xref: /openbmc/linux/arch/x86/kvm/hyperv.c (revision 2d091155)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * KVM Microsoft Hyper-V emulation
4  *
5  * derived from arch/x86/kvm/x86.c
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright (C) 2008 Qumranet, Inc.
9  * Copyright IBM Corporation, 2008
10  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11  * Copyright (C) 2015 Andrey Smetanin <asmetanin@virtuozzo.com>
12  *
13  * Authors:
14  *   Avi Kivity   <avi@qumranet.com>
15  *   Yaniv Kamay  <yaniv@qumranet.com>
16  *   Amit Shah    <amit.shah@qumranet.com>
17  *   Ben-Ami Yassour <benami@il.ibm.com>
18  *   Andrey Smetanin <asmetanin@virtuozzo.com>
19  */
20 
21 #include "x86.h"
22 #include "lapic.h"
23 #include "ioapic.h"
24 #include "cpuid.h"
25 #include "hyperv.h"
26 #include "xen.h"
27 
28 #include <linux/cpu.h>
29 #include <linux/kvm_host.h>
30 #include <linux/highmem.h>
31 #include <linux/sched/cputime.h>
32 #include <linux/eventfd.h>
33 
34 #include <asm/apicdef.h>
35 #include <trace/events/kvm.h>
36 
37 #include "trace.h"
38 #include "irq.h"
39 #include "fpu.h"
40 
41 /* "Hv#1" signature */
42 #define HYPERV_CPUID_SIGNATURE_EAX 0x31237648
43 
44 #define KVM_HV_MAX_SPARSE_VCPU_SET_BITS DIV_ROUND_UP(KVM_MAX_VCPUS, 64)
45 
46 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
47 				bool vcpu_kick);
48 
49 static inline u64 synic_read_sint(struct kvm_vcpu_hv_synic *synic, int sint)
50 {
51 	return atomic64_read(&synic->sint[sint]);
52 }
53 
54 static inline int synic_get_sint_vector(u64 sint_value)
55 {
56 	if (sint_value & HV_SYNIC_SINT_MASKED)
57 		return -1;
58 	return sint_value & HV_SYNIC_SINT_VECTOR_MASK;
59 }
60 
61 static bool synic_has_vector_connected(struct kvm_vcpu_hv_synic *synic,
62 				      int vector)
63 {
64 	int i;
65 
66 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
67 		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
68 			return true;
69 	}
70 	return false;
71 }
72 
73 static bool synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic *synic,
74 				     int vector)
75 {
76 	int i;
77 	u64 sint_value;
78 
79 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
80 		sint_value = synic_read_sint(synic, i);
81 		if (synic_get_sint_vector(sint_value) == vector &&
82 		    sint_value & HV_SYNIC_SINT_AUTO_EOI)
83 			return true;
84 	}
85 	return false;
86 }
87 
88 static void synic_update_vector(struct kvm_vcpu_hv_synic *synic,
89 				int vector)
90 {
91 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
92 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
93 	int auto_eoi_old, auto_eoi_new;
94 
95 	if (vector < HV_SYNIC_FIRST_VALID_VECTOR)
96 		return;
97 
98 	if (synic_has_vector_connected(synic, vector))
99 		__set_bit(vector, synic->vec_bitmap);
100 	else
101 		__clear_bit(vector, synic->vec_bitmap);
102 
103 	auto_eoi_old = bitmap_weight(synic->auto_eoi_bitmap, 256);
104 
105 	if (synic_has_vector_auto_eoi(synic, vector))
106 		__set_bit(vector, synic->auto_eoi_bitmap);
107 	else
108 		__clear_bit(vector, synic->auto_eoi_bitmap);
109 
110 	auto_eoi_new = bitmap_weight(synic->auto_eoi_bitmap, 256);
111 
112 	if (!!auto_eoi_old == !!auto_eoi_new)
113 		return;
114 
115 	if (!enable_apicv)
116 		return;
117 
118 	down_write(&vcpu->kvm->arch.apicv_update_lock);
119 
120 	if (auto_eoi_new)
121 		hv->synic_auto_eoi_used++;
122 	else
123 		hv->synic_auto_eoi_used--;
124 
125 	/*
126 	 * Inhibit APICv if any vCPU is using SynIC's AutoEOI, which relies on
127 	 * the hypervisor to manually inject IRQs.
128 	 */
129 	__kvm_set_or_clear_apicv_inhibit(vcpu->kvm,
130 					 APICV_INHIBIT_REASON_HYPERV,
131 					 !!hv->synic_auto_eoi_used);
132 
133 	up_write(&vcpu->kvm->arch.apicv_update_lock);
134 }
135 
136 static int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint,
137 			  u64 data, bool host)
138 {
139 	int vector, old_vector;
140 	bool masked;
141 
142 	vector = data & HV_SYNIC_SINT_VECTOR_MASK;
143 	masked = data & HV_SYNIC_SINT_MASKED;
144 
145 	/*
146 	 * Valid vectors are 16-255, however, nested Hyper-V attempts to write
147 	 * default '0x10000' value on boot and this should not #GP. We need to
148 	 * allow zero-initing the register from host as well.
149 	 */
150 	if (vector < HV_SYNIC_FIRST_VALID_VECTOR && !host && !masked)
151 		return 1;
152 	/*
153 	 * Guest may configure multiple SINTs to use the same vector, so
154 	 * we maintain a bitmap of vectors handled by synic, and a
155 	 * bitmap of vectors with auto-eoi behavior.  The bitmaps are
156 	 * updated here, and atomically queried on fast paths.
157 	 */
158 	old_vector = synic_read_sint(synic, sint) & HV_SYNIC_SINT_VECTOR_MASK;
159 
160 	atomic64_set(&synic->sint[sint], data);
161 
162 	synic_update_vector(synic, old_vector);
163 
164 	synic_update_vector(synic, vector);
165 
166 	/* Load SynIC vectors into EOI exit bitmap */
167 	kvm_make_request(KVM_REQ_SCAN_IOAPIC, hv_synic_to_vcpu(synic));
168 	return 0;
169 }
170 
171 static struct kvm_vcpu *get_vcpu_by_vpidx(struct kvm *kvm, u32 vpidx)
172 {
173 	struct kvm_vcpu *vcpu = NULL;
174 	unsigned long i;
175 
176 	if (vpidx >= KVM_MAX_VCPUS)
177 		return NULL;
178 
179 	vcpu = kvm_get_vcpu(kvm, vpidx);
180 	if (vcpu && kvm_hv_get_vpindex(vcpu) == vpidx)
181 		return vcpu;
182 	kvm_for_each_vcpu(i, vcpu, kvm)
183 		if (kvm_hv_get_vpindex(vcpu) == vpidx)
184 			return vcpu;
185 	return NULL;
186 }
187 
188 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vpidx)
189 {
190 	struct kvm_vcpu *vcpu;
191 	struct kvm_vcpu_hv_synic *synic;
192 
193 	vcpu = get_vcpu_by_vpidx(kvm, vpidx);
194 	if (!vcpu || !to_hv_vcpu(vcpu))
195 		return NULL;
196 	synic = to_hv_synic(vcpu);
197 	return (synic->active) ? synic : NULL;
198 }
199 
200 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
201 {
202 	struct kvm *kvm = vcpu->kvm;
203 	struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
204 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
205 	struct kvm_vcpu_hv_stimer *stimer;
206 	int gsi, idx;
207 
208 	trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
209 
210 	/* Try to deliver pending Hyper-V SynIC timers messages */
211 	for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
212 		stimer = &hv_vcpu->stimer[idx];
213 		if (stimer->msg_pending && stimer->config.enable &&
214 		    !stimer->config.direct_mode &&
215 		    stimer->config.sintx == sint)
216 			stimer_mark_pending(stimer, false);
217 	}
218 
219 	idx = srcu_read_lock(&kvm->irq_srcu);
220 	gsi = atomic_read(&synic->sint_to_gsi[sint]);
221 	if (gsi != -1)
222 		kvm_notify_acked_gsi(kvm, gsi);
223 	srcu_read_unlock(&kvm->irq_srcu, idx);
224 }
225 
226 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
227 {
228 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
229 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
230 
231 	hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
232 	hv_vcpu->exit.u.synic.msr = msr;
233 	hv_vcpu->exit.u.synic.control = synic->control;
234 	hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
235 	hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
236 
237 	kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
238 }
239 
240 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
241 			 u32 msr, u64 data, bool host)
242 {
243 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
244 	int ret;
245 
246 	if (!synic->active && (!host || data))
247 		return 1;
248 
249 	trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
250 
251 	ret = 0;
252 	switch (msr) {
253 	case HV_X64_MSR_SCONTROL:
254 		synic->control = data;
255 		if (!host)
256 			synic_exit(synic, msr);
257 		break;
258 	case HV_X64_MSR_SVERSION:
259 		if (!host) {
260 			ret = 1;
261 			break;
262 		}
263 		synic->version = data;
264 		break;
265 	case HV_X64_MSR_SIEFP:
266 		if ((data & HV_SYNIC_SIEFP_ENABLE) && !host &&
267 		    !synic->dont_zero_synic_pages)
268 			if (kvm_clear_guest(vcpu->kvm,
269 					    data & PAGE_MASK, PAGE_SIZE)) {
270 				ret = 1;
271 				break;
272 			}
273 		synic->evt_page = data;
274 		if (!host)
275 			synic_exit(synic, msr);
276 		break;
277 	case HV_X64_MSR_SIMP:
278 		if ((data & HV_SYNIC_SIMP_ENABLE) && !host &&
279 		    !synic->dont_zero_synic_pages)
280 			if (kvm_clear_guest(vcpu->kvm,
281 					    data & PAGE_MASK, PAGE_SIZE)) {
282 				ret = 1;
283 				break;
284 			}
285 		synic->msg_page = data;
286 		if (!host)
287 			synic_exit(synic, msr);
288 		break;
289 	case HV_X64_MSR_EOM: {
290 		int i;
291 
292 		if (!synic->active)
293 			break;
294 
295 		for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
296 			kvm_hv_notify_acked_sint(vcpu, i);
297 		break;
298 	}
299 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
300 		ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
301 		break;
302 	default:
303 		ret = 1;
304 		break;
305 	}
306 	return ret;
307 }
308 
309 static bool kvm_hv_is_syndbg_enabled(struct kvm_vcpu *vcpu)
310 {
311 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
312 
313 	return hv_vcpu->cpuid_cache.syndbg_cap_eax &
314 		HV_X64_SYNDBG_CAP_ALLOW_KERNEL_DEBUGGING;
315 }
316 
317 static int kvm_hv_syndbg_complete_userspace(struct kvm_vcpu *vcpu)
318 {
319 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
320 
321 	if (vcpu->run->hyperv.u.syndbg.msr == HV_X64_MSR_SYNDBG_CONTROL)
322 		hv->hv_syndbg.control.status =
323 			vcpu->run->hyperv.u.syndbg.status;
324 	return 1;
325 }
326 
327 static void syndbg_exit(struct kvm_vcpu *vcpu, u32 msr)
328 {
329 	struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
330 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
331 
332 	hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNDBG;
333 	hv_vcpu->exit.u.syndbg.msr = msr;
334 	hv_vcpu->exit.u.syndbg.control = syndbg->control.control;
335 	hv_vcpu->exit.u.syndbg.send_page = syndbg->control.send_page;
336 	hv_vcpu->exit.u.syndbg.recv_page = syndbg->control.recv_page;
337 	hv_vcpu->exit.u.syndbg.pending_page = syndbg->control.pending_page;
338 	vcpu->arch.complete_userspace_io =
339 			kvm_hv_syndbg_complete_userspace;
340 
341 	kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
342 }
343 
344 static int syndbg_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
345 {
346 	struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
347 
348 	if (!kvm_hv_is_syndbg_enabled(vcpu) && !host)
349 		return 1;
350 
351 	trace_kvm_hv_syndbg_set_msr(vcpu->vcpu_id,
352 				    to_hv_vcpu(vcpu)->vp_index, msr, data);
353 	switch (msr) {
354 	case HV_X64_MSR_SYNDBG_CONTROL:
355 		syndbg->control.control = data;
356 		if (!host)
357 			syndbg_exit(vcpu, msr);
358 		break;
359 	case HV_X64_MSR_SYNDBG_STATUS:
360 		syndbg->control.status = data;
361 		break;
362 	case HV_X64_MSR_SYNDBG_SEND_BUFFER:
363 		syndbg->control.send_page = data;
364 		break;
365 	case HV_X64_MSR_SYNDBG_RECV_BUFFER:
366 		syndbg->control.recv_page = data;
367 		break;
368 	case HV_X64_MSR_SYNDBG_PENDING_BUFFER:
369 		syndbg->control.pending_page = data;
370 		if (!host)
371 			syndbg_exit(vcpu, msr);
372 		break;
373 	case HV_X64_MSR_SYNDBG_OPTIONS:
374 		syndbg->options = data;
375 		break;
376 	default:
377 		break;
378 	}
379 
380 	return 0;
381 }
382 
383 static int syndbg_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
384 {
385 	struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
386 
387 	if (!kvm_hv_is_syndbg_enabled(vcpu) && !host)
388 		return 1;
389 
390 	switch (msr) {
391 	case HV_X64_MSR_SYNDBG_CONTROL:
392 		*pdata = syndbg->control.control;
393 		break;
394 	case HV_X64_MSR_SYNDBG_STATUS:
395 		*pdata = syndbg->control.status;
396 		break;
397 	case HV_X64_MSR_SYNDBG_SEND_BUFFER:
398 		*pdata = syndbg->control.send_page;
399 		break;
400 	case HV_X64_MSR_SYNDBG_RECV_BUFFER:
401 		*pdata = syndbg->control.recv_page;
402 		break;
403 	case HV_X64_MSR_SYNDBG_PENDING_BUFFER:
404 		*pdata = syndbg->control.pending_page;
405 		break;
406 	case HV_X64_MSR_SYNDBG_OPTIONS:
407 		*pdata = syndbg->options;
408 		break;
409 	default:
410 		break;
411 	}
412 
413 	trace_kvm_hv_syndbg_get_msr(vcpu->vcpu_id, kvm_hv_get_vpindex(vcpu), msr, *pdata);
414 
415 	return 0;
416 }
417 
418 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata,
419 			 bool host)
420 {
421 	int ret;
422 
423 	if (!synic->active && !host)
424 		return 1;
425 
426 	ret = 0;
427 	switch (msr) {
428 	case HV_X64_MSR_SCONTROL:
429 		*pdata = synic->control;
430 		break;
431 	case HV_X64_MSR_SVERSION:
432 		*pdata = synic->version;
433 		break;
434 	case HV_X64_MSR_SIEFP:
435 		*pdata = synic->evt_page;
436 		break;
437 	case HV_X64_MSR_SIMP:
438 		*pdata = synic->msg_page;
439 		break;
440 	case HV_X64_MSR_EOM:
441 		*pdata = 0;
442 		break;
443 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
444 		*pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
445 		break;
446 	default:
447 		ret = 1;
448 		break;
449 	}
450 	return ret;
451 }
452 
453 static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
454 {
455 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
456 	struct kvm_lapic_irq irq;
457 	int ret, vector;
458 
459 	if (KVM_BUG_ON(!lapic_in_kernel(vcpu), vcpu->kvm))
460 		return -EINVAL;
461 
462 	if (sint >= ARRAY_SIZE(synic->sint))
463 		return -EINVAL;
464 
465 	vector = synic_get_sint_vector(synic_read_sint(synic, sint));
466 	if (vector < 0)
467 		return -ENOENT;
468 
469 	memset(&irq, 0, sizeof(irq));
470 	irq.shorthand = APIC_DEST_SELF;
471 	irq.dest_mode = APIC_DEST_PHYSICAL;
472 	irq.delivery_mode = APIC_DM_FIXED;
473 	irq.vector = vector;
474 	irq.level = 1;
475 
476 	ret = kvm_irq_delivery_to_apic(vcpu->kvm, vcpu->arch.apic, &irq, NULL);
477 	trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
478 	return ret;
479 }
480 
481 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vpidx, u32 sint)
482 {
483 	struct kvm_vcpu_hv_synic *synic;
484 
485 	synic = synic_get(kvm, vpidx);
486 	if (!synic)
487 		return -EINVAL;
488 
489 	return synic_set_irq(synic, sint);
490 }
491 
492 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
493 {
494 	struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
495 	int i;
496 
497 	trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
498 
499 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
500 		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
501 			kvm_hv_notify_acked_sint(vcpu, i);
502 }
503 
504 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi)
505 {
506 	struct kvm_vcpu_hv_synic *synic;
507 
508 	synic = synic_get(kvm, vpidx);
509 	if (!synic)
510 		return -EINVAL;
511 
512 	if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
513 		return -EINVAL;
514 
515 	atomic_set(&synic->sint_to_gsi[sint], gsi);
516 	return 0;
517 }
518 
519 void kvm_hv_irq_routing_update(struct kvm *kvm)
520 {
521 	struct kvm_irq_routing_table *irq_rt;
522 	struct kvm_kernel_irq_routing_entry *e;
523 	u32 gsi;
524 
525 	irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
526 					lockdep_is_held(&kvm->irq_lock));
527 
528 	for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
529 		hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
530 			if (e->type == KVM_IRQ_ROUTING_HV_SINT)
531 				kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
532 						    e->hv_sint.sint, gsi);
533 		}
534 	}
535 }
536 
537 static void synic_init(struct kvm_vcpu_hv_synic *synic)
538 {
539 	int i;
540 
541 	memset(synic, 0, sizeof(*synic));
542 	synic->version = HV_SYNIC_VERSION_1;
543 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
544 		atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
545 		atomic_set(&synic->sint_to_gsi[i], -1);
546 	}
547 }
548 
549 static u64 get_time_ref_counter(struct kvm *kvm)
550 {
551 	struct kvm_hv *hv = to_kvm_hv(kvm);
552 	struct kvm_vcpu *vcpu;
553 	u64 tsc;
554 
555 	/*
556 	 * Fall back to get_kvmclock_ns() when TSC page hasn't been set up,
557 	 * is broken, disabled or being updated.
558 	 */
559 	if (hv->hv_tsc_page_status != HV_TSC_PAGE_SET)
560 		return div_u64(get_kvmclock_ns(kvm), 100);
561 
562 	vcpu = kvm_get_vcpu(kvm, 0);
563 	tsc = kvm_read_l1_tsc(vcpu, rdtsc());
564 	return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64)
565 		+ hv->tsc_ref.tsc_offset;
566 }
567 
568 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
569 				bool vcpu_kick)
570 {
571 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
572 
573 	set_bit(stimer->index,
574 		to_hv_vcpu(vcpu)->stimer_pending_bitmap);
575 	kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
576 	if (vcpu_kick)
577 		kvm_vcpu_kick(vcpu);
578 }
579 
580 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
581 {
582 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
583 
584 	trace_kvm_hv_stimer_cleanup(hv_stimer_to_vcpu(stimer)->vcpu_id,
585 				    stimer->index);
586 
587 	hrtimer_cancel(&stimer->timer);
588 	clear_bit(stimer->index,
589 		  to_hv_vcpu(vcpu)->stimer_pending_bitmap);
590 	stimer->msg_pending = false;
591 	stimer->exp_time = 0;
592 }
593 
594 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
595 {
596 	struct kvm_vcpu_hv_stimer *stimer;
597 
598 	stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
599 	trace_kvm_hv_stimer_callback(hv_stimer_to_vcpu(stimer)->vcpu_id,
600 				     stimer->index);
601 	stimer_mark_pending(stimer, true);
602 
603 	return HRTIMER_NORESTART;
604 }
605 
606 /*
607  * stimer_start() assumptions:
608  * a) stimer->count is not equal to 0
609  * b) stimer->config has HV_STIMER_ENABLE flag
610  */
611 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
612 {
613 	u64 time_now;
614 	ktime_t ktime_now;
615 
616 	time_now = get_time_ref_counter(hv_stimer_to_vcpu(stimer)->kvm);
617 	ktime_now = ktime_get();
618 
619 	if (stimer->config.periodic) {
620 		if (stimer->exp_time) {
621 			if (time_now >= stimer->exp_time) {
622 				u64 remainder;
623 
624 				div64_u64_rem(time_now - stimer->exp_time,
625 					      stimer->count, &remainder);
626 				stimer->exp_time =
627 					time_now + (stimer->count - remainder);
628 			}
629 		} else
630 			stimer->exp_time = time_now + stimer->count;
631 
632 		trace_kvm_hv_stimer_start_periodic(
633 					hv_stimer_to_vcpu(stimer)->vcpu_id,
634 					stimer->index,
635 					time_now, stimer->exp_time);
636 
637 		hrtimer_start(&stimer->timer,
638 			      ktime_add_ns(ktime_now,
639 					   100 * (stimer->exp_time - time_now)),
640 			      HRTIMER_MODE_ABS);
641 		return 0;
642 	}
643 	stimer->exp_time = stimer->count;
644 	if (time_now >= stimer->count) {
645 		/*
646 		 * Expire timer according to Hypervisor Top-Level Functional
647 		 * specification v4(15.3.1):
648 		 * "If a one shot is enabled and the specified count is in
649 		 * the past, it will expire immediately."
650 		 */
651 		stimer_mark_pending(stimer, false);
652 		return 0;
653 	}
654 
655 	trace_kvm_hv_stimer_start_one_shot(hv_stimer_to_vcpu(stimer)->vcpu_id,
656 					   stimer->index,
657 					   time_now, stimer->count);
658 
659 	hrtimer_start(&stimer->timer,
660 		      ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)),
661 		      HRTIMER_MODE_ABS);
662 	return 0;
663 }
664 
665 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
666 			     bool host)
667 {
668 	union hv_stimer_config new_config = {.as_uint64 = config},
669 		old_config = {.as_uint64 = stimer->config.as_uint64};
670 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
671 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
672 	struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
673 
674 	if (!synic->active && (!host || config))
675 		return 1;
676 
677 	if (unlikely(!host && hv_vcpu->enforce_cpuid && new_config.direct_mode &&
678 		     !(hv_vcpu->cpuid_cache.features_edx &
679 		       HV_STIMER_DIRECT_MODE_AVAILABLE)))
680 		return 1;
681 
682 	trace_kvm_hv_stimer_set_config(hv_stimer_to_vcpu(stimer)->vcpu_id,
683 				       stimer->index, config, host);
684 
685 	stimer_cleanup(stimer);
686 	if (old_config.enable &&
687 	    !new_config.direct_mode && new_config.sintx == 0)
688 		new_config.enable = 0;
689 	stimer->config.as_uint64 = new_config.as_uint64;
690 
691 	if (stimer->config.enable)
692 		stimer_mark_pending(stimer, false);
693 
694 	return 0;
695 }
696 
697 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
698 			    bool host)
699 {
700 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
701 	struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
702 
703 	if (!synic->active && (!host || count))
704 		return 1;
705 
706 	trace_kvm_hv_stimer_set_count(hv_stimer_to_vcpu(stimer)->vcpu_id,
707 				      stimer->index, count, host);
708 
709 	stimer_cleanup(stimer);
710 	stimer->count = count;
711 	if (stimer->count == 0)
712 		stimer->config.enable = 0;
713 	else if (stimer->config.auto_enable)
714 		stimer->config.enable = 1;
715 
716 	if (stimer->config.enable)
717 		stimer_mark_pending(stimer, false);
718 
719 	return 0;
720 }
721 
722 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
723 {
724 	*pconfig = stimer->config.as_uint64;
725 	return 0;
726 }
727 
728 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
729 {
730 	*pcount = stimer->count;
731 	return 0;
732 }
733 
734 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
735 			     struct hv_message *src_msg, bool no_retry)
736 {
737 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
738 	int msg_off = offsetof(struct hv_message_page, sint_message[sint]);
739 	gfn_t msg_page_gfn;
740 	struct hv_message_header hv_hdr;
741 	int r;
742 
743 	if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
744 		return -ENOENT;
745 
746 	msg_page_gfn = synic->msg_page >> PAGE_SHIFT;
747 
748 	/*
749 	 * Strictly following the spec-mandated ordering would assume setting
750 	 * .msg_pending before checking .message_type.  However, this function
751 	 * is only called in vcpu context so the entire update is atomic from
752 	 * guest POV and thus the exact order here doesn't matter.
753 	 */
754 	r = kvm_vcpu_read_guest_page(vcpu, msg_page_gfn, &hv_hdr.message_type,
755 				     msg_off + offsetof(struct hv_message,
756 							header.message_type),
757 				     sizeof(hv_hdr.message_type));
758 	if (r < 0)
759 		return r;
760 
761 	if (hv_hdr.message_type != HVMSG_NONE) {
762 		if (no_retry)
763 			return 0;
764 
765 		hv_hdr.message_flags.msg_pending = 1;
766 		r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn,
767 					      &hv_hdr.message_flags,
768 					      msg_off +
769 					      offsetof(struct hv_message,
770 						       header.message_flags),
771 					      sizeof(hv_hdr.message_flags));
772 		if (r < 0)
773 			return r;
774 		return -EAGAIN;
775 	}
776 
777 	r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn, src_msg, msg_off,
778 				      sizeof(src_msg->header) +
779 				      src_msg->header.payload_size);
780 	if (r < 0)
781 		return r;
782 
783 	r = synic_set_irq(synic, sint);
784 	if (r < 0)
785 		return r;
786 	if (r == 0)
787 		return -EFAULT;
788 	return 0;
789 }
790 
791 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
792 {
793 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
794 	struct hv_message *msg = &stimer->msg;
795 	struct hv_timer_message_payload *payload =
796 			(struct hv_timer_message_payload *)&msg->u.payload;
797 
798 	/*
799 	 * To avoid piling up periodic ticks, don't retry message
800 	 * delivery for them (within "lazy" lost ticks policy).
801 	 */
802 	bool no_retry = stimer->config.periodic;
803 
804 	payload->expiration_time = stimer->exp_time;
805 	payload->delivery_time = get_time_ref_counter(vcpu->kvm);
806 	return synic_deliver_msg(to_hv_synic(vcpu),
807 				 stimer->config.sintx, msg,
808 				 no_retry);
809 }
810 
811 static int stimer_notify_direct(struct kvm_vcpu_hv_stimer *stimer)
812 {
813 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
814 	struct kvm_lapic_irq irq = {
815 		.delivery_mode = APIC_DM_FIXED,
816 		.vector = stimer->config.apic_vector
817 	};
818 
819 	if (lapic_in_kernel(vcpu))
820 		return !kvm_apic_set_irq(vcpu, &irq, NULL);
821 	return 0;
822 }
823 
824 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
825 {
826 	int r, direct = stimer->config.direct_mode;
827 
828 	stimer->msg_pending = true;
829 	if (!direct)
830 		r = stimer_send_msg(stimer);
831 	else
832 		r = stimer_notify_direct(stimer);
833 	trace_kvm_hv_stimer_expiration(hv_stimer_to_vcpu(stimer)->vcpu_id,
834 				       stimer->index, direct, r);
835 	if (!r) {
836 		stimer->msg_pending = false;
837 		if (!(stimer->config.periodic))
838 			stimer->config.enable = 0;
839 	}
840 }
841 
842 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
843 {
844 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
845 	struct kvm_vcpu_hv_stimer *stimer;
846 	u64 time_now, exp_time;
847 	int i;
848 
849 	if (!hv_vcpu)
850 		return;
851 
852 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
853 		if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
854 			stimer = &hv_vcpu->stimer[i];
855 			if (stimer->config.enable) {
856 				exp_time = stimer->exp_time;
857 
858 				if (exp_time) {
859 					time_now =
860 						get_time_ref_counter(vcpu->kvm);
861 					if (time_now >= exp_time)
862 						stimer_expiration(stimer);
863 				}
864 
865 				if ((stimer->config.enable) &&
866 				    stimer->count) {
867 					if (!stimer->msg_pending)
868 						stimer_start(stimer);
869 				} else
870 					stimer_cleanup(stimer);
871 			}
872 		}
873 }
874 
875 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
876 {
877 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
878 	int i;
879 
880 	if (!hv_vcpu)
881 		return;
882 
883 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
884 		stimer_cleanup(&hv_vcpu->stimer[i]);
885 
886 	kfree(hv_vcpu);
887 	vcpu->arch.hyperv = NULL;
888 }
889 
890 bool kvm_hv_assist_page_enabled(struct kvm_vcpu *vcpu)
891 {
892 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
893 
894 	if (!hv_vcpu)
895 		return false;
896 
897 	if (!(hv_vcpu->hv_vapic & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE))
898 		return false;
899 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
900 }
901 EXPORT_SYMBOL_GPL(kvm_hv_assist_page_enabled);
902 
903 bool kvm_hv_get_assist_page(struct kvm_vcpu *vcpu,
904 			    struct hv_vp_assist_page *assist_page)
905 {
906 	if (!kvm_hv_assist_page_enabled(vcpu))
907 		return false;
908 	return !kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data,
909 				      assist_page, sizeof(*assist_page));
910 }
911 EXPORT_SYMBOL_GPL(kvm_hv_get_assist_page);
912 
913 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
914 {
915 	struct hv_message *msg = &stimer->msg;
916 	struct hv_timer_message_payload *payload =
917 			(struct hv_timer_message_payload *)&msg->u.payload;
918 
919 	memset(&msg->header, 0, sizeof(msg->header));
920 	msg->header.message_type = HVMSG_TIMER_EXPIRED;
921 	msg->header.payload_size = sizeof(*payload);
922 
923 	payload->timer_index = stimer->index;
924 	payload->expiration_time = 0;
925 	payload->delivery_time = 0;
926 }
927 
928 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
929 {
930 	memset(stimer, 0, sizeof(*stimer));
931 	stimer->index = timer_index;
932 	hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
933 	stimer->timer.function = stimer_timer_callback;
934 	stimer_prepare_msg(stimer);
935 }
936 
937 static int kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
938 {
939 	struct kvm_vcpu_hv *hv_vcpu;
940 	int i;
941 
942 	hv_vcpu = kzalloc(sizeof(struct kvm_vcpu_hv), GFP_KERNEL_ACCOUNT);
943 	if (!hv_vcpu)
944 		return -ENOMEM;
945 
946 	vcpu->arch.hyperv = hv_vcpu;
947 	hv_vcpu->vcpu = vcpu;
948 
949 	synic_init(&hv_vcpu->synic);
950 
951 	bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
952 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
953 		stimer_init(&hv_vcpu->stimer[i], i);
954 
955 	hv_vcpu->vp_index = vcpu->vcpu_idx;
956 
957 	return 0;
958 }
959 
960 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
961 {
962 	struct kvm_vcpu_hv_synic *synic;
963 	int r;
964 
965 	if (!to_hv_vcpu(vcpu)) {
966 		r = kvm_hv_vcpu_init(vcpu);
967 		if (r)
968 			return r;
969 	}
970 
971 	synic = to_hv_synic(vcpu);
972 
973 	synic->active = true;
974 	synic->dont_zero_synic_pages = dont_zero_synic_pages;
975 	synic->control = HV_SYNIC_CONTROL_ENABLE;
976 	return 0;
977 }
978 
979 static bool kvm_hv_msr_partition_wide(u32 msr)
980 {
981 	bool r = false;
982 
983 	switch (msr) {
984 	case HV_X64_MSR_GUEST_OS_ID:
985 	case HV_X64_MSR_HYPERCALL:
986 	case HV_X64_MSR_REFERENCE_TSC:
987 	case HV_X64_MSR_TIME_REF_COUNT:
988 	case HV_X64_MSR_CRASH_CTL:
989 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
990 	case HV_X64_MSR_RESET:
991 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
992 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
993 	case HV_X64_MSR_TSC_EMULATION_STATUS:
994 	case HV_X64_MSR_SYNDBG_OPTIONS:
995 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
996 		r = true;
997 		break;
998 	}
999 
1000 	return r;
1001 }
1002 
1003 static int kvm_hv_msr_get_crash_data(struct kvm *kvm, u32 index, u64 *pdata)
1004 {
1005 	struct kvm_hv *hv = to_kvm_hv(kvm);
1006 	size_t size = ARRAY_SIZE(hv->hv_crash_param);
1007 
1008 	if (WARN_ON_ONCE(index >= size))
1009 		return -EINVAL;
1010 
1011 	*pdata = hv->hv_crash_param[array_index_nospec(index, size)];
1012 	return 0;
1013 }
1014 
1015 static int kvm_hv_msr_get_crash_ctl(struct kvm *kvm, u64 *pdata)
1016 {
1017 	struct kvm_hv *hv = to_kvm_hv(kvm);
1018 
1019 	*pdata = hv->hv_crash_ctl;
1020 	return 0;
1021 }
1022 
1023 static int kvm_hv_msr_set_crash_ctl(struct kvm *kvm, u64 data)
1024 {
1025 	struct kvm_hv *hv = to_kvm_hv(kvm);
1026 
1027 	hv->hv_crash_ctl = data & HV_CRASH_CTL_CRASH_NOTIFY;
1028 
1029 	return 0;
1030 }
1031 
1032 static int kvm_hv_msr_set_crash_data(struct kvm *kvm, u32 index, u64 data)
1033 {
1034 	struct kvm_hv *hv = to_kvm_hv(kvm);
1035 	size_t size = ARRAY_SIZE(hv->hv_crash_param);
1036 
1037 	if (WARN_ON_ONCE(index >= size))
1038 		return -EINVAL;
1039 
1040 	hv->hv_crash_param[array_index_nospec(index, size)] = data;
1041 	return 0;
1042 }
1043 
1044 /*
1045  * The kvmclock and Hyper-V TSC page use similar formulas, and converting
1046  * between them is possible:
1047  *
1048  * kvmclock formula:
1049  *    nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
1050  *           + system_time
1051  *
1052  * Hyper-V formula:
1053  *    nsec/100 = ticks * scale / 2^64 + offset
1054  *
1055  * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
1056  * By dividing the kvmclock formula by 100 and equating what's left we get:
1057  *    ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1058  *            scale / 2^64 =         tsc_to_system_mul * 2^(tsc_shift-32) / 100
1059  *            scale        =         tsc_to_system_mul * 2^(32+tsc_shift) / 100
1060  *
1061  * Now expand the kvmclock formula and divide by 100:
1062  *    nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
1063  *           - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
1064  *           + system_time
1065  *    nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1066  *               - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1067  *               + system_time / 100
1068  *
1069  * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
1070  *    nsec/100 = ticks * scale / 2^64
1071  *               - tsc_timestamp * scale / 2^64
1072  *               + system_time / 100
1073  *
1074  * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
1075  *    offset = system_time / 100 - tsc_timestamp * scale / 2^64
1076  *
1077  * These two equivalencies are implemented in this function.
1078  */
1079 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
1080 					struct ms_hyperv_tsc_page *tsc_ref)
1081 {
1082 	u64 max_mul;
1083 
1084 	if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
1085 		return false;
1086 
1087 	/*
1088 	 * check if scale would overflow, if so we use the time ref counter
1089 	 *    tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
1090 	 *    tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
1091 	 *    tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
1092 	 */
1093 	max_mul = 100ull << (32 - hv_clock->tsc_shift);
1094 	if (hv_clock->tsc_to_system_mul >= max_mul)
1095 		return false;
1096 
1097 	/*
1098 	 * Otherwise compute the scale and offset according to the formulas
1099 	 * derived above.
1100 	 */
1101 	tsc_ref->tsc_scale =
1102 		mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
1103 				hv_clock->tsc_to_system_mul,
1104 				100);
1105 
1106 	tsc_ref->tsc_offset = hv_clock->system_time;
1107 	do_div(tsc_ref->tsc_offset, 100);
1108 	tsc_ref->tsc_offset -=
1109 		mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
1110 	return true;
1111 }
1112 
1113 /*
1114  * Don't touch TSC page values if the guest has opted for TSC emulation after
1115  * migration. KVM doesn't fully support reenlightenment notifications and TSC
1116  * access emulation and Hyper-V is known to expect the values in TSC page to
1117  * stay constant before TSC access emulation is disabled from guest side
1118  * (HV_X64_MSR_TSC_EMULATION_STATUS). KVM userspace is expected to preserve TSC
1119  * frequency and guest visible TSC value across migration (and prevent it when
1120  * TSC scaling is unsupported).
1121  */
1122 static inline bool tsc_page_update_unsafe(struct kvm_hv *hv)
1123 {
1124 	return (hv->hv_tsc_page_status != HV_TSC_PAGE_GUEST_CHANGED) &&
1125 		hv->hv_tsc_emulation_control;
1126 }
1127 
1128 void kvm_hv_setup_tsc_page(struct kvm *kvm,
1129 			   struct pvclock_vcpu_time_info *hv_clock)
1130 {
1131 	struct kvm_hv *hv = to_kvm_hv(kvm);
1132 	u32 tsc_seq;
1133 	u64 gfn;
1134 
1135 	BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
1136 	BUILD_BUG_ON(offsetof(struct ms_hyperv_tsc_page, tsc_sequence) != 0);
1137 
1138 	if (hv->hv_tsc_page_status == HV_TSC_PAGE_BROKEN ||
1139 	    hv->hv_tsc_page_status == HV_TSC_PAGE_UNSET)
1140 		return;
1141 
1142 	mutex_lock(&hv->hv_lock);
1143 	if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1144 		goto out_unlock;
1145 
1146 	gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
1147 	/*
1148 	 * Because the TSC parameters only vary when there is a
1149 	 * change in the master clock, do not bother with caching.
1150 	 */
1151 	if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
1152 				    &tsc_seq, sizeof(tsc_seq))))
1153 		goto out_err;
1154 
1155 	if (tsc_seq && tsc_page_update_unsafe(hv)) {
1156 		if (kvm_read_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
1157 			goto out_err;
1158 
1159 		hv->hv_tsc_page_status = HV_TSC_PAGE_SET;
1160 		goto out_unlock;
1161 	}
1162 
1163 	/*
1164 	 * While we're computing and writing the parameters, force the
1165 	 * guest to use the time reference count MSR.
1166 	 */
1167 	hv->tsc_ref.tsc_sequence = 0;
1168 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
1169 			    &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
1170 		goto out_err;
1171 
1172 	if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
1173 		goto out_err;
1174 
1175 	/* Ensure sequence is zero before writing the rest of the struct.  */
1176 	smp_wmb();
1177 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
1178 		goto out_err;
1179 
1180 	/*
1181 	 * Now switch to the TSC page mechanism by writing the sequence.
1182 	 */
1183 	tsc_seq++;
1184 	if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
1185 		tsc_seq = 1;
1186 
1187 	/* Write the struct entirely before the non-zero sequence.  */
1188 	smp_wmb();
1189 
1190 	hv->tsc_ref.tsc_sequence = tsc_seq;
1191 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
1192 			    &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
1193 		goto out_err;
1194 
1195 	hv->hv_tsc_page_status = HV_TSC_PAGE_SET;
1196 	goto out_unlock;
1197 
1198 out_err:
1199 	hv->hv_tsc_page_status = HV_TSC_PAGE_BROKEN;
1200 out_unlock:
1201 	mutex_unlock(&hv->hv_lock);
1202 }
1203 
1204 void kvm_hv_invalidate_tsc_page(struct kvm *kvm)
1205 {
1206 	struct kvm_hv *hv = to_kvm_hv(kvm);
1207 	u64 gfn;
1208 	int idx;
1209 
1210 	if (hv->hv_tsc_page_status == HV_TSC_PAGE_BROKEN ||
1211 	    hv->hv_tsc_page_status == HV_TSC_PAGE_UNSET ||
1212 	    tsc_page_update_unsafe(hv))
1213 		return;
1214 
1215 	mutex_lock(&hv->hv_lock);
1216 
1217 	if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1218 		goto out_unlock;
1219 
1220 	/* Preserve HV_TSC_PAGE_GUEST_CHANGED/HV_TSC_PAGE_HOST_CHANGED states */
1221 	if (hv->hv_tsc_page_status == HV_TSC_PAGE_SET)
1222 		hv->hv_tsc_page_status = HV_TSC_PAGE_UPDATING;
1223 
1224 	gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
1225 
1226 	hv->tsc_ref.tsc_sequence = 0;
1227 
1228 	/*
1229 	 * Take the srcu lock as memslots will be accessed to check the gfn
1230 	 * cache generation against the memslots generation.
1231 	 */
1232 	idx = srcu_read_lock(&kvm->srcu);
1233 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
1234 			    &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
1235 		hv->hv_tsc_page_status = HV_TSC_PAGE_BROKEN;
1236 	srcu_read_unlock(&kvm->srcu, idx);
1237 
1238 out_unlock:
1239 	mutex_unlock(&hv->hv_lock);
1240 }
1241 
1242 
1243 static bool hv_check_msr_access(struct kvm_vcpu_hv *hv_vcpu, u32 msr)
1244 {
1245 	if (!hv_vcpu->enforce_cpuid)
1246 		return true;
1247 
1248 	switch (msr) {
1249 	case HV_X64_MSR_GUEST_OS_ID:
1250 	case HV_X64_MSR_HYPERCALL:
1251 		return hv_vcpu->cpuid_cache.features_eax &
1252 			HV_MSR_HYPERCALL_AVAILABLE;
1253 	case HV_X64_MSR_VP_RUNTIME:
1254 		return hv_vcpu->cpuid_cache.features_eax &
1255 			HV_MSR_VP_RUNTIME_AVAILABLE;
1256 	case HV_X64_MSR_TIME_REF_COUNT:
1257 		return hv_vcpu->cpuid_cache.features_eax &
1258 			HV_MSR_TIME_REF_COUNT_AVAILABLE;
1259 	case HV_X64_MSR_VP_INDEX:
1260 		return hv_vcpu->cpuid_cache.features_eax &
1261 			HV_MSR_VP_INDEX_AVAILABLE;
1262 	case HV_X64_MSR_RESET:
1263 		return hv_vcpu->cpuid_cache.features_eax &
1264 			HV_MSR_RESET_AVAILABLE;
1265 	case HV_X64_MSR_REFERENCE_TSC:
1266 		return hv_vcpu->cpuid_cache.features_eax &
1267 			HV_MSR_REFERENCE_TSC_AVAILABLE;
1268 	case HV_X64_MSR_SCONTROL:
1269 	case HV_X64_MSR_SVERSION:
1270 	case HV_X64_MSR_SIEFP:
1271 	case HV_X64_MSR_SIMP:
1272 	case HV_X64_MSR_EOM:
1273 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1274 		return hv_vcpu->cpuid_cache.features_eax &
1275 			HV_MSR_SYNIC_AVAILABLE;
1276 	case HV_X64_MSR_STIMER0_CONFIG:
1277 	case HV_X64_MSR_STIMER1_CONFIG:
1278 	case HV_X64_MSR_STIMER2_CONFIG:
1279 	case HV_X64_MSR_STIMER3_CONFIG:
1280 	case HV_X64_MSR_STIMER0_COUNT:
1281 	case HV_X64_MSR_STIMER1_COUNT:
1282 	case HV_X64_MSR_STIMER2_COUNT:
1283 	case HV_X64_MSR_STIMER3_COUNT:
1284 		return hv_vcpu->cpuid_cache.features_eax &
1285 			HV_MSR_SYNTIMER_AVAILABLE;
1286 	case HV_X64_MSR_EOI:
1287 	case HV_X64_MSR_ICR:
1288 	case HV_X64_MSR_TPR:
1289 	case HV_X64_MSR_VP_ASSIST_PAGE:
1290 		return hv_vcpu->cpuid_cache.features_eax &
1291 			HV_MSR_APIC_ACCESS_AVAILABLE;
1292 		break;
1293 	case HV_X64_MSR_TSC_FREQUENCY:
1294 	case HV_X64_MSR_APIC_FREQUENCY:
1295 		return hv_vcpu->cpuid_cache.features_eax &
1296 			HV_ACCESS_FREQUENCY_MSRS;
1297 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1298 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1299 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1300 		return hv_vcpu->cpuid_cache.features_eax &
1301 			HV_ACCESS_REENLIGHTENMENT;
1302 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1303 	case HV_X64_MSR_CRASH_CTL:
1304 		return hv_vcpu->cpuid_cache.features_edx &
1305 			HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
1306 	case HV_X64_MSR_SYNDBG_OPTIONS:
1307 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1308 		return hv_vcpu->cpuid_cache.features_edx &
1309 			HV_FEATURE_DEBUG_MSRS_AVAILABLE;
1310 	default:
1311 		break;
1312 	}
1313 
1314 	return false;
1315 }
1316 
1317 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
1318 			     bool host)
1319 {
1320 	struct kvm *kvm = vcpu->kvm;
1321 	struct kvm_hv *hv = to_kvm_hv(kvm);
1322 
1323 	if (unlikely(!host && !hv_check_msr_access(to_hv_vcpu(vcpu), msr)))
1324 		return 1;
1325 
1326 	switch (msr) {
1327 	case HV_X64_MSR_GUEST_OS_ID:
1328 		hv->hv_guest_os_id = data;
1329 		/* setting guest os id to zero disables hypercall page */
1330 		if (!hv->hv_guest_os_id)
1331 			hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1332 		break;
1333 	case HV_X64_MSR_HYPERCALL: {
1334 		u8 instructions[9];
1335 		int i = 0;
1336 		u64 addr;
1337 
1338 		/* if guest os id is not set hypercall should remain disabled */
1339 		if (!hv->hv_guest_os_id)
1340 			break;
1341 		if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1342 			hv->hv_hypercall = data;
1343 			break;
1344 		}
1345 
1346 		/*
1347 		 * If Xen and Hyper-V hypercalls are both enabled, disambiguate
1348 		 * the same way Xen itself does, by setting the bit 31 of EAX
1349 		 * which is RsvdZ in the 32-bit Hyper-V hypercall ABI and just
1350 		 * going to be clobbered on 64-bit.
1351 		 */
1352 		if (kvm_xen_hypercall_enabled(kvm)) {
1353 			/* orl $0x80000000, %eax */
1354 			instructions[i++] = 0x0d;
1355 			instructions[i++] = 0x00;
1356 			instructions[i++] = 0x00;
1357 			instructions[i++] = 0x00;
1358 			instructions[i++] = 0x80;
1359 		}
1360 
1361 		/* vmcall/vmmcall */
1362 		static_call(kvm_x86_patch_hypercall)(vcpu, instructions + i);
1363 		i += 3;
1364 
1365 		/* ret */
1366 		((unsigned char *)instructions)[i++] = 0xc3;
1367 
1368 		addr = data & HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_MASK;
1369 		if (kvm_vcpu_write_guest(vcpu, addr, instructions, i))
1370 			return 1;
1371 		hv->hv_hypercall = data;
1372 		break;
1373 	}
1374 	case HV_X64_MSR_REFERENCE_TSC:
1375 		hv->hv_tsc_page = data;
1376 		if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE) {
1377 			if (!host)
1378 				hv->hv_tsc_page_status = HV_TSC_PAGE_GUEST_CHANGED;
1379 			else
1380 				hv->hv_tsc_page_status = HV_TSC_PAGE_HOST_CHANGED;
1381 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1382 		} else {
1383 			hv->hv_tsc_page_status = HV_TSC_PAGE_UNSET;
1384 		}
1385 		break;
1386 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1387 		return kvm_hv_msr_set_crash_data(kvm,
1388 						 msr - HV_X64_MSR_CRASH_P0,
1389 						 data);
1390 	case HV_X64_MSR_CRASH_CTL:
1391 		if (host)
1392 			return kvm_hv_msr_set_crash_ctl(kvm, data);
1393 
1394 		if (data & HV_CRASH_CTL_CRASH_NOTIFY) {
1395 			vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
1396 				   hv->hv_crash_param[0],
1397 				   hv->hv_crash_param[1],
1398 				   hv->hv_crash_param[2],
1399 				   hv->hv_crash_param[3],
1400 				   hv->hv_crash_param[4]);
1401 
1402 			/* Send notification about crash to user space */
1403 			kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
1404 		}
1405 		break;
1406 	case HV_X64_MSR_RESET:
1407 		if (data == 1) {
1408 			vcpu_debug(vcpu, "hyper-v reset requested\n");
1409 			kvm_make_request(KVM_REQ_HV_RESET, vcpu);
1410 		}
1411 		break;
1412 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1413 		hv->hv_reenlightenment_control = data;
1414 		break;
1415 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1416 		hv->hv_tsc_emulation_control = data;
1417 		break;
1418 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1419 		if (data && !host)
1420 			return 1;
1421 
1422 		hv->hv_tsc_emulation_status = data;
1423 		break;
1424 	case HV_X64_MSR_TIME_REF_COUNT:
1425 		/* read-only, but still ignore it if host-initiated */
1426 		if (!host)
1427 			return 1;
1428 		break;
1429 	case HV_X64_MSR_SYNDBG_OPTIONS:
1430 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1431 		return syndbg_set_msr(vcpu, msr, data, host);
1432 	default:
1433 		vcpu_unimpl(vcpu, "Hyper-V unhandled wrmsr: 0x%x data 0x%llx\n",
1434 			    msr, data);
1435 		return 1;
1436 	}
1437 	return 0;
1438 }
1439 
1440 /* Calculate cpu time spent by current task in 100ns units */
1441 static u64 current_task_runtime_100ns(void)
1442 {
1443 	u64 utime, stime;
1444 
1445 	task_cputime_adjusted(current, &utime, &stime);
1446 
1447 	return div_u64(utime + stime, 100);
1448 }
1449 
1450 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1451 {
1452 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
1453 
1454 	if (unlikely(!host && !hv_check_msr_access(hv_vcpu, msr)))
1455 		return 1;
1456 
1457 	switch (msr) {
1458 	case HV_X64_MSR_VP_INDEX: {
1459 		struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1460 		u32 new_vp_index = (u32)data;
1461 
1462 		if (!host || new_vp_index >= KVM_MAX_VCPUS)
1463 			return 1;
1464 
1465 		if (new_vp_index == hv_vcpu->vp_index)
1466 			return 0;
1467 
1468 		/*
1469 		 * The VP index is initialized to vcpu_index by
1470 		 * kvm_hv_vcpu_postcreate so they initially match.  Now the
1471 		 * VP index is changing, adjust num_mismatched_vp_indexes if
1472 		 * it now matches or no longer matches vcpu_idx.
1473 		 */
1474 		if (hv_vcpu->vp_index == vcpu->vcpu_idx)
1475 			atomic_inc(&hv->num_mismatched_vp_indexes);
1476 		else if (new_vp_index == vcpu->vcpu_idx)
1477 			atomic_dec(&hv->num_mismatched_vp_indexes);
1478 
1479 		hv_vcpu->vp_index = new_vp_index;
1480 		break;
1481 	}
1482 	case HV_X64_MSR_VP_ASSIST_PAGE: {
1483 		u64 gfn;
1484 		unsigned long addr;
1485 
1486 		if (!(data & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE)) {
1487 			hv_vcpu->hv_vapic = data;
1488 			if (kvm_lapic_set_pv_eoi(vcpu, 0, 0))
1489 				return 1;
1490 			break;
1491 		}
1492 		gfn = data >> HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT;
1493 		addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1494 		if (kvm_is_error_hva(addr))
1495 			return 1;
1496 
1497 		/*
1498 		 * Clear apic_assist portion of struct hv_vp_assist_page
1499 		 * only, there can be valuable data in the rest which needs
1500 		 * to be preserved e.g. on migration.
1501 		 */
1502 		if (__put_user(0, (u32 __user *)addr))
1503 			return 1;
1504 		hv_vcpu->hv_vapic = data;
1505 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
1506 		if (kvm_lapic_set_pv_eoi(vcpu,
1507 					    gfn_to_gpa(gfn) | KVM_MSR_ENABLED,
1508 					    sizeof(struct hv_vp_assist_page)))
1509 			return 1;
1510 		break;
1511 	}
1512 	case HV_X64_MSR_EOI:
1513 		return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1514 	case HV_X64_MSR_ICR:
1515 		return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1516 	case HV_X64_MSR_TPR:
1517 		return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1518 	case HV_X64_MSR_VP_RUNTIME:
1519 		if (!host)
1520 			return 1;
1521 		hv_vcpu->runtime_offset = data - current_task_runtime_100ns();
1522 		break;
1523 	case HV_X64_MSR_SCONTROL:
1524 	case HV_X64_MSR_SVERSION:
1525 	case HV_X64_MSR_SIEFP:
1526 	case HV_X64_MSR_SIMP:
1527 	case HV_X64_MSR_EOM:
1528 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1529 		return synic_set_msr(to_hv_synic(vcpu), msr, data, host);
1530 	case HV_X64_MSR_STIMER0_CONFIG:
1531 	case HV_X64_MSR_STIMER1_CONFIG:
1532 	case HV_X64_MSR_STIMER2_CONFIG:
1533 	case HV_X64_MSR_STIMER3_CONFIG: {
1534 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1535 
1536 		return stimer_set_config(to_hv_stimer(vcpu, timer_index),
1537 					 data, host);
1538 	}
1539 	case HV_X64_MSR_STIMER0_COUNT:
1540 	case HV_X64_MSR_STIMER1_COUNT:
1541 	case HV_X64_MSR_STIMER2_COUNT:
1542 	case HV_X64_MSR_STIMER3_COUNT: {
1543 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1544 
1545 		return stimer_set_count(to_hv_stimer(vcpu, timer_index),
1546 					data, host);
1547 	}
1548 	case HV_X64_MSR_TSC_FREQUENCY:
1549 	case HV_X64_MSR_APIC_FREQUENCY:
1550 		/* read-only, but still ignore it if host-initiated */
1551 		if (!host)
1552 			return 1;
1553 		break;
1554 	default:
1555 		vcpu_unimpl(vcpu, "Hyper-V unhandled wrmsr: 0x%x data 0x%llx\n",
1556 			    msr, data);
1557 		return 1;
1558 	}
1559 
1560 	return 0;
1561 }
1562 
1563 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1564 			     bool host)
1565 {
1566 	u64 data = 0;
1567 	struct kvm *kvm = vcpu->kvm;
1568 	struct kvm_hv *hv = to_kvm_hv(kvm);
1569 
1570 	if (unlikely(!host && !hv_check_msr_access(to_hv_vcpu(vcpu), msr)))
1571 		return 1;
1572 
1573 	switch (msr) {
1574 	case HV_X64_MSR_GUEST_OS_ID:
1575 		data = hv->hv_guest_os_id;
1576 		break;
1577 	case HV_X64_MSR_HYPERCALL:
1578 		data = hv->hv_hypercall;
1579 		break;
1580 	case HV_X64_MSR_TIME_REF_COUNT:
1581 		data = get_time_ref_counter(kvm);
1582 		break;
1583 	case HV_X64_MSR_REFERENCE_TSC:
1584 		data = hv->hv_tsc_page;
1585 		break;
1586 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1587 		return kvm_hv_msr_get_crash_data(kvm,
1588 						 msr - HV_X64_MSR_CRASH_P0,
1589 						 pdata);
1590 	case HV_X64_MSR_CRASH_CTL:
1591 		return kvm_hv_msr_get_crash_ctl(kvm, pdata);
1592 	case HV_X64_MSR_RESET:
1593 		data = 0;
1594 		break;
1595 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1596 		data = hv->hv_reenlightenment_control;
1597 		break;
1598 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1599 		data = hv->hv_tsc_emulation_control;
1600 		break;
1601 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1602 		data = hv->hv_tsc_emulation_status;
1603 		break;
1604 	case HV_X64_MSR_SYNDBG_OPTIONS:
1605 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1606 		return syndbg_get_msr(vcpu, msr, pdata, host);
1607 	default:
1608 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1609 		return 1;
1610 	}
1611 
1612 	*pdata = data;
1613 	return 0;
1614 }
1615 
1616 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1617 			  bool host)
1618 {
1619 	u64 data = 0;
1620 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
1621 
1622 	if (unlikely(!host && !hv_check_msr_access(hv_vcpu, msr)))
1623 		return 1;
1624 
1625 	switch (msr) {
1626 	case HV_X64_MSR_VP_INDEX:
1627 		data = hv_vcpu->vp_index;
1628 		break;
1629 	case HV_X64_MSR_EOI:
1630 		return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1631 	case HV_X64_MSR_ICR:
1632 		return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1633 	case HV_X64_MSR_TPR:
1634 		return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1635 	case HV_X64_MSR_VP_ASSIST_PAGE:
1636 		data = hv_vcpu->hv_vapic;
1637 		break;
1638 	case HV_X64_MSR_VP_RUNTIME:
1639 		data = current_task_runtime_100ns() + hv_vcpu->runtime_offset;
1640 		break;
1641 	case HV_X64_MSR_SCONTROL:
1642 	case HV_X64_MSR_SVERSION:
1643 	case HV_X64_MSR_SIEFP:
1644 	case HV_X64_MSR_SIMP:
1645 	case HV_X64_MSR_EOM:
1646 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1647 		return synic_get_msr(to_hv_synic(vcpu), msr, pdata, host);
1648 	case HV_X64_MSR_STIMER0_CONFIG:
1649 	case HV_X64_MSR_STIMER1_CONFIG:
1650 	case HV_X64_MSR_STIMER2_CONFIG:
1651 	case HV_X64_MSR_STIMER3_CONFIG: {
1652 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1653 
1654 		return stimer_get_config(to_hv_stimer(vcpu, timer_index),
1655 					 pdata);
1656 	}
1657 	case HV_X64_MSR_STIMER0_COUNT:
1658 	case HV_X64_MSR_STIMER1_COUNT:
1659 	case HV_X64_MSR_STIMER2_COUNT:
1660 	case HV_X64_MSR_STIMER3_COUNT: {
1661 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1662 
1663 		return stimer_get_count(to_hv_stimer(vcpu, timer_index),
1664 					pdata);
1665 	}
1666 	case HV_X64_MSR_TSC_FREQUENCY:
1667 		data = (u64)vcpu->arch.virtual_tsc_khz * 1000;
1668 		break;
1669 	case HV_X64_MSR_APIC_FREQUENCY:
1670 		data = APIC_BUS_FREQUENCY;
1671 		break;
1672 	default:
1673 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1674 		return 1;
1675 	}
1676 	*pdata = data;
1677 	return 0;
1678 }
1679 
1680 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1681 {
1682 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1683 
1684 	if (!host && !vcpu->arch.hyperv_enabled)
1685 		return 1;
1686 
1687 	if (!to_hv_vcpu(vcpu)) {
1688 		if (kvm_hv_vcpu_init(vcpu))
1689 			return 1;
1690 	}
1691 
1692 	if (kvm_hv_msr_partition_wide(msr)) {
1693 		int r;
1694 
1695 		mutex_lock(&hv->hv_lock);
1696 		r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1697 		mutex_unlock(&hv->hv_lock);
1698 		return r;
1699 	} else
1700 		return kvm_hv_set_msr(vcpu, msr, data, host);
1701 }
1702 
1703 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
1704 {
1705 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1706 
1707 	if (!host && !vcpu->arch.hyperv_enabled)
1708 		return 1;
1709 
1710 	if (!to_hv_vcpu(vcpu)) {
1711 		if (kvm_hv_vcpu_init(vcpu))
1712 			return 1;
1713 	}
1714 
1715 	if (kvm_hv_msr_partition_wide(msr)) {
1716 		int r;
1717 
1718 		mutex_lock(&hv->hv_lock);
1719 		r = kvm_hv_get_msr_pw(vcpu, msr, pdata, host);
1720 		mutex_unlock(&hv->hv_lock);
1721 		return r;
1722 	} else
1723 		return kvm_hv_get_msr(vcpu, msr, pdata, host);
1724 }
1725 
1726 static void sparse_set_to_vcpu_mask(struct kvm *kvm, u64 *sparse_banks,
1727 				    u64 valid_bank_mask, unsigned long *vcpu_mask)
1728 {
1729 	struct kvm_hv *hv = to_kvm_hv(kvm);
1730 	bool has_mismatch = atomic_read(&hv->num_mismatched_vp_indexes);
1731 	u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1732 	struct kvm_vcpu *vcpu;
1733 	int bank, sbank = 0;
1734 	unsigned long i;
1735 	u64 *bitmap;
1736 
1737 	BUILD_BUG_ON(sizeof(vp_bitmap) >
1738 		     sizeof(*vcpu_mask) * BITS_TO_LONGS(KVM_MAX_VCPUS));
1739 
1740 	/*
1741 	 * If vp_index == vcpu_idx for all vCPUs, fill vcpu_mask directly, else
1742 	 * fill a temporary buffer and manually test each vCPU's VP index.
1743 	 */
1744 	if (likely(!has_mismatch))
1745 		bitmap = (u64 *)vcpu_mask;
1746 	else
1747 		bitmap = vp_bitmap;
1748 
1749 	/*
1750 	 * Each set of 64 VPs is packed into sparse_banks, with valid_bank_mask
1751 	 * having a '1' for each bank that exists in sparse_banks.  Sets must
1752 	 * be in ascending order, i.e. bank0..bankN.
1753 	 */
1754 	memset(bitmap, 0, sizeof(vp_bitmap));
1755 	for_each_set_bit(bank, (unsigned long *)&valid_bank_mask,
1756 			 KVM_HV_MAX_SPARSE_VCPU_SET_BITS)
1757 		bitmap[bank] = sparse_banks[sbank++];
1758 
1759 	if (likely(!has_mismatch))
1760 		return;
1761 
1762 	bitmap_zero(vcpu_mask, KVM_MAX_VCPUS);
1763 	kvm_for_each_vcpu(i, vcpu, kvm) {
1764 		if (test_bit(kvm_hv_get_vpindex(vcpu), (unsigned long *)vp_bitmap))
1765 			__set_bit(i, vcpu_mask);
1766 	}
1767 }
1768 
1769 struct kvm_hv_hcall {
1770 	u64 param;
1771 	u64 ingpa;
1772 	u64 outgpa;
1773 	u16 code;
1774 	u16 var_cnt;
1775 	u16 rep_cnt;
1776 	u16 rep_idx;
1777 	bool fast;
1778 	bool rep;
1779 	sse128_t xmm[HV_HYPERCALL_MAX_XMM_REGISTERS];
1780 };
1781 
1782 static u64 kvm_get_sparse_vp_set(struct kvm *kvm, struct kvm_hv_hcall *hc,
1783 				 int consumed_xmm_halves,
1784 				 u64 *sparse_banks, gpa_t offset)
1785 {
1786 	u16 var_cnt;
1787 	int i;
1788 
1789 	if (hc->var_cnt > 64)
1790 		return -EINVAL;
1791 
1792 	/* Ignore banks that cannot possibly contain a legal VP index. */
1793 	var_cnt = min_t(u16, hc->var_cnt, KVM_HV_MAX_SPARSE_VCPU_SET_BITS);
1794 
1795 	if (hc->fast) {
1796 		/*
1797 		 * Each XMM holds two sparse banks, but do not count halves that
1798 		 * have already been consumed for hypercall parameters.
1799 		 */
1800 		if (hc->var_cnt > 2 * HV_HYPERCALL_MAX_XMM_REGISTERS - consumed_xmm_halves)
1801 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1802 		for (i = 0; i < var_cnt; i++) {
1803 			int j = i + consumed_xmm_halves;
1804 			if (j % 2)
1805 				sparse_banks[i] = sse128_hi(hc->xmm[j / 2]);
1806 			else
1807 				sparse_banks[i] = sse128_lo(hc->xmm[j / 2]);
1808 		}
1809 		return 0;
1810 	}
1811 
1812 	return kvm_read_guest(kvm, hc->ingpa + offset, sparse_banks,
1813 			      var_cnt * sizeof(*sparse_banks));
1814 }
1815 
1816 static u64 kvm_hv_flush_tlb(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc)
1817 {
1818 	struct kvm *kvm = vcpu->kvm;
1819 	struct hv_tlb_flush_ex flush_ex;
1820 	struct hv_tlb_flush flush;
1821 	DECLARE_BITMAP(vcpu_mask, KVM_MAX_VCPUS);
1822 	u64 valid_bank_mask;
1823 	u64 sparse_banks[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1824 	bool all_cpus;
1825 
1826 	/*
1827 	 * The Hyper-V TLFS doesn't allow more than 64 sparse banks, e.g. the
1828 	 * valid mask is a u64.  Fail the build if KVM's max allowed number of
1829 	 * vCPUs (>4096) would exceed this limit, KVM will additional changes
1830 	 * for Hyper-V support to avoid setting the guest up to fail.
1831 	 */
1832 	BUILD_BUG_ON(KVM_HV_MAX_SPARSE_VCPU_SET_BITS > 64);
1833 
1834 	if (hc->code == HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST ||
1835 	    hc->code == HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE) {
1836 		if (hc->fast) {
1837 			flush.address_space = hc->ingpa;
1838 			flush.flags = hc->outgpa;
1839 			flush.processor_mask = sse128_lo(hc->xmm[0]);
1840 		} else {
1841 			if (unlikely(kvm_read_guest(kvm, hc->ingpa,
1842 						    &flush, sizeof(flush))))
1843 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
1844 		}
1845 
1846 		trace_kvm_hv_flush_tlb(flush.processor_mask,
1847 				       flush.address_space, flush.flags);
1848 
1849 		valid_bank_mask = BIT_ULL(0);
1850 		sparse_banks[0] = flush.processor_mask;
1851 
1852 		/*
1853 		 * Work around possible WS2012 bug: it sends hypercalls
1854 		 * with processor_mask = 0x0 and HV_FLUSH_ALL_PROCESSORS clear,
1855 		 * while also expecting us to flush something and crashing if
1856 		 * we don't. Let's treat processor_mask == 0 same as
1857 		 * HV_FLUSH_ALL_PROCESSORS.
1858 		 */
1859 		all_cpus = (flush.flags & HV_FLUSH_ALL_PROCESSORS) ||
1860 			flush.processor_mask == 0;
1861 	} else {
1862 		if (hc->fast) {
1863 			flush_ex.address_space = hc->ingpa;
1864 			flush_ex.flags = hc->outgpa;
1865 			memcpy(&flush_ex.hv_vp_set,
1866 			       &hc->xmm[0], sizeof(hc->xmm[0]));
1867 		} else {
1868 			if (unlikely(kvm_read_guest(kvm, hc->ingpa, &flush_ex,
1869 						    sizeof(flush_ex))))
1870 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
1871 		}
1872 
1873 		trace_kvm_hv_flush_tlb_ex(flush_ex.hv_vp_set.valid_bank_mask,
1874 					  flush_ex.hv_vp_set.format,
1875 					  flush_ex.address_space,
1876 					  flush_ex.flags);
1877 
1878 		valid_bank_mask = flush_ex.hv_vp_set.valid_bank_mask;
1879 		all_cpus = flush_ex.hv_vp_set.format !=
1880 			HV_GENERIC_SET_SPARSE_4K;
1881 
1882 		if (hc->var_cnt != bitmap_weight((unsigned long *)&valid_bank_mask, 64))
1883 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1884 
1885 		if (all_cpus)
1886 			goto do_flush;
1887 
1888 		if (!hc->var_cnt)
1889 			goto ret_success;
1890 
1891 		if (kvm_get_sparse_vp_set(kvm, hc, 2, sparse_banks,
1892 					  offsetof(struct hv_tlb_flush_ex,
1893 						   hv_vp_set.bank_contents)))
1894 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1895 	}
1896 
1897 do_flush:
1898 	/*
1899 	 * vcpu->arch.cr3 may not be up-to-date for running vCPUs so we can't
1900 	 * analyze it here, flush TLB regardless of the specified address space.
1901 	 */
1902 	if (all_cpus) {
1903 		kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH_GUEST);
1904 	} else {
1905 		sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask, vcpu_mask);
1906 
1907 		kvm_make_vcpus_request_mask(kvm, KVM_REQ_TLB_FLUSH_GUEST, vcpu_mask);
1908 	}
1909 
1910 ret_success:
1911 	/* We always do full TLB flush, set 'Reps completed' = 'Rep Count' */
1912 	return (u64)HV_STATUS_SUCCESS |
1913 		((u64)hc->rep_cnt << HV_HYPERCALL_REP_COMP_OFFSET);
1914 }
1915 
1916 static void kvm_send_ipi_to_many(struct kvm *kvm, u32 vector,
1917 				 unsigned long *vcpu_bitmap)
1918 {
1919 	struct kvm_lapic_irq irq = {
1920 		.delivery_mode = APIC_DM_FIXED,
1921 		.vector = vector
1922 	};
1923 	struct kvm_vcpu *vcpu;
1924 	unsigned long i;
1925 
1926 	kvm_for_each_vcpu(i, vcpu, kvm) {
1927 		if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
1928 			continue;
1929 
1930 		/* We fail only when APIC is disabled */
1931 		kvm_apic_set_irq(vcpu, &irq, NULL);
1932 	}
1933 }
1934 
1935 static u64 kvm_hv_send_ipi(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc)
1936 {
1937 	struct kvm *kvm = vcpu->kvm;
1938 	struct hv_send_ipi_ex send_ipi_ex;
1939 	struct hv_send_ipi send_ipi;
1940 	DECLARE_BITMAP(vcpu_mask, KVM_MAX_VCPUS);
1941 	unsigned long valid_bank_mask;
1942 	u64 sparse_banks[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1943 	u32 vector;
1944 	bool all_cpus;
1945 
1946 	if (hc->code == HVCALL_SEND_IPI) {
1947 		if (!hc->fast) {
1948 			if (unlikely(kvm_read_guest(kvm, hc->ingpa, &send_ipi,
1949 						    sizeof(send_ipi))))
1950 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
1951 			sparse_banks[0] = send_ipi.cpu_mask;
1952 			vector = send_ipi.vector;
1953 		} else {
1954 			/* 'reserved' part of hv_send_ipi should be 0 */
1955 			if (unlikely(hc->ingpa >> 32 != 0))
1956 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
1957 			sparse_banks[0] = hc->outgpa;
1958 			vector = (u32)hc->ingpa;
1959 		}
1960 		all_cpus = false;
1961 		valid_bank_mask = BIT_ULL(0);
1962 
1963 		trace_kvm_hv_send_ipi(vector, sparse_banks[0]);
1964 	} else {
1965 		if (!hc->fast) {
1966 			if (unlikely(kvm_read_guest(kvm, hc->ingpa, &send_ipi_ex,
1967 						    sizeof(send_ipi_ex))))
1968 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
1969 		} else {
1970 			send_ipi_ex.vector = (u32)hc->ingpa;
1971 			send_ipi_ex.vp_set.format = hc->outgpa;
1972 			send_ipi_ex.vp_set.valid_bank_mask = sse128_lo(hc->xmm[0]);
1973 		}
1974 
1975 		trace_kvm_hv_send_ipi_ex(send_ipi_ex.vector,
1976 					 send_ipi_ex.vp_set.format,
1977 					 send_ipi_ex.vp_set.valid_bank_mask);
1978 
1979 		vector = send_ipi_ex.vector;
1980 		valid_bank_mask = send_ipi_ex.vp_set.valid_bank_mask;
1981 		all_cpus = send_ipi_ex.vp_set.format == HV_GENERIC_SET_ALL;
1982 
1983 		if (hc->var_cnt != bitmap_weight(&valid_bank_mask, 64))
1984 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1985 
1986 		if (all_cpus)
1987 			goto check_and_send_ipi;
1988 
1989 		if (!hc->var_cnt)
1990 			goto ret_success;
1991 
1992 		if (kvm_get_sparse_vp_set(kvm, hc, 1, sparse_banks,
1993 					  offsetof(struct hv_send_ipi_ex,
1994 						   vp_set.bank_contents)))
1995 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1996 	}
1997 
1998 check_and_send_ipi:
1999 	if ((vector < HV_IPI_LOW_VECTOR) || (vector > HV_IPI_HIGH_VECTOR))
2000 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
2001 
2002 	if (all_cpus) {
2003 		kvm_send_ipi_to_many(kvm, vector, NULL);
2004 	} else {
2005 		sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask, vcpu_mask);
2006 
2007 		kvm_send_ipi_to_many(kvm, vector, vcpu_mask);
2008 	}
2009 
2010 ret_success:
2011 	return HV_STATUS_SUCCESS;
2012 }
2013 
2014 void kvm_hv_set_cpuid(struct kvm_vcpu *vcpu)
2015 {
2016 	struct kvm_cpuid_entry2 *entry;
2017 	struct kvm_vcpu_hv *hv_vcpu;
2018 
2019 	entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_INTERFACE, 0);
2020 	if (entry && entry->eax == HYPERV_CPUID_SIGNATURE_EAX) {
2021 		vcpu->arch.hyperv_enabled = true;
2022 	} else {
2023 		vcpu->arch.hyperv_enabled = false;
2024 		return;
2025 	}
2026 
2027 	if (!to_hv_vcpu(vcpu) && kvm_hv_vcpu_init(vcpu))
2028 		return;
2029 
2030 	hv_vcpu = to_hv_vcpu(vcpu);
2031 
2032 	entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_FEATURES, 0);
2033 	if (entry) {
2034 		hv_vcpu->cpuid_cache.features_eax = entry->eax;
2035 		hv_vcpu->cpuid_cache.features_ebx = entry->ebx;
2036 		hv_vcpu->cpuid_cache.features_edx = entry->edx;
2037 	} else {
2038 		hv_vcpu->cpuid_cache.features_eax = 0;
2039 		hv_vcpu->cpuid_cache.features_ebx = 0;
2040 		hv_vcpu->cpuid_cache.features_edx = 0;
2041 	}
2042 
2043 	entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_ENLIGHTMENT_INFO, 0);
2044 	if (entry) {
2045 		hv_vcpu->cpuid_cache.enlightenments_eax = entry->eax;
2046 		hv_vcpu->cpuid_cache.enlightenments_ebx = entry->ebx;
2047 	} else {
2048 		hv_vcpu->cpuid_cache.enlightenments_eax = 0;
2049 		hv_vcpu->cpuid_cache.enlightenments_ebx = 0;
2050 	}
2051 
2052 	entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES, 0);
2053 	if (entry)
2054 		hv_vcpu->cpuid_cache.syndbg_cap_eax = entry->eax;
2055 	else
2056 		hv_vcpu->cpuid_cache.syndbg_cap_eax = 0;
2057 }
2058 
2059 int kvm_hv_set_enforce_cpuid(struct kvm_vcpu *vcpu, bool enforce)
2060 {
2061 	struct kvm_vcpu_hv *hv_vcpu;
2062 	int ret = 0;
2063 
2064 	if (!to_hv_vcpu(vcpu)) {
2065 		if (enforce) {
2066 			ret = kvm_hv_vcpu_init(vcpu);
2067 			if (ret)
2068 				return ret;
2069 		} else {
2070 			return 0;
2071 		}
2072 	}
2073 
2074 	hv_vcpu = to_hv_vcpu(vcpu);
2075 	hv_vcpu->enforce_cpuid = enforce;
2076 
2077 	return ret;
2078 }
2079 
2080 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
2081 {
2082 	bool longmode;
2083 
2084 	longmode = is_64_bit_hypercall(vcpu);
2085 	if (longmode)
2086 		kvm_rax_write(vcpu, result);
2087 	else {
2088 		kvm_rdx_write(vcpu, result >> 32);
2089 		kvm_rax_write(vcpu, result & 0xffffffff);
2090 	}
2091 }
2092 
2093 static int kvm_hv_hypercall_complete(struct kvm_vcpu *vcpu, u64 result)
2094 {
2095 	trace_kvm_hv_hypercall_done(result);
2096 	kvm_hv_hypercall_set_result(vcpu, result);
2097 	++vcpu->stat.hypercalls;
2098 	return kvm_skip_emulated_instruction(vcpu);
2099 }
2100 
2101 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
2102 {
2103 	return kvm_hv_hypercall_complete(vcpu, vcpu->run->hyperv.u.hcall.result);
2104 }
2105 
2106 static u16 kvm_hvcall_signal_event(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc)
2107 {
2108 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
2109 	struct eventfd_ctx *eventfd;
2110 
2111 	if (unlikely(!hc->fast)) {
2112 		int ret;
2113 		gpa_t gpa = hc->ingpa;
2114 
2115 		if ((gpa & (__alignof__(hc->ingpa) - 1)) ||
2116 		    offset_in_page(gpa) + sizeof(hc->ingpa) > PAGE_SIZE)
2117 			return HV_STATUS_INVALID_ALIGNMENT;
2118 
2119 		ret = kvm_vcpu_read_guest(vcpu, gpa,
2120 					  &hc->ingpa, sizeof(hc->ingpa));
2121 		if (ret < 0)
2122 			return HV_STATUS_INVALID_ALIGNMENT;
2123 	}
2124 
2125 	/*
2126 	 * Per spec, bits 32-47 contain the extra "flag number".  However, we
2127 	 * have no use for it, and in all known usecases it is zero, so just
2128 	 * report lookup failure if it isn't.
2129 	 */
2130 	if (hc->ingpa & 0xffff00000000ULL)
2131 		return HV_STATUS_INVALID_PORT_ID;
2132 	/* remaining bits are reserved-zero */
2133 	if (hc->ingpa & ~KVM_HYPERV_CONN_ID_MASK)
2134 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
2135 
2136 	/* the eventfd is protected by vcpu->kvm->srcu, but conn_to_evt isn't */
2137 	rcu_read_lock();
2138 	eventfd = idr_find(&hv->conn_to_evt, hc->ingpa);
2139 	rcu_read_unlock();
2140 	if (!eventfd)
2141 		return HV_STATUS_INVALID_PORT_ID;
2142 
2143 	eventfd_signal(eventfd, 1);
2144 	return HV_STATUS_SUCCESS;
2145 }
2146 
2147 static bool is_xmm_fast_hypercall(struct kvm_hv_hcall *hc)
2148 {
2149 	switch (hc->code) {
2150 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
2151 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
2152 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
2153 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
2154 	case HVCALL_SEND_IPI_EX:
2155 		return true;
2156 	}
2157 
2158 	return false;
2159 }
2160 
2161 static void kvm_hv_hypercall_read_xmm(struct kvm_hv_hcall *hc)
2162 {
2163 	int reg;
2164 
2165 	kvm_fpu_get();
2166 	for (reg = 0; reg < HV_HYPERCALL_MAX_XMM_REGISTERS; reg++)
2167 		_kvm_read_sse_reg(reg, &hc->xmm[reg]);
2168 	kvm_fpu_put();
2169 }
2170 
2171 static bool hv_check_hypercall_access(struct kvm_vcpu_hv *hv_vcpu, u16 code)
2172 {
2173 	if (!hv_vcpu->enforce_cpuid)
2174 		return true;
2175 
2176 	switch (code) {
2177 	case HVCALL_NOTIFY_LONG_SPIN_WAIT:
2178 		return hv_vcpu->cpuid_cache.enlightenments_ebx &&
2179 			hv_vcpu->cpuid_cache.enlightenments_ebx != U32_MAX;
2180 	case HVCALL_POST_MESSAGE:
2181 		return hv_vcpu->cpuid_cache.features_ebx & HV_POST_MESSAGES;
2182 	case HVCALL_SIGNAL_EVENT:
2183 		return hv_vcpu->cpuid_cache.features_ebx & HV_SIGNAL_EVENTS;
2184 	case HVCALL_POST_DEBUG_DATA:
2185 	case HVCALL_RETRIEVE_DEBUG_DATA:
2186 	case HVCALL_RESET_DEBUG_SESSION:
2187 		/*
2188 		 * Return 'true' when SynDBG is disabled so the resulting code
2189 		 * will be HV_STATUS_INVALID_HYPERCALL_CODE.
2190 		 */
2191 		return !kvm_hv_is_syndbg_enabled(hv_vcpu->vcpu) ||
2192 			hv_vcpu->cpuid_cache.features_ebx & HV_DEBUGGING;
2193 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
2194 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
2195 		if (!(hv_vcpu->cpuid_cache.enlightenments_eax &
2196 		      HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED))
2197 			return false;
2198 		fallthrough;
2199 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
2200 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
2201 		return hv_vcpu->cpuid_cache.enlightenments_eax &
2202 			HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
2203 	case HVCALL_SEND_IPI_EX:
2204 		if (!(hv_vcpu->cpuid_cache.enlightenments_eax &
2205 		      HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED))
2206 			return false;
2207 		fallthrough;
2208 	case HVCALL_SEND_IPI:
2209 		return hv_vcpu->cpuid_cache.enlightenments_eax &
2210 			HV_X64_CLUSTER_IPI_RECOMMENDED;
2211 	default:
2212 		break;
2213 	}
2214 
2215 	return true;
2216 }
2217 
2218 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
2219 {
2220 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
2221 	struct kvm_hv_hcall hc;
2222 	u64 ret = HV_STATUS_SUCCESS;
2223 
2224 	/*
2225 	 * hypercall generates UD from non zero cpl and real mode
2226 	 * per HYPER-V spec
2227 	 */
2228 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0 || !is_protmode(vcpu)) {
2229 		kvm_queue_exception(vcpu, UD_VECTOR);
2230 		return 1;
2231 	}
2232 
2233 #ifdef CONFIG_X86_64
2234 	if (is_64_bit_hypercall(vcpu)) {
2235 		hc.param = kvm_rcx_read(vcpu);
2236 		hc.ingpa = kvm_rdx_read(vcpu);
2237 		hc.outgpa = kvm_r8_read(vcpu);
2238 	} else
2239 #endif
2240 	{
2241 		hc.param = ((u64)kvm_rdx_read(vcpu) << 32) |
2242 			    (kvm_rax_read(vcpu) & 0xffffffff);
2243 		hc.ingpa = ((u64)kvm_rbx_read(vcpu) << 32) |
2244 			    (kvm_rcx_read(vcpu) & 0xffffffff);
2245 		hc.outgpa = ((u64)kvm_rdi_read(vcpu) << 32) |
2246 			     (kvm_rsi_read(vcpu) & 0xffffffff);
2247 	}
2248 
2249 	hc.code = hc.param & 0xffff;
2250 	hc.var_cnt = (hc.param & HV_HYPERCALL_VARHEAD_MASK) >> HV_HYPERCALL_VARHEAD_OFFSET;
2251 	hc.fast = !!(hc.param & HV_HYPERCALL_FAST_BIT);
2252 	hc.rep_cnt = (hc.param >> HV_HYPERCALL_REP_COMP_OFFSET) & 0xfff;
2253 	hc.rep_idx = (hc.param >> HV_HYPERCALL_REP_START_OFFSET) & 0xfff;
2254 	hc.rep = !!(hc.rep_cnt || hc.rep_idx);
2255 
2256 	trace_kvm_hv_hypercall(hc.code, hc.fast, hc.var_cnt, hc.rep_cnt,
2257 			       hc.rep_idx, hc.ingpa, hc.outgpa);
2258 
2259 	if (unlikely(!hv_check_hypercall_access(hv_vcpu, hc.code))) {
2260 		ret = HV_STATUS_ACCESS_DENIED;
2261 		goto hypercall_complete;
2262 	}
2263 
2264 	if (unlikely(hc.param & HV_HYPERCALL_RSVD_MASK)) {
2265 		ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2266 		goto hypercall_complete;
2267 	}
2268 
2269 	if (hc.fast && is_xmm_fast_hypercall(&hc)) {
2270 		if (unlikely(hv_vcpu->enforce_cpuid &&
2271 			     !(hv_vcpu->cpuid_cache.features_edx &
2272 			       HV_X64_HYPERCALL_XMM_INPUT_AVAILABLE))) {
2273 			kvm_queue_exception(vcpu, UD_VECTOR);
2274 			return 1;
2275 		}
2276 
2277 		kvm_hv_hypercall_read_xmm(&hc);
2278 	}
2279 
2280 	switch (hc.code) {
2281 	case HVCALL_NOTIFY_LONG_SPIN_WAIT:
2282 		if (unlikely(hc.rep || hc.var_cnt)) {
2283 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2284 			break;
2285 		}
2286 		kvm_vcpu_on_spin(vcpu, true);
2287 		break;
2288 	case HVCALL_SIGNAL_EVENT:
2289 		if (unlikely(hc.rep || hc.var_cnt)) {
2290 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2291 			break;
2292 		}
2293 		ret = kvm_hvcall_signal_event(vcpu, &hc);
2294 		if (ret != HV_STATUS_INVALID_PORT_ID)
2295 			break;
2296 		fallthrough;	/* maybe userspace knows this conn_id */
2297 	case HVCALL_POST_MESSAGE:
2298 		/* don't bother userspace if it has no way to handle it */
2299 		if (unlikely(hc.rep || hc.var_cnt || !to_hv_synic(vcpu)->active)) {
2300 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2301 			break;
2302 		}
2303 		vcpu->run->exit_reason = KVM_EXIT_HYPERV;
2304 		vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
2305 		vcpu->run->hyperv.u.hcall.input = hc.param;
2306 		vcpu->run->hyperv.u.hcall.params[0] = hc.ingpa;
2307 		vcpu->run->hyperv.u.hcall.params[1] = hc.outgpa;
2308 		vcpu->arch.complete_userspace_io =
2309 				kvm_hv_hypercall_complete_userspace;
2310 		return 0;
2311 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
2312 		if (unlikely(hc.var_cnt)) {
2313 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2314 			break;
2315 		}
2316 		fallthrough;
2317 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
2318 		if (unlikely(!hc.rep_cnt || hc.rep_idx)) {
2319 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2320 			break;
2321 		}
2322 		ret = kvm_hv_flush_tlb(vcpu, &hc);
2323 		break;
2324 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
2325 		if (unlikely(hc.var_cnt)) {
2326 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2327 			break;
2328 		}
2329 		fallthrough;
2330 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
2331 		if (unlikely(hc.rep)) {
2332 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2333 			break;
2334 		}
2335 		ret = kvm_hv_flush_tlb(vcpu, &hc);
2336 		break;
2337 	case HVCALL_SEND_IPI:
2338 		if (unlikely(hc.var_cnt)) {
2339 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2340 			break;
2341 		}
2342 		fallthrough;
2343 	case HVCALL_SEND_IPI_EX:
2344 		if (unlikely(hc.rep)) {
2345 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2346 			break;
2347 		}
2348 		ret = kvm_hv_send_ipi(vcpu, &hc);
2349 		break;
2350 	case HVCALL_POST_DEBUG_DATA:
2351 	case HVCALL_RETRIEVE_DEBUG_DATA:
2352 		if (unlikely(hc.fast)) {
2353 			ret = HV_STATUS_INVALID_PARAMETER;
2354 			break;
2355 		}
2356 		fallthrough;
2357 	case HVCALL_RESET_DEBUG_SESSION: {
2358 		struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
2359 
2360 		if (!kvm_hv_is_syndbg_enabled(vcpu)) {
2361 			ret = HV_STATUS_INVALID_HYPERCALL_CODE;
2362 			break;
2363 		}
2364 
2365 		if (!(syndbg->options & HV_X64_SYNDBG_OPTION_USE_HCALLS)) {
2366 			ret = HV_STATUS_OPERATION_DENIED;
2367 			break;
2368 		}
2369 		vcpu->run->exit_reason = KVM_EXIT_HYPERV;
2370 		vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
2371 		vcpu->run->hyperv.u.hcall.input = hc.param;
2372 		vcpu->run->hyperv.u.hcall.params[0] = hc.ingpa;
2373 		vcpu->run->hyperv.u.hcall.params[1] = hc.outgpa;
2374 		vcpu->arch.complete_userspace_io =
2375 				kvm_hv_hypercall_complete_userspace;
2376 		return 0;
2377 	}
2378 	default:
2379 		ret = HV_STATUS_INVALID_HYPERCALL_CODE;
2380 		break;
2381 	}
2382 
2383 hypercall_complete:
2384 	return kvm_hv_hypercall_complete(vcpu, ret);
2385 }
2386 
2387 void kvm_hv_init_vm(struct kvm *kvm)
2388 {
2389 	struct kvm_hv *hv = to_kvm_hv(kvm);
2390 
2391 	mutex_init(&hv->hv_lock);
2392 	idr_init(&hv->conn_to_evt);
2393 }
2394 
2395 void kvm_hv_destroy_vm(struct kvm *kvm)
2396 {
2397 	struct kvm_hv *hv = to_kvm_hv(kvm);
2398 	struct eventfd_ctx *eventfd;
2399 	int i;
2400 
2401 	idr_for_each_entry(&hv->conn_to_evt, eventfd, i)
2402 		eventfd_ctx_put(eventfd);
2403 	idr_destroy(&hv->conn_to_evt);
2404 }
2405 
2406 static int kvm_hv_eventfd_assign(struct kvm *kvm, u32 conn_id, int fd)
2407 {
2408 	struct kvm_hv *hv = to_kvm_hv(kvm);
2409 	struct eventfd_ctx *eventfd;
2410 	int ret;
2411 
2412 	eventfd = eventfd_ctx_fdget(fd);
2413 	if (IS_ERR(eventfd))
2414 		return PTR_ERR(eventfd);
2415 
2416 	mutex_lock(&hv->hv_lock);
2417 	ret = idr_alloc(&hv->conn_to_evt, eventfd, conn_id, conn_id + 1,
2418 			GFP_KERNEL_ACCOUNT);
2419 	mutex_unlock(&hv->hv_lock);
2420 
2421 	if (ret >= 0)
2422 		return 0;
2423 
2424 	if (ret == -ENOSPC)
2425 		ret = -EEXIST;
2426 	eventfd_ctx_put(eventfd);
2427 	return ret;
2428 }
2429 
2430 static int kvm_hv_eventfd_deassign(struct kvm *kvm, u32 conn_id)
2431 {
2432 	struct kvm_hv *hv = to_kvm_hv(kvm);
2433 	struct eventfd_ctx *eventfd;
2434 
2435 	mutex_lock(&hv->hv_lock);
2436 	eventfd = idr_remove(&hv->conn_to_evt, conn_id);
2437 	mutex_unlock(&hv->hv_lock);
2438 
2439 	if (!eventfd)
2440 		return -ENOENT;
2441 
2442 	synchronize_srcu(&kvm->srcu);
2443 	eventfd_ctx_put(eventfd);
2444 	return 0;
2445 }
2446 
2447 int kvm_vm_ioctl_hv_eventfd(struct kvm *kvm, struct kvm_hyperv_eventfd *args)
2448 {
2449 	if ((args->flags & ~KVM_HYPERV_EVENTFD_DEASSIGN) ||
2450 	    (args->conn_id & ~KVM_HYPERV_CONN_ID_MASK))
2451 		return -EINVAL;
2452 
2453 	if (args->flags == KVM_HYPERV_EVENTFD_DEASSIGN)
2454 		return kvm_hv_eventfd_deassign(kvm, args->conn_id);
2455 	return kvm_hv_eventfd_assign(kvm, args->conn_id, args->fd);
2456 }
2457 
2458 int kvm_get_hv_cpuid(struct kvm_vcpu *vcpu, struct kvm_cpuid2 *cpuid,
2459 		     struct kvm_cpuid_entry2 __user *entries)
2460 {
2461 	uint16_t evmcs_ver = 0;
2462 	struct kvm_cpuid_entry2 cpuid_entries[] = {
2463 		{ .function = HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS },
2464 		{ .function = HYPERV_CPUID_INTERFACE },
2465 		{ .function = HYPERV_CPUID_VERSION },
2466 		{ .function = HYPERV_CPUID_FEATURES },
2467 		{ .function = HYPERV_CPUID_ENLIGHTMENT_INFO },
2468 		{ .function = HYPERV_CPUID_IMPLEMENT_LIMITS },
2469 		{ .function = HYPERV_CPUID_SYNDBG_VENDOR_AND_MAX_FUNCTIONS },
2470 		{ .function = HYPERV_CPUID_SYNDBG_INTERFACE },
2471 		{ .function = HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES	},
2472 		{ .function = HYPERV_CPUID_NESTED_FEATURES },
2473 	};
2474 	int i, nent = ARRAY_SIZE(cpuid_entries);
2475 
2476 	if (kvm_x86_ops.nested_ops->get_evmcs_version)
2477 		evmcs_ver = kvm_x86_ops.nested_ops->get_evmcs_version(vcpu);
2478 
2479 	if (cpuid->nent < nent)
2480 		return -E2BIG;
2481 
2482 	if (cpuid->nent > nent)
2483 		cpuid->nent = nent;
2484 
2485 	for (i = 0; i < nent; i++) {
2486 		struct kvm_cpuid_entry2 *ent = &cpuid_entries[i];
2487 		u32 signature[3];
2488 
2489 		switch (ent->function) {
2490 		case HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS:
2491 			memcpy(signature, "Linux KVM Hv", 12);
2492 
2493 			ent->eax = HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES;
2494 			ent->ebx = signature[0];
2495 			ent->ecx = signature[1];
2496 			ent->edx = signature[2];
2497 			break;
2498 
2499 		case HYPERV_CPUID_INTERFACE:
2500 			ent->eax = HYPERV_CPUID_SIGNATURE_EAX;
2501 			break;
2502 
2503 		case HYPERV_CPUID_VERSION:
2504 			/*
2505 			 * We implement some Hyper-V 2016 functions so let's use
2506 			 * this version.
2507 			 */
2508 			ent->eax = 0x00003839;
2509 			ent->ebx = 0x000A0000;
2510 			break;
2511 
2512 		case HYPERV_CPUID_FEATURES:
2513 			ent->eax |= HV_MSR_VP_RUNTIME_AVAILABLE;
2514 			ent->eax |= HV_MSR_TIME_REF_COUNT_AVAILABLE;
2515 			ent->eax |= HV_MSR_SYNIC_AVAILABLE;
2516 			ent->eax |= HV_MSR_SYNTIMER_AVAILABLE;
2517 			ent->eax |= HV_MSR_APIC_ACCESS_AVAILABLE;
2518 			ent->eax |= HV_MSR_HYPERCALL_AVAILABLE;
2519 			ent->eax |= HV_MSR_VP_INDEX_AVAILABLE;
2520 			ent->eax |= HV_MSR_RESET_AVAILABLE;
2521 			ent->eax |= HV_MSR_REFERENCE_TSC_AVAILABLE;
2522 			ent->eax |= HV_ACCESS_FREQUENCY_MSRS;
2523 			ent->eax |= HV_ACCESS_REENLIGHTENMENT;
2524 
2525 			ent->ebx |= HV_POST_MESSAGES;
2526 			ent->ebx |= HV_SIGNAL_EVENTS;
2527 
2528 			ent->edx |= HV_X64_HYPERCALL_XMM_INPUT_AVAILABLE;
2529 			ent->edx |= HV_FEATURE_FREQUENCY_MSRS_AVAILABLE;
2530 			ent->edx |= HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
2531 
2532 			ent->ebx |= HV_DEBUGGING;
2533 			ent->edx |= HV_X64_GUEST_DEBUGGING_AVAILABLE;
2534 			ent->edx |= HV_FEATURE_DEBUG_MSRS_AVAILABLE;
2535 
2536 			/*
2537 			 * Direct Synthetic timers only make sense with in-kernel
2538 			 * LAPIC
2539 			 */
2540 			if (!vcpu || lapic_in_kernel(vcpu))
2541 				ent->edx |= HV_STIMER_DIRECT_MODE_AVAILABLE;
2542 
2543 			break;
2544 
2545 		case HYPERV_CPUID_ENLIGHTMENT_INFO:
2546 			ent->eax |= HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
2547 			ent->eax |= HV_X64_APIC_ACCESS_RECOMMENDED;
2548 			ent->eax |= HV_X64_RELAXED_TIMING_RECOMMENDED;
2549 			ent->eax |= HV_X64_CLUSTER_IPI_RECOMMENDED;
2550 			ent->eax |= HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED;
2551 			if (evmcs_ver)
2552 				ent->eax |= HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
2553 			if (!cpu_smt_possible())
2554 				ent->eax |= HV_X64_NO_NONARCH_CORESHARING;
2555 
2556 			ent->eax |= HV_DEPRECATING_AEOI_RECOMMENDED;
2557 			/*
2558 			 * Default number of spinlock retry attempts, matches
2559 			 * HyperV 2016.
2560 			 */
2561 			ent->ebx = 0x00000FFF;
2562 
2563 			break;
2564 
2565 		case HYPERV_CPUID_IMPLEMENT_LIMITS:
2566 			/* Maximum number of virtual processors */
2567 			ent->eax = KVM_MAX_VCPUS;
2568 			/*
2569 			 * Maximum number of logical processors, matches
2570 			 * HyperV 2016.
2571 			 */
2572 			ent->ebx = 64;
2573 
2574 			break;
2575 
2576 		case HYPERV_CPUID_NESTED_FEATURES:
2577 			ent->eax = evmcs_ver;
2578 			ent->eax |= HV_X64_NESTED_MSR_BITMAP;
2579 
2580 			break;
2581 
2582 		case HYPERV_CPUID_SYNDBG_VENDOR_AND_MAX_FUNCTIONS:
2583 			memcpy(signature, "Linux KVM Hv", 12);
2584 
2585 			ent->eax = 0;
2586 			ent->ebx = signature[0];
2587 			ent->ecx = signature[1];
2588 			ent->edx = signature[2];
2589 			break;
2590 
2591 		case HYPERV_CPUID_SYNDBG_INTERFACE:
2592 			memcpy(signature, "VS#1\0\0\0\0\0\0\0\0", 12);
2593 			ent->eax = signature[0];
2594 			break;
2595 
2596 		case HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES:
2597 			ent->eax |= HV_X64_SYNDBG_CAP_ALLOW_KERNEL_DEBUGGING;
2598 			break;
2599 
2600 		default:
2601 			break;
2602 		}
2603 	}
2604 
2605 	if (copy_to_user(entries, cpuid_entries,
2606 			 nent * sizeof(struct kvm_cpuid_entry2)))
2607 		return -EFAULT;
2608 
2609 	return 0;
2610 }
2611