 					    sizeof(send_ipi_ex))))
1465 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1466 
1467 		trace_kvm_hv_send_ipi_ex(send_ipi_ex.vector,
1468 					 send_ipi_ex.vp_set.format,
1469 					 send_ipi_ex.vp_set.valid_bank_mask);
1470 
1471 		vector = send_ipi_ex.vector;
1472 		valid_bank_mask = send_ipi_ex.vp_set.valid_bank_mask;
1473 		sparse_banks_len = bitmap_weight(&valid_bank_mask, 64) *
1474 			sizeof(sparse_banks[0]);
1475 
1476 		all_cpus = send_ipi_ex.vp_set.format == HV_GENERIC_SET_ALL;
1477 
1478 		if (!sparse_banks_len)
1479 			goto ret_success;
1480 
1481 		if (!all_cpus &&
1482 		    kvm_read_guest(kvm,
1483 				   ingpa + offsetof(struct hv_send_ipi_ex,
1484 						    vp_set.bank_contents),
1485 				   sparse_banks,
1486 				   sparse_banks_len))
1487 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1488 	}
1489 
1490 	if ((vector < HV_IPI_LOW_VECTOR) || (vector > HV_IPI_HIGH_VECTOR))
1491 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
1492 
1493 	vcpu_mask = all_cpus ? NULL :
1494 		sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1495 					vp_bitmap, vcpu_bitmap);
1496 
1497 	kvm_send_ipi_to_many(kvm, vector, vcpu_mask);
1498 
1499 ret_success:
1500 	return HV_STATUS_SUCCESS;
1501 }
1502 
1503 bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1504 {
1505 	return READ_ONCE(kvm->arch.hyperv.hv_hypercall) & HV_X64_MSR_HYPERCALL_ENABLE;
1506 }
1507 
1508 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
1509 {
1510 	bool longmode;
1511 
1512 	longmode = is_64_bit_mode(vcpu);
1513 	if (longmode)
1514 		kvm_register_write(vcpu, VCPU_REGS_RAX, result);
1515 	else {
1516 		kvm_register_write(vcpu, VCPU_REGS_RDX, result >> 32);
1517 		kvm_register_write(vcpu, VCPU_REGS_RAX, result & 0xffffffff);
1518 	}
1519 }
1520 
1521 static int kvm_hv_hypercall_complete(struct kvm_vcpu *vcpu, u64 result)
1522 {
1523 	kvm_hv_hypercall_set_result(vcpu, result);
1524 	++vcpu->stat.hypercalls;
1525 	return kvm_skip_emulated_instruction(vcpu);
1526 }
1527 
1528 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
1529 {
1530 	return kvm_hv_hypercall_complete(vcpu, vcpu->run->hyperv.u.hcall.result);
1531 }
1532 
1533 static u16 kvm_hvcall_signal_event(struct kvm_vcpu *vcpu, bool fast, u64 param)
1534 {
1535 	struct eventfd_ctx *eventfd;
1536 
1537 	if (unlikely(!fast)) {
1538 		int ret;
1539 		gpa_t gpa = param;
1540 
1541 		if ((gpa & (__alignof__(param) - 1)) ||
1542 		    offset_in_page(gpa) + sizeof(param) > PAGE_SIZE)
1543 			return HV_STATUS_INVALID_ALIGNMENT;
1544 
1545 		ret = kvm_vcpu_read_guest(vcpu, gpa, &param, sizeof(param));
1546 		if (ret < 0)
1547 			return HV_STATUS_INVALID_ALIGNMENT;
1548 	}
1549 
1550 	/*
1551 	 * Per spec, bits 32-47 contain the extra "flag number".  However, we
1552 	 * have no use for it, and in all known usecases it is zero, so just
1553 	 * report lookup failure if it isn't.
1554 	 */
1555 	if (param & 0xffff00000000ULL)
1556 		return HV_STATUS_INVALID_PORT_ID;
1557 	/* remaining bits are reserved-zero */
1558 	if (param & ~KVM_HYPERV_CONN_ID_MASK)
1559 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
1560 
1561 	/* the eventfd is protected by vcpu->kvm->srcu, but conn_to_evt isn't */
1562 	rcu_read_lock();
1563 	eventfd = idr_find(&vcpu->kvm->arch.hyperv.conn_to_evt, param);
1564 	rcu_read_unlock();
1565 	if (!eventfd)
1566 		return HV_STATUS_INVALID_PORT_ID;
1567 
1568 	eventfd_signal(eventfd, 1);
1569 	return HV_STATUS_SUCCESS;
1570 }
1571 
1572 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
1573 {
1574 	u64 param, ingpa, outgpa, ret = HV_STATUS_SUCCESS;
1575 	uint16_t code, rep_idx, rep_cnt;
1576 	bool fast, longmode, rep;
1577 
1578 	/*
1579 	 * hypercall generates UD from non zero cpl and real mode
1580 	 * per HYPER-V spec
1581 	 */
1582 	if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
1583 		kvm_queue_exception(vcpu, UD_VECTOR);
1584 		return 1;
1585 	}
1586 
1587 	longmode = is_64_bit_mode(vcpu);
1588 
1589 	if (!longmode) {
1590 		param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
1591 			(kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
1592 		ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
1593 			(kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
1594 		outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
1595 			(kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
1596 	}
1597 #ifdef CONFIG_X86_64
1598 	else {
1599 		param = kvm_register_read(vcpu, VCPU_REGS_RCX);
1600 		ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
1601 		outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
1602 	}
1603 #endif
1604 
1605 	code = param & 0xffff;
1606 	fast = !!(param & HV_HYPERCALL_FAST_BIT);
1607 	rep_cnt = (param >> HV_HYPERCALL_REP_COMP_OFFSET) & 0xfff;
1608 	rep_idx = (param >> HV_HYPERCALL_REP_START_OFFSET) & 0xfff;
1609 	rep = !!(rep_cnt || rep_idx);
1610 
1611 	trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
1612 
1613 	switch (code) {
1614 	case HVCALL_NOTIFY_LONG_SPIN_WAIT:
1615 		if (unlikely(rep)) {
1616 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1617 			break;
1618 		}
1619 		kvm_vcpu_on_spin(vcpu, true);
1620 		break;
1621 	case HVCALL_SIGNAL_EVENT:
1622 		if (unlikely(rep)) {
1623 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1624 			break;
1625 		}
1626 		ret = kvm_hvcall_signal_event(vcpu, fast, ingpa);
1627 		if (ret != HV_STATUS_INVALID_PORT_ID)
1628 			break;
1629 		/* maybe userspace knows this conn_id: fall through */
1630 	case HVCALL_POST_MESSAGE:
1631 		/* don't bother userspace if it has no way to handle it */
1632 		if (unlikely(rep || !vcpu_to_synic(vcpu)->active)) {
1633 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1634 			break;
1635 		}
1636 		vcpu->run->exit_reason = KVM_EXIT_HYPERV;
1637 		vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
1638 		vcpu->run->hyperv.u.hcall.input = param;
1639 		vcpu->run->hyperv.u.hcall.params[0] = ingpa;
1640 		vcpu->run->hyperv.u.hcall.params[1] = outgpa;
1641 		vcpu->arch.complete_userspace_io =
1642 				kvm_hv_hypercall_complete_userspace;
1643 		return 0;
1644 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
1645 		if (unlikely(fast || !rep_cnt || rep_idx)) {
1646 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1647 			break;
1648 		}
1649 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false);
1650 		break;
1651 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
1652 		if (unlikely(fast || rep)) {
1653 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1654 			break;
1655 		}
1656 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false);
1657 		break;
1658 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
1659 		if (unlikely(fast || !rep_cnt || rep_idx)) {
1660 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1661 			break;
1662 		}
1663 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
1664 		break;
1665 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
1666 		if (unlikely(fast || rep)) {
1667 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1668 			break;
1669 		}
1670 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
1671 		break;
1672 	case HVCALL_SEND_IPI:
1673 		if (unlikely(rep)) {
1674 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1675 			break;
1676 		}
1677 		ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, false, fast);
1678 		break;
1679 	case HVCALL_SEND_IPI_EX:
1680 		if (unlikely(fast || rep)) {
1681 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1682 			break;
1683 		}
1684 		ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, true, false);
1685 		break;
1686 	default:
1687 		ret = HV_STATUS_INVALID_HYPERCALL_CODE;
1688 		break;
1689 	}
1690 
1691 	return kvm_hv_hypercall_complete(vcpu, ret);
1692 }
1693 
1694 void kvm_hv_init_vm(struct kvm *kvm)
1695 {
1696 	mutex_init(&kvm->arch.hyperv.hv_lock);
1697 	idr_init(&kvm->arch.hyperv.conn_to_evt);
1698 }
1699 
1700 void kvm_hv_destroy_vm(struct kvm *kvm)
1701 {
1702 	struct eventfd_ctx *eventfd;
1703 	int i;
1704 
1705 	idr_for_each_entry(&kvm->arch.hyperv.conn_to_evt, eventfd, i)
1706 		eventfd_ctx_put(eventfd);
1707 	idr_destroy(&kvm->arch.hyperv.conn_to_evt);
1708 }
1709 
1710 static int kvm_hv_eventfd_assign(struct kvm *kvm, u32 conn_id, int fd)
1711 {
1712 	struct kvm_hv *hv = &kvm->arch.hyperv;
1713 	struct eventfd_ctx *eventfd;
1714 	int ret;
1715 
1716 	eventfd = eventfd_ctx_fdget(fd);
1717 	if (IS_ERR(eventfd))
1718 		return PTR_ERR(eventfd);
1719 
1720 	mutex_lock(&hv->hv_lock);
1721 	ret = idr_alloc(&hv->conn_to_evt, eventfd, conn_id, conn_id + 1,
1722 			GFP_KERNEL);
1723 	mutex_unlock(&hv->hv_lock);
1724 
1725 	if (ret >= 0)
1726 		return 0;
1727 
1728 	if (ret == -ENOSPC)
1729 		ret = -EEXIST;
1730 	eventfd_ctx_put(eventfd);
1731 	return ret;
1732 }
1733 
1734 static int kvm_hv_eventfd_deassign(struct kvm *kvm, u32 conn_id)
1735 {
1736 	struct kvm_hv *hv = &kvm->arch.hyperv;
1737 	struct eventfd_ctx *eventfd;
1738 
1739 	mutex_lock(&hv->hv_lock);
1740 	eventfd = idr_remove(&hv->conn_to_evt, conn_id);
1741 	mutex_unlock(&hv->hv_lock);
1742 
1743 	if (!eventfd)
1744 		return -ENOENT;
1745 
1746 	synchronize_srcu(&kvm->srcu);
1747 	eventfd_ctx_put(eventfd);
1748 	return 0;
1749 }
1750 
1751 int kvm_vm_ioctl_hv_eventfd(struct kvm *kvm, struct kvm_hyperv_eventfd *args)
1752 {
1753 	if ((args->flags & ~KVM_HYPERV_EVENTFD_DEASSIGN) ||
1754 	    (args->conn_id & ~KVM_HYPERV_CONN_ID_MASK))
1755 		return -EINVAL;
1756 
1757 	if (args->flags == KVM_HYPERV_EVENTFD_DEASSIGN)
1758 		return kvm_hv_eventfd_deassign(kvm, args->conn_id);
1759 	return kvm_hv_eventfd_assign(kvm, args->conn_id, args->fd);
1760 }
1761