xref: /openbmc/linux/arch/x86/kvm/hyperv.c (revision 84b102f5)
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 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
637 	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
638 
639 	if (!synic->active && !host)
640 		return 1;
641 
642 	trace_kvm_hv_stimer_set_config(stimer_to_vcpu(stimer)->vcpu_id,
643 				       stimer->index, config, host);
644 
645 	stimer_cleanup(stimer);
646 	if (old_config.enable &&
647 	    !new_config.direct_mode && new_config.sintx == 0)
648 		new_config.enable = 0;
649 	stimer->config.as_uint64 = new_config.as_uint64;
650 
651 	if (stimer->config.enable)
652 		stimer_mark_pending(stimer, false);
653 
654 	return 0;
655 }
656 
657 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
658 			    bool host)
659 {
660 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
661 	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
662 
663 	if (!synic->active && !host)
664 		return 1;
665 
666 	trace_kvm_hv_stimer_set_count(stimer_to_vcpu(stimer)->vcpu_id,
667 				      stimer->index, count, host);
668 
669 	stimer_cleanup(stimer);
670 	stimer->count = count;
671 	if (stimer->count == 0)
672 		stimer->config.enable = 0;
673 	else if (stimer->config.auto_enable)
674 		stimer->config.enable = 1;
675 
676 	if (stimer->config.enable)
677 		stimer_mark_pending(stimer, false);
678 
679 	return 0;
680 }
681 
682 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
683 {
684 	*pconfig = stimer->config.as_uint64;
685 	return 0;
686 }
687 
688 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
689 {
690 	*pcount = stimer->count;
691 	return 0;
692 }
693 
694 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
695 			     struct hv_message *src_msg, bool no_retry)
696 {
697 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
698 	int msg_off = offsetof(struct hv_message_page, sint_message[sint]);
699 	gfn_t msg_page_gfn;
700 	struct hv_message_header hv_hdr;
701 	int r;
702 
703 	if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
704 		return -ENOENT;
705 
706 	msg_page_gfn = synic->msg_page >> PAGE_SHIFT;
707 
708 	/*
709 	 * Strictly following the spec-mandated ordering would assume setting
710 	 * .msg_pending before checking .message_type.  However, this function
711 	 * is only called in vcpu context so the entire update is atomic from
712 	 * guest POV and thus the exact order here doesn't matter.
713 	 */
714 	r = kvm_vcpu_read_guest_page(vcpu, msg_page_gfn, &hv_hdr.message_type,
715 				     msg_off + offsetof(struct hv_message,
716 							header.message_type),
717 				     sizeof(hv_hdr.message_type));
718 	if (r < 0)
719 		return r;
720 
721 	if (hv_hdr.message_type != HVMSG_NONE) {
722 		if (no_retry)
723 			return 0;
724 
725 		hv_hdr.message_flags.msg_pending = 1;
726 		r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn,
727 					      &hv_hdr.message_flags,
728 					      msg_off +
729 					      offsetof(struct hv_message,
730 						       header.message_flags),
731 					      sizeof(hv_hdr.message_flags));
732 		if (r < 0)
733 			return r;
734 		return -EAGAIN;
735 	}
736 
737 	r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn, src_msg, msg_off,
738 				      sizeof(src_msg->header) +
739 				      src_msg->header.payload_size);
740 	if (r < 0)
741 		return r;
742 
743 	r = synic_set_irq(synic, sint);
744 	if (r < 0)
745 		return r;
746 	if (r == 0)
747 		return -EFAULT;
748 	return 0;
749 }
750 
751 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
752 {
753 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
754 	struct hv_message *msg = &stimer->msg;
755 	struct hv_timer_message_payload *payload =
756 			(struct hv_timer_message_payload *)&msg->u.payload;
757 
758 	/*
759 	 * To avoid piling up periodic ticks, don't retry message
760 	 * delivery for them (within "lazy" lost ticks policy).
761 	 */
762 	bool no_retry = stimer->config.periodic;
763 
764 	payload->expiration_time = stimer->exp_time;
765 	payload->delivery_time = get_time_ref_counter(vcpu->kvm);
766 	return synic_deliver_msg(vcpu_to_synic(vcpu),
767 				 stimer->config.sintx, msg,
768 				 no_retry);
769 }
770 
771 static int stimer_notify_direct(struct kvm_vcpu_hv_stimer *stimer)
772 {
773 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
774 	struct kvm_lapic_irq irq = {
775 		.delivery_mode = APIC_DM_FIXED,
776 		.vector = stimer->config.apic_vector
777 	};
778 
779 	if (lapic_in_kernel(vcpu))
780 		return !kvm_apic_set_irq(vcpu, &irq, NULL);
781 	return 0;
782 }
783 
784 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
785 {
786 	int r, direct = stimer->config.direct_mode;
787 
788 	stimer->msg_pending = true;
789 	if (!direct)
790 		r = stimer_send_msg(stimer);
791 	else
792 		r = stimer_notify_direct(stimer);
793 	trace_kvm_hv_stimer_expiration(stimer_to_vcpu(stimer)->vcpu_id,
794 				       stimer->index, direct, r);
795 	if (!r) {
796 		stimer->msg_pending = false;
797 		if (!(stimer->config.periodic))
798 			stimer->config.enable = 0;
799 	}
800 }
801 
802 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
803 {
804 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
805 	struct kvm_vcpu_hv_stimer *stimer;
806 	u64 time_now, exp_time;
807 	int i;
808 
809 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
810 		if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
811 			stimer = &hv_vcpu->stimer[i];
812 			if (stimer->config.enable) {
813 				exp_time = stimer->exp_time;
814 
815 				if (exp_time) {
816 					time_now =
817 						get_time_ref_counter(vcpu->kvm);
818 					if (time_now >= exp_time)
819 						stimer_expiration(stimer);
820 				}
821 
822 				if ((stimer->config.enable) &&
823 				    stimer->count) {
824 					if (!stimer->msg_pending)
825 						stimer_start(stimer);
826 				} else
827 					stimer_cleanup(stimer);
828 			}
829 		}
830 }
831 
832 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
833 {
834 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
835 	int i;
836 
837 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
838 		stimer_cleanup(&hv_vcpu->stimer[i]);
839 }
840 
841 bool kvm_hv_assist_page_enabled(struct kvm_vcpu *vcpu)
842 {
843 	if (!(vcpu->arch.hyperv.hv_vapic & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE))
844 		return false;
845 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
846 }
847 EXPORT_SYMBOL_GPL(kvm_hv_assist_page_enabled);
848 
849 bool kvm_hv_get_assist_page(struct kvm_vcpu *vcpu,
850 			    struct hv_vp_assist_page *assist_page)
851 {
852 	if (!kvm_hv_assist_page_enabled(vcpu))
853 		return false;
854 	return !kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data,
855 				      assist_page, sizeof(*assist_page));
856 }
857 EXPORT_SYMBOL_GPL(kvm_hv_get_assist_page);
858 
859 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
860 {
861 	struct hv_message *msg = &stimer->msg;
862 	struct hv_timer_message_payload *payload =
863 			(struct hv_timer_message_payload *)&msg->u.payload;
864 
865 	memset(&msg->header, 0, sizeof(msg->header));
866 	msg->header.message_type = HVMSG_TIMER_EXPIRED;
867 	msg->header.payload_size = sizeof(*payload);
868 
869 	payload->timer_index = stimer->index;
870 	payload->expiration_time = 0;
871 	payload->delivery_time = 0;
872 }
873 
874 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
875 {
876 	memset(stimer, 0, sizeof(*stimer));
877 	stimer->index = timer_index;
878 	hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
879 	stimer->timer.function = stimer_timer_callback;
880 	stimer_prepare_msg(stimer);
881 }
882 
883 void kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
884 {
885 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
886 	int i;
887 
888 	synic_init(&hv_vcpu->synic);
889 
890 	bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
891 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
892 		stimer_init(&hv_vcpu->stimer[i], i);
893 }
894 
895 void kvm_hv_vcpu_postcreate(struct kvm_vcpu *vcpu)
896 {
897 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
898 
899 	hv_vcpu->vp_index = kvm_vcpu_get_idx(vcpu);
900 }
901 
902 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
903 {
904 	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
905 
906 	/*
907 	 * Hyper-V SynIC auto EOI SINT's are
908 	 * not compatible with APICV, so request
909 	 * to deactivate APICV permanently.
910 	 */
911 	kvm_request_apicv_update(vcpu->kvm, false, APICV_INHIBIT_REASON_HYPERV);
912 	synic->active = true;
913 	synic->dont_zero_synic_pages = dont_zero_synic_pages;
914 	synic->control = HV_SYNIC_CONTROL_ENABLE;
915 	return 0;
916 }
917 
918 static bool kvm_hv_msr_partition_wide(u32 msr)
919 {
920 	bool r = false;
921 
922 	switch (msr) {
923 	case HV_X64_MSR_GUEST_OS_ID:
924 	case HV_X64_MSR_HYPERCALL:
925 	case HV_X64_MSR_REFERENCE_TSC:
926 	case HV_X64_MSR_TIME_REF_COUNT:
927 	case HV_X64_MSR_CRASH_CTL:
928 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
929 	case HV_X64_MSR_RESET:
930 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
931 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
932 	case HV_X64_MSR_TSC_EMULATION_STATUS:
933 	case HV_X64_MSR_SYNDBG_OPTIONS:
934 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
935 		r = true;
936 		break;
937 	}
938 
939 	return r;
940 }
941 
942 static int kvm_hv_msr_get_crash_data(struct kvm_vcpu *vcpu,
943 				     u32 index, u64 *pdata)
944 {
945 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
946 	size_t size = ARRAY_SIZE(hv->hv_crash_param);
947 
948 	if (WARN_ON_ONCE(index >= size))
949 		return -EINVAL;
950 
951 	*pdata = hv->hv_crash_param[array_index_nospec(index, size)];
952 	return 0;
953 }
954 
955 static int kvm_hv_msr_get_crash_ctl(struct kvm_vcpu *vcpu, u64 *pdata)
956 {
957 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
958 
959 	*pdata = hv->hv_crash_ctl;
960 	return 0;
961 }
962 
963 static int kvm_hv_msr_set_crash_ctl(struct kvm_vcpu *vcpu, u64 data, bool host)
964 {
965 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
966 
967 	if (host)
968 		hv->hv_crash_ctl = data & HV_CRASH_CTL_CRASH_NOTIFY;
969 
970 	if (!host && (data & HV_CRASH_CTL_CRASH_NOTIFY)) {
971 
972 		vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
973 			  hv->hv_crash_param[0],
974 			  hv->hv_crash_param[1],
975 			  hv->hv_crash_param[2],
976 			  hv->hv_crash_param[3],
977 			  hv->hv_crash_param[4]);
978 
979 		/* Send notification about crash to user space */
980 		kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
981 	}
982 
983 	return 0;
984 }
985 
986 static int kvm_hv_msr_set_crash_data(struct kvm_vcpu *vcpu,
987 				     u32 index, u64 data)
988 {
989 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
990 	size_t size = ARRAY_SIZE(hv->hv_crash_param);
991 
992 	if (WARN_ON_ONCE(index >= size))
993 		return -EINVAL;
994 
995 	hv->hv_crash_param[array_index_nospec(index, size)] = data;
996 	return 0;
997 }
998 
999 /*
1000  * The kvmclock and Hyper-V TSC page use similar formulas, and converting
1001  * between them is possible:
1002  *
1003  * kvmclock formula:
1004  *    nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
1005  *           + system_time
1006  *
1007  * Hyper-V formula:
1008  *    nsec/100 = ticks * scale / 2^64 + offset
1009  *
1010  * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
1011  * By dividing the kvmclock formula by 100 and equating what's left we get:
1012  *    ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1013  *            scale / 2^64 =         tsc_to_system_mul * 2^(tsc_shift-32) / 100
1014  *            scale        =         tsc_to_system_mul * 2^(32+tsc_shift) / 100
1015  *
1016  * Now expand the kvmclock formula and divide by 100:
1017  *    nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
1018  *           - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
1019  *           + system_time
1020  *    nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1021  *               - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1022  *               + system_time / 100
1023  *
1024  * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
1025  *    nsec/100 = ticks * scale / 2^64
1026  *               - tsc_timestamp * scale / 2^64
1027  *               + system_time / 100
1028  *
1029  * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
1030  *    offset = system_time / 100 - tsc_timestamp * scale / 2^64
1031  *
1032  * These two equivalencies are implemented in this function.
1033  */
1034 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
1035 					struct ms_hyperv_tsc_page *tsc_ref)
1036 {
1037 	u64 max_mul;
1038 
1039 	if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
1040 		return false;
1041 
1042 	/*
1043 	 * check if scale would overflow, if so we use the time ref counter
1044 	 *    tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
1045 	 *    tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
1046 	 *    tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
1047 	 */
1048 	max_mul = 100ull << (32 - hv_clock->tsc_shift);
1049 	if (hv_clock->tsc_to_system_mul >= max_mul)
1050 		return false;
1051 
1052 	/*
1053 	 * Otherwise compute the scale and offset according to the formulas
1054 	 * derived above.
1055 	 */
1056 	tsc_ref->tsc_scale =
1057 		mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
1058 				hv_clock->tsc_to_system_mul,
1059 				100);
1060 
1061 	tsc_ref->tsc_offset = hv_clock->system_time;
1062 	do_div(tsc_ref->tsc_offset, 100);
1063 	tsc_ref->tsc_offset -=
1064 		mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
1065 	return true;
1066 }
1067 
1068 void kvm_hv_setup_tsc_page(struct kvm *kvm,
1069 			   struct pvclock_vcpu_time_info *hv_clock)
1070 {
1071 	struct kvm_hv *hv = &kvm->arch.hyperv;
1072 	u32 tsc_seq;
1073 	u64 gfn;
1074 
1075 	BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
1076 	BUILD_BUG_ON(offsetof(struct ms_hyperv_tsc_page, tsc_sequence) != 0);
1077 
1078 	if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1079 		return;
1080 
1081 	mutex_lock(&kvm->arch.hyperv.hv_lock);
1082 	if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1083 		goto out_unlock;
1084 
1085 	gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
1086 	/*
1087 	 * Because the TSC parameters only vary when there is a
1088 	 * change in the master clock, do not bother with caching.
1089 	 */
1090 	if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
1091 				    &tsc_seq, sizeof(tsc_seq))))
1092 		goto out_unlock;
1093 
1094 	/*
1095 	 * While we're computing and writing the parameters, force the
1096 	 * guest to use the time reference count MSR.
1097 	 */
1098 	hv->tsc_ref.tsc_sequence = 0;
1099 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
1100 			    &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
1101 		goto out_unlock;
1102 
1103 	if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
1104 		goto out_unlock;
1105 
1106 	/* Ensure sequence is zero before writing the rest of the struct.  */
1107 	smp_wmb();
1108 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
1109 		goto out_unlock;
1110 
1111 	/*
1112 	 * Now switch to the TSC page mechanism by writing the sequence.
1113 	 */
1114 	tsc_seq++;
1115 	if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
1116 		tsc_seq = 1;
1117 
1118 	/* Write the struct entirely before the non-zero sequence.  */
1119 	smp_wmb();
1120 
1121 	hv->tsc_ref.tsc_sequence = tsc_seq;
1122 	kvm_write_guest(kvm, gfn_to_gpa(gfn),
1123 			&hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence));
1124 out_unlock:
1125 	mutex_unlock(&kvm->arch.hyperv.hv_lock);
1126 }
1127 
1128 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
1129 			     bool host)
1130 {
1131 	struct kvm *kvm = vcpu->kvm;
1132 	struct kvm_hv *hv = &kvm->arch.hyperv;
1133 
1134 	switch (msr) {
1135 	case HV_X64_MSR_GUEST_OS_ID:
1136 		hv->hv_guest_os_id = data;
1137 		/* setting guest os id to zero disables hypercall page */
1138 		if (!hv->hv_guest_os_id)
1139 			hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1140 		break;
1141 	case HV_X64_MSR_HYPERCALL: {
1142 		u64 gfn;
1143 		unsigned long addr;
1144 		u8 instructions[4];
1145 
1146 		/* if guest os id is not set hypercall should remain disabled */
1147 		if (!hv->hv_guest_os_id)
1148 			break;
1149 		if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1150 			hv->hv_hypercall = data;
1151 			break;
1152 		}
1153 		gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1154 		addr = gfn_to_hva(kvm, gfn);
1155 		if (kvm_is_error_hva(addr))
1156 			return 1;
1157 		kvm_x86_ops.patch_hypercall(vcpu, instructions);
1158 		((unsigned char *)instructions)[3] = 0xc3; /* ret */
1159 		if (__copy_to_user((void __user *)addr, instructions, 4))
1160 			return 1;
1161 		hv->hv_hypercall = data;
1162 		mark_page_dirty(kvm, gfn);
1163 		break;
1164 	}
1165 	case HV_X64_MSR_REFERENCE_TSC:
1166 		hv->hv_tsc_page = data;
1167 		if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE)
1168 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1169 		break;
1170 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1171 		return kvm_hv_msr_set_crash_data(vcpu,
1172 						 msr - HV_X64_MSR_CRASH_P0,
1173 						 data);
1174 	case HV_X64_MSR_CRASH_CTL:
1175 		return kvm_hv_msr_set_crash_ctl(vcpu, data, host);
1176 	case HV_X64_MSR_RESET:
1177 		if (data == 1) {
1178 			vcpu_debug(vcpu, "hyper-v reset requested\n");
1179 			kvm_make_request(KVM_REQ_HV_RESET, vcpu);
1180 		}
1181 		break;
1182 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1183 		hv->hv_reenlightenment_control = data;
1184 		break;
1185 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1186 		hv->hv_tsc_emulation_control = data;
1187 		break;
1188 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1189 		hv->hv_tsc_emulation_status = data;
1190 		break;
1191 	case HV_X64_MSR_TIME_REF_COUNT:
1192 		/* read-only, but still ignore it if host-initiated */
1193 		if (!host)
1194 			return 1;
1195 		break;
1196 	case HV_X64_MSR_SYNDBG_OPTIONS:
1197 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1198 		return syndbg_set_msr(vcpu, msr, data, host);
1199 	default:
1200 		vcpu_unimpl(vcpu, "Hyper-V unhandled wrmsr: 0x%x data 0x%llx\n",
1201 			    msr, data);
1202 		return 1;
1203 	}
1204 	return 0;
1205 }
1206 
1207 /* Calculate cpu time spent by current task in 100ns units */
1208 static u64 current_task_runtime_100ns(void)
1209 {
1210 	u64 utime, stime;
1211 
1212 	task_cputime_adjusted(current, &utime, &stime);
1213 
1214 	return div_u64(utime + stime, 100);
1215 }
1216 
1217 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1218 {
1219 	struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
1220 
1221 	switch (msr) {
1222 	case HV_X64_MSR_VP_INDEX: {
1223 		struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
1224 		int vcpu_idx = kvm_vcpu_get_idx(vcpu);
1225 		u32 new_vp_index = (u32)data;
1226 
1227 		if (!host || new_vp_index >= KVM_MAX_VCPUS)
1228 			return 1;
1229 
1230 		if (new_vp_index == hv_vcpu->vp_index)
1231 			return 0;
1232 
1233 		/*
1234 		 * The VP index is initialized to vcpu_index by
1235 		 * kvm_hv_vcpu_postcreate so they initially match.  Now the
1236 		 * VP index is changing, adjust num_mismatched_vp_indexes if
1237 		 * it now matches or no longer matches vcpu_idx.
1238 		 */
1239 		if (hv_vcpu->vp_index == vcpu_idx)
1240 			atomic_inc(&hv->num_mismatched_vp_indexes);
1241 		else if (new_vp_index == vcpu_idx)
1242 			atomic_dec(&hv->num_mismatched_vp_indexes);
1243 
1244 		hv_vcpu->vp_index = new_vp_index;
1245 		break;
1246 	}
1247 	case HV_X64_MSR_VP_ASSIST_PAGE: {
1248 		u64 gfn;
1249 		unsigned long addr;
1250 
1251 		if (!(data & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE)) {
1252 			hv_vcpu->hv_vapic = data;
1253 			if (kvm_lapic_enable_pv_eoi(vcpu, 0, 0))
1254 				return 1;
1255 			break;
1256 		}
1257 		gfn = data >> HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT;
1258 		addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1259 		if (kvm_is_error_hva(addr))
1260 			return 1;
1261 
1262 		/*
1263 		 * Clear apic_assist portion of struct hv_vp_assist_page
1264 		 * only, there can be valuable data in the rest which needs
1265 		 * to be preserved e.g. on migration.
1266 		 */
1267 		if (__put_user(0, (u32 __user *)addr))
1268 			return 1;
1269 		hv_vcpu->hv_vapic = data;
1270 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
1271 		if (kvm_lapic_enable_pv_eoi(vcpu,
1272 					    gfn_to_gpa(gfn) | KVM_MSR_ENABLED,
1273 					    sizeof(struct hv_vp_assist_page)))
1274 			return 1;
1275 		break;
1276 	}
1277 	case HV_X64_MSR_EOI:
1278 		return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1279 	case HV_X64_MSR_ICR:
1280 		return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1281 	case HV_X64_MSR_TPR:
1282 		return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1283 	case HV_X64_MSR_VP_RUNTIME:
1284 		if (!host)
1285 			return 1;
1286 		hv_vcpu->runtime_offset = data - current_task_runtime_100ns();
1287 		break;
1288 	case HV_X64_MSR_SCONTROL:
1289 	case HV_X64_MSR_SVERSION:
1290 	case HV_X64_MSR_SIEFP:
1291 	case HV_X64_MSR_SIMP:
1292 	case HV_X64_MSR_EOM:
1293 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1294 		return synic_set_msr(vcpu_to_synic(vcpu), msr, data, host);
1295 	case HV_X64_MSR_STIMER0_CONFIG:
1296 	case HV_X64_MSR_STIMER1_CONFIG:
1297 	case HV_X64_MSR_STIMER2_CONFIG:
1298 	case HV_X64_MSR_STIMER3_CONFIG: {
1299 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1300 
1301 		return stimer_set_config(vcpu_to_stimer(vcpu, timer_index),
1302 					 data, host);
1303 	}
1304 	case HV_X64_MSR_STIMER0_COUNT:
1305 	case HV_X64_MSR_STIMER1_COUNT:
1306 	case HV_X64_MSR_STIMER2_COUNT:
1307 	case HV_X64_MSR_STIMER3_COUNT: {
1308 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1309 
1310 		return stimer_set_count(vcpu_to_stimer(vcpu, timer_index),
1311 					data, host);
1312 	}
1313 	case HV_X64_MSR_TSC_FREQUENCY:
1314 	case HV_X64_MSR_APIC_FREQUENCY:
1315 		/* read-only, but still ignore it if host-initiated */
1316 		if (!host)
1317 			return 1;
1318 		break;
1319 	default:
1320 		vcpu_unimpl(vcpu, "Hyper-V unhandled wrmsr: 0x%x data 0x%llx\n",
1321 			    msr, data);
1322 		return 1;
1323 	}
1324 
1325 	return 0;
1326 }
1327 
1328 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1329 			     bool host)
1330 {
1331 	u64 data = 0;
1332 	struct kvm *kvm = vcpu->kvm;
1333 	struct kvm_hv *hv = &kvm->arch.hyperv;
1334 
1335 	switch (msr) {
1336 	case HV_X64_MSR_GUEST_OS_ID:
1337 		data = hv->hv_guest_os_id;
1338 		break;
1339 	case HV_X64_MSR_HYPERCALL:
1340 		data = hv->hv_hypercall;
1341 		break;
1342 	case HV_X64_MSR_TIME_REF_COUNT:
1343 		data = get_time_ref_counter(kvm);
1344 		break;
1345 	case HV_X64_MSR_REFERENCE_TSC:
1346 		data = hv->hv_tsc_page;
1347 		break;
1348 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1349 		return kvm_hv_msr_get_crash_data(vcpu,
1350 						 msr - HV_X64_MSR_CRASH_P0,
1351 						 pdata);
1352 	case HV_X64_MSR_CRASH_CTL:
1353 		return kvm_hv_msr_get_crash_ctl(vcpu, pdata);
1354 	case HV_X64_MSR_RESET:
1355 		data = 0;
1356 		break;
1357 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1358 		data = hv->hv_reenlightenment_control;
1359 		break;
1360 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1361 		data = hv->hv_tsc_emulation_control;
1362 		break;
1363 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1364 		data = hv->hv_tsc_emulation_status;
1365 		break;
1366 	case HV_X64_MSR_SYNDBG_OPTIONS:
1367 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1368 		return syndbg_get_msr(vcpu, msr, pdata, host);
1369 	default:
1370 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1371 		return 1;
1372 	}
1373 
1374 	*pdata = data;
1375 	return 0;
1376 }
1377 
1378 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1379 			  bool host)
1380 {
1381 	u64 data = 0;
1382 	struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
1383 
1384 	switch (msr) {
1385 	case HV_X64_MSR_VP_INDEX:
1386 		data = hv_vcpu->vp_index;
1387 		break;
1388 	case HV_X64_MSR_EOI:
1389 		return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1390 	case HV_X64_MSR_ICR:
1391 		return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1392 	case HV_X64_MSR_TPR:
1393 		return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1394 	case HV_X64_MSR_VP_ASSIST_PAGE:
1395 		data = hv_vcpu->hv_vapic;
1396 		break;
1397 	case HV_X64_MSR_VP_RUNTIME:
1398 		data = current_task_runtime_100ns() + hv_vcpu->runtime_offset;
1399 		break;
1400 	case HV_X64_MSR_SCONTROL:
1401 	case HV_X64_MSR_SVERSION:
1402 	case HV_X64_MSR_SIEFP:
1403 	case HV_X64_MSR_SIMP:
1404 	case HV_X64_MSR_EOM:
1405 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1406 		return synic_get_msr(vcpu_to_synic(vcpu), msr, pdata, host);
1407 	case HV_X64_MSR_STIMER0_CONFIG:
1408 	case HV_X64_MSR_STIMER1_CONFIG:
1409 	case HV_X64_MSR_STIMER2_CONFIG:
1410 	case HV_X64_MSR_STIMER3_CONFIG: {
1411 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1412 
1413 		return stimer_get_config(vcpu_to_stimer(vcpu, timer_index),
1414 					 pdata);
1415 	}
1416 	case HV_X64_MSR_STIMER0_COUNT:
1417 	case HV_X64_MSR_STIMER1_COUNT:
1418 	case HV_X64_MSR_STIMER2_COUNT:
1419 	case HV_X64_MSR_STIMER3_COUNT: {
1420 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1421 
1422 		return stimer_get_count(vcpu_to_stimer(vcpu, timer_index),
1423 					pdata);
1424 	}
1425 	case HV_X64_MSR_TSC_FREQUENCY:
1426 		data = (u64)vcpu->arch.virtual_tsc_khz * 1000;
1427 		break;
1428 	case HV_X64_MSR_APIC_FREQUENCY:
1429 		data = APIC_BUS_FREQUENCY;
1430 		break;
1431 	default:
1432 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1433 		return 1;
1434 	}
1435 	*pdata = data;
1436 	return 0;
1437 }
1438 
1439 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1440 {
1441 	if (kvm_hv_msr_partition_wide(msr)) {
1442 		int r;
1443 
1444 		mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1445 		r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1446 		mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1447 		return r;
1448 	} else
1449 		return kvm_hv_set_msr(vcpu, msr, data, host);
1450 }
1451 
1452 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
1453 {
1454 	if (kvm_hv_msr_partition_wide(msr)) {
1455 		int r;
1456 
1457 		mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1458 		r = kvm_hv_get_msr_pw(vcpu, msr, pdata, host);
1459 		mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1460 		return r;
1461 	} else
1462 		return kvm_hv_get_msr(vcpu, msr, pdata, host);
1463 }
1464 
1465 static __always_inline unsigned long *sparse_set_to_vcpu_mask(
1466 	struct kvm *kvm, u64 *sparse_banks, u64 valid_bank_mask,
1467 	u64 *vp_bitmap, unsigned long *vcpu_bitmap)
1468 {
1469 	struct kvm_hv *hv = &kvm->arch.hyperv;
1470 	struct kvm_vcpu *vcpu;
1471 	int i, bank, sbank = 0;
1472 
1473 	memset(vp_bitmap, 0,
1474 	       KVM_HV_MAX_SPARSE_VCPU_SET_BITS * sizeof(*vp_bitmap));
1475 	for_each_set_bit(bank, (unsigned long *)&valid_bank_mask,
1476 			 KVM_HV_MAX_SPARSE_VCPU_SET_BITS)
1477 		vp_bitmap[bank] = sparse_banks[sbank++];
1478 
1479 	if (likely(!atomic_read(&hv->num_mismatched_vp_indexes))) {
1480 		/* for all vcpus vp_index == vcpu_idx */
1481 		return (unsigned long *)vp_bitmap;
1482 	}
1483 
1484 	bitmap_zero(vcpu_bitmap, KVM_MAX_VCPUS);
1485 	kvm_for_each_vcpu(i, vcpu, kvm) {
1486 		if (test_bit(vcpu_to_hv_vcpu(vcpu)->vp_index,
1487 			     (unsigned long *)vp_bitmap))
1488 			__set_bit(i, vcpu_bitmap);
1489 	}
1490 	return vcpu_bitmap;
1491 }
1492 
1493 static u64 kvm_hv_flush_tlb(struct kvm_vcpu *current_vcpu, u64 ingpa,
1494 			    u16 rep_cnt, bool ex)
1495 {
1496 	struct kvm *kvm = current_vcpu->kvm;
1497 	struct kvm_vcpu_hv *hv_vcpu = &current_vcpu->arch.hyperv;
1498 	struct hv_tlb_flush_ex flush_ex;
1499 	struct hv_tlb_flush flush;
1500 	u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1501 	DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1502 	unsigned long *vcpu_mask;
1503 	u64 valid_bank_mask;
1504 	u64 sparse_banks[64];
1505 	int sparse_banks_len;
1506 	bool all_cpus;
1507 
1508 	if (!ex) {
1509 		if (unlikely(kvm_read_guest(kvm, ingpa, &flush, sizeof(flush))))
1510 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1511 
1512 		trace_kvm_hv_flush_tlb(flush.processor_mask,
1513 				       flush.address_space, flush.flags);
1514 
1515 		valid_bank_mask = BIT_ULL(0);
1516 		sparse_banks[0] = flush.processor_mask;
1517 
1518 		/*
1519 		 * Work around possible WS2012 bug: it sends hypercalls
1520 		 * with processor_mask = 0x0 and HV_FLUSH_ALL_PROCESSORS clear,
1521 		 * while also expecting us to flush something and crashing if
1522 		 * we don't. Let's treat processor_mask == 0 same as
1523 		 * HV_FLUSH_ALL_PROCESSORS.
1524 		 */
1525 		all_cpus = (flush.flags & HV_FLUSH_ALL_PROCESSORS) ||
1526 			flush.processor_mask == 0;
1527 	} else {
1528 		if (unlikely(kvm_read_guest(kvm, ingpa, &flush_ex,
1529 					    sizeof(flush_ex))))
1530 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1531 
1532 		trace_kvm_hv_flush_tlb_ex(flush_ex.hv_vp_set.valid_bank_mask,
1533 					  flush_ex.hv_vp_set.format,
1534 					  flush_ex.address_space,
1535 					  flush_ex.flags);
1536 
1537 		valid_bank_mask = flush_ex.hv_vp_set.valid_bank_mask;
1538 		all_cpus = flush_ex.hv_vp_set.format !=
1539 			HV_GENERIC_SET_SPARSE_4K;
1540 
1541 		sparse_banks_len =
1542 			bitmap_weight((unsigned long *)&valid_bank_mask, 64) *
1543 			sizeof(sparse_banks[0]);
1544 
1545 		if (!sparse_banks_len && !all_cpus)
1546 			goto ret_success;
1547 
1548 		if (!all_cpus &&
1549 		    kvm_read_guest(kvm,
1550 				   ingpa + offsetof(struct hv_tlb_flush_ex,
1551 						    hv_vp_set.bank_contents),
1552 				   sparse_banks,
1553 				   sparse_banks_len))
1554 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1555 	}
1556 
1557 	cpumask_clear(&hv_vcpu->tlb_flush);
1558 
1559 	vcpu_mask = all_cpus ? NULL :
1560 		sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1561 					vp_bitmap, vcpu_bitmap);
1562 
1563 	/*
1564 	 * vcpu->arch.cr3 may not be up-to-date for running vCPUs so we can't
1565 	 * analyze it here, flush TLB regardless of the specified address space.
1566 	 */
1567 	kvm_make_vcpus_request_mask(kvm, KVM_REQ_HV_TLB_FLUSH,
1568 				    NULL, vcpu_mask, &hv_vcpu->tlb_flush);
1569 
1570 ret_success:
1571 	/* We always do full TLB flush, set rep_done = rep_cnt. */
1572 	return (u64)HV_STATUS_SUCCESS |
1573 		((u64)rep_cnt << HV_HYPERCALL_REP_COMP_OFFSET);
1574 }
1575 
1576 static void kvm_send_ipi_to_many(struct kvm *kvm, u32 vector,
1577 				 unsigned long *vcpu_bitmap)
1578 {
1579 	struct kvm_lapic_irq irq = {
1580 		.delivery_mode = APIC_DM_FIXED,
1581 		.vector = vector
1582 	};
1583 	struct kvm_vcpu *vcpu;
1584 	int i;
1585 
1586 	kvm_for_each_vcpu(i, vcpu, kvm) {
1587 		if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
1588 			continue;
1589 
1590 		/* We fail only when APIC is disabled */
1591 		kvm_apic_set_irq(vcpu, &irq, NULL);
1592 	}
1593 }
1594 
1595 static u64 kvm_hv_send_ipi(struct kvm_vcpu *current_vcpu, u64 ingpa, u64 outgpa,
1596 			   bool ex, bool fast)
1597 {
1598 	struct kvm *kvm = current_vcpu->kvm;
1599 	struct hv_send_ipi_ex send_ipi_ex;
1600 	struct hv_send_ipi send_ipi;
1601 	u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1602 	DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1603 	unsigned long *vcpu_mask;
1604 	unsigned long valid_bank_mask;
1605 	u64 sparse_banks[64];
1606 	int sparse_banks_len;
1607 	u32 vector;
1608 	bool all_cpus;
1609 
1610 	if (!ex) {
1611 		if (!fast) {
1612 			if (unlikely(kvm_read_guest(kvm, ingpa, &send_ipi,
1613 						    sizeof(send_ipi))))
1614 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
1615 			sparse_banks[0] = send_ipi.cpu_mask;
1616 			vector = send_ipi.vector;
1617 		} else {
1618 			/* 'reserved' part of hv_send_ipi should be 0 */
1619 			if (unlikely(ingpa >> 32 != 0))
1620 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
1621 			sparse_banks[0] = outgpa;
1622 			vector = (u32)ingpa;
1623 		}
1624 		all_cpus = false;
1625 		valid_bank_mask = BIT_ULL(0);
1626 
1627 		trace_kvm_hv_send_ipi(vector, sparse_banks[0]);
1628 	} else {
1629 		if (unlikely(kvm_read_guest(kvm, ingpa, &send_ipi_ex,
1630 					    sizeof(send_ipi_ex))))
1631 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1632 
1633 		trace_kvm_hv_send_ipi_ex(send_ipi_ex.vector,
1634 					 send_ipi_ex.vp_set.format,
1635 					 send_ipi_ex.vp_set.valid_bank_mask);
1636 
1637 		vector = send_ipi_ex.vector;
1638 		valid_bank_mask = send_ipi_ex.vp_set.valid_bank_mask;
1639 		sparse_banks_len = bitmap_weight(&valid_bank_mask, 64) *
1640 			sizeof(sparse_banks[0]);
1641 
1642 		all_cpus = send_ipi_ex.vp_set.format == HV_GENERIC_SET_ALL;
1643 
1644 		if (!sparse_banks_len)
1645 			goto ret_success;
1646 
1647 		if (!all_cpus &&
1648 		    kvm_read_guest(kvm,
1649 				   ingpa + offsetof(struct hv_send_ipi_ex,
1650 						    vp_set.bank_contents),
1651 				   sparse_banks,
1652 				   sparse_banks_len))
1653 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1654 	}
1655 
1656 	if ((vector < HV_IPI_LOW_VECTOR) || (vector > HV_IPI_HIGH_VECTOR))
1657 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
1658 
1659 	vcpu_mask = all_cpus ? NULL :
1660 		sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1661 					vp_bitmap, vcpu_bitmap);
1662 
1663 	kvm_send_ipi_to_many(kvm, vector, vcpu_mask);
1664 
1665 ret_success:
1666 	return HV_STATUS_SUCCESS;
1667 }
1668 
1669 bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1670 {
1671 	return READ_ONCE(kvm->arch.hyperv.hv_guest_os_id) != 0;
1672 }
1673 
1674 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
1675 {
1676 	bool longmode;
1677 
1678 	longmode = is_64_bit_mode(vcpu);
1679 	if (longmode)
1680 		kvm_rax_write(vcpu, result);
1681 	else {
1682 		kvm_rdx_write(vcpu, result >> 32);
1683 		kvm_rax_write(vcpu, result & 0xffffffff);
1684 	}
1685 }
1686 
1687 static int kvm_hv_hypercall_complete(struct kvm_vcpu *vcpu, u64 result)
1688 {
1689 	kvm_hv_hypercall_set_result(vcpu, result);
1690 	++vcpu->stat.hypercalls;
1691 	return kvm_skip_emulated_instruction(vcpu);
1692 }
1693 
1694 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
1695 {
1696 	return kvm_hv_hypercall_complete(vcpu, vcpu->run->hyperv.u.hcall.result);
1697 }
1698 
1699 static u16 kvm_hvcall_signal_event(struct kvm_vcpu *vcpu, bool fast, u64 param)
1700 {
1701 	struct eventfd_ctx *eventfd;
1702 
1703 	if (unlikely(!fast)) {
1704 		int ret;
1705 		gpa_t gpa = param;
1706 
1707 		if ((gpa & (__alignof__(param) - 1)) ||
1708 		    offset_in_page(gpa) + sizeof(param) > PAGE_SIZE)
1709 			return HV_STATUS_INVALID_ALIGNMENT;
1710 
1711 		ret = kvm_vcpu_read_guest(vcpu, gpa, &param, sizeof(param));
1712 		if (ret < 0)
1713 			return HV_STATUS_INVALID_ALIGNMENT;
1714 	}
1715 
1716 	/*
1717 	 * Per spec, bits 32-47 contain the extra "flag number".  However, we
1718 	 * have no use for it, and in all known usecases it is zero, so just
1719 	 * report lookup failure if it isn't.
1720 	 */
1721 	if (param & 0xffff00000000ULL)
1722 		return HV_STATUS_INVALID_PORT_ID;
1723 	/* remaining bits are reserved-zero */
1724 	if (param & ~KVM_HYPERV_CONN_ID_MASK)
1725 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
1726 
1727 	/* the eventfd is protected by vcpu->kvm->srcu, but conn_to_evt isn't */
1728 	rcu_read_lock();
1729 	eventfd = idr_find(&vcpu->kvm->arch.hyperv.conn_to_evt, param);
1730 	rcu_read_unlock();
1731 	if (!eventfd)
1732 		return HV_STATUS_INVALID_PORT_ID;
1733 
1734 	eventfd_signal(eventfd, 1);
1735 	return HV_STATUS_SUCCESS;
1736 }
1737 
1738 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
1739 {
1740 	u64 param, ingpa, outgpa, ret = HV_STATUS_SUCCESS;
1741 	uint16_t code, rep_idx, rep_cnt;
1742 	bool fast, rep;
1743 
1744 	/*
1745 	 * hypercall generates UD from non zero cpl and real mode
1746 	 * per HYPER-V spec
1747 	 */
1748 	if (kvm_x86_ops.get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
1749 		kvm_queue_exception(vcpu, UD_VECTOR);
1750 		return 1;
1751 	}
1752 
1753 #ifdef CONFIG_X86_64
1754 	if (is_64_bit_mode(vcpu)) {
1755 		param = kvm_rcx_read(vcpu);
1756 		ingpa = kvm_rdx_read(vcpu);
1757 		outgpa = kvm_r8_read(vcpu);
1758 	} else
1759 #endif
1760 	{
1761 		param = ((u64)kvm_rdx_read(vcpu) << 32) |
1762 			(kvm_rax_read(vcpu) & 0xffffffff);
1763 		ingpa = ((u64)kvm_rbx_read(vcpu) << 32) |
1764 			(kvm_rcx_read(vcpu) & 0xffffffff);
1765 		outgpa = ((u64)kvm_rdi_read(vcpu) << 32) |
1766 			(kvm_rsi_read(vcpu) & 0xffffffff);
1767 	}
1768 
1769 	code = param & 0xffff;
1770 	fast = !!(param & HV_HYPERCALL_FAST_BIT);
1771 	rep_cnt = (param >> HV_HYPERCALL_REP_COMP_OFFSET) & 0xfff;
1772 	rep_idx = (param >> HV_HYPERCALL_REP_START_OFFSET) & 0xfff;
1773 	rep = !!(rep_cnt || rep_idx);
1774 
1775 	trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
1776 
1777 	switch (code) {
1778 	case HVCALL_NOTIFY_LONG_SPIN_WAIT:
1779 		if (unlikely(rep)) {
1780 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1781 			break;
1782 		}
1783 		kvm_vcpu_on_spin(vcpu, true);
1784 		break;
1785 	case HVCALL_SIGNAL_EVENT:
1786 		if (unlikely(rep)) {
1787 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1788 			break;
1789 		}
1790 		ret = kvm_hvcall_signal_event(vcpu, fast, ingpa);
1791 		if (ret != HV_STATUS_INVALID_PORT_ID)
1792 			break;
1793 		fallthrough;	/* maybe userspace knows this conn_id */
1794 	case HVCALL_POST_MESSAGE:
1795 		/* don't bother userspace if it has no way to handle it */
1796 		if (unlikely(rep || !vcpu_to_synic(vcpu)->active)) {
1797 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1798 			break;
1799 		}
1800 		vcpu->run->exit_reason = KVM_EXIT_HYPERV;
1801 		vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
1802 		vcpu->run->hyperv.u.hcall.input = param;
1803 		vcpu->run->hyperv.u.hcall.params[0] = ingpa;
1804 		vcpu->run->hyperv.u.hcall.params[1] = outgpa;
1805 		vcpu->arch.complete_userspace_io =
1806 				kvm_hv_hypercall_complete_userspace;
1807 		return 0;
1808 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
1809 		if (unlikely(fast || !rep_cnt || rep_idx)) {
1810 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1811 			break;
1812 		}
1813 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false);
1814 		break;
1815 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
1816 		if (unlikely(fast || rep)) {
1817 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1818 			break;
1819 		}
1820 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false);
1821 		break;
1822 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
1823 		if (unlikely(fast || !rep_cnt || rep_idx)) {
1824 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1825 			break;
1826 		}
1827 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
1828 		break;
1829 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
1830 		if (unlikely(fast || rep)) {
1831 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1832 			break;
1833 		}
1834 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
1835 		break;
1836 	case HVCALL_SEND_IPI:
1837 		if (unlikely(rep)) {
1838 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1839 			break;
1840 		}
1841 		ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, false, fast);
1842 		break;
1843 	case HVCALL_SEND_IPI_EX:
1844 		if (unlikely(fast || rep)) {
1845 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1846 			break;
1847 		}
1848 		ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, true, false);
1849 		break;
1850 	case HVCALL_POST_DEBUG_DATA:
1851 	case HVCALL_RETRIEVE_DEBUG_DATA:
1852 		if (unlikely(fast)) {
1853 			ret = HV_STATUS_INVALID_PARAMETER;
1854 			break;
1855 		}
1856 		fallthrough;
1857 	case HVCALL_RESET_DEBUG_SESSION: {
1858 		struct kvm_hv_syndbg *syndbg = vcpu_to_hv_syndbg(vcpu);
1859 
1860 		if (!kvm_hv_is_syndbg_enabled(vcpu)) {
1861 			ret = HV_STATUS_INVALID_HYPERCALL_CODE;
1862 			break;
1863 		}
1864 
1865 		if (!(syndbg->options & HV_X64_SYNDBG_OPTION_USE_HCALLS)) {
1866 			ret = HV_STATUS_OPERATION_DENIED;
1867 			break;
1868 		}
1869 		vcpu->run->exit_reason = KVM_EXIT_HYPERV;
1870 		vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
1871 		vcpu->run->hyperv.u.hcall.input = param;
1872 		vcpu->run->hyperv.u.hcall.params[0] = ingpa;
1873 		vcpu->run->hyperv.u.hcall.params[1] = outgpa;
1874 		vcpu->arch.complete_userspace_io =
1875 				kvm_hv_hypercall_complete_userspace;
1876 		return 0;
1877 	}
1878 	default:
1879 		ret = HV_STATUS_INVALID_HYPERCALL_CODE;
1880 		break;
1881 	}
1882 
1883 	return kvm_hv_hypercall_complete(vcpu, ret);
1884 }
1885 
1886 void kvm_hv_init_vm(struct kvm *kvm)
1887 {
1888 	mutex_init(&kvm->arch.hyperv.hv_lock);
1889 	idr_init(&kvm->arch.hyperv.conn_to_evt);
1890 }
1891 
1892 void kvm_hv_destroy_vm(struct kvm *kvm)
1893 {
1894 	struct eventfd_ctx *eventfd;
1895 	int i;
1896 
1897 	idr_for_each_entry(&kvm->arch.hyperv.conn_to_evt, eventfd, i)
1898 		eventfd_ctx_put(eventfd);
1899 	idr_destroy(&kvm->arch.hyperv.conn_to_evt);
1900 }
1901 
1902 static int kvm_hv_eventfd_assign(struct kvm *kvm, u32 conn_id, int fd)
1903 {
1904 	struct kvm_hv *hv = &kvm->arch.hyperv;
1905 	struct eventfd_ctx *eventfd;
1906 	int ret;
1907 
1908 	eventfd = eventfd_ctx_fdget(fd);
1909 	if (IS_ERR(eventfd))
1910 		return PTR_ERR(eventfd);
1911 
1912 	mutex_lock(&hv->hv_lock);
1913 	ret = idr_alloc(&hv->conn_to_evt, eventfd, conn_id, conn_id + 1,
1914 			GFP_KERNEL_ACCOUNT);
1915 	mutex_unlock(&hv->hv_lock);
1916 
1917 	if (ret >= 0)
1918 		return 0;
1919 
1920 	if (ret == -ENOSPC)
1921 		ret = -EEXIST;
1922 	eventfd_ctx_put(eventfd);
1923 	return ret;
1924 }
1925 
1926 static int kvm_hv_eventfd_deassign(struct kvm *kvm, u32 conn_id)
1927 {
1928 	struct kvm_hv *hv = &kvm->arch.hyperv;
1929 	struct eventfd_ctx *eventfd;
1930 
1931 	mutex_lock(&hv->hv_lock);
1932 	eventfd = idr_remove(&hv->conn_to_evt, conn_id);
1933 	mutex_unlock(&hv->hv_lock);
1934 
1935 	if (!eventfd)
1936 		return -ENOENT;
1937 
1938 	synchronize_srcu(&kvm->srcu);
1939 	eventfd_ctx_put(eventfd);
1940 	return 0;
1941 }
1942 
1943 int kvm_vm_ioctl_hv_eventfd(struct kvm *kvm, struct kvm_hyperv_eventfd *args)
1944 {
1945 	if ((args->flags & ~KVM_HYPERV_EVENTFD_DEASSIGN) ||
1946 	    (args->conn_id & ~KVM_HYPERV_CONN_ID_MASK))
1947 		return -EINVAL;
1948 
1949 	if (args->flags == KVM_HYPERV_EVENTFD_DEASSIGN)
1950 		return kvm_hv_eventfd_deassign(kvm, args->conn_id);
1951 	return kvm_hv_eventfd_assign(kvm, args->conn_id, args->fd);
1952 }
1953 
1954 int kvm_get_hv_cpuid(struct kvm_vcpu *vcpu, struct kvm_cpuid2 *cpuid,
1955 		     struct kvm_cpuid_entry2 __user *entries)
1956 {
1957 	uint16_t evmcs_ver = 0;
1958 	struct kvm_cpuid_entry2 cpuid_entries[] = {
1959 		{ .function = HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS },
1960 		{ .function = HYPERV_CPUID_INTERFACE },
1961 		{ .function = HYPERV_CPUID_VERSION },
1962 		{ .function = HYPERV_CPUID_FEATURES },
1963 		{ .function = HYPERV_CPUID_ENLIGHTMENT_INFO },
1964 		{ .function = HYPERV_CPUID_IMPLEMENT_LIMITS },
1965 		{ .function = HYPERV_CPUID_SYNDBG_VENDOR_AND_MAX_FUNCTIONS },
1966 		{ .function = HYPERV_CPUID_SYNDBG_INTERFACE },
1967 		{ .function = HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES	},
1968 		{ .function = HYPERV_CPUID_NESTED_FEATURES },
1969 	};
1970 	int i, nent = ARRAY_SIZE(cpuid_entries);
1971 
1972 	if (kvm_x86_ops.nested_ops->get_evmcs_version)
1973 		evmcs_ver = kvm_x86_ops.nested_ops->get_evmcs_version(vcpu);
1974 
1975 	/* Skip NESTED_FEATURES if eVMCS is not supported */
1976 	if (!evmcs_ver)
1977 		--nent;
1978 
1979 	if (cpuid->nent < nent)
1980 		return -E2BIG;
1981 
1982 	if (cpuid->nent > nent)
1983 		cpuid->nent = nent;
1984 
1985 	for (i = 0; i < nent; i++) {
1986 		struct kvm_cpuid_entry2 *ent = &cpuid_entries[i];
1987 		u32 signature[3];
1988 
1989 		switch (ent->function) {
1990 		case HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS:
1991 			memcpy(signature, "Linux KVM Hv", 12);
1992 
1993 			ent->eax = HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES;
1994 			ent->ebx = signature[0];
1995 			ent->ecx = signature[1];
1996 			ent->edx = signature[2];
1997 			break;
1998 
1999 		case HYPERV_CPUID_INTERFACE:
2000 			memcpy(signature, "Hv#1\0\0\0\0\0\0\0\0", 12);
2001 			ent->eax = signature[0];
2002 			break;
2003 
2004 		case HYPERV_CPUID_VERSION:
2005 			/*
2006 			 * We implement some Hyper-V 2016 functions so let's use
2007 			 * this version.
2008 			 */
2009 			ent->eax = 0x00003839;
2010 			ent->ebx = 0x000A0000;
2011 			break;
2012 
2013 		case HYPERV_CPUID_FEATURES:
2014 			ent->eax |= HV_MSR_VP_RUNTIME_AVAILABLE;
2015 			ent->eax |= HV_MSR_TIME_REF_COUNT_AVAILABLE;
2016 			ent->eax |= HV_MSR_SYNIC_AVAILABLE;
2017 			ent->eax |= HV_MSR_SYNTIMER_AVAILABLE;
2018 			ent->eax |= HV_MSR_APIC_ACCESS_AVAILABLE;
2019 			ent->eax |= HV_MSR_HYPERCALL_AVAILABLE;
2020 			ent->eax |= HV_MSR_VP_INDEX_AVAILABLE;
2021 			ent->eax |= HV_MSR_RESET_AVAILABLE;
2022 			ent->eax |= HV_MSR_REFERENCE_TSC_AVAILABLE;
2023 			ent->eax |= HV_ACCESS_FREQUENCY_MSRS;
2024 			ent->eax |= HV_ACCESS_REENLIGHTENMENT;
2025 
2026 			ent->ebx |= HV_POST_MESSAGES;
2027 			ent->ebx |= HV_SIGNAL_EVENTS;
2028 
2029 			ent->edx |= HV_FEATURE_FREQUENCY_MSRS_AVAILABLE;
2030 			ent->edx |= HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
2031 
2032 			ent->ebx |= HV_DEBUGGING;
2033 			ent->edx |= HV_X64_GUEST_DEBUGGING_AVAILABLE;
2034 			ent->edx |= HV_FEATURE_DEBUG_MSRS_AVAILABLE;
2035 
2036 			/*
2037 			 * Direct Synthetic timers only make sense with in-kernel
2038 			 * LAPIC
2039 			 */
2040 			if (!vcpu || lapic_in_kernel(vcpu))
2041 				ent->edx |= HV_STIMER_DIRECT_MODE_AVAILABLE;
2042 
2043 			break;
2044 
2045 		case HYPERV_CPUID_ENLIGHTMENT_INFO:
2046 			ent->eax |= HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
2047 			ent->eax |= HV_X64_APIC_ACCESS_RECOMMENDED;
2048 			ent->eax |= HV_X64_RELAXED_TIMING_RECOMMENDED;
2049 			ent->eax |= HV_X64_CLUSTER_IPI_RECOMMENDED;
2050 			ent->eax |= HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED;
2051 			if (evmcs_ver)
2052 				ent->eax |= HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
2053 			if (!cpu_smt_possible())
2054 				ent->eax |= HV_X64_NO_NONARCH_CORESHARING;
2055 			/*
2056 			 * Default number of spinlock retry attempts, matches
2057 			 * HyperV 2016.
2058 			 */
2059 			ent->ebx = 0x00000FFF;
2060 
2061 			break;
2062 
2063 		case HYPERV_CPUID_IMPLEMENT_LIMITS:
2064 			/* Maximum number of virtual processors */
2065 			ent->eax = KVM_MAX_VCPUS;
2066 			/*
2067 			 * Maximum number of logical processors, matches
2068 			 * HyperV 2016.
2069 			 */
2070 			ent->ebx = 64;
2071 
2072 			break;
2073 
2074 		case HYPERV_CPUID_NESTED_FEATURES:
2075 			ent->eax = evmcs_ver;
2076 
2077 			break;
2078 
2079 		case HYPERV_CPUID_SYNDBG_VENDOR_AND_MAX_FUNCTIONS:
2080 			memcpy(signature, "Linux KVM Hv", 12);
2081 
2082 			ent->eax = 0;
2083 			ent->ebx = signature[0];
2084 			ent->ecx = signature[1];
2085 			ent->edx = signature[2];
2086 			break;
2087 
2088 		case HYPERV_CPUID_SYNDBG_INTERFACE:
2089 			memcpy(signature, "VS#1\0\0\0\0\0\0\0\0", 12);
2090 			ent->eax = signature[0];
2091 			break;
2092 
2093 		case HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES:
2094 			ent->eax |= HV_X64_SYNDBG_CAP_ALLOW_KERNEL_DEBUGGING;
2095 			break;
2096 
2097 		default:
2098 			break;
2099 		}
2100 	}
2101 
2102 	if (copy_to_user(entries, cpuid_entries,
2103 			 nent * sizeof(struct kvm_cpuid_entry2)))
2104 		return -EFAULT;
2105 
2106 	return 0;
2107 }
2108