xref: /openbmc/linux/arch/x86/kvm/hyperv.c (revision f97769fd)
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 
27 #include <linux/cpu.h>
28 #include <linux/kvm_host.h>
29 #include <linux/highmem.h>
30 #include <linux/sched/cputime.h>
31 #include <linux/eventfd.h>
32 
33 #include <asm/apicdef.h>
34 #include <trace/events/kvm.h>
35 
36 #include "trace.h"
37 #include "irq.h"
38 
39 #define KVM_HV_MAX_SPARSE_VCPU_SET_BITS DIV_ROUND_UP(KVM_MAX_VCPUS, 64)
40 
41 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
42 				bool vcpu_kick);
43 
44 static inline u64 synic_read_sint(struct kvm_vcpu_hv_synic *synic, int sint)
45 {
46 	return atomic64_read(&synic->sint[sint]);
47 }
48 
49 static inline int synic_get_sint_vector(u64 sint_value)
50 {
51 	if (sint_value & HV_SYNIC_SINT_MASKED)
52 		return -1;
53 	return sint_value & HV_SYNIC_SINT_VECTOR_MASK;
54 }
55 
56 static bool synic_has_vector_connected(struct kvm_vcpu_hv_synic *synic,
57 				      int vector)
58 {
59 	int i;
60 
61 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
62 		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
63 			return true;
64 	}
65 	return false;
66 }
67 
68 static bool synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic *synic,
69 				     int vector)
70 {
71 	int i;
72 	u64 sint_value;
73 
74 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
75 		sint_value = synic_read_sint(synic, i);
76 		if (synic_get_sint_vector(sint_value) == vector &&
77 		    sint_value & HV_SYNIC_SINT_AUTO_EOI)
78 			return true;
79 	}
80 	return false;
81 }
82 
83 static void synic_update_vector(struct kvm_vcpu_hv_synic *synic,
84 				int vector)
85 {
86 	if (vector < HV_SYNIC_FIRST_VALID_VECTOR)
87 		return;
88 
89 	if (synic_has_vector_connected(synic, vector))
90 		__set_bit(vector, synic->vec_bitmap);
91 	else
92 		__clear_bit(vector, synic->vec_bitmap);
93 
94 	if (synic_has_vector_auto_eoi(synic, vector))
95 		__set_bit(vector, synic->auto_eoi_bitmap);
96 	else
97 		__clear_bit(vector, synic->auto_eoi_bitmap);
98 }
99 
100 static int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint,
101 			  u64 data, bool host)
102 {
103 	int vector, old_vector;
104 	bool masked;
105 
106 	vector = data & HV_SYNIC_SINT_VECTOR_MASK;
107 	masked = data & HV_SYNIC_SINT_MASKED;
108 
109 	/*
110 	 * Valid vectors are 16-255, however, nested Hyper-V attempts to write
111 	 * default '0x10000' value on boot and this should not #GP. We need to
112 	 * allow zero-initing the register from host as well.
113 	 */
114 	if (vector < HV_SYNIC_FIRST_VALID_VECTOR && !host && !masked)
115 		return 1;
116 	/*
117 	 * Guest may configure multiple SINTs to use the same vector, so
118 	 * we maintain a bitmap of vectors handled by synic, and a
119 	 * bitmap of vectors with auto-eoi behavior.  The bitmaps are
120 	 * updated here, and atomically queried on fast paths.
121 	 */
122 	old_vector = synic_read_sint(synic, sint) & HV_SYNIC_SINT_VECTOR_MASK;
123 
124 	atomic64_set(&synic->sint[sint], data);
125 
126 	synic_update_vector(synic, old_vector);
127 
128 	synic_update_vector(synic, vector);
129 
130 	/* Load SynIC vectors into EOI exit bitmap */
131 	kvm_make_request(KVM_REQ_SCAN_IOAPIC, synic_to_vcpu(synic));
132 	return 0;
133 }
134 
135 static struct kvm_vcpu *get_vcpu_by_vpidx(struct kvm *kvm, u32 vpidx)
136 {
137 	struct kvm_vcpu *vcpu = NULL;
138 	int i;
139 
140 	if (vpidx >= KVM_MAX_VCPUS)
141 		return NULL;
142 
143 	vcpu = kvm_get_vcpu(kvm, vpidx);
144 	if (vcpu && vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
145 		return vcpu;
146 	kvm_for_each_vcpu(i, vcpu, kvm)
147 		if (vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
148 			return vcpu;
149 	return NULL;
150 }
151 
152 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vpidx)
153 {
154 	struct kvm_vcpu *vcpu;
155 	struct kvm_vcpu_hv_synic *synic;
156 
157 	vcpu = get_vcpu_by_vpidx(kvm, vpidx);
158 	if (!vcpu)
159 		return NULL;
160 	synic = vcpu_to_synic(vcpu);
161 	return (synic->active) ? synic : NULL;
162 }
163 
164 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
165 {
166 	struct kvm *kvm = vcpu->kvm;
167 	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
168 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
169 	struct kvm_vcpu_hv_stimer *stimer;
170 	int gsi, idx;
171 
172 	trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
173 
174 	/* Try to deliver pending Hyper-V SynIC timers messages */
175 	for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
176 		stimer = &hv_vcpu->stimer[idx];
177 		if (stimer->msg_pending && stimer->config.enable &&
178 		    !stimer->config.direct_mode &&
179 		    stimer->config.sintx == sint)
180 			stimer_mark_pending(stimer, false);
181 	}
182 
183 	idx = srcu_read_lock(&kvm->irq_srcu);
184 	gsi = atomic_read(&synic->sint_to_gsi[sint]);
185 	if (gsi != -1)
186 		kvm_notify_acked_gsi(kvm, gsi);
187 	srcu_read_unlock(&kvm->irq_srcu, idx);
188 }
189 
190 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
191 {
192 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
193 	struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
194 
195 	hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
196 	hv_vcpu->exit.u.synic.msr = msr;
197 	hv_vcpu->exit.u.synic.control = synic->control;
198 	hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
199 	hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
200 
201 	kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
202 }
203 
204 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
205 			 u32 msr, u64 data, bool host)
206 {
207 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
208 	int ret;
209 
210 	if (!synic->active && !host)
211 		return 1;
212 
213 	trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
214 
215 	ret = 0;
216 	switch (msr) {
217 	case HV_X64_MSR_SCONTROL:
218 		synic->control = data;
219 		if (!host)
220 			synic_exit(synic, msr);
221 		break;
222 	case HV_X64_MSR_SVERSION:
223 		if (!host) {
224 			ret = 1;
225 			break;
226 		}
227 		synic->version = data;
228 		break;
229 	case HV_X64_MSR_SIEFP:
230 		if ((data & HV_SYNIC_SIEFP_ENABLE) && !host &&
231 		    !synic->dont_zero_synic_pages)
232 			if (kvm_clear_guest(vcpu->kvm,
233 					    data & PAGE_MASK, PAGE_SIZE)) {
234 				ret = 1;
235 				break;
236 			}
237 		synic->evt_page = data;
238 		if (!host)
239 			synic_exit(synic, msr);
240 		break;
241 	case HV_X64_MSR_SIMP:
242 		if ((data & HV_SYNIC_SIMP_ENABLE) && !host &&
243 		    !synic->dont_zero_synic_pages)
244 			if (kvm_clear_guest(vcpu->kvm,
245 					    data & PAGE_MASK, PAGE_SIZE)) {
246 				ret = 1;
247 				break;
248 			}
249 		synic->msg_page = data;
250 		if (!host)
251 			synic_exit(synic, msr);
252 		break;
253 	case HV_X64_MSR_EOM: {
254 		int i;
255 
256 		for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
257 			kvm_hv_notify_acked_sint(vcpu, i);
258 		break;
259 	}
260 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
261 		ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
262 		break;
263 	default:
264 		ret = 1;
265 		break;
266 	}
267 	return ret;
268 }
269 
270 static bool kvm_hv_is_syndbg_enabled(struct kvm_vcpu *vcpu)
271 {
272 	struct kvm_cpuid_entry2 *entry;
273 
274 	entry = kvm_find_cpuid_entry(vcpu,
275 				     HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES,
276 				     0);
277 	if (!entry)
278 		return false;
279 
280 	return entry->eax & HV_X64_SYNDBG_CAP_ALLOW_KERNEL_DEBUGGING;
281 }
282 
283 static int kvm_hv_syndbg_complete_userspace(struct kvm_vcpu *vcpu)
284 {
285 	struct kvm *kvm = vcpu->kvm;
286 	struct kvm_hv *hv = &kvm->arch.hyperv;
287 
288 	if (vcpu->run->hyperv.u.syndbg.msr == HV_X64_MSR_SYNDBG_CONTROL)
289 		hv->hv_syndbg.control.status =
290 			vcpu->run->hyperv.u.syndbg.status;
291 	return 1;
292 }
293 
294 static void syndbg_exit(struct kvm_vcpu *vcpu, u32 msr)
295 {
296 	struct kvm_hv_syndbg *syndbg = vcpu_to_hv_syndbg(vcpu);
297 	struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
298 
299 	hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNDBG;
300 	hv_vcpu->exit.u.syndbg.msr = msr;
301 	hv_vcpu->exit.u.syndbg.control = syndbg->control.control;
302 	hv_vcpu->exit.u.syndbg.send_page = syndbg->control.send_page;
303 	hv_vcpu->exit.u.syndbg.recv_page = syndbg->control.recv_page;
304 	hv_vcpu->exit.u.syndbg.pending_page = syndbg->control.pending_page;
305 	vcpu->arch.complete_userspace_io =
306 			kvm_hv_syndbg_complete_userspace;
307 
308 	kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
309 }
310 
311 static int syndbg_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
312 {
313 	struct kvm_hv_syndbg *syndbg = vcpu_to_hv_syndbg(vcpu);
314 
315 	if (!kvm_hv_is_syndbg_enabled(vcpu) && !host)
316 		return 1;
317 
318 	trace_kvm_hv_syndbg_set_msr(vcpu->vcpu_id,
319 				    vcpu_to_hv_vcpu(vcpu)->vp_index, msr, data);
320 	switch (msr) {
321 	case HV_X64_MSR_SYNDBG_CONTROL:
322 		syndbg->control.control = data;
323 		if (!host)
324 			syndbg_exit(vcpu, msr);
325 		break;
326 	case HV_X64_MSR_SYNDBG_STATUS:
327 		syndbg->control.status = data;
328 		break;
329 	case HV_X64_MSR_SYNDBG_SEND_BUFFER:
330 		syndbg->control.send_page = data;
331 		break;
332 	case HV_X64_MSR_SYNDBG_RECV_BUFFER:
333 		syndbg->control.recv_page = data;
334 		break;
335 	case HV_X64_MSR_SYNDBG_PENDING_BUFFER:
336 		syndbg->control.pending_page = data;
337 		if (!host)
338 			syndbg_exit(vcpu, msr);
339 		break;
340 	case HV_X64_MSR_SYNDBG_OPTIONS:
341 		syndbg->options = data;
342 		break;
343 	default:
344 		break;
345 	}
346 
347 	return 0;
348 }
349 
350 static int syndbg_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
351 {
352 	struct kvm_hv_syndbg *syndbg = vcpu_to_hv_syndbg(vcpu);
353 
354 	if (!kvm_hv_is_syndbg_enabled(vcpu) && !host)
355 		return 1;
356 
357 	switch (msr) {
358 	case HV_X64_MSR_SYNDBG_CONTROL:
359 		*pdata = syndbg->control.control;
360 		break;
361 	case HV_X64_MSR_SYNDBG_STATUS:
362 		*pdata = syndbg->control.status;
363 		break;
364 	case HV_X64_MSR_SYNDBG_SEND_BUFFER:
365 		*pdata = syndbg->control.send_page;
366 		break;
367 	case HV_X64_MSR_SYNDBG_RECV_BUFFER:
368 		*pdata = syndbg->control.recv_page;
369 		break;
370 	case HV_X64_MSR_SYNDBG_PENDING_BUFFER:
371 		*pdata = syndbg->control.pending_page;
372 		break;
373 	case HV_X64_MSR_SYNDBG_OPTIONS:
374 		*pdata = syndbg->options;
375 		break;
376 	default:
377 		break;
378 	}
379 
380 	trace_kvm_hv_syndbg_get_msr(vcpu->vcpu_id,
381 				    vcpu_to_hv_vcpu(vcpu)->vp_index, msr,
382 				    *pdata);
383 
384 	return 0;
385 }
386 
387 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata,
388 			 bool host)
389 {
390 	int ret;
391 
392 	if (!synic->active && !host)
393 		return 1;
394 
395 	ret = 0;
396 	switch (msr) {
397 	case HV_X64_MSR_SCONTROL:
398 		*pdata = synic->control;
399 		break;
400 	case HV_X64_MSR_SVERSION:
401 		*pdata = synic->version;
402 		break;
403 	case HV_X64_MSR_SIEFP:
404 		*pdata = synic->evt_page;
405 		break;
406 	case HV_X64_MSR_SIMP:
407 		*pdata = synic->msg_page;
408 		break;
409 	case HV_X64_MSR_EOM:
410 		*pdata = 0;
411 		break;
412 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
413 		*pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
414 		break;
415 	default:
416 		ret = 1;
417 		break;
418 	}
419 	return ret;
420 }
421 
422 static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
423 {
424 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
425 	struct kvm_lapic_irq irq;
426 	int ret, vector;
427 
428 	if (sint >= ARRAY_SIZE(synic->sint))
429 		return -EINVAL;
430 
431 	vector = synic_get_sint_vector(synic_read_sint(synic, sint));
432 	if (vector < 0)
433 		return -ENOENT;
434 
435 	memset(&irq, 0, sizeof(irq));
436 	irq.shorthand = APIC_DEST_SELF;
437 	irq.dest_mode = APIC_DEST_PHYSICAL;
438 	irq.delivery_mode = APIC_DM_FIXED;
439 	irq.vector = vector;
440 	irq.level = 1;
441 
442 	ret = kvm_irq_delivery_to_apic(vcpu->kvm, vcpu->arch.apic, &irq, NULL);
443 	trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
444 	return ret;
445 }
446 
447 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vpidx, u32 sint)
448 {
449 	struct kvm_vcpu_hv_synic *synic;
450 
451 	synic = synic_get(kvm, vpidx);
452 	if (!synic)
453 		return -EINVAL;
454 
455 	return synic_set_irq(synic, sint);
456 }
457 
458 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
459 {
460 	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
461 	int i;
462 
463 	trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
464 
465 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
466 		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
467 			kvm_hv_notify_acked_sint(vcpu, i);
468 }
469 
470 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi)
471 {
472 	struct kvm_vcpu_hv_synic *synic;
473 
474 	synic = synic_get(kvm, vpidx);
475 	if (!synic)
476 		return -EINVAL;
477 
478 	if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
479 		return -EINVAL;
480 
481 	atomic_set(&synic->sint_to_gsi[sint], gsi);
482 	return 0;
483 }
484 
485 void kvm_hv_irq_routing_update(struct kvm *kvm)
486 {
487 	struct kvm_irq_routing_table *irq_rt;
488 	struct kvm_kernel_irq_routing_entry *e;
489 	u32 gsi;
490 
491 	irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
492 					lockdep_is_held(&kvm->irq_lock));
493 
494 	for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
495 		hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
496 			if (e->type == KVM_IRQ_ROUTING_HV_SINT)
497 				kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
498 						    e->hv_sint.sint, gsi);
499 		}
500 	}
501 }
502 
503 static void synic_init(struct kvm_vcpu_hv_synic *synic)
504 {
505 	int i;
506 
507 	memset(synic, 0, sizeof(*synic));
508 	synic->version = HV_SYNIC_VERSION_1;
509 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
510 		atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
511 		atomic_set(&synic->sint_to_gsi[i], -1);
512 	}
513 }
514 
515 static u64 get_time_ref_counter(struct kvm *kvm)
516 {
517 	struct kvm_hv *hv = &kvm->arch.hyperv;
518 	struct kvm_vcpu *vcpu;
519 	u64 tsc;
520 
521 	/*
522 	 * The guest has not set up the TSC page or the clock isn't
523 	 * stable, fall back to get_kvmclock_ns.
524 	 */
525 	if (!hv->tsc_ref.tsc_sequence)
526 		return div_u64(get_kvmclock_ns(kvm), 100);
527 
528 	vcpu = kvm_get_vcpu(kvm, 0);
529 	tsc = kvm_read_l1_tsc(vcpu, rdtsc());
530 	return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64)
531 		+ hv->tsc_ref.tsc_offset;
532 }
533 
534 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
535 				bool vcpu_kick)
536 {
537 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
538 
539 	set_bit(stimer->index,
540 		vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
541 	kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
542 	if (vcpu_kick)
543 		kvm_vcpu_kick(vcpu);
544 }
545 
546 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
547 {
548 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
549 
550 	trace_kvm_hv_stimer_cleanup(stimer_to_vcpu(stimer)->vcpu_id,
551 				    stimer->index);
552 
553 	hrtimer_cancel(&stimer->timer);
554 	clear_bit(stimer->index,
555 		  vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
556 	stimer->msg_pending = false;
557 	stimer->exp_time = 0;
558 }
559 
560 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
561 {
562 	struct kvm_vcpu_hv_stimer *stimer;
563 
564 	stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
565 	trace_kvm_hv_stimer_callback(stimer_to_vcpu(stimer)->vcpu_id,
566 				     stimer->index);
567 	stimer_mark_pending(stimer, true);
568 
569 	return HRTIMER_NORESTART;
570 }
571 
572 /*
573  * stimer_start() assumptions:
574  * a) stimer->count is not equal to 0
575  * b) stimer->config has HV_STIMER_ENABLE flag
576  */
577 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
578 {
579 	u64 time_now;
580 	ktime_t ktime_now;
581 
582 	time_now = get_time_ref_counter(stimer_to_vcpu(stimer)->kvm);
583 	ktime_now = ktime_get();
584 
585 	if (stimer->config.periodic) {
586 		if (stimer->exp_time) {
587 			if (time_now >= stimer->exp_time) {
588 				u64 remainder;
589 
590 				div64_u64_rem(time_now - stimer->exp_time,
591 					      stimer->count, &remainder);
592 				stimer->exp_time =
593 					time_now + (stimer->count - remainder);
594 			}
595 		} else
596 			stimer->exp_time = time_now + stimer->count;
597 
598 		trace_kvm_hv_stimer_start_periodic(
599 					stimer_to_vcpu(stimer)->vcpu_id,
600 					stimer->index,
601 					time_now, stimer->exp_time);
602 
603 		hrtimer_start(&stimer->timer,
604 			      ktime_add_ns(ktime_now,
605 					   100 * (stimer->exp_time - time_now)),
606 			      HRTIMER_MODE_ABS);
607 		return 0;
608 	}
609 	stimer->exp_time = stimer->count;
610 	if (time_now >= stimer->count) {
611 		/*
612 		 * Expire timer according to Hypervisor Top-Level Functional
613 		 * specification v4(15.3.1):
614 		 * "If a one shot is enabled and the specified count is in
615 		 * the past, it will expire immediately."
616 		 */
617 		stimer_mark_pending(stimer, false);
618 		return 0;
619 	}
620 
621 	trace_kvm_hv_stimer_start_one_shot(stimer_to_vcpu(stimer)->vcpu_id,
622 					   stimer->index,
623 					   time_now, stimer->count);
624 
625 	hrtimer_start(&stimer->timer,
626 		      ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)),
627 		      HRTIMER_MODE_ABS);
628 	return 0;
629 }
630 
631 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
632 			     bool host)
633 {
634 	union hv_stimer_config new_config = {.as_uint64 = config},
635 		old_config = {.as_uint64 = stimer->config.as_uint64};
636 
637 	trace_kvm_hv_stimer_set_config(stimer_to_vcpu(stimer)->vcpu_id,
638 				       stimer->index, config, host);
639 
640 	stimer_cleanup(stimer);
641 	if (old_config.enable &&
642 	    !new_config.direct_mode && new_config.sintx == 0)
643 		new_config.enable = 0;
644 	stimer->config.as_uint64 = new_config.as_uint64;
645 
646 	if (stimer->config.enable)
647 		stimer_mark_pending(stimer, false);
648 
649 	return 0;
650 }
651 
652 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
653 			    bool host)
654 {
655 	trace_kvm_hv_stimer_set_count(stimer_to_vcpu(stimer)->vcpu_id,
656 				      stimer->index, count, host);
657 
658 	stimer_cleanup(stimer);
659 	stimer->count = count;
660 	if (stimer->count == 0)
661 		stimer->config.enable = 0;
662 	else if (stimer->config.auto_enable)
663 		stimer->config.enable = 1;
664 
665 	if (stimer->config.enable)
666 		stimer_mark_pending(stimer, false);
667 
668 	return 0;
669 }
670 
671 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
672 {
673 	*pconfig = stimer->config.as_uint64;
674 	return 0;
675 }
676 
677 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
678 {
679 	*pcount = stimer->count;
680 	return 0;
681 }
682 
683 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
684 			     struct hv_message *src_msg, bool no_retry)
685 {
686 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
687 	int msg_off = offsetof(struct hv_message_page, sint_message[sint]);
688 	gfn_t msg_page_gfn;
689 	struct hv_message_header hv_hdr;
690 	int r;
691 
692 	if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
693 		return -ENOENT;
694 
695 	msg_page_gfn = synic->msg_page >> PAGE_SHIFT;
696 
697 	/*
698 	 * Strictly following the spec-mandated ordering would assume setting
699 	 * .msg_pending before checking .message_type.  However, this function
700 	 * is only called in vcpu context so the entire update is atomic from
701 	 * guest POV and thus the exact order here doesn't matter.
702 	 */
703 	r = kvm_vcpu_read_guest_page(vcpu, msg_page_gfn, &hv_hdr.message_type,
704 				     msg_off + offsetof(struct hv_message,
705 							header.message_type),
706 				     sizeof(hv_hdr.message_type));
707 	if (r < 0)
708 		return r;
709 
710 	if (hv_hdr.message_type != HVMSG_NONE) {
711 		if (no_retry)
712 			return 0;
713 
714 		hv_hdr.message_flags.msg_pending = 1;
715 		r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn,
716 					      &hv_hdr.message_flags,
717 					      msg_off +
718 					      offsetof(struct hv_message,
719 						       header.message_flags),
720 					      sizeof(hv_hdr.message_flags));
721 		if (r < 0)
722 			return r;
723 		return -EAGAIN;
724 	}
725 
726 	r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn, src_msg, msg_off,
727 				      sizeof(src_msg->header) +
728 				      src_msg->header.payload_size);
729 	if (r < 0)
730 		return r;
731 
732 	r = synic_set_irq(synic, sint);
733 	if (r < 0)
734 		return r;
735 	if (r == 0)
736 		return -EFAULT;
737 	return 0;
738 }
739 
740 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
741 {
742 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
743 	struct hv_message *msg = &stimer->msg;
744 	struct hv_timer_message_payload *payload =
745 			(struct hv_timer_message_payload *)&msg->u.payload;
746 
747 	/*
748 	 * To avoid piling up periodic ticks, don't retry message
749 	 * delivery for them (within "lazy" lost ticks policy).
750 	 */
751 	bool no_retry = stimer->config.periodic;
752 
753 	payload->expiration_time = stimer->exp_time;
754 	payload->delivery_time = get_time_ref_counter(vcpu->kvm);
755 	return synic_deliver_msg(vcpu_to_synic(vcpu),
756 				 stimer->config.sintx, msg,
757 				 no_retry);
758 }
759 
760 static int stimer_notify_direct(struct kvm_vcpu_hv_stimer *stimer)
761 {
762 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
763 	struct kvm_lapic_irq irq = {
764 		.delivery_mode = APIC_DM_FIXED,
765 		.vector = stimer->config.apic_vector
766 	};
767 
768 	if (lapic_in_kernel(vcpu))
769 		return !kvm_apic_set_irq(vcpu, &irq, NULL);
770 	return 0;
771 }
772 
773 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
774 {
775 	int r, direct = stimer->config.direct_mode;
776 
777 	stimer->msg_pending = true;
778 	if (!direct)
779 		r = stimer_send_msg(stimer);
780 	else
781 		r = stimer_notify_direct(stimer);
782 	trace_kvm_hv_stimer_expiration(stimer_to_vcpu(stimer)->vcpu_id,
783 				       stimer->index, direct, r);
784 	if (!r) {
785 		stimer->msg_pending = false;
786 		if (!(stimer->config.periodic))
787 			stimer->config.enable = 0;
788 	}
789 }
790 
791 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
792 {
793 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
794 	struct kvm_vcpu_hv_stimer *stimer;
795 	u64 time_now, exp_time;
796 	int i;
797 
798 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
799 		if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
800 			stimer = &hv_vcpu->stimer[i];
801 			if (stimer->config.enable) {
802 				exp_time = stimer->exp_time;
803 
804 				if (exp_time) {
805 					time_now =
806 						get_time_ref_counter(vcpu->kvm);
807 					if (time_now >= exp_time)
808 						stimer_expiration(stimer);
809 				}
810 
811 				if ((stimer->config.enable) &&
812 				    stimer->count) {
813 					if (!stimer->msg_pending)
814 						stimer_start(stimer);
815 				} else
816 					stimer_cleanup(stimer);
817 			}
818 		}
819 }
820 
821 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
822 {
823 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
824 	int i;
825 
826 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
827 		stimer_cleanup(&hv_vcpu->stimer[i]);
828 }
829 
830 bool kvm_hv_assist_page_enabled(struct kvm_vcpu *vcpu)
831 {
832 	if (!(vcpu->arch.hyperv.hv_vapic & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE))
833 		return false;
834 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
835 }
836 EXPORT_SYMBOL_GPL(kvm_hv_assist_page_enabled);
837 
838 bool kvm_hv_get_assist_page(struct kvm_vcpu *vcpu,
839 			    struct hv_vp_assist_page *assist_page)
840 {
841 	if (!kvm_hv_assist_page_enabled(vcpu))
842 		return false;
843 	return !kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data,
844 				      assist_page, sizeof(*assist_page));
845 }
846 EXPORT_SYMBOL_GPL(kvm_hv_get_assist_page);
847 
848 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
849 {
850 	struct hv_message *msg = &stimer->msg;
851 	struct hv_timer_message_payload *payload =
852 			(struct hv_timer_message_payload *)&msg->u.payload;
853 
854 	memset(&msg->header, 0, sizeof(msg->header));
855 	msg->header.message_type = HVMSG_TIMER_EXPIRED;
856 	msg->header.payload_size = sizeof(*payload);
857 
858 	payload->timer_index = stimer->index;
859 	payload->expiration_time = 0;
860 	payload->delivery_time = 0;
861 }
862 
863 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
864 {
865 	memset(stimer, 0, sizeof(*stimer));
866 	stimer->index = timer_index;
867 	hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
868 	stimer->timer.function = stimer_timer_callback;
869 	stimer_prepare_msg(stimer);
870 }
871 
872 void kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
873 {
874 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
875 	int i;
876 
877 	synic_init(&hv_vcpu->synic);
878 
879 	bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
880 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
881 		stimer_init(&hv_vcpu->stimer[i], i);
882 }
883 
884 void kvm_hv_vcpu_postcreate(struct kvm_vcpu *vcpu)
885 {
886 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
887 
888 	hv_vcpu->vp_index = kvm_vcpu_get_idx(vcpu);
889 }
890 
891 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
892 {
893 	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
894 
895 	/*
896 	 * Hyper-V SynIC auto EOI SINT's are
897 	 * not compatible with APICV, so request
898 	 * to deactivate APICV permanently.
899 	 */
900 	kvm_request_apicv_update(vcpu->kvm, false, APICV_INHIBIT_REASON_HYPERV);
901 	synic->active = true;
902 	synic->dont_zero_synic_pages = dont_zero_synic_pages;
903 	synic->control = HV_SYNIC_CONTROL_ENABLE;
904 	return 0;
905 }
906 
907 static bool kvm_hv_msr_partition_wide(u32 msr)
908 {
909 	bool r = false;
910 
911 	switch (msr) {
912 	case HV_X64_MSR_GUEST_OS_ID:
913 	case HV_X64_MSR_HYPERCALL:
914 	case HV_X64_MSR_REFERENCE_TSC:
915 	case HV_X64_MSR_TIME_REF_COUNT:
916 	case HV_X64_MSR_CRASH_CTL:
917 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
918 	case HV_X64_MSR_RESET:
919 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
920 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
921 	case HV_X64_MSR_TSC_EMULATION_STATUS:
922 	case HV_X64_MSR_SYNDBG_OPTIONS:
923 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
924 		r = true;
925 		break;
926 	}
927 
928 	return r;
929 }
930 
931 static int kvm_hv_msr_get_crash_data(struct kvm_vcpu *vcpu,
932 				     u32 index, u64 *pdata)
933 {
934 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
935 	size_t size = ARRAY_SIZE(hv->hv_crash_param);
936 
937 	if (WARN_ON_ONCE(index >= size))
938 		return -EINVAL;
939 
940 	*pdata = hv->hv_crash_param[array_index_nospec(index, size)];
941 	return 0;
942 }
943 
944 static int kvm_hv_msr_get_crash_ctl(struct kvm_vcpu *vcpu, u64 *pdata)
945 {
946 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
947 
948 	*pdata = hv->hv_crash_ctl;
949 	return 0;
950 }
951 
952 static int kvm_hv_msr_set_crash_ctl(struct kvm_vcpu *vcpu, u64 data, bool host)
953 {
954 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
955 
956 	if (host)
957 		hv->hv_crash_ctl = data & HV_CRASH_CTL_CRASH_NOTIFY;
958 
959 	if (!host && (data & HV_CRASH_CTL_CRASH_NOTIFY)) {
960 
961 		vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
962 			  hv->hv_crash_param[0],
963 			  hv->hv_crash_param[1],
964 			  hv->hv_crash_param[2],
965 			  hv->hv_crash_param[3],
966 			  hv->hv_crash_param[4]);
967 
968 		/* Send notification about crash to user space */
969 		kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
970 	}
971 
972 	return 0;
973 }
974 
975 static int kvm_hv_msr_set_crash_data(struct kvm_vcpu *vcpu,
976 				     u32 index, u64 data)
977 {
978 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
979 	size_t size = ARRAY_SIZE(hv->hv_crash_param);
980 
981 	if (WARN_ON_ONCE(index >= size))
982 		return -EINVAL;
983 
984 	hv->hv_crash_param[array_index_nospec(index, size)] = data;
985 	return 0;
986 }
987 
988 /*
989  * The kvmclock and Hyper-V TSC page use similar formulas, and converting
990  * between them is possible:
991  *
992  * kvmclock formula:
993  *    nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
994  *           + system_time
995  *
996  * Hyper-V formula:
997  *    nsec/100 = ticks * scale / 2^64 + offset
998  *
999  * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
1000  * By dividing the kvmclock formula by 100 and equating what's left we get:
1001  *    ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1002  *            scale / 2^64 =         tsc_to_system_mul * 2^(tsc_shift-32) / 100
1003  *            scale        =         tsc_to_system_mul * 2^(32+tsc_shift) / 100
1004  *
1005  * Now expand the kvmclock formula and divide by 100:
1006  *    nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
1007  *           - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
1008  *           + system_time
1009  *    nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1010  *               - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1011  *               + system_time / 100
1012  *
1013  * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
1014  *    nsec/100 = ticks * scale / 2^64
1015  *               - tsc_timestamp * scale / 2^64
1016  *               + system_time / 100
1017  *
1018  * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
1019  *    offset = system_time / 100 - tsc_timestamp * scale / 2^64
1020  *
1021  * These two equivalencies are implemented in this function.
1022  */
1023 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
1024 					struct ms_hyperv_tsc_page *tsc_ref)
1025 {
1026 	u64 max_mul;
1027 
1028 	if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
1029 		return false;
1030 
1031 	/*
1032 	 * check if scale would overflow, if so we use the time ref counter
1033 	 *    tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
1034 	 *    tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
1035 	 *    tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
1036 	 */
1037 	max_mul = 100ull << (32 - hv_clock->tsc_shift);
1038 	if (hv_clock->tsc_to_system_mul >= max_mul)
1039 		return false;
1040 
1041 	/*
1042 	 * Otherwise compute the scale and offset according to the formulas
1043 	 * derived above.
1044 	 */
1045 	tsc_ref->tsc_scale =
1046 		mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
1047 				hv_clock->tsc_to_system_mul,
1048 				100);
1049 
1050 	tsc_ref->tsc_offset = hv_clock->system_time;
1051 	do_div(tsc_ref->tsc_offset, 100);
1052 	tsc_ref->tsc_offset -=
1053 		mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
1054 	return true;
1055 }
1056 
1057 void kvm_hv_setup_tsc_page(struct kvm *kvm,
1058 			   struct pvclock_vcpu_time_info *hv_clock)
1059 {
1060 	struct kvm_hv *hv = &kvm->arch.hyperv;
1061 	u32 tsc_seq;
1062 	u64 gfn;
1063 
1064 	BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
1065 	BUILD_BUG_ON(offsetof(struct ms_hyperv_tsc_page, tsc_sequence) != 0);
1066 
1067 	if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1068 		return;
1069 
1070 	mutex_lock(&kvm->arch.hyperv.hv_lock);
1071 	if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1072 		goto out_unlock;
1073 
1074 	gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
1075 	/*
1076 	 * Because the TSC parameters only vary when there is a
1077 	 * change in the master clock, do not bother with caching.
1078 	 */
1079 	if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
1080 				    &tsc_seq, sizeof(tsc_seq))))
1081 		goto out_unlock;
1082 
1083 	/*
1084 	 * While we're computing and writing the parameters, force the
1085 	 * guest to use the time reference count MSR.
1086 	 */
1087 	hv->tsc_ref.tsc_sequence = 0;
1088 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
1089 			    &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
1090 		goto out_unlock;
1091 
1092 	if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
1093 		goto out_unlock;
1094 
1095 	/* Ensure sequence is zero before writing the rest of the struct.  */
1096 	smp_wmb();
1097 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
1098 		goto out_unlock;
1099 
1100 	/*
1101 	 * Now switch to the TSC page mechanism by writing the sequence.
1102 	 */
1103 	tsc_seq++;
1104 	if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
1105 		tsc_seq = 1;
1106 
1107 	/* Write the struct entirely before the non-zero sequence.  */
1108 	smp_wmb();
1109 
1110 	hv->tsc_ref.tsc_sequence = tsc_seq;
1111 	kvm_write_guest(kvm, gfn_to_gpa(gfn),
1112 			&hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence));
1113 out_unlock:
1114 	mutex_unlock(&kvm->arch.hyperv.hv_lock);
1115 }
1116 
1117 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
1118 			     bool host)
1119 {
1120 	struct kvm *kvm = vcpu->kvm;
1121 	struct kvm_hv *hv = &kvm->arch.hyperv;
1122 
1123 	switch (msr) {
1124 	case HV_X64_MSR_GUEST_OS_ID:
1125 		hv->hv_guest_os_id = data;
1126 		/* setting guest os id to zero disables hypercall page */
1127 		if (!hv->hv_guest_os_id)
1128 			hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1129 		break;
1130 	case HV_X64_MSR_HYPERCALL: {
1131 		u64 gfn;
1132 		unsigned long addr;
1133 		u8 instructions[4];
1134 
1135 		/* if guest os id is not set hypercall should remain disabled */
1136 		if (!hv->hv_guest_os_id)
1137 			break;
1138 		if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1139 			hv->hv_hypercall = data;
1140 			break;
1141 		}
1142 		gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1143 		addr = gfn_to_hva(kvm, gfn);
1144 		if (kvm_is_error_hva(addr))
1145 			return 1;
1146 		kvm_x86_ops.patch_hypercall(vcpu, instructions);
1147 		((unsigned char *)instructions)[3] = 0xc3; /* ret */
1148 		if (__copy_to_user((void __user *)addr, instructions, 4))
1149 			return 1;
1150 		hv->hv_hypercall = data;
1151 		mark_page_dirty(kvm, gfn);
1152 		break;
1153 	}
1154 	case HV_X64_MSR_REFERENCE_TSC:
1155 		hv->hv_tsc_page = data;
1156 		if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE)
1157 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1158 		break;
1159 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1160 		return kvm_hv_msr_set_crash_data(vcpu,
1161 						 msr - HV_X64_MSR_CRASH_P0,
1162 						 data);
1163 	case HV_X64_MSR_CRASH_CTL:
1164 		return kvm_hv_msr_set_crash_ctl(vcpu, data, host);
1165 	case HV_X64_MSR_RESET:
1166 		if (data == 1) {
1167 			vcpu_debug(vcpu, "hyper-v reset requested\n");
1168 			kvm_make_request(KVM_REQ_HV_RESET, vcpu);
1169 		}
1170 		break;
1171 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1172 		hv->hv_reenlightenment_control = data;
1173 		break;
1174 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1175 		hv->hv_tsc_emulation_control = data;
1176 		break;
1177 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1178 		hv->hv_tsc_emulation_status = data;
1179 		break;
1180 	case HV_X64_MSR_TIME_REF_COUNT:
1181 		/* read-only, but still ignore it if host-initiated */
1182 		if (!host)
1183 			return 1;
1184 		break;
1185 	case HV_X64_MSR_SYNDBG_OPTIONS:
1186 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1187 		return syndbg_set_msr(vcpu, msr, data, host);
1188 	default:
1189 		vcpu_unimpl(vcpu, "Hyper-V unhandled wrmsr: 0x%x data 0x%llx\n",
1190 			    msr, data);
1191 		return 1;
1192 	}
1193 	return 0;
1194 }
1195 
1196 /* Calculate cpu time spent by current task in 100ns units */
1197 static u64 current_task_runtime_100ns(void)
1198 {
1199 	u64 utime, stime;
1200 
1201 	task_cputime_adjusted(current, &utime, &stime);
1202 
1203 	return div_u64(utime + stime, 100);
1204 }
1205 
1206 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1207 {
1208 	struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
1209 
1210 	switch (msr) {
1211 	case HV_X64_MSR_VP_INDEX: {
1212 		struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
1213 		int vcpu_idx = kvm_vcpu_get_idx(vcpu);
1214 		u32 new_vp_index = (u32)data;
1215 
1216 		if (!host || new_vp_index >= KVM_MAX_VCPUS)
1217 			return 1;
1218 
1219 		if (new_vp_index == hv_vcpu->vp_index)
1220 			return 0;
1221 
1222 		/*
1223 		 * The VP index is initialized to vcpu_index by
1224 		 * kvm_hv_vcpu_postcreate so they initially match.  Now the
1225 		 * VP index is changing, adjust num_mismatched_vp_indexes if
1226 		 * it now matches or no longer matches vcpu_idx.
1227 		 */
1228 		if (hv_vcpu->vp_index == vcpu_idx)
1229 			atomic_inc(&hv->num_mismatched_vp_indexes);
1230 		else if (new_vp_index == vcpu_idx)
1231 			atomic_dec(&hv->num_mismatched_vp_indexes);
1232 
1233 		hv_vcpu->vp_index = new_vp_index;
1234 		break;
1235 	}
1236 	case HV_X64_MSR_VP_ASSIST_PAGE: {
1237 		u64 gfn;
1238 		unsigned long addr;
1239 
1240 		if (!(data & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE)) {
1241 			hv_vcpu->hv_vapic = data;
1242 			if (kvm_lapic_enable_pv_eoi(vcpu, 0, 0))
1243 				return 1;
1244 			break;
1245 		}
1246 		gfn = data >> HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT;
1247 		addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1248 		if (kvm_is_error_hva(addr))
1249 			return 1;
1250 
1251 		/*
1252 		 * Clear apic_assist portion of struct hv_vp_assist_page
1253 		 * only, there can be valuable data in the rest which needs
1254 		 * to be preserved e.g. on migration.
1255 		 */
1256 		if (__put_user(0, (u32 __user *)addr))
1257 			return 1;
1258 		hv_vcpu->hv_vapic = data;
1259 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
1260 		if (kvm_lapic_enable_pv_eoi(vcpu,
1261 					    gfn_to_gpa(gfn) | KVM_MSR_ENABLED,
1262 					    sizeof(struct hv_vp_assist_page)))
1263 			return 1;
1264 		break;
1265 	}
1266 	case HV_X64_MSR_EOI:
1267 		return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1268 	case HV_X64_MSR_ICR:
1269 		return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1270 	case HV_X64_MSR_TPR:
1271 		return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1272 	case HV_X64_MSR_VP_RUNTIME:
1273 		if (!host)
1274 			return 1;
1275 		hv_vcpu->runtime_offset = data - current_task_runtime_100ns();
1276 		break;
1277 	case HV_X64_MSR_SCONTROL:
1278 	case HV_X64_MSR_SVERSION:
1279 	case HV_X64_MSR_SIEFP:
1280 	case HV_X64_MSR_SIMP:
1281 	case HV_X64_MSR_EOM:
1282 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1283 		return synic_set_msr(vcpu_to_synic(vcpu), msr, data, host);
1284 	case HV_X64_MSR_STIMER0_CONFIG:
1285 	case HV_X64_MSR_STIMER1_CONFIG:
1286 	case HV_X64_MSR_STIMER2_CONFIG:
1287 	case HV_X64_MSR_STIMER3_CONFIG: {
1288 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1289 
1290 		return stimer_set_config(vcpu_to_stimer(vcpu, timer_index),
1291 					 data, host);
1292 	}
1293 	case HV_X64_MSR_STIMER0_COUNT:
1294 	case HV_X64_MSR_STIMER1_COUNT:
1295 	case HV_X64_MSR_STIMER2_COUNT:
1296 	case HV_X64_MSR_STIMER3_COUNT: {
1297 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1298 
1299 		return stimer_set_count(vcpu_to_stimer(vcpu, timer_index),
1300 					data, host);
1301 	}
1302 	case HV_X64_MSR_TSC_FREQUENCY:
1303 	case HV_X64_MSR_APIC_FREQUENCY:
1304 		/* read-only, but still ignore it if host-initiated */
1305 		if (!host)
1306 			return 1;
1307 		break;
1308 	default:
1309 		vcpu_unimpl(vcpu, "Hyper-V unhandled wrmsr: 0x%x data 0x%llx\n",
1310 			    msr, data);
1311 		return 1;
1312 	}
1313 
1314 	return 0;
1315 }
1316 
1317 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1318 			     bool host)
1319 {
1320 	u64 data = 0;
1321 	struct kvm *kvm = vcpu->kvm;
1322 	struct kvm_hv *hv = &kvm->arch.hyperv;
1323 
1324 	switch (msr) {
1325 	case HV_X64_MSR_GUEST_OS_ID:
1326 		data = hv->hv_guest_os_id;
1327 		break;
1328 	case HV_X64_MSR_HYPERCALL:
1329 		data = hv->hv_hypercall;
1330 		break;
1331 	case HV_X64_MSR_TIME_REF_COUNT:
1332 		data = get_time_ref_counter(kvm);
1333 		break;
1334 	case HV_X64_MSR_REFERENCE_TSC:
1335 		data = hv->hv_tsc_page;
1336 		break;
1337 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1338 		return kvm_hv_msr_get_crash_data(vcpu,
1339 						 msr - HV_X64_MSR_CRASH_P0,
1340 						 pdata);
1341 	case HV_X64_MSR_CRASH_CTL:
1342 		return kvm_hv_msr_get_crash_ctl(vcpu, pdata);
1343 	case HV_X64_MSR_RESET:
1344 		data = 0;
1345 		break;
1346 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1347 		data = hv->hv_reenlightenment_control;
1348 		break;
1349 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1350 		data = hv->hv_tsc_emulation_control;
1351 		break;
1352 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1353 		data = hv->hv_tsc_emulation_status;
1354 		break;
1355 	case HV_X64_MSR_SYNDBG_OPTIONS:
1356 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1357 		return syndbg_get_msr(vcpu, msr, pdata, host);
1358 	default:
1359 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1360 		return 1;
1361 	}
1362 
1363 	*pdata = data;
1364 	return 0;
1365 }
1366 
1367 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1368 			  bool host)
1369 {
1370 	u64 data = 0;
1371 	struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
1372 
1373 	switch (msr) {
1374 	case HV_X64_MSR_VP_INDEX:
1375 		data = hv_vcpu->vp_index;
1376 		break;
1377 	case HV_X64_MSR_EOI:
1378 		return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1379 	case HV_X64_MSR_ICR:
1380 		return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1381 	case HV_X64_MSR_TPR:
1382 		return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1383 	case HV_X64_MSR_VP_ASSIST_PAGE:
1384 		data = hv_vcpu->hv_vapic;
1385 		break;
1386 	case HV_X64_MSR_VP_RUNTIME:
1387 		data = current_task_runtime_100ns() + hv_vcpu->runtime_offset;
1388 		break;
1389 	case HV_X64_MSR_SCONTROL:
1390 	case HV_X64_MSR_SVERSION:
1391 	case HV_X64_MSR_SIEFP:
1392 	case HV_X64_MSR_SIMP:
1393 	case HV_X64_MSR_EOM:
1394 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1395 		return synic_get_msr(vcpu_to_synic(vcpu), msr, pdata, host);
1396 	case HV_X64_MSR_STIMER0_CONFIG:
1397 	case HV_X64_MSR_STIMER1_CONFIG:
1398 	case HV_X64_MSR_STIMER2_CONFIG:
1399 	case HV_X64_MSR_STIMER3_CONFIG: {
1400 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1401 
1402 		return stimer_get_config(vcpu_to_stimer(vcpu, timer_index),
1403 					 pdata);
1404 	}
1405 	case HV_X64_MSR_STIMER0_COUNT:
1406 	case HV_X64_MSR_STIMER1_COUNT:
1407 	case HV_X64_MSR_STIMER2_COUNT:
1408 	case HV_X64_MSR_STIMER3_COUNT: {
1409 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1410 
1411 		return stimer_get_count(vcpu_to_stimer(vcpu, timer_index),
1412 					pdata);
1413 	}
1414 	case HV_X64_MSR_TSC_FREQUENCY:
1415 		data = (u64)vcpu->arch.virtual_tsc_khz * 1000;
1416 		break;
1417 	case HV_X64_MSR_APIC_FREQUENCY:
1418 		data = APIC_BUS_FREQUENCY;
1419 		break;
1420 	default:
1421 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1422 		return 1;
1423 	}
1424 	*pdata = data;
1425 	return 0;
1426 }
1427 
1428 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1429 {
1430 	if (kvm_hv_msr_partition_wide(msr)) {
1431 		int r;
1432 
1433 		mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1434 		r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1435 		mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1436 		return r;
1437 	} else
1438 		return kvm_hv_set_msr(vcpu, msr, data, host);
1439 }
1440 
1441 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
1442 {
1443 	if (kvm_hv_msr_partition_wide(msr)) {
1444 		int r;
1445 
1446 		mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1447 		r = kvm_hv_get_msr_pw(vcpu, msr, pdata, host);
1448 		mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1449 		return r;
1450 	} else
1451 		return kvm_hv_get_msr(vcpu, msr, pdata, host);
1452 }
1453 
1454 static __always_inline unsigned long *sparse_set_to_vcpu_mask(
1455 	struct kvm *kvm, u64 *sparse_banks, u64 valid_bank_mask,
1456 	u64 *vp_bitmap, unsigned long *vcpu_bitmap)
1457 {
1458 	struct kvm_hv *hv = &kvm->arch.hyperv;
1459 	struct kvm_vcpu *vcpu;
1460 	int i, bank, sbank = 0;
1461 
1462 	memset(vp_bitmap, 0,
1463 	       KVM_HV_MAX_SPARSE_VCPU_SET_BITS * sizeof(*vp_bitmap));
1464 	for_each_set_bit(bank, (unsigned long *)&valid_bank_mask,
1465 			 KVM_HV_MAX_SPARSE_VCPU_SET_BITS)
1466 		vp_bitmap[bank] = sparse_banks[sbank++];
1467 
1468 	if (likely(!atomic_read(&hv->num_mismatched_vp_indexes))) {
1469 		/* for all vcpus vp_index == vcpu_idx */
1470 		return (unsigned long *)vp_bitmap;
1471 	}
1472 
1473 	bitmap_zero(vcpu_bitmap, KVM_MAX_VCPUS);
1474 	kvm_for_each_vcpu(i, vcpu, kvm) {
1475 		if (test_bit(vcpu_to_hv_vcpu(vcpu)->vp_index,
1476 			     (unsigned long *)vp_bitmap))
1477 			__set_bit(i, vcpu_bitmap);
1478 	}
1479 	return vcpu_bitmap;
1480 }
1481 
1482 static u64 kvm_hv_flush_tlb(struct kvm_vcpu *current_vcpu, u64 ingpa,
1483 			    u16 rep_cnt, bool ex)
1484 {
1485 	struct kvm *kvm = current_vcpu->kvm;
1486 	struct kvm_vcpu_hv *hv_vcpu = &current_vcpu->arch.hyperv;
1487 	struct hv_tlb_flush_ex flush_ex;
1488 	struct hv_tlb_flush flush;
1489 	u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1490 	DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1491 	unsigned long *vcpu_mask;
1492 	u64 valid_bank_mask;
1493 	u64 sparse_banks[64];
1494 	int sparse_banks_len;
1495 	bool all_cpus;
1496 
1497 	if (!ex) {
1498 		if (unlikely(kvm_read_guest(kvm, ingpa, &flush, sizeof(flush))))
1499 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1500 
1501 		trace_kvm_hv_flush_tlb(flush.processor_mask,
1502 				       flush.address_space, flush.flags);
1503 
1504 		valid_bank_mask = BIT_ULL(0);
1505 		sparse_banks[0] = flush.processor_mask;
1506 
1507 		/*
1508 		 * Work around possible WS2012 bug: it sends hypercalls
1509 		 * with processor_mask = 0x0 and HV_FLUSH_ALL_PROCESSORS clear,
1510 		 * while also expecting us to flush something and crashing if
1511 		 * we don't. Let's treat processor_mask == 0 same as
1512 		 * HV_FLUSH_ALL_PROCESSORS.
1513 		 */
1514 		all_cpus = (flush.flags & HV_FLUSH_ALL_PROCESSORS) ||
1515 			flush.processor_mask == 0;
1516 	} else {
1517 		if (unlikely(kvm_read_guest(kvm, ingpa, &flush_ex,
1518 					    sizeof(flush_ex))))
1519 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1520 
1521 		trace_kvm_hv_flush_tlb_ex(flush_ex.hv_vp_set.valid_bank_mask,
1522 					  flush_ex.hv_vp_set.format,
1523 					  flush_ex.address_space,
1524 					  flush_ex.flags);
1525 
1526 		valid_bank_mask = flush_ex.hv_vp_set.valid_bank_mask;
1527 		all_cpus = flush_ex.hv_vp_set.format !=
1528 			HV_GENERIC_SET_SPARSE_4K;
1529 
1530 		sparse_banks_len =
1531 			bitmap_weight((unsigned long *)&valid_bank_mask, 64) *
1532 			sizeof(sparse_banks[0]);
1533 
1534 		if (!sparse_banks_len && !all_cpus)
1535 			goto ret_success;
1536 
1537 		if (!all_cpus &&
1538 		    kvm_read_guest(kvm,
1539 				   ingpa + offsetof(struct hv_tlb_flush_ex,
1540 						    hv_vp_set.bank_contents),
1541 				   sparse_banks,
1542 				   sparse_banks_len))
1543 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1544 	}
1545 
1546 	cpumask_clear(&hv_vcpu->tlb_flush);
1547 
1548 	vcpu_mask = all_cpus ? NULL :
1549 		sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1550 					vp_bitmap, vcpu_bitmap);
1551 
1552 	/*
1553 	 * vcpu->arch.cr3 may not be up-to-date for running vCPUs so we can't
1554 	 * analyze it here, flush TLB regardless of the specified address space.
1555 	 */
1556 	kvm_make_vcpus_request_mask(kvm, KVM_REQ_HV_TLB_FLUSH,
1557 				    NULL, vcpu_mask, &hv_vcpu->tlb_flush);
1558 
1559 ret_success:
1560 	/* We always do full TLB flush, set rep_done = rep_cnt. */
1561 	return (u64)HV_STATUS_SUCCESS |
1562 		((u64)rep_cnt << HV_HYPERCALL_REP_COMP_OFFSET);
1563 }
1564 
1565 static void kvm_send_ipi_to_many(struct kvm *kvm, u32 vector,
1566 				 unsigned long *vcpu_bitmap)
1567 {
1568 	struct kvm_lapic_irq irq = {
1569 		.delivery_mode = APIC_DM_FIXED,
1570 		.vector = vector
1571 	};
1572 	struct kvm_vcpu *vcpu;
1573 	int i;
1574 
1575 	kvm_for_each_vcpu(i, vcpu, kvm) {
1576 		if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
1577 			continue;
1578 
1579 		/* We fail only when APIC is disabled */
1580 		kvm_apic_set_irq(vcpu, &irq, NULL);
1581 	}
1582 }
1583 
1584 static u64 kvm_hv_send_ipi(struct kvm_vcpu *current_vcpu, u64 ingpa, u64 outgpa,
1585 			   bool ex, bool fast)
1586 {
1587 	struct kvm *kvm = current_vcpu->kvm;
1588 	struct hv_send_ipi_ex send_ipi_ex;
1589 	struct hv_send_ipi send_ipi;
1590 	u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1591 	DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1592 	unsigned long *vcpu_mask;
1593 	unsigned long valid_bank_mask;
1594 	u64 sparse_banks[64];
1595 	int sparse_banks_len;
1596 	u32 vector;
1597 	bool all_cpus;
1598 
1599 	if (!ex) {
1600 		if (!fast) {
1601 			if (unlikely(kvm_read_guest(kvm, ingpa, &send_ipi,
1602 						    sizeof(send_ipi))))
1603 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
1604 			sparse_banks[0] = send_ipi.cpu_mask;
1605 			vector = send_ipi.vector;
1606 		} else {
1607 			/* 'reserved' part of hv_send_ipi should be 0 */
1608 			if (unlikely(ingpa >> 32 != 0))
1609 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
1610 			sparse_banks[0] = outgpa;
1611 			vector = (u32)ingpa;
1612 		}
1613 		all_cpus = false;
1614 		valid_bank_mask = BIT_ULL(0);
1615 
1616 		trace_kvm_hv_send_ipi(vector, sparse_banks[0]);
1617 	} else {
1618 		if (unlikely(kvm_read_guest(kvm, ingpa, &send_ipi_ex,
1619 					    sizeof(send_ipi_ex))))
1620 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1621 
1622 		trace_kvm_hv_send_ipi_ex(send_ipi_ex.vector,
1623 					 send_ipi_ex.vp_set.format,
1624 					 send_ipi_ex.vp_set.valid_bank_mask);
1625 
1626 		vector = send_ipi_ex.vector;
1627 		valid_bank_mask = send_ipi_ex.vp_set.valid_bank_mask;
1628 		sparse_banks_len = bitmap_weight(&valid_bank_mask, 64) *
1629 			sizeof(sparse_banks[0]);
1630 
1631 		all_cpus = send_ipi_ex.vp_set.format == HV_GENERIC_SET_ALL;
1632 
1633 		if (!sparse_banks_len)
1634 			goto ret_success;
1635 
1636 		if (!all_cpus &&
1637 		    kvm_read_guest(kvm,
1638 				   ingpa + offsetof(struct hv_send_ipi_ex,
1639 						    vp_set.bank_contents),
1640 				   sparse_banks,
1641 				   sparse_banks_len))
1642 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1643 	}
1644 
1645 	if ((vector < HV_IPI_LOW_VECTOR) || (vector > HV_IPI_HIGH_VECTOR))
1646 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
1647 
1648 	vcpu_mask = all_cpus ? NULL :
1649 		sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1650 					vp_bitmap, vcpu_bitmap);
1651 
1652 	kvm_send_ipi_to_many(kvm, vector, vcpu_mask);
1653 
1654 ret_success:
1655 	return HV_STATUS_SUCCESS;
1656 }
1657 
1658 bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1659 {
1660 	return READ_ONCE(kvm->arch.hyperv.hv_guest_os_id) != 0;
1661 }
1662 
1663 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
1664 {
1665 	bool longmode;
1666 
1667 	longmode = is_64_bit_mode(vcpu);
1668 	if (longmode)
1669 		kvm_rax_write(vcpu, result);
1670 	else {
1671 		kvm_rdx_write(vcpu, result >> 32);
1672 		kvm_rax_write(vcpu, result & 0xffffffff);
1673 	}
1674 }
1675 
1676 static int kvm_hv_hypercall_complete(struct kvm_vcpu *vcpu, u64 result)
1677 {
1678 	kvm_hv_hypercall_set_result(vcpu, result);
1679 	++vcpu->stat.hypercalls;
1680 	return kvm_skip_emulated_instruction(vcpu);
1681 }
1682 
1683 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
1684 {
1685 	return kvm_hv_hypercall_complete(vcpu, vcpu->run->hyperv.u.hcall.result);
1686 }
1687 
1688 static u16 kvm_hvcall_signal_event(struct kvm_vcpu *vcpu, bool fast, u64 param)
1689 {
1690 	struct eventfd_ctx *eventfd;
1691 
1692 	if (unlikely(!fast)) {
1693 		int ret;
1694 		gpa_t gpa = param;
1695 
1696 		if ((gpa & (__alignof__(param) - 1)) ||
1697 		    offset_in_page(gpa) + sizeof(param) > PAGE_SIZE)
1698 			return HV_STATUS_INVALID_ALIGNMENT;
1699 
1700 		ret = kvm_vcpu_read_guest(vcpu, gpa, &param, sizeof(param));
1701 		if (ret < 0)
1702 			return HV_STATUS_INVALID_ALIGNMENT;
1703 	}
1704 
1705 	/*
1706 	 * Per spec, bits 32-47 contain the extra "flag number".  However, we
1707 	 * have no use for it, and in all known usecases it is zero, so just
1708 	 * report lookup failure if it isn't.
1709 	 */
1710 	if (param & 0xffff00000000ULL)
1711 		return HV_STATUS_INVALID_PORT_ID;
1712 	/* remaining bits are reserved-zero */
1713 	if (param & ~KVM_HYPERV_CONN_ID_MASK)
1714 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
1715 
1716 	/* the eventfd is protected by vcpu->kvm->srcu, but conn_to_evt isn't */
1717 	rcu_read_lock();
1718 	eventfd = idr_find(&vcpu->kvm->arch.hyperv.conn_to_evt, param);
1719 	rcu_read_unlock();
1720 	if (!eventfd)
1721 		return HV_STATUS_INVALID_PORT_ID;
1722 
1723 	eventfd_signal(eventfd, 1);
1724 	return HV_STATUS_SUCCESS;
1725 }
1726 
1727 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
1728 {
1729 	u64 param, ingpa, outgpa, ret = HV_STATUS_SUCCESS;
1730 	uint16_t code, rep_idx, rep_cnt;
1731 	bool fast, rep;
1732 
1733 	/*
1734 	 * hypercall generates UD from non zero cpl and real mode
1735 	 * per HYPER-V spec
1736 	 */
1737 	if (kvm_x86_ops.get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
1738 		kvm_queue_exception(vcpu, UD_VECTOR);
1739 		return 1;
1740 	}
1741 
1742 #ifdef CONFIG_X86_64
1743 	if (is_64_bit_mode(vcpu)) {
1744 		param = kvm_rcx_read(vcpu);
1745 		ingpa = kvm_rdx_read(vcpu);
1746 		outgpa = kvm_r8_read(vcpu);
1747 	} else
1748 #endif
1749 	{
1750 		param = ((u64)kvm_rdx_read(vcpu) << 32) |
1751 			(kvm_rax_read(vcpu) & 0xffffffff);
1752 		ingpa = ((u64)kvm_rbx_read(vcpu) << 32) |
1753 			(kvm_rcx_read(vcpu) & 0xffffffff);
1754 		outgpa = ((u64)kvm_rdi_read(vcpu) << 32) |
1755 			(kvm_rsi_read(vcpu) & 0xffffffff);
1756 	}
1757 
1758 	code = param & 0xffff;
1759 	fast = !!(param & HV_HYPERCALL_FAST_BIT);
1760 	rep_cnt = (param >> HV_HYPERCALL_REP_COMP_OFFSET) & 0xfff;
1761 	rep_idx = (param >> HV_HYPERCALL_REP_START_OFFSET) & 0xfff;
1762 	rep = !!(rep_cnt || rep_idx);
1763 
1764 	trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
1765 
1766 	switch (code) {
1767 	case HVCALL_NOTIFY_LONG_SPIN_WAIT:
1768 		if (unlikely(rep)) {
1769 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1770 			break;
1771 		}
1772 		kvm_vcpu_on_spin(vcpu, true);
1773 		break;
1774 	case HVCALL_SIGNAL_EVENT:
1775 		if (unlikely(rep)) {
1776 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1777 			break;
1778 		}
1779 		ret = kvm_hvcall_signal_event(vcpu, fast, ingpa);
1780 		if (ret != HV_STATUS_INVALID_PORT_ID)
1781 			break;
1782 		fallthrough;	/* maybe userspace knows this conn_id */
1783 	case HVCALL_POST_MESSAGE:
1784 		/* don't bother userspace if it has no way to handle it */
1785 		if (unlikely(rep || !vcpu_to_synic(vcpu)->active)) {
1786 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1787 			break;
1788 		}
1789 		vcpu->run->exit_reason = KVM_EXIT_HYPERV;
1790 		vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
1791 		vcpu->run->hyperv.u.hcall.input = param;
1792 		vcpu->run->hyperv.u.hcall.params[0] = ingpa;
1793 		vcpu->run->hyperv.u.hcall.params[1] = outgpa;
1794 		vcpu->arch.complete_userspace_io =
1795 				kvm_hv_hypercall_complete_userspace;
1796 		return 0;
1797 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
1798 		if (unlikely(fast || !rep_cnt || rep_idx)) {
1799 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1800 			break;
1801 		}
1802 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false);
1803 		break;
1804 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
1805 		if (unlikely(fast || rep)) {
1806 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1807 			break;
1808 		}
1809 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false);
1810 		break;
1811 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
1812 		if (unlikely(fast || !rep_cnt || rep_idx)) {
1813 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1814 			break;
1815 		}
1816 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
1817 		break;
1818 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
1819 		if (unlikely(fast || rep)) {
1820 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1821 			break;
1822 		}
1823 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
1824 		break;
1825 	case HVCALL_SEND_IPI:
1826 		if (unlikely(rep)) {
1827 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1828 			break;
1829 		}
1830 		ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, false, fast);
1831 		break;
1832 	case HVCALL_SEND_IPI_EX:
1833 		if (unlikely(fast || rep)) {
1834 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1835 			break;
1836 		}
1837 		ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, true, false);
1838 		break;
1839 	case HVCALL_POST_DEBUG_DATA:
1840 	case HVCALL_RETRIEVE_DEBUG_DATA:
1841 		if (unlikely(fast)) {
1842 			ret = HV_STATUS_INVALID_PARAMETER;
1843 			break;
1844 		}
1845 		fallthrough;
1846 	case HVCALL_RESET_DEBUG_SESSION: {
1847 		struct kvm_hv_syndbg *syndbg = vcpu_to_hv_syndbg(vcpu);
1848 
1849 		if (!kvm_hv_is_syndbg_enabled(vcpu)) {
1850 			ret = HV_STATUS_INVALID_HYPERCALL_CODE;
1851 			break;
1852 		}
1853 
1854 		if (!(syndbg->options & HV_X64_SYNDBG_OPTION_USE_HCALLS)) {
1855 			ret = HV_STATUS_OPERATION_DENIED;
1856 			break;
1857 		}
1858 		vcpu->run->exit_reason = KVM_EXIT_HYPERV;
1859 		vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
1860 		vcpu->run->hyperv.u.hcall.input = param;
1861 		vcpu->run->hyperv.u.hcall.params[0] = ingpa;
1862 		vcpu->run->hyperv.u.hcall.params[1] = outgpa;
1863 		vcpu->arch.complete_userspace_io =
1864 				kvm_hv_hypercall_complete_userspace;
1865 		return 0;
1866 	}
1867 	default:
1868 		ret = HV_STATUS_INVALID_HYPERCALL_CODE;
1869 		break;
1870 	}
1871 
1872 	return kvm_hv_hypercall_complete(vcpu, ret);
1873 }
1874 
1875 void kvm_hv_init_vm(struct kvm *kvm)
1876 {
1877 	mutex_init(&kvm->arch.hyperv.hv_lock);
1878 	idr_init(&kvm->arch.hyperv.conn_to_evt);
1879 }
1880 
1881 void kvm_hv_destroy_vm(struct kvm *kvm)
1882 {
1883 	struct eventfd_ctx *eventfd;
1884 	int i;
1885 
1886 	idr_for_each_entry(&kvm->arch.hyperv.conn_to_evt, eventfd, i)
1887 		eventfd_ctx_put(eventfd);
1888 	idr_destroy(&kvm->arch.hyperv.conn_to_evt);
1889 }
1890 
1891 static int kvm_hv_eventfd_assign(struct kvm *kvm, u32 conn_id, int fd)
1892 {
1893 	struct kvm_hv *hv = &kvm->arch.hyperv;
1894 	struct eventfd_ctx *eventfd;
1895 	int ret;
1896 
1897 	eventfd = eventfd_ctx_fdget(fd);
1898 	if (IS_ERR(eventfd))
1899 		return PTR_ERR(eventfd);
1900 
1901 	mutex_lock(&hv->hv_lock);
1902 	ret = idr_alloc(&hv->conn_to_evt, eventfd, conn_id, conn_id + 1,
1903 			GFP_KERNEL_ACCOUNT);
1904 	mutex_unlock(&hv->hv_lock);
1905 
1906 	if (ret >= 0)
1907 		return 0;
1908 
1909 	if (ret == -ENOSPC)
1910 		ret = -EEXIST;
1911 	eventfd_ctx_put(eventfd);
1912 	return ret;
1913 }
1914 
1915 static int kvm_hv_eventfd_deassign(struct kvm *kvm, u32 conn_id)
1916 {
1917 	struct kvm_hv *hv = &kvm->arch.hyperv;
1918 	struct eventfd_ctx *eventfd;
1919 
1920 	mutex_lock(&hv->hv_lock);
1921 	eventfd = idr_remove(&hv->conn_to_evt, conn_id);
1922 	mutex_unlock(&hv->hv_lock);
1923 
1924 	if (!eventfd)
1925 		return -ENOENT;
1926 
1927 	synchronize_srcu(&kvm->srcu);
1928 	eventfd_ctx_put(eventfd);
1929 	return 0;
1930 }
1931 
1932 int kvm_vm_ioctl_hv_eventfd(struct kvm *kvm, struct kvm_hyperv_eventfd *args)
1933 {
1934 	if ((args->flags & ~KVM_HYPERV_EVENTFD_DEASSIGN) ||
1935 	    (args->conn_id & ~KVM_HYPERV_CONN_ID_MASK))
1936 		return -EINVAL;
1937 
1938 	if (args->flags == KVM_HYPERV_EVENTFD_DEASSIGN)
1939 		return kvm_hv_eventfd_deassign(kvm, args->conn_id);
1940 	return kvm_hv_eventfd_assign(kvm, args->conn_id, args->fd);
1941 }
1942 
1943 int kvm_vcpu_ioctl_get_hv_cpuid(struct kvm_vcpu *vcpu, struct kvm_cpuid2 *cpuid,
1944 				struct kvm_cpuid_entry2 __user *entries)
1945 {
1946 	uint16_t evmcs_ver = 0;
1947 	struct kvm_cpuid_entry2 cpuid_entries[] = {
1948 		{ .function = HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS },
1949 		{ .function = HYPERV_CPUID_INTERFACE },
1950 		{ .function = HYPERV_CPUID_VERSION },
1951 		{ .function = HYPERV_CPUID_FEATURES },
1952 		{ .function = HYPERV_CPUID_ENLIGHTMENT_INFO },
1953 		{ .function = HYPERV_CPUID_IMPLEMENT_LIMITS },
1954 		{ .function = HYPERV_CPUID_SYNDBG_VENDOR_AND_MAX_FUNCTIONS },
1955 		{ .function = HYPERV_CPUID_SYNDBG_INTERFACE },
1956 		{ .function = HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES	},
1957 		{ .function = HYPERV_CPUID_NESTED_FEATURES },
1958 	};
1959 	int i, nent = ARRAY_SIZE(cpuid_entries);
1960 
1961 	if (kvm_x86_ops.nested_ops->get_evmcs_version)
1962 		evmcs_ver = kvm_x86_ops.nested_ops->get_evmcs_version(vcpu);
1963 
1964 	/* Skip NESTED_FEATURES if eVMCS is not supported */
1965 	if (!evmcs_ver)
1966 		--nent;
1967 
1968 	if (cpuid->nent < nent)
1969 		return -E2BIG;
1970 
1971 	if (cpuid->nent > nent)
1972 		cpuid->nent = nent;
1973 
1974 	for (i = 0; i < nent; i++) {
1975 		struct kvm_cpuid_entry2 *ent = &cpuid_entries[i];
1976 		u32 signature[3];
1977 
1978 		switch (ent->function) {
1979 		case HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS:
1980 			memcpy(signature, "Linux KVM Hv", 12);
1981 
1982 			ent->eax = HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES;
1983 			ent->ebx = signature[0];
1984 			ent->ecx = signature[1];
1985 			ent->edx = signature[2];
1986 			break;
1987 
1988 		case HYPERV_CPUID_INTERFACE:
1989 			memcpy(signature, "Hv#1\0\0\0\0\0\0\0\0", 12);
1990 			ent->eax = signature[0];
1991 			break;
1992 
1993 		case HYPERV_CPUID_VERSION:
1994 			/*
1995 			 * We implement some Hyper-V 2016 functions so let's use
1996 			 * this version.
1997 			 */
1998 			ent->eax = 0x00003839;
1999 			ent->ebx = 0x000A0000;
2000 			break;
2001 
2002 		case HYPERV_CPUID_FEATURES:
2003 			ent->eax |= HV_X64_MSR_VP_RUNTIME_AVAILABLE;
2004 			ent->eax |= HV_MSR_TIME_REF_COUNT_AVAILABLE;
2005 			ent->eax |= HV_X64_MSR_SYNIC_AVAILABLE;
2006 			ent->eax |= HV_MSR_SYNTIMER_AVAILABLE;
2007 			ent->eax |= HV_X64_MSR_APIC_ACCESS_AVAILABLE;
2008 			ent->eax |= HV_X64_MSR_HYPERCALL_AVAILABLE;
2009 			ent->eax |= HV_X64_MSR_VP_INDEX_AVAILABLE;
2010 			ent->eax |= HV_X64_MSR_RESET_AVAILABLE;
2011 			ent->eax |= HV_MSR_REFERENCE_TSC_AVAILABLE;
2012 			ent->eax |= HV_X64_ACCESS_FREQUENCY_MSRS;
2013 			ent->eax |= HV_X64_ACCESS_REENLIGHTENMENT;
2014 
2015 			ent->ebx |= HV_X64_POST_MESSAGES;
2016 			ent->ebx |= HV_X64_SIGNAL_EVENTS;
2017 
2018 			ent->edx |= HV_FEATURE_FREQUENCY_MSRS_AVAILABLE;
2019 			ent->edx |= HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
2020 
2021 			ent->ebx |= HV_DEBUGGING;
2022 			ent->edx |= HV_X64_GUEST_DEBUGGING_AVAILABLE;
2023 			ent->edx |= HV_FEATURE_DEBUG_MSRS_AVAILABLE;
2024 
2025 			/*
2026 			 * Direct Synthetic timers only make sense with in-kernel
2027 			 * LAPIC
2028 			 */
2029 			if (lapic_in_kernel(vcpu))
2030 				ent->edx |= HV_STIMER_DIRECT_MODE_AVAILABLE;
2031 
2032 			break;
2033 
2034 		case HYPERV_CPUID_ENLIGHTMENT_INFO:
2035 			ent->eax |= HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
2036 			ent->eax |= HV_X64_APIC_ACCESS_RECOMMENDED;
2037 			ent->eax |= HV_X64_RELAXED_TIMING_RECOMMENDED;
2038 			ent->eax |= HV_X64_CLUSTER_IPI_RECOMMENDED;
2039 			ent->eax |= HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED;
2040 			if (evmcs_ver)
2041 				ent->eax |= HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
2042 			if (!cpu_smt_possible())
2043 				ent->eax |= HV_X64_NO_NONARCH_CORESHARING;
2044 			/*
2045 			 * Default number of spinlock retry attempts, matches
2046 			 * HyperV 2016.
2047 			 */
2048 			ent->ebx = 0x00000FFF;
2049 
2050 			break;
2051 
2052 		case HYPERV_CPUID_IMPLEMENT_LIMITS:
2053 			/* Maximum number of virtual processors */
2054 			ent->eax = KVM_MAX_VCPUS;
2055 			/*
2056 			 * Maximum number of logical processors, matches
2057 			 * HyperV 2016.
2058 			 */
2059 			ent->ebx = 64;
2060 
2061 			break;
2062 
2063 		case HYPERV_CPUID_NESTED_FEATURES:
2064 			ent->eax = evmcs_ver;
2065 
2066 			break;
2067 
2068 		case HYPERV_CPUID_SYNDBG_VENDOR_AND_MAX_FUNCTIONS:
2069 			memcpy(signature, "Linux KVM Hv", 12);
2070 
2071 			ent->eax = 0;
2072 			ent->ebx = signature[0];
2073 			ent->ecx = signature[1];
2074 			ent->edx = signature[2];
2075 			break;
2076 
2077 		case HYPERV_CPUID_SYNDBG_INTERFACE:
2078 			memcpy(signature, "VS#1\0\0\0\0\0\0\0\0", 12);
2079 			ent->eax = signature[0];
2080 			break;
2081 
2082 		case HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES:
2083 			ent->eax |= HV_X64_SYNDBG_CAP_ALLOW_KERNEL_DEBUGGING;
2084 			break;
2085 
2086 		default:
2087 			break;
2088 		}
2089 	}
2090 
2091 	if (copy_to_user(entries, cpuid_entries,
2092 			 nent * sizeof(struct kvm_cpuid_entry2)))
2093 		return -EFAULT;
2094 
2095 	return 0;
2096 }
2097