xref: /openbmc/linux/arch/x86/kvm/hyperv.c (revision a44e4f3a)
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 
37 #define KVM_HV_MAX_SPARSE_VCPU_SET_BITS DIV_ROUND_UP(KVM_MAX_VCPUS, 64)
38 
39 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
40 				bool vcpu_kick);
41 
42 static inline u64 synic_read_sint(struct kvm_vcpu_hv_synic *synic, int sint)
43 {
44 	return atomic64_read(&synic->sint[sint]);
45 }
46 
47 static inline int synic_get_sint_vector(u64 sint_value)
48 {
49 	if (sint_value & HV_SYNIC_SINT_MASKED)
50 		return -1;
51 	return sint_value & HV_SYNIC_SINT_VECTOR_MASK;
52 }
53 
54 static bool synic_has_vector_connected(struct kvm_vcpu_hv_synic *synic,
55 				      int vector)
56 {
57 	int i;
58 
59 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
60 		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
61 			return true;
62 	}
63 	return false;
64 }
65 
66 static bool synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic *synic,
67 				     int vector)
68 {
69 	int i;
70 	u64 sint_value;
71 
72 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
73 		sint_value = synic_read_sint(synic, i);
74 		if (synic_get_sint_vector(sint_value) == vector &&
75 		    sint_value & HV_SYNIC_SINT_AUTO_EOI)
76 			return true;
77 	}
78 	return false;
79 }
80 
81 static void synic_update_vector(struct kvm_vcpu_hv_synic *synic,
82 				int vector)
83 {
84 	if (vector < HV_SYNIC_FIRST_VALID_VECTOR)
85 		return;
86 
87 	if (synic_has_vector_connected(synic, vector))
88 		__set_bit(vector, synic->vec_bitmap);
89 	else
90 		__clear_bit(vector, synic->vec_bitmap);
91 
92 	if (synic_has_vector_auto_eoi(synic, vector))
93 		__set_bit(vector, synic->auto_eoi_bitmap);
94 	else
95 		__clear_bit(vector, synic->auto_eoi_bitmap);
96 }
97 
98 static int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint,
99 			  u64 data, bool host)
100 {
101 	int vector, old_vector;
102 	bool masked;
103 
104 	vector = data & HV_SYNIC_SINT_VECTOR_MASK;
105 	masked = data & HV_SYNIC_SINT_MASKED;
106 
107 	/*
108 	 * Valid vectors are 16-255, however, nested Hyper-V attempts to write
109 	 * default '0x10000' value on boot and this should not #GP. We need to
110 	 * allow zero-initing the register from host as well.
111 	 */
112 	if (vector < HV_SYNIC_FIRST_VALID_VECTOR && !host && !masked)
113 		return 1;
114 	/*
115 	 * Guest may configure multiple SINTs to use the same vector, so
116 	 * we maintain a bitmap of vectors handled by synic, and a
117 	 * bitmap of vectors with auto-eoi behavior.  The bitmaps are
118 	 * updated here, and atomically queried on fast paths.
119 	 */
120 	old_vector = synic_read_sint(synic, sint) & HV_SYNIC_SINT_VECTOR_MASK;
121 
122 	atomic64_set(&synic->sint[sint], data);
123 
124 	synic_update_vector(synic, old_vector);
125 
126 	synic_update_vector(synic, vector);
127 
128 	/* Load SynIC vectors into EOI exit bitmap */
129 	kvm_make_request(KVM_REQ_SCAN_IOAPIC, synic_to_vcpu(synic));
130 	return 0;
131 }
132 
133 static struct kvm_vcpu *get_vcpu_by_vpidx(struct kvm *kvm, u32 vpidx)
134 {
135 	struct kvm_vcpu *vcpu = NULL;
136 	int i;
137 
138 	if (vpidx >= KVM_MAX_VCPUS)
139 		return NULL;
140 
141 	vcpu = kvm_get_vcpu(kvm, vpidx);
142 	if (vcpu && vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
143 		return vcpu;
144 	kvm_for_each_vcpu(i, vcpu, kvm)
145 		if (vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
146 			return vcpu;
147 	return NULL;
148 }
149 
150 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vpidx)
151 {
152 	struct kvm_vcpu *vcpu;
153 	struct kvm_vcpu_hv_synic *synic;
154 
155 	vcpu = get_vcpu_by_vpidx(kvm, vpidx);
156 	if (!vcpu)
157 		return NULL;
158 	synic = vcpu_to_synic(vcpu);
159 	return (synic->active) ? synic : NULL;
160 }
161 
162 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
163 {
164 	struct kvm *kvm = vcpu->kvm;
165 	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
166 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
167 	struct kvm_vcpu_hv_stimer *stimer;
168 	int gsi, idx;
169 
170 	trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
171 
172 	/* Try to deliver pending Hyper-V SynIC timers messages */
173 	for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
174 		stimer = &hv_vcpu->stimer[idx];
175 		if (stimer->msg_pending && stimer->config.enable &&
176 		    !stimer->config.direct_mode &&
177 		    stimer->config.sintx == sint)
178 			stimer_mark_pending(stimer, false);
179 	}
180 
181 	idx = srcu_read_lock(&kvm->irq_srcu);
182 	gsi = atomic_read(&synic->sint_to_gsi[sint]);
183 	if (gsi != -1)
184 		kvm_notify_acked_gsi(kvm, gsi);
185 	srcu_read_unlock(&kvm->irq_srcu, idx);
186 }
187 
188 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
189 {
190 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
191 	struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
192 
193 	hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
194 	hv_vcpu->exit.u.synic.msr = msr;
195 	hv_vcpu->exit.u.synic.control = synic->control;
196 	hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
197 	hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
198 
199 	kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
200 }
201 
202 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
203 			 u32 msr, u64 data, bool host)
204 {
205 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
206 	int ret;
207 
208 	if (!synic->active && !host)
209 		return 1;
210 
211 	trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
212 
213 	ret = 0;
214 	switch (msr) {
215 	case HV_X64_MSR_SCONTROL:
216 		synic->control = data;
217 		if (!host)
218 			synic_exit(synic, msr);
219 		break;
220 	case HV_X64_MSR_SVERSION:
221 		if (!host) {
222 			ret = 1;
223 			break;
224 		}
225 		synic->version = data;
226 		break;
227 	case HV_X64_MSR_SIEFP:
228 		if ((data & HV_SYNIC_SIEFP_ENABLE) && !host &&
229 		    !synic->dont_zero_synic_pages)
230 			if (kvm_clear_guest(vcpu->kvm,
231 					    data & PAGE_MASK, PAGE_SIZE)) {
232 				ret = 1;
233 				break;
234 			}
235 		synic->evt_page = data;
236 		if (!host)
237 			synic_exit(synic, msr);
238 		break;
239 	case HV_X64_MSR_SIMP:
240 		if ((data & HV_SYNIC_SIMP_ENABLE) && !host &&
241 		    !synic->dont_zero_synic_pages)
242 			if (kvm_clear_guest(vcpu->kvm,
243 					    data & PAGE_MASK, PAGE_SIZE)) {
244 				ret = 1;
245 				break;
246 			}
247 		synic->msg_page = data;
248 		if (!host)
249 			synic_exit(synic, msr);
250 		break;
251 	case HV_X64_MSR_EOM: {
252 		int i;
253 
254 		for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
255 			kvm_hv_notify_acked_sint(vcpu, i);
256 		break;
257 	}
258 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
259 		ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
260 		break;
261 	default:
262 		ret = 1;
263 		break;
264 	}
265 	return ret;
266 }
267 
268 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata,
269 			 bool host)
270 {
271 	int ret;
272 
273 	if (!synic->active && !host)
274 		return 1;
275 
276 	ret = 0;
277 	switch (msr) {
278 	case HV_X64_MSR_SCONTROL:
279 		*pdata = synic->control;
280 		break;
281 	case HV_X64_MSR_SVERSION:
282 		*pdata = synic->version;
283 		break;
284 	case HV_X64_MSR_SIEFP:
285 		*pdata = synic->evt_page;
286 		break;
287 	case HV_X64_MSR_SIMP:
288 		*pdata = synic->msg_page;
289 		break;
290 	case HV_X64_MSR_EOM:
291 		*pdata = 0;
292 		break;
293 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
294 		*pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
295 		break;
296 	default:
297 		ret = 1;
298 		break;
299 	}
300 	return ret;
301 }
302 
303 static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
304 {
305 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
306 	struct kvm_lapic_irq irq;
307 	int ret, vector;
308 
309 	if (sint >= ARRAY_SIZE(synic->sint))
310 		return -EINVAL;
311 
312 	vector = synic_get_sint_vector(synic_read_sint(synic, sint));
313 	if (vector < 0)
314 		return -ENOENT;
315 
316 	memset(&irq, 0, sizeof(irq));
317 	irq.shorthand = APIC_DEST_SELF;
318 	irq.dest_mode = APIC_DEST_PHYSICAL;
319 	irq.delivery_mode = APIC_DM_FIXED;
320 	irq.vector = vector;
321 	irq.level = 1;
322 
323 	ret = kvm_irq_delivery_to_apic(vcpu->kvm, vcpu->arch.apic, &irq, NULL);
324 	trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
325 	return ret;
326 }
327 
328 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vpidx, u32 sint)
329 {
330 	struct kvm_vcpu_hv_synic *synic;
331 
332 	synic = synic_get(kvm, vpidx);
333 	if (!synic)
334 		return -EINVAL;
335 
336 	return synic_set_irq(synic, sint);
337 }
338 
339 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
340 {
341 	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
342 	int i;
343 
344 	trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
345 
346 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
347 		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
348 			kvm_hv_notify_acked_sint(vcpu, i);
349 }
350 
351 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi)
352 {
353 	struct kvm_vcpu_hv_synic *synic;
354 
355 	synic = synic_get(kvm, vpidx);
356 	if (!synic)
357 		return -EINVAL;
358 
359 	if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
360 		return -EINVAL;
361 
362 	atomic_set(&synic->sint_to_gsi[sint], gsi);
363 	return 0;
364 }
365 
366 void kvm_hv_irq_routing_update(struct kvm *kvm)
367 {
368 	struct kvm_irq_routing_table *irq_rt;
369 	struct kvm_kernel_irq_routing_entry *e;
370 	u32 gsi;
371 
372 	irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
373 					lockdep_is_held(&kvm->irq_lock));
374 
375 	for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
376 		hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
377 			if (e->type == KVM_IRQ_ROUTING_HV_SINT)
378 				kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
379 						    e->hv_sint.sint, gsi);
380 		}
381 	}
382 }
383 
384 static void synic_init(struct kvm_vcpu_hv_synic *synic)
385 {
386 	int i;
387 
388 	memset(synic, 0, sizeof(*synic));
389 	synic->version = HV_SYNIC_VERSION_1;
390 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
391 		atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
392 		atomic_set(&synic->sint_to_gsi[i], -1);
393 	}
394 }
395 
396 static u64 get_time_ref_counter(struct kvm *kvm)
397 {
398 	struct kvm_hv *hv = &kvm->arch.hyperv;
399 	struct kvm_vcpu *vcpu;
400 	u64 tsc;
401 
402 	/*
403 	 * The guest has not set up the TSC page or the clock isn't
404 	 * stable, fall back to get_kvmclock_ns.
405 	 */
406 	if (!hv->tsc_ref.tsc_sequence)
407 		return div_u64(get_kvmclock_ns(kvm), 100);
408 
409 	vcpu = kvm_get_vcpu(kvm, 0);
410 	tsc = kvm_read_l1_tsc(vcpu, rdtsc());
411 	return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64)
412 		+ hv->tsc_ref.tsc_offset;
413 }
414 
415 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
416 				bool vcpu_kick)
417 {
418 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
419 
420 	set_bit(stimer->index,
421 		vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
422 	kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
423 	if (vcpu_kick)
424 		kvm_vcpu_kick(vcpu);
425 }
426 
427 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
428 {
429 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
430 
431 	trace_kvm_hv_stimer_cleanup(stimer_to_vcpu(stimer)->vcpu_id,
432 				    stimer->index);
433 
434 	hrtimer_cancel(&stimer->timer);
435 	clear_bit(stimer->index,
436 		  vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
437 	stimer->msg_pending = false;
438 	stimer->exp_time = 0;
439 }
440 
441 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
442 {
443 	struct kvm_vcpu_hv_stimer *stimer;
444 
445 	stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
446 	trace_kvm_hv_stimer_callback(stimer_to_vcpu(stimer)->vcpu_id,
447 				     stimer->index);
448 	stimer_mark_pending(stimer, true);
449 
450 	return HRTIMER_NORESTART;
451 }
452 
453 /*
454  * stimer_start() assumptions:
455  * a) stimer->count is not equal to 0
456  * b) stimer->config has HV_STIMER_ENABLE flag
457  */
458 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
459 {
460 	u64 time_now;
461 	ktime_t ktime_now;
462 
463 	time_now = get_time_ref_counter(stimer_to_vcpu(stimer)->kvm);
464 	ktime_now = ktime_get();
465 
466 	if (stimer->config.periodic) {
467 		if (stimer->exp_time) {
468 			if (time_now >= stimer->exp_time) {
469 				u64 remainder;
470 
471 				div64_u64_rem(time_now - stimer->exp_time,
472 					      stimer->count, &remainder);
473 				stimer->exp_time =
474 					time_now + (stimer->count - remainder);
475 			}
476 		} else
477 			stimer->exp_time = time_now + stimer->count;
478 
479 		trace_kvm_hv_stimer_start_periodic(
480 					stimer_to_vcpu(stimer)->vcpu_id,
481 					stimer->index,
482 					time_now, stimer->exp_time);
483 
484 		hrtimer_start(&stimer->timer,
485 			      ktime_add_ns(ktime_now,
486 					   100 * (stimer->exp_time - time_now)),
487 			      HRTIMER_MODE_ABS);
488 		return 0;
489 	}
490 	stimer->exp_time = stimer->count;
491 	if (time_now >= stimer->count) {
492 		/*
493 		 * Expire timer according to Hypervisor Top-Level Functional
494 		 * specification v4(15.3.1):
495 		 * "If a one shot is enabled and the specified count is in
496 		 * the past, it will expire immediately."
497 		 */
498 		stimer_mark_pending(stimer, false);
499 		return 0;
500 	}
501 
502 	trace_kvm_hv_stimer_start_one_shot(stimer_to_vcpu(stimer)->vcpu_id,
503 					   stimer->index,
504 					   time_now, stimer->count);
505 
506 	hrtimer_start(&stimer->timer,
507 		      ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)),
508 		      HRTIMER_MODE_ABS);
509 	return 0;
510 }
511 
512 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
513 			     bool host)
514 {
515 	union hv_stimer_config new_config = {.as_uint64 = config},
516 		old_config = {.as_uint64 = stimer->config.as_uint64};
517 
518 	trace_kvm_hv_stimer_set_config(stimer_to_vcpu(stimer)->vcpu_id,
519 				       stimer->index, config, host);
520 
521 	stimer_cleanup(stimer);
522 	if (old_config.enable &&
523 	    !new_config.direct_mode && new_config.sintx == 0)
524 		new_config.enable = 0;
525 	stimer->config.as_uint64 = new_config.as_uint64;
526 
527 	if (stimer->config.enable)
528 		stimer_mark_pending(stimer, false);
529 
530 	return 0;
531 }
532 
533 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
534 			    bool host)
535 {
536 	trace_kvm_hv_stimer_set_count(stimer_to_vcpu(stimer)->vcpu_id,
537 				      stimer->index, count, host);
538 
539 	stimer_cleanup(stimer);
540 	stimer->count = count;
541 	if (stimer->count == 0)
542 		stimer->config.enable = 0;
543 	else if (stimer->config.auto_enable)
544 		stimer->config.enable = 1;
545 
546 	if (stimer->config.enable)
547 		stimer_mark_pending(stimer, false);
548 
549 	return 0;
550 }
551 
552 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
553 {
554 	*pconfig = stimer->config.as_uint64;
555 	return 0;
556 }
557 
558 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
559 {
560 	*pcount = stimer->count;
561 	return 0;
562 }
563 
564 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
565 			     struct hv_message *src_msg, bool no_retry)
566 {
567 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
568 	int msg_off = offsetof(struct hv_message_page, sint_message[sint]);
569 	gfn_t msg_page_gfn;
570 	struct hv_message_header hv_hdr;
571 	int r;
572 
573 	if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
574 		return -ENOENT;
575 
576 	msg_page_gfn = synic->msg_page >> PAGE_SHIFT;
577 
578 	/*
579 	 * Strictly following the spec-mandated ordering would assume setting
580 	 * .msg_pending before checking .message_type.  However, this function
581 	 * is only called in vcpu context so the entire update is atomic from
582 	 * guest POV and thus the exact order here doesn't matter.
583 	 */
584 	r = kvm_vcpu_read_guest_page(vcpu, msg_page_gfn, &hv_hdr.message_type,
585 				     msg_off + offsetof(struct hv_message,
586 							header.message_type),
587 				     sizeof(hv_hdr.message_type));
588 	if (r < 0)
589 		return r;
590 
591 	if (hv_hdr.message_type != HVMSG_NONE) {
592 		if (no_retry)
593 			return 0;
594 
595 		hv_hdr.message_flags.msg_pending = 1;
596 		r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn,
597 					      &hv_hdr.message_flags,
598 					      msg_off +
599 					      offsetof(struct hv_message,
600 						       header.message_flags),
601 					      sizeof(hv_hdr.message_flags));
602 		if (r < 0)
603 			return r;
604 		return -EAGAIN;
605 	}
606 
607 	r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn, src_msg, msg_off,
608 				      sizeof(src_msg->header) +
609 				      src_msg->header.payload_size);
610 	if (r < 0)
611 		return r;
612 
613 	r = synic_set_irq(synic, sint);
614 	if (r < 0)
615 		return r;
616 	if (r == 0)
617 		return -EFAULT;
618 	return 0;
619 }
620 
621 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
622 {
623 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
624 	struct hv_message *msg = &stimer->msg;
625 	struct hv_timer_message_payload *payload =
626 			(struct hv_timer_message_payload *)&msg->u.payload;
627 
628 	/*
629 	 * To avoid piling up periodic ticks, don't retry message
630 	 * delivery for them (within "lazy" lost ticks policy).
631 	 */
632 	bool no_retry = stimer->config.periodic;
633 
634 	payload->expiration_time = stimer->exp_time;
635 	payload->delivery_time = get_time_ref_counter(vcpu->kvm);
636 	return synic_deliver_msg(vcpu_to_synic(vcpu),
637 				 stimer->config.sintx, msg,
638 				 no_retry);
639 }
640 
641 static int stimer_notify_direct(struct kvm_vcpu_hv_stimer *stimer)
642 {
643 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
644 	struct kvm_lapic_irq irq = {
645 		.delivery_mode = APIC_DM_FIXED,
646 		.vector = stimer->config.apic_vector
647 	};
648 
649 	if (lapic_in_kernel(vcpu))
650 		return !kvm_apic_set_irq(vcpu, &irq, NULL);
651 	return 0;
652 }
653 
654 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
655 {
656 	int r, direct = stimer->config.direct_mode;
657 
658 	stimer->msg_pending = true;
659 	if (!direct)
660 		r = stimer_send_msg(stimer);
661 	else
662 		r = stimer_notify_direct(stimer);
663 	trace_kvm_hv_stimer_expiration(stimer_to_vcpu(stimer)->vcpu_id,
664 				       stimer->index, direct, r);
665 	if (!r) {
666 		stimer->msg_pending = false;
667 		if (!(stimer->config.periodic))
668 			stimer->config.enable = 0;
669 	}
670 }
671 
672 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
673 {
674 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
675 	struct kvm_vcpu_hv_stimer *stimer;
676 	u64 time_now, exp_time;
677 	int i;
678 
679 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
680 		if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
681 			stimer = &hv_vcpu->stimer[i];
682 			if (stimer->config.enable) {
683 				exp_time = stimer->exp_time;
684 
685 				if (exp_time) {
686 					time_now =
687 						get_time_ref_counter(vcpu->kvm);
688 					if (time_now >= exp_time)
689 						stimer_expiration(stimer);
690 				}
691 
692 				if ((stimer->config.enable) &&
693 				    stimer->count) {
694 					if (!stimer->msg_pending)
695 						stimer_start(stimer);
696 				} else
697 					stimer_cleanup(stimer);
698 			}
699 		}
700 }
701 
702 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
703 {
704 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
705 	int i;
706 
707 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
708 		stimer_cleanup(&hv_vcpu->stimer[i]);
709 }
710 
711 bool kvm_hv_assist_page_enabled(struct kvm_vcpu *vcpu)
712 {
713 	if (!(vcpu->arch.hyperv.hv_vapic & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE))
714 		return false;
715 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
716 }
717 EXPORT_SYMBOL_GPL(kvm_hv_assist_page_enabled);
718 
719 bool kvm_hv_get_assist_page(struct kvm_vcpu *vcpu,
720 			    struct hv_vp_assist_page *assist_page)
721 {
722 	if (!kvm_hv_assist_page_enabled(vcpu))
723 		return false;
724 	return !kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data,
725 				      assist_page, sizeof(*assist_page));
726 }
727 EXPORT_SYMBOL_GPL(kvm_hv_get_assist_page);
728 
729 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
730 {
731 	struct hv_message *msg = &stimer->msg;
732 	struct hv_timer_message_payload *payload =
733 			(struct hv_timer_message_payload *)&msg->u.payload;
734 
735 	memset(&msg->header, 0, sizeof(msg->header));
736 	msg->header.message_type = HVMSG_TIMER_EXPIRED;
737 	msg->header.payload_size = sizeof(*payload);
738 
739 	payload->timer_index = stimer->index;
740 	payload->expiration_time = 0;
741 	payload->delivery_time = 0;
742 }
743 
744 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
745 {
746 	memset(stimer, 0, sizeof(*stimer));
747 	stimer->index = timer_index;
748 	hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
749 	stimer->timer.function = stimer_timer_callback;
750 	stimer_prepare_msg(stimer);
751 }
752 
753 void kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
754 {
755 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
756 	int i;
757 
758 	synic_init(&hv_vcpu->synic);
759 
760 	bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
761 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
762 		stimer_init(&hv_vcpu->stimer[i], i);
763 }
764 
765 void kvm_hv_vcpu_postcreate(struct kvm_vcpu *vcpu)
766 {
767 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
768 
769 	hv_vcpu->vp_index = kvm_vcpu_get_idx(vcpu);
770 }
771 
772 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
773 {
774 	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
775 
776 	/*
777 	 * Hyper-V SynIC auto EOI SINT's are
778 	 * not compatible with APICV, so deactivate APICV
779 	 */
780 	kvm_vcpu_deactivate_apicv(vcpu);
781 	synic->active = true;
782 	synic->dont_zero_synic_pages = dont_zero_synic_pages;
783 	return 0;
784 }
785 
786 static bool kvm_hv_msr_partition_wide(u32 msr)
787 {
788 	bool r = false;
789 
790 	switch (msr) {
791 	case HV_X64_MSR_GUEST_OS_ID:
792 	case HV_X64_MSR_HYPERCALL:
793 	case HV_X64_MSR_REFERENCE_TSC:
794 	case HV_X64_MSR_TIME_REF_COUNT:
795 	case HV_X64_MSR_CRASH_CTL:
796 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
797 	case HV_X64_MSR_RESET:
798 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
799 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
800 	case HV_X64_MSR_TSC_EMULATION_STATUS:
801 		r = true;
802 		break;
803 	}
804 
805 	return r;
806 }
807 
808 static int kvm_hv_msr_get_crash_data(struct kvm_vcpu *vcpu,
809 				     u32 index, u64 *pdata)
810 {
811 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
812 
813 	if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param)))
814 		return -EINVAL;
815 
816 	*pdata = hv->hv_crash_param[index];
817 	return 0;
818 }
819 
820 static int kvm_hv_msr_get_crash_ctl(struct kvm_vcpu *vcpu, u64 *pdata)
821 {
822 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
823 
824 	*pdata = hv->hv_crash_ctl;
825 	return 0;
826 }
827 
828 static int kvm_hv_msr_set_crash_ctl(struct kvm_vcpu *vcpu, u64 data, bool host)
829 {
830 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
831 
832 	if (host)
833 		hv->hv_crash_ctl = data & HV_CRASH_CTL_CRASH_NOTIFY;
834 
835 	if (!host && (data & HV_CRASH_CTL_CRASH_NOTIFY)) {
836 
837 		vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
838 			  hv->hv_crash_param[0],
839 			  hv->hv_crash_param[1],
840 			  hv->hv_crash_param[2],
841 			  hv->hv_crash_param[3],
842 			  hv->hv_crash_param[4]);
843 
844 		/* Send notification about crash to user space */
845 		kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
846 	}
847 
848 	return 0;
849 }
850 
851 static int kvm_hv_msr_set_crash_data(struct kvm_vcpu *vcpu,
852 				     u32 index, u64 data)
853 {
854 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
855 
856 	if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param)))
857 		return -EINVAL;
858 
859 	hv->hv_crash_param[index] = data;
860 	return 0;
861 }
862 
863 /*
864  * The kvmclock and Hyper-V TSC page use similar formulas, and converting
865  * between them is possible:
866  *
867  * kvmclock formula:
868  *    nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
869  *           + system_time
870  *
871  * Hyper-V formula:
872  *    nsec/100 = ticks * scale / 2^64 + offset
873  *
874  * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
875  * By dividing the kvmclock formula by 100 and equating what's left we get:
876  *    ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
877  *            scale / 2^64 =         tsc_to_system_mul * 2^(tsc_shift-32) / 100
878  *            scale        =         tsc_to_system_mul * 2^(32+tsc_shift) / 100
879  *
880  * Now expand the kvmclock formula and divide by 100:
881  *    nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
882  *           - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
883  *           + system_time
884  *    nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
885  *               - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
886  *               + system_time / 100
887  *
888  * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
889  *    nsec/100 = ticks * scale / 2^64
890  *               - tsc_timestamp * scale / 2^64
891  *               + system_time / 100
892  *
893  * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
894  *    offset = system_time / 100 - tsc_timestamp * scale / 2^64
895  *
896  * These two equivalencies are implemented in this function.
897  */
898 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
899 					HV_REFERENCE_TSC_PAGE *tsc_ref)
900 {
901 	u64 max_mul;
902 
903 	if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
904 		return false;
905 
906 	/*
907 	 * check if scale would overflow, if so we use the time ref counter
908 	 *    tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
909 	 *    tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
910 	 *    tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
911 	 */
912 	max_mul = 100ull << (32 - hv_clock->tsc_shift);
913 	if (hv_clock->tsc_to_system_mul >= max_mul)
914 		return false;
915 
916 	/*
917 	 * Otherwise compute the scale and offset according to the formulas
918 	 * derived above.
919 	 */
920 	tsc_ref->tsc_scale =
921 		mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
922 				hv_clock->tsc_to_system_mul,
923 				100);
924 
925 	tsc_ref->tsc_offset = hv_clock->system_time;
926 	do_div(tsc_ref->tsc_offset, 100);
927 	tsc_ref->tsc_offset -=
928 		mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
929 	return true;
930 }
931 
932 void kvm_hv_setup_tsc_page(struct kvm *kvm,
933 			   struct pvclock_vcpu_time_info *hv_clock)
934 {
935 	struct kvm_hv *hv = &kvm->arch.hyperv;
936 	u32 tsc_seq;
937 	u64 gfn;
938 
939 	BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
940 	BUILD_BUG_ON(offsetof(HV_REFERENCE_TSC_PAGE, tsc_sequence) != 0);
941 
942 	if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
943 		return;
944 
945 	mutex_lock(&kvm->arch.hyperv.hv_lock);
946 	if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
947 		goto out_unlock;
948 
949 	gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
950 	/*
951 	 * Because the TSC parameters only vary when there is a
952 	 * change in the master clock, do not bother with caching.
953 	 */
954 	if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
955 				    &tsc_seq, sizeof(tsc_seq))))
956 		goto out_unlock;
957 
958 	/*
959 	 * While we're computing and writing the parameters, force the
960 	 * guest to use the time reference count MSR.
961 	 */
962 	hv->tsc_ref.tsc_sequence = 0;
963 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
964 			    &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
965 		goto out_unlock;
966 
967 	if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
968 		goto out_unlock;
969 
970 	/* Ensure sequence is zero before writing the rest of the struct.  */
971 	smp_wmb();
972 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
973 		goto out_unlock;
974 
975 	/*
976 	 * Now switch to the TSC page mechanism by writing the sequence.
977 	 */
978 	tsc_seq++;
979 	if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
980 		tsc_seq = 1;
981 
982 	/* Write the struct entirely before the non-zero sequence.  */
983 	smp_wmb();
984 
985 	hv->tsc_ref.tsc_sequence = tsc_seq;
986 	kvm_write_guest(kvm, gfn_to_gpa(gfn),
987 			&hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence));
988 out_unlock:
989 	mutex_unlock(&kvm->arch.hyperv.hv_lock);
990 }
991 
992 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
993 			     bool host)
994 {
995 	struct kvm *kvm = vcpu->kvm;
996 	struct kvm_hv *hv = &kvm->arch.hyperv;
997 
998 	switch (msr) {
999 	case HV_X64_MSR_GUEST_OS_ID:
1000 		hv->hv_guest_os_id = data;
1001 		/* setting guest os id to zero disables hypercall page */
1002 		if (!hv->hv_guest_os_id)
1003 			hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1004 		break;
1005 	case HV_X64_MSR_HYPERCALL: {
1006 		u64 gfn;
1007 		unsigned long addr;
1008 		u8 instructions[4];
1009 
1010 		/* if guest os id is not set hypercall should remain disabled */
1011 		if (!hv->hv_guest_os_id)
1012 			break;
1013 		if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1014 			hv->hv_hypercall = data;
1015 			break;
1016 		}
1017 		gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1018 		addr = gfn_to_hva(kvm, gfn);
1019 		if (kvm_is_error_hva(addr))
1020 			return 1;
1021 		kvm_x86_ops->patch_hypercall(vcpu, instructions);
1022 		((unsigned char *)instructions)[3] = 0xc3; /* ret */
1023 		if (__copy_to_user((void __user *)addr, instructions, 4))
1024 			return 1;
1025 		hv->hv_hypercall = data;
1026 		mark_page_dirty(kvm, gfn);
1027 		break;
1028 	}
1029 	case HV_X64_MSR_REFERENCE_TSC:
1030 		hv->hv_tsc_page = data;
1031 		if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE)
1032 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1033 		break;
1034 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1035 		return kvm_hv_msr_set_crash_data(vcpu,
1036 						 msr - HV_X64_MSR_CRASH_P0,
1037 						 data);
1038 	case HV_X64_MSR_CRASH_CTL:
1039 		return kvm_hv_msr_set_crash_ctl(vcpu, data, host);
1040 	case HV_X64_MSR_RESET:
1041 		if (data == 1) {
1042 			vcpu_debug(vcpu, "hyper-v reset requested\n");
1043 			kvm_make_request(KVM_REQ_HV_RESET, vcpu);
1044 		}
1045 		break;
1046 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1047 		hv->hv_reenlightenment_control = data;
1048 		break;
1049 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1050 		hv->hv_tsc_emulation_control = data;
1051 		break;
1052 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1053 		hv->hv_tsc_emulation_status = data;
1054 		break;
1055 	case HV_X64_MSR_TIME_REF_COUNT:
1056 		/* read-only, but still ignore it if host-initiated */
1057 		if (!host)
1058 			return 1;
1059 		break;
1060 	default:
1061 		vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
1062 			    msr, data);
1063 		return 1;
1064 	}
1065 	return 0;
1066 }
1067 
1068 /* Calculate cpu time spent by current task in 100ns units */
1069 static u64 current_task_runtime_100ns(void)
1070 {
1071 	u64 utime, stime;
1072 
1073 	task_cputime_adjusted(current, &utime, &stime);
1074 
1075 	return div_u64(utime + stime, 100);
1076 }
1077 
1078 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1079 {
1080 	struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
1081 
1082 	switch (msr) {
1083 	case HV_X64_MSR_VP_INDEX: {
1084 		struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
1085 		int vcpu_idx = kvm_vcpu_get_idx(vcpu);
1086 		u32 new_vp_index = (u32)data;
1087 
1088 		if (!host || new_vp_index >= KVM_MAX_VCPUS)
1089 			return 1;
1090 
1091 		if (new_vp_index == hv_vcpu->vp_index)
1092 			return 0;
1093 
1094 		/*
1095 		 * The VP index is initialized to vcpu_index by
1096 		 * kvm_hv_vcpu_postcreate so they initially match.  Now the
1097 		 * VP index is changing, adjust num_mismatched_vp_indexes if
1098 		 * it now matches or no longer matches vcpu_idx.
1099 		 */
1100 		if (hv_vcpu->vp_index == vcpu_idx)
1101 			atomic_inc(&hv->num_mismatched_vp_indexes);
1102 		else if (new_vp_index == vcpu_idx)
1103 			atomic_dec(&hv->num_mismatched_vp_indexes);
1104 
1105 		hv_vcpu->vp_index = new_vp_index;
1106 		break;
1107 	}
1108 	case HV_X64_MSR_VP_ASSIST_PAGE: {
1109 		u64 gfn;
1110 		unsigned long addr;
1111 
1112 		if (!(data & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE)) {
1113 			hv_vcpu->hv_vapic = data;
1114 			if (kvm_lapic_enable_pv_eoi(vcpu, 0, 0))
1115 				return 1;
1116 			break;
1117 		}
1118 		gfn = data >> HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT;
1119 		addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1120 		if (kvm_is_error_hva(addr))
1121 			return 1;
1122 
1123 		/*
1124 		 * Clear apic_assist portion of f(struct hv_vp_assist_page
1125 		 * only, there can be valuable data in the rest which needs
1126 		 * to be preserved e.g. on migration.
1127 		 */
1128 		if (__clear_user((void __user *)addr, sizeof(u32)))
1129 			return 1;
1130 		hv_vcpu->hv_vapic = data;
1131 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
1132 		if (kvm_lapic_enable_pv_eoi(vcpu,
1133 					    gfn_to_gpa(gfn) | KVM_MSR_ENABLED,
1134 					    sizeof(struct hv_vp_assist_page)))
1135 			return 1;
1136 		break;
1137 	}
1138 	case HV_X64_MSR_EOI:
1139 		return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1140 	case HV_X64_MSR_ICR:
1141 		return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1142 	case HV_X64_MSR_TPR:
1143 		return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1144 	case HV_X64_MSR_VP_RUNTIME:
1145 		if (!host)
1146 			return 1;
1147 		hv_vcpu->runtime_offset = data - current_task_runtime_100ns();
1148 		break;
1149 	case HV_X64_MSR_SCONTROL:
1150 	case HV_X64_MSR_SVERSION:
1151 	case HV_X64_MSR_SIEFP:
1152 	case HV_X64_MSR_SIMP:
1153 	case HV_X64_MSR_EOM:
1154 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1155 		return synic_set_msr(vcpu_to_synic(vcpu), msr, data, host);
1156 	case HV_X64_MSR_STIMER0_CONFIG:
1157 	case HV_X64_MSR_STIMER1_CONFIG:
1158 	case HV_X64_MSR_STIMER2_CONFIG:
1159 	case HV_X64_MSR_STIMER3_CONFIG: {
1160 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1161 
1162 		return stimer_set_config(vcpu_to_stimer(vcpu, timer_index),
1163 					 data, host);
1164 	}
1165 	case HV_X64_MSR_STIMER0_COUNT:
1166 	case HV_X64_MSR_STIMER1_COUNT:
1167 	case HV_X64_MSR_STIMER2_COUNT:
1168 	case HV_X64_MSR_STIMER3_COUNT: {
1169 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1170 
1171 		return stimer_set_count(vcpu_to_stimer(vcpu, timer_index),
1172 					data, host);
1173 	}
1174 	case HV_X64_MSR_TSC_FREQUENCY:
1175 	case HV_X64_MSR_APIC_FREQUENCY:
1176 		/* read-only, but still ignore it if host-initiated */
1177 		if (!host)
1178 			return 1;
1179 		break;
1180 	default:
1181 		vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
1182 			    msr, data);
1183 		return 1;
1184 	}
1185 
1186 	return 0;
1187 }
1188 
1189 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1190 {
1191 	u64 data = 0;
1192 	struct kvm *kvm = vcpu->kvm;
1193 	struct kvm_hv *hv = &kvm->arch.hyperv;
1194 
1195 	switch (msr) {
1196 	case HV_X64_MSR_GUEST_OS_ID:
1197 		data = hv->hv_guest_os_id;
1198 		break;
1199 	case HV_X64_MSR_HYPERCALL:
1200 		data = hv->hv_hypercall;
1201 		break;
1202 	case HV_X64_MSR_TIME_REF_COUNT:
1203 		data = get_time_ref_counter(kvm);
1204 		break;
1205 	case HV_X64_MSR_REFERENCE_TSC:
1206 		data = hv->hv_tsc_page;
1207 		break;
1208 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1209 		return kvm_hv_msr_get_crash_data(vcpu,
1210 						 msr - HV_X64_MSR_CRASH_P0,
1211 						 pdata);
1212 	case HV_X64_MSR_CRASH_CTL:
1213 		return kvm_hv_msr_get_crash_ctl(vcpu, pdata);
1214 	case HV_X64_MSR_RESET:
1215 		data = 0;
1216 		break;
1217 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1218 		data = hv->hv_reenlightenment_control;
1219 		break;
1220 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1221 		data = hv->hv_tsc_emulation_control;
1222 		break;
1223 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1224 		data = hv->hv_tsc_emulation_status;
1225 		break;
1226 	default:
1227 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1228 		return 1;
1229 	}
1230 
1231 	*pdata = data;
1232 	return 0;
1233 }
1234 
1235 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1236 			  bool host)
1237 {
1238 	u64 data = 0;
1239 	struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
1240 
1241 	switch (msr) {
1242 	case HV_X64_MSR_VP_INDEX:
1243 		data = hv_vcpu->vp_index;
1244 		break;
1245 	case HV_X64_MSR_EOI:
1246 		return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1247 	case HV_X64_MSR_ICR:
1248 		return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1249 	case HV_X64_MSR_TPR:
1250 		return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1251 	case HV_X64_MSR_VP_ASSIST_PAGE:
1252 		data = hv_vcpu->hv_vapic;
1253 		break;
1254 	case HV_X64_MSR_VP_RUNTIME:
1255 		data = current_task_runtime_100ns() + hv_vcpu->runtime_offset;
1256 		break;
1257 	case HV_X64_MSR_SCONTROL:
1258 	case HV_X64_MSR_SVERSION:
1259 	case HV_X64_MSR_SIEFP:
1260 	case HV_X64_MSR_SIMP:
1261 	case HV_X64_MSR_EOM:
1262 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1263 		return synic_get_msr(vcpu_to_synic(vcpu), msr, pdata, host);
1264 	case HV_X64_MSR_STIMER0_CONFIG:
1265 	case HV_X64_MSR_STIMER1_CONFIG:
1266 	case HV_X64_MSR_STIMER2_CONFIG:
1267 	case HV_X64_MSR_STIMER3_CONFIG: {
1268 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1269 
1270 		return stimer_get_config(vcpu_to_stimer(vcpu, timer_index),
1271 					 pdata);
1272 	}
1273 	case HV_X64_MSR_STIMER0_COUNT:
1274 	case HV_X64_MSR_STIMER1_COUNT:
1275 	case HV_X64_MSR_STIMER2_COUNT:
1276 	case HV_X64_MSR_STIMER3_COUNT: {
1277 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1278 
1279 		return stimer_get_count(vcpu_to_stimer(vcpu, timer_index),
1280 					pdata);
1281 	}
1282 	case HV_X64_MSR_TSC_FREQUENCY:
1283 		data = (u64)vcpu->arch.virtual_tsc_khz * 1000;
1284 		break;
1285 	case HV_X64_MSR_APIC_FREQUENCY:
1286 		data = APIC_BUS_FREQUENCY;
1287 		break;
1288 	default:
1289 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1290 		return 1;
1291 	}
1292 	*pdata = data;
1293 	return 0;
1294 }
1295 
1296 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1297 {
1298 	if (kvm_hv_msr_partition_wide(msr)) {
1299 		int r;
1300 
1301 		mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1302 		r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1303 		mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1304 		return r;
1305 	} else
1306 		return kvm_hv_set_msr(vcpu, msr, data, host);
1307 }
1308 
1309 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
1310 {
1311 	if (kvm_hv_msr_partition_wide(msr)) {
1312 		int r;
1313 
1314 		mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1315 		r = kvm_hv_get_msr_pw(vcpu, msr, pdata);
1316 		mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1317 		return r;
1318 	} else
1319 		return kvm_hv_get_msr(vcpu, msr, pdata, host);
1320 }
1321 
1322 static __always_inline unsigned long *sparse_set_to_vcpu_mask(
1323 	struct kvm *kvm, u64 *sparse_banks, u64 valid_bank_mask,
1324 	u64 *vp_bitmap, unsigned long *vcpu_bitmap)
1325 {
1326 	struct kvm_hv *hv = &kvm->arch.hyperv;
1327 	struct kvm_vcpu *vcpu;
1328 	int i, bank, sbank = 0;
1329 
1330 	memset(vp_bitmap, 0,
1331 	       KVM_HV_MAX_SPARSE_VCPU_SET_BITS * sizeof(*vp_bitmap));
1332 	for_each_set_bit(bank, (unsigned long *)&valid_bank_mask,
1333 			 KVM_HV_MAX_SPARSE_VCPU_SET_BITS)
1334 		vp_bitmap[bank] = sparse_banks[sbank++];
1335 
1336 	if (likely(!atomic_read(&hv->num_mismatched_vp_indexes))) {
1337 		/* for all vcpus vp_index == vcpu_idx */
1338 		return (unsigned long *)vp_bitmap;
1339 	}
1340 
1341 	bitmap_zero(vcpu_bitmap, KVM_MAX_VCPUS);
1342 	kvm_for_each_vcpu(i, vcpu, kvm) {
1343 		if (test_bit(vcpu_to_hv_vcpu(vcpu)->vp_index,
1344 			     (unsigned long *)vp_bitmap))
1345 			__set_bit(i, vcpu_bitmap);
1346 	}
1347 	return vcpu_bitmap;
1348 }
1349 
1350 static u64 kvm_hv_flush_tlb(struct kvm_vcpu *current_vcpu, u64 ingpa,
1351 			    u16 rep_cnt, bool ex)
1352 {
1353 	struct kvm *kvm = current_vcpu->kvm;
1354 	struct kvm_vcpu_hv *hv_vcpu = &current_vcpu->arch.hyperv;
1355 	struct hv_tlb_flush_ex flush_ex;
1356 	struct hv_tlb_flush flush;
1357 	u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1358 	DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1359 	unsigned long *vcpu_mask;
1360 	u64 valid_bank_mask;
1361 	u64 sparse_banks[64];
1362 	int sparse_banks_len;
1363 	bool all_cpus;
1364 
1365 	if (!ex) {
1366 		if (unlikely(kvm_read_guest(kvm, ingpa, &flush, sizeof(flush))))
1367 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1368 
1369 		trace_kvm_hv_flush_tlb(flush.processor_mask,
1370 				       flush.address_space, flush.flags);
1371 
1372 		valid_bank_mask = BIT_ULL(0);
1373 		sparse_banks[0] = flush.processor_mask;
1374 
1375 		/*
1376 		 * Work around possible WS2012 bug: it sends hypercalls
1377 		 * with processor_mask = 0x0 and HV_FLUSH_ALL_PROCESSORS clear,
1378 		 * while also expecting us to flush something and crashing if
1379 		 * we don't. Let's treat processor_mask == 0 same as
1380 		 * HV_FLUSH_ALL_PROCESSORS.
1381 		 */
1382 		all_cpus = (flush.flags & HV_FLUSH_ALL_PROCESSORS) ||
1383 			flush.processor_mask == 0;
1384 	} else {
1385 		if (unlikely(kvm_read_guest(kvm, ingpa, &flush_ex,
1386 					    sizeof(flush_ex))))
1387 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1388 
1389 		trace_kvm_hv_flush_tlb_ex(flush_ex.hv_vp_set.valid_bank_mask,
1390 					  flush_ex.hv_vp_set.format,
1391 					  flush_ex.address_space,
1392 					  flush_ex.flags);
1393 
1394 		valid_bank_mask = flush_ex.hv_vp_set.valid_bank_mask;
1395 		all_cpus = flush_ex.hv_vp_set.format !=
1396 			HV_GENERIC_SET_SPARSE_4K;
1397 
1398 		sparse_banks_len =
1399 			bitmap_weight((unsigned long *)&valid_bank_mask, 64) *
1400 			sizeof(sparse_banks[0]);
1401 
1402 		if (!sparse_banks_len && !all_cpus)
1403 			goto ret_success;
1404 
1405 		if (!all_cpus &&
1406 		    kvm_read_guest(kvm,
1407 				   ingpa + offsetof(struct hv_tlb_flush_ex,
1408 						    hv_vp_set.bank_contents),
1409 				   sparse_banks,
1410 				   sparse_banks_len))
1411 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1412 	}
1413 
1414 	cpumask_clear(&hv_vcpu->tlb_flush);
1415 
1416 	vcpu_mask = all_cpus ? NULL :
1417 		sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1418 					vp_bitmap, vcpu_bitmap);
1419 
1420 	/*
1421 	 * vcpu->arch.cr3 may not be up-to-date for running vCPUs so we can't
1422 	 * analyze it here, flush TLB regardless of the specified address space.
1423 	 */
1424 	kvm_make_vcpus_request_mask(kvm,
1425 				    KVM_REQ_TLB_FLUSH | KVM_REQUEST_NO_WAKEUP,
1426 				    vcpu_mask, &hv_vcpu->tlb_flush);
1427 
1428 ret_success:
1429 	/* We always do full TLB flush, set rep_done = rep_cnt. */
1430 	return (u64)HV_STATUS_SUCCESS |
1431 		((u64)rep_cnt << HV_HYPERCALL_REP_COMP_OFFSET);
1432 }
1433 
1434 static void kvm_send_ipi_to_many(struct kvm *kvm, u32 vector,
1435 				 unsigned long *vcpu_bitmap)
1436 {
1437 	struct kvm_lapic_irq irq = {
1438 		.delivery_mode = APIC_DM_FIXED,
1439 		.vector = vector
1440 	};
1441 	struct kvm_vcpu *vcpu;
1442 	int i;
1443 
1444 	kvm_for_each_vcpu(i, vcpu, kvm) {
1445 		if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
1446 			continue;
1447 
1448 		/* We fail only when APIC is disabled */
1449 		kvm_apic_set_irq(vcpu, &irq, NULL);
1450 	}
1451 }
1452 
1453 static u64 kvm_hv_send_ipi(struct kvm_vcpu *current_vcpu, u64 ingpa, u64 outgpa,
1454 			   bool ex, bool fast)
1455 {
1456 	struct kvm *kvm = current_vcpu->kvm;
1457 	struct hv_send_ipi_ex send_ipi_ex;
1458 	struct hv_send_ipi send_ipi;
1459 	u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1460 	DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1461 	unsigned long *vcpu_mask;
1462 	unsigned long valid_bank_mask;
1463 	u64 sparse_banks[64];
1464 	int sparse_banks_len;
1465 	u32 vector;
1466 	bool all_cpus;
1467 
1468 	if (!ex) {
1469 		if (!fast) {
1470 			if (unlikely(kvm_read_guest(kvm, ingpa, &send_ipi,
1471 						    sizeof(send_ipi))))
1472 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
1473 			sparse_banks[0] = send_ipi.cpu_mask;
1474 			vector = send_ipi.vector;
1475 		} else {
1476 			/* 'reserved' part of hv_send_ipi should be 0 */
1477 			if (unlikely(ingpa >> 32 != 0))
1478 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
1479 			sparse_banks[0] = outgpa;
1480 			vector = (u32)ingpa;
1481 		}
1482 		all_cpus = false;
1483 		valid_bank_mask = BIT_ULL(0);
1484 
1485 		trace_kvm_hv_send_ipi(vector, sparse_banks[0]);
1486 	} else {
1487 		if (unlikely(kvm_read_guest(kvm, ingpa, &send_ipi_ex,
1488 					    sizeof(send_ipi_ex))))
1489 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1490 
1491 		trace_kvm_hv_send_ipi_ex(send_ipi_ex.vector,
1492 					 send_ipi_ex.vp_set.format,
1493 					 send_ipi_ex.vp_set.valid_bank_mask);
1494 
1495 		vector = send_ipi_ex.vector;
1496 		valid_bank_mask = send_ipi_ex.vp_set.valid_bank_mask;
1497 		sparse_banks_len = bitmap_weight(&valid_bank_mask, 64) *
1498 			sizeof(sparse_banks[0]);
1499 
1500 		all_cpus = send_ipi_ex.vp_set.format == HV_GENERIC_SET_ALL;
1501 
1502 		if (!sparse_banks_len)
1503 			goto ret_success;
1504 
1505 		if (!all_cpus &&
1506 		    kvm_read_guest(kvm,
1507 				   ingpa + offsetof(struct hv_send_ipi_ex,
1508 						    vp_set.bank_contents),
1509 				   sparse_banks,
1510 				   sparse_banks_len))
1511 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1512 	}
1513 
1514 	if ((vector < HV_IPI_LOW_VECTOR) || (vector > HV_IPI_HIGH_VECTOR))
1515 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
1516 
1517 	vcpu_mask = all_cpus ? NULL :
1518 		sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1519 					vp_bitmap, vcpu_bitmap);
1520 
1521 	kvm_send_ipi_to_many(kvm, vector, vcpu_mask);
1522 
1523 ret_success:
1524 	return HV_STATUS_SUCCESS;
1525 }
1526 
1527 bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1528 {
1529 	return READ_ONCE(kvm->arch.hyperv.hv_hypercall) & HV_X64_MSR_HYPERCALL_ENABLE;
1530 }
1531 
1532 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
1533 {
1534 	bool longmode;
1535 
1536 	longmode = is_64_bit_mode(vcpu);
1537 	if (longmode)
1538 		kvm_rax_write(vcpu, result);
1539 	else {
1540 		kvm_rdx_write(vcpu, result >> 32);
1541 		kvm_rax_write(vcpu, result & 0xffffffff);
1542 	}
1543 }
1544 
1545 static int kvm_hv_hypercall_complete(struct kvm_vcpu *vcpu, u64 result)
1546 {
1547 	kvm_hv_hypercall_set_result(vcpu, result);
1548 	++vcpu->stat.hypercalls;
1549 	return kvm_skip_emulated_instruction(vcpu);
1550 }
1551 
1552 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
1553 {
1554 	return kvm_hv_hypercall_complete(vcpu, vcpu->run->hyperv.u.hcall.result);
1555 }
1556 
1557 static u16 kvm_hvcall_signal_event(struct kvm_vcpu *vcpu, bool fast, u64 param)
1558 {
1559 	struct eventfd_ctx *eventfd;
1560 
1561 	if (unlikely(!fast)) {
1562 		int ret;
1563 		gpa_t gpa = param;
1564 
1565 		if ((gpa & (__alignof__(param) - 1)) ||
1566 		    offset_in_page(gpa) + sizeof(param) > PAGE_SIZE)
1567 			return HV_STATUS_INVALID_ALIGNMENT;
1568 
1569 		ret = kvm_vcpu_read_guest(vcpu, gpa, &param, sizeof(param));
1570 		if (ret < 0)
1571 			return HV_STATUS_INVALID_ALIGNMENT;
1572 	}
1573 
1574 	/*
1575 	 * Per spec, bits 32-47 contain the extra "flag number".  However, we
1576 	 * have no use for it, and in all known usecases it is zero, so just
1577 	 * report lookup failure if it isn't.
1578 	 */
1579 	if (param & 0xffff00000000ULL)
1580 		return HV_STATUS_INVALID_PORT_ID;
1581 	/* remaining bits are reserved-zero */
1582 	if (param & ~KVM_HYPERV_CONN_ID_MASK)
1583 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
1584 
1585 	/* the eventfd is protected by vcpu->kvm->srcu, but conn_to_evt isn't */
1586 	rcu_read_lock();
1587 	eventfd = idr_find(&vcpu->kvm->arch.hyperv.conn_to_evt, param);
1588 	rcu_read_unlock();
1589 	if (!eventfd)
1590 		return HV_STATUS_INVALID_PORT_ID;
1591 
1592 	eventfd_signal(eventfd, 1);
1593 	return HV_STATUS_SUCCESS;
1594 }
1595 
1596 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
1597 {
1598 	u64 param, ingpa, outgpa, ret = HV_STATUS_SUCCESS;
1599 	uint16_t code, rep_idx, rep_cnt;
1600 	bool fast, rep;
1601 
1602 	/*
1603 	 * hypercall generates UD from non zero cpl and real mode
1604 	 * per HYPER-V spec
1605 	 */
1606 	if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
1607 		kvm_queue_exception(vcpu, UD_VECTOR);
1608 		return 1;
1609 	}
1610 
1611 #ifdef CONFIG_X86_64
1612 	if (is_64_bit_mode(vcpu)) {
1613 		param = kvm_rcx_read(vcpu);
1614 		ingpa = kvm_rdx_read(vcpu);
1615 		outgpa = kvm_r8_read(vcpu);
1616 	} else
1617 #endif
1618 	{
1619 		param = ((u64)kvm_rdx_read(vcpu) << 32) |
1620 			(kvm_rax_read(vcpu) & 0xffffffff);
1621 		ingpa = ((u64)kvm_rbx_read(vcpu) << 32) |
1622 			(kvm_rcx_read(vcpu) & 0xffffffff);
1623 		outgpa = ((u64)kvm_rdi_read(vcpu) << 32) |
1624 			(kvm_rsi_read(vcpu) & 0xffffffff);
1625 	}
1626 
1627 	code = param & 0xffff;
1628 	fast = !!(param & HV_HYPERCALL_FAST_BIT);
1629 	rep_cnt = (param >> HV_HYPERCALL_REP_COMP_OFFSET) & 0xfff;
1630 	rep_idx = (param >> HV_HYPERCALL_REP_START_OFFSET) & 0xfff;
1631 	rep = !!(rep_cnt || rep_idx);
1632 
1633 	trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
1634 
1635 	switch (code) {
1636 	case HVCALL_NOTIFY_LONG_SPIN_WAIT:
1637 		if (unlikely(rep)) {
1638 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1639 			break;
1640 		}
1641 		kvm_vcpu_on_spin(vcpu, true);
1642 		break;
1643 	case HVCALL_SIGNAL_EVENT:
1644 		if (unlikely(rep)) {
1645 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1646 			break;
1647 		}
1648 		ret = kvm_hvcall_signal_event(vcpu, fast, ingpa);
1649 		if (ret != HV_STATUS_INVALID_PORT_ID)
1650 			break;
1651 		/* fall through - maybe userspace knows this conn_id. */
1652 	case HVCALL_POST_MESSAGE:
1653 		/* don't bother userspace if it has no way to handle it */
1654 		if (unlikely(rep || !vcpu_to_synic(vcpu)->active)) {
1655 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1656 			break;
1657 		}
1658 		vcpu->run->exit_reason = KVM_EXIT_HYPERV;
1659 		vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
1660 		vcpu->run->hyperv.u.hcall.input = param;
1661 		vcpu->run->hyperv.u.hcall.params[0] = ingpa;
1662 		vcpu->run->hyperv.u.hcall.params[1] = outgpa;
1663 		vcpu->arch.complete_userspace_io =
1664 				kvm_hv_hypercall_complete_userspace;
1665 		return 0;
1666 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
1667 		if (unlikely(fast || !rep_cnt || rep_idx)) {
1668 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1669 			break;
1670 		}
1671 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false);
1672 		break;
1673 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
1674 		if (unlikely(fast || rep)) {
1675 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1676 			break;
1677 		}
1678 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false);
1679 		break;
1680 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
1681 		if (unlikely(fast || !rep_cnt || rep_idx)) {
1682 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1683 			break;
1684 		}
1685 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
1686 		break;
1687 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
1688 		if (unlikely(fast || rep)) {
1689 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1690 			break;
1691 		}
1692 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
1693 		break;
1694 	case HVCALL_SEND_IPI:
1695 		if (unlikely(rep)) {
1696 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1697 			break;
1698 		}
1699 		ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, false, fast);
1700 		break;
1701 	case HVCALL_SEND_IPI_EX:
1702 		if (unlikely(fast || rep)) {
1703 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1704 			break;
1705 		}
1706 		ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, true, false);
1707 		break;
1708 	default:
1709 		ret = HV_STATUS_INVALID_HYPERCALL_CODE;
1710 		break;
1711 	}
1712 
1713 	return kvm_hv_hypercall_complete(vcpu, ret);
1714 }
1715 
1716 void kvm_hv_init_vm(struct kvm *kvm)
1717 {
1718 	mutex_init(&kvm->arch.hyperv.hv_lock);
1719 	idr_init(&kvm->arch.hyperv.conn_to_evt);
1720 }
1721 
1722 void kvm_hv_destroy_vm(struct kvm *kvm)
1723 {
1724 	struct eventfd_ctx *eventfd;
1725 	int i;
1726 
1727 	idr_for_each_entry(&kvm->arch.hyperv.conn_to_evt, eventfd, i)
1728 		eventfd_ctx_put(eventfd);
1729 	idr_destroy(&kvm->arch.hyperv.conn_to_evt);
1730 }
1731 
1732 static int kvm_hv_eventfd_assign(struct kvm *kvm, u32 conn_id, int fd)
1733 {
1734 	struct kvm_hv *hv = &kvm->arch.hyperv;
1735 	struct eventfd_ctx *eventfd;
1736 	int ret;
1737 
1738 	eventfd = eventfd_ctx_fdget(fd);
1739 	if (IS_ERR(eventfd))
1740 		return PTR_ERR(eventfd);
1741 
1742 	mutex_lock(&hv->hv_lock);
1743 	ret = idr_alloc(&hv->conn_to_evt, eventfd, conn_id, conn_id + 1,
1744 			GFP_KERNEL_ACCOUNT);
1745 	mutex_unlock(&hv->hv_lock);
1746 
1747 	if (ret >= 0)
1748 		return 0;
1749 
1750 	if (ret == -ENOSPC)
1751 		ret = -EEXIST;
1752 	eventfd_ctx_put(eventfd);
1753 	return ret;
1754 }
1755 
1756 static int kvm_hv_eventfd_deassign(struct kvm *kvm, u32 conn_id)
1757 {
1758 	struct kvm_hv *hv = &kvm->arch.hyperv;
1759 	struct eventfd_ctx *eventfd;
1760 
1761 	mutex_lock(&hv->hv_lock);
1762 	eventfd = idr_remove(&hv->conn_to_evt, conn_id);
1763 	mutex_unlock(&hv->hv_lock);
1764 
1765 	if (!eventfd)
1766 		return -ENOENT;
1767 
1768 	synchronize_srcu(&kvm->srcu);
1769 	eventfd_ctx_put(eventfd);
1770 	return 0;
1771 }
1772 
1773 int kvm_vm_ioctl_hv_eventfd(struct kvm *kvm, struct kvm_hyperv_eventfd *args)
1774 {
1775 	if ((args->flags & ~KVM_HYPERV_EVENTFD_DEASSIGN) ||
1776 	    (args->conn_id & ~KVM_HYPERV_CONN_ID_MASK))
1777 		return -EINVAL;
1778 
1779 	if (args->flags == KVM_HYPERV_EVENTFD_DEASSIGN)
1780 		return kvm_hv_eventfd_deassign(kvm, args->conn_id);
1781 	return kvm_hv_eventfd_assign(kvm, args->conn_id, args->fd);
1782 }
1783 
1784 int kvm_vcpu_ioctl_get_hv_cpuid(struct kvm_vcpu *vcpu, struct kvm_cpuid2 *cpuid,
1785 				struct kvm_cpuid_entry2 __user *entries)
1786 {
1787 	uint16_t evmcs_ver = 0;
1788 	struct kvm_cpuid_entry2 cpuid_entries[] = {
1789 		{ .function = HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS },
1790 		{ .function = HYPERV_CPUID_INTERFACE },
1791 		{ .function = HYPERV_CPUID_VERSION },
1792 		{ .function = HYPERV_CPUID_FEATURES },
1793 		{ .function = HYPERV_CPUID_ENLIGHTMENT_INFO },
1794 		{ .function = HYPERV_CPUID_IMPLEMENT_LIMITS },
1795 		{ .function = HYPERV_CPUID_NESTED_FEATURES },
1796 	};
1797 	int i, nent = ARRAY_SIZE(cpuid_entries);
1798 
1799 	if (kvm_x86_ops->nested_get_evmcs_version)
1800 		evmcs_ver = kvm_x86_ops->nested_get_evmcs_version(vcpu);
1801 
1802 	/* Skip NESTED_FEATURES if eVMCS is not supported */
1803 	if (!evmcs_ver)
1804 		--nent;
1805 
1806 	if (cpuid->nent < nent)
1807 		return -E2BIG;
1808 
1809 	if (cpuid->nent > nent)
1810 		cpuid->nent = nent;
1811 
1812 	for (i = 0; i < nent; i++) {
1813 		struct kvm_cpuid_entry2 *ent = &cpuid_entries[i];
1814 		u32 signature[3];
1815 
1816 		switch (ent->function) {
1817 		case HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS:
1818 			memcpy(signature, "Linux KVM Hv", 12);
1819 
1820 			ent->eax = HYPERV_CPUID_NESTED_FEATURES;
1821 			ent->ebx = signature[0];
1822 			ent->ecx = signature[1];
1823 			ent->edx = signature[2];
1824 			break;
1825 
1826 		case HYPERV_CPUID_INTERFACE:
1827 			memcpy(signature, "Hv#1\0\0\0\0\0\0\0\0", 12);
1828 			ent->eax = signature[0];
1829 			break;
1830 
1831 		case HYPERV_CPUID_VERSION:
1832 			/*
1833 			 * We implement some Hyper-V 2016 functions so let's use
1834 			 * this version.
1835 			 */
1836 			ent->eax = 0x00003839;
1837 			ent->ebx = 0x000A0000;
1838 			break;
1839 
1840 		case HYPERV_CPUID_FEATURES:
1841 			ent->eax |= HV_X64_MSR_VP_RUNTIME_AVAILABLE;
1842 			ent->eax |= HV_MSR_TIME_REF_COUNT_AVAILABLE;
1843 			ent->eax |= HV_X64_MSR_SYNIC_AVAILABLE;
1844 			ent->eax |= HV_MSR_SYNTIMER_AVAILABLE;
1845 			ent->eax |= HV_X64_MSR_APIC_ACCESS_AVAILABLE;
1846 			ent->eax |= HV_X64_MSR_HYPERCALL_AVAILABLE;
1847 			ent->eax |= HV_X64_MSR_VP_INDEX_AVAILABLE;
1848 			ent->eax |= HV_X64_MSR_RESET_AVAILABLE;
1849 			ent->eax |= HV_MSR_REFERENCE_TSC_AVAILABLE;
1850 			ent->eax |= HV_X64_ACCESS_FREQUENCY_MSRS;
1851 			ent->eax |= HV_X64_ACCESS_REENLIGHTENMENT;
1852 
1853 			ent->ebx |= HV_X64_POST_MESSAGES;
1854 			ent->ebx |= HV_X64_SIGNAL_EVENTS;
1855 
1856 			ent->edx |= HV_FEATURE_FREQUENCY_MSRS_AVAILABLE;
1857 			ent->edx |= HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
1858 
1859 			/*
1860 			 * Direct Synthetic timers only make sense with in-kernel
1861 			 * LAPIC
1862 			 */
1863 			if (lapic_in_kernel(vcpu))
1864 				ent->edx |= HV_STIMER_DIRECT_MODE_AVAILABLE;
1865 
1866 			break;
1867 
1868 		case HYPERV_CPUID_ENLIGHTMENT_INFO:
1869 			ent->eax |= HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
1870 			ent->eax |= HV_X64_APIC_ACCESS_RECOMMENDED;
1871 			ent->eax |= HV_X64_RELAXED_TIMING_RECOMMENDED;
1872 			ent->eax |= HV_X64_CLUSTER_IPI_RECOMMENDED;
1873 			ent->eax |= HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED;
1874 			if (evmcs_ver)
1875 				ent->eax |= HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
1876 			if (!cpu_smt_possible())
1877 				ent->eax |= HV_X64_NO_NONARCH_CORESHARING;
1878 			/*
1879 			 * Default number of spinlock retry attempts, matches
1880 			 * HyperV 2016.
1881 			 */
1882 			ent->ebx = 0x00000FFF;
1883 
1884 			break;
1885 
1886 		case HYPERV_CPUID_IMPLEMENT_LIMITS:
1887 			/* Maximum number of virtual processors */
1888 			ent->eax = KVM_MAX_VCPUS;
1889 			/*
1890 			 * Maximum number of logical processors, matches
1891 			 * HyperV 2016.
1892 			 */
1893 			ent->ebx = 64;
1894 
1895 			break;
1896 
1897 		case HYPERV_CPUID_NESTED_FEATURES:
1898 			ent->eax = evmcs_ver;
1899 
1900 			break;
1901 
1902 		default:
1903 			break;
1904 		}
1905 	}
1906 
1907 	if (copy_to_user(entries, cpuid_entries,
1908 			 nent * sizeof(struct kvm_cpuid_entry2)))
1909 		return -EFAULT;
1910 
1911 	return 0;
1912 }
1913