xref: /openbmc/linux/arch/x86/kvm/hyperv.c (revision 78700c0a)
1 /*
2  * KVM Microsoft Hyper-V emulation
3  *
4  * derived from arch/x86/kvm/x86.c
5  *
6  * Copyright (C) 2006 Qumranet, Inc.
7  * Copyright (C) 2008 Qumranet, Inc.
8  * Copyright IBM Corporation, 2008
9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10  * Copyright (C) 2015 Andrey Smetanin <asmetanin@virtuozzo.com>
11  *
12  * Authors:
13  *   Avi Kivity   <avi@qumranet.com>
14  *   Yaniv Kamay  <yaniv@qumranet.com>
15  *   Amit Shah    <amit.shah@qumranet.com>
16  *   Ben-Ami Yassour <benami@il.ibm.com>
17  *   Andrey Smetanin <asmetanin@virtuozzo.com>
18  *
19  * This work is licensed under the terms of the GNU GPL, version 2.  See
20  * the COPYING file in the top-level directory.
21  *
22  */
23 
24 #include "x86.h"
25 #include "lapic.h"
26 #include "ioapic.h"
27 #include "hyperv.h"
28 
29 #include <linux/kvm_host.h>
30 #include <linux/highmem.h>
31 #include <asm/apicdef.h>
32 #include <trace/events/kvm.h>
33 
34 #include "trace.h"
35 
36 static inline u64 synic_read_sint(struct kvm_vcpu_hv_synic *synic, int sint)
37 {
38 	return atomic64_read(&synic->sint[sint]);
39 }
40 
41 static inline int synic_get_sint_vector(u64 sint_value)
42 {
43 	if (sint_value & HV_SYNIC_SINT_MASKED)
44 		return -1;
45 	return sint_value & HV_SYNIC_SINT_VECTOR_MASK;
46 }
47 
48 static bool synic_has_vector_connected(struct kvm_vcpu_hv_synic *synic,
49 				      int vector)
50 {
51 	int i;
52 
53 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
54 		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
55 			return true;
56 	}
57 	return false;
58 }
59 
60 static bool synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic *synic,
61 				     int vector)
62 {
63 	int i;
64 	u64 sint_value;
65 
66 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
67 		sint_value = synic_read_sint(synic, i);
68 		if (synic_get_sint_vector(sint_value) == vector &&
69 		    sint_value & HV_SYNIC_SINT_AUTO_EOI)
70 			return true;
71 	}
72 	return false;
73 }
74 
75 static int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint,
76 			  u64 data, bool host)
77 {
78 	int vector;
79 
80 	vector = data & HV_SYNIC_SINT_VECTOR_MASK;
81 	if (vector < 16 && !host)
82 		return 1;
83 	/*
84 	 * Guest may configure multiple SINTs to use the same vector, so
85 	 * we maintain a bitmap of vectors handled by synic, and a
86 	 * bitmap of vectors with auto-eoi behavior.  The bitmaps are
87 	 * updated here, and atomically queried on fast paths.
88 	 */
89 
90 	atomic64_set(&synic->sint[sint], data);
91 
92 	if (synic_has_vector_connected(synic, vector))
93 		__set_bit(vector, synic->vec_bitmap);
94 	else
95 		__clear_bit(vector, synic->vec_bitmap);
96 
97 	if (synic_has_vector_auto_eoi(synic, vector))
98 		__set_bit(vector, synic->auto_eoi_bitmap);
99 	else
100 		__clear_bit(vector, synic->auto_eoi_bitmap);
101 
102 	/* Load SynIC vectors into EOI exit bitmap */
103 	kvm_make_request(KVM_REQ_SCAN_IOAPIC, synic_to_vcpu(synic));
104 	return 0;
105 }
106 
107 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vcpu_id)
108 {
109 	struct kvm_vcpu *vcpu;
110 	struct kvm_vcpu_hv_synic *synic;
111 
112 	if (vcpu_id >= atomic_read(&kvm->online_vcpus))
113 		return NULL;
114 	vcpu = kvm_get_vcpu(kvm, vcpu_id);
115 	if (!vcpu)
116 		return NULL;
117 	synic = vcpu_to_synic(vcpu);
118 	return (synic->active) ? synic : NULL;
119 }
120 
121 static void synic_clear_sint_msg_pending(struct kvm_vcpu_hv_synic *synic,
122 					u32 sint)
123 {
124 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
125 	struct page *page;
126 	gpa_t gpa;
127 	struct hv_message *msg;
128 	struct hv_message_page *msg_page;
129 
130 	gpa = synic->msg_page & PAGE_MASK;
131 	page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
132 	if (is_error_page(page)) {
133 		vcpu_err(vcpu, "Hyper-V SynIC can't get msg page, gpa 0x%llx\n",
134 			 gpa);
135 		return;
136 	}
137 	msg_page = kmap_atomic(page);
138 
139 	msg = &msg_page->sint_message[sint];
140 	msg->header.message_flags.msg_pending = 0;
141 
142 	kunmap_atomic(msg_page);
143 	kvm_release_page_dirty(page);
144 	kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
145 }
146 
147 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
148 {
149 	struct kvm *kvm = vcpu->kvm;
150 	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
151 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
152 	struct kvm_vcpu_hv_stimer *stimer;
153 	int gsi, idx, stimers_pending;
154 
155 	trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
156 
157 	if (synic->msg_page & HV_SYNIC_SIMP_ENABLE)
158 		synic_clear_sint_msg_pending(synic, sint);
159 
160 	/* Try to deliver pending Hyper-V SynIC timers messages */
161 	stimers_pending = 0;
162 	for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
163 		stimer = &hv_vcpu->stimer[idx];
164 		if (stimer->msg_pending &&
165 		    (stimer->config & HV_STIMER_ENABLE) &&
166 		    HV_STIMER_SINT(stimer->config) == sint) {
167 			set_bit(stimer->index,
168 				hv_vcpu->stimer_pending_bitmap);
169 			stimers_pending++;
170 		}
171 	}
172 	if (stimers_pending)
173 		kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
174 
175 	idx = srcu_read_lock(&kvm->irq_srcu);
176 	gsi = atomic_read(&synic->sint_to_gsi[sint]);
177 	if (gsi != -1)
178 		kvm_notify_acked_gsi(kvm, gsi);
179 	srcu_read_unlock(&kvm->irq_srcu, idx);
180 }
181 
182 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
183 {
184 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
185 	struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
186 
187 	hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
188 	hv_vcpu->exit.u.synic.msr = msr;
189 	hv_vcpu->exit.u.synic.control = synic->control;
190 	hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
191 	hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
192 
193 	kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
194 }
195 
196 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
197 			 u32 msr, u64 data, bool host)
198 {
199 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
200 	int ret;
201 
202 	if (!synic->active)
203 		return 1;
204 
205 	trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
206 
207 	ret = 0;
208 	switch (msr) {
209 	case HV_X64_MSR_SCONTROL:
210 		synic->control = data;
211 		if (!host)
212 			synic_exit(synic, msr);
213 		break;
214 	case HV_X64_MSR_SVERSION:
215 		if (!host) {
216 			ret = 1;
217 			break;
218 		}
219 		synic->version = data;
220 		break;
221 	case HV_X64_MSR_SIEFP:
222 		if (data & HV_SYNIC_SIEFP_ENABLE)
223 			if (kvm_clear_guest(vcpu->kvm,
224 					    data & PAGE_MASK, PAGE_SIZE)) {
225 				ret = 1;
226 				break;
227 			}
228 		synic->evt_page = data;
229 		if (!host)
230 			synic_exit(synic, msr);
231 		break;
232 	case HV_X64_MSR_SIMP:
233 		if (data & HV_SYNIC_SIMP_ENABLE)
234 			if (kvm_clear_guest(vcpu->kvm,
235 					    data & PAGE_MASK, PAGE_SIZE)) {
236 				ret = 1;
237 				break;
238 			}
239 		synic->msg_page = data;
240 		if (!host)
241 			synic_exit(synic, msr);
242 		break;
243 	case HV_X64_MSR_EOM: {
244 		int i;
245 
246 		for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
247 			kvm_hv_notify_acked_sint(vcpu, i);
248 		break;
249 	}
250 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
251 		ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
252 		break;
253 	default:
254 		ret = 1;
255 		break;
256 	}
257 	return ret;
258 }
259 
260 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata)
261 {
262 	int ret;
263 
264 	if (!synic->active)
265 		return 1;
266 
267 	ret = 0;
268 	switch (msr) {
269 	case HV_X64_MSR_SCONTROL:
270 		*pdata = synic->control;
271 		break;
272 	case HV_X64_MSR_SVERSION:
273 		*pdata = synic->version;
274 		break;
275 	case HV_X64_MSR_SIEFP:
276 		*pdata = synic->evt_page;
277 		break;
278 	case HV_X64_MSR_SIMP:
279 		*pdata = synic->msg_page;
280 		break;
281 	case HV_X64_MSR_EOM:
282 		*pdata = 0;
283 		break;
284 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
285 		*pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
286 		break;
287 	default:
288 		ret = 1;
289 		break;
290 	}
291 	return ret;
292 }
293 
294 int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
295 {
296 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
297 	struct kvm_lapic_irq irq;
298 	int ret, vector;
299 
300 	if (sint >= ARRAY_SIZE(synic->sint))
301 		return -EINVAL;
302 
303 	vector = synic_get_sint_vector(synic_read_sint(synic, sint));
304 	if (vector < 0)
305 		return -ENOENT;
306 
307 	memset(&irq, 0, sizeof(irq));
308 	irq.dest_id = kvm_apic_id(vcpu->arch.apic);
309 	irq.dest_mode = APIC_DEST_PHYSICAL;
310 	irq.delivery_mode = APIC_DM_FIXED;
311 	irq.vector = vector;
312 	irq.level = 1;
313 
314 	ret = kvm_irq_delivery_to_apic(vcpu->kvm, NULL, &irq, NULL);
315 	trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
316 	return ret;
317 }
318 
319 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vcpu_id, u32 sint)
320 {
321 	struct kvm_vcpu_hv_synic *synic;
322 
323 	synic = synic_get(kvm, vcpu_id);
324 	if (!synic)
325 		return -EINVAL;
326 
327 	return synic_set_irq(synic, sint);
328 }
329 
330 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
331 {
332 	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
333 	int i;
334 
335 	trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
336 
337 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
338 		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
339 			kvm_hv_notify_acked_sint(vcpu, i);
340 }
341 
342 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vcpu_id, u32 sint, int gsi)
343 {
344 	struct kvm_vcpu_hv_synic *synic;
345 
346 	synic = synic_get(kvm, vcpu_id);
347 	if (!synic)
348 		return -EINVAL;
349 
350 	if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
351 		return -EINVAL;
352 
353 	atomic_set(&synic->sint_to_gsi[sint], gsi);
354 	return 0;
355 }
356 
357 void kvm_hv_irq_routing_update(struct kvm *kvm)
358 {
359 	struct kvm_irq_routing_table *irq_rt;
360 	struct kvm_kernel_irq_routing_entry *e;
361 	u32 gsi;
362 
363 	irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
364 					lockdep_is_held(&kvm->irq_lock));
365 
366 	for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
367 		hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
368 			if (e->type == KVM_IRQ_ROUTING_HV_SINT)
369 				kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
370 						    e->hv_sint.sint, gsi);
371 		}
372 	}
373 }
374 
375 static void synic_init(struct kvm_vcpu_hv_synic *synic)
376 {
377 	int i;
378 
379 	memset(synic, 0, sizeof(*synic));
380 	synic->version = HV_SYNIC_VERSION_1;
381 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
382 		atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
383 		atomic_set(&synic->sint_to_gsi[i], -1);
384 	}
385 }
386 
387 static u64 get_time_ref_counter(struct kvm *kvm)
388 {
389 	return div_u64(get_kernel_ns() + kvm->arch.kvmclock_offset, 100);
390 }
391 
392 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
393 				bool vcpu_kick)
394 {
395 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
396 
397 	set_bit(stimer->index,
398 		vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
399 	kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
400 	if (vcpu_kick)
401 		kvm_vcpu_kick(vcpu);
402 }
403 
404 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
405 {
406 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
407 
408 	trace_kvm_hv_stimer_cleanup(stimer_to_vcpu(stimer)->vcpu_id,
409 				    stimer->index);
410 
411 	hrtimer_cancel(&stimer->timer);
412 	clear_bit(stimer->index,
413 		  vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
414 	stimer->msg_pending = false;
415 	stimer->exp_time = 0;
416 }
417 
418 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
419 {
420 	struct kvm_vcpu_hv_stimer *stimer;
421 
422 	stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
423 	trace_kvm_hv_stimer_callback(stimer_to_vcpu(stimer)->vcpu_id,
424 				     stimer->index);
425 	stimer_mark_pending(stimer, true);
426 
427 	return HRTIMER_NORESTART;
428 }
429 
430 /*
431  * stimer_start() assumptions:
432  * a) stimer->count is not equal to 0
433  * b) stimer->config has HV_STIMER_ENABLE flag
434  */
435 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
436 {
437 	u64 time_now;
438 	ktime_t ktime_now;
439 
440 	time_now = get_time_ref_counter(stimer_to_vcpu(stimer)->kvm);
441 	ktime_now = ktime_get();
442 
443 	if (stimer->config & HV_STIMER_PERIODIC) {
444 		if (stimer->exp_time) {
445 			if (time_now >= stimer->exp_time) {
446 				u64 remainder;
447 
448 				div64_u64_rem(time_now - stimer->exp_time,
449 					      stimer->count, &remainder);
450 				stimer->exp_time =
451 					time_now + (stimer->count - remainder);
452 			}
453 		} else
454 			stimer->exp_time = time_now + stimer->count;
455 
456 		trace_kvm_hv_stimer_start_periodic(
457 					stimer_to_vcpu(stimer)->vcpu_id,
458 					stimer->index,
459 					time_now, stimer->exp_time);
460 
461 		hrtimer_start(&stimer->timer,
462 			      ktime_add_ns(ktime_now,
463 					   100 * (stimer->exp_time - time_now)),
464 			      HRTIMER_MODE_ABS);
465 		return 0;
466 	}
467 	stimer->exp_time = stimer->count;
468 	if (time_now >= stimer->count) {
469 		/*
470 		 * Expire timer according to Hypervisor Top-Level Functional
471 		 * specification v4(15.3.1):
472 		 * "If a one shot is enabled and the specified count is in
473 		 * the past, it will expire immediately."
474 		 */
475 		stimer_mark_pending(stimer, false);
476 		return 0;
477 	}
478 
479 	trace_kvm_hv_stimer_start_one_shot(stimer_to_vcpu(stimer)->vcpu_id,
480 					   stimer->index,
481 					   time_now, stimer->count);
482 
483 	hrtimer_start(&stimer->timer,
484 		      ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)),
485 		      HRTIMER_MODE_ABS);
486 	return 0;
487 }
488 
489 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
490 			     bool host)
491 {
492 	trace_kvm_hv_stimer_set_config(stimer_to_vcpu(stimer)->vcpu_id,
493 				       stimer->index, config, host);
494 
495 	stimer_cleanup(stimer);
496 	if ((stimer->config & HV_STIMER_ENABLE) && HV_STIMER_SINT(config) == 0)
497 		config &= ~HV_STIMER_ENABLE;
498 	stimer->config = config;
499 	stimer_mark_pending(stimer, false);
500 	return 0;
501 }
502 
503 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
504 			    bool host)
505 {
506 	trace_kvm_hv_stimer_set_count(stimer_to_vcpu(stimer)->vcpu_id,
507 				      stimer->index, count, host);
508 
509 	stimer_cleanup(stimer);
510 	stimer->count = count;
511 	if (stimer->count == 0)
512 		stimer->config &= ~HV_STIMER_ENABLE;
513 	else if (stimer->config & HV_STIMER_AUTOENABLE)
514 		stimer->config |= HV_STIMER_ENABLE;
515 	stimer_mark_pending(stimer, false);
516 	return 0;
517 }
518 
519 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
520 {
521 	*pconfig = stimer->config;
522 	return 0;
523 }
524 
525 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
526 {
527 	*pcount = stimer->count;
528 	return 0;
529 }
530 
531 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
532 			     struct hv_message *src_msg)
533 {
534 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
535 	struct page *page;
536 	gpa_t gpa;
537 	struct hv_message *dst_msg;
538 	int r;
539 	struct hv_message_page *msg_page;
540 
541 	if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
542 		return -ENOENT;
543 
544 	gpa = synic->msg_page & PAGE_MASK;
545 	page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
546 	if (is_error_page(page))
547 		return -EFAULT;
548 
549 	msg_page = kmap_atomic(page);
550 	dst_msg = &msg_page->sint_message[sint];
551 	if (sync_cmpxchg(&dst_msg->header.message_type, HVMSG_NONE,
552 			 src_msg->header.message_type) != HVMSG_NONE) {
553 		dst_msg->header.message_flags.msg_pending = 1;
554 		r = -EAGAIN;
555 	} else {
556 		memcpy(&dst_msg->u.payload, &src_msg->u.payload,
557 		       src_msg->header.payload_size);
558 		dst_msg->header.message_type = src_msg->header.message_type;
559 		dst_msg->header.payload_size = src_msg->header.payload_size;
560 		r = synic_set_irq(synic, sint);
561 		if (r >= 1)
562 			r = 0;
563 		else if (r == 0)
564 			r = -EFAULT;
565 	}
566 	kunmap_atomic(msg_page);
567 	kvm_release_page_dirty(page);
568 	kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
569 	return r;
570 }
571 
572 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
573 {
574 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
575 	struct hv_message *msg = &stimer->msg;
576 	struct hv_timer_message_payload *payload =
577 			(struct hv_timer_message_payload *)&msg->u.payload;
578 
579 	payload->expiration_time = stimer->exp_time;
580 	payload->delivery_time = get_time_ref_counter(vcpu->kvm);
581 	return synic_deliver_msg(vcpu_to_synic(vcpu),
582 				 HV_STIMER_SINT(stimer->config), msg);
583 }
584 
585 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
586 {
587 	int r;
588 
589 	stimer->msg_pending = true;
590 	r = stimer_send_msg(stimer);
591 	trace_kvm_hv_stimer_expiration(stimer_to_vcpu(stimer)->vcpu_id,
592 				       stimer->index, r);
593 	if (!r) {
594 		stimer->msg_pending = false;
595 		if (!(stimer->config & HV_STIMER_PERIODIC))
596 			stimer->config &= ~HV_STIMER_ENABLE;
597 	}
598 }
599 
600 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
601 {
602 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
603 	struct kvm_vcpu_hv_stimer *stimer;
604 	u64 time_now, exp_time;
605 	int i;
606 
607 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
608 		if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
609 			stimer = &hv_vcpu->stimer[i];
610 			if (stimer->config & HV_STIMER_ENABLE) {
611 				exp_time = stimer->exp_time;
612 
613 				if (exp_time) {
614 					time_now =
615 						get_time_ref_counter(vcpu->kvm);
616 					if (time_now >= exp_time)
617 						stimer_expiration(stimer);
618 				}
619 
620 				if ((stimer->config & HV_STIMER_ENABLE) &&
621 				    stimer->count)
622 					stimer_start(stimer);
623 				else
624 					stimer_cleanup(stimer);
625 			}
626 		}
627 }
628 
629 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
630 {
631 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
632 	int i;
633 
634 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
635 		stimer_cleanup(&hv_vcpu->stimer[i]);
636 }
637 
638 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
639 {
640 	struct hv_message *msg = &stimer->msg;
641 	struct hv_timer_message_payload *payload =
642 			(struct hv_timer_message_payload *)&msg->u.payload;
643 
644 	memset(&msg->header, 0, sizeof(msg->header));
645 	msg->header.message_type = HVMSG_TIMER_EXPIRED;
646 	msg->header.payload_size = sizeof(*payload);
647 
648 	payload->timer_index = stimer->index;
649 	payload->expiration_time = 0;
650 	payload->delivery_time = 0;
651 }
652 
653 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
654 {
655 	memset(stimer, 0, sizeof(*stimer));
656 	stimer->index = timer_index;
657 	hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
658 	stimer->timer.function = stimer_timer_callback;
659 	stimer_prepare_msg(stimer);
660 }
661 
662 void kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
663 {
664 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
665 	int i;
666 
667 	synic_init(&hv_vcpu->synic);
668 
669 	bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
670 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
671 		stimer_init(&hv_vcpu->stimer[i], i);
672 }
673 
674 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu)
675 {
676 	/*
677 	 * Hyper-V SynIC auto EOI SINT's are
678 	 * not compatible with APICV, so deactivate APICV
679 	 */
680 	kvm_vcpu_deactivate_apicv(vcpu);
681 	vcpu_to_synic(vcpu)->active = true;
682 	return 0;
683 }
684 
685 static bool kvm_hv_msr_partition_wide(u32 msr)
686 {
687 	bool r = false;
688 
689 	switch (msr) {
690 	case HV_X64_MSR_GUEST_OS_ID:
691 	case HV_X64_MSR_HYPERCALL:
692 	case HV_X64_MSR_REFERENCE_TSC:
693 	case HV_X64_MSR_TIME_REF_COUNT:
694 	case HV_X64_MSR_CRASH_CTL:
695 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
696 	case HV_X64_MSR_RESET:
697 		r = true;
698 		break;
699 	}
700 
701 	return r;
702 }
703 
704 static int kvm_hv_msr_get_crash_data(struct kvm_vcpu *vcpu,
705 				     u32 index, u64 *pdata)
706 {
707 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
708 
709 	if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param)))
710 		return -EINVAL;
711 
712 	*pdata = hv->hv_crash_param[index];
713 	return 0;
714 }
715 
716 static int kvm_hv_msr_get_crash_ctl(struct kvm_vcpu *vcpu, u64 *pdata)
717 {
718 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
719 
720 	*pdata = hv->hv_crash_ctl;
721 	return 0;
722 }
723 
724 static int kvm_hv_msr_set_crash_ctl(struct kvm_vcpu *vcpu, u64 data, bool host)
725 {
726 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
727 
728 	if (host)
729 		hv->hv_crash_ctl = data & HV_X64_MSR_CRASH_CTL_NOTIFY;
730 
731 	if (!host && (data & HV_X64_MSR_CRASH_CTL_NOTIFY)) {
732 
733 		vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
734 			  hv->hv_crash_param[0],
735 			  hv->hv_crash_param[1],
736 			  hv->hv_crash_param[2],
737 			  hv->hv_crash_param[3],
738 			  hv->hv_crash_param[4]);
739 
740 		/* Send notification about crash to user space */
741 		kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
742 	}
743 
744 	return 0;
745 }
746 
747 static int kvm_hv_msr_set_crash_data(struct kvm_vcpu *vcpu,
748 				     u32 index, u64 data)
749 {
750 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
751 
752 	if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param)))
753 		return -EINVAL;
754 
755 	hv->hv_crash_param[index] = data;
756 	return 0;
757 }
758 
759 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
760 			     bool host)
761 {
762 	struct kvm *kvm = vcpu->kvm;
763 	struct kvm_hv *hv = &kvm->arch.hyperv;
764 
765 	switch (msr) {
766 	case HV_X64_MSR_GUEST_OS_ID:
767 		hv->hv_guest_os_id = data;
768 		/* setting guest os id to zero disables hypercall page */
769 		if (!hv->hv_guest_os_id)
770 			hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
771 		break;
772 	case HV_X64_MSR_HYPERCALL: {
773 		u64 gfn;
774 		unsigned long addr;
775 		u8 instructions[4];
776 
777 		/* if guest os id is not set hypercall should remain disabled */
778 		if (!hv->hv_guest_os_id)
779 			break;
780 		if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
781 			hv->hv_hypercall = data;
782 			break;
783 		}
784 		gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
785 		addr = gfn_to_hva(kvm, gfn);
786 		if (kvm_is_error_hva(addr))
787 			return 1;
788 		kvm_x86_ops->patch_hypercall(vcpu, instructions);
789 		((unsigned char *)instructions)[3] = 0xc3; /* ret */
790 		if (__copy_to_user((void __user *)addr, instructions, 4))
791 			return 1;
792 		hv->hv_hypercall = data;
793 		mark_page_dirty(kvm, gfn);
794 		break;
795 	}
796 	case HV_X64_MSR_REFERENCE_TSC: {
797 		u64 gfn;
798 		HV_REFERENCE_TSC_PAGE tsc_ref;
799 
800 		memset(&tsc_ref, 0, sizeof(tsc_ref));
801 		hv->hv_tsc_page = data;
802 		if (!(data & HV_X64_MSR_TSC_REFERENCE_ENABLE))
803 			break;
804 		gfn = data >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
805 		if (kvm_write_guest(
806 				kvm,
807 				gfn << HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT,
808 				&tsc_ref, sizeof(tsc_ref)))
809 			return 1;
810 		mark_page_dirty(kvm, gfn);
811 		break;
812 	}
813 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
814 		return kvm_hv_msr_set_crash_data(vcpu,
815 						 msr - HV_X64_MSR_CRASH_P0,
816 						 data);
817 	case HV_X64_MSR_CRASH_CTL:
818 		return kvm_hv_msr_set_crash_ctl(vcpu, data, host);
819 	case HV_X64_MSR_RESET:
820 		if (data == 1) {
821 			vcpu_debug(vcpu, "hyper-v reset requested\n");
822 			kvm_make_request(KVM_REQ_HV_RESET, vcpu);
823 		}
824 		break;
825 	default:
826 		vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
827 			    msr, data);
828 		return 1;
829 	}
830 	return 0;
831 }
832 
833 /* Calculate cpu time spent by current task in 100ns units */
834 static u64 current_task_runtime_100ns(void)
835 {
836 	cputime_t utime, stime;
837 
838 	task_cputime_adjusted(current, &utime, &stime);
839 	return div_u64(cputime_to_nsecs(utime + stime), 100);
840 }
841 
842 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
843 {
844 	struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
845 
846 	switch (msr) {
847 	case HV_X64_MSR_APIC_ASSIST_PAGE: {
848 		u64 gfn;
849 		unsigned long addr;
850 
851 		if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
852 			hv->hv_vapic = data;
853 			if (kvm_lapic_enable_pv_eoi(vcpu, 0))
854 				return 1;
855 			break;
856 		}
857 		gfn = data >> HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT;
858 		addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
859 		if (kvm_is_error_hva(addr))
860 			return 1;
861 		if (__clear_user((void __user *)addr, PAGE_SIZE))
862 			return 1;
863 		hv->hv_vapic = data;
864 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
865 		if (kvm_lapic_enable_pv_eoi(vcpu,
866 					    gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
867 			return 1;
868 		break;
869 	}
870 	case HV_X64_MSR_EOI:
871 		return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
872 	case HV_X64_MSR_ICR:
873 		return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
874 	case HV_X64_MSR_TPR:
875 		return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
876 	case HV_X64_MSR_VP_RUNTIME:
877 		if (!host)
878 			return 1;
879 		hv->runtime_offset = data - current_task_runtime_100ns();
880 		break;
881 	case HV_X64_MSR_SCONTROL:
882 	case HV_X64_MSR_SVERSION:
883 	case HV_X64_MSR_SIEFP:
884 	case HV_X64_MSR_SIMP:
885 	case HV_X64_MSR_EOM:
886 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
887 		return synic_set_msr(vcpu_to_synic(vcpu), msr, data, host);
888 	case HV_X64_MSR_STIMER0_CONFIG:
889 	case HV_X64_MSR_STIMER1_CONFIG:
890 	case HV_X64_MSR_STIMER2_CONFIG:
891 	case HV_X64_MSR_STIMER3_CONFIG: {
892 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
893 
894 		return stimer_set_config(vcpu_to_stimer(vcpu, timer_index),
895 					 data, host);
896 	}
897 	case HV_X64_MSR_STIMER0_COUNT:
898 	case HV_X64_MSR_STIMER1_COUNT:
899 	case HV_X64_MSR_STIMER2_COUNT:
900 	case HV_X64_MSR_STIMER3_COUNT: {
901 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
902 
903 		return stimer_set_count(vcpu_to_stimer(vcpu, timer_index),
904 					data, host);
905 	}
906 	default:
907 		vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
908 			    msr, data);
909 		return 1;
910 	}
911 
912 	return 0;
913 }
914 
915 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
916 {
917 	u64 data = 0;
918 	struct kvm *kvm = vcpu->kvm;
919 	struct kvm_hv *hv = &kvm->arch.hyperv;
920 
921 	switch (msr) {
922 	case HV_X64_MSR_GUEST_OS_ID:
923 		data = hv->hv_guest_os_id;
924 		break;
925 	case HV_X64_MSR_HYPERCALL:
926 		data = hv->hv_hypercall;
927 		break;
928 	case HV_X64_MSR_TIME_REF_COUNT:
929 		data = get_time_ref_counter(kvm);
930 		break;
931 	case HV_X64_MSR_REFERENCE_TSC:
932 		data = hv->hv_tsc_page;
933 		break;
934 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
935 		return kvm_hv_msr_get_crash_data(vcpu,
936 						 msr - HV_X64_MSR_CRASH_P0,
937 						 pdata);
938 	case HV_X64_MSR_CRASH_CTL:
939 		return kvm_hv_msr_get_crash_ctl(vcpu, pdata);
940 	case HV_X64_MSR_RESET:
941 		data = 0;
942 		break;
943 	default:
944 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
945 		return 1;
946 	}
947 
948 	*pdata = data;
949 	return 0;
950 }
951 
952 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
953 {
954 	u64 data = 0;
955 	struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
956 
957 	switch (msr) {
958 	case HV_X64_MSR_VP_INDEX: {
959 		int r;
960 		struct kvm_vcpu *v;
961 
962 		kvm_for_each_vcpu(r, v, vcpu->kvm) {
963 			if (v == vcpu) {
964 				data = r;
965 				break;
966 			}
967 		}
968 		break;
969 	}
970 	case HV_X64_MSR_EOI:
971 		return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
972 	case HV_X64_MSR_ICR:
973 		return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
974 	case HV_X64_MSR_TPR:
975 		return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
976 	case HV_X64_MSR_APIC_ASSIST_PAGE:
977 		data = hv->hv_vapic;
978 		break;
979 	case HV_X64_MSR_VP_RUNTIME:
980 		data = current_task_runtime_100ns() + hv->runtime_offset;
981 		break;
982 	case HV_X64_MSR_SCONTROL:
983 	case HV_X64_MSR_SVERSION:
984 	case HV_X64_MSR_SIEFP:
985 	case HV_X64_MSR_SIMP:
986 	case HV_X64_MSR_EOM:
987 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
988 		return synic_get_msr(vcpu_to_synic(vcpu), msr, pdata);
989 	case HV_X64_MSR_STIMER0_CONFIG:
990 	case HV_X64_MSR_STIMER1_CONFIG:
991 	case HV_X64_MSR_STIMER2_CONFIG:
992 	case HV_X64_MSR_STIMER3_CONFIG: {
993 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
994 
995 		return stimer_get_config(vcpu_to_stimer(vcpu, timer_index),
996 					 pdata);
997 	}
998 	case HV_X64_MSR_STIMER0_COUNT:
999 	case HV_X64_MSR_STIMER1_COUNT:
1000 	case HV_X64_MSR_STIMER2_COUNT:
1001 	case HV_X64_MSR_STIMER3_COUNT: {
1002 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1003 
1004 		return stimer_get_count(vcpu_to_stimer(vcpu, timer_index),
1005 					pdata);
1006 	}
1007 	default:
1008 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1009 		return 1;
1010 	}
1011 	*pdata = data;
1012 	return 0;
1013 }
1014 
1015 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1016 {
1017 	if (kvm_hv_msr_partition_wide(msr)) {
1018 		int r;
1019 
1020 		mutex_lock(&vcpu->kvm->lock);
1021 		r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1022 		mutex_unlock(&vcpu->kvm->lock);
1023 		return r;
1024 	} else
1025 		return kvm_hv_set_msr(vcpu, msr, data, host);
1026 }
1027 
1028 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1029 {
1030 	if (kvm_hv_msr_partition_wide(msr)) {
1031 		int r;
1032 
1033 		mutex_lock(&vcpu->kvm->lock);
1034 		r = kvm_hv_get_msr_pw(vcpu, msr, pdata);
1035 		mutex_unlock(&vcpu->kvm->lock);
1036 		return r;
1037 	} else
1038 		return kvm_hv_get_msr(vcpu, msr, pdata);
1039 }
1040 
1041 bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1042 {
1043 	return kvm->arch.hyperv.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1044 }
1045 
1046 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
1047 {
1048 	bool longmode;
1049 
1050 	longmode = is_64_bit_mode(vcpu);
1051 	if (longmode)
1052 		kvm_register_write(vcpu, VCPU_REGS_RAX, result);
1053 	else {
1054 		kvm_register_write(vcpu, VCPU_REGS_RDX, result >> 32);
1055 		kvm_register_write(vcpu, VCPU_REGS_RAX, result & 0xffffffff);
1056 	}
1057 }
1058 
1059 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
1060 {
1061 	struct kvm_run *run = vcpu->run;
1062 
1063 	kvm_hv_hypercall_set_result(vcpu, run->hyperv.u.hcall.result);
1064 	return 1;
1065 }
1066 
1067 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
1068 {
1069 	u64 param, ingpa, outgpa, ret;
1070 	uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
1071 	bool fast, longmode;
1072 
1073 	/*
1074 	 * hypercall generates UD from non zero cpl and real mode
1075 	 * per HYPER-V spec
1076 	 */
1077 	if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
1078 		kvm_queue_exception(vcpu, UD_VECTOR);
1079 		return 1;
1080 	}
1081 
1082 	longmode = is_64_bit_mode(vcpu);
1083 
1084 	if (!longmode) {
1085 		param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
1086 			(kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
1087 		ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
1088 			(kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
1089 		outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
1090 			(kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
1091 	}
1092 #ifdef CONFIG_X86_64
1093 	else {
1094 		param = kvm_register_read(vcpu, VCPU_REGS_RCX);
1095 		ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
1096 		outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
1097 	}
1098 #endif
1099 
1100 	code = param & 0xffff;
1101 	fast = (param >> 16) & 0x1;
1102 	rep_cnt = (param >> 32) & 0xfff;
1103 	rep_idx = (param >> 48) & 0xfff;
1104 
1105 	trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
1106 
1107 	/* Hypercall continuation is not supported yet */
1108 	if (rep_cnt || rep_idx) {
1109 		res = HV_STATUS_INVALID_HYPERCALL_CODE;
1110 		goto set_result;
1111 	}
1112 
1113 	switch (code) {
1114 	case HVCALL_NOTIFY_LONG_SPIN_WAIT:
1115 		kvm_vcpu_on_spin(vcpu);
1116 		break;
1117 	case HVCALL_POST_MESSAGE:
1118 	case HVCALL_SIGNAL_EVENT:
1119 		/* don't bother userspace if it has no way to handle it */
1120 		if (!vcpu_to_synic(vcpu)->active) {
1121 			res = HV_STATUS_INVALID_HYPERCALL_CODE;
1122 			break;
1123 		}
1124 		vcpu->run->exit_reason = KVM_EXIT_HYPERV;
1125 		vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
1126 		vcpu->run->hyperv.u.hcall.input = param;
1127 		vcpu->run->hyperv.u.hcall.params[0] = ingpa;
1128 		vcpu->run->hyperv.u.hcall.params[1] = outgpa;
1129 		vcpu->arch.complete_userspace_io =
1130 				kvm_hv_hypercall_complete_userspace;
1131 		return 0;
1132 	default:
1133 		res = HV_STATUS_INVALID_HYPERCALL_CODE;
1134 		break;
1135 	}
1136 
1137 set_result:
1138 	ret = res | (((u64)rep_done & 0xfff) << 32);
1139 	kvm_hv_hypercall_set_result(vcpu, ret);
1140 	return 1;
1141 }
1142