?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
mm: fix exec activate_mm vs TLB shootdown and lazy tlb switching race
[linux/fpc-iii.git] / arch / x86 / kvm / hyperv.c
blob2fba82b06c2d5d0448705098bb3806b938b84035
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 if (!stimer->msg_pending)
654 stimer_start(stimer);
655 } else
656 stimer_cleanup(stimer);
657 }
658 }
659 }
660
661 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
662 {
663 struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
664 int i;
665
666 for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
667 stimer_cleanup(&hv_vcpu->stimer[i]);
668 }
669
670 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
671 {
672 struct hv_message *msg = &stimer->msg;
673 struct hv_timer_message_payload *payload =
674 (struct hv_timer_message_payload *)&msg->u.payload;
675
676 memset(&msg->header, 0, sizeof(msg->header));
677 msg->header.message_type = HVMSG_TIMER_EXPIRED;
678 msg->header.payload_size = sizeof(*payload);
679
680 payload->timer_index = stimer->index;
681 payload->expiration_time = 0;
682 payload->delivery_time = 0;
683 }
684
685 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
686 {
687 memset(stimer, 0, sizeof(*stimer));
688 stimer->index = timer_index;
689 hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
690 stimer->timer.function = stimer_timer_callback;
691 stimer_prepare_msg(stimer);
692 }
693
694 void kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
695 {
696 struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
697 int i;
698
699 synic_init(&hv_vcpu->synic);
700
701 bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
702 for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
703 stimer_init(&hv_vcpu->stimer[i], i);
704 }
705
706 void kvm_hv_vcpu_postcreate(struct kvm_vcpu *vcpu)
707 {
708 struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
709
710 hv_vcpu->vp_index = kvm_vcpu_get_idx(vcpu);
711 }
712
713 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
714 {
715 struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
716
717 /*
718 * Hyper-V SynIC auto EOI SINT's are
719 * not compatible with APICV, so deactivate APICV
720 */
721 kvm_vcpu_deactivate_apicv(vcpu);
722 synic->active = true;
723 synic->dont_zero_synic_pages = dont_zero_synic_pages;
724 return 0;
725 }
726
727 static bool kvm_hv_msr_partition_wide(u32 msr)
728 {
729 bool r = false;
730
731 switch (msr) {
732 case HV_X64_MSR_GUEST_OS_ID:
733 case HV_X64_MSR_HYPERCALL:
734 case HV_X64_MSR_REFERENCE_TSC:
735 case HV_X64_MSR_TIME_REF_COUNT:
736 case HV_X64_MSR_CRASH_CTL:
737 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
738 case HV_X64_MSR_RESET:
739 r = true;
740 break;
741 }
742
743 return r;
744 }
745
746 static int kvm_hv_msr_get_crash_data(struct kvm_vcpu *vcpu,
747 u32 index, u64 *pdata)
748 {
749 struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
750 size_t size = ARRAY_SIZE(hv->hv_crash_param);
751
752 if (WARN_ON_ONCE(index >= size))
753 return -EINVAL;
754
755 *pdata = hv->hv_crash_param[array_index_nospec(index, size)];
756 return 0;
757 }
758
759 static int kvm_hv_msr_get_crash_ctl(struct kvm_vcpu *vcpu, u64 *pdata)
760 {
761 struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
762
763 *pdata = hv->hv_crash_ctl;
764 return 0;
765 }
766
767 static int kvm_hv_msr_set_crash_ctl(struct kvm_vcpu *vcpu, u64 data, bool host)
768 {
769 struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
770
771 if (host)
772 hv->hv_crash_ctl = data & HV_X64_MSR_CRASH_CTL_NOTIFY;
773
774 if (!host && (data & HV_X64_MSR_CRASH_CTL_NOTIFY)) {
775
776 vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
777 hv->hv_crash_param[0],
778 hv->hv_crash_param[1],
779 hv->hv_crash_param[2],
780 hv->hv_crash_param[3],
781 hv->hv_crash_param[4]);
782
783 /* Send notification about crash to user space */
784 kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
785 }
786
787 return 0;
788 }
789
790 static int kvm_hv_msr_set_crash_data(struct kvm_vcpu *vcpu,
791 u32 index, u64 data)
792 {
793 struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
794 size_t size = ARRAY_SIZE(hv->hv_crash_param);
795
796 if (WARN_ON_ONCE(index >= size))
797 return -EINVAL;
798
799 hv->hv_crash_param[array_index_nospec(index, size)] = data;
800 return 0;
801 }
802
803 /*
804 * The kvmclock and Hyper-V TSC page use similar formulas, and converting
805 * between them is possible:
806 *
807 * kvmclock formula:
808 * nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
809 * + system_time
810 *
811 * Hyper-V formula:
812 * nsec/100 = ticks * scale / 2^64 + offset
813 *
814 * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
815 * By dividing the kvmclock formula by 100 and equating what's left we get:
816 * ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
817 * scale / 2^64 = tsc_to_system_mul * 2^(tsc_shift-32) / 100
818 * scale = tsc_to_system_mul * 2^(32+tsc_shift) / 100
819 *
820 * Now expand the kvmclock formula and divide by 100:
821 * nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
822 * - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
823 * + system_time
824 * nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
825 * - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
826 * + system_time / 100
827 *
828 * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
829 * nsec/100 = ticks * scale / 2^64
830 * - tsc_timestamp * scale / 2^64
831 * + system_time / 100
832 *
833 * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
834 * offset = system_time / 100 - tsc_timestamp * scale / 2^64
835 *
836 * These two equivalencies are implemented in this function.
837 */
838 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
839 HV_REFERENCE_TSC_PAGE *tsc_ref)
840 {
841 u64 max_mul;
842
843 if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
844 return false;
845
846 /*
847 * check if scale would overflow, if so we use the time ref counter
848 * tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
849 * tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
850 * tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
851 */
852 max_mul = 100ull << (32 - hv_clock->tsc_shift);
853 if (hv_clock->tsc_to_system_mul >= max_mul)
854 return false;
855
856 /*
857 * Otherwise compute the scale and offset according to the formulas
858 * derived above.
859 */
860 tsc_ref->tsc_scale =
861 mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
862 hv_clock->tsc_to_system_mul,
863 100);
864
865 tsc_ref->tsc_offset = hv_clock->system_time;
866 do_div(tsc_ref->tsc_offset, 100);
867 tsc_ref->tsc_offset -=
868 mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
869 return true;
870 }
871
872 void kvm_hv_setup_tsc_page(struct kvm *kvm,
873 struct pvclock_vcpu_time_info *hv_clock)
874 {
875 struct kvm_hv *hv = &kvm->arch.hyperv;
876 u32 tsc_seq;
877 u64 gfn;
878
879 BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
880 BUILD_BUG_ON(offsetof(HV_REFERENCE_TSC_PAGE, tsc_sequence) != 0);
881
882 if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
883 return;
884
885 mutex_lock(&kvm->arch.hyperv.hv_lock);
886 if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
887 goto out_unlock;
888
889 gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
890 /*
891 * Because the TSC parameters only vary when there is a
892 * change in the master clock, do not bother with caching.
893 */
894 if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
895 &tsc_seq, sizeof(tsc_seq))))
896 goto out_unlock;
897
898 /*
899 * While we're computing and writing the parameters, force the
900 * guest to use the time reference count MSR.
901 */
902 hv->tsc_ref.tsc_sequence = 0;
903 if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
904 &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
905 goto out_unlock;
906
907 if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
908 goto out_unlock;
909
910 /* Ensure sequence is zero before writing the rest of the struct. */
911 smp_wmb();
912 if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
913 goto out_unlock;
914
915 /*
916 * Now switch to the TSC page mechanism by writing the sequence.
917 */
918 tsc_seq++;
919 if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
920 tsc_seq = 1;
921
922 /* Write the struct entirely before the non-zero sequence. */
923 smp_wmb();
924
925 hv->tsc_ref.tsc_sequence = tsc_seq;
926 kvm_write_guest(kvm, gfn_to_gpa(gfn),
927 &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence));
928 out_unlock:
929 mutex_unlock(&kvm->arch.hyperv.hv_lock);
930 }
931
932 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
933 bool host)
934 {
935 struct kvm *kvm = vcpu->kvm;
936 struct kvm_hv *hv = &kvm->arch.hyperv;
937
938 switch (msr) {
939 case HV_X64_MSR_GUEST_OS_ID:
940 hv->hv_guest_os_id = data;
941 /* setting guest os id to zero disables hypercall page */
942 if (!hv->hv_guest_os_id)
943 hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
944 break;
945 case HV_X64_MSR_HYPERCALL: {
946 u64 gfn;
947 unsigned long addr;
948 u8 instructions[4];
949
950 /* if guest os id is not set hypercall should remain disabled */
951 if (!hv->hv_guest_os_id)
952 break;
953 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
954 hv->hv_hypercall = data;
955 break;
956 }
957 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
958 addr = gfn_to_hva(kvm, gfn);
959 if (kvm_is_error_hva(addr))
960 return 1;
961 kvm_x86_ops->patch_hypercall(vcpu, instructions);
962 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
963 if (__copy_to_user((void __user *)addr, instructions, 4))
964 return 1;
965 hv->hv_hypercall = data;
966 mark_page_dirty(kvm, gfn);
967 break;
968 }
969 case HV_X64_MSR_REFERENCE_TSC:
970 hv->hv_tsc_page = data;
971 if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE)
972 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
973 break;
974 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
975 return kvm_hv_msr_set_crash_data(vcpu,
976 msr - HV_X64_MSR_CRASH_P0,
977 data);
978 case HV_X64_MSR_CRASH_CTL:
979 return kvm_hv_msr_set_crash_ctl(vcpu, data, host);
980 case HV_X64_MSR_RESET:
981 if (data == 1) {
982 vcpu_debug(vcpu, "hyper-v reset requested\n");
983 kvm_make_request(KVM_REQ_HV_RESET, vcpu);
984 }
985 break;
986 default:
987 vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
988 msr, data);
989 return 1;
990 }
991 return 0;
992 }
993
994 /* Calculate cpu time spent by current task in 100ns units */
995 static u64 current_task_runtime_100ns(void)
996 {
997 u64 utime, stime;
998
999 task_cputime_adjusted(current, &utime, &stime);
1000
1001 return div_u64(utime + stime, 100);
1002 }
1003
1004 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1005 {
1006 struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
1007
1008 switch (msr) {
1009 case HV_X64_MSR_VP_INDEX:
1010 if (!host)
1011 return 1;
1012 hv->vp_index = (u32)data;
1013 break;
1014 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1015 u64 gfn;
1016 unsigned long addr;
1017
1018 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1019 hv->hv_vapic = data;
1020 if (kvm_lapic_enable_pv_eoi(vcpu, 0))
1021 return 1;
1022 break;
1023 }
1024 gfn = data >> HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT;
1025 addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1026 if (kvm_is_error_hva(addr))
1027 return 1;
1028 if (__clear_user((void __user *)addr, PAGE_SIZE))
1029 return 1;
1030 hv->hv_vapic = data;
1031 kvm_vcpu_mark_page_dirty(vcpu, gfn);
1032 if (kvm_lapic_enable_pv_eoi(vcpu,
1033 gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
1034 return 1;
1035 break;
1036 }
1037 case HV_X64_MSR_EOI:
1038 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1039 case HV_X64_MSR_ICR:
1040 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1041 case HV_X64_MSR_TPR:
1042 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1043 case HV_X64_MSR_VP_RUNTIME:
1044 if (!host)
1045 return 1;
1046 hv->runtime_offset = data - current_task_runtime_100ns();
1047 break;
1048 case HV_X64_MSR_SCONTROL:
1049 case HV_X64_MSR_SVERSION:
1050 case HV_X64_MSR_SIEFP:
1051 case HV_X64_MSR_SIMP:
1052 case HV_X64_MSR_EOM:
1053 case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1054 return synic_set_msr(vcpu_to_synic(vcpu), msr, data, host);
1055 case HV_X64_MSR_STIMER0_CONFIG:
1056 case HV_X64_MSR_STIMER1_CONFIG:
1057 case HV_X64_MSR_STIMER2_CONFIG:
1058 case HV_X64_MSR_STIMER3_CONFIG: {
1059 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1060
1061 return stimer_set_config(vcpu_to_stimer(vcpu, timer_index),
1062 data, host);
1063 }
1064 case HV_X64_MSR_STIMER0_COUNT:
1065 case HV_X64_MSR_STIMER1_COUNT:
1066 case HV_X64_MSR_STIMER2_COUNT:
1067 case HV_X64_MSR_STIMER3_COUNT: {
1068 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1069
1070 return stimer_set_count(vcpu_to_stimer(vcpu, timer_index),
1071 data, host);
1072 }
1073 default:
1074 vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
1075 msr, data);
1076 return 1;
1077 }
1078
1079 return 0;
1080 }
1081
1082 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1083 {
1084 u64 data = 0;
1085 struct kvm *kvm = vcpu->kvm;
1086 struct kvm_hv *hv = &kvm->arch.hyperv;
1087
1088 switch (msr) {
1089 case HV_X64_MSR_GUEST_OS_ID:
1090 data = hv->hv_guest_os_id;
1091 break;
1092 case HV_X64_MSR_HYPERCALL:
1093 data = hv->hv_hypercall;
1094 break;
1095 case HV_X64_MSR_TIME_REF_COUNT:
1096 data = get_time_ref_counter(kvm);
1097 break;
1098 case HV_X64_MSR_REFERENCE_TSC:
1099 data = hv->hv_tsc_page;
1100 break;
1101 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1102 return kvm_hv_msr_get_crash_data(vcpu,
1103 msr - HV_X64_MSR_CRASH_P0,
1104 pdata);
1105 case HV_X64_MSR_CRASH_CTL:
1106 return kvm_hv_msr_get_crash_ctl(vcpu, pdata);
1107 case HV_X64_MSR_RESET:
1108 data = 0;
1109 break;
1110 default:
1111 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1112 return 1;
1113 }
1114
1115 *pdata = data;
1116 return 0;
1117 }
1118
1119 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1120 {
1121 u64 data = 0;
1122 struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
1123
1124 switch (msr) {
1125 case HV_X64_MSR_VP_INDEX:
1126 data = hv->vp_index;
1127 break;
1128 case HV_X64_MSR_EOI:
1129 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1130 case HV_X64_MSR_ICR:
1131 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1132 case HV_X64_MSR_TPR:
1133 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1134 case HV_X64_MSR_APIC_ASSIST_PAGE:
1135 data = hv->hv_vapic;
1136 break;
1137 case HV_X64_MSR_VP_RUNTIME:
1138 data = current_task_runtime_100ns() + hv->runtime_offset;
1139 break;
1140 case HV_X64_MSR_SCONTROL:
1141 case HV_X64_MSR_SVERSION:
1142 case HV_X64_MSR_SIEFP:
1143 case HV_X64_MSR_SIMP:
1144 case HV_X64_MSR_EOM:
1145 case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1146 return synic_get_msr(vcpu_to_synic(vcpu), msr, pdata);
1147 case HV_X64_MSR_STIMER0_CONFIG:
1148 case HV_X64_MSR_STIMER1_CONFIG:
1149 case HV_X64_MSR_STIMER2_CONFIG:
1150 case HV_X64_MSR_STIMER3_CONFIG: {
1151 int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1152
1153 return stimer_get_config(vcpu_to_stimer(vcpu, timer_index),
1154 pdata);
1155 }
1156 case HV_X64_MSR_STIMER0_COUNT:
1157 case HV_X64_MSR_STIMER1_COUNT:
1158 case HV_X64_MSR_STIMER2_COUNT:
1159 case HV_X64_MSR_STIMER3_COUNT: {
1160 int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1161
1162 return stimer_get_count(vcpu_to_stimer(vcpu, timer_index),
1163 pdata);
1164 }
1165 case HV_X64_MSR_TSC_FREQUENCY:
1166 data = (u64)vcpu->arch.virtual_tsc_khz * 1000;
1167 break;
1168 case HV_X64_MSR_APIC_FREQUENCY:
1169 data = APIC_BUS_FREQUENCY;
1170 break;
1171 default:
1172 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1173 return 1;
1174 }
1175 *pdata = data;
1176 return 0;
1177 }
1178
1179 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1180 {
1181 if (kvm_hv_msr_partition_wide(msr)) {
1182 int r;
1183
1184 mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1185 r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1186 mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1187 return r;
1188 } else
1189 return kvm_hv_set_msr(vcpu, msr, data, host);
1190 }
1191
1192 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1193 {
1194 if (kvm_hv_msr_partition_wide(msr)) {
1195 int r;
1196
1197 mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1198 r = kvm_hv_get_msr_pw(vcpu, msr, pdata);
1199 mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1200 return r;
1201 } else
1202 return kvm_hv_get_msr(vcpu, msr, pdata);
1203 }
1204
1205 bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1206 {
1207 return READ_ONCE(kvm->arch.hyperv.hv_hypercall) & HV_X64_MSR_HYPERCALL_ENABLE;
1208 }
1209
1210 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
1211 {
1212 bool longmode;
1213
1214 longmode = is_64_bit_mode(vcpu);
1215 if (longmode)
1216 kvm_register_write(vcpu, VCPU_REGS_RAX, result);
1217 else {
1218 kvm_register_write(vcpu, VCPU_REGS_RDX, result >> 32);
1219 kvm_register_write(vcpu, VCPU_REGS_RAX, result & 0xffffffff);
1220 }
1221 }
1222
1223 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
1224 {
1225 struct kvm_run *run = vcpu->run;
1226
1227 kvm_hv_hypercall_set_result(vcpu, run->hyperv.u.hcall.result);
1228 return kvm_skip_emulated_instruction(vcpu);
1229 }
1230
1231 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
1232 {
1233 u64 param, ingpa, outgpa, ret;
1234 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
1235 bool fast, longmode;
1236
1237 /*
1238 * hypercall generates UD from non zero cpl and real mode
1239 * per HYPER-V spec
1240 */
1241 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
1242 kvm_queue_exception(vcpu, UD_VECTOR);
1243 return 1;
1244 }
1245
1246 longmode = is_64_bit_mode(vcpu);
1247
1248 if (!longmode) {
1249 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
1250 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
1251 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
1252 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
1253 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
1254 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
1255 }
1256 #ifdef CONFIG_X86_64
1257 else {
1258 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
1259 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
1260 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
1261 }
1262 #endif
1263
1264 code = param & 0xffff;
1265 fast = (param >> 16) & 0x1;
1266 rep_cnt = (param >> 32) & 0xfff;
1267 rep_idx = (param >> 48) & 0xfff;
1268
1269 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
1270
1271 /* Hypercall continuation is not supported yet */
1272 if (rep_cnt || rep_idx) {
1273 res = HV_STATUS_INVALID_HYPERCALL_CODE;
1274 goto set_result;
1275 }
1276
1277 switch (code) {
1278 case HVCALL_NOTIFY_LONG_SPIN_WAIT:
1279 kvm_vcpu_on_spin(vcpu, true);
1280 break;
1281 case HVCALL_POST_MESSAGE:
1282 case HVCALL_SIGNAL_EVENT:
1283 /* don't bother userspace if it has no way to handle it */
1284 if (!vcpu_to_synic(vcpu)->active) {
1285 res = HV_STATUS_INVALID_HYPERCALL_CODE;
1286 break;
1287 }
1288 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
1289 vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
1290 vcpu->run->hyperv.u.hcall.input = param;
1291 vcpu->run->hyperv.u.hcall.params[0] = ingpa;
1292 vcpu->run->hyperv.u.hcall.params[1] = outgpa;
1293 vcpu->arch.complete_userspace_io =
1294 kvm_hv_hypercall_complete_userspace;
1295 return 0;
1296 default:
1297 res = HV_STATUS_INVALID_HYPERCALL_CODE;
1298 break;
1299 }
1300
1301 set_result:
1302 ret = res | (((u64)rep_done & 0xfff) << 32);
1303 kvm_hv_hypercall_set_result(vcpu, ret);
1304 return 1;
1305 }
Linux with added FPC-III support