]> asedeno.scripts.mit.edu Git - linux.git/blob - arch/x86/kvm/hyperv.c
kvm: x86: hyperv: make VP_INDEX managed by userspace
[linux.git] / arch / x86 / kvm / hyperv.c
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 <linux/sched/cputime.h>
32
33 #include <asm/apicdef.h>
34 #include <trace/events/kvm.h>
35
36 #include "trace.h"
37
38 static inline u64 synic_read_sint(struct kvm_vcpu_hv_synic *synic, int sint)
39 {
40         return atomic64_read(&synic->sint[sint]);
41 }
42
43 static inline int synic_get_sint_vector(u64 sint_value)
44 {
45         if (sint_value & HV_SYNIC_SINT_MASKED)
46                 return -1;
47         return sint_value & HV_SYNIC_SINT_VECTOR_MASK;
48 }
49
50 static bool synic_has_vector_connected(struct kvm_vcpu_hv_synic *synic,
51                                       int vector)
52 {
53         int i;
54
55         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
56                 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
57                         return true;
58         }
59         return false;
60 }
61
62 static bool synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic *synic,
63                                      int vector)
64 {
65         int i;
66         u64 sint_value;
67
68         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
69                 sint_value = synic_read_sint(synic, i);
70                 if (synic_get_sint_vector(sint_value) == vector &&
71                     sint_value & HV_SYNIC_SINT_AUTO_EOI)
72                         return true;
73         }
74         return false;
75 }
76
77 static int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint,
78                           u64 data, bool host)
79 {
80         int vector;
81
82         vector = data & HV_SYNIC_SINT_VECTOR_MASK;
83         if (vector < 16 && !host)
84                 return 1;
85         /*
86          * Guest may configure multiple SINTs to use the same vector, so
87          * we maintain a bitmap of vectors handled by synic, and a
88          * bitmap of vectors with auto-eoi behavior.  The bitmaps are
89          * updated here, and atomically queried on fast paths.
90          */
91
92         atomic64_set(&synic->sint[sint], data);
93
94         if (synic_has_vector_connected(synic, vector))
95                 __set_bit(vector, synic->vec_bitmap);
96         else
97                 __clear_bit(vector, synic->vec_bitmap);
98
99         if (synic_has_vector_auto_eoi(synic, vector))
100                 __set_bit(vector, synic->auto_eoi_bitmap);
101         else
102                 __clear_bit(vector, synic->auto_eoi_bitmap);
103
104         /* Load SynIC vectors into EOI exit bitmap */
105         kvm_make_request(KVM_REQ_SCAN_IOAPIC, synic_to_vcpu(synic));
106         return 0;
107 }
108
109 static struct kvm_vcpu *get_vcpu_by_vpidx(struct kvm *kvm, u32 vpidx)
110 {
111         struct kvm_vcpu *vcpu = NULL;
112         int i;
113
114         if (vpidx < KVM_MAX_VCPUS)
115                 vcpu = kvm_get_vcpu(kvm, vpidx);
116         if (vcpu && vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
117                 return vcpu;
118         kvm_for_each_vcpu(i, vcpu, kvm)
119                 if (vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
120                         return vcpu;
121         return NULL;
122 }
123
124 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vpidx)
125 {
126         struct kvm_vcpu *vcpu;
127         struct kvm_vcpu_hv_synic *synic;
128
129         vcpu = get_vcpu_by_vpidx(kvm, vpidx);
130         if (!vcpu)
131                 return NULL;
132         synic = vcpu_to_synic(vcpu);
133         return (synic->active) ? synic : NULL;
134 }
135
136 static void synic_clear_sint_msg_pending(struct kvm_vcpu_hv_synic *synic,
137                                         u32 sint)
138 {
139         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
140         struct page *page;
141         gpa_t gpa;
142         struct hv_message *msg;
143         struct hv_message_page *msg_page;
144
145         gpa = synic->msg_page & PAGE_MASK;
146         page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
147         if (is_error_page(page)) {
148                 vcpu_err(vcpu, "Hyper-V SynIC can't get msg page, gpa 0x%llx\n",
149                          gpa);
150                 return;
151         }
152         msg_page = kmap_atomic(page);
153
154         msg = &msg_page->sint_message[sint];
155         msg->header.message_flags.msg_pending = 0;
156
157         kunmap_atomic(msg_page);
158         kvm_release_page_dirty(page);
159         kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
160 }
161
162 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
163 {
164         struct kvm *kvm = vcpu->kvm;
165         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
166         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
167         struct kvm_vcpu_hv_stimer *stimer;
168         int gsi, idx, stimers_pending;
169
170         trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
171
172         if (synic->msg_page & HV_SYNIC_SIMP_ENABLE)
173                 synic_clear_sint_msg_pending(synic, sint);
174
175         /* Try to deliver pending Hyper-V SynIC timers messages */
176         stimers_pending = 0;
177         for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
178                 stimer = &hv_vcpu->stimer[idx];
179                 if (stimer->msg_pending &&
180                     (stimer->config & HV_STIMER_ENABLE) &&
181                     HV_STIMER_SINT(stimer->config) == sint) {
182                         set_bit(stimer->index,
183                                 hv_vcpu->stimer_pending_bitmap);
184                         stimers_pending++;
185                 }
186         }
187         if (stimers_pending)
188                 kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
189
190         idx = srcu_read_lock(&kvm->irq_srcu);
191         gsi = atomic_read(&synic->sint_to_gsi[sint]);
192         if (gsi != -1)
193                 kvm_notify_acked_gsi(kvm, gsi);
194         srcu_read_unlock(&kvm->irq_srcu, idx);
195 }
196
197 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
198 {
199         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
200         struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
201
202         hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
203         hv_vcpu->exit.u.synic.msr = msr;
204         hv_vcpu->exit.u.synic.control = synic->control;
205         hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
206         hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
207
208         kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
209 }
210
211 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
212                          u32 msr, u64 data, bool host)
213 {
214         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
215         int ret;
216
217         if (!synic->active)
218                 return 1;
219
220         trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
221
222         ret = 0;
223         switch (msr) {
224         case HV_X64_MSR_SCONTROL:
225                 synic->control = data;
226                 if (!host)
227                         synic_exit(synic, msr);
228                 break;
229         case HV_X64_MSR_SVERSION:
230                 if (!host) {
231                         ret = 1;
232                         break;
233                 }
234                 synic->version = data;
235                 break;
236         case HV_X64_MSR_SIEFP:
237                 if ((data & HV_SYNIC_SIEFP_ENABLE) && !host &&
238                     !synic->dont_zero_synic_pages)
239                         if (kvm_clear_guest(vcpu->kvm,
240                                             data & PAGE_MASK, PAGE_SIZE)) {
241                                 ret = 1;
242                                 break;
243                         }
244                 synic->evt_page = data;
245                 if (!host)
246                         synic_exit(synic, msr);
247                 break;
248         case HV_X64_MSR_SIMP:
249                 if ((data & HV_SYNIC_SIMP_ENABLE) && !host &&
250                     !synic->dont_zero_synic_pages)
251                         if (kvm_clear_guest(vcpu->kvm,
252                                             data & PAGE_MASK, PAGE_SIZE)) {
253                                 ret = 1;
254                                 break;
255                         }
256                 synic->msg_page = data;
257                 if (!host)
258                         synic_exit(synic, msr);
259                 break;
260         case HV_X64_MSR_EOM: {
261                 int i;
262
263                 for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
264                         kvm_hv_notify_acked_sint(vcpu, i);
265                 break;
266         }
267         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
268                 ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
269                 break;
270         default:
271                 ret = 1;
272                 break;
273         }
274         return ret;
275 }
276
277 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata)
278 {
279         int ret;
280
281         if (!synic->active)
282                 return 1;
283
284         ret = 0;
285         switch (msr) {
286         case HV_X64_MSR_SCONTROL:
287                 *pdata = synic->control;
288                 break;
289         case HV_X64_MSR_SVERSION:
290                 *pdata = synic->version;
291                 break;
292         case HV_X64_MSR_SIEFP:
293                 *pdata = synic->evt_page;
294                 break;
295         case HV_X64_MSR_SIMP:
296                 *pdata = synic->msg_page;
297                 break;
298         case HV_X64_MSR_EOM:
299                 *pdata = 0;
300                 break;
301         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
302                 *pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
303                 break;
304         default:
305                 ret = 1;
306                 break;
307         }
308         return ret;
309 }
310
311 static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
312 {
313         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
314         struct kvm_lapic_irq irq;
315         int ret, vector;
316
317         if (sint >= ARRAY_SIZE(synic->sint))
318                 return -EINVAL;
319
320         vector = synic_get_sint_vector(synic_read_sint(synic, sint));
321         if (vector < 0)
322                 return -ENOENT;
323
324         memset(&irq, 0, sizeof(irq));
325         irq.shorthand = APIC_DEST_SELF;
326         irq.dest_mode = APIC_DEST_PHYSICAL;
327         irq.delivery_mode = APIC_DM_FIXED;
328         irq.vector = vector;
329         irq.level = 1;
330
331         ret = kvm_irq_delivery_to_apic(vcpu->kvm, vcpu->arch.apic, &irq, NULL);
332         trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
333         return ret;
334 }
335
336 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vpidx, u32 sint)
337 {
338         struct kvm_vcpu_hv_synic *synic;
339
340         synic = synic_get(kvm, vpidx);
341         if (!synic)
342                 return -EINVAL;
343
344         return synic_set_irq(synic, sint);
345 }
346
347 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
348 {
349         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
350         int i;
351
352         trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
353
354         for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
355                 if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
356                         kvm_hv_notify_acked_sint(vcpu, i);
357 }
358
359 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi)
360 {
361         struct kvm_vcpu_hv_synic *synic;
362
363         synic = synic_get(kvm, vpidx);
364         if (!synic)
365                 return -EINVAL;
366
367         if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
368                 return -EINVAL;
369
370         atomic_set(&synic->sint_to_gsi[sint], gsi);
371         return 0;
372 }
373
374 void kvm_hv_irq_routing_update(struct kvm *kvm)
375 {
376         struct kvm_irq_routing_table *irq_rt;
377         struct kvm_kernel_irq_routing_entry *e;
378         u32 gsi;
379
380         irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
381                                         lockdep_is_held(&kvm->irq_lock));
382
383         for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
384                 hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
385                         if (e->type == KVM_IRQ_ROUTING_HV_SINT)
386                                 kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
387                                                     e->hv_sint.sint, gsi);
388                 }
389         }
390 }
391
392 static void synic_init(struct kvm_vcpu_hv_synic *synic)
393 {
394         int i;
395
396         memset(synic, 0, sizeof(*synic));
397         synic->version = HV_SYNIC_VERSION_1;
398         for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
399                 atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
400                 atomic_set(&synic->sint_to_gsi[i], -1);
401         }
402 }
403
404 static u64 get_time_ref_counter(struct kvm *kvm)
405 {
406         struct kvm_hv *hv = &kvm->arch.hyperv;
407         struct kvm_vcpu *vcpu;
408         u64 tsc;
409
410         /*
411          * The guest has not set up the TSC page or the clock isn't
412          * stable, fall back to get_kvmclock_ns.
413          */
414         if (!hv->tsc_ref.tsc_sequence)
415                 return div_u64(get_kvmclock_ns(kvm), 100);
416
417         vcpu = kvm_get_vcpu(kvm, 0);
418         tsc = kvm_read_l1_tsc(vcpu, rdtsc());
419         return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64)
420                 + hv->tsc_ref.tsc_offset;
421 }
422
423 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
424                                 bool vcpu_kick)
425 {
426         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
427
428         set_bit(stimer->index,
429                 vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
430         kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
431         if (vcpu_kick)
432                 kvm_vcpu_kick(vcpu);
433 }
434
435 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
436 {
437         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
438
439         trace_kvm_hv_stimer_cleanup(stimer_to_vcpu(stimer)->vcpu_id,
440                                     stimer->index);
441
442         hrtimer_cancel(&stimer->timer);
443         clear_bit(stimer->index,
444                   vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
445         stimer->msg_pending = false;
446         stimer->exp_time = 0;
447 }
448
449 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
450 {
451         struct kvm_vcpu_hv_stimer *stimer;
452
453         stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
454         trace_kvm_hv_stimer_callback(stimer_to_vcpu(stimer)->vcpu_id,
455                                      stimer->index);
456         stimer_mark_pending(stimer, true);
457
458         return HRTIMER_NORESTART;
459 }
460
461 /*
462  * stimer_start() assumptions:
463  * a) stimer->count is not equal to 0
464  * b) stimer->config has HV_STIMER_ENABLE flag
465  */
466 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
467 {
468         u64 time_now;
469         ktime_t ktime_now;
470
471         time_now = get_time_ref_counter(stimer_to_vcpu(stimer)->kvm);
472         ktime_now = ktime_get();
473
474         if (stimer->config & HV_STIMER_PERIODIC) {
475                 if (stimer->exp_time) {
476                         if (time_now >= stimer->exp_time) {
477                                 u64 remainder;
478
479                                 div64_u64_rem(time_now - stimer->exp_time,
480                                               stimer->count, &remainder);
481                                 stimer->exp_time =
482                                         time_now + (stimer->count - remainder);
483                         }
484                 } else
485                         stimer->exp_time = time_now + stimer->count;
486
487                 trace_kvm_hv_stimer_start_periodic(
488                                         stimer_to_vcpu(stimer)->vcpu_id,
489                                         stimer->index,
490                                         time_now, stimer->exp_time);
491
492                 hrtimer_start(&stimer->timer,
493                               ktime_add_ns(ktime_now,
494                                            100 * (stimer->exp_time - time_now)),
495                               HRTIMER_MODE_ABS);
496                 return 0;
497         }
498         stimer->exp_time = stimer->count;
499         if (time_now >= stimer->count) {
500                 /*
501                  * Expire timer according to Hypervisor Top-Level Functional
502                  * specification v4(15.3.1):
503                  * "If a one shot is enabled and the specified count is in
504                  * the past, it will expire immediately."
505                  */
506                 stimer_mark_pending(stimer, false);
507                 return 0;
508         }
509
510         trace_kvm_hv_stimer_start_one_shot(stimer_to_vcpu(stimer)->vcpu_id,
511                                            stimer->index,
512                                            time_now, stimer->count);
513
514         hrtimer_start(&stimer->timer,
515                       ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)),
516                       HRTIMER_MODE_ABS);
517         return 0;
518 }
519
520 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
521                              bool host)
522 {
523         trace_kvm_hv_stimer_set_config(stimer_to_vcpu(stimer)->vcpu_id,
524                                        stimer->index, config, host);
525
526         stimer_cleanup(stimer);
527         if ((stimer->config & HV_STIMER_ENABLE) && HV_STIMER_SINT(config) == 0)
528                 config &= ~HV_STIMER_ENABLE;
529         stimer->config = config;
530         stimer_mark_pending(stimer, false);
531         return 0;
532 }
533
534 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
535                             bool host)
536 {
537         trace_kvm_hv_stimer_set_count(stimer_to_vcpu(stimer)->vcpu_id,
538                                       stimer->index, count, host);
539
540         stimer_cleanup(stimer);
541         stimer->count = count;
542         if (stimer->count == 0)
543                 stimer->config &= ~HV_STIMER_ENABLE;
544         else if (stimer->config & HV_STIMER_AUTOENABLE)
545                 stimer->config |= HV_STIMER_ENABLE;
546         stimer_mark_pending(stimer, false);
547         return 0;
548 }
549
550 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
551 {
552         *pconfig = stimer->config;
553         return 0;
554 }
555
556 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
557 {
558         *pcount = stimer->count;
559         return 0;
560 }
561
562 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
563                              struct hv_message *src_msg)
564 {
565         struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
566         struct page *page;
567         gpa_t gpa;
568         struct hv_message *dst_msg;
569         int r;
570         struct hv_message_page *msg_page;
571
572         if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
573                 return -ENOENT;
574
575         gpa = synic->msg_page & PAGE_MASK;
576         page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
577         if (is_error_page(page))
578                 return -EFAULT;
579
580         msg_page = kmap_atomic(page);
581         dst_msg = &msg_page->sint_message[sint];
582         if (sync_cmpxchg(&dst_msg->header.message_type, HVMSG_NONE,
583                          src_msg->header.message_type) != HVMSG_NONE) {
584                 dst_msg->header.message_flags.msg_pending = 1;
585                 r = -EAGAIN;
586         } else {
587                 memcpy(&dst_msg->u.payload, &src_msg->u.payload,
588                        src_msg->header.payload_size);
589                 dst_msg->header.message_type = src_msg->header.message_type;
590                 dst_msg->header.payload_size = src_msg->header.payload_size;
591                 r = synic_set_irq(synic, sint);
592                 if (r >= 1)
593                         r = 0;
594                 else if (r == 0)
595                         r = -EFAULT;
596         }
597         kunmap_atomic(msg_page);
598         kvm_release_page_dirty(page);
599         kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
600         return r;
601 }
602
603 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
604 {
605         struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
606         struct hv_message *msg = &stimer->msg;
607         struct hv_timer_message_payload *payload =
608                         (struct hv_timer_message_payload *)&msg->u.payload;
609
610         payload->expiration_time = stimer->exp_time;
611         payload->delivery_time = get_time_ref_counter(vcpu->kvm);
612         return synic_deliver_msg(vcpu_to_synic(vcpu),
613                                  HV_STIMER_SINT(stimer->config), msg);
614 }
615
616 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
617 {
618         int r;
619
620         stimer->msg_pending = true;
621         r = stimer_send_msg(stimer);
622         trace_kvm_hv_stimer_expiration(stimer_to_vcpu(stimer)->vcpu_id,
623                                        stimer->index, r);
624         if (!r) {
625                 stimer->msg_pending = false;
626                 if (!(stimer->config & HV_STIMER_PERIODIC))
627                         stimer->config &= ~HV_STIMER_ENABLE;
628         }
629 }
630
631 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
632 {
633         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
634         struct kvm_vcpu_hv_stimer *stimer;
635         u64 time_now, exp_time;
636         int i;
637
638         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
639                 if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
640                         stimer = &hv_vcpu->stimer[i];
641                         if (stimer->config & HV_STIMER_ENABLE) {
642                                 exp_time = stimer->exp_time;
643
644                                 if (exp_time) {
645                                         time_now =
646                                                 get_time_ref_counter(vcpu->kvm);
647                                         if (time_now >= exp_time)
648                                                 stimer_expiration(stimer);
649                                 }
650
651                                 if ((stimer->config & HV_STIMER_ENABLE) &&
652                                     stimer->count)
653                                         stimer_start(stimer);
654                                 else
655                                         stimer_cleanup(stimer);
656                         }
657                 }
658 }
659
660 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
661 {
662         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
663         int i;
664
665         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
666                 stimer_cleanup(&hv_vcpu->stimer[i]);
667 }
668
669 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
670 {
671         struct hv_message *msg = &stimer->msg;
672         struct hv_timer_message_payload *payload =
673                         (struct hv_timer_message_payload *)&msg->u.payload;
674
675         memset(&msg->header, 0, sizeof(msg->header));
676         msg->header.message_type = HVMSG_TIMER_EXPIRED;
677         msg->header.payload_size = sizeof(*payload);
678
679         payload->timer_index = stimer->index;
680         payload->expiration_time = 0;
681         payload->delivery_time = 0;
682 }
683
684 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
685 {
686         memset(stimer, 0, sizeof(*stimer));
687         stimer->index = timer_index;
688         hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
689         stimer->timer.function = stimer_timer_callback;
690         stimer_prepare_msg(stimer);
691 }
692
693 void kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
694 {
695         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
696         int i;
697
698         synic_init(&hv_vcpu->synic);
699
700         bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
701         for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
702                 stimer_init(&hv_vcpu->stimer[i], i);
703 }
704
705 void kvm_hv_vcpu_postcreate(struct kvm_vcpu *vcpu)
706 {
707         struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
708
709         hv_vcpu->vp_index = kvm_vcpu_get_idx(vcpu);
710 }
711
712 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
713 {
714         struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
715
716         /*
717          * Hyper-V SynIC auto EOI SINT's are
718          * not compatible with APICV, so deactivate APICV
719          */
720         kvm_vcpu_deactivate_apicv(vcpu);
721         synic->active = true;
722         synic->dont_zero_synic_pages = dont_zero_synic_pages;
723         return 0;
724 }
725
726 static bool kvm_hv_msr_partition_wide(u32 msr)
727 {
728         bool r = false;
729
730         switch (msr) {
731         case HV_X64_MSR_GUEST_OS_ID:
732         case HV_X64_MSR_HYPERCALL:
733         case HV_X64_MSR_REFERENCE_TSC:
734         case HV_X64_MSR_TIME_REF_COUNT:
735         case HV_X64_MSR_CRASH_CTL:
736         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
737         case HV_X64_MSR_RESET:
738                 r = true;
739                 break;
740         }
741
742         return r;
743 }
744
745 static int kvm_hv_msr_get_crash_data(struct kvm_vcpu *vcpu,
746                                      u32 index, u64 *pdata)
747 {
748         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
749
750         if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param)))
751                 return -EINVAL;
752
753         *pdata = hv->hv_crash_param[index];
754         return 0;
755 }
756
757 static int kvm_hv_msr_get_crash_ctl(struct kvm_vcpu *vcpu, u64 *pdata)
758 {
759         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
760
761         *pdata = hv->hv_crash_ctl;
762         return 0;
763 }
764
765 static int kvm_hv_msr_set_crash_ctl(struct kvm_vcpu *vcpu, u64 data, bool host)
766 {
767         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
768
769         if (host)
770                 hv->hv_crash_ctl = data & HV_X64_MSR_CRASH_CTL_NOTIFY;
771
772         if (!host && (data & HV_X64_MSR_CRASH_CTL_NOTIFY)) {
773
774                 vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
775                           hv->hv_crash_param[0],
776                           hv->hv_crash_param[1],
777                           hv->hv_crash_param[2],
778                           hv->hv_crash_param[3],
779                           hv->hv_crash_param[4]);
780
781                 /* Send notification about crash to user space */
782                 kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
783         }
784
785         return 0;
786 }
787
788 static int kvm_hv_msr_set_crash_data(struct kvm_vcpu *vcpu,
789                                      u32 index, u64 data)
790 {
791         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
792
793         if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param)))
794                 return -EINVAL;
795
796         hv->hv_crash_param[index] = data;
797         return 0;
798 }
799
800 /*
801  * The kvmclock and Hyper-V TSC page use similar formulas, and converting
802  * between them is possible:
803  *
804  * kvmclock formula:
805  *    nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
806  *           + system_time
807  *
808  * Hyper-V formula:
809  *    nsec/100 = ticks * scale / 2^64 + offset
810  *
811  * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
812  * By dividing the kvmclock formula by 100 and equating what's left we get:
813  *    ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
814  *            scale / 2^64 =         tsc_to_system_mul * 2^(tsc_shift-32) / 100
815  *            scale        =         tsc_to_system_mul * 2^(32+tsc_shift) / 100
816  *
817  * Now expand the kvmclock formula and divide by 100:
818  *    nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
819  *           - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
820  *           + system_time
821  *    nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
822  *               - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
823  *               + system_time / 100
824  *
825  * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
826  *    nsec/100 = ticks * scale / 2^64
827  *               - tsc_timestamp * scale / 2^64
828  *               + system_time / 100
829  *
830  * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
831  *    offset = system_time / 100 - tsc_timestamp * scale / 2^64
832  *
833  * These two equivalencies are implemented in this function.
834  */
835 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
836                                         HV_REFERENCE_TSC_PAGE *tsc_ref)
837 {
838         u64 max_mul;
839
840         if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
841                 return false;
842
843         /*
844          * check if scale would overflow, if so we use the time ref counter
845          *    tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
846          *    tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
847          *    tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
848          */
849         max_mul = 100ull << (32 - hv_clock->tsc_shift);
850         if (hv_clock->tsc_to_system_mul >= max_mul)
851                 return false;
852
853         /*
854          * Otherwise compute the scale and offset according to the formulas
855          * derived above.
856          */
857         tsc_ref->tsc_scale =
858                 mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
859                                 hv_clock->tsc_to_system_mul,
860                                 100);
861
862         tsc_ref->tsc_offset = hv_clock->system_time;
863         do_div(tsc_ref->tsc_offset, 100);
864         tsc_ref->tsc_offset -=
865                 mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
866         return true;
867 }
868
869 void kvm_hv_setup_tsc_page(struct kvm *kvm,
870                            struct pvclock_vcpu_time_info *hv_clock)
871 {
872         struct kvm_hv *hv = &kvm->arch.hyperv;
873         u32 tsc_seq;
874         u64 gfn;
875
876         BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
877         BUILD_BUG_ON(offsetof(HV_REFERENCE_TSC_PAGE, tsc_sequence) != 0);
878
879         if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
880                 return;
881
882         mutex_lock(&kvm->arch.hyperv.hv_lock);
883         if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
884                 goto out_unlock;
885
886         gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
887         /*
888          * Because the TSC parameters only vary when there is a
889          * change in the master clock, do not bother with caching.
890          */
891         if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
892                                     &tsc_seq, sizeof(tsc_seq))))
893                 goto out_unlock;
894
895         /*
896          * While we're computing and writing the parameters, force the
897          * guest to use the time reference count MSR.
898          */
899         hv->tsc_ref.tsc_sequence = 0;
900         if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
901                             &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
902                 goto out_unlock;
903
904         if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
905                 goto out_unlock;
906
907         /* Ensure sequence is zero before writing the rest of the struct.  */
908         smp_wmb();
909         if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
910                 goto out_unlock;
911
912         /*
913          * Now switch to the TSC page mechanism by writing the sequence.
914          */
915         tsc_seq++;
916         if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
917                 tsc_seq = 1;
918
919         /* Write the struct entirely before the non-zero sequence.  */
920         smp_wmb();
921
922         hv->tsc_ref.tsc_sequence = tsc_seq;
923         kvm_write_guest(kvm, gfn_to_gpa(gfn),
924                         &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence));
925 out_unlock:
926         mutex_unlock(&kvm->arch.hyperv.hv_lock);
927 }
928
929 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
930                              bool host)
931 {
932         struct kvm *kvm = vcpu->kvm;
933         struct kvm_hv *hv = &kvm->arch.hyperv;
934
935         switch (msr) {
936         case HV_X64_MSR_GUEST_OS_ID:
937                 hv->hv_guest_os_id = data;
938                 /* setting guest os id to zero disables hypercall page */
939                 if (!hv->hv_guest_os_id)
940                         hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
941                 break;
942         case HV_X64_MSR_HYPERCALL: {
943                 u64 gfn;
944                 unsigned long addr;
945                 u8 instructions[4];
946
947                 /* if guest os id is not set hypercall should remain disabled */
948                 if (!hv->hv_guest_os_id)
949                         break;
950                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
951                         hv->hv_hypercall = data;
952                         break;
953                 }
954                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
955                 addr = gfn_to_hva(kvm, gfn);
956                 if (kvm_is_error_hva(addr))
957                         return 1;
958                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
959                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
960                 if (__copy_to_user((void __user *)addr, instructions, 4))
961                         return 1;
962                 hv->hv_hypercall = data;
963                 mark_page_dirty(kvm, gfn);
964                 break;
965         }
966         case HV_X64_MSR_REFERENCE_TSC:
967                 hv->hv_tsc_page = data;
968                 if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE)
969                         kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
970                 break;
971         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
972                 return kvm_hv_msr_set_crash_data(vcpu,
973                                                  msr - HV_X64_MSR_CRASH_P0,
974                                                  data);
975         case HV_X64_MSR_CRASH_CTL:
976                 return kvm_hv_msr_set_crash_ctl(vcpu, data, host);
977         case HV_X64_MSR_RESET:
978                 if (data == 1) {
979                         vcpu_debug(vcpu, "hyper-v reset requested\n");
980                         kvm_make_request(KVM_REQ_HV_RESET, vcpu);
981                 }
982                 break;
983         default:
984                 vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
985                             msr, data);
986                 return 1;
987         }
988         return 0;
989 }
990
991 /* Calculate cpu time spent by current task in 100ns units */
992 static u64 current_task_runtime_100ns(void)
993 {
994         u64 utime, stime;
995
996         task_cputime_adjusted(current, &utime, &stime);
997
998         return div_u64(utime + stime, 100);
999 }
1000
1001 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1002 {
1003         struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
1004
1005         switch (msr) {
1006         case HV_X64_MSR_VP_INDEX:
1007                 if (!host)
1008                         return 1;
1009                 hv->vp_index = (u32)data;
1010                 break;
1011         case HV_X64_MSR_APIC_ASSIST_PAGE: {
1012                 u64 gfn;
1013                 unsigned long addr;
1014
1015                 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1016                         hv->hv_vapic = data;
1017                         if (kvm_lapic_enable_pv_eoi(vcpu, 0))
1018                                 return 1;
1019                         break;
1020                 }
1021                 gfn = data >> HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT;
1022                 addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1023                 if (kvm_is_error_hva(addr))
1024                         return 1;
1025                 if (__clear_user((void __user *)addr, PAGE_SIZE))
1026                         return 1;
1027                 hv->hv_vapic = data;
1028                 kvm_vcpu_mark_page_dirty(vcpu, gfn);
1029                 if (kvm_lapic_enable_pv_eoi(vcpu,
1030                                             gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
1031                         return 1;
1032                 break;
1033         }
1034         case HV_X64_MSR_EOI:
1035                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1036         case HV_X64_MSR_ICR:
1037                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1038         case HV_X64_MSR_TPR:
1039                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1040         case HV_X64_MSR_VP_RUNTIME:
1041                 if (!host)
1042                         return 1;
1043                 hv->runtime_offset = data - current_task_runtime_100ns();
1044                 break;
1045         case HV_X64_MSR_SCONTROL:
1046         case HV_X64_MSR_SVERSION:
1047         case HV_X64_MSR_SIEFP:
1048         case HV_X64_MSR_SIMP:
1049         case HV_X64_MSR_EOM:
1050         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1051                 return synic_set_msr(vcpu_to_synic(vcpu), msr, data, host);
1052         case HV_X64_MSR_STIMER0_CONFIG:
1053         case HV_X64_MSR_STIMER1_CONFIG:
1054         case HV_X64_MSR_STIMER2_CONFIG:
1055         case HV_X64_MSR_STIMER3_CONFIG: {
1056                 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1057
1058                 return stimer_set_config(vcpu_to_stimer(vcpu, timer_index),
1059                                          data, host);
1060         }
1061         case HV_X64_MSR_STIMER0_COUNT:
1062         case HV_X64_MSR_STIMER1_COUNT:
1063         case HV_X64_MSR_STIMER2_COUNT:
1064         case HV_X64_MSR_STIMER3_COUNT: {
1065                 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1066
1067                 return stimer_set_count(vcpu_to_stimer(vcpu, timer_index),
1068                                         data, host);
1069         }
1070         default:
1071                 vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
1072                             msr, data);
1073                 return 1;
1074         }
1075
1076         return 0;
1077 }
1078
1079 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1080 {
1081         u64 data = 0;
1082         struct kvm *kvm = vcpu->kvm;
1083         struct kvm_hv *hv = &kvm->arch.hyperv;
1084
1085         switch (msr) {
1086         case HV_X64_MSR_GUEST_OS_ID:
1087                 data = hv->hv_guest_os_id;
1088                 break;
1089         case HV_X64_MSR_HYPERCALL:
1090                 data = hv->hv_hypercall;
1091                 break;
1092         case HV_X64_MSR_TIME_REF_COUNT:
1093                 data = get_time_ref_counter(kvm);
1094                 break;
1095         case HV_X64_MSR_REFERENCE_TSC:
1096                 data = hv->hv_tsc_page;
1097                 break;
1098         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1099                 return kvm_hv_msr_get_crash_data(vcpu,
1100                                                  msr - HV_X64_MSR_CRASH_P0,
1101                                                  pdata);
1102         case HV_X64_MSR_CRASH_CTL:
1103                 return kvm_hv_msr_get_crash_ctl(vcpu, pdata);
1104         case HV_X64_MSR_RESET:
1105                 data = 0;
1106                 break;
1107         default:
1108                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1109                 return 1;
1110         }
1111
1112         *pdata = data;
1113         return 0;
1114 }
1115
1116 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1117 {
1118         u64 data = 0;
1119         struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
1120
1121         switch (msr) {
1122         case HV_X64_MSR_VP_INDEX:
1123                 data = hv->vp_index;
1124                 break;
1125         case HV_X64_MSR_EOI:
1126                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1127         case HV_X64_MSR_ICR:
1128                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1129         case HV_X64_MSR_TPR:
1130                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1131         case HV_X64_MSR_APIC_ASSIST_PAGE:
1132                 data = hv->hv_vapic;
1133                 break;
1134         case HV_X64_MSR_VP_RUNTIME:
1135                 data = current_task_runtime_100ns() + hv->runtime_offset;
1136                 break;
1137         case HV_X64_MSR_SCONTROL:
1138         case HV_X64_MSR_SVERSION:
1139         case HV_X64_MSR_SIEFP:
1140         case HV_X64_MSR_SIMP:
1141         case HV_X64_MSR_EOM:
1142         case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1143                 return synic_get_msr(vcpu_to_synic(vcpu), msr, pdata);
1144         case HV_X64_MSR_STIMER0_CONFIG:
1145         case HV_X64_MSR_STIMER1_CONFIG:
1146         case HV_X64_MSR_STIMER2_CONFIG:
1147         case HV_X64_MSR_STIMER3_CONFIG: {
1148                 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1149
1150                 return stimer_get_config(vcpu_to_stimer(vcpu, timer_index),
1151                                          pdata);
1152         }
1153         case HV_X64_MSR_STIMER0_COUNT:
1154         case HV_X64_MSR_STIMER1_COUNT:
1155         case HV_X64_MSR_STIMER2_COUNT:
1156         case HV_X64_MSR_STIMER3_COUNT: {
1157                 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1158
1159                 return stimer_get_count(vcpu_to_stimer(vcpu, timer_index),
1160                                         pdata);
1161         }
1162         default:
1163                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1164                 return 1;
1165         }
1166         *pdata = data;
1167         return 0;
1168 }
1169
1170 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1171 {
1172         if (kvm_hv_msr_partition_wide(msr)) {
1173                 int r;
1174
1175                 mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1176                 r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1177                 mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1178                 return r;
1179         } else
1180                 return kvm_hv_set_msr(vcpu, msr, data, host);
1181 }
1182
1183 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1184 {
1185         if (kvm_hv_msr_partition_wide(msr)) {
1186                 int r;
1187
1188                 mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1189                 r = kvm_hv_get_msr_pw(vcpu, msr, pdata);
1190                 mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1191                 return r;
1192         } else
1193                 return kvm_hv_get_msr(vcpu, msr, pdata);
1194 }
1195
1196 bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1197 {
1198         return READ_ONCE(kvm->arch.hyperv.hv_hypercall) & HV_X64_MSR_HYPERCALL_ENABLE;
1199 }
1200
1201 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
1202 {
1203         bool longmode;
1204
1205         longmode = is_64_bit_mode(vcpu);
1206         if (longmode)
1207                 kvm_register_write(vcpu, VCPU_REGS_RAX, result);
1208         else {
1209                 kvm_register_write(vcpu, VCPU_REGS_RDX, result >> 32);
1210                 kvm_register_write(vcpu, VCPU_REGS_RAX, result & 0xffffffff);
1211         }
1212 }
1213
1214 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
1215 {
1216         struct kvm_run *run = vcpu->run;
1217
1218         kvm_hv_hypercall_set_result(vcpu, run->hyperv.u.hcall.result);
1219         return 1;
1220 }
1221
1222 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
1223 {
1224         u64 param, ingpa, outgpa, ret;
1225         uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
1226         bool fast, longmode;
1227
1228         /*
1229          * hypercall generates UD from non zero cpl and real mode
1230          * per HYPER-V spec
1231          */
1232         if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
1233                 kvm_queue_exception(vcpu, UD_VECTOR);
1234                 return 1;
1235         }
1236
1237         longmode = is_64_bit_mode(vcpu);
1238
1239         if (!longmode) {
1240                 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
1241                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
1242                 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
1243                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
1244                 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
1245                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
1246         }
1247 #ifdef CONFIG_X86_64
1248         else {
1249                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
1250                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
1251                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
1252         }
1253 #endif
1254
1255         code = param & 0xffff;
1256         fast = (param >> 16) & 0x1;
1257         rep_cnt = (param >> 32) & 0xfff;
1258         rep_idx = (param >> 48) & 0xfff;
1259
1260         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
1261
1262         /* Hypercall continuation is not supported yet */
1263         if (rep_cnt || rep_idx) {
1264                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
1265                 goto set_result;
1266         }
1267
1268         switch (code) {
1269         case HVCALL_NOTIFY_LONG_SPIN_WAIT:
1270                 kvm_vcpu_on_spin(vcpu);
1271                 break;
1272         case HVCALL_POST_MESSAGE:
1273         case HVCALL_SIGNAL_EVENT:
1274                 /* don't bother userspace if it has no way to handle it */
1275                 if (!vcpu_to_synic(vcpu)->active) {
1276                         res = HV_STATUS_INVALID_HYPERCALL_CODE;
1277                         break;
1278                 }
1279                 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
1280                 vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
1281                 vcpu->run->hyperv.u.hcall.input = param;
1282                 vcpu->run->hyperv.u.hcall.params[0] = ingpa;
1283                 vcpu->run->hyperv.u.hcall.params[1] = outgpa;
1284                 vcpu->arch.complete_userspace_io =
1285                                 kvm_hv_hypercall_complete_userspace;
1286                 return 0;
1287         default:
1288                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
1289                 break;
1290         }
1291
1292 set_result:
1293         ret = res | (((u64)rep_done & 0xfff) << 32);
1294         kvm_hv_hypercall_set_result(vcpu, ret);
1295         return 1;
1296 }