xref: /linux/arch/x86/kvm/hyperv.c (revision ba199dc909a20fe62270ae4e93f263987bb9d119)
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 *vcpu;
210 	struct kvm_vcpu_hv_synic *synic;
211 
212 	vcpu = get_vcpu_by_vpidx(kvm, vpidx);
213 	if (!vcpu || !to_hv_vcpu(vcpu))
214 		return NULL;
215 	synic = to_hv_synic(vcpu);
216 	return (synic->active) ? synic : NULL;
217 }
218 
219 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
220 {
221 	struct kvm *kvm = vcpu->kvm;
222 	struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
223 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
224 	struct kvm_vcpu_hv_stimer *stimer;
225 	int gsi, idx;
226 
227 	trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
228 
229 	/* Try to deliver pending Hyper-V SynIC timers messages */
230 	for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
231 		stimer = &hv_vcpu->stimer[idx];
232 		if (stimer->msg_pending && stimer->config.enable &&
233 		    !stimer->config.direct_mode &&
234 		    stimer->config.sintx == sint)
235 			stimer_mark_pending(stimer, false);
236 	}
237 
238 	idx = srcu_read_lock(&kvm->irq_srcu);
239 	gsi = atomic_read(&synic->sint_to_gsi[sint]);
240 	if (gsi != -1)
241 		kvm_notify_acked_gsi(kvm, gsi);
242 	srcu_read_unlock(&kvm->irq_srcu, idx);
243 }
244 
245 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
246 {
247 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
248 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
249 
250 	hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
251 	hv_vcpu->exit.u.synic.msr = msr;
252 	hv_vcpu->exit.u.synic.control = synic->control;
253 	hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
254 	hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
255 
256 	kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
257 }
258 
259 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
260 			 u32 msr, u64 data, bool host)
261 {
262 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
263 	int ret;
264 
265 	if (!synic->active && (!host || data))
266 		return 1;
267 
268 	trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
269 
270 	ret = 0;
271 	switch (msr) {
272 	case HV_X64_MSR_SCONTROL:
273 		synic->control = data;
274 		if (!host)
275 			synic_exit(synic, msr);
276 		break;
277 	case HV_X64_MSR_SVERSION:
278 		if (!host) {
279 			ret = 1;
280 			break;
281 		}
282 		synic->version = data;
283 		break;
284 	case HV_X64_MSR_SIEFP:
285 		if ((data & HV_SYNIC_SIEFP_ENABLE) && !host &&
286 		    !synic->dont_zero_synic_pages)
287 			if (kvm_clear_guest(vcpu->kvm,
288 					    data & PAGE_MASK, PAGE_SIZE)) {
289 				ret = 1;
290 				break;
291 			}
292 		synic->evt_page = data;
293 		if (!host)
294 			synic_exit(synic, msr);
295 		break;
296 	case HV_X64_MSR_SIMP:
297 		if ((data & HV_SYNIC_SIMP_ENABLE) && !host &&
298 		    !synic->dont_zero_synic_pages)
299 			if (kvm_clear_guest(vcpu->kvm,
300 					    data & PAGE_MASK, PAGE_SIZE)) {
301 				ret = 1;
302 				break;
303 			}
304 		synic->msg_page = data;
305 		if (!host)
306 			synic_exit(synic, msr);
307 		break;
308 	case HV_X64_MSR_EOM: {
309 		int i;
310 
311 		if (!synic->active)
312 			break;
313 
314 		for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
315 			kvm_hv_notify_acked_sint(vcpu, i);
316 		break;
317 	}
318 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
319 		ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
320 		break;
321 	default:
322 		ret = 1;
323 		break;
324 	}
325 	return ret;
326 }
327 
328 static bool kvm_hv_is_syndbg_enabled(struct kvm_vcpu *vcpu)
329 {
330 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
331 
332 	return hv_vcpu->cpuid_cache.syndbg_cap_eax &
333 		HV_X64_SYNDBG_CAP_ALLOW_KERNEL_DEBUGGING;
334 }
335 
336 static int kvm_hv_syndbg_complete_userspace(struct kvm_vcpu *vcpu)
337 {
338 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
339 
340 	if (vcpu->run->hyperv.u.syndbg.msr == HV_X64_MSR_SYNDBG_CONTROL)
341 		hv->hv_syndbg.control.status =
342 			vcpu->run->hyperv.u.syndbg.status;
343 	return 1;
344 }
345 
346 static void syndbg_exit(struct kvm_vcpu *vcpu, u32 msr)
347 {
348 	struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
349 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
350 
351 	hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNDBG;
352 	hv_vcpu->exit.u.syndbg.msr = msr;
353 	hv_vcpu->exit.u.syndbg.control = syndbg->control.control;
354 	hv_vcpu->exit.u.syndbg.send_page = syndbg->control.send_page;
355 	hv_vcpu->exit.u.syndbg.recv_page = syndbg->control.recv_page;
356 	hv_vcpu->exit.u.syndbg.pending_page = syndbg->control.pending_page;
357 	vcpu->arch.complete_userspace_io =
358 			kvm_hv_syndbg_complete_userspace;
359 
360 	kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
361 }
362 
363 static int syndbg_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
364 {
365 	struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
366 
367 	if (!kvm_hv_is_syndbg_enabled(vcpu) && !host)
368 		return 1;
369 
370 	trace_kvm_hv_syndbg_set_msr(vcpu->vcpu_id,
371 				    to_hv_vcpu(vcpu)->vp_index, msr, data);
372 	switch (msr) {
373 	case HV_X64_MSR_SYNDBG_CONTROL:
374 		syndbg->control.control = data;
375 		if (!host)
376 			syndbg_exit(vcpu, msr);
377 		break;
378 	case HV_X64_MSR_SYNDBG_STATUS:
379 		syndbg->control.status = data;
380 		break;
381 	case HV_X64_MSR_SYNDBG_SEND_BUFFER:
382 		syndbg->control.send_page = data;
383 		break;
384 	case HV_X64_MSR_SYNDBG_RECV_BUFFER:
385 		syndbg->control.recv_page = data;
386 		break;
387 	case HV_X64_MSR_SYNDBG_PENDING_BUFFER:
388 		syndbg->control.pending_page = data;
389 		if (!host)
390 			syndbg_exit(vcpu, msr);
391 		break;
392 	case HV_X64_MSR_SYNDBG_OPTIONS:
393 		syndbg->options = data;
394 		break;
395 	default:
396 		break;
397 	}
398 
399 	return 0;
400 }
401 
402 static int syndbg_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
403 {
404 	struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
405 
406 	if (!kvm_hv_is_syndbg_enabled(vcpu) && !host)
407 		return 1;
408 
409 	switch (msr) {
410 	case HV_X64_MSR_SYNDBG_CONTROL:
411 		*pdata = syndbg->control.control;
412 		break;
413 	case HV_X64_MSR_SYNDBG_STATUS:
414 		*pdata = syndbg->control.status;
415 		break;
416 	case HV_X64_MSR_SYNDBG_SEND_BUFFER:
417 		*pdata = syndbg->control.send_page;
418 		break;
419 	case HV_X64_MSR_SYNDBG_RECV_BUFFER:
420 		*pdata = syndbg->control.recv_page;
421 		break;
422 	case HV_X64_MSR_SYNDBG_PENDING_BUFFER:
423 		*pdata = syndbg->control.pending_page;
424 		break;
425 	case HV_X64_MSR_SYNDBG_OPTIONS:
426 		*pdata = syndbg->options;
427 		break;
428 	default:
429 		break;
430 	}
431 
432 	trace_kvm_hv_syndbg_get_msr(vcpu->vcpu_id, kvm_hv_get_vpindex(vcpu), msr, *pdata);
433 
434 	return 0;
435 }
436 
437 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata,
438 			 bool host)
439 {
440 	int ret;
441 
442 	if (!synic->active && !host)
443 		return 1;
444 
445 	ret = 0;
446 	switch (msr) {
447 	case HV_X64_MSR_SCONTROL:
448 		*pdata = synic->control;
449 		break;
450 	case HV_X64_MSR_SVERSION:
451 		*pdata = synic->version;
452 		break;
453 	case HV_X64_MSR_SIEFP:
454 		*pdata = synic->evt_page;
455 		break;
456 	case HV_X64_MSR_SIMP:
457 		*pdata = synic->msg_page;
458 		break;
459 	case HV_X64_MSR_EOM:
460 		*pdata = 0;
461 		break;
462 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
463 		*pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
464 		break;
465 	default:
466 		ret = 1;
467 		break;
468 	}
469 	return ret;
470 }
471 
472 static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
473 {
474 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
475 	struct kvm_lapic_irq irq;
476 	int ret, vector;
477 
478 	if (KVM_BUG_ON(!lapic_in_kernel(vcpu), vcpu->kvm))
479 		return -EINVAL;
480 
481 	if (sint >= ARRAY_SIZE(synic->sint))
482 		return -EINVAL;
483 
484 	vector = synic_get_sint_vector(synic_read_sint(synic, sint));
485 	if (vector < 0)
486 		return -ENOENT;
487 
488 	memset(&irq, 0, sizeof(irq));
489 	irq.shorthand = APIC_DEST_SELF;
490 	irq.dest_mode = APIC_DEST_PHYSICAL;
491 	irq.delivery_mode = APIC_DM_FIXED;
492 	irq.vector = vector;
493 	irq.level = 1;
494 
495 	ret = kvm_irq_delivery_to_apic(vcpu->kvm, vcpu->arch.apic, &irq, NULL);
496 	trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
497 	return ret;
498 }
499 
500 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vpidx, u32 sint)
501 {
502 	struct kvm_vcpu_hv_synic *synic;
503 
504 	synic = synic_get(kvm, vpidx);
505 	if (!synic)
506 		return -EINVAL;
507 
508 	return synic_set_irq(synic, sint);
509 }
510 
511 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
512 {
513 	struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
514 	int i;
515 
516 	trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
517 
518 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
519 		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
520 			kvm_hv_notify_acked_sint(vcpu, i);
521 }
522 
523 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi)
524 {
525 	struct kvm_vcpu_hv_synic *synic;
526 
527 	synic = synic_get(kvm, vpidx);
528 	if (!synic)
529 		return -EINVAL;
530 
531 	if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
532 		return -EINVAL;
533 
534 	atomic_set(&synic->sint_to_gsi[sint], gsi);
535 	return 0;
536 }
537 
538 void kvm_hv_irq_routing_update(struct kvm *kvm)
539 {
540 	struct kvm_irq_routing_table *irq_rt;
541 	struct kvm_kernel_irq_routing_entry *e;
542 	u32 gsi;
543 
544 	irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
545 					lockdep_is_held(&kvm->irq_lock));
546 
547 	for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
548 		hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
549 			if (e->type == KVM_IRQ_ROUTING_HV_SINT)
550 				kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
551 						    e->hv_sint.sint, gsi);
552 		}
553 	}
554 }
555 
556 static void synic_init(struct kvm_vcpu_hv_synic *synic)
557 {
558 	int i;
559 
560 	memset(synic, 0, sizeof(*synic));
561 	synic->version = HV_SYNIC_VERSION_1;
562 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
563 		atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
564 		atomic_set(&synic->sint_to_gsi[i], -1);
565 	}
566 }
567 
568 static u64 get_time_ref_counter(struct kvm *kvm)
569 {
570 	struct kvm_hv *hv = to_kvm_hv(kvm);
571 	struct kvm_vcpu *vcpu;
572 	u64 tsc;
573 
574 	/*
575 	 * Fall back to get_kvmclock_ns() when TSC page hasn't been set up,
576 	 * is broken, disabled or being updated.
577 	 */
578 	if (hv->hv_tsc_page_status != HV_TSC_PAGE_SET)
579 		return div_u64(get_kvmclock_ns(kvm), 100);
580 
581 	vcpu = kvm_get_vcpu(kvm, 0);
582 	tsc = kvm_read_l1_tsc(vcpu, rdtsc());
583 	return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64)
584 		+ hv->tsc_ref.tsc_offset;
585 }
586 
587 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
588 				bool vcpu_kick)
589 {
590 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
591 
592 	set_bit(stimer->index,
593 		to_hv_vcpu(vcpu)->stimer_pending_bitmap);
594 	kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
595 	if (vcpu_kick)
596 		kvm_vcpu_kick(vcpu);
597 }
598 
599 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
600 {
601 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
602 
603 	trace_kvm_hv_stimer_cleanup(hv_stimer_to_vcpu(stimer)->vcpu_id,
604 				    stimer->index);
605 
606 	hrtimer_cancel(&stimer->timer);
607 	clear_bit(stimer->index,
608 		  to_hv_vcpu(vcpu)->stimer_pending_bitmap);
609 	stimer->msg_pending = false;
610 	stimer->exp_time = 0;
611 }
612 
613 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
614 {
615 	struct kvm_vcpu_hv_stimer *stimer;
616 
617 	stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
618 	trace_kvm_hv_stimer_callback(hv_stimer_to_vcpu(stimer)->vcpu_id,
619 				     stimer->index);
620 	stimer_mark_pending(stimer, true);
621 
622 	return HRTIMER_NORESTART;
623 }
624 
625 /*
626  * stimer_start() assumptions:
627  * a) stimer->count is not equal to 0
628  * b) stimer->config has HV_STIMER_ENABLE flag
629  */
630 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
631 {
632 	u64 time_now;
633 	ktime_t ktime_now;
634 
635 	time_now = get_time_ref_counter(hv_stimer_to_vcpu(stimer)->kvm);
636 	ktime_now = ktime_get();
637 
638 	if (stimer->config.periodic) {
639 		if (stimer->exp_time) {
640 			if (time_now >= stimer->exp_time) {
641 				u64 remainder;
642 
643 				div64_u64_rem(time_now - stimer->exp_time,
644 					      stimer->count, &remainder);
645 				stimer->exp_time =
646 					time_now + (stimer->count - remainder);
647 			}
648 		} else
649 			stimer->exp_time = time_now + stimer->count;
650 
651 		trace_kvm_hv_stimer_start_periodic(
652 					hv_stimer_to_vcpu(stimer)->vcpu_id,
653 					stimer->index,
654 					time_now, stimer->exp_time);
655 
656 		hrtimer_start(&stimer->timer,
657 			      ktime_add_ns(ktime_now,
658 					   100 * (stimer->exp_time - time_now)),
659 			      HRTIMER_MODE_ABS);
660 		return 0;
661 	}
662 	stimer->exp_time = stimer->count;
663 	if (time_now >= stimer->count) {
664 		/*
665 		 * Expire timer according to Hypervisor Top-Level Functional
666 		 * specification v4(15.3.1):
667 		 * "If a one shot is enabled and the specified count is in
668 		 * the past, it will expire immediately."
669 		 */
670 		stimer_mark_pending(stimer, false);
671 		return 0;
672 	}
673 
674 	trace_kvm_hv_stimer_start_one_shot(hv_stimer_to_vcpu(stimer)->vcpu_id,
675 					   stimer->index,
676 					   time_now, stimer->count);
677 
678 	hrtimer_start(&stimer->timer,
679 		      ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)),
680 		      HRTIMER_MODE_ABS);
681 	return 0;
682 }
683 
684 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
685 			     bool host)
686 {
687 	union hv_stimer_config new_config = {.as_uint64 = config},
688 		old_config = {.as_uint64 = stimer->config.as_uint64};
689 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
690 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
691 	struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
692 
693 	if (!synic->active && (!host || config))
694 		return 1;
695 
696 	if (unlikely(!host && hv_vcpu->enforce_cpuid && new_config.direct_mode &&
697 		     !(hv_vcpu->cpuid_cache.features_edx &
698 		       HV_STIMER_DIRECT_MODE_AVAILABLE)))
699 		return 1;
700 
701 	trace_kvm_hv_stimer_set_config(hv_stimer_to_vcpu(stimer)->vcpu_id,
702 				       stimer->index, config, host);
703 
704 	stimer_cleanup(stimer);
705 	if (old_config.enable &&
706 	    !new_config.direct_mode && new_config.sintx == 0)
707 		new_config.enable = 0;
708 	stimer->config.as_uint64 = new_config.as_uint64;
709 
710 	if (stimer->config.enable)
711 		stimer_mark_pending(stimer, false);
712 
713 	return 0;
714 }
715 
716 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
717 			    bool host)
718 {
719 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
720 	struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
721 
722 	if (!synic->active && (!host || count))
723 		return 1;
724 
725 	trace_kvm_hv_stimer_set_count(hv_stimer_to_vcpu(stimer)->vcpu_id,
726 				      stimer->index, count, host);
727 
728 	stimer_cleanup(stimer);
729 	stimer->count = count;
730 	if (!host) {
731 		if (stimer->count == 0)
732 			stimer->config.enable = 0;
733 		else if (stimer->config.auto_enable)
734 			stimer->config.enable = 1;
735 	}
736 
737 	if (stimer->config.enable)
738 		stimer_mark_pending(stimer, false);
739 
740 	return 0;
741 }
742 
743 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
744 {
745 	*pconfig = stimer->config.as_uint64;
746 	return 0;
747 }
748 
749 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
750 {
751 	*pcount = stimer->count;
752 	return 0;
753 }
754 
755 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
756 			     struct hv_message *src_msg, bool no_retry)
757 {
758 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
759 	int msg_off = offsetof(struct hv_message_page, sint_message[sint]);
760 	gfn_t msg_page_gfn;
761 	struct hv_message_header hv_hdr;
762 	int r;
763 
764 	if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
765 		return -ENOENT;
766 
767 	msg_page_gfn = synic->msg_page >> PAGE_SHIFT;
768 
769 	/*
770 	 * Strictly following the spec-mandated ordering would assume setting
771 	 * .msg_pending before checking .message_type.  However, this function
772 	 * is only called in vcpu context so the entire update is atomic from
773 	 * guest POV and thus the exact order here doesn't matter.
774 	 */
775 	r = kvm_vcpu_read_guest_page(vcpu, msg_page_gfn, &hv_hdr.message_type,
776 				     msg_off + offsetof(struct hv_message,
777 							header.message_type),
778 				     sizeof(hv_hdr.message_type));
779 	if (r < 0)
780 		return r;
781 
782 	if (hv_hdr.message_type != HVMSG_NONE) {
783 		if (no_retry)
784 			return 0;
785 
786 		hv_hdr.message_flags.msg_pending = 1;
787 		r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn,
788 					      &hv_hdr.message_flags,
789 					      msg_off +
790 					      offsetof(struct hv_message,
791 						       header.message_flags),
792 					      sizeof(hv_hdr.message_flags));
793 		if (r < 0)
794 			return r;
795 		return -EAGAIN;
796 	}
797 
798 	r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn, src_msg, msg_off,
799 				      sizeof(src_msg->header) +
800 				      src_msg->header.payload_size);
801 	if (r < 0)
802 		return r;
803 
804 	r = synic_set_irq(synic, sint);
805 	if (r < 0)
806 		return r;
807 	if (r == 0)
808 		return -EFAULT;
809 	return 0;
810 }
811 
812 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
813 {
814 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
815 	struct hv_message *msg = &stimer->msg;
816 	struct hv_timer_message_payload *payload =
817 			(struct hv_timer_message_payload *)&msg->u.payload;
818 
819 	/*
820 	 * To avoid piling up periodic ticks, don't retry message
821 	 * delivery for them (within "lazy" lost ticks policy).
822 	 */
823 	bool no_retry = stimer->config.periodic;
824 
825 	payload->expiration_time = stimer->exp_time;
826 	payload->delivery_time = get_time_ref_counter(vcpu->kvm);
827 	return synic_deliver_msg(to_hv_synic(vcpu),
828 				 stimer->config.sintx, msg,
829 				 no_retry);
830 }
831 
832 static int stimer_notify_direct(struct kvm_vcpu_hv_stimer *stimer)
833 {
834 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
835 	struct kvm_lapic_irq irq = {
836 		.delivery_mode = APIC_DM_FIXED,
837 		.vector = stimer->config.apic_vector
838 	};
839 
840 	if (lapic_in_kernel(vcpu))
841 		return !kvm_apic_set_irq(vcpu, &irq, NULL);
842 	return 0;
843 }
844 
845 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
846 {
847 	int r, direct = stimer->config.direct_mode;
848 
849 	stimer->msg_pending = true;
850 	if (!direct)
851 		r = stimer_send_msg(stimer);
852 	else
853 		r = stimer_notify_direct(stimer);
854 	trace_kvm_hv_stimer_expiration(hv_stimer_to_vcpu(stimer)->vcpu_id,
855 				       stimer->index, direct, r);
856 	if (!r) {
857 		stimer->msg_pending = false;
858 		if (!(stimer->config.periodic))
859 			stimer->config.enable = 0;
860 	}
861 }
862 
863 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
864 {
865 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
866 	struct kvm_vcpu_hv_stimer *stimer;
867 	u64 time_now, exp_time;
868 	int i;
869 
870 	if (!hv_vcpu)
871 		return;
872 
873 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
874 		if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
875 			stimer = &hv_vcpu->stimer[i];
876 			if (stimer->config.enable) {
877 				exp_time = stimer->exp_time;
878 
879 				if (exp_time) {
880 					time_now =
881 						get_time_ref_counter(vcpu->kvm);
882 					if (time_now >= exp_time)
883 						stimer_expiration(stimer);
884 				}
885 
886 				if ((stimer->config.enable) &&
887 				    stimer->count) {
888 					if (!stimer->msg_pending)
889 						stimer_start(stimer);
890 				} else
891 					stimer_cleanup(stimer);
892 			}
893 		}
894 }
895 
896 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
897 {
898 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
899 	int i;
900 
901 	if (!hv_vcpu)
902 		return;
903 
904 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
905 		stimer_cleanup(&hv_vcpu->stimer[i]);
906 
907 	kfree(hv_vcpu);
908 	vcpu->arch.hyperv = NULL;
909 }
910 
911 bool kvm_hv_assist_page_enabled(struct kvm_vcpu *vcpu)
912 {
913 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
914 
915 	if (!hv_vcpu)
916 		return false;
917 
918 	if (!(hv_vcpu->hv_vapic & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE))
919 		return false;
920 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
921 }
922 EXPORT_SYMBOL_GPL(kvm_hv_assist_page_enabled);
923 
924 int kvm_hv_get_assist_page(struct kvm_vcpu *vcpu)
925 {
926 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
927 
928 	if (!hv_vcpu || !kvm_hv_assist_page_enabled(vcpu))
929 		return -EFAULT;
930 
931 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data,
932 				     &hv_vcpu->vp_assist_page, sizeof(struct hv_vp_assist_page));
933 }
934 EXPORT_SYMBOL_GPL(kvm_hv_get_assist_page);
935 
936 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
937 {
938 	struct hv_message *msg = &stimer->msg;
939 	struct hv_timer_message_payload *payload =
940 			(struct hv_timer_message_payload *)&msg->u.payload;
941 
942 	memset(&msg->header, 0, sizeof(msg->header));
943 	msg->header.message_type = HVMSG_TIMER_EXPIRED;
944 	msg->header.payload_size = sizeof(*payload);
945 
946 	payload->timer_index = stimer->index;
947 	payload->expiration_time = 0;
948 	payload->delivery_time = 0;
949 }
950 
951 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
952 {
953 	memset(stimer, 0, sizeof(*stimer));
954 	stimer->index = timer_index;
955 	hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
956 	stimer->timer.function = stimer_timer_callback;
957 	stimer_prepare_msg(stimer);
958 }
959 
960 int kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
961 {
962 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
963 	int i;
964 
965 	if (hv_vcpu)
966 		return 0;
967 
968 	hv_vcpu = kzalloc(sizeof(struct kvm_vcpu_hv), GFP_KERNEL_ACCOUNT);
969 	if (!hv_vcpu)
970 		return -ENOMEM;
971 
972 	vcpu->arch.hyperv = hv_vcpu;
973 	hv_vcpu->vcpu = vcpu;
974 
975 	synic_init(&hv_vcpu->synic);
976 
977 	bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
978 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
979 		stimer_init(&hv_vcpu->stimer[i], i);
980 
981 	hv_vcpu->vp_index = vcpu->vcpu_idx;
982 
983 	for (i = 0; i < HV_NR_TLB_FLUSH_FIFOS; i++) {
984 		INIT_KFIFO(hv_vcpu->tlb_flush_fifo[i].entries);
985 		spin_lock_init(&hv_vcpu->tlb_flush_fifo[i].write_lock);
986 	}
987 
988 	return 0;
989 }
990 
991 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
992 {
993 	struct kvm_vcpu_hv_synic *synic;
994 	int r;
995 
996 	r = kvm_hv_vcpu_init(vcpu);
997 	if (r)
998 		return r;
999 
1000 	synic = to_hv_synic(vcpu);
1001 
1002 	synic->active = true;
1003 	synic->dont_zero_synic_pages = dont_zero_synic_pages;
1004 	synic->control = HV_SYNIC_CONTROL_ENABLE;
1005 	return 0;
1006 }
1007 
1008 static bool kvm_hv_msr_partition_wide(u32 msr)
1009 {
1010 	bool r = false;
1011 
1012 	switch (msr) {
1013 	case HV_X64_MSR_GUEST_OS_ID:
1014 	case HV_X64_MSR_HYPERCALL:
1015 	case HV_X64_MSR_REFERENCE_TSC:
1016 	case HV_X64_MSR_TIME_REF_COUNT:
1017 	case HV_X64_MSR_CRASH_CTL:
1018 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1019 	case HV_X64_MSR_RESET:
1020 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1021 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1022 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1023 	case HV_X64_MSR_TSC_INVARIANT_CONTROL:
1024 	case HV_X64_MSR_SYNDBG_OPTIONS:
1025 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1026 		r = true;
1027 		break;
1028 	}
1029 
1030 	return r;
1031 }
1032 
1033 static int kvm_hv_msr_get_crash_data(struct kvm *kvm, u32 index, u64 *pdata)
1034 {
1035 	struct kvm_hv *hv = to_kvm_hv(kvm);
1036 	size_t size = ARRAY_SIZE(hv->hv_crash_param);
1037 
1038 	if (WARN_ON_ONCE(index >= size))
1039 		return -EINVAL;
1040 
1041 	*pdata = hv->hv_crash_param[array_index_nospec(index, size)];
1042 	return 0;
1043 }
1044 
1045 static int kvm_hv_msr_get_crash_ctl(struct kvm *kvm, u64 *pdata)
1046 {
1047 	struct kvm_hv *hv = to_kvm_hv(kvm);
1048 
1049 	*pdata = hv->hv_crash_ctl;
1050 	return 0;
1051 }
1052 
1053 static int kvm_hv_msr_set_crash_ctl(struct kvm *kvm, u64 data)
1054 {
1055 	struct kvm_hv *hv = to_kvm_hv(kvm);
1056 
1057 	hv->hv_crash_ctl = data & HV_CRASH_CTL_CRASH_NOTIFY;
1058 
1059 	return 0;
1060 }
1061 
1062 static int kvm_hv_msr_set_crash_data(struct kvm *kvm, u32 index, u64 data)
1063 {
1064 	struct kvm_hv *hv = to_kvm_hv(kvm);
1065 	size_t size = ARRAY_SIZE(hv->hv_crash_param);
1066 
1067 	if (WARN_ON_ONCE(index >= size))
1068 		return -EINVAL;
1069 
1070 	hv->hv_crash_param[array_index_nospec(index, size)] = data;
1071 	return 0;
1072 }
1073 
1074 /*
1075  * The kvmclock and Hyper-V TSC page use similar formulas, and converting
1076  * between them is possible:
1077  *
1078  * kvmclock formula:
1079  *    nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
1080  *           + system_time
1081  *
1082  * Hyper-V formula:
1083  *    nsec/100 = ticks * scale / 2^64 + offset
1084  *
1085  * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
1086  * By dividing the kvmclock formula by 100 and equating what's left we get:
1087  *    ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1088  *            scale / 2^64 =         tsc_to_system_mul * 2^(tsc_shift-32) / 100
1089  *            scale        =         tsc_to_system_mul * 2^(32+tsc_shift) / 100
1090  *
1091  * Now expand the kvmclock formula and divide by 100:
1092  *    nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
1093  *           - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
1094  *           + system_time
1095  *    nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1096  *               - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1097  *               + system_time / 100
1098  *
1099  * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
1100  *    nsec/100 = ticks * scale / 2^64
1101  *               - tsc_timestamp * scale / 2^64
1102  *               + system_time / 100
1103  *
1104  * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
1105  *    offset = system_time / 100 - tsc_timestamp * scale / 2^64
1106  *
1107  * These two equivalencies are implemented in this function.
1108  */
1109 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
1110 					struct ms_hyperv_tsc_page *tsc_ref)
1111 {
1112 	u64 max_mul;
1113 
1114 	if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
1115 		return false;
1116 
1117 	/*
1118 	 * check if scale would overflow, if so we use the time ref counter
1119 	 *    tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
1120 	 *    tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
1121 	 *    tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
1122 	 */
1123 	max_mul = 100ull << (32 - hv_clock->tsc_shift);
1124 	if (hv_clock->tsc_to_system_mul >= max_mul)
1125 		return false;
1126 
1127 	/*
1128 	 * Otherwise compute the scale and offset according to the formulas
1129 	 * derived above.
1130 	 */
1131 	tsc_ref->tsc_scale =
1132 		mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
1133 				hv_clock->tsc_to_system_mul,
1134 				100);
1135 
1136 	tsc_ref->tsc_offset = hv_clock->system_time;
1137 	do_div(tsc_ref->tsc_offset, 100);
1138 	tsc_ref->tsc_offset -=
1139 		mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
1140 	return true;
1141 }
1142 
1143 /*
1144  * Don't touch TSC page values if the guest has opted for TSC emulation after
1145  * migration. KVM doesn't fully support reenlightenment notifications and TSC
1146  * access emulation and Hyper-V is known to expect the values in TSC page to
1147  * stay constant before TSC access emulation is disabled from guest side
1148  * (HV_X64_MSR_TSC_EMULATION_STATUS). KVM userspace is expected to preserve TSC
1149  * frequency and guest visible TSC value across migration (and prevent it when
1150  * TSC scaling is unsupported).
1151  */
1152 static inline bool tsc_page_update_unsafe(struct kvm_hv *hv)
1153 {
1154 	return (hv->hv_tsc_page_status != HV_TSC_PAGE_GUEST_CHANGED) &&
1155 		hv->hv_tsc_emulation_control;
1156 }
1157 
1158 void kvm_hv_setup_tsc_page(struct kvm *kvm,
1159 			   struct pvclock_vcpu_time_info *hv_clock)
1160 {
1161 	struct kvm_hv *hv = to_kvm_hv(kvm);
1162 	u32 tsc_seq;
1163 	u64 gfn;
1164 
1165 	BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
1166 	BUILD_BUG_ON(offsetof(struct ms_hyperv_tsc_page, tsc_sequence) != 0);
1167 
1168 	mutex_lock(&hv->hv_lock);
1169 
1170 	if (hv->hv_tsc_page_status == HV_TSC_PAGE_BROKEN ||
1171 	    hv->hv_tsc_page_status == HV_TSC_PAGE_SET ||
1172 	    hv->hv_tsc_page_status == HV_TSC_PAGE_UNSET)
1173 		goto out_unlock;
1174 
1175 	if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1176 		goto out_unlock;
1177 
1178 	gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
1179 	/*
1180 	 * Because the TSC parameters only vary when there is a
1181 	 * change in the master clock, do not bother with caching.
1182 	 */
1183 	if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
1184 				    &tsc_seq, sizeof(tsc_seq))))
1185 		goto out_err;
1186 
1187 	if (tsc_seq && tsc_page_update_unsafe(hv)) {
1188 		if (kvm_read_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
1189 			goto out_err;
1190 
1191 		hv->hv_tsc_page_status = HV_TSC_PAGE_SET;
1192 		goto out_unlock;
1193 	}
1194 
1195 	/*
1196 	 * While we're computing and writing the parameters, force the
1197 	 * guest to use the time reference count MSR.
1198 	 */
1199 	hv->tsc_ref.tsc_sequence = 0;
1200 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
1201 			    &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
1202 		goto out_err;
1203 
1204 	if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
1205 		goto out_err;
1206 
1207 	/* Ensure sequence is zero before writing the rest of the struct.  */
1208 	smp_wmb();
1209 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
1210 		goto out_err;
1211 
1212 	/*
1213 	 * Now switch to the TSC page mechanism by writing the sequence.
1214 	 */
1215 	tsc_seq++;
1216 	if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
1217 		tsc_seq = 1;
1218 
1219 	/* Write the struct entirely before the non-zero sequence.  */
1220 	smp_wmb();
1221 
1222 	hv->tsc_ref.tsc_sequence = tsc_seq;
1223 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
1224 			    &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
1225 		goto out_err;
1226 
1227 	hv->hv_tsc_page_status = HV_TSC_PAGE_SET;
1228 	goto out_unlock;
1229 
1230 out_err:
1231 	hv->hv_tsc_page_status = HV_TSC_PAGE_BROKEN;
1232 out_unlock:
1233 	mutex_unlock(&hv->hv_lock);
1234 }
1235 
1236 void kvm_hv_request_tsc_page_update(struct kvm *kvm)
1237 {
1238 	struct kvm_hv *hv = to_kvm_hv(kvm);
1239 
1240 	mutex_lock(&hv->hv_lock);
1241 
1242 	if (hv->hv_tsc_page_status == HV_TSC_PAGE_SET &&
1243 	    !tsc_page_update_unsafe(hv))
1244 		hv->hv_tsc_page_status = HV_TSC_PAGE_HOST_CHANGED;
1245 
1246 	mutex_unlock(&hv->hv_lock);
1247 }
1248 
1249 static bool hv_check_msr_access(struct kvm_vcpu_hv *hv_vcpu, u32 msr)
1250 {
1251 	if (!hv_vcpu->enforce_cpuid)
1252 		return true;
1253 
1254 	switch (msr) {
1255 	case HV_X64_MSR_GUEST_OS_ID:
1256 	case HV_X64_MSR_HYPERCALL:
1257 		return hv_vcpu->cpuid_cache.features_eax &
1258 			HV_MSR_HYPERCALL_AVAILABLE;
1259 	case HV_X64_MSR_VP_RUNTIME:
1260 		return hv_vcpu->cpuid_cache.features_eax &
1261 			HV_MSR_VP_RUNTIME_AVAILABLE;
1262 	case HV_X64_MSR_TIME_REF_COUNT:
1263 		return hv_vcpu->cpuid_cache.features_eax &
1264 			HV_MSR_TIME_REF_COUNT_AVAILABLE;
1265 	case HV_X64_MSR_VP_INDEX:
1266 		return hv_vcpu->cpuid_cache.features_eax &
1267 			HV_MSR_VP_INDEX_AVAILABLE;
1268 	case HV_X64_MSR_RESET:
1269 		return hv_vcpu->cpuid_cache.features_eax &
1270 			HV_MSR_RESET_AVAILABLE;
1271 	case HV_X64_MSR_REFERENCE_TSC:
1272 		return hv_vcpu->cpuid_cache.features_eax &
1273 			HV_MSR_REFERENCE_TSC_AVAILABLE;
1274 	case HV_X64_MSR_SCONTROL:
1275 	case HV_X64_MSR_SVERSION:
1276 	case HV_X64_MSR_SIEFP:
1277 	case HV_X64_MSR_SIMP:
1278 	case HV_X64_MSR_EOM:
1279 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1280 		return hv_vcpu->cpuid_cache.features_eax &
1281 			HV_MSR_SYNIC_AVAILABLE;
1282 	case HV_X64_MSR_STIMER0_CONFIG:
1283 	case HV_X64_MSR_STIMER1_CONFIG:
1284 	case HV_X64_MSR_STIMER2_CONFIG:
1285 	case HV_X64_MSR_STIMER3_CONFIG:
1286 	case HV_X64_MSR_STIMER0_COUNT:
1287 	case HV_X64_MSR_STIMER1_COUNT:
1288 	case HV_X64_MSR_STIMER2_COUNT:
1289 	case HV_X64_MSR_STIMER3_COUNT:
1290 		return hv_vcpu->cpuid_cache.features_eax &
1291 			HV_MSR_SYNTIMER_AVAILABLE;
1292 	case HV_X64_MSR_EOI:
1293 	case HV_X64_MSR_ICR:
1294 	case HV_X64_MSR_TPR:
1295 	case HV_X64_MSR_VP_ASSIST_PAGE:
1296 		return hv_vcpu->cpuid_cache.features_eax &
1297 			HV_MSR_APIC_ACCESS_AVAILABLE;
1298 	case HV_X64_MSR_TSC_FREQUENCY:
1299 	case HV_X64_MSR_APIC_FREQUENCY:
1300 		return hv_vcpu->cpuid_cache.features_eax &
1301 			HV_ACCESS_FREQUENCY_MSRS;
1302 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1303 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1304 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1305 		return hv_vcpu->cpuid_cache.features_eax &
1306 			HV_ACCESS_REENLIGHTENMENT;
1307 	case HV_X64_MSR_TSC_INVARIANT_CONTROL:
1308 		return hv_vcpu->cpuid_cache.features_eax &
1309 			HV_ACCESS_TSC_INVARIANT;
1310 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1311 	case HV_X64_MSR_CRASH_CTL:
1312 		return hv_vcpu->cpuid_cache.features_edx &
1313 			HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
1314 	case HV_X64_MSR_SYNDBG_OPTIONS:
1315 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1316 		return hv_vcpu->cpuid_cache.features_edx &
1317 			HV_FEATURE_DEBUG_MSRS_AVAILABLE;
1318 	default:
1319 		break;
1320 	}
1321 
1322 	return false;
1323 }
1324 
1325 #define KVM_HV_WIN2016_GUEST_ID 0x1040a00003839
1326 #define KVM_HV_WIN2016_GUEST_ID_MASK (~GENMASK_ULL(23, 16)) /* mask out the service version */
1327 
1328 /*
1329  * Hyper-V enabled Windows Server 2016 SMP VMs fail to boot in !XSAVES && XSAVEC
1330  * configuration.
1331  * Such configuration can result from, for example, AMD Erratum 1386 workaround.
1332  *
1333  * Print a notice so users aren't left wondering what's suddenly gone wrong.
1334  */
1335 static void __kvm_hv_xsaves_xsavec_maybe_warn(struct kvm_vcpu *vcpu)
1336 {
1337 	struct kvm *kvm = vcpu->kvm;
1338 	struct kvm_hv *hv = to_kvm_hv(kvm);
1339 
1340 	/* Check again under the hv_lock.  */
1341 	if (hv->xsaves_xsavec_checked)
1342 		return;
1343 
1344 	if ((hv->hv_guest_os_id & KVM_HV_WIN2016_GUEST_ID_MASK) !=
1345 	    KVM_HV_WIN2016_GUEST_ID)
1346 		return;
1347 
1348 	hv->xsaves_xsavec_checked = true;
1349 
1350 	/* UP configurations aren't affected */
1351 	if (atomic_read(&kvm->online_vcpus) < 2)
1352 		return;
1353 
1354 	if (guest_cpuid_has(vcpu, X86_FEATURE_XSAVES) ||
1355 	    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVEC))
1356 		return;
1357 
1358 	pr_notice_ratelimited("Booting SMP Windows KVM VM with !XSAVES && XSAVEC. "
1359 			      "If it fails to boot try disabling XSAVEC in the VM config.\n");
1360 }
1361 
1362 void kvm_hv_xsaves_xsavec_maybe_warn(struct kvm_vcpu *vcpu)
1363 {
1364 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1365 
1366 	if (!vcpu->arch.hyperv_enabled ||
1367 	    hv->xsaves_xsavec_checked)
1368 		return;
1369 
1370 	mutex_lock(&hv->hv_lock);
1371 	__kvm_hv_xsaves_xsavec_maybe_warn(vcpu);
1372 	mutex_unlock(&hv->hv_lock);
1373 }
1374 
1375 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
1376 			     bool host)
1377 {
1378 	struct kvm *kvm = vcpu->kvm;
1379 	struct kvm_hv *hv = to_kvm_hv(kvm);
1380 
1381 	if (unlikely(!host && !hv_check_msr_access(to_hv_vcpu(vcpu), msr)))
1382 		return 1;
1383 
1384 	switch (msr) {
1385 	case HV_X64_MSR_GUEST_OS_ID:
1386 		hv->hv_guest_os_id = data;
1387 		/* setting guest os id to zero disables hypercall page */
1388 		if (!hv->hv_guest_os_id)
1389 			hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1390 		break;
1391 	case HV_X64_MSR_HYPERCALL: {
1392 		u8 instructions[9];
1393 		int i = 0;
1394 		u64 addr;
1395 
1396 		/* if guest os id is not set hypercall should remain disabled */
1397 		if (!hv->hv_guest_os_id)
1398 			break;
1399 		if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1400 			hv->hv_hypercall = data;
1401 			break;
1402 		}
1403 
1404 		/*
1405 		 * If Xen and Hyper-V hypercalls are both enabled, disambiguate
1406 		 * the same way Xen itself does, by setting the bit 31 of EAX
1407 		 * which is RsvdZ in the 32-bit Hyper-V hypercall ABI and just
1408 		 * going to be clobbered on 64-bit.
1409 		 */
1410 		if (kvm_xen_hypercall_enabled(kvm)) {
1411 			/* orl $0x80000000, %eax */
1412 			instructions[i++] = 0x0d;
1413 			instructions[i++] = 0x00;
1414 			instructions[i++] = 0x00;
1415 			instructions[i++] = 0x00;
1416 			instructions[i++] = 0x80;
1417 		}
1418 
1419 		/* vmcall/vmmcall */
1420 		kvm_x86_call(patch_hypercall)(vcpu, instructions + i);
1421 		i += 3;
1422 
1423 		/* ret */
1424 		((unsigned char *)instructions)[i++] = 0xc3;
1425 
1426 		addr = data & HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_MASK;
1427 		if (kvm_vcpu_write_guest(vcpu, addr, instructions, i))
1428 			return 1;
1429 		hv->hv_hypercall = data;
1430 		break;
1431 	}
1432 	case HV_X64_MSR_REFERENCE_TSC:
1433 		hv->hv_tsc_page = data;
1434 		if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE) {
1435 			if (!host)
1436 				hv->hv_tsc_page_status = HV_TSC_PAGE_GUEST_CHANGED;
1437 			else
1438 				hv->hv_tsc_page_status = HV_TSC_PAGE_HOST_CHANGED;
1439 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1440 		} else {
1441 			hv->hv_tsc_page_status = HV_TSC_PAGE_UNSET;
1442 		}
1443 		break;
1444 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1445 		return kvm_hv_msr_set_crash_data(kvm,
1446 						 msr - HV_X64_MSR_CRASH_P0,
1447 						 data);
1448 	case HV_X64_MSR_CRASH_CTL:
1449 		if (host)
1450 			return kvm_hv_msr_set_crash_ctl(kvm, data);
1451 
1452 		if (data & HV_CRASH_CTL_CRASH_NOTIFY) {
1453 			vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
1454 				   hv->hv_crash_param[0],
1455 				   hv->hv_crash_param[1],
1456 				   hv->hv_crash_param[2],
1457 				   hv->hv_crash_param[3],
1458 				   hv->hv_crash_param[4]);
1459 
1460 			/* Send notification about crash to user space */
1461 			kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
1462 		}
1463 		break;
1464 	case HV_X64_MSR_RESET:
1465 		if (data == 1) {
1466 			vcpu_debug(vcpu, "hyper-v reset requested\n");
1467 			kvm_make_request(KVM_REQ_HV_RESET, vcpu);
1468 		}
1469 		break;
1470 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1471 		hv->hv_reenlightenment_control = data;
1472 		break;
1473 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1474 		hv->hv_tsc_emulation_control = data;
1475 		break;
1476 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1477 		if (data && !host)
1478 			return 1;
1479 
1480 		hv->hv_tsc_emulation_status = data;
1481 		break;
1482 	case HV_X64_MSR_TIME_REF_COUNT:
1483 		/* read-only, but still ignore it if host-initiated */
1484 		if (!host)
1485 			return 1;
1486 		break;
1487 	case HV_X64_MSR_TSC_INVARIANT_CONTROL:
1488 		/* Only bit 0 is supported */
1489 		if (data & ~HV_EXPOSE_INVARIANT_TSC)
1490 			return 1;
1491 
1492 		/* The feature can't be disabled from the guest */
1493 		if (!host && hv->hv_invtsc_control && !data)
1494 			return 1;
1495 
1496 		hv->hv_invtsc_control = data;
1497 		break;
1498 	case HV_X64_MSR_SYNDBG_OPTIONS:
1499 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1500 		return syndbg_set_msr(vcpu, msr, data, host);
1501 	default:
1502 		kvm_pr_unimpl_wrmsr(vcpu, msr, data);
1503 		return 1;
1504 	}
1505 	return 0;
1506 }
1507 
1508 /* Calculate cpu time spent by current task in 100ns units */
1509 static u64 current_task_runtime_100ns(void)
1510 {
1511 	u64 utime, stime;
1512 
1513 	task_cputime_adjusted(current, &utime, &stime);
1514 
1515 	return div_u64(utime + stime, 100);
1516 }
1517 
1518 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1519 {
1520 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
1521 
1522 	if (unlikely(!host && !hv_check_msr_access(hv_vcpu, msr)))
1523 		return 1;
1524 
1525 	switch (msr) {
1526 	case HV_X64_MSR_VP_INDEX: {
1527 		struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1528 		u32 new_vp_index = (u32)data;
1529 
1530 		if (!host || new_vp_index >= KVM_MAX_VCPUS)
1531 			return 1;
1532 
1533 		if (new_vp_index == hv_vcpu->vp_index)
1534 			return 0;
1535 
1536 		/*
1537 		 * The VP index is initialized to vcpu_index by
1538 		 * kvm_hv_vcpu_postcreate so they initially match.  Now the
1539 		 * VP index is changing, adjust num_mismatched_vp_indexes if
1540 		 * it now matches or no longer matches vcpu_idx.
1541 		 */
1542 		if (hv_vcpu->vp_index == vcpu->vcpu_idx)
1543 			atomic_inc(&hv->num_mismatched_vp_indexes);
1544 		else if (new_vp_index == vcpu->vcpu_idx)
1545 			atomic_dec(&hv->num_mismatched_vp_indexes);
1546 
1547 		hv_vcpu->vp_index = new_vp_index;
1548 		break;
1549 	}
1550 	case HV_X64_MSR_VP_ASSIST_PAGE: {
1551 		u64 gfn;
1552 		unsigned long addr;
1553 
1554 		if (!(data & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE)) {
1555 			hv_vcpu->hv_vapic = data;
1556 			if (kvm_lapic_set_pv_eoi(vcpu, 0, 0))
1557 				return 1;
1558 			break;
1559 		}
1560 		gfn = data >> HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT;
1561 		addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1562 		if (kvm_is_error_hva(addr))
1563 			return 1;
1564 
1565 		/*
1566 		 * Clear apic_assist portion of struct hv_vp_assist_page
1567 		 * only, there can be valuable data in the rest which needs
1568 		 * to be preserved e.g. on migration.
1569 		 */
1570 		if (__put_user(0, (u32 __user *)addr))
1571 			return 1;
1572 		hv_vcpu->hv_vapic = data;
1573 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
1574 		if (kvm_lapic_set_pv_eoi(vcpu,
1575 					    gfn_to_gpa(gfn) | KVM_MSR_ENABLED,
1576 					    sizeof(struct hv_vp_assist_page)))
1577 			return 1;
1578 		break;
1579 	}
1580 	case HV_X64_MSR_EOI:
1581 		return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1582 	case HV_X64_MSR_ICR:
1583 		return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1584 	case HV_X64_MSR_TPR:
1585 		return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1586 	case HV_X64_MSR_VP_RUNTIME:
1587 		if (!host)
1588 			return 1;
1589 		hv_vcpu->runtime_offset = data - current_task_runtime_100ns();
1590 		break;
1591 	case HV_X64_MSR_SCONTROL:
1592 	case HV_X64_MSR_SVERSION:
1593 	case HV_X64_MSR_SIEFP:
1594 	case HV_X64_MSR_SIMP:
1595 	case HV_X64_MSR_EOM:
1596 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1597 		return synic_set_msr(to_hv_synic(vcpu), msr, data, host);
1598 	case HV_X64_MSR_STIMER0_CONFIG:
1599 	case HV_X64_MSR_STIMER1_CONFIG:
1600 	case HV_X64_MSR_STIMER2_CONFIG:
1601 	case HV_X64_MSR_STIMER3_CONFIG: {
1602 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1603 
1604 		return stimer_set_config(to_hv_stimer(vcpu, timer_index),
1605 					 data, host);
1606 	}
1607 	case HV_X64_MSR_STIMER0_COUNT:
1608 	case HV_X64_MSR_STIMER1_COUNT:
1609 	case HV_X64_MSR_STIMER2_COUNT:
1610 	case HV_X64_MSR_STIMER3_COUNT: {
1611 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1612 
1613 		return stimer_set_count(to_hv_stimer(vcpu, timer_index),
1614 					data, host);
1615 	}
1616 	case HV_X64_MSR_TSC_FREQUENCY:
1617 	case HV_X64_MSR_APIC_FREQUENCY:
1618 		/* read-only, but still ignore it if host-initiated */
1619 		if (!host)
1620 			return 1;
1621 		break;
1622 	default:
1623 		kvm_pr_unimpl_wrmsr(vcpu, msr, data);
1624 		return 1;
1625 	}
1626 
1627 	return 0;
1628 }
1629 
1630 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1631 			     bool host)
1632 {
1633 	u64 data = 0;
1634 	struct kvm *kvm = vcpu->kvm;
1635 	struct kvm_hv *hv = to_kvm_hv(kvm);
1636 
1637 	if (unlikely(!host && !hv_check_msr_access(to_hv_vcpu(vcpu), msr)))
1638 		return 1;
1639 
1640 	switch (msr) {
1641 	case HV_X64_MSR_GUEST_OS_ID:
1642 		data = hv->hv_guest_os_id;
1643 		break;
1644 	case HV_X64_MSR_HYPERCALL:
1645 		data = hv->hv_hypercall;
1646 		break;
1647 	case HV_X64_MSR_TIME_REF_COUNT:
1648 		data = get_time_ref_counter(kvm);
1649 		break;
1650 	case HV_X64_MSR_REFERENCE_TSC:
1651 		data = hv->hv_tsc_page;
1652 		break;
1653 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1654 		return kvm_hv_msr_get_crash_data(kvm,
1655 						 msr - HV_X64_MSR_CRASH_P0,
1656 						 pdata);
1657 	case HV_X64_MSR_CRASH_CTL:
1658 		return kvm_hv_msr_get_crash_ctl(kvm, pdata);
1659 	case HV_X64_MSR_RESET:
1660 		data = 0;
1661 		break;
1662 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1663 		data = hv->hv_reenlightenment_control;
1664 		break;
1665 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1666 		data = hv->hv_tsc_emulation_control;
1667 		break;
1668 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1669 		data = hv->hv_tsc_emulation_status;
1670 		break;
1671 	case HV_X64_MSR_TSC_INVARIANT_CONTROL:
1672 		data = hv->hv_invtsc_control;
1673 		break;
1674 	case HV_X64_MSR_SYNDBG_OPTIONS:
1675 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1676 		return syndbg_get_msr(vcpu, msr, pdata, host);
1677 	default:
1678 		kvm_pr_unimpl_rdmsr(vcpu, msr);
1679 		return 1;
1680 	}
1681 
1682 	*pdata = data;
1683 	return 0;
1684 }
1685 
1686 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1687 			  bool host)
1688 {
1689 	u64 data = 0;
1690 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
1691 
1692 	if (unlikely(!host && !hv_check_msr_access(hv_vcpu, msr)))
1693 		return 1;
1694 
1695 	switch (msr) {
1696 	case HV_X64_MSR_VP_INDEX:
1697 		data = hv_vcpu->vp_index;
1698 		break;
1699 	case HV_X64_MSR_EOI:
1700 		return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1701 	case HV_X64_MSR_ICR:
1702 		return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1703 	case HV_X64_MSR_TPR:
1704 		return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1705 	case HV_X64_MSR_VP_ASSIST_PAGE:
1706 		data = hv_vcpu->hv_vapic;
1707 		break;
1708 	case HV_X64_MSR_VP_RUNTIME:
1709 		data = current_task_runtime_100ns() + hv_vcpu->runtime_offset;
1710 		break;
1711 	case HV_X64_MSR_SCONTROL:
1712 	case HV_X64_MSR_SVERSION:
1713 	case HV_X64_MSR_SIEFP:
1714 	case HV_X64_MSR_SIMP:
1715 	case HV_X64_MSR_EOM:
1716 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1717 		return synic_get_msr(to_hv_synic(vcpu), msr, pdata, host);
1718 	case HV_X64_MSR_STIMER0_CONFIG:
1719 	case HV_X64_MSR_STIMER1_CONFIG:
1720 	case HV_X64_MSR_STIMER2_CONFIG:
1721 	case HV_X64_MSR_STIMER3_CONFIG: {
1722 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1723 
1724 		return stimer_get_config(to_hv_stimer(vcpu, timer_index),
1725 					 pdata);
1726 	}
1727 	case HV_X64_MSR_STIMER0_COUNT:
1728 	case HV_X64_MSR_STIMER1_COUNT:
1729 	case HV_X64_MSR_STIMER2_COUNT:
1730 	case HV_X64_MSR_STIMER3_COUNT: {
1731 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1732 
1733 		return stimer_get_count(to_hv_stimer(vcpu, timer_index),
1734 					pdata);
1735 	}
1736 	case HV_X64_MSR_TSC_FREQUENCY:
1737 		data = (u64)vcpu->arch.virtual_tsc_khz * 1000;
1738 		break;
1739 	case HV_X64_MSR_APIC_FREQUENCY:
1740 		data = div64_u64(1000000000ULL,
1741 				 vcpu->kvm->arch.apic_bus_cycle_ns);
1742 		break;
1743 	default:
1744 		kvm_pr_unimpl_rdmsr(vcpu, msr);
1745 		return 1;
1746 	}
1747 	*pdata = data;
1748 	return 0;
1749 }
1750 
1751 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1752 {
1753 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1754 
1755 	if (!host && !vcpu->arch.hyperv_enabled)
1756 		return 1;
1757 
1758 	if (kvm_hv_vcpu_init(vcpu))
1759 		return 1;
1760 
1761 	if (kvm_hv_msr_partition_wide(msr)) {
1762 		int r;
1763 
1764 		mutex_lock(&hv->hv_lock);
1765 		r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1766 		mutex_unlock(&hv->hv_lock);
1767 		return r;
1768 	} else
1769 		return kvm_hv_set_msr(vcpu, msr, data, host);
1770 }
1771 
1772 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
1773 {
1774 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1775 
1776 	if (!host && !vcpu->arch.hyperv_enabled)
1777 		return 1;
1778 
1779 	if (kvm_hv_vcpu_init(vcpu))
1780 		return 1;
1781 
1782 	if (kvm_hv_msr_partition_wide(msr)) {
1783 		int r;
1784 
1785 		mutex_lock(&hv->hv_lock);
1786 		r = kvm_hv_get_msr_pw(vcpu, msr, pdata, host);
1787 		mutex_unlock(&hv->hv_lock);
1788 		return r;
1789 	} else
1790 		return kvm_hv_get_msr(vcpu, msr, pdata, host);
1791 }
1792 
1793 static void sparse_set_to_vcpu_mask(struct kvm *kvm, u64 *sparse_banks,
1794 				    u64 valid_bank_mask, unsigned long *vcpu_mask)
1795 {
1796 	struct kvm_hv *hv = to_kvm_hv(kvm);
1797 	bool has_mismatch = atomic_read(&hv->num_mismatched_vp_indexes);
1798 	u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1799 	struct kvm_vcpu *vcpu;
1800 	int bank, sbank = 0;
1801 	unsigned long i;
1802 	u64 *bitmap;
1803 
1804 	BUILD_BUG_ON(sizeof(vp_bitmap) >
1805 		     sizeof(*vcpu_mask) * BITS_TO_LONGS(KVM_MAX_VCPUS));
1806 
1807 	/*
1808 	 * If vp_index == vcpu_idx for all vCPUs, fill vcpu_mask directly, else
1809 	 * fill a temporary buffer and manually test each vCPU's VP index.
1810 	 */
1811 	if (likely(!has_mismatch))
1812 		bitmap = (u64 *)vcpu_mask;
1813 	else
1814 		bitmap = vp_bitmap;
1815 
1816 	/*
1817 	 * Each set of 64 VPs is packed into sparse_banks, with valid_bank_mask
1818 	 * having a '1' for each bank that exists in sparse_banks.  Sets must
1819 	 * be in ascending order, i.e. bank0..bankN.
1820 	 */
1821 	memset(bitmap, 0, sizeof(vp_bitmap));
1822 	for_each_set_bit(bank, (unsigned long *)&valid_bank_mask,
1823 			 KVM_HV_MAX_SPARSE_VCPU_SET_BITS)
1824 		bitmap[bank] = sparse_banks[sbank++];
1825 
1826 	if (likely(!has_mismatch))
1827 		return;
1828 
1829 	bitmap_zero(vcpu_mask, KVM_MAX_VCPUS);
1830 	kvm_for_each_vcpu(i, vcpu, kvm) {
1831 		if (test_bit(kvm_hv_get_vpindex(vcpu), (unsigned long *)vp_bitmap))
1832 			__set_bit(i, vcpu_mask);
1833 	}
1834 }
1835 
1836 static bool hv_is_vp_in_sparse_set(u32 vp_id, u64 valid_bank_mask, u64 sparse_banks[])
1837 {
1838 	int valid_bit_nr = vp_id / HV_VCPUS_PER_SPARSE_BANK;
1839 	unsigned long sbank;
1840 
1841 	if (!test_bit(valid_bit_nr, (unsigned long *)&valid_bank_mask))
1842 		return false;
1843 
1844 	/*
1845 	 * The index into the sparse bank is the number of preceding bits in
1846 	 * the valid mask.  Optimize for VMs with <64 vCPUs by skipping the
1847 	 * fancy math if there can't possibly be preceding bits.
1848 	 */
1849 	if (valid_bit_nr)
1850 		sbank = hweight64(valid_bank_mask & GENMASK_ULL(valid_bit_nr - 1, 0));
1851 	else
1852 		sbank = 0;
1853 
1854 	return test_bit(vp_id % HV_VCPUS_PER_SPARSE_BANK,
1855 			(unsigned long *)&sparse_banks[sbank]);
1856 }
1857 
1858 struct kvm_hv_hcall {
1859 	/* Hypercall input data */
1860 	u64 param;
1861 	u64 ingpa;
1862 	u64 outgpa;
1863 	u16 code;
1864 	u16 var_cnt;
1865 	u16 rep_cnt;
1866 	u16 rep_idx;
1867 	bool fast;
1868 	bool rep;
1869 	sse128_t xmm[HV_HYPERCALL_MAX_XMM_REGISTERS];
1870 
1871 	/*
1872 	 * Current read offset when KVM reads hypercall input data gradually,
1873 	 * either offset in bytes from 'ingpa' for regular hypercalls or the
1874 	 * number of already consumed 'XMM halves' for 'fast' hypercalls.
1875 	 */
1876 	union {
1877 		gpa_t data_offset;
1878 		int consumed_xmm_halves;
1879 	};
1880 };
1881 
1882 
1883 static int kvm_hv_get_hc_data(struct kvm *kvm, struct kvm_hv_hcall *hc,
1884 			      u16 orig_cnt, u16 cnt_cap, u64 *data)
1885 {
1886 	/*
1887 	 * Preserve the original count when ignoring entries via a "cap", KVM
1888 	 * still needs to validate the guest input (though the non-XMM path
1889 	 * punts on the checks).
1890 	 */
1891 	u16 cnt = min(orig_cnt, cnt_cap);
1892 	int i, j;
1893 
1894 	if (hc->fast) {
1895 		/*
1896 		 * Each XMM holds two sparse banks, but do not count halves that
1897 		 * have already been consumed for hypercall parameters.
1898 		 */
1899 		if (orig_cnt > 2 * HV_HYPERCALL_MAX_XMM_REGISTERS - hc->consumed_xmm_halves)
1900 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1901 
1902 		for (i = 0; i < cnt; i++) {
1903 			j = i + hc->consumed_xmm_halves;
1904 			if (j % 2)
1905 				data[i] = sse128_hi(hc->xmm[j / 2]);
1906 			else
1907 				data[i] = sse128_lo(hc->xmm[j / 2]);
1908 		}
1909 		return 0;
1910 	}
1911 
1912 	return kvm_read_guest(kvm, hc->ingpa + hc->data_offset, data,
1913 			      cnt * sizeof(*data));
1914 }
1915 
1916 static u64 kvm_get_sparse_vp_set(struct kvm *kvm, struct kvm_hv_hcall *hc,
1917 				 u64 *sparse_banks)
1918 {
1919 	if (hc->var_cnt > HV_MAX_SPARSE_VCPU_BANKS)
1920 		return -EINVAL;
1921 
1922 	/* Cap var_cnt to ignore banks that cannot contain a legal VP index. */
1923 	return kvm_hv_get_hc_data(kvm, hc, hc->var_cnt, KVM_HV_MAX_SPARSE_VCPU_SET_BITS,
1924 				  sparse_banks);
1925 }
1926 
1927 static int kvm_hv_get_tlb_flush_entries(struct kvm *kvm, struct kvm_hv_hcall *hc, u64 entries[])
1928 {
1929 	return kvm_hv_get_hc_data(kvm, hc, hc->rep_cnt, hc->rep_cnt, entries);
1930 }
1931 
1932 static void hv_tlb_flush_enqueue(struct kvm_vcpu *vcpu,
1933 				 struct kvm_vcpu_hv_tlb_flush_fifo *tlb_flush_fifo,
1934 				 u64 *entries, int count)
1935 {
1936 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
1937 	u64 flush_all_entry = KVM_HV_TLB_FLUSHALL_ENTRY;
1938 
1939 	if (!hv_vcpu)
1940 		return;
1941 
1942 	spin_lock(&tlb_flush_fifo->write_lock);
1943 
1944 	/*
1945 	 * All entries should fit on the fifo leaving one free for 'flush all'
1946 	 * entry in case another request comes in. In case there's not enough
1947 	 * space, just put 'flush all' entry there.
1948 	 */
1949 	if (count && entries && count < kfifo_avail(&tlb_flush_fifo->entries)) {
1950 		WARN_ON(kfifo_in(&tlb_flush_fifo->entries, entries, count) != count);
1951 		goto out_unlock;
1952 	}
1953 
1954 	/*
1955 	 * Note: full fifo always contains 'flush all' entry, no need to check the
1956 	 * return value.
1957 	 */
1958 	kfifo_in(&tlb_flush_fifo->entries, &flush_all_entry, 1);
1959 
1960 out_unlock:
1961 	spin_unlock(&tlb_flush_fifo->write_lock);
1962 }
1963 
1964 int kvm_hv_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
1965 {
1966 	struct kvm_vcpu_hv_tlb_flush_fifo *tlb_flush_fifo;
1967 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
1968 	u64 entries[KVM_HV_TLB_FLUSH_FIFO_SIZE];
1969 	int i, j, count;
1970 	gva_t gva;
1971 
1972 	if (!tdp_enabled || !hv_vcpu)
1973 		return -EINVAL;
1974 
1975 	tlb_flush_fifo = kvm_hv_get_tlb_flush_fifo(vcpu, is_guest_mode(vcpu));
1976 
1977 	count = kfifo_out(&tlb_flush_fifo->entries, entries, KVM_HV_TLB_FLUSH_FIFO_SIZE);
1978 
1979 	for (i = 0; i < count; i++) {
1980 		if (entries[i] == KVM_HV_TLB_FLUSHALL_ENTRY)
1981 			goto out_flush_all;
1982 
1983 		/*
1984 		 * Lower 12 bits of 'address' encode the number of additional
1985 		 * pages to flush.
1986 		 */
1987 		gva = entries[i] & PAGE_MASK;
1988 		for (j = 0; j < (entries[i] & ~PAGE_MASK) + 1; j++)
1989 			kvm_x86_call(flush_tlb_gva)(vcpu, gva + j * PAGE_SIZE);
1990 
1991 		++vcpu->stat.tlb_flush;
1992 	}
1993 	return 0;
1994 
1995 out_flush_all:
1996 	kfifo_reset_out(&tlb_flush_fifo->entries);
1997 
1998 	/* Fall back to full flush. */
1999 	return -ENOSPC;
2000 }
2001 
2002 static u64 kvm_hv_flush_tlb(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc)
2003 {
2004 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
2005 	u64 *sparse_banks = hv_vcpu->sparse_banks;
2006 	struct kvm *kvm = vcpu->kvm;
2007 	struct hv_tlb_flush_ex flush_ex;
2008 	struct hv_tlb_flush flush;
2009 	DECLARE_BITMAP(vcpu_mask, KVM_MAX_VCPUS);
2010 	struct kvm_vcpu_hv_tlb_flush_fifo *tlb_flush_fifo;
2011 	/*
2012 	 * Normally, there can be no more than 'KVM_HV_TLB_FLUSH_FIFO_SIZE'
2013 	 * entries on the TLB flush fifo. The last entry, however, needs to be
2014 	 * always left free for 'flush all' entry which gets placed when
2015 	 * there is not enough space to put all the requested entries.
2016 	 */
2017 	u64 __tlb_flush_entries[KVM_HV_TLB_FLUSH_FIFO_SIZE - 1];
2018 	u64 *tlb_flush_entries;
2019 	u64 valid_bank_mask;
2020 	struct kvm_vcpu *v;
2021 	unsigned long i;
2022 	bool all_cpus;
2023 
2024 	/*
2025 	 * The Hyper-V TLFS doesn't allow more than HV_MAX_SPARSE_VCPU_BANKS
2026 	 * sparse banks. Fail the build if KVM's max allowed number of
2027 	 * vCPUs (>4096) exceeds this limit.
2028 	 */
2029 	BUILD_BUG_ON(KVM_HV_MAX_SPARSE_VCPU_SET_BITS > HV_MAX_SPARSE_VCPU_BANKS);
2030 
2031 	/*
2032 	 * 'Slow' hypercall's first parameter is the address in guest's memory
2033 	 * where hypercall parameters are placed. This is either a GPA or a
2034 	 * nested GPA when KVM is handling the call from L2 ('direct' TLB
2035 	 * flush).  Translate the address here so the memory can be uniformly
2036 	 * read with kvm_read_guest().
2037 	 */
2038 	if (!hc->fast && is_guest_mode(vcpu)) {
2039 		hc->ingpa = translate_nested_gpa(vcpu, hc->ingpa, 0, NULL);
2040 		if (unlikely(hc->ingpa == INVALID_GPA))
2041 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
2042 	}
2043 
2044 	if (hc->code == HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST ||
2045 	    hc->code == HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE) {
2046 		if (hc->fast) {
2047 			flush.address_space = hc->ingpa;
2048 			flush.flags = hc->outgpa;
2049 			flush.processor_mask = sse128_lo(hc->xmm[0]);
2050 			hc->consumed_xmm_halves = 1;
2051 		} else {
2052 			if (unlikely(kvm_read_guest(kvm, hc->ingpa,
2053 						    &flush, sizeof(flush))))
2054 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
2055 			hc->data_offset = sizeof(flush);
2056 		}
2057 
2058 		trace_kvm_hv_flush_tlb(flush.processor_mask,
2059 				       flush.address_space, flush.flags,
2060 				       is_guest_mode(vcpu));
2061 
2062 		valid_bank_mask = BIT_ULL(0);
2063 		sparse_banks[0] = flush.processor_mask;
2064 
2065 		/*
2066 		 * Work around possible WS2012 bug: it sends hypercalls
2067 		 * with processor_mask = 0x0 and HV_FLUSH_ALL_PROCESSORS clear,
2068 		 * while also expecting us to flush something and crashing if
2069 		 * we don't. Let's treat processor_mask == 0 same as
2070 		 * HV_FLUSH_ALL_PROCESSORS.
2071 		 */
2072 		all_cpus = (flush.flags & HV_FLUSH_ALL_PROCESSORS) ||
2073 			flush.processor_mask == 0;
2074 	} else {
2075 		if (hc->fast) {
2076 			flush_ex.address_space = hc->ingpa;
2077 			flush_ex.flags = hc->outgpa;
2078 			memcpy(&flush_ex.hv_vp_set,
2079 			       &hc->xmm[0], sizeof(hc->xmm[0]));
2080 			hc->consumed_xmm_halves = 2;
2081 		} else {
2082 			if (unlikely(kvm_read_guest(kvm, hc->ingpa, &flush_ex,
2083 						    sizeof(flush_ex))))
2084 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
2085 			hc->data_offset = sizeof(flush_ex);
2086 		}
2087 
2088 		trace_kvm_hv_flush_tlb_ex(flush_ex.hv_vp_set.valid_bank_mask,
2089 					  flush_ex.hv_vp_set.format,
2090 					  flush_ex.address_space,
2091 					  flush_ex.flags, is_guest_mode(vcpu));
2092 
2093 		valid_bank_mask = flush_ex.hv_vp_set.valid_bank_mask;
2094 		all_cpus = flush_ex.hv_vp_set.format !=
2095 			HV_GENERIC_SET_SPARSE_4K;
2096 
2097 		if (hc->var_cnt != hweight64(valid_bank_mask))
2098 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
2099 
2100 		if (!all_cpus) {
2101 			if (!hc->var_cnt)
2102 				goto ret_success;
2103 
2104 			if (kvm_get_sparse_vp_set(kvm, hc, sparse_banks))
2105 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
2106 		}
2107 
2108 		/*
2109 		 * Hyper-V TLFS doesn't explicitly forbid non-empty sparse vCPU
2110 		 * banks (and, thus, non-zero 'var_cnt') for the 'all vCPUs'
2111 		 * case (HV_GENERIC_SET_ALL).  Always adjust data_offset and
2112 		 * consumed_xmm_halves to make sure TLB flush entries are read
2113 		 * from the correct offset.
2114 		 */
2115 		if (hc->fast)
2116 			hc->consumed_xmm_halves += hc->var_cnt;
2117 		else
2118 			hc->data_offset += hc->var_cnt * sizeof(sparse_banks[0]);
2119 	}
2120 
2121 	if (hc->code == HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE ||
2122 	    hc->code == HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX ||
2123 	    hc->rep_cnt > ARRAY_SIZE(__tlb_flush_entries)) {
2124 		tlb_flush_entries = NULL;
2125 	} else {
2126 		if (kvm_hv_get_tlb_flush_entries(kvm, hc, __tlb_flush_entries))
2127 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
2128 		tlb_flush_entries = __tlb_flush_entries;
2129 	}
2130 
2131 	/*
2132 	 * vcpu->arch.cr3 may not be up-to-date for running vCPUs so we can't
2133 	 * analyze it here, flush TLB regardless of the specified address space.
2134 	 */
2135 	if (all_cpus && !is_guest_mode(vcpu)) {
2136 		kvm_for_each_vcpu(i, v, kvm) {
2137 			tlb_flush_fifo = kvm_hv_get_tlb_flush_fifo(v, false);
2138 			hv_tlb_flush_enqueue(v, tlb_flush_fifo,
2139 					     tlb_flush_entries, hc->rep_cnt);
2140 		}
2141 
2142 		kvm_make_all_cpus_request(kvm, KVM_REQ_HV_TLB_FLUSH);
2143 	} else if (!is_guest_mode(vcpu)) {
2144 		sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask, vcpu_mask);
2145 
2146 		for_each_set_bit(i, vcpu_mask, KVM_MAX_VCPUS) {
2147 			v = kvm_get_vcpu(kvm, i);
2148 			if (!v)
2149 				continue;
2150 			tlb_flush_fifo = kvm_hv_get_tlb_flush_fifo(v, false);
2151 			hv_tlb_flush_enqueue(v, tlb_flush_fifo,
2152 					     tlb_flush_entries, hc->rep_cnt);
2153 		}
2154 
2155 		kvm_make_vcpus_request_mask(kvm, KVM_REQ_HV_TLB_FLUSH, vcpu_mask);
2156 	} else {
2157 		struct kvm_vcpu_hv *hv_v;
2158 
2159 		bitmap_zero(vcpu_mask, KVM_MAX_VCPUS);
2160 
2161 		kvm_for_each_vcpu(i, v, kvm) {
2162 			hv_v = to_hv_vcpu(v);
2163 
2164 			/*
2165 			 * The following check races with nested vCPUs entering/exiting
2166 			 * and/or migrating between L1's vCPUs, however the only case when
2167 			 * KVM *must* flush the TLB is when the target L2 vCPU keeps
2168 			 * running on the same L1 vCPU from the moment of the request until
2169 			 * kvm_hv_flush_tlb() returns. TLB is fully flushed in all other
2170 			 * cases, e.g. when the target L2 vCPU migrates to a different L1
2171 			 * vCPU or when the corresponding L1 vCPU temporary switches to a
2172 			 * different L2 vCPU while the request is being processed.
2173 			 */
2174 			if (!hv_v || hv_v->nested.vm_id != hv_vcpu->nested.vm_id)
2175 				continue;
2176 
2177 			if (!all_cpus &&
2178 			    !hv_is_vp_in_sparse_set(hv_v->nested.vp_id, valid_bank_mask,
2179 						    sparse_banks))
2180 				continue;
2181 
2182 			__set_bit(i, vcpu_mask);
2183 			tlb_flush_fifo = kvm_hv_get_tlb_flush_fifo(v, true);
2184 			hv_tlb_flush_enqueue(v, tlb_flush_fifo,
2185 					     tlb_flush_entries, hc->rep_cnt);
2186 		}
2187 
2188 		kvm_make_vcpus_request_mask(kvm, KVM_REQ_HV_TLB_FLUSH, vcpu_mask);
2189 	}
2190 
2191 ret_success:
2192 	/* We always do full TLB flush, set 'Reps completed' = 'Rep Count' */
2193 	return (u64)HV_STATUS_SUCCESS |
2194 		((u64)hc->rep_cnt << HV_HYPERCALL_REP_COMP_OFFSET);
2195 }
2196 
2197 static void kvm_hv_send_ipi_to_many(struct kvm *kvm, u32 vector,
2198 				    u64 *sparse_banks, u64 valid_bank_mask)
2199 {
2200 	struct kvm_lapic_irq irq = {
2201 		.delivery_mode = APIC_DM_FIXED,
2202 		.vector = vector
2203 	};
2204 	struct kvm_vcpu *vcpu;
2205 	unsigned long i;
2206 
2207 	kvm_for_each_vcpu(i, vcpu, kvm) {
2208 		if (sparse_banks &&
2209 		    !hv_is_vp_in_sparse_set(kvm_hv_get_vpindex(vcpu),
2210 					    valid_bank_mask, sparse_banks))
2211 			continue;
2212 
2213 		/* We fail only when APIC is disabled */
2214 		kvm_apic_set_irq(vcpu, &irq, NULL);
2215 	}
2216 }
2217 
2218 static u64 kvm_hv_send_ipi(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc)
2219 {
2220 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
2221 	u64 *sparse_banks = hv_vcpu->sparse_banks;
2222 	struct kvm *kvm = vcpu->kvm;
2223 	struct hv_send_ipi_ex send_ipi_ex;
2224 	struct hv_send_ipi send_ipi;
2225 	u64 valid_bank_mask;
2226 	u32 vector;
2227 	bool all_cpus;
2228 
2229 	if (hc->code == HVCALL_SEND_IPI) {
2230 		if (!hc->fast) {
2231 			if (unlikely(kvm_read_guest(kvm, hc->ingpa, &send_ipi,
2232 						    sizeof(send_ipi))))
2233 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
2234 			sparse_banks[0] = send_ipi.cpu_mask;
2235 			vector = send_ipi.vector;
2236 		} else {
2237 			/* 'reserved' part of hv_send_ipi should be 0 */
2238 			if (unlikely(hc->ingpa >> 32 != 0))
2239 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
2240 			sparse_banks[0] = hc->outgpa;
2241 			vector = (u32)hc->ingpa;
2242 		}
2243 		all_cpus = false;
2244 		valid_bank_mask = BIT_ULL(0);
2245 
2246 		trace_kvm_hv_send_ipi(vector, sparse_banks[0]);
2247 	} else {
2248 		if (!hc->fast) {
2249 			if (unlikely(kvm_read_guest(kvm, hc->ingpa, &send_ipi_ex,
2250 						    sizeof(send_ipi_ex))))
2251 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
2252 		} else {
2253 			send_ipi_ex.vector = (u32)hc->ingpa;
2254 			send_ipi_ex.vp_set.format = hc->outgpa;
2255 			send_ipi_ex.vp_set.valid_bank_mask = sse128_lo(hc->xmm[0]);
2256 		}
2257 
2258 		trace_kvm_hv_send_ipi_ex(send_ipi_ex.vector,
2259 					 send_ipi_ex.vp_set.format,
2260 					 send_ipi_ex.vp_set.valid_bank_mask);
2261 
2262 		vector = send_ipi_ex.vector;
2263 		valid_bank_mask = send_ipi_ex.vp_set.valid_bank_mask;
2264 		all_cpus = send_ipi_ex.vp_set.format == HV_GENERIC_SET_ALL;
2265 
2266 		if (hc->var_cnt != hweight64(valid_bank_mask))
2267 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
2268 
2269 		if (all_cpus)
2270 			goto check_and_send_ipi;
2271 
2272 		if (!hc->var_cnt)
2273 			goto ret_success;
2274 
2275 		if (!hc->fast)
2276 			hc->data_offset = offsetof(struct hv_send_ipi_ex,
2277 						   vp_set.bank_contents);
2278 		else
2279 			hc->consumed_xmm_halves = 1;
2280 
2281 		if (kvm_get_sparse_vp_set(kvm, hc, sparse_banks))
2282 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
2283 	}
2284 
2285 check_and_send_ipi:
2286 	if ((vector < HV_IPI_LOW_VECTOR) || (vector > HV_IPI_HIGH_VECTOR))
2287 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
2288 
2289 	if (all_cpus)
2290 		kvm_hv_send_ipi_to_many(kvm, vector, NULL, 0);
2291 	else
2292 		kvm_hv_send_ipi_to_many(kvm, vector, sparse_banks, valid_bank_mask);
2293 
2294 ret_success:
2295 	return HV_STATUS_SUCCESS;
2296 }
2297 
2298 void kvm_hv_set_cpuid(struct kvm_vcpu *vcpu, bool hyperv_enabled)
2299 {
2300 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
2301 	struct kvm_cpuid_entry2 *entry;
2302 
2303 	vcpu->arch.hyperv_enabled = hyperv_enabled;
2304 
2305 	if (!hv_vcpu) {
2306 		/*
2307 		 * KVM should have already allocated kvm_vcpu_hv if Hyper-V is
2308 		 * enabled in CPUID.
2309 		 */
2310 		WARN_ON_ONCE(vcpu->arch.hyperv_enabled);
2311 		return;
2312 	}
2313 
2314 	memset(&hv_vcpu->cpuid_cache, 0, sizeof(hv_vcpu->cpuid_cache));
2315 
2316 	if (!vcpu->arch.hyperv_enabled)
2317 		return;
2318 
2319 	entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_FEATURES);
2320 	if (entry) {
2321 		hv_vcpu->cpuid_cache.features_eax = entry->eax;
2322 		hv_vcpu->cpuid_cache.features_ebx = entry->ebx;
2323 		hv_vcpu->cpuid_cache.features_edx = entry->edx;
2324 	}
2325 
2326 	entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_ENLIGHTMENT_INFO);
2327 	if (entry) {
2328 		hv_vcpu->cpuid_cache.enlightenments_eax = entry->eax;
2329 		hv_vcpu->cpuid_cache.enlightenments_ebx = entry->ebx;
2330 	}
2331 
2332 	entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES);
2333 	if (entry)
2334 		hv_vcpu->cpuid_cache.syndbg_cap_eax = entry->eax;
2335 
2336 	entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_NESTED_FEATURES);
2337 	if (entry) {
2338 		hv_vcpu->cpuid_cache.nested_eax = entry->eax;
2339 		hv_vcpu->cpuid_cache.nested_ebx = entry->ebx;
2340 	}
2341 }
2342 
2343 int kvm_hv_set_enforce_cpuid(struct kvm_vcpu *vcpu, bool enforce)
2344 {
2345 	struct kvm_vcpu_hv *hv_vcpu;
2346 	int ret = 0;
2347 
2348 	if (!to_hv_vcpu(vcpu)) {
2349 		if (enforce) {
2350 			ret = kvm_hv_vcpu_init(vcpu);
2351 			if (ret)
2352 				return ret;
2353 		} else {
2354 			return 0;
2355 		}
2356 	}
2357 
2358 	hv_vcpu = to_hv_vcpu(vcpu);
2359 	hv_vcpu->enforce_cpuid = enforce;
2360 
2361 	return ret;
2362 }
2363 
2364 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
2365 {
2366 	bool longmode;
2367 
2368 	longmode = is_64_bit_hypercall(vcpu);
2369 	if (longmode)
2370 		kvm_rax_write(vcpu, result);
2371 	else {
2372 		kvm_rdx_write(vcpu, result >> 32);
2373 		kvm_rax_write(vcpu, result & 0xffffffff);
2374 	}
2375 }
2376 
2377 static int kvm_hv_hypercall_complete(struct kvm_vcpu *vcpu, u64 result)
2378 {
2379 	u32 tlb_lock_count = 0;
2380 	int ret;
2381 
2382 	if (hv_result_success(result) && is_guest_mode(vcpu) &&
2383 	    kvm_hv_is_tlb_flush_hcall(vcpu) &&
2384 	    kvm_read_guest(vcpu->kvm, to_hv_vcpu(vcpu)->nested.pa_page_gpa,
2385 			   &tlb_lock_count, sizeof(tlb_lock_count)))
2386 		result = HV_STATUS_INVALID_HYPERCALL_INPUT;
2387 
2388 	trace_kvm_hv_hypercall_done(result);
2389 	kvm_hv_hypercall_set_result(vcpu, result);
2390 	++vcpu->stat.hypercalls;
2391 
2392 	ret = kvm_skip_emulated_instruction(vcpu);
2393 
2394 	if (tlb_lock_count)
2395 		kvm_x86_ops.nested_ops->hv_inject_synthetic_vmexit_post_tlb_flush(vcpu);
2396 
2397 	return ret;
2398 }
2399 
2400 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
2401 {
2402 	return kvm_hv_hypercall_complete(vcpu, vcpu->run->hyperv.u.hcall.result);
2403 }
2404 
2405 static u16 kvm_hvcall_signal_event(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc)
2406 {
2407 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
2408 	struct eventfd_ctx *eventfd;
2409 
2410 	if (unlikely(!hc->fast)) {
2411 		int ret;
2412 		gpa_t gpa = hc->ingpa;
2413 
2414 		if ((gpa & (__alignof__(hc->ingpa) - 1)) ||
2415 		    offset_in_page(gpa) + sizeof(hc->ingpa) > PAGE_SIZE)
2416 			return HV_STATUS_INVALID_ALIGNMENT;
2417 
2418 		ret = kvm_vcpu_read_guest(vcpu, gpa,
2419 					  &hc->ingpa, sizeof(hc->ingpa));
2420 		if (ret < 0)
2421 			return HV_STATUS_INVALID_ALIGNMENT;
2422 	}
2423 
2424 	/*
2425 	 * Per spec, bits 32-47 contain the extra "flag number".  However, we
2426 	 * have no use for it, and in all known usecases it is zero, so just
2427 	 * report lookup failure if it isn't.
2428 	 */
2429 	if (hc->ingpa & 0xffff00000000ULL)
2430 		return HV_STATUS_INVALID_PORT_ID;
2431 	/* remaining bits are reserved-zero */
2432 	if (hc->ingpa & ~KVM_HYPERV_CONN_ID_MASK)
2433 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
2434 
2435 	/* the eventfd is protected by vcpu->kvm->srcu, but conn_to_evt isn't */
2436 	rcu_read_lock();
2437 	eventfd = idr_find(&hv->conn_to_evt, hc->ingpa);
2438 	rcu_read_unlock();
2439 	if (!eventfd)
2440 		return HV_STATUS_INVALID_PORT_ID;
2441 
2442 	eventfd_signal(eventfd);
2443 	return HV_STATUS_SUCCESS;
2444 }
2445 
2446 static bool is_xmm_fast_hypercall(struct kvm_hv_hcall *hc)
2447 {
2448 	switch (hc->code) {
2449 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
2450 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
2451 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
2452 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
2453 	case HVCALL_SEND_IPI_EX:
2454 		return true;
2455 	}
2456 
2457 	return false;
2458 }
2459 
2460 static void kvm_hv_hypercall_read_xmm(struct kvm_hv_hcall *hc)
2461 {
2462 	int reg;
2463 
2464 	kvm_fpu_get();
2465 	for (reg = 0; reg < HV_HYPERCALL_MAX_XMM_REGISTERS; reg++)
2466 		_kvm_read_sse_reg(reg, &hc->xmm[reg]);
2467 	kvm_fpu_put();
2468 }
2469 
2470 static bool hv_check_hypercall_access(struct kvm_vcpu_hv *hv_vcpu, u16 code)
2471 {
2472 	if (!hv_vcpu->enforce_cpuid)
2473 		return true;
2474 
2475 	switch (code) {
2476 	case HVCALL_NOTIFY_LONG_SPIN_WAIT:
2477 		return hv_vcpu->cpuid_cache.enlightenments_ebx &&
2478 			hv_vcpu->cpuid_cache.enlightenments_ebx != U32_MAX;
2479 	case HVCALL_POST_MESSAGE:
2480 		return hv_vcpu->cpuid_cache.features_ebx & HV_POST_MESSAGES;
2481 	case HVCALL_SIGNAL_EVENT:
2482 		return hv_vcpu->cpuid_cache.features_ebx & HV_SIGNAL_EVENTS;
2483 	case HVCALL_POST_DEBUG_DATA:
2484 	case HVCALL_RETRIEVE_DEBUG_DATA:
2485 	case HVCALL_RESET_DEBUG_SESSION:
2486 		/*
2487 		 * Return 'true' when SynDBG is disabled so the resulting code
2488 		 * will be HV_STATUS_INVALID_HYPERCALL_CODE.
2489 		 */
2490 		return !kvm_hv_is_syndbg_enabled(hv_vcpu->vcpu) ||
2491 			hv_vcpu->cpuid_cache.features_ebx & HV_DEBUGGING;
2492 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
2493 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
2494 		if (!(hv_vcpu->cpuid_cache.enlightenments_eax &
2495 		      HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED))
2496 			return false;
2497 		fallthrough;
2498 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
2499 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
2500 		return hv_vcpu->cpuid_cache.enlightenments_eax &
2501 			HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
2502 	case HVCALL_SEND_IPI_EX:
2503 		if (!(hv_vcpu->cpuid_cache.enlightenments_eax &
2504 		      HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED))
2505 			return false;
2506 		fallthrough;
2507 	case HVCALL_SEND_IPI:
2508 		return hv_vcpu->cpuid_cache.enlightenments_eax &
2509 			HV_X64_CLUSTER_IPI_RECOMMENDED;
2510 	case HV_EXT_CALL_QUERY_CAPABILITIES ... HV_EXT_CALL_MAX:
2511 		return hv_vcpu->cpuid_cache.features_ebx &
2512 			HV_ENABLE_EXTENDED_HYPERCALLS;
2513 	default:
2514 		break;
2515 	}
2516 
2517 	return true;
2518 }
2519 
2520 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
2521 {
2522 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
2523 	struct kvm_hv_hcall hc;
2524 	u64 ret = HV_STATUS_SUCCESS;
2525 
2526 	/*
2527 	 * hypercall generates UD from non zero cpl and real mode
2528 	 * per HYPER-V spec
2529 	 */
2530 	if (kvm_x86_call(get_cpl)(vcpu) != 0 || !is_protmode(vcpu)) {
2531 		kvm_queue_exception(vcpu, UD_VECTOR);
2532 		return 1;
2533 	}
2534 
2535 #ifdef CONFIG_X86_64
2536 	if (is_64_bit_hypercall(vcpu)) {
2537 		hc.param = kvm_rcx_read(vcpu);
2538 		hc.ingpa = kvm_rdx_read(vcpu);
2539 		hc.outgpa = kvm_r8_read(vcpu);
2540 	} else
2541 #endif
2542 	{
2543 		hc.param = ((u64)kvm_rdx_read(vcpu) << 32) |
2544 			    (kvm_rax_read(vcpu) & 0xffffffff);
2545 		hc.ingpa = ((u64)kvm_rbx_read(vcpu) << 32) |
2546 			    (kvm_rcx_read(vcpu) & 0xffffffff);
2547 		hc.outgpa = ((u64)kvm_rdi_read(vcpu) << 32) |
2548 			     (kvm_rsi_read(vcpu) & 0xffffffff);
2549 	}
2550 
2551 	hc.code = hc.param & 0xffff;
2552 	hc.var_cnt = (hc.param & HV_HYPERCALL_VARHEAD_MASK) >> HV_HYPERCALL_VARHEAD_OFFSET;
2553 	hc.fast = !!(hc.param & HV_HYPERCALL_FAST_BIT);
2554 	hc.rep_cnt = (hc.param >> HV_HYPERCALL_REP_COMP_OFFSET) & 0xfff;
2555 	hc.rep_idx = (hc.param >> HV_HYPERCALL_REP_START_OFFSET) & 0xfff;
2556 	hc.rep = !!(hc.rep_cnt || hc.rep_idx);
2557 
2558 	trace_kvm_hv_hypercall(hc.code, hc.fast, hc.var_cnt, hc.rep_cnt,
2559 			       hc.rep_idx, hc.ingpa, hc.outgpa);
2560 
2561 	if (unlikely(!hv_check_hypercall_access(hv_vcpu, hc.code))) {
2562 		ret = HV_STATUS_ACCESS_DENIED;
2563 		goto hypercall_complete;
2564 	}
2565 
2566 	if (unlikely(hc.param & HV_HYPERCALL_RSVD_MASK)) {
2567 		ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2568 		goto hypercall_complete;
2569 	}
2570 
2571 	if (hc.fast && is_xmm_fast_hypercall(&hc)) {
2572 		if (unlikely(hv_vcpu->enforce_cpuid &&
2573 			     !(hv_vcpu->cpuid_cache.features_edx &
2574 			       HV_X64_HYPERCALL_XMM_INPUT_AVAILABLE))) {
2575 			kvm_queue_exception(vcpu, UD_VECTOR);
2576 			return 1;
2577 		}
2578 
2579 		kvm_hv_hypercall_read_xmm(&hc);
2580 	}
2581 
2582 	switch (hc.code) {
2583 	case HVCALL_NOTIFY_LONG_SPIN_WAIT:
2584 		if (unlikely(hc.rep || hc.var_cnt)) {
2585 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2586 			break;
2587 		}
2588 		kvm_vcpu_on_spin(vcpu, true);
2589 		break;
2590 	case HVCALL_SIGNAL_EVENT:
2591 		if (unlikely(hc.rep || hc.var_cnt)) {
2592 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2593 			break;
2594 		}
2595 		ret = kvm_hvcall_signal_event(vcpu, &hc);
2596 		if (ret != HV_STATUS_INVALID_PORT_ID)
2597 			break;
2598 		fallthrough;	/* maybe userspace knows this conn_id */
2599 	case HVCALL_POST_MESSAGE:
2600 		/* don't bother userspace if it has no way to handle it */
2601 		if (unlikely(hc.rep || hc.var_cnt || !to_hv_synic(vcpu)->active)) {
2602 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2603 			break;
2604 		}
2605 		goto hypercall_userspace_exit;
2606 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
2607 		if (unlikely(hc.var_cnt)) {
2608 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2609 			break;
2610 		}
2611 		fallthrough;
2612 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
2613 		if (unlikely(!hc.rep_cnt || hc.rep_idx)) {
2614 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2615 			break;
2616 		}
2617 		ret = kvm_hv_flush_tlb(vcpu, &hc);
2618 		break;
2619 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
2620 		if (unlikely(hc.var_cnt)) {
2621 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2622 			break;
2623 		}
2624 		fallthrough;
2625 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
2626 		if (unlikely(hc.rep)) {
2627 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2628 			break;
2629 		}
2630 		ret = kvm_hv_flush_tlb(vcpu, &hc);
2631 		break;
2632 	case HVCALL_SEND_IPI:
2633 		if (unlikely(hc.var_cnt)) {
2634 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2635 			break;
2636 		}
2637 		fallthrough;
2638 	case HVCALL_SEND_IPI_EX:
2639 		if (unlikely(hc.rep)) {
2640 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2641 			break;
2642 		}
2643 		ret = kvm_hv_send_ipi(vcpu, &hc);
2644 		break;
2645 	case HVCALL_POST_DEBUG_DATA:
2646 	case HVCALL_RETRIEVE_DEBUG_DATA:
2647 		if (unlikely(hc.fast)) {
2648 			ret = HV_STATUS_INVALID_PARAMETER;
2649 			break;
2650 		}
2651 		fallthrough;
2652 	case HVCALL_RESET_DEBUG_SESSION: {
2653 		struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
2654 
2655 		if (!kvm_hv_is_syndbg_enabled(vcpu)) {
2656 			ret = HV_STATUS_INVALID_HYPERCALL_CODE;
2657 			break;
2658 		}
2659 
2660 		if (!(syndbg->options & HV_X64_SYNDBG_OPTION_USE_HCALLS)) {
2661 			ret = HV_STATUS_OPERATION_DENIED;
2662 			break;
2663 		}
2664 		goto hypercall_userspace_exit;
2665 	}
2666 	case HV_EXT_CALL_QUERY_CAPABILITIES ... HV_EXT_CALL_MAX:
2667 		if (unlikely(hc.fast)) {
2668 			ret = HV_STATUS_INVALID_PARAMETER;
2669 			break;
2670 		}
2671 		goto hypercall_userspace_exit;
2672 	default:
2673 		ret = HV_STATUS_INVALID_HYPERCALL_CODE;
2674 		break;
2675 	}
2676 
2677 hypercall_complete:
2678 	return kvm_hv_hypercall_complete(vcpu, ret);
2679 
2680 hypercall_userspace_exit:
2681 	vcpu->run->exit_reason = KVM_EXIT_HYPERV;
2682 	vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
2683 	vcpu->run->hyperv.u.hcall.input = hc.param;
2684 	vcpu->run->hyperv.u.hcall.params[0] = hc.ingpa;
2685 	vcpu->run->hyperv.u.hcall.params[1] = hc.outgpa;
2686 	vcpu->arch.complete_userspace_io = kvm_hv_hypercall_complete_userspace;
2687 	return 0;
2688 }
2689 
2690 void kvm_hv_init_vm(struct kvm *kvm)
2691 {
2692 	struct kvm_hv *hv = to_kvm_hv(kvm);
2693 
2694 	mutex_init(&hv->hv_lock);
2695 	idr_init(&hv->conn_to_evt);
2696 }
2697 
2698 void kvm_hv_destroy_vm(struct kvm *kvm)
2699 {
2700 	struct kvm_hv *hv = to_kvm_hv(kvm);
2701 	struct eventfd_ctx *eventfd;
2702 	int i;
2703 
2704 	idr_for_each_entry(&hv->conn_to_evt, eventfd, i)
2705 		eventfd_ctx_put(eventfd);
2706 	idr_destroy(&hv->conn_to_evt);
2707 }
2708 
2709 static int kvm_hv_eventfd_assign(struct kvm *kvm, u32 conn_id, int fd)
2710 {
2711 	struct kvm_hv *hv = to_kvm_hv(kvm);
2712 	struct eventfd_ctx *eventfd;
2713 	int ret;
2714 
2715 	eventfd = eventfd_ctx_fdget(fd);
2716 	if (IS_ERR(eventfd))
2717 		return PTR_ERR(eventfd);
2718 
2719 	mutex_lock(&hv->hv_lock);
2720 	ret = idr_alloc(&hv->conn_to_evt, eventfd, conn_id, conn_id + 1,
2721 			GFP_KERNEL_ACCOUNT);
2722 	mutex_unlock(&hv->hv_lock);
2723 
2724 	if (ret >= 0)
2725 		return 0;
2726 
2727 	if (ret == -ENOSPC)
2728 		ret = -EEXIST;
2729 	eventfd_ctx_put(eventfd);
2730 	return ret;
2731 }
2732 
2733 static int kvm_hv_eventfd_deassign(struct kvm *kvm, u32 conn_id)
2734 {
2735 	struct kvm_hv *hv = to_kvm_hv(kvm);
2736 	struct eventfd_ctx *eventfd;
2737 
2738 	mutex_lock(&hv->hv_lock);
2739 	eventfd = idr_remove(&hv->conn_to_evt, conn_id);
2740 	mutex_unlock(&hv->hv_lock);
2741 
2742 	if (!eventfd)
2743 		return -ENOENT;
2744 
2745 	synchronize_srcu(&kvm->srcu);
2746 	eventfd_ctx_put(eventfd);
2747 	return 0;
2748 }
2749 
2750 int kvm_vm_ioctl_hv_eventfd(struct kvm *kvm, struct kvm_hyperv_eventfd *args)
2751 {
2752 	if ((args->flags & ~KVM_HYPERV_EVENTFD_DEASSIGN) ||
2753 	    (args->conn_id & ~KVM_HYPERV_CONN_ID_MASK))
2754 		return -EINVAL;
2755 
2756 	if (args->flags == KVM_HYPERV_EVENTFD_DEASSIGN)
2757 		return kvm_hv_eventfd_deassign(kvm, args->conn_id);
2758 	return kvm_hv_eventfd_assign(kvm, args->conn_id, args->fd);
2759 }
2760 
2761 int kvm_get_hv_cpuid(struct kvm_vcpu *vcpu, struct kvm_cpuid2 *cpuid,
2762 		     struct kvm_cpuid_entry2 __user *entries)
2763 {
2764 	uint16_t evmcs_ver = 0;
2765 	struct kvm_cpuid_entry2 cpuid_entries[] = {
2766 		{ .function = HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS },
2767 		{ .function = HYPERV_CPUID_INTERFACE },
2768 		{ .function = HYPERV_CPUID_VERSION },
2769 		{ .function = HYPERV_CPUID_FEATURES },
2770 		{ .function = HYPERV_CPUID_ENLIGHTMENT_INFO },
2771 		{ .function = HYPERV_CPUID_IMPLEMENT_LIMITS },
2772 		{ .function = HYPERV_CPUID_SYNDBG_VENDOR_AND_MAX_FUNCTIONS },
2773 		{ .function = HYPERV_CPUID_SYNDBG_INTERFACE },
2774 		{ .function = HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES	},
2775 		{ .function = HYPERV_CPUID_NESTED_FEATURES },
2776 	};
2777 	int i, nent = ARRAY_SIZE(cpuid_entries);
2778 
2779 	if (kvm_x86_ops.nested_ops->get_evmcs_version)
2780 		evmcs_ver = kvm_x86_ops.nested_ops->get_evmcs_version(vcpu);
2781 
2782 	if (cpuid->nent < nent)
2783 		return -E2BIG;
2784 
2785 	if (cpuid->nent > nent)
2786 		cpuid->nent = nent;
2787 
2788 	for (i = 0; i < nent; i++) {
2789 		struct kvm_cpuid_entry2 *ent = &cpuid_entries[i];
2790 		u32 signature[3];
2791 
2792 		switch (ent->function) {
2793 		case HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS:
2794 			memcpy(signature, "Linux KVM Hv", 12);
2795 
2796 			ent->eax = HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES;
2797 			ent->ebx = signature[0];
2798 			ent->ecx = signature[1];
2799 			ent->edx = signature[2];
2800 			break;
2801 
2802 		case HYPERV_CPUID_INTERFACE:
2803 			ent->eax = HYPERV_CPUID_SIGNATURE_EAX;
2804 			break;
2805 
2806 		case HYPERV_CPUID_VERSION:
2807 			/*
2808 			 * We implement some Hyper-V 2016 functions so let's use
2809 			 * this version.
2810 			 */
2811 			ent->eax = 0x00003839;
2812 			ent->ebx = 0x000A0000;
2813 			break;
2814 
2815 		case HYPERV_CPUID_FEATURES:
2816 			ent->eax |= HV_MSR_VP_RUNTIME_AVAILABLE;
2817 			ent->eax |= HV_MSR_TIME_REF_COUNT_AVAILABLE;
2818 			ent->eax |= HV_MSR_SYNIC_AVAILABLE;
2819 			ent->eax |= HV_MSR_SYNTIMER_AVAILABLE;
2820 			ent->eax |= HV_MSR_APIC_ACCESS_AVAILABLE;
2821 			ent->eax |= HV_MSR_HYPERCALL_AVAILABLE;
2822 			ent->eax |= HV_MSR_VP_INDEX_AVAILABLE;
2823 			ent->eax |= HV_MSR_RESET_AVAILABLE;
2824 			ent->eax |= HV_MSR_REFERENCE_TSC_AVAILABLE;
2825 			ent->eax |= HV_ACCESS_FREQUENCY_MSRS;
2826 			ent->eax |= HV_ACCESS_REENLIGHTENMENT;
2827 			ent->eax |= HV_ACCESS_TSC_INVARIANT;
2828 
2829 			ent->ebx |= HV_POST_MESSAGES;
2830 			ent->ebx |= HV_SIGNAL_EVENTS;
2831 			ent->ebx |= HV_ENABLE_EXTENDED_HYPERCALLS;
2832 
2833 			ent->edx |= HV_X64_HYPERCALL_XMM_INPUT_AVAILABLE;
2834 			ent->edx |= HV_FEATURE_FREQUENCY_MSRS_AVAILABLE;
2835 			ent->edx |= HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
2836 
2837 			ent->ebx |= HV_DEBUGGING;
2838 			ent->edx |= HV_X64_GUEST_DEBUGGING_AVAILABLE;
2839 			ent->edx |= HV_FEATURE_DEBUG_MSRS_AVAILABLE;
2840 			ent->edx |= HV_FEATURE_EXT_GVA_RANGES_FLUSH;
2841 
2842 			/*
2843 			 * Direct Synthetic timers only make sense with in-kernel
2844 			 * LAPIC
2845 			 */
2846 			if (!vcpu || lapic_in_kernel(vcpu))
2847 				ent->edx |= HV_STIMER_DIRECT_MODE_AVAILABLE;
2848 
2849 			break;
2850 
2851 		case HYPERV_CPUID_ENLIGHTMENT_INFO:
2852 			ent->eax |= HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
2853 			ent->eax |= HV_X64_APIC_ACCESS_RECOMMENDED;
2854 			ent->eax |= HV_X64_RELAXED_TIMING_RECOMMENDED;
2855 			ent->eax |= HV_X64_CLUSTER_IPI_RECOMMENDED;
2856 			ent->eax |= HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED;
2857 			if (evmcs_ver)
2858 				ent->eax |= HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
2859 			if (!cpu_smt_possible())
2860 				ent->eax |= HV_X64_NO_NONARCH_CORESHARING;
2861 
2862 			ent->eax |= HV_DEPRECATING_AEOI_RECOMMENDED;
2863 			/*
2864 			 * Default number of spinlock retry attempts, matches
2865 			 * HyperV 2016.
2866 			 */
2867 			ent->ebx = 0x00000FFF;
2868 
2869 			break;
2870 
2871 		case HYPERV_CPUID_IMPLEMENT_LIMITS:
2872 			/* Maximum number of virtual processors */
2873 			ent->eax = KVM_MAX_VCPUS;
2874 			/*
2875 			 * Maximum number of logical processors, matches
2876 			 * HyperV 2016.
2877 			 */
2878 			ent->ebx = 64;
2879 
2880 			break;
2881 
2882 		case HYPERV_CPUID_NESTED_FEATURES:
2883 			ent->eax = evmcs_ver;
2884 			ent->eax |= HV_X64_NESTED_DIRECT_FLUSH;
2885 			ent->eax |= HV_X64_NESTED_MSR_BITMAP;
2886 			ent->ebx |= HV_X64_NESTED_EVMCS1_PERF_GLOBAL_CTRL;
2887 			break;
2888 
2889 		case HYPERV_CPUID_SYNDBG_VENDOR_AND_MAX_FUNCTIONS:
2890 			memcpy(signature, "Linux KVM Hv", 12);
2891 
2892 			ent->eax = 0;
2893 			ent->ebx = signature[0];
2894 			ent->ecx = signature[1];
2895 			ent->edx = signature[2];
2896 			break;
2897 
2898 		case HYPERV_CPUID_SYNDBG_INTERFACE:
2899 			memcpy(signature, "VS#1\0\0\0\0\0\0\0\0", 12);
2900 			ent->eax = signature[0];
2901 			break;
2902 
2903 		case HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES:
2904 			ent->eax |= HV_X64_SYNDBG_CAP_ALLOW_KERNEL_DEBUGGING;
2905 			break;
2906 
2907 		default:
2908 			break;
2909 		}
2910 	}
2911 
2912 	if (copy_to_user(entries, cpuid_entries,
2913 			 nent * sizeof(struct kvm_cpuid_entry2)))
2914 		return -EFAULT;
2915 
2916 	return 0;
2917 }
2918