xref: /linux/arch/x86/kvm/hyperv.c (revision 7d1f84f1b94e79cb95d974ab3ec54cb3a4d64f5b)
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 
22 #include "x86.h"
23 #include "lapic.h"
24 #include "ioapic.h"
25 #include "cpuid.h"
26 #include "hyperv.h"
27 #include "mmu.h"
28 #include "xen.h"
29 
30 #include <linux/cpu.h>
31 #include <linux/kvm_host.h>
32 #include <linux/highmem.h>
33 #include <linux/sched/cputime.h>
34 #include <linux/spinlock.h>
35 #include <linux/eventfd.h>
36 
37 #include <asm/apicdef.h>
38 #include <asm/mshyperv.h>
39 #include <trace/events/kvm.h>
40 
41 #include "trace.h"
42 #include "irq.h"
43 #include "fpu.h"
44 
45 #define KVM_HV_MAX_SPARSE_VCPU_SET_BITS DIV_ROUND_UP(KVM_MAX_VCPUS, HV_VCPUS_PER_SPARSE_BANK)
46 
47 /*
48  * As per Hyper-V TLFS, extended hypercalls start from 0x8001
49  * (HvExtCallQueryCapabilities). Response of this hypercalls is a 64 bit value
50  * where each bit tells which extended hypercall is available besides
51  * HvExtCallQueryCapabilities.
52  *
53  * 0x8001 - First extended hypercall, HvExtCallQueryCapabilities, no bit
54  * assigned.
55  *
56  * 0x8002 - Bit 0
57  * 0x8003 - Bit 1
58  * ..
59  * 0x8041 - Bit 63
60  *
61  * Therefore, HV_EXT_CALL_MAX = 0x8001 + 64
62  */
63 #define HV_EXT_CALL_MAX (HV_EXT_CALL_QUERY_CAPABILITIES + 64)
64 
65 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
66 				bool vcpu_kick);
67 
68 static inline u64 synic_read_sint(struct kvm_vcpu_hv_synic *synic, int sint)
69 {
70 	return atomic64_read(&synic->sint[sint]);
71 }
72 
73 static inline int synic_get_sint_vector(u64 sint_value)
74 {
75 	if (sint_value & HV_SYNIC_SINT_MASKED)
76 		return -1;
77 	return sint_value & HV_SYNIC_SINT_VECTOR_MASK;
78 }
79 
80 static bool synic_has_vector_connected(struct kvm_vcpu_hv_synic *synic,
81 				      int vector)
82 {
83 	int i;
84 
85 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
86 		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
87 			return true;
88 	}
89 	return false;
90 }
91 
92 static bool synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic *synic,
93 				     int vector)
94 {
95 	int i;
96 	u64 sint_value;
97 
98 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
99 		sint_value = synic_read_sint(synic, i);
100 		if (synic_get_sint_vector(sint_value) == vector &&
101 		    sint_value & HV_SYNIC_SINT_AUTO_EOI)
102 			return true;
103 	}
104 	return false;
105 }
106 
107 static void synic_update_vector(struct kvm_vcpu_hv_synic *synic,
108 				int vector)
109 {
110 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
111 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
112 	bool auto_eoi_old, auto_eoi_new;
113 
114 	if (vector < HV_SYNIC_FIRST_VALID_VECTOR)
115 		return;
116 
117 	if (synic_has_vector_connected(synic, vector))
118 		__set_bit(vector, synic->vec_bitmap);
119 	else
120 		__clear_bit(vector, synic->vec_bitmap);
121 
122 	auto_eoi_old = !bitmap_empty(synic->auto_eoi_bitmap, 256);
123 
124 	if (synic_has_vector_auto_eoi(synic, vector))
125 		__set_bit(vector, synic->auto_eoi_bitmap);
126 	else
127 		__clear_bit(vector, synic->auto_eoi_bitmap);
128 
129 	auto_eoi_new = !bitmap_empty(synic->auto_eoi_bitmap, 256);
130 
131 	if (auto_eoi_old == auto_eoi_new)
132 		return;
133 
134 	if (!enable_apicv)
135 		return;
136 
137 	down_write(&vcpu->kvm->arch.apicv_update_lock);
138 
139 	if (auto_eoi_new)
140 		hv->synic_auto_eoi_used++;
141 	else
142 		hv->synic_auto_eoi_used--;
143 
144 	/*
145 	 * Inhibit APICv if any vCPU is using SynIC's AutoEOI, which relies on
146 	 * the hypervisor to manually inject IRQs.
147 	 */
148 	__kvm_set_or_clear_apicv_inhibit(vcpu->kvm,
149 					 APICV_INHIBIT_REASON_HYPERV,
150 					 !!hv->synic_auto_eoi_used);
151 
152 	up_write(&vcpu->kvm->arch.apicv_update_lock);
153 }
154 
155 static int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint,
156 			  u64 data, bool host)
157 {
158 	int vector, old_vector;
159 	bool masked;
160 
161 	vector = data & HV_SYNIC_SINT_VECTOR_MASK;
162 	masked = data & HV_SYNIC_SINT_MASKED;
163 
164 	/*
165 	 * Valid vectors are 16-255, however, nested Hyper-V attempts to write
166 	 * default '0x10000' value on boot and this should not #GP. We need to
167 	 * allow zero-initing the register from host as well.
168 	 */
169 	if (vector < HV_SYNIC_FIRST_VALID_VECTOR && !host && !masked)
170 		return 1;
171 	/*
172 	 * Guest may configure multiple SINTs to use the same vector, so
173 	 * we maintain a bitmap of vectors handled by synic, and a
174 	 * bitmap of vectors with auto-eoi behavior.  The bitmaps are
175 	 * updated here, and atomically queried on fast paths.
176 	 */
177 	old_vector = synic_read_sint(synic, sint) & HV_SYNIC_SINT_VECTOR_MASK;
178 
179 	atomic64_set(&synic->sint[sint], data);
180 
181 	synic_update_vector(synic, old_vector);
182 
183 	synic_update_vector(synic, vector);
184 
185 	/* Load SynIC vectors into EOI exit bitmap */
186 	kvm_make_request(KVM_REQ_SCAN_IOAPIC, hv_synic_to_vcpu(synic));
187 	return 0;
188 }
189 
190 static struct kvm_vcpu *get_vcpu_by_vpidx(struct kvm *kvm, u32 vpidx)
191 {
192 	struct kvm_vcpu *vcpu = NULL;
193 	unsigned long i;
194 
195 	if (vpidx >= KVM_MAX_VCPUS)
196 		return NULL;
197 
198 	vcpu = kvm_get_vcpu(kvm, vpidx);
199 	if (vcpu && kvm_hv_get_vpindex(vcpu) == vpidx)
200 		return vcpu;
201 	kvm_for_each_vcpu(i, vcpu, kvm)
202 		if (kvm_hv_get_vpindex(vcpu) == vpidx)
203 			return vcpu;
204 	return NULL;
205 }
206 
207 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vpidx)
208 {
209 	struct kvm_vcpu_hv_synic *synic;
210 	struct kvm_vcpu_hv *hv_vcpu;
211 	struct kvm_vcpu *vcpu;
212 
213 	vcpu = get_vcpu_by_vpidx(kvm, vpidx);
214 	if (!vcpu)
215 		return NULL;
216 
217 	hv_vcpu = to_hv_vcpu_safe(vcpu);
218 	if (!hv_vcpu)
219 		return NULL;
220 
221 	synic = &hv_vcpu->synic;
222 	return READ_ONCE(synic->active) ? synic : NULL;
223 }
224 
225 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
226 {
227 	struct kvm *kvm = vcpu->kvm;
228 	struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
229 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
230 	struct kvm_vcpu_hv_stimer *stimer;
231 	int gsi, idx;
232 
233 	trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
234 
235 	/* Try to deliver pending Hyper-V SynIC timers messages */
236 	for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
237 		stimer = &hv_vcpu->stimer[idx];
238 		if (stimer->msg_pending && stimer->config.enable &&
239 		    !stimer->config.direct_mode &&
240 		    stimer->config.sintx == sint)
241 			stimer_mark_pending(stimer, false);
242 	}
243 
244 	idx = srcu_read_lock(&kvm->irq_srcu);
245 	gsi = atomic_read(&synic->sint_to_gsi[sint]);
246 	if (gsi != -1)
247 		kvm_notify_acked_gsi(kvm, gsi);
248 	srcu_read_unlock(&kvm->irq_srcu, idx);
249 }
250 
251 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
252 {
253 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
254 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
255 
256 	hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
257 	hv_vcpu->exit.u.synic.msr = msr;
258 	hv_vcpu->exit.u.synic.control = synic->control;
259 	hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
260 	hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
261 
262 	kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
263 }
264 
265 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
266 			 u32 msr, u64 data, bool host)
267 {
268 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
269 	int ret;
270 
271 	if (!synic->active && (!host || data))
272 		return 1;
273 
274 	trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
275 
276 	ret = 0;
277 	switch (msr) {
278 	case HV_X64_MSR_SCONTROL:
279 		synic->control = data;
280 		if (!host)
281 			synic_exit(synic, msr);
282 		break;
283 	case HV_X64_MSR_SVERSION:
284 		if (!host) {
285 			ret = 1;
286 			break;
287 		}
288 		synic->version = data;
289 		break;
290 	case HV_X64_MSR_SIEFP:
291 		if ((data & HV_SYNIC_SIEFP_ENABLE) && !host &&
292 		    !synic->dont_zero_synic_pages)
293 			if (kvm_clear_guest(vcpu->kvm,
294 					    data & PAGE_MASK, PAGE_SIZE)) {
295 				ret = 1;
296 				break;
297 			}
298 		synic->evt_page = data;
299 		if (!host)
300 			synic_exit(synic, msr);
301 		break;
302 	case HV_X64_MSR_SIMP:
303 		if ((data & HV_SYNIC_SIMP_ENABLE) && !host &&
304 		    !synic->dont_zero_synic_pages)
305 			if (kvm_clear_guest(vcpu->kvm,
306 					    data & PAGE_MASK, PAGE_SIZE)) {
307 				ret = 1;
308 				break;
309 			}
310 		synic->msg_page = data;
311 		if (!host)
312 			synic_exit(synic, msr);
313 		break;
314 	case HV_X64_MSR_EOM: {
315 		int i;
316 
317 		if (!synic->active)
318 			break;
319 
320 		for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
321 			kvm_hv_notify_acked_sint(vcpu, i);
322 		break;
323 	}
324 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
325 		ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
326 		break;
327 	default:
328 		ret = 1;
329 		break;
330 	}
331 	return ret;
332 }
333 
334 static bool kvm_hv_is_syndbg_enabled(struct kvm_vcpu *vcpu)
335 {
336 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
337 
338 	return hv_vcpu->cpuid_cache.syndbg_cap_eax &
339 		HV_X64_SYNDBG_CAP_ALLOW_KERNEL_DEBUGGING;
340 }
341 
342 static int kvm_hv_syndbg_complete_userspace(struct kvm_vcpu *vcpu)
343 {
344 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
345 
346 	if (vcpu->run->hyperv.u.syndbg.msr == HV_X64_MSR_SYNDBG_CONTROL)
347 		hv->hv_syndbg.control.status =
348 			vcpu->run->hyperv.u.syndbg.status;
349 	return 1;
350 }
351 
352 static void syndbg_exit(struct kvm_vcpu *vcpu, u32 msr)
353 {
354 	struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
355 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
356 
357 	hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNDBG;
358 	hv_vcpu->exit.u.syndbg.msr = msr;
359 	hv_vcpu->exit.u.syndbg.control = syndbg->control.control;
360 	hv_vcpu->exit.u.syndbg.send_page = syndbg->control.send_page;
361 	hv_vcpu->exit.u.syndbg.recv_page = syndbg->control.recv_page;
362 	hv_vcpu->exit.u.syndbg.pending_page = syndbg->control.pending_page;
363 	vcpu->arch.complete_userspace_io =
364 			kvm_hv_syndbg_complete_userspace;
365 
366 	kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
367 }
368 
369 static int syndbg_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
370 {
371 	struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
372 
373 	if (!kvm_hv_is_syndbg_enabled(vcpu) && !host)
374 		return 1;
375 
376 	trace_kvm_hv_syndbg_set_msr(vcpu->vcpu_id,
377 				    to_hv_vcpu(vcpu)->vp_index, msr, data);
378 	switch (msr) {
379 	case HV_X64_MSR_SYNDBG_CONTROL:
380 		syndbg->control.control = data;
381 		if (!host)
382 			syndbg_exit(vcpu, msr);
383 		break;
384 	case HV_X64_MSR_SYNDBG_STATUS:
385 		syndbg->control.status = data;
386 		break;
387 	case HV_X64_MSR_SYNDBG_SEND_BUFFER:
388 		syndbg->control.send_page = data;
389 		break;
390 	case HV_X64_MSR_SYNDBG_RECV_BUFFER:
391 		syndbg->control.recv_page = data;
392 		break;
393 	case HV_X64_MSR_SYNDBG_PENDING_BUFFER:
394 		syndbg->control.pending_page = data;
395 		if (!host)
396 			syndbg_exit(vcpu, msr);
397 		break;
398 	case HV_X64_MSR_SYNDBG_OPTIONS:
399 		syndbg->options = data;
400 		break;
401 	default:
402 		break;
403 	}
404 
405 	return 0;
406 }
407 
408 static int syndbg_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
409 {
410 	struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
411 
412 	if (!kvm_hv_is_syndbg_enabled(vcpu) && !host)
413 		return 1;
414 
415 	switch (msr) {
416 	case HV_X64_MSR_SYNDBG_CONTROL:
417 		*pdata = syndbg->control.control;
418 		break;
419 	case HV_X64_MSR_SYNDBG_STATUS:
420 		*pdata = syndbg->control.status;
421 		break;
422 	case HV_X64_MSR_SYNDBG_SEND_BUFFER:
423 		*pdata = syndbg->control.send_page;
424 		break;
425 	case HV_X64_MSR_SYNDBG_RECV_BUFFER:
426 		*pdata = syndbg->control.recv_page;
427 		break;
428 	case HV_X64_MSR_SYNDBG_PENDING_BUFFER:
429 		*pdata = syndbg->control.pending_page;
430 		break;
431 	case HV_X64_MSR_SYNDBG_OPTIONS:
432 		*pdata = syndbg->options;
433 		break;
434 	default:
435 		break;
436 	}
437 
438 	trace_kvm_hv_syndbg_get_msr(vcpu->vcpu_id, kvm_hv_get_vpindex(vcpu), msr, *pdata);
439 
440 	return 0;
441 }
442 
443 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata,
444 			 bool host)
445 {
446 	int ret;
447 
448 	if (!synic->active && !host)
449 		return 1;
450 
451 	ret = 0;
452 	switch (msr) {
453 	case HV_X64_MSR_SCONTROL:
454 		*pdata = synic->control;
455 		break;
456 	case HV_X64_MSR_SVERSION:
457 		*pdata = synic->version;
458 		break;
459 	case HV_X64_MSR_SIEFP:
460 		*pdata = synic->evt_page;
461 		break;
462 	case HV_X64_MSR_SIMP:
463 		*pdata = synic->msg_page;
464 		break;
465 	case HV_X64_MSR_EOM:
466 		*pdata = 0;
467 		break;
468 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
469 		*pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
470 		break;
471 	default:
472 		ret = 1;
473 		break;
474 	}
475 	return ret;
476 }
477 
478 static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
479 {
480 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
481 	struct kvm_lapic_irq irq;
482 	int ret, vector;
483 
484 	if (KVM_BUG_ON(!lapic_in_kernel(vcpu), vcpu->kvm))
485 		return -EINVAL;
486 
487 	if (sint >= ARRAY_SIZE(synic->sint))
488 		return -EINVAL;
489 
490 	vector = synic_get_sint_vector(synic_read_sint(synic, sint));
491 	if (vector < 0)
492 		return -ENOENT;
493 
494 	memset(&irq, 0, sizeof(irq));
495 	irq.shorthand = APIC_DEST_SELF;
496 	irq.dest_mode = APIC_DEST_PHYSICAL;
497 	irq.delivery_mode = APIC_DM_FIXED;
498 	irq.vector = vector;
499 	irq.level = 1;
500 
501 	ret = kvm_irq_delivery_to_apic(vcpu->kvm, vcpu->arch.apic, &irq);
502 	trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
503 	return ret;
504 }
505 
506 int kvm_hv_synic_set_irq(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm,
507 			 int irq_source_id, int level, bool line_status)
508 {
509 	struct kvm_vcpu_hv_synic *synic;
510 
511 	if (!level)
512 		return -1;
513 
514 	synic = synic_get(kvm, e->hv_sint.vcpu);
515 	if (!synic)
516 		return -EINVAL;
517 
518 	return synic_set_irq(synic, e->hv_sint.sint);
519 }
520 
521 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
522 {
523 	struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
524 	int i;
525 
526 	trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
527 
528 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
529 		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
530 			kvm_hv_notify_acked_sint(vcpu, i);
531 }
532 
533 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi)
534 {
535 	struct kvm_vcpu_hv_synic *synic;
536 
537 	synic = synic_get(kvm, vpidx);
538 	if (!synic)
539 		return -EINVAL;
540 
541 	if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
542 		return -EINVAL;
543 
544 	atomic_set(&synic->sint_to_gsi[sint], gsi);
545 	return 0;
546 }
547 
548 void kvm_hv_irq_routing_update(struct kvm *kvm)
549 {
550 	struct kvm_irq_routing_table *irq_rt;
551 	struct kvm_kernel_irq_routing_entry *e;
552 	u32 gsi;
553 
554 	irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
555 					lockdep_is_held(&kvm->irq_lock));
556 
557 	for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
558 		hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
559 			if (e->type == KVM_IRQ_ROUTING_HV_SINT)
560 				kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
561 						    e->hv_sint.sint, gsi);
562 		}
563 	}
564 }
565 
566 static void synic_init(struct kvm_vcpu_hv_synic *synic)
567 {
568 	int i;
569 
570 	memset(synic, 0, sizeof(*synic));
571 	synic->version = HV_SYNIC_VERSION_1;
572 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
573 		atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
574 		atomic_set(&synic->sint_to_gsi[i], -1);
575 	}
576 }
577 
578 static u64 get_time_ref_counter(struct kvm *kvm)
579 {
580 	struct kvm_hv *hv = to_kvm_hv(kvm);
581 	struct kvm_vcpu *vcpu;
582 	u64 tsc;
583 
584 	/*
585 	 * Fall back to get_kvmclock_ns() when TSC page hasn't been set up,
586 	 * is broken, disabled or being updated.
587 	 */
588 	if (hv->hv_tsc_page_status != HV_TSC_PAGE_SET)
589 		return div_u64(get_kvmclock_ns(kvm), 100);
590 
591 	vcpu = kvm_get_vcpu(kvm, 0);
592 	tsc = kvm_read_l1_tsc(vcpu, rdtsc());
593 	return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64)
594 		+ hv->tsc_ref.tsc_offset;
595 }
596 
597 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
598 				bool vcpu_kick)
599 {
600 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
601 
602 	set_bit(stimer->index, vcpu->arch.hyperv->stimer_pending_bitmap);
603 	kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
604 	if (vcpu_kick)
605 		kvm_vcpu_kick(vcpu);
606 }
607 
608 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
609 {
610 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
611 
612 	trace_kvm_hv_stimer_cleanup(hv_stimer_to_vcpu(stimer)->vcpu_id,
613 				    stimer->index);
614 
615 	hrtimer_cancel(&stimer->timer);
616 	clear_bit(stimer->index, vcpu->arch.hyperv->stimer_pending_bitmap);
617 	stimer->msg_pending = false;
618 	stimer->exp_time = 0;
619 }
620 
621 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
622 {
623 	struct kvm_vcpu_hv_stimer *stimer;
624 
625 	stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
626 	trace_kvm_hv_stimer_callback(hv_stimer_to_vcpu(stimer)->vcpu_id,
627 				     stimer->index);
628 	stimer_mark_pending(stimer, true);
629 
630 	return HRTIMER_NORESTART;
631 }
632 
633 /*
634  * Translate a stimer expiry given in 100ns reference ticks into an
635  * an absolute deadline. Saturates on overflow.
636  */
637 static ktime_t stimer_add_delta(ktime_t now, u64 delta_100ns)
638 {
639 	if (delta_100ns >= KTIME_MAX / 100)
640 		return KTIME_MAX;
641 
642 	return ktime_add_safe(now, 100 * delta_100ns);
643 }
644 
645 /*
646  * stimer_start() assumptions:
647  * a) stimer->count is not equal to 0
648  * b) stimer->config has HV_STIMER_ENABLE flag
649  */
650 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
651 {
652 	u64 time_now;
653 	ktime_t ktime_now;
654 	ktime_t deadline;
655 
656 	time_now = get_time_ref_counter(hv_stimer_to_vcpu(stimer)->kvm);
657 	ktime_now = ktime_get();
658 
659 	if (stimer->config.periodic) {
660 		if (stimer->exp_time) {
661 			if (time_now >= stimer->exp_time) {
662 				u64 remainder;
663 
664 				div64_u64_rem(time_now - stimer->exp_time,
665 					      stimer->count, &remainder);
666 				stimer->exp_time =
667 					time_now + (stimer->count - remainder);
668 			}
669 		} else
670 			stimer->exp_time = time_now + stimer->count;
671 
672 		trace_kvm_hv_stimer_start_periodic(
673 					hv_stimer_to_vcpu(stimer)->vcpu_id,
674 					stimer->index,
675 					time_now, stimer->exp_time);
676 
677 		deadline = stimer_add_delta(ktime_now, stimer->exp_time - time_now);
678 		hrtimer_start(&stimer->timer, deadline, HRTIMER_MODE_ABS);
679 		return 0;
680 	}
681 	stimer->exp_time = stimer->count;
682 	if (time_now >= stimer->count) {
683 		/*
684 		 * Expire timer according to Hypervisor Top-Level Functional
685 		 * specification v4(15.3.1):
686 		 * "If a one shot is enabled and the specified count is in
687 		 * the past, it will expire immediately."
688 		 */
689 		stimer_mark_pending(stimer, false);
690 		return 0;
691 	}
692 
693 	trace_kvm_hv_stimer_start_one_shot(hv_stimer_to_vcpu(stimer)->vcpu_id,
694 					   stimer->index,
695 					   time_now, stimer->count);
696 
697 	deadline = stimer_add_delta(ktime_now, stimer->count - time_now);
698 	hrtimer_start(&stimer->timer, deadline, HRTIMER_MODE_ABS);
699 
700 	return 0;
701 }
702 
703 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
704 			     bool host)
705 {
706 	union hv_stimer_config new_config = {.as_uint64 = config},
707 		old_config = {.as_uint64 = stimer->config.as_uint64};
708 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
709 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
710 	struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
711 
712 	if (!synic->active && (!host || config))
713 		return 1;
714 
715 	if (unlikely(!host && hv_vcpu->enforce_cpuid && new_config.direct_mode &&
716 		     !(hv_vcpu->cpuid_cache.features_edx &
717 		       HV_STIMER_DIRECT_MODE_AVAILABLE)))
718 		return 1;
719 
720 	trace_kvm_hv_stimer_set_config(hv_stimer_to_vcpu(stimer)->vcpu_id,
721 				       stimer->index, config, host);
722 
723 	stimer_cleanup(stimer);
724 	if (old_config.enable &&
725 	    !new_config.direct_mode && new_config.sintx == 0)
726 		new_config.enable = 0;
727 	stimer->config.as_uint64 = new_config.as_uint64;
728 
729 	if (stimer->config.enable)
730 		stimer_mark_pending(stimer, false);
731 
732 	return 0;
733 }
734 
735 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
736 			    bool host)
737 {
738 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
739 	struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
740 
741 	if (!synic->active && (!host || count))
742 		return 1;
743 
744 	trace_kvm_hv_stimer_set_count(hv_stimer_to_vcpu(stimer)->vcpu_id,
745 				      stimer->index, count, host);
746 
747 	stimer_cleanup(stimer);
748 	stimer->count = count;
749 	if (!host) {
750 		if (stimer->count == 0)
751 			stimer->config.enable = 0;
752 		else if (stimer->config.auto_enable)
753 			stimer->config.enable = 1;
754 	}
755 
756 	if (stimer->config.enable)
757 		stimer_mark_pending(stimer, false);
758 
759 	return 0;
760 }
761 
762 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
763 {
764 	*pconfig = stimer->config.as_uint64;
765 	return 0;
766 }
767 
768 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
769 {
770 	*pcount = stimer->count;
771 	return 0;
772 }
773 
774 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
775 			     struct hv_message *src_msg, bool no_retry)
776 {
777 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
778 	int msg_off = offsetof(struct hv_message_page, sint_message[sint]);
779 	gfn_t msg_page_gfn;
780 	struct hv_message_header hv_hdr;
781 	int r;
782 
783 	if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
784 		return -ENOENT;
785 
786 	msg_page_gfn = synic->msg_page >> PAGE_SHIFT;
787 
788 	/*
789 	 * Strictly following the spec-mandated ordering would assume setting
790 	 * .msg_pending before checking .message_type.  However, this function
791 	 * is only called in vcpu context so the entire update is atomic from
792 	 * guest POV and thus the exact order here doesn't matter.
793 	 */
794 	r = kvm_vcpu_read_guest_page(vcpu, msg_page_gfn, &hv_hdr.message_type,
795 				     msg_off + offsetof(struct hv_message,
796 							header.message_type),
797 				     sizeof(hv_hdr.message_type));
798 	if (r < 0)
799 		return r;
800 
801 	if (hv_hdr.message_type != HVMSG_NONE) {
802 		if (no_retry)
803 			return 0;
804 
805 		hv_hdr.message_flags.asu8 = 0;
806 		hv_hdr.message_flags.msg_pending = 1;
807 
808 		r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn,
809 					      &hv_hdr.message_flags,
810 					      msg_off +
811 					      offsetof(struct hv_message,
812 						       header.message_flags),
813 					      sizeof(hv_hdr.message_flags));
814 		if (r < 0)
815 			return r;
816 		return -EAGAIN;
817 	}
818 
819 	r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn, src_msg, msg_off,
820 				      sizeof(src_msg->header) +
821 				      src_msg->header.payload_size);
822 	if (r < 0)
823 		return r;
824 
825 	r = synic_set_irq(synic, sint);
826 	if (r < 0)
827 		return r;
828 	if (r == 0)
829 		return -EFAULT;
830 	return 0;
831 }
832 
833 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
834 {
835 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
836 	struct hv_message *msg = &stimer->msg;
837 	struct hv_timer_message_payload *payload =
838 			(struct hv_timer_message_payload *)&msg->u.payload;
839 
840 	/*
841 	 * To avoid piling up periodic ticks, don't retry message
842 	 * delivery for them (within "lazy" lost ticks policy).
843 	 */
844 	bool no_retry = stimer->config.periodic;
845 
846 	payload->expiration_time = stimer->exp_time;
847 	payload->delivery_time = get_time_ref_counter(vcpu->kvm);
848 	return synic_deliver_msg(to_hv_synic(vcpu),
849 				 stimer->config.sintx, msg,
850 				 no_retry);
851 }
852 
853 static int stimer_notify_direct(struct kvm_vcpu_hv_stimer *stimer)
854 {
855 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
856 	struct kvm_lapic_irq irq = {
857 		.delivery_mode = APIC_DM_FIXED,
858 		.vector = stimer->config.apic_vector
859 	};
860 
861 	if (lapic_in_kernel(vcpu))
862 		return !kvm_apic_set_irq(vcpu, &irq, NULL);
863 	return 0;
864 }
865 
866 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
867 {
868 	int r, direct = stimer->config.direct_mode;
869 
870 	stimer->msg_pending = true;
871 	if (!direct)
872 		r = stimer_send_msg(stimer);
873 	else
874 		r = stimer_notify_direct(stimer);
875 	trace_kvm_hv_stimer_expiration(hv_stimer_to_vcpu(stimer)->vcpu_id,
876 				       stimer->index, direct, r);
877 	if (!r) {
878 		stimer->msg_pending = false;
879 		if (!(stimer->config.periodic))
880 			stimer->config.enable = 0;
881 	}
882 }
883 
884 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
885 {
886 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
887 	struct kvm_vcpu_hv_stimer *stimer;
888 	u64 time_now, exp_time;
889 	int i;
890 
891 	if (!hv_vcpu)
892 		return;
893 
894 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
895 		if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
896 			stimer = &hv_vcpu->stimer[i];
897 			if (stimer->config.enable) {
898 				exp_time = stimer->exp_time;
899 
900 				if (exp_time) {
901 					time_now =
902 						get_time_ref_counter(vcpu->kvm);
903 					if (time_now >= exp_time)
904 						stimer_expiration(stimer);
905 				}
906 
907 				if ((stimer->config.enable) &&
908 				    stimer->count) {
909 					if (!stimer->msg_pending)
910 						stimer_start(stimer);
911 				} else
912 					stimer_cleanup(stimer);
913 			}
914 		}
915 }
916 
917 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
918 {
919 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
920 	int i;
921 
922 	if (!hv_vcpu)
923 		return;
924 
925 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
926 		stimer_cleanup(&hv_vcpu->stimer[i]);
927 
928 	kfree(hv_vcpu);
929 	vcpu->arch.hyperv = NULL;
930 }
931 
932 bool kvm_hv_assist_page_enabled(struct kvm_vcpu *vcpu)
933 {
934 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
935 
936 	if (!hv_vcpu)
937 		return false;
938 
939 	if (!(hv_vcpu->hv_vapic & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE))
940 		return false;
941 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
942 }
943 EXPORT_SYMBOL_FOR_KVM_INTERNAL(kvm_hv_assist_page_enabled);
944 
945 int kvm_hv_get_assist_page(struct kvm_vcpu *vcpu)
946 {
947 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
948 
949 	if (!hv_vcpu || !kvm_hv_assist_page_enabled(vcpu))
950 		return -EFAULT;
951 
952 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data,
953 				     &hv_vcpu->vp_assist_page, sizeof(struct hv_vp_assist_page));
954 }
955 EXPORT_SYMBOL_FOR_KVM_INTERNAL(kvm_hv_get_assist_page);
956 
957 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
958 {
959 	struct hv_message *msg = &stimer->msg;
960 	struct hv_timer_message_payload *payload =
961 			(struct hv_timer_message_payload *)&msg->u.payload;
962 
963 	memset(&msg->header, 0, sizeof(msg->header));
964 	msg->header.message_type = HVMSG_TIMER_EXPIRED;
965 	msg->header.payload_size = sizeof(*payload);
966 
967 	payload->timer_index = stimer->index;
968 	payload->expiration_time = 0;
969 	payload->delivery_time = 0;
970 }
971 
972 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
973 {
974 	memset(stimer, 0, sizeof(*stimer));
975 	stimer->index = timer_index;
976 	hrtimer_setup(&stimer->timer, stimer_timer_callback, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
977 	stimer_prepare_msg(stimer);
978 }
979 
980 int kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
981 {
982 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
983 	int i;
984 
985 	if (hv_vcpu)
986 		return 0;
987 
988 	hv_vcpu = kzalloc_obj(struct kvm_vcpu_hv, GFP_KERNEL_ACCOUNT);
989 	if (!hv_vcpu)
990 		return -ENOMEM;
991 
992 	hv_vcpu->vcpu = vcpu;
993 
994 	synic_init(&hv_vcpu->synic);
995 
996 	bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
997 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
998 		stimer_init(&hv_vcpu->stimer[i], i);
999 
1000 	hv_vcpu->vp_index = vcpu->vcpu_idx;
1001 
1002 	for (i = 0; i < HV_NR_TLB_FLUSH_FIFOS; i++) {
1003 		INIT_KFIFO(hv_vcpu->tlb_flush_fifo[i].entries);
1004 		spin_lock_init(&hv_vcpu->tlb_flush_fifo[i].write_lock);
1005 	}
1006 
1007 	/*
1008 	 * Ensure the structure is fully initialized before it's visible to
1009 	 * other tasks, as much of the state can be legally accessed without
1010 	 * holding vcpu->mutex.
1011 	 *
1012 	 * Pairs with the smp_load_acquire() in to_hv_vcpu_safe().
1013 	 */
1014 	smp_store_release(&vcpu->arch.hyperv, hv_vcpu);
1015 	return 0;
1016 }
1017 
1018 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
1019 {
1020 	struct kvm_vcpu_hv_synic *synic;
1021 	int r;
1022 
1023 	r = kvm_hv_vcpu_init(vcpu);
1024 	if (r)
1025 		return r;
1026 
1027 	synic = to_hv_synic(vcpu);
1028 
1029 	WRITE_ONCE(synic->active, true);
1030 	synic->dont_zero_synic_pages = dont_zero_synic_pages;
1031 	synic->control = HV_SYNIC_CONTROL_ENABLE;
1032 	return 0;
1033 }
1034 
1035 static bool kvm_hv_msr_partition_wide(u32 msr)
1036 {
1037 	bool r = false;
1038 
1039 	switch (msr) {
1040 	case HV_X64_MSR_GUEST_OS_ID:
1041 	case HV_X64_MSR_HYPERCALL:
1042 	case HV_X64_MSR_REFERENCE_TSC:
1043 	case HV_X64_MSR_TIME_REF_COUNT:
1044 	case HV_X64_MSR_CRASH_CTL:
1045 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1046 	case HV_X64_MSR_RESET:
1047 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1048 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1049 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1050 	case HV_X64_MSR_TSC_INVARIANT_CONTROL:
1051 	case HV_X64_MSR_SYNDBG_OPTIONS:
1052 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1053 		r = true;
1054 		break;
1055 	}
1056 
1057 	return r;
1058 }
1059 
1060 static int kvm_hv_msr_get_crash_data(struct kvm *kvm, u32 index, u64 *pdata)
1061 {
1062 	struct kvm_hv *hv = to_kvm_hv(kvm);
1063 	size_t size = ARRAY_SIZE(hv->hv_crash_param);
1064 
1065 	if (WARN_ON_ONCE(index >= size))
1066 		return -EINVAL;
1067 
1068 	*pdata = hv->hv_crash_param[array_index_nospec(index, size)];
1069 	return 0;
1070 }
1071 
1072 static int kvm_hv_msr_get_crash_ctl(struct kvm *kvm, u64 *pdata)
1073 {
1074 	struct kvm_hv *hv = to_kvm_hv(kvm);
1075 
1076 	*pdata = hv->hv_crash_ctl;
1077 	return 0;
1078 }
1079 
1080 static int kvm_hv_msr_set_crash_ctl(struct kvm *kvm, u64 data)
1081 {
1082 	struct kvm_hv *hv = to_kvm_hv(kvm);
1083 
1084 	hv->hv_crash_ctl = data & HV_CRASH_CTL_CRASH_NOTIFY;
1085 
1086 	return 0;
1087 }
1088 
1089 static int kvm_hv_msr_set_crash_data(struct kvm *kvm, u32 index, u64 data)
1090 {
1091 	struct kvm_hv *hv = to_kvm_hv(kvm);
1092 	size_t size = ARRAY_SIZE(hv->hv_crash_param);
1093 
1094 	if (WARN_ON_ONCE(index >= size))
1095 		return -EINVAL;
1096 
1097 	hv->hv_crash_param[array_index_nospec(index, size)] = data;
1098 	return 0;
1099 }
1100 
1101 /*
1102  * The kvmclock and Hyper-V TSC page use similar formulas, and converting
1103  * between them is possible:
1104  *
1105  * kvmclock formula:
1106  *    nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
1107  *           + system_time
1108  *
1109  * Hyper-V formula:
1110  *    nsec/100 = ticks * scale / 2^64 + offset
1111  *
1112  * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
1113  * By dividing the kvmclock formula by 100 and equating what's left we get:
1114  *    ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1115  *            scale / 2^64 =         tsc_to_system_mul * 2^(tsc_shift-32) / 100
1116  *            scale        =         tsc_to_system_mul * 2^(32+tsc_shift) / 100
1117  *
1118  * Now expand the kvmclock formula and divide by 100:
1119  *    nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
1120  *           - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
1121  *           + system_time
1122  *    nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1123  *               - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1124  *               + system_time / 100
1125  *
1126  * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
1127  *    nsec/100 = ticks * scale / 2^64
1128  *               - tsc_timestamp * scale / 2^64
1129  *               + system_time / 100
1130  *
1131  * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
1132  *    offset = system_time / 100 - tsc_timestamp * scale / 2^64
1133  *
1134  * These two equivalencies are implemented in this function.
1135  */
1136 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
1137 					struct ms_hyperv_tsc_page *tsc_ref)
1138 {
1139 	u64 max_mul;
1140 
1141 	if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
1142 		return false;
1143 
1144 	/*
1145 	 * check if scale would overflow, if so we use the time ref counter
1146 	 *    tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
1147 	 *    tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
1148 	 *    tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
1149 	 */
1150 	max_mul = 100ull << (32 - hv_clock->tsc_shift);
1151 	if (hv_clock->tsc_to_system_mul >= max_mul)
1152 		return false;
1153 
1154 	/*
1155 	 * Otherwise compute the scale and offset according to the formulas
1156 	 * derived above.
1157 	 */
1158 	tsc_ref->tsc_scale =
1159 		mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
1160 				hv_clock->tsc_to_system_mul,
1161 				100);
1162 
1163 	tsc_ref->tsc_offset = hv_clock->system_time;
1164 	do_div(tsc_ref->tsc_offset, 100);
1165 	tsc_ref->tsc_offset -=
1166 		mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
1167 	return true;
1168 }
1169 
1170 /*
1171  * Don't touch TSC page values if the guest has opted for TSC emulation after
1172  * migration. KVM doesn't fully support reenlightenment notifications and TSC
1173  * access emulation and Hyper-V is known to expect the values in TSC page to
1174  * stay constant before TSC access emulation is disabled from guest side
1175  * (HV_X64_MSR_TSC_EMULATION_STATUS). KVM userspace is expected to preserve TSC
1176  * frequency and guest visible TSC value across migration (and prevent it when
1177  * TSC scaling is unsupported).
1178  */
1179 static inline bool tsc_page_update_unsafe(struct kvm_hv *hv)
1180 {
1181 	return (hv->hv_tsc_page_status != HV_TSC_PAGE_GUEST_CHANGED) &&
1182 		hv->hv_tsc_emulation_control;
1183 }
1184 
1185 void kvm_hv_setup_tsc_page(struct kvm *kvm,
1186 			   struct pvclock_vcpu_time_info *hv_clock)
1187 {
1188 	struct kvm_hv *hv = to_kvm_hv(kvm);
1189 	u32 tsc_seq;
1190 	u64 gfn;
1191 
1192 	BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
1193 	BUILD_BUG_ON(offsetof(struct ms_hyperv_tsc_page, tsc_sequence) != 0);
1194 
1195 	guard(mutex)(&hv->hv_lock);
1196 
1197 	if (hv->hv_tsc_page_status == HV_TSC_PAGE_BROKEN ||
1198 	    hv->hv_tsc_page_status == HV_TSC_PAGE_SET ||
1199 	    hv->hv_tsc_page_status == HV_TSC_PAGE_UNSET)
1200 		return;
1201 
1202 	if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1203 		return;
1204 
1205 	gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
1206 	/*
1207 	 * Because the TSC parameters only vary when there is a
1208 	 * change in the master clock, do not bother with caching.
1209 	 */
1210 	if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
1211 				    &tsc_seq, sizeof(tsc_seq))))
1212 		goto out_err;
1213 
1214 	if (tsc_seq && tsc_page_update_unsafe(hv)) {
1215 		if (kvm_read_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
1216 			goto out_err;
1217 
1218 		hv->hv_tsc_page_status = HV_TSC_PAGE_SET;
1219 		return;
1220 	}
1221 
1222 	/*
1223 	 * While we're computing and writing the parameters, force the
1224 	 * guest to use the time reference count MSR.
1225 	 */
1226 	hv->tsc_ref.tsc_sequence = 0;
1227 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
1228 			    &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
1229 		goto out_err;
1230 
1231 	if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
1232 		goto out_err;
1233 
1234 	/* Ensure sequence is zero before writing the rest of the struct.  */
1235 	smp_wmb();
1236 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
1237 		goto out_err;
1238 
1239 	/*
1240 	 * Now switch to the TSC page mechanism by writing the sequence.
1241 	 */
1242 	tsc_seq++;
1243 	if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
1244 		tsc_seq = 1;
1245 
1246 	/* Write the struct entirely before the non-zero sequence.  */
1247 	smp_wmb();
1248 
1249 	hv->tsc_ref.tsc_sequence = tsc_seq;
1250 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
1251 			    &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
1252 		goto out_err;
1253 
1254 	hv->hv_tsc_page_status = HV_TSC_PAGE_SET;
1255 	return;
1256 
1257 out_err:
1258 	hv->hv_tsc_page_status = HV_TSC_PAGE_BROKEN;
1259 }
1260 
1261 void kvm_hv_request_tsc_page_update(struct kvm *kvm)
1262 {
1263 	struct kvm_hv *hv = to_kvm_hv(kvm);
1264 
1265 	mutex_lock(&hv->hv_lock);
1266 
1267 	if (hv->hv_tsc_page_status == HV_TSC_PAGE_SET &&
1268 	    !tsc_page_update_unsafe(hv))
1269 		hv->hv_tsc_page_status = HV_TSC_PAGE_HOST_CHANGED;
1270 
1271 	mutex_unlock(&hv->hv_lock);
1272 }
1273 
1274 static bool hv_check_msr_access(struct kvm_vcpu_hv *hv_vcpu, u32 msr)
1275 {
1276 	if (!hv_vcpu->enforce_cpuid)
1277 		return true;
1278 
1279 	switch (msr) {
1280 	case HV_X64_MSR_GUEST_OS_ID:
1281 	case HV_X64_MSR_HYPERCALL:
1282 		return hv_vcpu->cpuid_cache.features_eax &
1283 			HV_MSR_HYPERCALL_AVAILABLE;
1284 	case HV_X64_MSR_VP_RUNTIME:
1285 		return hv_vcpu->cpuid_cache.features_eax &
1286 			HV_MSR_VP_RUNTIME_AVAILABLE;
1287 	case HV_X64_MSR_TIME_REF_COUNT:
1288 		return hv_vcpu->cpuid_cache.features_eax &
1289 			HV_MSR_TIME_REF_COUNT_AVAILABLE;
1290 	case HV_X64_MSR_VP_INDEX:
1291 		return hv_vcpu->cpuid_cache.features_eax &
1292 			HV_MSR_VP_INDEX_AVAILABLE;
1293 	case HV_X64_MSR_RESET:
1294 		return hv_vcpu->cpuid_cache.features_eax &
1295 			HV_MSR_RESET_AVAILABLE;
1296 	case HV_X64_MSR_REFERENCE_TSC:
1297 		return hv_vcpu->cpuid_cache.features_eax &
1298 			HV_MSR_REFERENCE_TSC_AVAILABLE;
1299 	case HV_X64_MSR_SCONTROL:
1300 	case HV_X64_MSR_SVERSION:
1301 	case HV_X64_MSR_SIEFP:
1302 	case HV_X64_MSR_SIMP:
1303 	case HV_X64_MSR_EOM:
1304 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1305 		return hv_vcpu->cpuid_cache.features_eax &
1306 			HV_MSR_SYNIC_AVAILABLE;
1307 	case HV_X64_MSR_STIMER0_CONFIG:
1308 	case HV_X64_MSR_STIMER1_CONFIG:
1309 	case HV_X64_MSR_STIMER2_CONFIG:
1310 	case HV_X64_MSR_STIMER3_CONFIG:
1311 	case HV_X64_MSR_STIMER0_COUNT:
1312 	case HV_X64_MSR_STIMER1_COUNT:
1313 	case HV_X64_MSR_STIMER2_COUNT:
1314 	case HV_X64_MSR_STIMER3_COUNT:
1315 		return hv_vcpu->cpuid_cache.features_eax &
1316 			HV_MSR_SYNTIMER_AVAILABLE;
1317 	case HV_X64_MSR_EOI:
1318 	case HV_X64_MSR_ICR:
1319 	case HV_X64_MSR_TPR:
1320 	case HV_X64_MSR_VP_ASSIST_PAGE:
1321 		return hv_vcpu->cpuid_cache.features_eax &
1322 			HV_MSR_APIC_ACCESS_AVAILABLE;
1323 	case HV_X64_MSR_TSC_FREQUENCY:
1324 	case HV_X64_MSR_APIC_FREQUENCY:
1325 		return hv_vcpu->cpuid_cache.features_eax &
1326 			HV_ACCESS_FREQUENCY_MSRS;
1327 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1328 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1329 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1330 		return hv_vcpu->cpuid_cache.features_eax &
1331 			HV_ACCESS_REENLIGHTENMENT;
1332 	case HV_X64_MSR_TSC_INVARIANT_CONTROL:
1333 		return hv_vcpu->cpuid_cache.features_eax &
1334 			HV_ACCESS_TSC_INVARIANT;
1335 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1336 	case HV_X64_MSR_CRASH_CTL:
1337 		return hv_vcpu->cpuid_cache.features_edx &
1338 			HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
1339 	case HV_X64_MSR_SYNDBG_OPTIONS:
1340 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1341 		return hv_vcpu->cpuid_cache.features_edx &
1342 			HV_FEATURE_DEBUG_MSRS_AVAILABLE;
1343 	default:
1344 		break;
1345 	}
1346 
1347 	return false;
1348 }
1349 
1350 #define KVM_HV_WIN2016_GUEST_ID 0x1040a00003839
1351 #define KVM_HV_WIN2016_GUEST_ID_MASK (~GENMASK_ULL(23, 16)) /* mask out the service version */
1352 
1353 /*
1354  * Hyper-V enabled Windows Server 2016 SMP VMs fail to boot in !XSAVES && XSAVEC
1355  * configuration.
1356  * Such configuration can result from, for example, AMD Erratum 1386 workaround.
1357  *
1358  * Print a notice so users aren't left wondering what's suddenly gone wrong.
1359  */
1360 static void __kvm_hv_xsaves_xsavec_maybe_warn(struct kvm_vcpu *vcpu)
1361 {
1362 	struct kvm *kvm = vcpu->kvm;
1363 	struct kvm_hv *hv = to_kvm_hv(kvm);
1364 
1365 	/* Check again under the hv_lock.  */
1366 	if (hv->xsaves_xsavec_checked)
1367 		return;
1368 
1369 	if ((hv->hv_guest_os_id & KVM_HV_WIN2016_GUEST_ID_MASK) !=
1370 	    KVM_HV_WIN2016_GUEST_ID)
1371 		return;
1372 
1373 	hv->xsaves_xsavec_checked = true;
1374 
1375 	/* UP configurations aren't affected */
1376 	if (atomic_read(&kvm->online_vcpus) < 2)
1377 		return;
1378 
1379 	if (guest_cpuid_has(vcpu, X86_FEATURE_XSAVES) ||
1380 	    !guest_cpu_cap_has(vcpu, X86_FEATURE_XSAVEC))
1381 		return;
1382 
1383 	pr_notice_ratelimited("Booting SMP Windows KVM VM with !XSAVES && XSAVEC. "
1384 			      "If it fails to boot try disabling XSAVEC in the VM config.\n");
1385 }
1386 
1387 void kvm_hv_xsaves_xsavec_maybe_warn(struct kvm_vcpu *vcpu)
1388 {
1389 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1390 
1391 	if (!vcpu->arch.hyperv_enabled ||
1392 	    hv->xsaves_xsavec_checked)
1393 		return;
1394 
1395 	mutex_lock(&hv->hv_lock);
1396 	__kvm_hv_xsaves_xsavec_maybe_warn(vcpu);
1397 	mutex_unlock(&hv->hv_lock);
1398 }
1399 
1400 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
1401 			     bool host)
1402 {
1403 	struct kvm *kvm = vcpu->kvm;
1404 	struct kvm_hv *hv = to_kvm_hv(kvm);
1405 
1406 	if (unlikely(!host && !hv_check_msr_access(to_hv_vcpu(vcpu), msr)))
1407 		return 1;
1408 
1409 	switch (msr) {
1410 	case HV_X64_MSR_GUEST_OS_ID:
1411 		hv->hv_guest_os_id = data;
1412 		/* setting guest os id to zero disables hypercall page */
1413 		if (!hv->hv_guest_os_id)
1414 			hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1415 		break;
1416 	case HV_X64_MSR_HYPERCALL: {
1417 		u8 instructions[9];
1418 		int i = 0;
1419 		u64 addr;
1420 
1421 		/* if guest os id is not set hypercall should remain disabled */
1422 		if (!hv->hv_guest_os_id)
1423 			break;
1424 		if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1425 			hv->hv_hypercall = data;
1426 			break;
1427 		}
1428 
1429 		/*
1430 		 * If Xen and Hyper-V hypercalls are both enabled, disambiguate
1431 		 * the same way Xen itself does, by setting the bit 31 of EAX
1432 		 * which is RsvdZ in the 32-bit Hyper-V hypercall ABI and just
1433 		 * going to be clobbered on 64-bit.
1434 		 */
1435 		if (kvm_xen_hypercall_enabled(kvm)) {
1436 			/* orl $0x80000000, %eax */
1437 			instructions[i++] = 0x0d;
1438 			instructions[i++] = 0x00;
1439 			instructions[i++] = 0x00;
1440 			instructions[i++] = 0x00;
1441 			instructions[i++] = 0x80;
1442 		}
1443 
1444 		/* vmcall/vmmcall */
1445 		kvm_x86_call(patch_hypercall)(vcpu, instructions + i);
1446 		i += 3;
1447 
1448 		/* ret */
1449 		((unsigned char *)instructions)[i++] = 0xc3;
1450 
1451 		addr = data & HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_MASK;
1452 		if (kvm_vcpu_write_guest(vcpu, addr, instructions, i))
1453 			return 1;
1454 		hv->hv_hypercall = data;
1455 		break;
1456 	}
1457 	case HV_X64_MSR_REFERENCE_TSC:
1458 		hv->hv_tsc_page = data;
1459 		if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE) {
1460 			if (!host)
1461 				hv->hv_tsc_page_status = HV_TSC_PAGE_GUEST_CHANGED;
1462 			else
1463 				hv->hv_tsc_page_status = HV_TSC_PAGE_HOST_CHANGED;
1464 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1465 		} else {
1466 			hv->hv_tsc_page_status = HV_TSC_PAGE_UNSET;
1467 		}
1468 		break;
1469 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1470 		return kvm_hv_msr_set_crash_data(kvm,
1471 						 msr - HV_X64_MSR_CRASH_P0,
1472 						 data);
1473 	case HV_X64_MSR_CRASH_CTL:
1474 		if (host)
1475 			return kvm_hv_msr_set_crash_ctl(kvm, data);
1476 
1477 		if (data & HV_CRASH_CTL_CRASH_NOTIFY) {
1478 			vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
1479 				   hv->hv_crash_param[0],
1480 				   hv->hv_crash_param[1],
1481 				   hv->hv_crash_param[2],
1482 				   hv->hv_crash_param[3],
1483 				   hv->hv_crash_param[4]);
1484 
1485 			/* Send notification about crash to user space */
1486 			kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
1487 		}
1488 		break;
1489 	case HV_X64_MSR_RESET:
1490 		if (data == 1) {
1491 			vcpu_debug(vcpu, "hyper-v reset requested\n");
1492 			kvm_make_request(KVM_REQ_HV_RESET, vcpu);
1493 		}
1494 		break;
1495 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1496 		hv->hv_reenlightenment_control = data;
1497 		break;
1498 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1499 		hv->hv_tsc_emulation_control = data;
1500 		break;
1501 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1502 		if (data && !host)
1503 			return 1;
1504 
1505 		hv->hv_tsc_emulation_status = data;
1506 		break;
1507 	case HV_X64_MSR_TIME_REF_COUNT:
1508 		/* read-only, but still ignore it if host-initiated */
1509 		if (!host)
1510 			return 1;
1511 		break;
1512 	case HV_X64_MSR_TSC_INVARIANT_CONTROL:
1513 		/* Only bit 0 is supported */
1514 		if (data & ~HV_EXPOSE_INVARIANT_TSC)
1515 			return 1;
1516 
1517 		/* The feature can't be disabled from the guest */
1518 		if (!host && hv->hv_invtsc_control && !data)
1519 			return 1;
1520 
1521 		hv->hv_invtsc_control = data;
1522 		break;
1523 	case HV_X64_MSR_SYNDBG_OPTIONS:
1524 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1525 		return syndbg_set_msr(vcpu, msr, data, host);
1526 	default:
1527 		kvm_pr_unimpl_wrmsr(vcpu, msr, data);
1528 		return 1;
1529 	}
1530 	return 0;
1531 }
1532 
1533 /* Calculate cpu time spent by current task in 100ns units */
1534 static u64 current_task_runtime_100ns(void)
1535 {
1536 	u64 utime, stime;
1537 
1538 	task_cputime_adjusted(current, &utime, &stime);
1539 
1540 	return div_u64(utime + stime, 100);
1541 }
1542 
1543 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1544 {
1545 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
1546 
1547 	if (unlikely(!host && !hv_check_msr_access(hv_vcpu, msr)))
1548 		return 1;
1549 
1550 	switch (msr) {
1551 	case HV_X64_MSR_VP_INDEX: {
1552 		struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1553 		u32 new_vp_index = (u32)data;
1554 
1555 		if (!host || new_vp_index >= KVM_MAX_VCPUS)
1556 			return 1;
1557 
1558 		if (new_vp_index == hv_vcpu->vp_index)
1559 			return 0;
1560 
1561 		/*
1562 		 * The VP index is initialized to vcpu_index by
1563 		 * kvm_hv_vcpu_postcreate so they initially match.  Now the
1564 		 * VP index is changing, adjust num_mismatched_vp_indexes if
1565 		 * it now matches or no longer matches vcpu_idx.
1566 		 */
1567 		if (hv_vcpu->vp_index == vcpu->vcpu_idx)
1568 			atomic_inc(&hv->num_mismatched_vp_indexes);
1569 		else if (new_vp_index == vcpu->vcpu_idx)
1570 			atomic_dec(&hv->num_mismatched_vp_indexes);
1571 
1572 		hv_vcpu->vp_index = new_vp_index;
1573 		break;
1574 	}
1575 	case HV_X64_MSR_VP_ASSIST_PAGE: {
1576 		u64 gfn;
1577 		unsigned long addr;
1578 
1579 		if (!(data & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE)) {
1580 			hv_vcpu->hv_vapic = data;
1581 			if (kvm_lapic_set_pv_eoi(vcpu, 0, 0))
1582 				return 1;
1583 			break;
1584 		}
1585 		gfn = data >> HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT;
1586 		addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1587 		if (kvm_is_error_hva(addr))
1588 			return 1;
1589 
1590 		/*
1591 		 * Clear apic_assist portion of struct hv_vp_assist_page
1592 		 * only, there can be valuable data in the rest which needs
1593 		 * to be preserved e.g. on migration.
1594 		 */
1595 		if (put_user(0, (u32 __user *)addr))
1596 			return 1;
1597 		hv_vcpu->hv_vapic = data;
1598 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
1599 		if (kvm_lapic_set_pv_eoi(vcpu,
1600 					    gfn_to_gpa(gfn) | KVM_MSR_ENABLED,
1601 					    sizeof(struct hv_vp_assist_page)))
1602 			return 1;
1603 		break;
1604 	}
1605 	case HV_X64_MSR_EOI:
1606 		return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1607 	case HV_X64_MSR_ICR:
1608 		return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1609 	case HV_X64_MSR_TPR:
1610 		return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1611 	case HV_X64_MSR_VP_RUNTIME:
1612 		if (!host)
1613 			return 1;
1614 		hv_vcpu->runtime_offset = data - current_task_runtime_100ns();
1615 		break;
1616 	case HV_X64_MSR_SCONTROL:
1617 	case HV_X64_MSR_SVERSION:
1618 	case HV_X64_MSR_SIEFP:
1619 	case HV_X64_MSR_SIMP:
1620 	case HV_X64_MSR_EOM:
1621 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1622 		return synic_set_msr(to_hv_synic(vcpu), msr, data, host);
1623 	case HV_X64_MSR_STIMER0_CONFIG:
1624 	case HV_X64_MSR_STIMER1_CONFIG:
1625 	case HV_X64_MSR_STIMER2_CONFIG:
1626 	case HV_X64_MSR_STIMER3_CONFIG: {
1627 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1628 
1629 		return stimer_set_config(to_hv_stimer(vcpu, timer_index),
1630 					 data, host);
1631 	}
1632 	case HV_X64_MSR_STIMER0_COUNT:
1633 	case HV_X64_MSR_STIMER1_COUNT:
1634 	case HV_X64_MSR_STIMER2_COUNT:
1635 	case HV_X64_MSR_STIMER3_COUNT: {
1636 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1637 
1638 		return stimer_set_count(to_hv_stimer(vcpu, timer_index),
1639 					data, host);
1640 	}
1641 	case HV_X64_MSR_TSC_FREQUENCY:
1642 	case HV_X64_MSR_APIC_FREQUENCY:
1643 		/* read-only, but still ignore it if host-initiated */
1644 		if (!host)
1645 			return 1;
1646 		break;
1647 	default:
1648 		kvm_pr_unimpl_wrmsr(vcpu, msr, data);
1649 		return 1;
1650 	}
1651 
1652 	return 0;
1653 }
1654 
1655 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1656 			     bool host)
1657 {
1658 	u64 data = 0;
1659 	struct kvm *kvm = vcpu->kvm;
1660 	struct kvm_hv *hv = to_kvm_hv(kvm);
1661 
1662 	if (unlikely(!host && !hv_check_msr_access(to_hv_vcpu(vcpu), msr)))
1663 		return 1;
1664 
1665 	switch (msr) {
1666 	case HV_X64_MSR_GUEST_OS_ID:
1667 		data = hv->hv_guest_os_id;
1668 		break;
1669 	case HV_X64_MSR_HYPERCALL:
1670 		data = hv->hv_hypercall;
1671 		break;
1672 	case HV_X64_MSR_TIME_REF_COUNT:
1673 		data = get_time_ref_counter(kvm);
1674 		break;
1675 	case HV_X64_MSR_REFERENCE_TSC:
1676 		data = hv->hv_tsc_page;
1677 		break;
1678 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1679 		return kvm_hv_msr_get_crash_data(kvm,
1680 						 msr - HV_X64_MSR_CRASH_P0,
1681 						 pdata);
1682 	case HV_X64_MSR_CRASH_CTL:
1683 		return kvm_hv_msr_get_crash_ctl(kvm, pdata);
1684 	case HV_X64_MSR_RESET:
1685 		data = 0;
1686 		break;
1687 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1688 		data = hv->hv_reenlightenment_control;
1689 		break;
1690 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1691 		data = hv->hv_tsc_emulation_control;
1692 		break;
1693 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1694 		data = hv->hv_tsc_emulation_status;
1695 		break;
1696 	case HV_X64_MSR_TSC_INVARIANT_CONTROL:
1697 		data = hv->hv_invtsc_control;
1698 		break;
1699 	case HV_X64_MSR_SYNDBG_OPTIONS:
1700 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1701 		return syndbg_get_msr(vcpu, msr, pdata, host);
1702 	default:
1703 		kvm_pr_unimpl_rdmsr(vcpu, msr);
1704 		return 1;
1705 	}
1706 
1707 	*pdata = data;
1708 	return 0;
1709 }
1710 
1711 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1712 			  bool host)
1713 {
1714 	u64 data = 0;
1715 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
1716 
1717 	if (unlikely(!host && !hv_check_msr_access(hv_vcpu, msr)))
1718 		return 1;
1719 
1720 	switch (msr) {
1721 	case HV_X64_MSR_VP_INDEX:
1722 		data = hv_vcpu->vp_index;
1723 		break;
1724 	case HV_X64_MSR_EOI:
1725 		return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1726 	case HV_X64_MSR_ICR:
1727 		return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1728 	case HV_X64_MSR_TPR:
1729 		return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1730 	case HV_X64_MSR_VP_ASSIST_PAGE:
1731 		data = hv_vcpu->hv_vapic;
1732 		break;
1733 	case HV_X64_MSR_VP_RUNTIME:
1734 		data = current_task_runtime_100ns() + hv_vcpu->runtime_offset;
1735 		break;
1736 	case HV_X64_MSR_SCONTROL:
1737 	case HV_X64_MSR_SVERSION:
1738 	case HV_X64_MSR_SIEFP:
1739 	case HV_X64_MSR_SIMP:
1740 	case HV_X64_MSR_EOM:
1741 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1742 		return synic_get_msr(to_hv_synic(vcpu), msr, pdata, host);
1743 	case HV_X64_MSR_STIMER0_CONFIG:
1744 	case HV_X64_MSR_STIMER1_CONFIG:
1745 	case HV_X64_MSR_STIMER2_CONFIG:
1746 	case HV_X64_MSR_STIMER3_CONFIG: {
1747 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1748 
1749 		return stimer_get_config(to_hv_stimer(vcpu, timer_index),
1750 					 pdata);
1751 	}
1752 	case HV_X64_MSR_STIMER0_COUNT:
1753 	case HV_X64_MSR_STIMER1_COUNT:
1754 	case HV_X64_MSR_STIMER2_COUNT:
1755 	case HV_X64_MSR_STIMER3_COUNT: {
1756 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1757 
1758 		return stimer_get_count(to_hv_stimer(vcpu, timer_index),
1759 					pdata);
1760 	}
1761 	case HV_X64_MSR_TSC_FREQUENCY:
1762 		data = (u64)vcpu->arch.virtual_tsc_khz * 1000;
1763 		break;
1764 	case HV_X64_MSR_APIC_FREQUENCY:
1765 		data = div64_u64(1000000000ULL,
1766 				 vcpu->kvm->arch.apic_bus_cycle_ns);
1767 		break;
1768 	default:
1769 		kvm_pr_unimpl_rdmsr(vcpu, msr);
1770 		return 1;
1771 	}
1772 	*pdata = data;
1773 	return 0;
1774 }
1775 
1776 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1777 {
1778 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1779 
1780 	if (!host && !vcpu->arch.hyperv_enabled)
1781 		return 1;
1782 
1783 	if (kvm_hv_vcpu_init(vcpu))
1784 		return 1;
1785 
1786 	if (kvm_hv_msr_partition_wide(msr)) {
1787 		int r;
1788 
1789 		mutex_lock(&hv->hv_lock);
1790 		r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1791 		mutex_unlock(&hv->hv_lock);
1792 		return r;
1793 	} else
1794 		return kvm_hv_set_msr(vcpu, msr, data, host);
1795 }
1796 
1797 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
1798 {
1799 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1800 
1801 	if (!host && !vcpu->arch.hyperv_enabled)
1802 		return 1;
1803 
1804 	if (kvm_hv_vcpu_init(vcpu))
1805 		return 1;
1806 
1807 	if (kvm_hv_msr_partition_wide(msr)) {
1808 		int r;
1809 
1810 		mutex_lock(&hv->hv_lock);
1811 		r = kvm_hv_get_msr_pw(vcpu, msr, pdata, host);
1812 		mutex_unlock(&hv->hv_lock);
1813 		return r;
1814 	} else
1815 		return kvm_hv_get_msr(vcpu, msr, pdata, host);
1816 }
1817 
1818 static void sparse_set_to_vcpu_mask(struct kvm *kvm, u64 *sparse_banks,
1819 				    u64 valid_bank_mask, unsigned long *vcpu_mask)
1820 {
1821 	struct kvm_hv *hv = to_kvm_hv(kvm);
1822 	bool has_mismatch = atomic_read(&hv->num_mismatched_vp_indexes);
1823 	u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1824 	struct kvm_vcpu *vcpu;
1825 	int bank, sbank = 0;
1826 	unsigned long i;
1827 	u64 *bitmap;
1828 
1829 	BUILD_BUG_ON(sizeof(vp_bitmap) >
1830 		     sizeof(*vcpu_mask) * BITS_TO_LONGS(KVM_MAX_VCPUS));
1831 
1832 	/*
1833 	 * If vp_index == vcpu_idx for all vCPUs, fill vcpu_mask directly, else
1834 	 * fill a temporary buffer and manually test each vCPU's VP index.
1835 	 */
1836 	if (likely(!has_mismatch))
1837 		bitmap = (u64 *)vcpu_mask;
1838 	else
1839 		bitmap = vp_bitmap;
1840 
1841 	/*
1842 	 * Each set of 64 VPs is packed into sparse_banks, with valid_bank_mask
1843 	 * having a '1' for each bank that exists in sparse_banks.  Sets must
1844 	 * be in ascending order, i.e. bank0..bankN.
1845 	 */
1846 	memset(bitmap, 0, sizeof(vp_bitmap));
1847 	for_each_set_bit(bank, (unsigned long *)&valid_bank_mask,
1848 			 KVM_HV_MAX_SPARSE_VCPU_SET_BITS)
1849 		bitmap[bank] = sparse_banks[sbank++];
1850 
1851 	if (likely(!has_mismatch))
1852 		return;
1853 
1854 	bitmap_zero(vcpu_mask, KVM_MAX_VCPUS);
1855 	kvm_for_each_vcpu(i, vcpu, kvm) {
1856 		if (test_bit(kvm_hv_get_vpindex(vcpu), (unsigned long *)vp_bitmap))
1857 			__set_bit(i, vcpu_mask);
1858 	}
1859 }
1860 
1861 static bool hv_is_vp_in_sparse_set(u32 vp_id, u64 valid_bank_mask, u64 sparse_banks[])
1862 {
1863 	int valid_bit_nr = vp_id / HV_VCPUS_PER_SPARSE_BANK;
1864 	unsigned long sbank;
1865 
1866 	BUILD_BUG_ON(BITS_PER_TYPE(valid_bank_mask) != HV_MAX_SPARSE_VCPU_BANKS);
1867 
1868 	if (valid_bit_nr >= HV_MAX_SPARSE_VCPU_BANKS)
1869 		return false;
1870 
1871 	if (!test_bit(valid_bit_nr, (unsigned long *)&valid_bank_mask))
1872 		return false;
1873 
1874 	/*
1875 	 * The index into the sparse bank is the number of preceding bits in
1876 	 * the valid mask.  Optimize for VMs with <64 vCPUs by skipping the
1877 	 * fancy math if there can't possibly be preceding bits.
1878 	 */
1879 	if (valid_bit_nr)
1880 		sbank = hweight64(valid_bank_mask & GENMASK_ULL(valid_bit_nr - 1, 0));
1881 	else
1882 		sbank = 0;
1883 
1884 	return test_bit(vp_id % HV_VCPUS_PER_SPARSE_BANK,
1885 			(unsigned long *)&sparse_banks[sbank]);
1886 }
1887 
1888 struct kvm_hv_hcall {
1889 	/* Hypercall input data */
1890 	u64 param;
1891 	u64 ingpa;
1892 	u64 outgpa;
1893 	u16 code;
1894 	u16 var_cnt;
1895 	u16 rep_cnt;
1896 	u16 rep_idx;
1897 	bool fast;
1898 	bool rep;
1899 	sse128_t xmm[HV_HYPERCALL_MAX_XMM_REGISTERS];
1900 
1901 	/*
1902 	 * Current read offset when KVM reads hypercall input data gradually,
1903 	 * either offset in bytes from 'ingpa' for regular hypercalls or the
1904 	 * number of already consumed 'XMM halves' for 'fast' hypercalls.
1905 	 */
1906 	union {
1907 		gpa_t data_offset;
1908 		int consumed_xmm_halves;
1909 	};
1910 };
1911 
1912 
1913 static int kvm_hv_get_hc_data(struct kvm *kvm, struct kvm_hv_hcall *hc,
1914 			      u16 orig_cnt, u16 cnt_cap, u64 *data)
1915 {
1916 	/*
1917 	 * Preserve the original count when ignoring entries via a "cap", KVM
1918 	 * still needs to validate the guest input (though the non-XMM path
1919 	 * punts on the checks).
1920 	 */
1921 	u16 cnt = min(orig_cnt, cnt_cap);
1922 	int i, j;
1923 
1924 	if (hc->fast) {
1925 		/*
1926 		 * Each XMM holds two sparse banks, but do not count halves that
1927 		 * have already been consumed for hypercall parameters.
1928 		 */
1929 		if (orig_cnt > 2 * HV_HYPERCALL_MAX_XMM_REGISTERS - hc->consumed_xmm_halves)
1930 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1931 
1932 		for (i = 0; i < cnt; i++) {
1933 			j = i + hc->consumed_xmm_halves;
1934 			if (j % 2)
1935 				data[i] = sse128_hi(hc->xmm[j / 2]);
1936 			else
1937 				data[i] = sse128_lo(hc->xmm[j / 2]);
1938 		}
1939 		return 0;
1940 	}
1941 
1942 	return kvm_read_guest(kvm, hc->ingpa + hc->data_offset, data,
1943 			      cnt * sizeof(*data));
1944 }
1945 
1946 static u64 kvm_get_sparse_vp_set(struct kvm *kvm, struct kvm_hv_hcall *hc,
1947 				 u64 *sparse_banks)
1948 {
1949 	if (hc->var_cnt > HV_MAX_SPARSE_VCPU_BANKS)
1950 		return -EINVAL;
1951 
1952 	/* Cap var_cnt to ignore banks that cannot contain a legal VP index. */
1953 	return kvm_hv_get_hc_data(kvm, hc, hc->var_cnt, KVM_HV_MAX_SPARSE_VCPU_SET_BITS,
1954 				  sparse_banks);
1955 }
1956 
1957 static int kvm_hv_get_tlb_flush_entries(struct kvm *kvm, struct kvm_hv_hcall *hc, u64 entries[])
1958 {
1959 	return kvm_hv_get_hc_data(kvm, hc, hc->rep_cnt, hc->rep_cnt, entries);
1960 }
1961 
1962 static void hv_tlb_flush_enqueue(struct kvm_vcpu *vcpu, u64 *entries, int count,
1963 				 bool is_guest_mode)
1964 {
1965 	struct kvm_vcpu_hv_tlb_flush_fifo *tlb_flush_fifo;
1966 	u64 flush_all_entry = KVM_HV_TLB_FLUSHALL_ENTRY;
1967 
1968 	tlb_flush_fifo = kvm_hv_get_tlb_flush_fifo(vcpu, is_guest_mode);
1969 	if (!tlb_flush_fifo)
1970 		return;
1971 
1972 	spin_lock(&tlb_flush_fifo->write_lock);
1973 
1974 	/*
1975 	 * All entries should fit on the fifo leaving one free for 'flush all'
1976 	 * entry in case another request comes in. In case there's not enough
1977 	 * space, just put 'flush all' entry there.
1978 	 */
1979 	if (count && entries && count < kfifo_avail(&tlb_flush_fifo->entries)) {
1980 		WARN_ON(kfifo_in(&tlb_flush_fifo->entries, entries, count) != count);
1981 		goto out_unlock;
1982 	}
1983 
1984 	/*
1985 	 * Note: full fifo always contains 'flush all' entry, no need to check the
1986 	 * return value.
1987 	 */
1988 	kfifo_in(&tlb_flush_fifo->entries, &flush_all_entry, 1);
1989 
1990 out_unlock:
1991 	spin_unlock(&tlb_flush_fifo->write_lock);
1992 }
1993 
1994 int kvm_hv_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
1995 {
1996 	struct kvm_vcpu_hv_tlb_flush_fifo *tlb_flush_fifo;
1997 	u64 entries[KVM_HV_TLB_FLUSH_FIFO_SIZE];
1998 	int i, j, count;
1999 	gva_t gva;
2000 
2001 	if (!tdp_enabled)
2002 		return -EINVAL;
2003 
2004 	tlb_flush_fifo = kvm_hv_get_tlb_flush_fifo(vcpu, is_guest_mode(vcpu));
2005 	if (!tlb_flush_fifo)
2006 		return -EINVAL;
2007 
2008 	count = kfifo_out(&tlb_flush_fifo->entries, entries, KVM_HV_TLB_FLUSH_FIFO_SIZE);
2009 
2010 	for (i = 0; i < count; i++) {
2011 		if (entries[i] == KVM_HV_TLB_FLUSHALL_ENTRY)
2012 			goto out_flush_all;
2013 
2014 		/*
2015 		 * Lower 12 bits of 'address' encode the number of additional
2016 		 * pages to flush.
2017 		 */
2018 		gva = entries[i] & PAGE_MASK;
2019 		for (j = 0; j < (entries[i] & ~PAGE_MASK) + 1; j++) {
2020 			if (is_noncanonical_invlpg_address(gva + j * PAGE_SIZE, vcpu))
2021 				continue;
2022 
2023 			kvm_x86_call(flush_tlb_gva)(vcpu, gva + j * PAGE_SIZE);
2024 		}
2025 
2026 		++vcpu->stat.tlb_flush;
2027 	}
2028 	return 0;
2029 
2030 out_flush_all:
2031 	kfifo_reset_out(&tlb_flush_fifo->entries);
2032 
2033 	/* Fall back to full flush. */
2034 	return -ENOSPC;
2035 }
2036 
2037 static u64 kvm_hv_flush_tlb(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc)
2038 {
2039 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
2040 	unsigned long *vcpu_mask = hv_vcpu->vcpu_mask;
2041 	u64 *sparse_banks = hv_vcpu->sparse_banks;
2042 	struct kvm *kvm = vcpu->kvm;
2043 	struct hv_tlb_flush_ex flush_ex;
2044 	struct hv_tlb_flush flush;
2045 	/*
2046 	 * Normally, there can be no more than 'KVM_HV_TLB_FLUSH_FIFO_SIZE'
2047 	 * entries on the TLB flush fifo. The last entry, however, needs to be
2048 	 * always left free for 'flush all' entry which gets placed when
2049 	 * there is not enough space to put all the requested entries.
2050 	 */
2051 	u64 __tlb_flush_entries[KVM_HV_TLB_FLUSH_FIFO_SIZE - 1];
2052 	u64 *tlb_flush_entries;
2053 	u64 valid_bank_mask;
2054 	struct kvm_vcpu *v;
2055 	unsigned long i;
2056 	bool all_cpus;
2057 
2058 	/*
2059 	 * The Hyper-V TLFS doesn't allow more than HV_MAX_SPARSE_VCPU_BANKS
2060 	 * sparse banks. Fail the build if KVM's max allowed number of
2061 	 * vCPUs (>4096) exceeds this limit.
2062 	 */
2063 	BUILD_BUG_ON(KVM_HV_MAX_SPARSE_VCPU_SET_BITS > HV_MAX_SPARSE_VCPU_BANKS);
2064 
2065 	/*
2066 	 * 'Slow' hypercall's first parameter is the address in guest's memory
2067 	 * where hypercall parameters are placed. This is either a GPA or a
2068 	 * nested GPA when KVM is handling the call from L2 ('direct' TLB
2069 	 * flush).  Translate the address here so the memory can be uniformly
2070 	 * read with kvm_read_guest().
2071 	 */
2072 	if (!hc->fast) {
2073 		hc->ingpa = kvm_translate_gpa(vcpu, &vcpu->arch.gva_walk, hc->ingpa,
2074 					      PFERR_GUEST_FINAL_MASK, NULL, 0);
2075 		if (unlikely(hc->ingpa == INVALID_GPA))
2076 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
2077 	}
2078 
2079 	if (hc->code == HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST ||
2080 	    hc->code == HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE) {
2081 		if (hc->fast) {
2082 			flush.address_space = hc->ingpa;
2083 			flush.flags = hc->outgpa;
2084 			flush.processor_mask = sse128_lo(hc->xmm[0]);
2085 			hc->consumed_xmm_halves = 1;
2086 		} else {
2087 			if (unlikely(kvm_read_guest(kvm, hc->ingpa,
2088 						    &flush, sizeof(flush))))
2089 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
2090 			hc->data_offset = sizeof(flush);
2091 		}
2092 
2093 		trace_kvm_hv_flush_tlb(flush.processor_mask,
2094 				       flush.address_space, flush.flags,
2095 				       is_guest_mode(vcpu));
2096 
2097 		valid_bank_mask = BIT_ULL(0);
2098 		sparse_banks[0] = flush.processor_mask;
2099 
2100 		/*
2101 		 * Work around possible WS2012 bug: it sends hypercalls
2102 		 * with processor_mask = 0x0 and HV_FLUSH_ALL_PROCESSORS clear,
2103 		 * while also expecting us to flush something and crashing if
2104 		 * we don't. Let's treat processor_mask == 0 same as
2105 		 * HV_FLUSH_ALL_PROCESSORS.
2106 		 */
2107 		all_cpus = (flush.flags & HV_FLUSH_ALL_PROCESSORS) ||
2108 			flush.processor_mask == 0;
2109 	} else {
2110 		if (hc->fast) {
2111 			flush_ex.address_space = hc->ingpa;
2112 			flush_ex.flags = hc->outgpa;
2113 			memcpy(&flush_ex.hv_vp_set,
2114 			       &hc->xmm[0], sizeof(hc->xmm[0]));
2115 			hc->consumed_xmm_halves = 2;
2116 		} else {
2117 			if (unlikely(kvm_read_guest(kvm, hc->ingpa, &flush_ex,
2118 						    sizeof(flush_ex))))
2119 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
2120 			hc->data_offset = sizeof(flush_ex);
2121 		}
2122 
2123 		trace_kvm_hv_flush_tlb_ex(flush_ex.hv_vp_set.valid_bank_mask,
2124 					  flush_ex.hv_vp_set.format,
2125 					  flush_ex.address_space,
2126 					  flush_ex.flags, is_guest_mode(vcpu));
2127 
2128 		valid_bank_mask = flush_ex.hv_vp_set.valid_bank_mask;
2129 		all_cpus = flush_ex.hv_vp_set.format !=
2130 			HV_GENERIC_SET_SPARSE_4K;
2131 
2132 		if (hc->var_cnt != hweight64(valid_bank_mask))
2133 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
2134 
2135 		if (!all_cpus) {
2136 			if (!hc->var_cnt)
2137 				goto ret_success;
2138 
2139 			if (kvm_get_sparse_vp_set(kvm, hc, sparse_banks))
2140 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
2141 		}
2142 
2143 		/*
2144 		 * Hyper-V TLFS doesn't explicitly forbid non-empty sparse vCPU
2145 		 * banks (and, thus, non-zero 'var_cnt') for the 'all vCPUs'
2146 		 * case (HV_GENERIC_SET_ALL).  Always adjust data_offset and
2147 		 * consumed_xmm_halves to make sure TLB flush entries are read
2148 		 * from the correct offset.
2149 		 */
2150 		if (hc->fast)
2151 			hc->consumed_xmm_halves += hc->var_cnt;
2152 		else
2153 			hc->data_offset += hc->var_cnt * sizeof(sparse_banks[0]);
2154 	}
2155 
2156 	if (hc->code == HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE ||
2157 	    hc->code == HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX ||
2158 	    hc->rep_cnt > ARRAY_SIZE(__tlb_flush_entries)) {
2159 		tlb_flush_entries = NULL;
2160 	} else {
2161 		if (kvm_hv_get_tlb_flush_entries(kvm, hc, __tlb_flush_entries))
2162 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
2163 		tlb_flush_entries = __tlb_flush_entries;
2164 	}
2165 
2166 	/*
2167 	 * vcpu->arch.cr3 may not be up-to-date for running vCPUs so we can't
2168 	 * analyze it here, flush TLB regardless of the specified address space.
2169 	 */
2170 	if (all_cpus && !is_guest_mode(vcpu)) {
2171 		kvm_for_each_vcpu(i, v, kvm)
2172 			hv_tlb_flush_enqueue(v, tlb_flush_entries, hc->rep_cnt, false);
2173 
2174 		kvm_make_all_cpus_request(kvm, KVM_REQ_HV_TLB_FLUSH);
2175 	} else if (!is_guest_mode(vcpu)) {
2176 		sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask, vcpu_mask);
2177 
2178 		for_each_set_bit(i, vcpu_mask, KVM_MAX_VCPUS) {
2179 			v = kvm_get_vcpu(kvm, i);
2180 			if (!v)
2181 				continue;
2182 			hv_tlb_flush_enqueue(v, tlb_flush_entries, hc->rep_cnt, false);
2183 		}
2184 
2185 		kvm_make_vcpus_request_mask(kvm, KVM_REQ_HV_TLB_FLUSH, vcpu_mask);
2186 	} else {
2187 		struct kvm_vcpu_hv *hv_v;
2188 
2189 		bitmap_zero(vcpu_mask, KVM_MAX_VCPUS);
2190 
2191 		kvm_for_each_vcpu(i, v, kvm) {
2192 			hv_v = to_hv_vcpu_safe(v);
2193 
2194 			/*
2195 			 * The following check races with nested vCPUs entering/exiting
2196 			 * and/or migrating between L1's vCPUs, however the only case when
2197 			 * KVM *must* flush the TLB is when the target L2 vCPU keeps
2198 			 * running on the same L1 vCPU from the moment of the request until
2199 			 * kvm_hv_flush_tlb() returns. TLB is fully flushed in all other
2200 			 * cases, e.g. when the target L2 vCPU migrates to a different L1
2201 			 * vCPU or when the corresponding L1 vCPU temporary switches to a
2202 			 * different L2 vCPU while the request is being processed.
2203 			 */
2204 			if (!hv_v || hv_v->nested.vm_id != hv_vcpu->nested.vm_id)
2205 				continue;
2206 
2207 			if (!all_cpus &&
2208 			    !hv_is_vp_in_sparse_set(hv_v->nested.vp_id, valid_bank_mask,
2209 						    sparse_banks))
2210 				continue;
2211 
2212 			__set_bit(i, vcpu_mask);
2213 			hv_tlb_flush_enqueue(v, tlb_flush_entries, hc->rep_cnt, true);
2214 		}
2215 
2216 		kvm_make_vcpus_request_mask(kvm, KVM_REQ_HV_TLB_FLUSH, vcpu_mask);
2217 	}
2218 
2219 ret_success:
2220 	/* We always do full TLB flush, set 'Reps completed' = 'Rep Count' */
2221 	return (u64)HV_STATUS_SUCCESS |
2222 		((u64)hc->rep_cnt << HV_HYPERCALL_REP_COMP_OFFSET);
2223 }
2224 
2225 static void kvm_hv_send_ipi_to_many(struct kvm *kvm, u32 vector,
2226 				    u64 *sparse_banks, u64 valid_bank_mask)
2227 {
2228 	struct kvm_lapic_irq irq = {
2229 		.delivery_mode = APIC_DM_FIXED,
2230 		.vector = vector
2231 	};
2232 	struct kvm_vcpu *vcpu;
2233 	unsigned long i;
2234 
2235 	kvm_for_each_vcpu(i, vcpu, kvm) {
2236 		if (sparse_banks &&
2237 		    !hv_is_vp_in_sparse_set(kvm_hv_get_vpindex(vcpu),
2238 					    valid_bank_mask, sparse_banks))
2239 			continue;
2240 
2241 		/* We fail only when APIC is disabled */
2242 		kvm_apic_set_irq(vcpu, &irq, NULL);
2243 	}
2244 }
2245 
2246 static u64 kvm_hv_send_ipi(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc)
2247 {
2248 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
2249 	u64 *sparse_banks = hv_vcpu->sparse_banks;
2250 	struct kvm *kvm = vcpu->kvm;
2251 	struct hv_send_ipi_ex send_ipi_ex;
2252 	struct hv_send_ipi send_ipi;
2253 	u64 valid_bank_mask;
2254 	u32 vector;
2255 	bool all_cpus;
2256 
2257 	if (!lapic_in_kernel(vcpu))
2258 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
2259 
2260 	if (hc->code == HVCALL_SEND_IPI) {
2261 		if (!hc->fast) {
2262 			if (unlikely(kvm_read_guest(kvm, hc->ingpa, &send_ipi,
2263 						    sizeof(send_ipi))))
2264 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
2265 			sparse_banks[0] = send_ipi.cpu_mask;
2266 			vector = send_ipi.vector;
2267 		} else {
2268 			/* 'reserved' part of hv_send_ipi should be 0 */
2269 			if (unlikely(hc->ingpa >> 32 != 0))
2270 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
2271 			sparse_banks[0] = hc->outgpa;
2272 			vector = (u32)hc->ingpa;
2273 		}
2274 		all_cpus = false;
2275 		valid_bank_mask = BIT_ULL(0);
2276 
2277 		trace_kvm_hv_send_ipi(vector, sparse_banks[0]);
2278 	} else {
2279 		if (!hc->fast) {
2280 			if (unlikely(kvm_read_guest(kvm, hc->ingpa, &send_ipi_ex,
2281 						    sizeof(send_ipi_ex))))
2282 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
2283 		} else {
2284 			send_ipi_ex.vector = (u32)hc->ingpa;
2285 			send_ipi_ex.vp_set.format = hc->outgpa;
2286 			send_ipi_ex.vp_set.valid_bank_mask = sse128_lo(hc->xmm[0]);
2287 		}
2288 
2289 		trace_kvm_hv_send_ipi_ex(send_ipi_ex.vector,
2290 					 send_ipi_ex.vp_set.format,
2291 					 send_ipi_ex.vp_set.valid_bank_mask);
2292 
2293 		vector = send_ipi_ex.vector;
2294 		valid_bank_mask = send_ipi_ex.vp_set.valid_bank_mask;
2295 		all_cpus = send_ipi_ex.vp_set.format == HV_GENERIC_SET_ALL;
2296 
2297 		if (hc->var_cnt != hweight64(valid_bank_mask))
2298 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
2299 
2300 		if (all_cpus)
2301 			goto check_and_send_ipi;
2302 
2303 		if (!hc->var_cnt)
2304 			goto ret_success;
2305 
2306 		if (!hc->fast)
2307 			hc->data_offset = offsetof(struct hv_send_ipi_ex,
2308 						   vp_set.bank_contents);
2309 		else
2310 			hc->consumed_xmm_halves = 1;
2311 
2312 		if (kvm_get_sparse_vp_set(kvm, hc, sparse_banks))
2313 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
2314 	}
2315 
2316 check_and_send_ipi:
2317 	if ((vector < HV_IPI_LOW_VECTOR) || (vector > HV_IPI_HIGH_VECTOR))
2318 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
2319 
2320 	if (all_cpus)
2321 		kvm_hv_send_ipi_to_many(kvm, vector, NULL, 0);
2322 	else
2323 		kvm_hv_send_ipi_to_many(kvm, vector, sparse_banks, valid_bank_mask);
2324 
2325 ret_success:
2326 	return HV_STATUS_SUCCESS;
2327 }
2328 
2329 void kvm_hv_set_cpuid(struct kvm_vcpu *vcpu, bool hyperv_enabled)
2330 {
2331 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
2332 	struct kvm_cpuid_entry2 *entry;
2333 
2334 	vcpu->arch.hyperv_enabled = hyperv_enabled;
2335 
2336 	if (!hv_vcpu) {
2337 		/*
2338 		 * KVM should have already allocated kvm_vcpu_hv if Hyper-V is
2339 		 * enabled in CPUID.
2340 		 */
2341 		WARN_ON_ONCE(vcpu->arch.hyperv_enabled);
2342 		return;
2343 	}
2344 
2345 	memset(&hv_vcpu->cpuid_cache, 0, sizeof(hv_vcpu->cpuid_cache));
2346 
2347 	if (!vcpu->arch.hyperv_enabled)
2348 		return;
2349 
2350 	entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_FEATURES);
2351 	if (entry) {
2352 		hv_vcpu->cpuid_cache.features_eax = entry->eax;
2353 		hv_vcpu->cpuid_cache.features_ebx = entry->ebx;
2354 		hv_vcpu->cpuid_cache.features_edx = entry->edx;
2355 	}
2356 
2357 	entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_ENLIGHTMENT_INFO);
2358 	if (entry) {
2359 		hv_vcpu->cpuid_cache.enlightenments_eax = entry->eax;
2360 		hv_vcpu->cpuid_cache.enlightenments_ebx = entry->ebx;
2361 	}
2362 
2363 	entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES);
2364 	if (entry)
2365 		hv_vcpu->cpuid_cache.syndbg_cap_eax = entry->eax;
2366 
2367 	entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_NESTED_FEATURES);
2368 	if (entry) {
2369 		hv_vcpu->cpuid_cache.nested_eax = entry->eax;
2370 		hv_vcpu->cpuid_cache.nested_ebx = entry->ebx;
2371 	}
2372 }
2373 
2374 int kvm_hv_set_enforce_cpuid(struct kvm_vcpu *vcpu, bool enforce)
2375 {
2376 	struct kvm_vcpu_hv *hv_vcpu;
2377 	int ret = 0;
2378 
2379 	if (!to_hv_vcpu(vcpu)) {
2380 		if (enforce) {
2381 			ret = kvm_hv_vcpu_init(vcpu);
2382 			if (ret)
2383 				return ret;
2384 		} else {
2385 			return 0;
2386 		}
2387 	}
2388 
2389 	hv_vcpu = to_hv_vcpu(vcpu);
2390 	hv_vcpu->enforce_cpuid = enforce;
2391 
2392 	return ret;
2393 }
2394 
2395 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
2396 {
2397 	bool longmode;
2398 
2399 	longmode = is_64_bit_hypercall(vcpu);
2400 	if (longmode)
2401 		kvm_rax_write_raw(vcpu, result);
2402 	else {
2403 		kvm_edx_write(vcpu, result >> 32);
2404 		kvm_eax_write(vcpu, result);
2405 	}
2406 }
2407 
2408 static int kvm_hv_hypercall_complete(struct kvm_vcpu *vcpu, u64 result)
2409 {
2410 	u32 tlb_lock_count = 0;
2411 	int ret;
2412 
2413 	if (hv_result_success(result) && is_guest_mode(vcpu) &&
2414 	    kvm_hv_is_tlb_flush_hcall(vcpu) &&
2415 	    kvm_read_guest(vcpu->kvm, to_hv_vcpu(vcpu)->nested.pa_page_gpa,
2416 			   &tlb_lock_count, sizeof(tlb_lock_count)))
2417 		result = HV_STATUS_INVALID_HYPERCALL_INPUT;
2418 
2419 	trace_kvm_hv_hypercall_done(result);
2420 	kvm_hv_hypercall_set_result(vcpu, result);
2421 	++vcpu->stat.hypercalls;
2422 
2423 	ret = kvm_skip_emulated_instruction(vcpu);
2424 
2425 	if (tlb_lock_count)
2426 		kvm_nested_call(hv_inject_synthetic_vmexit_post_tlb_flush)(vcpu);
2427 
2428 	return ret;
2429 }
2430 
2431 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
2432 {
2433 	return kvm_hv_hypercall_complete(vcpu, vcpu->run->hyperv.u.hcall.result);
2434 }
2435 
2436 static u16 kvm_hvcall_signal_event(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc)
2437 {
2438 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
2439 	struct eventfd_ctx *eventfd;
2440 
2441 	if (unlikely(!hc->fast)) {
2442 		int ret;
2443 		gpa_t gpa = hc->ingpa;
2444 
2445 		if ((gpa & (__alignof__(hc->ingpa) - 1)) ||
2446 		    offset_in_page(gpa) + sizeof(hc->ingpa) > PAGE_SIZE)
2447 			return HV_STATUS_INVALID_ALIGNMENT;
2448 
2449 		ret = kvm_vcpu_read_guest(vcpu, gpa,
2450 					  &hc->ingpa, sizeof(hc->ingpa));
2451 		if (ret < 0)
2452 			return HV_STATUS_INVALID_ALIGNMENT;
2453 	}
2454 
2455 	/*
2456 	 * Per spec, bits 32-47 contain the extra "flag number".  However, we
2457 	 * have no use for it, and in all known usecases it is zero, so just
2458 	 * report lookup failure if it isn't.
2459 	 */
2460 	if (hc->ingpa & 0xffff00000000ULL)
2461 		return HV_STATUS_INVALID_PORT_ID;
2462 	/* remaining bits are reserved-zero */
2463 	if (hc->ingpa & ~KVM_HYPERV_CONN_ID_MASK)
2464 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
2465 
2466 	/* the eventfd is protected by vcpu->kvm->srcu, but conn_to_evt isn't */
2467 	rcu_read_lock();
2468 	eventfd = idr_find(&hv->conn_to_evt, hc->ingpa);
2469 	rcu_read_unlock();
2470 	if (!eventfd)
2471 		return HV_STATUS_INVALID_PORT_ID;
2472 
2473 	eventfd_signal(eventfd);
2474 	return HV_STATUS_SUCCESS;
2475 }
2476 
2477 static bool is_xmm_fast_hypercall(struct kvm_hv_hcall *hc)
2478 {
2479 	switch (hc->code) {
2480 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
2481 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
2482 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
2483 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
2484 	case HVCALL_SEND_IPI_EX:
2485 		return true;
2486 	}
2487 
2488 	return false;
2489 }
2490 
2491 static void kvm_hv_hypercall_read_xmm(struct kvm_hv_hcall *hc)
2492 {
2493 	int reg;
2494 
2495 	kvm_fpu_get();
2496 	for (reg = 0; reg < HV_HYPERCALL_MAX_XMM_REGISTERS; reg++)
2497 		_kvm_read_sse_reg(reg, &hc->xmm[reg]);
2498 	kvm_fpu_put();
2499 }
2500 
2501 static bool hv_check_hypercall_access(struct kvm_vcpu_hv *hv_vcpu, u16 code)
2502 {
2503 	if (!hv_vcpu->enforce_cpuid)
2504 		return true;
2505 
2506 	switch (code) {
2507 	case HVCALL_NOTIFY_LONG_SPIN_WAIT:
2508 		return hv_vcpu->cpuid_cache.enlightenments_ebx &&
2509 			hv_vcpu->cpuid_cache.enlightenments_ebx != U32_MAX;
2510 	case HVCALL_POST_MESSAGE:
2511 		return hv_vcpu->cpuid_cache.features_ebx & HV_POST_MESSAGES;
2512 	case HVCALL_SIGNAL_EVENT:
2513 		return hv_vcpu->cpuid_cache.features_ebx & HV_SIGNAL_EVENTS;
2514 	case HVCALL_POST_DEBUG_DATA:
2515 	case HVCALL_RETRIEVE_DEBUG_DATA:
2516 	case HVCALL_RESET_DEBUG_SESSION:
2517 		/*
2518 		 * Return 'true' when SynDBG is disabled so the resulting code
2519 		 * will be HV_STATUS_INVALID_HYPERCALL_CODE.
2520 		 */
2521 		return !kvm_hv_is_syndbg_enabled(hv_vcpu->vcpu) ||
2522 			hv_vcpu->cpuid_cache.features_ebx & HV_DEBUGGING;
2523 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
2524 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
2525 		if (!(hv_vcpu->cpuid_cache.enlightenments_eax &
2526 		      HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED))
2527 			return false;
2528 		fallthrough;
2529 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
2530 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
2531 		return hv_vcpu->cpuid_cache.enlightenments_eax &
2532 			HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
2533 	case HVCALL_SEND_IPI_EX:
2534 		if (!(hv_vcpu->cpuid_cache.enlightenments_eax &
2535 		      HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED))
2536 			return false;
2537 		fallthrough;
2538 	case HVCALL_SEND_IPI:
2539 		return hv_vcpu->cpuid_cache.enlightenments_eax &
2540 			HV_X64_CLUSTER_IPI_RECOMMENDED;
2541 	case HV_EXT_CALL_QUERY_CAPABILITIES ... HV_EXT_CALL_MAX:
2542 		return hv_vcpu->cpuid_cache.features_ebx &
2543 			HV_ENABLE_EXTENDED_HYPERCALLS;
2544 	default:
2545 		break;
2546 	}
2547 
2548 	return true;
2549 }
2550 
2551 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
2552 {
2553 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
2554 	struct kvm_hv_hcall hc;
2555 	u64 ret = HV_STATUS_SUCCESS;
2556 
2557 	/*
2558 	 * hypercall generates UD from non zero cpl and real mode
2559 	 * per HYPER-V spec
2560 	 */
2561 	if (kvm_x86_call(get_cpl)(vcpu) != 0 || !is_protmode(vcpu)) {
2562 		kvm_queue_exception(vcpu, UD_VECTOR);
2563 		return 1;
2564 	}
2565 
2566 #ifdef CONFIG_X86_64
2567 	if (is_64_bit_hypercall(vcpu)) {
2568 		hc.param = kvm_rcx_read_raw(vcpu);
2569 		hc.ingpa = kvm_rdx_read_raw(vcpu);
2570 		hc.outgpa = kvm_r8_read_raw(vcpu);
2571 	} else
2572 #endif
2573 	{
2574 		hc.param = ((u64)kvm_edx_read(vcpu) << 32) | kvm_eax_read(vcpu);
2575 		hc.ingpa = ((u64)kvm_ebx_read(vcpu) << 32) | kvm_ecx_read(vcpu);
2576 		hc.outgpa = ((u64)kvm_edi_read(vcpu) << 32) | kvm_esi_read(vcpu);
2577 	}
2578 
2579 	hc.code = hc.param & 0xffff;
2580 	hc.var_cnt = (hc.param & HV_HYPERCALL_VARHEAD_MASK) >> HV_HYPERCALL_VARHEAD_OFFSET;
2581 	hc.fast = !!(hc.param & HV_HYPERCALL_FAST_BIT);
2582 	hc.rep_cnt = (hc.param >> HV_HYPERCALL_REP_COMP_OFFSET) & 0xfff;
2583 	hc.rep_idx = (hc.param >> HV_HYPERCALL_REP_START_OFFSET) & 0xfff;
2584 	hc.rep = !!(hc.rep_cnt || hc.rep_idx);
2585 
2586 	trace_kvm_hv_hypercall(hc.code, hc.fast, hc.var_cnt, hc.rep_cnt,
2587 			       hc.rep_idx, hc.ingpa, hc.outgpa);
2588 
2589 	if (unlikely(!hv_check_hypercall_access(hv_vcpu, hc.code))) {
2590 		ret = HV_STATUS_ACCESS_DENIED;
2591 		goto hypercall_complete;
2592 	}
2593 
2594 	if (unlikely(hc.param & HV_HYPERCALL_RSVD_MASK)) {
2595 		ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2596 		goto hypercall_complete;
2597 	}
2598 
2599 	if (hc.fast && is_xmm_fast_hypercall(&hc)) {
2600 		if (unlikely(hv_vcpu->enforce_cpuid &&
2601 			     !(hv_vcpu->cpuid_cache.features_edx &
2602 			       HV_X64_HYPERCALL_XMM_INPUT_AVAILABLE))) {
2603 			kvm_queue_exception(vcpu, UD_VECTOR);
2604 			return 1;
2605 		}
2606 
2607 		kvm_hv_hypercall_read_xmm(&hc);
2608 	}
2609 
2610 	switch (hc.code) {
2611 	case HVCALL_NOTIFY_LONG_SPIN_WAIT:
2612 		if (unlikely(hc.rep || hc.var_cnt)) {
2613 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2614 			break;
2615 		}
2616 		kvm_vcpu_on_spin(vcpu, true);
2617 		break;
2618 	case HVCALL_SIGNAL_EVENT:
2619 		if (unlikely(hc.rep || hc.var_cnt)) {
2620 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2621 			break;
2622 		}
2623 		ret = kvm_hvcall_signal_event(vcpu, &hc);
2624 		if (ret != HV_STATUS_INVALID_PORT_ID)
2625 			break;
2626 		fallthrough;	/* maybe userspace knows this conn_id */
2627 	case HVCALL_POST_MESSAGE:
2628 		/* don't bother userspace if it has no way to handle it */
2629 		if (unlikely(hc.rep || hc.var_cnt || !to_hv_synic(vcpu)->active)) {
2630 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2631 			break;
2632 		}
2633 		goto hypercall_userspace_exit;
2634 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
2635 		if (unlikely(hc.var_cnt)) {
2636 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2637 			break;
2638 		}
2639 		fallthrough;
2640 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
2641 		if (unlikely(!hc.rep_cnt || hc.rep_idx)) {
2642 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2643 			break;
2644 		}
2645 		ret = kvm_hv_flush_tlb(vcpu, &hc);
2646 		break;
2647 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
2648 		if (unlikely(hc.var_cnt)) {
2649 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2650 			break;
2651 		}
2652 		fallthrough;
2653 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
2654 		if (unlikely(hc.rep)) {
2655 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2656 			break;
2657 		}
2658 		ret = kvm_hv_flush_tlb(vcpu, &hc);
2659 		break;
2660 	case HVCALL_SEND_IPI:
2661 		if (unlikely(hc.var_cnt)) {
2662 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2663 			break;
2664 		}
2665 		fallthrough;
2666 	case HVCALL_SEND_IPI_EX:
2667 		if (unlikely(hc.rep)) {
2668 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2669 			break;
2670 		}
2671 		ret = kvm_hv_send_ipi(vcpu, &hc);
2672 		break;
2673 	case HVCALL_POST_DEBUG_DATA:
2674 	case HVCALL_RETRIEVE_DEBUG_DATA:
2675 		if (unlikely(hc.fast)) {
2676 			ret = HV_STATUS_INVALID_PARAMETER;
2677 			break;
2678 		}
2679 		fallthrough;
2680 	case HVCALL_RESET_DEBUG_SESSION: {
2681 		struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
2682 
2683 		if (!kvm_hv_is_syndbg_enabled(vcpu)) {
2684 			ret = HV_STATUS_INVALID_HYPERCALL_CODE;
2685 			break;
2686 		}
2687 
2688 		if (!(syndbg->options & HV_X64_SYNDBG_OPTION_USE_HCALLS)) {
2689 			ret = HV_STATUS_OPERATION_DENIED;
2690 			break;
2691 		}
2692 		goto hypercall_userspace_exit;
2693 	}
2694 	case HV_EXT_CALL_QUERY_CAPABILITIES ... HV_EXT_CALL_MAX:
2695 		if (unlikely(hc.fast)) {
2696 			ret = HV_STATUS_INVALID_PARAMETER;
2697 			break;
2698 		}
2699 		goto hypercall_userspace_exit;
2700 	default:
2701 		ret = HV_STATUS_INVALID_HYPERCALL_CODE;
2702 		break;
2703 	}
2704 
2705 hypercall_complete:
2706 	return kvm_hv_hypercall_complete(vcpu, ret);
2707 
2708 hypercall_userspace_exit:
2709 	vcpu->run->exit_reason = KVM_EXIT_HYPERV;
2710 	vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
2711 	vcpu->run->hyperv.u.hcall.input = hc.param;
2712 	vcpu->run->hyperv.u.hcall.params[0] = hc.ingpa;
2713 	vcpu->run->hyperv.u.hcall.params[1] = hc.outgpa;
2714 	vcpu->arch.complete_userspace_io = kvm_hv_hypercall_complete_userspace;
2715 	return 0;
2716 }
2717 
2718 void kvm_hv_init_vm(struct kvm *kvm)
2719 {
2720 	struct kvm_hv *hv = to_kvm_hv(kvm);
2721 
2722 	mutex_init(&hv->hv_lock);
2723 	idr_init(&hv->conn_to_evt);
2724 }
2725 
2726 void kvm_hv_destroy_vm(struct kvm *kvm)
2727 {
2728 	struct kvm_hv *hv = to_kvm_hv(kvm);
2729 	struct eventfd_ctx *eventfd;
2730 	int i;
2731 
2732 	idr_for_each_entry(&hv->conn_to_evt, eventfd, i)
2733 		eventfd_ctx_put(eventfd);
2734 	idr_destroy(&hv->conn_to_evt);
2735 }
2736 
2737 static int kvm_hv_eventfd_assign(struct kvm *kvm, u32 conn_id, int fd)
2738 {
2739 	struct kvm_hv *hv = to_kvm_hv(kvm);
2740 	struct eventfd_ctx *eventfd;
2741 	int ret;
2742 
2743 	eventfd = eventfd_ctx_fdget(fd);
2744 	if (IS_ERR(eventfd))
2745 		return PTR_ERR(eventfd);
2746 
2747 	mutex_lock(&hv->hv_lock);
2748 	ret = idr_alloc(&hv->conn_to_evt, eventfd, conn_id, conn_id + 1,
2749 			GFP_KERNEL_ACCOUNT);
2750 	mutex_unlock(&hv->hv_lock);
2751 
2752 	if (ret >= 0)
2753 		return 0;
2754 
2755 	if (ret == -ENOSPC)
2756 		ret = -EEXIST;
2757 	eventfd_ctx_put(eventfd);
2758 	return ret;
2759 }
2760 
2761 static int kvm_hv_eventfd_deassign(struct kvm *kvm, u32 conn_id)
2762 {
2763 	struct kvm_hv *hv = to_kvm_hv(kvm);
2764 	struct eventfd_ctx *eventfd;
2765 
2766 	mutex_lock(&hv->hv_lock);
2767 	eventfd = idr_remove(&hv->conn_to_evt, conn_id);
2768 	mutex_unlock(&hv->hv_lock);
2769 
2770 	if (!eventfd)
2771 		return -ENOENT;
2772 
2773 	synchronize_srcu(&kvm->srcu);
2774 	eventfd_ctx_put(eventfd);
2775 	return 0;
2776 }
2777 
2778 int kvm_vm_ioctl_hv_eventfd(struct kvm *kvm, struct kvm_hyperv_eventfd *args)
2779 {
2780 	if ((args->flags & ~KVM_HYPERV_EVENTFD_DEASSIGN) ||
2781 	    (args->conn_id & ~KVM_HYPERV_CONN_ID_MASK))
2782 		return -EINVAL;
2783 
2784 	if (args->flags == KVM_HYPERV_EVENTFD_DEASSIGN)
2785 		return kvm_hv_eventfd_deassign(kvm, args->conn_id);
2786 	return kvm_hv_eventfd_assign(kvm, args->conn_id, args->fd);
2787 }
2788 
2789 int kvm_get_hv_cpuid(struct kvm_vcpu *vcpu, struct kvm_cpuid2 *cpuid,
2790 		     struct kvm_cpuid_entry2 __user *entries)
2791 {
2792 	uint16_t evmcs_ver = 0;
2793 	struct kvm_cpuid_entry2 cpuid_entries[] = {
2794 		{ .function = HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS },
2795 		{ .function = HYPERV_CPUID_INTERFACE },
2796 		{ .function = HYPERV_CPUID_VERSION },
2797 		{ .function = HYPERV_CPUID_FEATURES },
2798 		{ .function = HYPERV_CPUID_ENLIGHTMENT_INFO },
2799 		{ .function = HYPERV_CPUID_IMPLEMENT_LIMITS },
2800 		{ .function = HYPERV_CPUID_SYNDBG_VENDOR_AND_MAX_FUNCTIONS },
2801 		{ .function = HYPERV_CPUID_SYNDBG_INTERFACE },
2802 		{ .function = HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES	},
2803 		{ .function = HYPERV_CPUID_NESTED_FEATURES },
2804 	};
2805 	int i, nent = ARRAY_SIZE(cpuid_entries);
2806 
2807 	if (kvm_nested_ops.enabled)
2808 		evmcs_ver = kvm_nested_call(get_evmcs_version)(vcpu);
2809 
2810 	if (cpuid->nent < nent)
2811 		return -E2BIG;
2812 
2813 	if (cpuid->nent > nent)
2814 		cpuid->nent = nent;
2815 
2816 	for (i = 0; i < nent; i++) {
2817 		struct kvm_cpuid_entry2 *ent = &cpuid_entries[i];
2818 		u32 signature[3];
2819 
2820 		switch (ent->function) {
2821 		case HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS:
2822 			memcpy(signature, "Linux KVM Hv", 12);
2823 
2824 			ent->eax = HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES;
2825 			ent->ebx = signature[0];
2826 			ent->ecx = signature[1];
2827 			ent->edx = signature[2];
2828 			break;
2829 
2830 		case HYPERV_CPUID_INTERFACE:
2831 			ent->eax = HYPERV_CPUID_SIGNATURE_EAX;
2832 			break;
2833 
2834 		case HYPERV_CPUID_VERSION:
2835 			/*
2836 			 * We implement some Hyper-V 2016 functions so let's use
2837 			 * this version.
2838 			 */
2839 			ent->eax = 0x00003839;
2840 			ent->ebx = 0x000A0000;
2841 			break;
2842 
2843 		case HYPERV_CPUID_FEATURES:
2844 			ent->eax |= HV_MSR_VP_RUNTIME_AVAILABLE;
2845 			ent->eax |= HV_MSR_TIME_REF_COUNT_AVAILABLE;
2846 			ent->eax |= HV_MSR_SYNIC_AVAILABLE;
2847 			ent->eax |= HV_MSR_SYNTIMER_AVAILABLE;
2848 			ent->eax |= HV_MSR_APIC_ACCESS_AVAILABLE;
2849 			ent->eax |= HV_MSR_HYPERCALL_AVAILABLE;
2850 			ent->eax |= HV_MSR_VP_INDEX_AVAILABLE;
2851 			ent->eax |= HV_MSR_RESET_AVAILABLE;
2852 			ent->eax |= HV_MSR_REFERENCE_TSC_AVAILABLE;
2853 			ent->eax |= HV_ACCESS_FREQUENCY_MSRS;
2854 			ent->eax |= HV_ACCESS_REENLIGHTENMENT;
2855 			ent->eax |= HV_ACCESS_TSC_INVARIANT;
2856 
2857 			ent->ebx |= HV_POST_MESSAGES;
2858 			ent->ebx |= HV_SIGNAL_EVENTS;
2859 			ent->ebx |= HV_ENABLE_EXTENDED_HYPERCALLS;
2860 
2861 			ent->edx |= HV_X64_HYPERCALL_XMM_INPUT_AVAILABLE;
2862 			ent->edx |= HV_FEATURE_FREQUENCY_MSRS_AVAILABLE;
2863 			ent->edx |= HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
2864 
2865 			ent->ebx |= HV_DEBUGGING;
2866 			ent->edx |= HV_X64_GUEST_DEBUGGING_AVAILABLE;
2867 			ent->edx |= HV_FEATURE_DEBUG_MSRS_AVAILABLE;
2868 			ent->edx |= HV_FEATURE_EXT_GVA_RANGES_FLUSH;
2869 
2870 			/*
2871 			 * Direct Synthetic timers only make sense with in-kernel
2872 			 * LAPIC
2873 			 */
2874 			if (!vcpu || lapic_in_kernel(vcpu))
2875 				ent->edx |= HV_STIMER_DIRECT_MODE_AVAILABLE;
2876 
2877 			break;
2878 
2879 		case HYPERV_CPUID_ENLIGHTMENT_INFO:
2880 			ent->eax |= HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
2881 			ent->eax |= HV_X64_APIC_ACCESS_RECOMMENDED;
2882 			ent->eax |= HV_X64_RELAXED_TIMING_RECOMMENDED;
2883 			if (!vcpu || lapic_in_kernel(vcpu))
2884 				ent->eax |= HV_X64_CLUSTER_IPI_RECOMMENDED;
2885 			ent->eax |= HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED;
2886 			if (evmcs_ver)
2887 				ent->eax |= HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
2888 			if (!cpu_smt_possible())
2889 				ent->eax |= HV_X64_NO_NONARCH_CORESHARING;
2890 
2891 			ent->eax |= HV_DEPRECATING_AEOI_RECOMMENDED;
2892 			/*
2893 			 * Default number of spinlock retry attempts, matches
2894 			 * HyperV 2016.
2895 			 */
2896 			ent->ebx = 0x00000FFF;
2897 
2898 			break;
2899 
2900 		case HYPERV_CPUID_IMPLEMENT_LIMITS:
2901 			/* Maximum number of virtual processors */
2902 			ent->eax = KVM_MAX_VCPUS;
2903 			/*
2904 			 * Maximum number of logical processors, matches
2905 			 * HyperV 2016.
2906 			 */
2907 			ent->ebx = 64;
2908 
2909 			break;
2910 
2911 		case HYPERV_CPUID_NESTED_FEATURES:
2912 			ent->eax = evmcs_ver;
2913 			ent->eax |= HV_X64_NESTED_DIRECT_FLUSH;
2914 			ent->eax |= HV_X64_NESTED_MSR_BITMAP;
2915 			ent->ebx |= HV_X64_NESTED_EVMCS1_PERF_GLOBAL_CTRL;
2916 			break;
2917 
2918 		case HYPERV_CPUID_SYNDBG_VENDOR_AND_MAX_FUNCTIONS:
2919 			memcpy(signature, "Linux KVM Hv", 12);
2920 
2921 			ent->eax = 0;
2922 			ent->ebx = signature[0];
2923 			ent->ecx = signature[1];
2924 			ent->edx = signature[2];
2925 			break;
2926 
2927 		case HYPERV_CPUID_SYNDBG_INTERFACE:
2928 			memcpy(signature, "VS#1\0\0\0\0\0\0\0\0", 12);
2929 			ent->eax = signature[0];
2930 			break;
2931 
2932 		case HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES:
2933 			ent->eax |= HV_X64_SYNDBG_CAP_ALLOW_KERNEL_DEBUGGING;
2934 			break;
2935 
2936 		default:
2937 			break;
2938 		}
2939 	}
2940 
2941 	if (copy_to_user(entries, cpuid_entries,
2942 			 nent * sizeof(struct kvm_cpuid_entry2)))
2943 		return -EFAULT;
2944 
2945 	return 0;
2946 }
2947