xref: /linux/arch/x86/kvm/hyperv.c (revision c5951e7c8ee5cb04b8b41c32bf567b90117a2124)
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 "hyperv.h"
25 
26 #include <linux/cpu.h>
27 #include <linux/kvm_host.h>
28 #include <linux/highmem.h>
29 #include <linux/sched/cputime.h>
30 #include <linux/eventfd.h>
31 
32 #include <asm/apicdef.h>
33 #include <trace/events/kvm.h>
34 
35 #include "trace.h"
36 #include "irq.h"
37 
38 #define KVM_HV_MAX_SPARSE_VCPU_SET_BITS DIV_ROUND_UP(KVM_MAX_VCPUS, 64)
39 
40 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
41 				bool vcpu_kick);
42 
43 static inline u64 synic_read_sint(struct kvm_vcpu_hv_synic *synic, int sint)
44 {
45 	return atomic64_read(&synic->sint[sint]);
46 }
47 
48 static inline int synic_get_sint_vector(u64 sint_value)
49 {
50 	if (sint_value & HV_SYNIC_SINT_MASKED)
51 		return -1;
52 	return sint_value & HV_SYNIC_SINT_VECTOR_MASK;
53 }
54 
55 static bool synic_has_vector_connected(struct kvm_vcpu_hv_synic *synic,
56 				      int vector)
57 {
58 	int i;
59 
60 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
61 		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
62 			return true;
63 	}
64 	return false;
65 }
66 
67 static bool synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic *synic,
68 				     int vector)
69 {
70 	int i;
71 	u64 sint_value;
72 
73 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
74 		sint_value = synic_read_sint(synic, i);
75 		if (synic_get_sint_vector(sint_value) == vector &&
76 		    sint_value & HV_SYNIC_SINT_AUTO_EOI)
77 			return true;
78 	}
79 	return false;
80 }
81 
82 static void synic_update_vector(struct kvm_vcpu_hv_synic *synic,
83 				int vector)
84 {
85 	if (vector < HV_SYNIC_FIRST_VALID_VECTOR)
86 		return;
87 
88 	if (synic_has_vector_connected(synic, vector))
89 		__set_bit(vector, synic->vec_bitmap);
90 	else
91 		__clear_bit(vector, synic->vec_bitmap);
92 
93 	if (synic_has_vector_auto_eoi(synic, vector))
94 		__set_bit(vector, synic->auto_eoi_bitmap);
95 	else
96 		__clear_bit(vector, synic->auto_eoi_bitmap);
97 }
98 
99 static int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint,
100 			  u64 data, bool host)
101 {
102 	int vector, old_vector;
103 	bool masked;
104 
105 	vector = data & HV_SYNIC_SINT_VECTOR_MASK;
106 	masked = data & HV_SYNIC_SINT_MASKED;
107 
108 	/*
109 	 * Valid vectors are 16-255, however, nested Hyper-V attempts to write
110 	 * default '0x10000' value on boot and this should not #GP. We need to
111 	 * allow zero-initing the register from host as well.
112 	 */
113 	if (vector < HV_SYNIC_FIRST_VALID_VECTOR && !host && !masked)
114 		return 1;
115 	/*
116 	 * Guest may configure multiple SINTs to use the same vector, so
117 	 * we maintain a bitmap of vectors handled by synic, and a
118 	 * bitmap of vectors with auto-eoi behavior.  The bitmaps are
119 	 * updated here, and atomically queried on fast paths.
120 	 */
121 	old_vector = synic_read_sint(synic, sint) & HV_SYNIC_SINT_VECTOR_MASK;
122 
123 	atomic64_set(&synic->sint[sint], data);
124 
125 	synic_update_vector(synic, old_vector);
126 
127 	synic_update_vector(synic, vector);
128 
129 	/* Load SynIC vectors into EOI exit bitmap */
130 	kvm_make_request(KVM_REQ_SCAN_IOAPIC, synic_to_vcpu(synic));
131 	return 0;
132 }
133 
134 static struct kvm_vcpu *get_vcpu_by_vpidx(struct kvm *kvm, u32 vpidx)
135 {
136 	struct kvm_vcpu *vcpu = NULL;
137 	int i;
138 
139 	if (vpidx >= KVM_MAX_VCPUS)
140 		return NULL;
141 
142 	vcpu = kvm_get_vcpu(kvm, vpidx);
143 	if (vcpu && vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
144 		return vcpu;
145 	kvm_for_each_vcpu(i, vcpu, kvm)
146 		if (vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
147 			return vcpu;
148 	return NULL;
149 }
150 
151 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vpidx)
152 {
153 	struct kvm_vcpu *vcpu;
154 	struct kvm_vcpu_hv_synic *synic;
155 
156 	vcpu = get_vcpu_by_vpidx(kvm, vpidx);
157 	if (!vcpu)
158 		return NULL;
159 	synic = vcpu_to_synic(vcpu);
160 	return (synic->active) ? synic : NULL;
161 }
162 
163 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
164 {
165 	struct kvm *kvm = vcpu->kvm;
166 	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
167 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
168 	struct kvm_vcpu_hv_stimer *stimer;
169 	int gsi, idx;
170 
171 	trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
172 
173 	/* Try to deliver pending Hyper-V SynIC timers messages */
174 	for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
175 		stimer = &hv_vcpu->stimer[idx];
176 		if (stimer->msg_pending && stimer->config.enable &&
177 		    !stimer->config.direct_mode &&
178 		    stimer->config.sintx == sint)
179 			stimer_mark_pending(stimer, false);
180 	}
181 
182 	idx = srcu_read_lock(&kvm->irq_srcu);
183 	gsi = atomic_read(&synic->sint_to_gsi[sint]);
184 	if (gsi != -1)
185 		kvm_notify_acked_gsi(kvm, gsi);
186 	srcu_read_unlock(&kvm->irq_srcu, idx);
187 }
188 
189 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
190 {
191 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
192 	struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
193 
194 	hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
195 	hv_vcpu->exit.u.synic.msr = msr;
196 	hv_vcpu->exit.u.synic.control = synic->control;
197 	hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
198 	hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
199 
200 	kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
201 }
202 
203 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
204 			 u32 msr, u64 data, bool host)
205 {
206 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
207 	int ret;
208 
209 	if (!synic->active && !host)
210 		return 1;
211 
212 	trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
213 
214 	ret = 0;
215 	switch (msr) {
216 	case HV_X64_MSR_SCONTROL:
217 		synic->control = data;
218 		if (!host)
219 			synic_exit(synic, msr);
220 		break;
221 	case HV_X64_MSR_SVERSION:
222 		if (!host) {
223 			ret = 1;
224 			break;
225 		}
226 		synic->version = data;
227 		break;
228 	case HV_X64_MSR_SIEFP:
229 		if ((data & HV_SYNIC_SIEFP_ENABLE) && !host &&
230 		    !synic->dont_zero_synic_pages)
231 			if (kvm_clear_guest(vcpu->kvm,
232 					    data & PAGE_MASK, PAGE_SIZE)) {
233 				ret = 1;
234 				break;
235 			}
236 		synic->evt_page = data;
237 		if (!host)
238 			synic_exit(synic, msr);
239 		break;
240 	case HV_X64_MSR_SIMP:
241 		if ((data & HV_SYNIC_SIMP_ENABLE) && !host &&
242 		    !synic->dont_zero_synic_pages)
243 			if (kvm_clear_guest(vcpu->kvm,
244 					    data & PAGE_MASK, PAGE_SIZE)) {
245 				ret = 1;
246 				break;
247 			}
248 		synic->msg_page = data;
249 		if (!host)
250 			synic_exit(synic, msr);
251 		break;
252 	case HV_X64_MSR_EOM: {
253 		int i;
254 
255 		for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
256 			kvm_hv_notify_acked_sint(vcpu, i);
257 		break;
258 	}
259 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
260 		ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
261 		break;
262 	default:
263 		ret = 1;
264 		break;
265 	}
266 	return ret;
267 }
268 
269 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata,
270 			 bool host)
271 {
272 	int ret;
273 
274 	if (!synic->active && !host)
275 		return 1;
276 
277 	ret = 0;
278 	switch (msr) {
279 	case HV_X64_MSR_SCONTROL:
280 		*pdata = synic->control;
281 		break;
282 	case HV_X64_MSR_SVERSION:
283 		*pdata = synic->version;
284 		break;
285 	case HV_X64_MSR_SIEFP:
286 		*pdata = synic->evt_page;
287 		break;
288 	case HV_X64_MSR_SIMP:
289 		*pdata = synic->msg_page;
290 		break;
291 	case HV_X64_MSR_EOM:
292 		*pdata = 0;
293 		break;
294 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
295 		*pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
296 		break;
297 	default:
298 		ret = 1;
299 		break;
300 	}
301 	return ret;
302 }
303 
304 static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
305 {
306 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
307 	struct kvm_lapic_irq irq;
308 	int ret, vector;
309 
310 	if (sint >= ARRAY_SIZE(synic->sint))
311 		return -EINVAL;
312 
313 	vector = synic_get_sint_vector(synic_read_sint(synic, sint));
314 	if (vector < 0)
315 		return -ENOENT;
316 
317 	memset(&irq, 0, sizeof(irq));
318 	irq.shorthand = APIC_DEST_SELF;
319 	irq.dest_mode = APIC_DEST_PHYSICAL;
320 	irq.delivery_mode = APIC_DM_FIXED;
321 	irq.vector = vector;
322 	irq.level = 1;
323 
324 	ret = kvm_irq_delivery_to_apic(vcpu->kvm, vcpu->arch.apic, &irq, NULL);
325 	trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
326 	return ret;
327 }
328 
329 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vpidx, u32 sint)
330 {
331 	struct kvm_vcpu_hv_synic *synic;
332 
333 	synic = synic_get(kvm, vpidx);
334 	if (!synic)
335 		return -EINVAL;
336 
337 	return synic_set_irq(synic, sint);
338 }
339 
340 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
341 {
342 	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
343 	int i;
344 
345 	trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
346 
347 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
348 		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
349 			kvm_hv_notify_acked_sint(vcpu, i);
350 }
351 
352 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi)
353 {
354 	struct kvm_vcpu_hv_synic *synic;
355 
356 	synic = synic_get(kvm, vpidx);
357 	if (!synic)
358 		return -EINVAL;
359 
360 	if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
361 		return -EINVAL;
362 
363 	atomic_set(&synic->sint_to_gsi[sint], gsi);
364 	return 0;
365 }
366 
367 void kvm_hv_irq_routing_update(struct kvm *kvm)
368 {
369 	struct kvm_irq_routing_table *irq_rt;
370 	struct kvm_kernel_irq_routing_entry *e;
371 	u32 gsi;
372 
373 	irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
374 					lockdep_is_held(&kvm->irq_lock));
375 
376 	for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
377 		hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
378 			if (e->type == KVM_IRQ_ROUTING_HV_SINT)
379 				kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
380 						    e->hv_sint.sint, gsi);
381 		}
382 	}
383 }
384 
385 static void synic_init(struct kvm_vcpu_hv_synic *synic)
386 {
387 	int i;
388 
389 	memset(synic, 0, sizeof(*synic));
390 	synic->version = HV_SYNIC_VERSION_1;
391 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
392 		atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
393 		atomic_set(&synic->sint_to_gsi[i], -1);
394 	}
395 }
396 
397 static u64 get_time_ref_counter(struct kvm *kvm)
398 {
399 	struct kvm_hv *hv = &kvm->arch.hyperv;
400 	struct kvm_vcpu *vcpu;
401 	u64 tsc;
402 
403 	/*
404 	 * The guest has not set up the TSC page or the clock isn't
405 	 * stable, fall back to get_kvmclock_ns.
406 	 */
407 	if (!hv->tsc_ref.tsc_sequence)
408 		return div_u64(get_kvmclock_ns(kvm), 100);
409 
410 	vcpu = kvm_get_vcpu(kvm, 0);
411 	tsc = kvm_read_l1_tsc(vcpu, rdtsc());
412 	return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64)
413 		+ hv->tsc_ref.tsc_offset;
414 }
415 
416 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
417 				bool vcpu_kick)
418 {
419 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
420 
421 	set_bit(stimer->index,
422 		vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
423 	kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
424 	if (vcpu_kick)
425 		kvm_vcpu_kick(vcpu);
426 }
427 
428 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
429 {
430 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
431 
432 	trace_kvm_hv_stimer_cleanup(stimer_to_vcpu(stimer)->vcpu_id,
433 				    stimer->index);
434 
435 	hrtimer_cancel(&stimer->timer);
436 	clear_bit(stimer->index,
437 		  vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
438 	stimer->msg_pending = false;
439 	stimer->exp_time = 0;
440 }
441 
442 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
443 {
444 	struct kvm_vcpu_hv_stimer *stimer;
445 
446 	stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
447 	trace_kvm_hv_stimer_callback(stimer_to_vcpu(stimer)->vcpu_id,
448 				     stimer->index);
449 	stimer_mark_pending(stimer, true);
450 
451 	return HRTIMER_NORESTART;
452 }
453 
454 /*
455  * stimer_start() assumptions:
456  * a) stimer->count is not equal to 0
457  * b) stimer->config has HV_STIMER_ENABLE flag
458  */
459 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
460 {
461 	u64 time_now;
462 	ktime_t ktime_now;
463 
464 	time_now = get_time_ref_counter(stimer_to_vcpu(stimer)->kvm);
465 	ktime_now = ktime_get();
466 
467 	if (stimer->config.periodic) {
468 		if (stimer->exp_time) {
469 			if (time_now >= stimer->exp_time) {
470 				u64 remainder;
471 
472 				div64_u64_rem(time_now - stimer->exp_time,
473 					      stimer->count, &remainder);
474 				stimer->exp_time =
475 					time_now + (stimer->count - remainder);
476 			}
477 		} else
478 			stimer->exp_time = time_now + stimer->count;
479 
480 		trace_kvm_hv_stimer_start_periodic(
481 					stimer_to_vcpu(stimer)->vcpu_id,
482 					stimer->index,
483 					time_now, stimer->exp_time);
484 
485 		hrtimer_start(&stimer->timer,
486 			      ktime_add_ns(ktime_now,
487 					   100 * (stimer->exp_time - time_now)),
488 			      HRTIMER_MODE_ABS);
489 		return 0;
490 	}
491 	stimer->exp_time = stimer->count;
492 	if (time_now >= stimer->count) {
493 		/*
494 		 * Expire timer according to Hypervisor Top-Level Functional
495 		 * specification v4(15.3.1):
496 		 * "If a one shot is enabled and the specified count is in
497 		 * the past, it will expire immediately."
498 		 */
499 		stimer_mark_pending(stimer, false);
500 		return 0;
501 	}
502 
503 	trace_kvm_hv_stimer_start_one_shot(stimer_to_vcpu(stimer)->vcpu_id,
504 					   stimer->index,
505 					   time_now, stimer->count);
506 
507 	hrtimer_start(&stimer->timer,
508 		      ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)),
509 		      HRTIMER_MODE_ABS);
510 	return 0;
511 }
512 
513 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
514 			     bool host)
515 {
516 	union hv_stimer_config new_config = {.as_uint64 = config},
517 		old_config = {.as_uint64 = stimer->config.as_uint64};
518 
519 	trace_kvm_hv_stimer_set_config(stimer_to_vcpu(stimer)->vcpu_id,
520 				       stimer->index, config, host);
521 
522 	stimer_cleanup(stimer);
523 	if (old_config.enable &&
524 	    !new_config.direct_mode && new_config.sintx == 0)
525 		new_config.enable = 0;
526 	stimer->config.as_uint64 = new_config.as_uint64;
527 
528 	if (stimer->config.enable)
529 		stimer_mark_pending(stimer, false);
530 
531 	return 0;
532 }
533 
534 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
535 			    bool host)
536 {
537 	trace_kvm_hv_stimer_set_count(stimer_to_vcpu(stimer)->vcpu_id,
538 				      stimer->index, count, host);
539 
540 	stimer_cleanup(stimer);
541 	stimer->count = count;
542 	if (stimer->count == 0)
543 		stimer->config.enable = 0;
544 	else if (stimer->config.auto_enable)
545 		stimer->config.enable = 1;
546 
547 	if (stimer->config.enable)
548 		stimer_mark_pending(stimer, false);
549 
550 	return 0;
551 }
552 
553 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
554 {
555 	*pconfig = stimer->config.as_uint64;
556 	return 0;
557 }
558 
559 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
560 {
561 	*pcount = stimer->count;
562 	return 0;
563 }
564 
565 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
566 			     struct hv_message *src_msg, bool no_retry)
567 {
568 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
569 	int msg_off = offsetof(struct hv_message_page, sint_message[sint]);
570 	gfn_t msg_page_gfn;
571 	struct hv_message_header hv_hdr;
572 	int r;
573 
574 	if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
575 		return -ENOENT;
576 
577 	msg_page_gfn = synic->msg_page >> PAGE_SHIFT;
578 
579 	/*
580 	 * Strictly following the spec-mandated ordering would assume setting
581 	 * .msg_pending before checking .message_type.  However, this function
582 	 * is only called in vcpu context so the entire update is atomic from
583 	 * guest POV and thus the exact order here doesn't matter.
584 	 */
585 	r = kvm_vcpu_read_guest_page(vcpu, msg_page_gfn, &hv_hdr.message_type,
586 				     msg_off + offsetof(struct hv_message,
587 							header.message_type),
588 				     sizeof(hv_hdr.message_type));
589 	if (r < 0)
590 		return r;
591 
592 	if (hv_hdr.message_type != HVMSG_NONE) {
593 		if (no_retry)
594 			return 0;
595 
596 		hv_hdr.message_flags.msg_pending = 1;
597 		r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn,
598 					      &hv_hdr.message_flags,
599 					      msg_off +
600 					      offsetof(struct hv_message,
601 						       header.message_flags),
602 					      sizeof(hv_hdr.message_flags));
603 		if (r < 0)
604 			return r;
605 		return -EAGAIN;
606 	}
607 
608 	r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn, src_msg, msg_off,
609 				      sizeof(src_msg->header) +
610 				      src_msg->header.payload_size);
611 	if (r < 0)
612 		return r;
613 
614 	r = synic_set_irq(synic, sint);
615 	if (r < 0)
616 		return r;
617 	if (r == 0)
618 		return -EFAULT;
619 	return 0;
620 }
621 
622 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
623 {
624 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
625 	struct hv_message *msg = &stimer->msg;
626 	struct hv_timer_message_payload *payload =
627 			(struct hv_timer_message_payload *)&msg->u.payload;
628 
629 	/*
630 	 * To avoid piling up periodic ticks, don't retry message
631 	 * delivery for them (within "lazy" lost ticks policy).
632 	 */
633 	bool no_retry = stimer->config.periodic;
634 
635 	payload->expiration_time = stimer->exp_time;
636 	payload->delivery_time = get_time_ref_counter(vcpu->kvm);
637 	return synic_deliver_msg(vcpu_to_synic(vcpu),
638 				 stimer->config.sintx, msg,
639 				 no_retry);
640 }
641 
642 static int stimer_notify_direct(struct kvm_vcpu_hv_stimer *stimer)
643 {
644 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
645 	struct kvm_lapic_irq irq = {
646 		.delivery_mode = APIC_DM_FIXED,
647 		.vector = stimer->config.apic_vector
648 	};
649 
650 	if (lapic_in_kernel(vcpu))
651 		return !kvm_apic_set_irq(vcpu, &irq, NULL);
652 	return 0;
653 }
654 
655 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
656 {
657 	int r, direct = stimer->config.direct_mode;
658 
659 	stimer->msg_pending = true;
660 	if (!direct)
661 		r = stimer_send_msg(stimer);
662 	else
663 		r = stimer_notify_direct(stimer);
664 	trace_kvm_hv_stimer_expiration(stimer_to_vcpu(stimer)->vcpu_id,
665 				       stimer->index, direct, r);
666 	if (!r) {
667 		stimer->msg_pending = false;
668 		if (!(stimer->config.periodic))
669 			stimer->config.enable = 0;
670 	}
671 }
672 
673 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
674 {
675 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
676 	struct kvm_vcpu_hv_stimer *stimer;
677 	u64 time_now, exp_time;
678 	int i;
679 
680 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
681 		if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
682 			stimer = &hv_vcpu->stimer[i];
683 			if (stimer->config.enable) {
684 				exp_time = stimer->exp_time;
685 
686 				if (exp_time) {
687 					time_now =
688 						get_time_ref_counter(vcpu->kvm);
689 					if (time_now >= exp_time)
690 						stimer_expiration(stimer);
691 				}
692 
693 				if ((stimer->config.enable) &&
694 				    stimer->count) {
695 					if (!stimer->msg_pending)
696 						stimer_start(stimer);
697 				} else
698 					stimer_cleanup(stimer);
699 			}
700 		}
701 }
702 
703 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
704 {
705 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
706 	int i;
707 
708 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
709 		stimer_cleanup(&hv_vcpu->stimer[i]);
710 }
711 
712 bool kvm_hv_assist_page_enabled(struct kvm_vcpu *vcpu)
713 {
714 	if (!(vcpu->arch.hyperv.hv_vapic & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE))
715 		return false;
716 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
717 }
718 EXPORT_SYMBOL_GPL(kvm_hv_assist_page_enabled);
719 
720 bool kvm_hv_get_assist_page(struct kvm_vcpu *vcpu,
721 			    struct hv_vp_assist_page *assist_page)
722 {
723 	if (!kvm_hv_assist_page_enabled(vcpu))
724 		return false;
725 	return !kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data,
726 				      assist_page, sizeof(*assist_page));
727 }
728 EXPORT_SYMBOL_GPL(kvm_hv_get_assist_page);
729 
730 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
731 {
732 	struct hv_message *msg = &stimer->msg;
733 	struct hv_timer_message_payload *payload =
734 			(struct hv_timer_message_payload *)&msg->u.payload;
735 
736 	memset(&msg->header, 0, sizeof(msg->header));
737 	msg->header.message_type = HVMSG_TIMER_EXPIRED;
738 	msg->header.payload_size = sizeof(*payload);
739 
740 	payload->timer_index = stimer->index;
741 	payload->expiration_time = 0;
742 	payload->delivery_time = 0;
743 }
744 
745 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
746 {
747 	memset(stimer, 0, sizeof(*stimer));
748 	stimer->index = timer_index;
749 	hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
750 	stimer->timer.function = stimer_timer_callback;
751 	stimer_prepare_msg(stimer);
752 }
753 
754 void kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
755 {
756 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
757 	int i;
758 
759 	synic_init(&hv_vcpu->synic);
760 
761 	bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
762 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
763 		stimer_init(&hv_vcpu->stimer[i], i);
764 }
765 
766 void kvm_hv_vcpu_postcreate(struct kvm_vcpu *vcpu)
767 {
768 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
769 
770 	hv_vcpu->vp_index = kvm_vcpu_get_idx(vcpu);
771 }
772 
773 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
774 {
775 	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
776 
777 	/*
778 	 * Hyper-V SynIC auto EOI SINT's are
779 	 * not compatible with APICV, so deactivate APICV
780 	 */
781 	kvm_vcpu_deactivate_apicv(vcpu);
782 	synic->active = true;
783 	synic->dont_zero_synic_pages = dont_zero_synic_pages;
784 	return 0;
785 }
786 
787 static bool kvm_hv_msr_partition_wide(u32 msr)
788 {
789 	bool r = false;
790 
791 	switch (msr) {
792 	case HV_X64_MSR_GUEST_OS_ID:
793 	case HV_X64_MSR_HYPERCALL:
794 	case HV_X64_MSR_REFERENCE_TSC:
795 	case HV_X64_MSR_TIME_REF_COUNT:
796 	case HV_X64_MSR_CRASH_CTL:
797 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
798 	case HV_X64_MSR_RESET:
799 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
800 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
801 	case HV_X64_MSR_TSC_EMULATION_STATUS:
802 		r = true;
803 		break;
804 	}
805 
806 	return r;
807 }
808 
809 static int kvm_hv_msr_get_crash_data(struct kvm_vcpu *vcpu,
810 				     u32 index, u64 *pdata)
811 {
812 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
813 	size_t size = ARRAY_SIZE(hv->hv_crash_param);
814 
815 	if (WARN_ON_ONCE(index >= size))
816 		return -EINVAL;
817 
818 	*pdata = hv->hv_crash_param[array_index_nospec(index, size)];
819 	return 0;
820 }
821 
822 static int kvm_hv_msr_get_crash_ctl(struct kvm_vcpu *vcpu, u64 *pdata)
823 {
824 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
825 
826 	*pdata = hv->hv_crash_ctl;
827 	return 0;
828 }
829 
830 static int kvm_hv_msr_set_crash_ctl(struct kvm_vcpu *vcpu, u64 data, bool host)
831 {
832 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
833 
834 	if (host)
835 		hv->hv_crash_ctl = data & HV_CRASH_CTL_CRASH_NOTIFY;
836 
837 	if (!host && (data & HV_CRASH_CTL_CRASH_NOTIFY)) {
838 
839 		vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
840 			  hv->hv_crash_param[0],
841 			  hv->hv_crash_param[1],
842 			  hv->hv_crash_param[2],
843 			  hv->hv_crash_param[3],
844 			  hv->hv_crash_param[4]);
845 
846 		/* Send notification about crash to user space */
847 		kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
848 	}
849 
850 	return 0;
851 }
852 
853 static int kvm_hv_msr_set_crash_data(struct kvm_vcpu *vcpu,
854 				     u32 index, u64 data)
855 {
856 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
857 	size_t size = ARRAY_SIZE(hv->hv_crash_param);
858 
859 	if (WARN_ON_ONCE(index >= size))
860 		return -EINVAL;
861 
862 	hv->hv_crash_param[array_index_nospec(index, size)] = data;
863 	return 0;
864 }
865 
866 /*
867  * The kvmclock and Hyper-V TSC page use similar formulas, and converting
868  * between them is possible:
869  *
870  * kvmclock formula:
871  *    nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
872  *           + system_time
873  *
874  * Hyper-V formula:
875  *    nsec/100 = ticks * scale / 2^64 + offset
876  *
877  * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
878  * By dividing the kvmclock formula by 100 and equating what's left we get:
879  *    ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
880  *            scale / 2^64 =         tsc_to_system_mul * 2^(tsc_shift-32) / 100
881  *            scale        =         tsc_to_system_mul * 2^(32+tsc_shift) / 100
882  *
883  * Now expand the kvmclock formula and divide by 100:
884  *    nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
885  *           - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
886  *           + system_time
887  *    nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
888  *               - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
889  *               + system_time / 100
890  *
891  * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
892  *    nsec/100 = ticks * scale / 2^64
893  *               - tsc_timestamp * scale / 2^64
894  *               + system_time / 100
895  *
896  * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
897  *    offset = system_time / 100 - tsc_timestamp * scale / 2^64
898  *
899  * These two equivalencies are implemented in this function.
900  */
901 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
902 					HV_REFERENCE_TSC_PAGE *tsc_ref)
903 {
904 	u64 max_mul;
905 
906 	if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
907 		return false;
908 
909 	/*
910 	 * check if scale would overflow, if so we use the time ref counter
911 	 *    tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
912 	 *    tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
913 	 *    tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
914 	 */
915 	max_mul = 100ull << (32 - hv_clock->tsc_shift);
916 	if (hv_clock->tsc_to_system_mul >= max_mul)
917 		return false;
918 
919 	/*
920 	 * Otherwise compute the scale and offset according to the formulas
921 	 * derived above.
922 	 */
923 	tsc_ref->tsc_scale =
924 		mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
925 				hv_clock->tsc_to_system_mul,
926 				100);
927 
928 	tsc_ref->tsc_offset = hv_clock->system_time;
929 	do_div(tsc_ref->tsc_offset, 100);
930 	tsc_ref->tsc_offset -=
931 		mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
932 	return true;
933 }
934 
935 void kvm_hv_setup_tsc_page(struct kvm *kvm,
936 			   struct pvclock_vcpu_time_info *hv_clock)
937 {
938 	struct kvm_hv *hv = &kvm->arch.hyperv;
939 	u32 tsc_seq;
940 	u64 gfn;
941 
942 	BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
943 	BUILD_BUG_ON(offsetof(HV_REFERENCE_TSC_PAGE, tsc_sequence) != 0);
944 
945 	if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
946 		return;
947 
948 	mutex_lock(&kvm->arch.hyperv.hv_lock);
949 	if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
950 		goto out_unlock;
951 
952 	gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
953 	/*
954 	 * Because the TSC parameters only vary when there is a
955 	 * change in the master clock, do not bother with caching.
956 	 */
957 	if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
958 				    &tsc_seq, sizeof(tsc_seq))))
959 		goto out_unlock;
960 
961 	/*
962 	 * While we're computing and writing the parameters, force the
963 	 * guest to use the time reference count MSR.
964 	 */
965 	hv->tsc_ref.tsc_sequence = 0;
966 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
967 			    &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
968 		goto out_unlock;
969 
970 	if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
971 		goto out_unlock;
972 
973 	/* Ensure sequence is zero before writing the rest of the struct.  */
974 	smp_wmb();
975 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
976 		goto out_unlock;
977 
978 	/*
979 	 * Now switch to the TSC page mechanism by writing the sequence.
980 	 */
981 	tsc_seq++;
982 	if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
983 		tsc_seq = 1;
984 
985 	/* Write the struct entirely before the non-zero sequence.  */
986 	smp_wmb();
987 
988 	hv->tsc_ref.tsc_sequence = tsc_seq;
989 	kvm_write_guest(kvm, gfn_to_gpa(gfn),
990 			&hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence));
991 out_unlock:
992 	mutex_unlock(&kvm->arch.hyperv.hv_lock);
993 }
994 
995 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
996 			     bool host)
997 {
998 	struct kvm *kvm = vcpu->kvm;
999 	struct kvm_hv *hv = &kvm->arch.hyperv;
1000 
1001 	switch (msr) {
1002 	case HV_X64_MSR_GUEST_OS_ID:
1003 		hv->hv_guest_os_id = data;
1004 		/* setting guest os id to zero disables hypercall page */
1005 		if (!hv->hv_guest_os_id)
1006 			hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1007 		break;
1008 	case HV_X64_MSR_HYPERCALL: {
1009 		u64 gfn;
1010 		unsigned long addr;
1011 		u8 instructions[4];
1012 
1013 		/* if guest os id is not set hypercall should remain disabled */
1014 		if (!hv->hv_guest_os_id)
1015 			break;
1016 		if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1017 			hv->hv_hypercall = data;
1018 			break;
1019 		}
1020 		gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1021 		addr = gfn_to_hva(kvm, gfn);
1022 		if (kvm_is_error_hva(addr))
1023 			return 1;
1024 		kvm_x86_ops->patch_hypercall(vcpu, instructions);
1025 		((unsigned char *)instructions)[3] = 0xc3; /* ret */
1026 		if (__copy_to_user((void __user *)addr, instructions, 4))
1027 			return 1;
1028 		hv->hv_hypercall = data;
1029 		mark_page_dirty(kvm, gfn);
1030 		break;
1031 	}
1032 	case HV_X64_MSR_REFERENCE_TSC:
1033 		hv->hv_tsc_page = data;
1034 		if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE)
1035 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1036 		break;
1037 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1038 		return kvm_hv_msr_set_crash_data(vcpu,
1039 						 msr - HV_X64_MSR_CRASH_P0,
1040 						 data);
1041 	case HV_X64_MSR_CRASH_CTL:
1042 		return kvm_hv_msr_set_crash_ctl(vcpu, data, host);
1043 	case HV_X64_MSR_RESET:
1044 		if (data == 1) {
1045 			vcpu_debug(vcpu, "hyper-v reset requested\n");
1046 			kvm_make_request(KVM_REQ_HV_RESET, vcpu);
1047 		}
1048 		break;
1049 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1050 		hv->hv_reenlightenment_control = data;
1051 		break;
1052 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1053 		hv->hv_tsc_emulation_control = data;
1054 		break;
1055 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1056 		hv->hv_tsc_emulation_status = data;
1057 		break;
1058 	case HV_X64_MSR_TIME_REF_COUNT:
1059 		/* read-only, but still ignore it if host-initiated */
1060 		if (!host)
1061 			return 1;
1062 		break;
1063 	default:
1064 		vcpu_unimpl(vcpu, "Hyper-V unhandled wrmsr: 0x%x data 0x%llx\n",
1065 			    msr, data);
1066 		return 1;
1067 	}
1068 	return 0;
1069 }
1070 
1071 /* Calculate cpu time spent by current task in 100ns units */
1072 static u64 current_task_runtime_100ns(void)
1073 {
1074 	u64 utime, stime;
1075 
1076 	task_cputime_adjusted(current, &utime, &stime);
1077 
1078 	return div_u64(utime + stime, 100);
1079 }
1080 
1081 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1082 {
1083 	struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
1084 
1085 	switch (msr) {
1086 	case HV_X64_MSR_VP_INDEX: {
1087 		struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
1088 		int vcpu_idx = kvm_vcpu_get_idx(vcpu);
1089 		u32 new_vp_index = (u32)data;
1090 
1091 		if (!host || new_vp_index >= KVM_MAX_VCPUS)
1092 			return 1;
1093 
1094 		if (new_vp_index == hv_vcpu->vp_index)
1095 			return 0;
1096 
1097 		/*
1098 		 * The VP index is initialized to vcpu_index by
1099 		 * kvm_hv_vcpu_postcreate so they initially match.  Now the
1100 		 * VP index is changing, adjust num_mismatched_vp_indexes if
1101 		 * it now matches or no longer matches vcpu_idx.
1102 		 */
1103 		if (hv_vcpu->vp_index == vcpu_idx)
1104 			atomic_inc(&hv->num_mismatched_vp_indexes);
1105 		else if (new_vp_index == vcpu_idx)
1106 			atomic_dec(&hv->num_mismatched_vp_indexes);
1107 
1108 		hv_vcpu->vp_index = new_vp_index;
1109 		break;
1110 	}
1111 	case HV_X64_MSR_VP_ASSIST_PAGE: {
1112 		u64 gfn;
1113 		unsigned long addr;
1114 
1115 		if (!(data & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE)) {
1116 			hv_vcpu->hv_vapic = data;
1117 			if (kvm_lapic_enable_pv_eoi(vcpu, 0, 0))
1118 				return 1;
1119 			break;
1120 		}
1121 		gfn = data >> HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT;
1122 		addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1123 		if (kvm_is_error_hva(addr))
1124 			return 1;
1125 
1126 		/*
1127 		 * Clear apic_assist portion of struct hv_vp_assist_page
1128 		 * only, there can be valuable data in the rest which needs
1129 		 * to be preserved e.g. on migration.
1130 		 */
1131 		if (__clear_user((void __user *)addr, sizeof(u32)))
1132 			return 1;
1133 		hv_vcpu->hv_vapic = data;
1134 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
1135 		if (kvm_lapic_enable_pv_eoi(vcpu,
1136 					    gfn_to_gpa(gfn) | KVM_MSR_ENABLED,
1137 					    sizeof(struct hv_vp_assist_page)))
1138 			return 1;
1139 		break;
1140 	}
1141 	case HV_X64_MSR_EOI:
1142 		return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1143 	case HV_X64_MSR_ICR:
1144 		return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1145 	case HV_X64_MSR_TPR:
1146 		return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1147 	case HV_X64_MSR_VP_RUNTIME:
1148 		if (!host)
1149 			return 1;
1150 		hv_vcpu->runtime_offset = data - current_task_runtime_100ns();
1151 		break;
1152 	case HV_X64_MSR_SCONTROL:
1153 	case HV_X64_MSR_SVERSION:
1154 	case HV_X64_MSR_SIEFP:
1155 	case HV_X64_MSR_SIMP:
1156 	case HV_X64_MSR_EOM:
1157 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1158 		return synic_set_msr(vcpu_to_synic(vcpu), msr, data, host);
1159 	case HV_X64_MSR_STIMER0_CONFIG:
1160 	case HV_X64_MSR_STIMER1_CONFIG:
1161 	case HV_X64_MSR_STIMER2_CONFIG:
1162 	case HV_X64_MSR_STIMER3_CONFIG: {
1163 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1164 
1165 		return stimer_set_config(vcpu_to_stimer(vcpu, timer_index),
1166 					 data, host);
1167 	}
1168 	case HV_X64_MSR_STIMER0_COUNT:
1169 	case HV_X64_MSR_STIMER1_COUNT:
1170 	case HV_X64_MSR_STIMER2_COUNT:
1171 	case HV_X64_MSR_STIMER3_COUNT: {
1172 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1173 
1174 		return stimer_set_count(vcpu_to_stimer(vcpu, timer_index),
1175 					data, host);
1176 	}
1177 	case HV_X64_MSR_TSC_FREQUENCY:
1178 	case HV_X64_MSR_APIC_FREQUENCY:
1179 		/* read-only, but still ignore it if host-initiated */
1180 		if (!host)
1181 			return 1;
1182 		break;
1183 	default:
1184 		vcpu_unimpl(vcpu, "Hyper-V unhandled wrmsr: 0x%x data 0x%llx\n",
1185 			    msr, data);
1186 		return 1;
1187 	}
1188 
1189 	return 0;
1190 }
1191 
1192 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1193 {
1194 	u64 data = 0;
1195 	struct kvm *kvm = vcpu->kvm;
1196 	struct kvm_hv *hv = &kvm->arch.hyperv;
1197 
1198 	switch (msr) {
1199 	case HV_X64_MSR_GUEST_OS_ID:
1200 		data = hv->hv_guest_os_id;
1201 		break;
1202 	case HV_X64_MSR_HYPERCALL:
1203 		data = hv->hv_hypercall;
1204 		break;
1205 	case HV_X64_MSR_TIME_REF_COUNT:
1206 		data = get_time_ref_counter(kvm);
1207 		break;
1208 	case HV_X64_MSR_REFERENCE_TSC:
1209 		data = hv->hv_tsc_page;
1210 		break;
1211 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1212 		return kvm_hv_msr_get_crash_data(vcpu,
1213 						 msr - HV_X64_MSR_CRASH_P0,
1214 						 pdata);
1215 	case HV_X64_MSR_CRASH_CTL:
1216 		return kvm_hv_msr_get_crash_ctl(vcpu, pdata);
1217 	case HV_X64_MSR_RESET:
1218 		data = 0;
1219 		break;
1220 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1221 		data = hv->hv_reenlightenment_control;
1222 		break;
1223 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1224 		data = hv->hv_tsc_emulation_control;
1225 		break;
1226 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1227 		data = hv->hv_tsc_emulation_status;
1228 		break;
1229 	default:
1230 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1231 		return 1;
1232 	}
1233 
1234 	*pdata = data;
1235 	return 0;
1236 }
1237 
1238 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1239 			  bool host)
1240 {
1241 	u64 data = 0;
1242 	struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
1243 
1244 	switch (msr) {
1245 	case HV_X64_MSR_VP_INDEX:
1246 		data = hv_vcpu->vp_index;
1247 		break;
1248 	case HV_X64_MSR_EOI:
1249 		return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1250 	case HV_X64_MSR_ICR:
1251 		return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1252 	case HV_X64_MSR_TPR:
1253 		return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1254 	case HV_X64_MSR_VP_ASSIST_PAGE:
1255 		data = hv_vcpu->hv_vapic;
1256 		break;
1257 	case HV_X64_MSR_VP_RUNTIME:
1258 		data = current_task_runtime_100ns() + hv_vcpu->runtime_offset;
1259 		break;
1260 	case HV_X64_MSR_SCONTROL:
1261 	case HV_X64_MSR_SVERSION:
1262 	case HV_X64_MSR_SIEFP:
1263 	case HV_X64_MSR_SIMP:
1264 	case HV_X64_MSR_EOM:
1265 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1266 		return synic_get_msr(vcpu_to_synic(vcpu), msr, pdata, host);
1267 	case HV_X64_MSR_STIMER0_CONFIG:
1268 	case HV_X64_MSR_STIMER1_CONFIG:
1269 	case HV_X64_MSR_STIMER2_CONFIG:
1270 	case HV_X64_MSR_STIMER3_CONFIG: {
1271 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1272 
1273 		return stimer_get_config(vcpu_to_stimer(vcpu, timer_index),
1274 					 pdata);
1275 	}
1276 	case HV_X64_MSR_STIMER0_COUNT:
1277 	case HV_X64_MSR_STIMER1_COUNT:
1278 	case HV_X64_MSR_STIMER2_COUNT:
1279 	case HV_X64_MSR_STIMER3_COUNT: {
1280 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1281 
1282 		return stimer_get_count(vcpu_to_stimer(vcpu, timer_index),
1283 					pdata);
1284 	}
1285 	case HV_X64_MSR_TSC_FREQUENCY:
1286 		data = (u64)vcpu->arch.virtual_tsc_khz * 1000;
1287 		break;
1288 	case HV_X64_MSR_APIC_FREQUENCY:
1289 		data = APIC_BUS_FREQUENCY;
1290 		break;
1291 	default:
1292 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1293 		return 1;
1294 	}
1295 	*pdata = data;
1296 	return 0;
1297 }
1298 
1299 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1300 {
1301 	if (kvm_hv_msr_partition_wide(msr)) {
1302 		int r;
1303 
1304 		mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1305 		r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1306 		mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1307 		return r;
1308 	} else
1309 		return kvm_hv_set_msr(vcpu, msr, data, host);
1310 }
1311 
1312 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
1313 {
1314 	if (kvm_hv_msr_partition_wide(msr)) {
1315 		int r;
1316 
1317 		mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1318 		r = kvm_hv_get_msr_pw(vcpu, msr, pdata);
1319 		mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1320 		return r;
1321 	} else
1322 		return kvm_hv_get_msr(vcpu, msr, pdata, host);
1323 }
1324 
1325 static __always_inline unsigned long *sparse_set_to_vcpu_mask(
1326 	struct kvm *kvm, u64 *sparse_banks, u64 valid_bank_mask,
1327 	u64 *vp_bitmap, unsigned long *vcpu_bitmap)
1328 {
1329 	struct kvm_hv *hv = &kvm->arch.hyperv;
1330 	struct kvm_vcpu *vcpu;
1331 	int i, bank, sbank = 0;
1332 
1333 	memset(vp_bitmap, 0,
1334 	       KVM_HV_MAX_SPARSE_VCPU_SET_BITS * sizeof(*vp_bitmap));
1335 	for_each_set_bit(bank, (unsigned long *)&valid_bank_mask,
1336 			 KVM_HV_MAX_SPARSE_VCPU_SET_BITS)
1337 		vp_bitmap[bank] = sparse_banks[sbank++];
1338 
1339 	if (likely(!atomic_read(&hv->num_mismatched_vp_indexes))) {
1340 		/* for all vcpus vp_index == vcpu_idx */
1341 		return (unsigned long *)vp_bitmap;
1342 	}
1343 
1344 	bitmap_zero(vcpu_bitmap, KVM_MAX_VCPUS);
1345 	kvm_for_each_vcpu(i, vcpu, kvm) {
1346 		if (test_bit(vcpu_to_hv_vcpu(vcpu)->vp_index,
1347 			     (unsigned long *)vp_bitmap))
1348 			__set_bit(i, vcpu_bitmap);
1349 	}
1350 	return vcpu_bitmap;
1351 }
1352 
1353 static u64 kvm_hv_flush_tlb(struct kvm_vcpu *current_vcpu, u64 ingpa,
1354 			    u16 rep_cnt, bool ex)
1355 {
1356 	struct kvm *kvm = current_vcpu->kvm;
1357 	struct kvm_vcpu_hv *hv_vcpu = &current_vcpu->arch.hyperv;
1358 	struct hv_tlb_flush_ex flush_ex;
1359 	struct hv_tlb_flush flush;
1360 	u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1361 	DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1362 	unsigned long *vcpu_mask;
1363 	u64 valid_bank_mask;
1364 	u64 sparse_banks[64];
1365 	int sparse_banks_len;
1366 	bool all_cpus;
1367 
1368 	if (!ex) {
1369 		if (unlikely(kvm_read_guest(kvm, ingpa, &flush, sizeof(flush))))
1370 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1371 
1372 		trace_kvm_hv_flush_tlb(flush.processor_mask,
1373 				       flush.address_space, flush.flags);
1374 
1375 		valid_bank_mask = BIT_ULL(0);
1376 		sparse_banks[0] = flush.processor_mask;
1377 
1378 		/*
1379 		 * Work around possible WS2012 bug: it sends hypercalls
1380 		 * with processor_mask = 0x0 and HV_FLUSH_ALL_PROCESSORS clear,
1381 		 * while also expecting us to flush something and crashing if
1382 		 * we don't. Let's treat processor_mask == 0 same as
1383 		 * HV_FLUSH_ALL_PROCESSORS.
1384 		 */
1385 		all_cpus = (flush.flags & HV_FLUSH_ALL_PROCESSORS) ||
1386 			flush.processor_mask == 0;
1387 	} else {
1388 		if (unlikely(kvm_read_guest(kvm, ingpa, &flush_ex,
1389 					    sizeof(flush_ex))))
1390 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1391 
1392 		trace_kvm_hv_flush_tlb_ex(flush_ex.hv_vp_set.valid_bank_mask,
1393 					  flush_ex.hv_vp_set.format,
1394 					  flush_ex.address_space,
1395 					  flush_ex.flags);
1396 
1397 		valid_bank_mask = flush_ex.hv_vp_set.valid_bank_mask;
1398 		all_cpus = flush_ex.hv_vp_set.format !=
1399 			HV_GENERIC_SET_SPARSE_4K;
1400 
1401 		sparse_banks_len =
1402 			bitmap_weight((unsigned long *)&valid_bank_mask, 64) *
1403 			sizeof(sparse_banks[0]);
1404 
1405 		if (!sparse_banks_len && !all_cpus)
1406 			goto ret_success;
1407 
1408 		if (!all_cpus &&
1409 		    kvm_read_guest(kvm,
1410 				   ingpa + offsetof(struct hv_tlb_flush_ex,
1411 						    hv_vp_set.bank_contents),
1412 				   sparse_banks,
1413 				   sparse_banks_len))
1414 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1415 	}
1416 
1417 	cpumask_clear(&hv_vcpu->tlb_flush);
1418 
1419 	vcpu_mask = all_cpus ? NULL :
1420 		sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1421 					vp_bitmap, vcpu_bitmap);
1422 
1423 	/*
1424 	 * vcpu->arch.cr3 may not be up-to-date for running vCPUs so we can't
1425 	 * analyze it here, flush TLB regardless of the specified address space.
1426 	 */
1427 	kvm_make_vcpus_request_mask(kvm,
1428 				    KVM_REQ_TLB_FLUSH | KVM_REQUEST_NO_WAKEUP,
1429 				    vcpu_mask, &hv_vcpu->tlb_flush);
1430 
1431 ret_success:
1432 	/* We always do full TLB flush, set rep_done = rep_cnt. */
1433 	return (u64)HV_STATUS_SUCCESS |
1434 		((u64)rep_cnt << HV_HYPERCALL_REP_COMP_OFFSET);
1435 }
1436 
1437 static void kvm_send_ipi_to_many(struct kvm *kvm, u32 vector,
1438 				 unsigned long *vcpu_bitmap)
1439 {
1440 	struct kvm_lapic_irq irq = {
1441 		.delivery_mode = APIC_DM_FIXED,
1442 		.vector = vector
1443 	};
1444 	struct kvm_vcpu *vcpu;
1445 	int i;
1446 
1447 	kvm_for_each_vcpu(i, vcpu, kvm) {
1448 		if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
1449 			continue;
1450 
1451 		/* We fail only when APIC is disabled */
1452 		kvm_apic_set_irq(vcpu, &irq, NULL);
1453 	}
1454 }
1455 
1456 static u64 kvm_hv_send_ipi(struct kvm_vcpu *current_vcpu, u64 ingpa, u64 outgpa,
1457 			   bool ex, bool fast)
1458 {
1459 	struct kvm *kvm = current_vcpu->kvm;
1460 	struct hv_send_ipi_ex send_ipi_ex;
1461 	struct hv_send_ipi send_ipi;
1462 	u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1463 	DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1464 	unsigned long *vcpu_mask;
1465 	unsigned long valid_bank_mask;
1466 	u64 sparse_banks[64];
1467 	int sparse_banks_len;
1468 	u32 vector;
1469 	bool all_cpus;
1470 
1471 	if (!ex) {
1472 		if (!fast) {
1473 			if (unlikely(kvm_read_guest(kvm, ingpa, &send_ipi,
1474 						    sizeof(send_ipi))))
1475 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
1476 			sparse_banks[0] = send_ipi.cpu_mask;
1477 			vector = send_ipi.vector;
1478 		} else {
1479 			/* 'reserved' part of hv_send_ipi should be 0 */
1480 			if (unlikely(ingpa >> 32 != 0))
1481 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
1482 			sparse_banks[0] = outgpa;
1483 			vector = (u32)ingpa;
1484 		}
1485 		all_cpus = false;
1486 		valid_bank_mask = BIT_ULL(0);
1487 
1488 		trace_kvm_hv_send_ipi(vector, sparse_banks[0]);
1489 	} else {
1490 		if (unlikely(kvm_read_guest(kvm, ingpa, &send_ipi_ex,
1491 					    sizeof(send_ipi_ex))))
1492 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1493 
1494 		trace_kvm_hv_send_ipi_ex(send_ipi_ex.vector,
1495 					 send_ipi_ex.vp_set.format,
1496 					 send_ipi_ex.vp_set.valid_bank_mask);
1497 
1498 		vector = send_ipi_ex.vector;
1499 		valid_bank_mask = send_ipi_ex.vp_set.valid_bank_mask;
1500 		sparse_banks_len = bitmap_weight(&valid_bank_mask, 64) *
1501 			sizeof(sparse_banks[0]);
1502 
1503 		all_cpus = send_ipi_ex.vp_set.format == HV_GENERIC_SET_ALL;
1504 
1505 		if (!sparse_banks_len)
1506 			goto ret_success;
1507 
1508 		if (!all_cpus &&
1509 		    kvm_read_guest(kvm,
1510 				   ingpa + offsetof(struct hv_send_ipi_ex,
1511 						    vp_set.bank_contents),
1512 				   sparse_banks,
1513 				   sparse_banks_len))
1514 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1515 	}
1516 
1517 	if ((vector < HV_IPI_LOW_VECTOR) || (vector > HV_IPI_HIGH_VECTOR))
1518 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
1519 
1520 	vcpu_mask = all_cpus ? NULL :
1521 		sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1522 					vp_bitmap, vcpu_bitmap);
1523 
1524 	kvm_send_ipi_to_many(kvm, vector, vcpu_mask);
1525 
1526 ret_success:
1527 	return HV_STATUS_SUCCESS;
1528 }
1529 
1530 bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1531 {
1532 	return READ_ONCE(kvm->arch.hyperv.hv_hypercall) & HV_X64_MSR_HYPERCALL_ENABLE;
1533 }
1534 
1535 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
1536 {
1537 	bool longmode;
1538 
1539 	longmode = is_64_bit_mode(vcpu);
1540 	if (longmode)
1541 		kvm_rax_write(vcpu, result);
1542 	else {
1543 		kvm_rdx_write(vcpu, result >> 32);
1544 		kvm_rax_write(vcpu, result & 0xffffffff);
1545 	}
1546 }
1547 
1548 static int kvm_hv_hypercall_complete(struct kvm_vcpu *vcpu, u64 result)
1549 {
1550 	kvm_hv_hypercall_set_result(vcpu, result);
1551 	++vcpu->stat.hypercalls;
1552 	return kvm_skip_emulated_instruction(vcpu);
1553 }
1554 
1555 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
1556 {
1557 	return kvm_hv_hypercall_complete(vcpu, vcpu->run->hyperv.u.hcall.result);
1558 }
1559 
1560 static u16 kvm_hvcall_signal_event(struct kvm_vcpu *vcpu, bool fast, u64 param)
1561 {
1562 	struct eventfd_ctx *eventfd;
1563 
1564 	if (unlikely(!fast)) {
1565 		int ret;
1566 		gpa_t gpa = param;
1567 
1568 		if ((gpa & (__alignof__(param) - 1)) ||
1569 		    offset_in_page(gpa) + sizeof(param) > PAGE_SIZE)
1570 			return HV_STATUS_INVALID_ALIGNMENT;
1571 
1572 		ret = kvm_vcpu_read_guest(vcpu, gpa, &param, sizeof(param));
1573 		if (ret < 0)
1574 			return HV_STATUS_INVALID_ALIGNMENT;
1575 	}
1576 
1577 	/*
1578 	 * Per spec, bits 32-47 contain the extra "flag number".  However, we
1579 	 * have no use for it, and in all known usecases it is zero, so just
1580 	 * report lookup failure if it isn't.
1581 	 */
1582 	if (param & 0xffff00000000ULL)
1583 		return HV_STATUS_INVALID_PORT_ID;
1584 	/* remaining bits are reserved-zero */
1585 	if (param & ~KVM_HYPERV_CONN_ID_MASK)
1586 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
1587 
1588 	/* the eventfd is protected by vcpu->kvm->srcu, but conn_to_evt isn't */
1589 	rcu_read_lock();
1590 	eventfd = idr_find(&vcpu->kvm->arch.hyperv.conn_to_evt, param);
1591 	rcu_read_unlock();
1592 	if (!eventfd)
1593 		return HV_STATUS_INVALID_PORT_ID;
1594 
1595 	eventfd_signal(eventfd, 1);
1596 	return HV_STATUS_SUCCESS;
1597 }
1598 
1599 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
1600 {
1601 	u64 param, ingpa, outgpa, ret = HV_STATUS_SUCCESS;
1602 	uint16_t code, rep_idx, rep_cnt;
1603 	bool fast, rep;
1604 
1605 	/*
1606 	 * hypercall generates UD from non zero cpl and real mode
1607 	 * per HYPER-V spec
1608 	 */
1609 	if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
1610 		kvm_queue_exception(vcpu, UD_VECTOR);
1611 		return 1;
1612 	}
1613 
1614 #ifdef CONFIG_X86_64
1615 	if (is_64_bit_mode(vcpu)) {
1616 		param = kvm_rcx_read(vcpu);
1617 		ingpa = kvm_rdx_read(vcpu);
1618 		outgpa = kvm_r8_read(vcpu);
1619 	} else
1620 #endif
1621 	{
1622 		param = ((u64)kvm_rdx_read(vcpu) << 32) |
1623 			(kvm_rax_read(vcpu) & 0xffffffff);
1624 		ingpa = ((u64)kvm_rbx_read(vcpu) << 32) |
1625 			(kvm_rcx_read(vcpu) & 0xffffffff);
1626 		outgpa = ((u64)kvm_rdi_read(vcpu) << 32) |
1627 			(kvm_rsi_read(vcpu) & 0xffffffff);
1628 	}
1629 
1630 	code = param & 0xffff;
1631 	fast = !!(param & HV_HYPERCALL_FAST_BIT);
1632 	rep_cnt = (param >> HV_HYPERCALL_REP_COMP_OFFSET) & 0xfff;
1633 	rep_idx = (param >> HV_HYPERCALL_REP_START_OFFSET) & 0xfff;
1634 	rep = !!(rep_cnt || rep_idx);
1635 
1636 	trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
1637 
1638 	switch (code) {
1639 	case HVCALL_NOTIFY_LONG_SPIN_WAIT:
1640 		if (unlikely(rep)) {
1641 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1642 			break;
1643 		}
1644 		kvm_vcpu_on_spin(vcpu, true);
1645 		break;
1646 	case HVCALL_SIGNAL_EVENT:
1647 		if (unlikely(rep)) {
1648 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1649 			break;
1650 		}
1651 		ret = kvm_hvcall_signal_event(vcpu, fast, ingpa);
1652 		if (ret != HV_STATUS_INVALID_PORT_ID)
1653 			break;
1654 		/* fall through - maybe userspace knows this conn_id. */
1655 	case HVCALL_POST_MESSAGE:
1656 		/* don't bother userspace if it has no way to handle it */
1657 		if (unlikely(rep || !vcpu_to_synic(vcpu)->active)) {
1658 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1659 			break;
1660 		}
1661 		vcpu->run->exit_reason = KVM_EXIT_HYPERV;
1662 		vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
1663 		vcpu->run->hyperv.u.hcall.input = param;
1664 		vcpu->run->hyperv.u.hcall.params[0] = ingpa;
1665 		vcpu->run->hyperv.u.hcall.params[1] = outgpa;
1666 		vcpu->arch.complete_userspace_io =
1667 				kvm_hv_hypercall_complete_userspace;
1668 		return 0;
1669 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
1670 		if (unlikely(fast || !rep_cnt || rep_idx)) {
1671 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1672 			break;
1673 		}
1674 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false);
1675 		break;
1676 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
1677 		if (unlikely(fast || rep)) {
1678 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1679 			break;
1680 		}
1681 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false);
1682 		break;
1683 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
1684 		if (unlikely(fast || !rep_cnt || rep_idx)) {
1685 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1686 			break;
1687 		}
1688 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
1689 		break;
1690 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
1691 		if (unlikely(fast || rep)) {
1692 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1693 			break;
1694 		}
1695 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
1696 		break;
1697 	case HVCALL_SEND_IPI:
1698 		if (unlikely(rep)) {
1699 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1700 			break;
1701 		}
1702 		ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, false, fast);
1703 		break;
1704 	case HVCALL_SEND_IPI_EX:
1705 		if (unlikely(fast || rep)) {
1706 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1707 			break;
1708 		}
1709 		ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, true, false);
1710 		break;
1711 	default:
1712 		ret = HV_STATUS_INVALID_HYPERCALL_CODE;
1713 		break;
1714 	}
1715 
1716 	return kvm_hv_hypercall_complete(vcpu, ret);
1717 }
1718 
1719 void kvm_hv_init_vm(struct kvm *kvm)
1720 {
1721 	mutex_init(&kvm->arch.hyperv.hv_lock);
1722 	idr_init(&kvm->arch.hyperv.conn_to_evt);
1723 }
1724 
1725 void kvm_hv_destroy_vm(struct kvm *kvm)
1726 {
1727 	struct eventfd_ctx *eventfd;
1728 	int i;
1729 
1730 	idr_for_each_entry(&kvm->arch.hyperv.conn_to_evt, eventfd, i)
1731 		eventfd_ctx_put(eventfd);
1732 	idr_destroy(&kvm->arch.hyperv.conn_to_evt);
1733 }
1734 
1735 static int kvm_hv_eventfd_assign(struct kvm *kvm, u32 conn_id, int fd)
1736 {
1737 	struct kvm_hv *hv = &kvm->arch.hyperv;
1738 	struct eventfd_ctx *eventfd;
1739 	int ret;
1740 
1741 	eventfd = eventfd_ctx_fdget(fd);
1742 	if (IS_ERR(eventfd))
1743 		return PTR_ERR(eventfd);
1744 
1745 	mutex_lock(&hv->hv_lock);
1746 	ret = idr_alloc(&hv->conn_to_evt, eventfd, conn_id, conn_id + 1,
1747 			GFP_KERNEL_ACCOUNT);
1748 	mutex_unlock(&hv->hv_lock);
1749 
1750 	if (ret >= 0)
1751 		return 0;
1752 
1753 	if (ret == -ENOSPC)
1754 		ret = -EEXIST;
1755 	eventfd_ctx_put(eventfd);
1756 	return ret;
1757 }
1758 
1759 static int kvm_hv_eventfd_deassign(struct kvm *kvm, u32 conn_id)
1760 {
1761 	struct kvm_hv *hv = &kvm->arch.hyperv;
1762 	struct eventfd_ctx *eventfd;
1763 
1764 	mutex_lock(&hv->hv_lock);
1765 	eventfd = idr_remove(&hv->conn_to_evt, conn_id);
1766 	mutex_unlock(&hv->hv_lock);
1767 
1768 	if (!eventfd)
1769 		return -ENOENT;
1770 
1771 	synchronize_srcu(&kvm->srcu);
1772 	eventfd_ctx_put(eventfd);
1773 	return 0;
1774 }
1775 
1776 int kvm_vm_ioctl_hv_eventfd(struct kvm *kvm, struct kvm_hyperv_eventfd *args)
1777 {
1778 	if ((args->flags & ~KVM_HYPERV_EVENTFD_DEASSIGN) ||
1779 	    (args->conn_id & ~KVM_HYPERV_CONN_ID_MASK))
1780 		return -EINVAL;
1781 
1782 	if (args->flags == KVM_HYPERV_EVENTFD_DEASSIGN)
1783 		return kvm_hv_eventfd_deassign(kvm, args->conn_id);
1784 	return kvm_hv_eventfd_assign(kvm, args->conn_id, args->fd);
1785 }
1786 
1787 int kvm_vcpu_ioctl_get_hv_cpuid(struct kvm_vcpu *vcpu, struct kvm_cpuid2 *cpuid,
1788 				struct kvm_cpuid_entry2 __user *entries)
1789 {
1790 	uint16_t evmcs_ver = 0;
1791 	struct kvm_cpuid_entry2 cpuid_entries[] = {
1792 		{ .function = HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS },
1793 		{ .function = HYPERV_CPUID_INTERFACE },
1794 		{ .function = HYPERV_CPUID_VERSION },
1795 		{ .function = HYPERV_CPUID_FEATURES },
1796 		{ .function = HYPERV_CPUID_ENLIGHTMENT_INFO },
1797 		{ .function = HYPERV_CPUID_IMPLEMENT_LIMITS },
1798 		{ .function = HYPERV_CPUID_NESTED_FEATURES },
1799 	};
1800 	int i, nent = ARRAY_SIZE(cpuid_entries);
1801 
1802 	if (kvm_x86_ops->nested_get_evmcs_version)
1803 		evmcs_ver = kvm_x86_ops->nested_get_evmcs_version(vcpu);
1804 
1805 	/* Skip NESTED_FEATURES if eVMCS is not supported */
1806 	if (!evmcs_ver)
1807 		--nent;
1808 
1809 	if (cpuid->nent < nent)
1810 		return -E2BIG;
1811 
1812 	if (cpuid->nent > nent)
1813 		cpuid->nent = nent;
1814 
1815 	for (i = 0; i < nent; i++) {
1816 		struct kvm_cpuid_entry2 *ent = &cpuid_entries[i];
1817 		u32 signature[3];
1818 
1819 		switch (ent->function) {
1820 		case HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS:
1821 			memcpy(signature, "Linux KVM Hv", 12);
1822 
1823 			ent->eax = HYPERV_CPUID_NESTED_FEATURES;
1824 			ent->ebx = signature[0];
1825 			ent->ecx = signature[1];
1826 			ent->edx = signature[2];
1827 			break;
1828 
1829 		case HYPERV_CPUID_INTERFACE:
1830 			memcpy(signature, "Hv#1\0\0\0\0\0\0\0\0", 12);
1831 			ent->eax = signature[0];
1832 			break;
1833 
1834 		case HYPERV_CPUID_VERSION:
1835 			/*
1836 			 * We implement some Hyper-V 2016 functions so let's use
1837 			 * this version.
1838 			 */
1839 			ent->eax = 0x00003839;
1840 			ent->ebx = 0x000A0000;
1841 			break;
1842 
1843 		case HYPERV_CPUID_FEATURES:
1844 			ent->eax |= HV_X64_MSR_VP_RUNTIME_AVAILABLE;
1845 			ent->eax |= HV_MSR_TIME_REF_COUNT_AVAILABLE;
1846 			ent->eax |= HV_X64_MSR_SYNIC_AVAILABLE;
1847 			ent->eax |= HV_MSR_SYNTIMER_AVAILABLE;
1848 			ent->eax |= HV_X64_MSR_APIC_ACCESS_AVAILABLE;
1849 			ent->eax |= HV_X64_MSR_HYPERCALL_AVAILABLE;
1850 			ent->eax |= HV_X64_MSR_VP_INDEX_AVAILABLE;
1851 			ent->eax |= HV_X64_MSR_RESET_AVAILABLE;
1852 			ent->eax |= HV_MSR_REFERENCE_TSC_AVAILABLE;
1853 			ent->eax |= HV_X64_ACCESS_FREQUENCY_MSRS;
1854 			ent->eax |= HV_X64_ACCESS_REENLIGHTENMENT;
1855 
1856 			ent->ebx |= HV_X64_POST_MESSAGES;
1857 			ent->ebx |= HV_X64_SIGNAL_EVENTS;
1858 
1859 			ent->edx |= HV_FEATURE_FREQUENCY_MSRS_AVAILABLE;
1860 			ent->edx |= HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
1861 
1862 			/*
1863 			 * Direct Synthetic timers only make sense with in-kernel
1864 			 * LAPIC
1865 			 */
1866 			if (lapic_in_kernel(vcpu))
1867 				ent->edx |= HV_STIMER_DIRECT_MODE_AVAILABLE;
1868 
1869 			break;
1870 
1871 		case HYPERV_CPUID_ENLIGHTMENT_INFO:
1872 			ent->eax |= HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
1873 			ent->eax |= HV_X64_APIC_ACCESS_RECOMMENDED;
1874 			ent->eax |= HV_X64_RELAXED_TIMING_RECOMMENDED;
1875 			ent->eax |= HV_X64_CLUSTER_IPI_RECOMMENDED;
1876 			ent->eax |= HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED;
1877 			if (evmcs_ver)
1878 				ent->eax |= HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
1879 			if (!cpu_smt_possible())
1880 				ent->eax |= HV_X64_NO_NONARCH_CORESHARING;
1881 			/*
1882 			 * Default number of spinlock retry attempts, matches
1883 			 * HyperV 2016.
1884 			 */
1885 			ent->ebx = 0x00000FFF;
1886 
1887 			break;
1888 
1889 		case HYPERV_CPUID_IMPLEMENT_LIMITS:
1890 			/* Maximum number of virtual processors */
1891 			ent->eax = KVM_MAX_VCPUS;
1892 			/*
1893 			 * Maximum number of logical processors, matches
1894 			 * HyperV 2016.
1895 			 */
1896 			ent->ebx = 64;
1897 
1898 			break;
1899 
1900 		case HYPERV_CPUID_NESTED_FEATURES:
1901 			ent->eax = evmcs_ver;
1902 
1903 			break;
1904 
1905 		default:
1906 			break;
1907 		}
1908 	}
1909 
1910 	if (copy_to_user(entries, cpuid_entries,
1911 			 nent * sizeof(struct kvm_cpuid_entry2)))
1912 		return -EFAULT;
1913 
1914 	return 0;
1915 }
1916