• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * KVM Microsoft Hyper-V emulation
4  *
5  * derived from arch/x86/kvm/x86.c
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright (C) 2008 Qumranet, Inc.
9  * Copyright IBM Corporation, 2008
10  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11  * Copyright (C) 2015 Andrey Smetanin <asmetanin@virtuozzo.com>
12  *
13  * Authors:
14  *   Avi Kivity   <avi@qumranet.com>
15  *   Yaniv Kamay  <yaniv@qumranet.com>
16  *   Amit Shah    <amit.shah@qumranet.com>
17  *   Ben-Ami Yassour <benami@il.ibm.com>
18  *   Andrey Smetanin <asmetanin@virtuozzo.com>
19  */
20 
21 #include "x86.h"
22 #include "lapic.h"
23 #include "ioapic.h"
24 #include "hyperv.h"
25 
26 #include <linux/cpu.h>
27 #include <linux/kvm_host.h>
28 #include <linux/highmem.h>
29 #include <linux/sched/cputime.h>
30 #include <linux/eventfd.h>
31 
32 #include <asm/apicdef.h>
33 #include <trace/events/kvm.h>
34 
35 #include "trace.h"
36 
37 #define KVM_HV_MAX_SPARSE_VCPU_SET_BITS DIV_ROUND_UP(KVM_MAX_VCPUS, 64)
38 
39 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
40 				bool vcpu_kick);
41 
synic_read_sint(struct kvm_vcpu_hv_synic * synic,int sint)42 static inline u64 synic_read_sint(struct kvm_vcpu_hv_synic *synic, int sint)
43 {
44 	return atomic64_read(&synic->sint[sint]);
45 }
46 
synic_get_sint_vector(u64 sint_value)47 static inline int synic_get_sint_vector(u64 sint_value)
48 {
49 	if (sint_value & HV_SYNIC_SINT_MASKED)
50 		return -1;
51 	return sint_value & HV_SYNIC_SINT_VECTOR_MASK;
52 }
53 
synic_has_vector_connected(struct kvm_vcpu_hv_synic * synic,int vector)54 static bool synic_has_vector_connected(struct kvm_vcpu_hv_synic *synic,
55 				      int vector)
56 {
57 	int i;
58 
59 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
60 		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
61 			return true;
62 	}
63 	return false;
64 }
65 
synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic * synic,int vector)66 static bool synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic *synic,
67 				     int vector)
68 {
69 	int i;
70 	u64 sint_value;
71 
72 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
73 		sint_value = synic_read_sint(synic, i);
74 		if (synic_get_sint_vector(sint_value) == vector &&
75 		    sint_value & HV_SYNIC_SINT_AUTO_EOI)
76 			return true;
77 	}
78 	return false;
79 }
80 
synic_update_vector(struct kvm_vcpu_hv_synic * synic,int vector)81 static void synic_update_vector(struct kvm_vcpu_hv_synic *synic,
82 				int vector)
83 {
84 	if (vector < HV_SYNIC_FIRST_VALID_VECTOR)
85 		return;
86 
87 	if (synic_has_vector_connected(synic, vector))
88 		__set_bit(vector, synic->vec_bitmap);
89 	else
90 		__clear_bit(vector, synic->vec_bitmap);
91 
92 	if (synic_has_vector_auto_eoi(synic, vector))
93 		__set_bit(vector, synic->auto_eoi_bitmap);
94 	else
95 		__clear_bit(vector, synic->auto_eoi_bitmap);
96 }
97 
synic_set_sint(struct kvm_vcpu_hv_synic * synic,int sint,u64 data,bool host)98 static int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint,
99 			  u64 data, bool host)
100 {
101 	int vector, old_vector;
102 	bool masked;
103 
104 	vector = data & HV_SYNIC_SINT_VECTOR_MASK;
105 	masked = data & HV_SYNIC_SINT_MASKED;
106 
107 	/*
108 	 * Valid vectors are 16-255, however, nested Hyper-V attempts to write
109 	 * default '0x10000' value on boot and this should not #GP. We need to
110 	 * allow zero-initing the register from host as well.
111 	 */
112 	if (vector < HV_SYNIC_FIRST_VALID_VECTOR && !host && !masked)
113 		return 1;
114 	/*
115 	 * Guest may configure multiple SINTs to use the same vector, so
116 	 * we maintain a bitmap of vectors handled by synic, and a
117 	 * bitmap of vectors with auto-eoi behavior.  The bitmaps are
118 	 * updated here, and atomically queried on fast paths.
119 	 */
120 	old_vector = synic_read_sint(synic, sint) & HV_SYNIC_SINT_VECTOR_MASK;
121 
122 	atomic64_set(&synic->sint[sint], data);
123 
124 	synic_update_vector(synic, old_vector);
125 
126 	synic_update_vector(synic, vector);
127 
128 	/* Load SynIC vectors into EOI exit bitmap */
129 	kvm_make_request(KVM_REQ_SCAN_IOAPIC, synic_to_vcpu(synic));
130 	return 0;
131 }
132 
get_vcpu_by_vpidx(struct kvm * kvm,u32 vpidx)133 static struct kvm_vcpu *get_vcpu_by_vpidx(struct kvm *kvm, u32 vpidx)
134 {
135 	struct kvm_vcpu *vcpu = NULL;
136 	int i;
137 
138 	if (vpidx >= KVM_MAX_VCPUS)
139 		return NULL;
140 
141 	vcpu = kvm_get_vcpu(kvm, vpidx);
142 	if (vcpu && vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
143 		return vcpu;
144 	kvm_for_each_vcpu(i, vcpu, kvm)
145 		if (vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
146 			return vcpu;
147 	return NULL;
148 }
149 
synic_get(struct kvm * kvm,u32 vpidx)150 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vpidx)
151 {
152 	struct kvm_vcpu *vcpu;
153 	struct kvm_vcpu_hv_synic *synic;
154 
155 	vcpu = get_vcpu_by_vpidx(kvm, vpidx);
156 	if (!vcpu)
157 		return NULL;
158 	synic = vcpu_to_synic(vcpu);
159 	return (synic->active) ? synic : NULL;
160 }
161 
kvm_hv_notify_acked_sint(struct kvm_vcpu * vcpu,u32 sint)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;
169 
170 	trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
171 
172 	/* Try to deliver pending Hyper-V SynIC timers messages */
173 	for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
174 		stimer = &hv_vcpu->stimer[idx];
175 		if (stimer->msg_pending && stimer->config.enable &&
176 		    !stimer->config.direct_mode &&
177 		    stimer->config.sintx == sint)
178 			stimer_mark_pending(stimer, false);
179 	}
180 
181 	idx = srcu_read_lock(&kvm->irq_srcu);
182 	gsi = atomic_read(&synic->sint_to_gsi[sint]);
183 	if (gsi != -1)
184 		kvm_notify_acked_gsi(kvm, gsi);
185 	srcu_read_unlock(&kvm->irq_srcu, idx);
186 }
187 
synic_exit(struct kvm_vcpu_hv_synic * synic,u32 msr)188 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
189 {
190 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
191 	struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
192 
193 	hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
194 	hv_vcpu->exit.u.synic.msr = msr;
195 	hv_vcpu->exit.u.synic.control = synic->control;
196 	hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
197 	hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
198 
199 	kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
200 }
201 
synic_set_msr(struct kvm_vcpu_hv_synic * synic,u32 msr,u64 data,bool host)202 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
203 			 u32 msr, u64 data, bool host)
204 {
205 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
206 	int ret;
207 
208 	if (!synic->active && (!host || data))
209 		return 1;
210 
211 	trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
212 
213 	ret = 0;
214 	switch (msr) {
215 	case HV_X64_MSR_SCONTROL:
216 		synic->control = data;
217 		if (!host)
218 			synic_exit(synic, msr);
219 		break;
220 	case HV_X64_MSR_SVERSION:
221 		if (!host) {
222 			ret = 1;
223 			break;
224 		}
225 		synic->version = data;
226 		break;
227 	case HV_X64_MSR_SIEFP:
228 		if ((data & HV_SYNIC_SIEFP_ENABLE) && !host &&
229 		    !synic->dont_zero_synic_pages)
230 			if (kvm_clear_guest(vcpu->kvm,
231 					    data & PAGE_MASK, PAGE_SIZE)) {
232 				ret = 1;
233 				break;
234 			}
235 		synic->evt_page = data;
236 		if (!host)
237 			synic_exit(synic, msr);
238 		break;
239 	case HV_X64_MSR_SIMP:
240 		if ((data & HV_SYNIC_SIMP_ENABLE) && !host &&
241 		    !synic->dont_zero_synic_pages)
242 			if (kvm_clear_guest(vcpu->kvm,
243 					    data & PAGE_MASK, PAGE_SIZE)) {
244 				ret = 1;
245 				break;
246 			}
247 		synic->msg_page = data;
248 		if (!host)
249 			synic_exit(synic, msr);
250 		break;
251 	case HV_X64_MSR_EOM: {
252 		int i;
253 
254 		if (!synic->active)
255 			break;
256 
257 		for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
258 			kvm_hv_notify_acked_sint(vcpu, i);
259 		break;
260 	}
261 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
262 		ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
263 		break;
264 	default:
265 		ret = 1;
266 		break;
267 	}
268 	return ret;
269 }
270 
synic_get_msr(struct kvm_vcpu_hv_synic * synic,u32 msr,u64 * pdata,bool host)271 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata,
272 			 bool host)
273 {
274 	int ret;
275 
276 	if (!synic->active && !host)
277 		return 1;
278 
279 	ret = 0;
280 	switch (msr) {
281 	case HV_X64_MSR_SCONTROL:
282 		*pdata = synic->control;
283 		break;
284 	case HV_X64_MSR_SVERSION:
285 		*pdata = synic->version;
286 		break;
287 	case HV_X64_MSR_SIEFP:
288 		*pdata = synic->evt_page;
289 		break;
290 	case HV_X64_MSR_SIMP:
291 		*pdata = synic->msg_page;
292 		break;
293 	case HV_X64_MSR_EOM:
294 		*pdata = 0;
295 		break;
296 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
297 		*pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
298 		break;
299 	default:
300 		ret = 1;
301 		break;
302 	}
303 	return ret;
304 }
305 
synic_set_irq(struct kvm_vcpu_hv_synic * synic,u32 sint)306 static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
307 {
308 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
309 	struct kvm_lapic_irq irq;
310 	int ret, vector;
311 
312 	if (KVM_BUG_ON(!lapic_in_kernel(vcpu), vcpu->kvm))
313 		return -EINVAL;
314 
315 	if (sint >= ARRAY_SIZE(synic->sint))
316 		return -EINVAL;
317 
318 	vector = synic_get_sint_vector(synic_read_sint(synic, sint));
319 	if (vector < 0)
320 		return -ENOENT;
321 
322 	memset(&irq, 0, sizeof(irq));
323 	irq.shorthand = APIC_DEST_SELF;
324 	irq.dest_mode = APIC_DEST_PHYSICAL;
325 	irq.delivery_mode = APIC_DM_FIXED;
326 	irq.vector = vector;
327 	irq.level = 1;
328 
329 	ret = kvm_irq_delivery_to_apic(vcpu->kvm, vcpu->arch.apic, &irq, NULL);
330 	trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
331 	return ret;
332 }
333 
kvm_hv_synic_set_irq(struct kvm * kvm,u32 vpidx,u32 sint)334 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vpidx, u32 sint)
335 {
336 	struct kvm_vcpu_hv_synic *synic;
337 
338 	synic = synic_get(kvm, vpidx);
339 	if (!synic)
340 		return -EINVAL;
341 
342 	return synic_set_irq(synic, sint);
343 }
344 
kvm_hv_synic_send_eoi(struct kvm_vcpu * vcpu,int vector)345 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
346 {
347 	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
348 	int i;
349 
350 	trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
351 
352 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
353 		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
354 			kvm_hv_notify_acked_sint(vcpu, i);
355 }
356 
kvm_hv_set_sint_gsi(struct kvm * kvm,u32 vpidx,u32 sint,int gsi)357 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi)
358 {
359 	struct kvm_vcpu_hv_synic *synic;
360 
361 	synic = synic_get(kvm, vpidx);
362 	if (!synic)
363 		return -EINVAL;
364 
365 	if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
366 		return -EINVAL;
367 
368 	atomic_set(&synic->sint_to_gsi[sint], gsi);
369 	return 0;
370 }
371 
kvm_hv_irq_routing_update(struct kvm * kvm)372 void kvm_hv_irq_routing_update(struct kvm *kvm)
373 {
374 	struct kvm_irq_routing_table *irq_rt;
375 	struct kvm_kernel_irq_routing_entry *e;
376 	u32 gsi;
377 
378 	irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
379 					lockdep_is_held(&kvm->irq_lock));
380 
381 	for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
382 		hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
383 			if (e->type == KVM_IRQ_ROUTING_HV_SINT)
384 				kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
385 						    e->hv_sint.sint, gsi);
386 		}
387 	}
388 }
389 
synic_init(struct kvm_vcpu_hv_synic * synic)390 static void synic_init(struct kvm_vcpu_hv_synic *synic)
391 {
392 	int i;
393 
394 	memset(synic, 0, sizeof(*synic));
395 	synic->version = HV_SYNIC_VERSION_1;
396 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
397 		atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
398 		atomic_set(&synic->sint_to_gsi[i], -1);
399 	}
400 }
401 
get_time_ref_counter(struct kvm * kvm)402 static u64 get_time_ref_counter(struct kvm *kvm)
403 {
404 	struct kvm_hv *hv = &kvm->arch.hyperv;
405 	struct kvm_vcpu *vcpu;
406 	u64 tsc;
407 
408 	/*
409 	 * The guest has not set up the TSC page or the clock isn't
410 	 * stable, fall back to get_kvmclock_ns.
411 	 */
412 	if (!hv->tsc_ref.tsc_sequence)
413 		return div_u64(get_kvmclock_ns(kvm), 100);
414 
415 	vcpu = kvm_get_vcpu(kvm, 0);
416 	tsc = kvm_read_l1_tsc(vcpu, rdtsc());
417 	return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64)
418 		+ hv->tsc_ref.tsc_offset;
419 }
420 
stimer_mark_pending(struct kvm_vcpu_hv_stimer * stimer,bool vcpu_kick)421 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
422 				bool vcpu_kick)
423 {
424 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
425 
426 	set_bit(stimer->index,
427 		vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
428 	kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
429 	if (vcpu_kick)
430 		kvm_vcpu_kick(vcpu);
431 }
432 
stimer_cleanup(struct kvm_vcpu_hv_stimer * stimer)433 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
434 {
435 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
436 
437 	trace_kvm_hv_stimer_cleanup(stimer_to_vcpu(stimer)->vcpu_id,
438 				    stimer->index);
439 
440 	hrtimer_cancel(&stimer->timer);
441 	clear_bit(stimer->index,
442 		  vcpu_to_hv_vcpu(vcpu)->stimer_pending_bitmap);
443 	stimer->msg_pending = false;
444 	stimer->exp_time = 0;
445 }
446 
stimer_timer_callback(struct hrtimer * timer)447 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
448 {
449 	struct kvm_vcpu_hv_stimer *stimer;
450 
451 	stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
452 	trace_kvm_hv_stimer_callback(stimer_to_vcpu(stimer)->vcpu_id,
453 				     stimer->index);
454 	stimer_mark_pending(stimer, true);
455 
456 	return HRTIMER_NORESTART;
457 }
458 
459 /*
460  * stimer_start() assumptions:
461  * a) stimer->count is not equal to 0
462  * b) stimer->config has HV_STIMER_ENABLE flag
463  */
stimer_start(struct kvm_vcpu_hv_stimer * stimer)464 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
465 {
466 	u64 time_now;
467 	ktime_t ktime_now;
468 
469 	time_now = get_time_ref_counter(stimer_to_vcpu(stimer)->kvm);
470 	ktime_now = ktime_get();
471 
472 	if (stimer->config.periodic) {
473 		if (stimer->exp_time) {
474 			if (time_now >= stimer->exp_time) {
475 				u64 remainder;
476 
477 				div64_u64_rem(time_now - stimer->exp_time,
478 					      stimer->count, &remainder);
479 				stimer->exp_time =
480 					time_now + (stimer->count - remainder);
481 			}
482 		} else
483 			stimer->exp_time = time_now + stimer->count;
484 
485 		trace_kvm_hv_stimer_start_periodic(
486 					stimer_to_vcpu(stimer)->vcpu_id,
487 					stimer->index,
488 					time_now, stimer->exp_time);
489 
490 		hrtimer_start(&stimer->timer,
491 			      ktime_add_ns(ktime_now,
492 					   100 * (stimer->exp_time - time_now)),
493 			      HRTIMER_MODE_ABS);
494 		return 0;
495 	}
496 	stimer->exp_time = stimer->count;
497 	if (time_now >= stimer->count) {
498 		/*
499 		 * Expire timer according to Hypervisor Top-Level Functional
500 		 * specification v4(15.3.1):
501 		 * "If a one shot is enabled and the specified count is in
502 		 * the past, it will expire immediately."
503 		 */
504 		stimer_mark_pending(stimer, false);
505 		return 0;
506 	}
507 
508 	trace_kvm_hv_stimer_start_one_shot(stimer_to_vcpu(stimer)->vcpu_id,
509 					   stimer->index,
510 					   time_now, stimer->count);
511 
512 	hrtimer_start(&stimer->timer,
513 		      ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)),
514 		      HRTIMER_MODE_ABS);
515 	return 0;
516 }
517 
stimer_set_config(struct kvm_vcpu_hv_stimer * stimer,u64 config,bool host)518 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
519 			     bool host)
520 {
521 	union hv_stimer_config new_config = {.as_uint64 = config},
522 		old_config = {.as_uint64 = stimer->config.as_uint64};
523 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
524 	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
525 
526 	if (!synic->active && (!host || config))
527 		return 1;
528 
529 	trace_kvm_hv_stimer_set_config(stimer_to_vcpu(stimer)->vcpu_id,
530 				       stimer->index, config, host);
531 
532 	stimer_cleanup(stimer);
533 	if (old_config.enable &&
534 	    !new_config.direct_mode && new_config.sintx == 0)
535 		new_config.enable = 0;
536 	stimer->config.as_uint64 = new_config.as_uint64;
537 
538 	if (stimer->config.enable)
539 		stimer_mark_pending(stimer, false);
540 
541 	return 0;
542 }
543 
stimer_set_count(struct kvm_vcpu_hv_stimer * stimer,u64 count,bool host)544 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
545 			    bool host)
546 {
547 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
548 	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
549 
550 	if (!synic->active && (!host || count))
551 		return 1;
552 
553 	trace_kvm_hv_stimer_set_count(stimer_to_vcpu(stimer)->vcpu_id,
554 				      stimer->index, count, host);
555 
556 	stimer_cleanup(stimer);
557 	stimer->count = count;
558 	if (!host) {
559 		if (stimer->count == 0)
560 			stimer->config.enable = 0;
561 		else if (stimer->config.auto_enable)
562 			stimer->config.enable = 1;
563 	}
564 
565 	if (stimer->config.enable)
566 		stimer_mark_pending(stimer, false);
567 
568 	return 0;
569 }
570 
stimer_get_config(struct kvm_vcpu_hv_stimer * stimer,u64 * pconfig)571 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
572 {
573 	*pconfig = stimer->config.as_uint64;
574 	return 0;
575 }
576 
stimer_get_count(struct kvm_vcpu_hv_stimer * stimer,u64 * pcount)577 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
578 {
579 	*pcount = stimer->count;
580 	return 0;
581 }
582 
synic_deliver_msg(struct kvm_vcpu_hv_synic * synic,u32 sint,struct hv_message * src_msg,bool no_retry)583 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
584 			     struct hv_message *src_msg, bool no_retry)
585 {
586 	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
587 	int msg_off = offsetof(struct hv_message_page, sint_message[sint]);
588 	gfn_t msg_page_gfn;
589 	struct hv_message_header hv_hdr;
590 	int r;
591 
592 	if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
593 		return -ENOENT;
594 
595 	msg_page_gfn = synic->msg_page >> PAGE_SHIFT;
596 
597 	/*
598 	 * Strictly following the spec-mandated ordering would assume setting
599 	 * .msg_pending before checking .message_type.  However, this function
600 	 * is only called in vcpu context so the entire update is atomic from
601 	 * guest POV and thus the exact order here doesn't matter.
602 	 */
603 	r = kvm_vcpu_read_guest_page(vcpu, msg_page_gfn, &hv_hdr.message_type,
604 				     msg_off + offsetof(struct hv_message,
605 							header.message_type),
606 				     sizeof(hv_hdr.message_type));
607 	if (r < 0)
608 		return r;
609 
610 	if (hv_hdr.message_type != HVMSG_NONE) {
611 		if (no_retry)
612 			return 0;
613 
614 		hv_hdr.message_flags.msg_pending = 1;
615 		r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn,
616 					      &hv_hdr.message_flags,
617 					      msg_off +
618 					      offsetof(struct hv_message,
619 						       header.message_flags),
620 					      sizeof(hv_hdr.message_flags));
621 		if (r < 0)
622 			return r;
623 		return -EAGAIN;
624 	}
625 
626 	r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn, src_msg, msg_off,
627 				      sizeof(src_msg->header) +
628 				      src_msg->header.payload_size);
629 	if (r < 0)
630 		return r;
631 
632 	r = synic_set_irq(synic, sint);
633 	if (r < 0)
634 		return r;
635 	if (r == 0)
636 		return -EFAULT;
637 	return 0;
638 }
639 
stimer_send_msg(struct kvm_vcpu_hv_stimer * stimer)640 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
641 {
642 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
643 	struct hv_message *msg = &stimer->msg;
644 	struct hv_timer_message_payload *payload =
645 			(struct hv_timer_message_payload *)&msg->u.payload;
646 
647 	/*
648 	 * To avoid piling up periodic ticks, don't retry message
649 	 * delivery for them (within "lazy" lost ticks policy).
650 	 */
651 	bool no_retry = stimer->config.periodic;
652 
653 	payload->expiration_time = stimer->exp_time;
654 	payload->delivery_time = get_time_ref_counter(vcpu->kvm);
655 	return synic_deliver_msg(vcpu_to_synic(vcpu),
656 				 stimer->config.sintx, msg,
657 				 no_retry);
658 }
659 
stimer_notify_direct(struct kvm_vcpu_hv_stimer * stimer)660 static int stimer_notify_direct(struct kvm_vcpu_hv_stimer *stimer)
661 {
662 	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
663 	struct kvm_lapic_irq irq = {
664 		.delivery_mode = APIC_DM_FIXED,
665 		.vector = stimer->config.apic_vector
666 	};
667 
668 	if (lapic_in_kernel(vcpu))
669 		return !kvm_apic_set_irq(vcpu, &irq, NULL);
670 	return 0;
671 }
672 
stimer_expiration(struct kvm_vcpu_hv_stimer * stimer)673 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
674 {
675 	int r, direct = stimer->config.direct_mode;
676 
677 	stimer->msg_pending = true;
678 	if (!direct)
679 		r = stimer_send_msg(stimer);
680 	else
681 		r = stimer_notify_direct(stimer);
682 	trace_kvm_hv_stimer_expiration(stimer_to_vcpu(stimer)->vcpu_id,
683 				       stimer->index, direct, r);
684 	if (!r) {
685 		stimer->msg_pending = false;
686 		if (!(stimer->config.periodic))
687 			stimer->config.enable = 0;
688 	}
689 }
690 
kvm_hv_process_stimers(struct kvm_vcpu * vcpu)691 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
692 {
693 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
694 	struct kvm_vcpu_hv_stimer *stimer;
695 	u64 time_now, exp_time;
696 	int i;
697 
698 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
699 		if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
700 			stimer = &hv_vcpu->stimer[i];
701 			if (stimer->config.enable) {
702 				exp_time = stimer->exp_time;
703 
704 				if (exp_time) {
705 					time_now =
706 						get_time_ref_counter(vcpu->kvm);
707 					if (time_now >= exp_time)
708 						stimer_expiration(stimer);
709 				}
710 
711 				if ((stimer->config.enable) &&
712 				    stimer->count) {
713 					if (!stimer->msg_pending)
714 						stimer_start(stimer);
715 				} else
716 					stimer_cleanup(stimer);
717 			}
718 		}
719 }
720 
kvm_hv_vcpu_uninit(struct kvm_vcpu * vcpu)721 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
722 {
723 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
724 	int i;
725 
726 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
727 		stimer_cleanup(&hv_vcpu->stimer[i]);
728 }
729 
kvm_hv_assist_page_enabled(struct kvm_vcpu * vcpu)730 bool kvm_hv_assist_page_enabled(struct kvm_vcpu *vcpu)
731 {
732 	if (!(vcpu->arch.hyperv.hv_vapic & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE))
733 		return false;
734 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
735 }
736 EXPORT_SYMBOL_GPL(kvm_hv_assist_page_enabled);
737 
kvm_hv_get_assist_page(struct kvm_vcpu * vcpu,struct hv_vp_assist_page * assist_page)738 bool kvm_hv_get_assist_page(struct kvm_vcpu *vcpu,
739 			    struct hv_vp_assist_page *assist_page)
740 {
741 	if (!kvm_hv_assist_page_enabled(vcpu))
742 		return false;
743 	return !kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data,
744 				      assist_page, sizeof(*assist_page));
745 }
746 EXPORT_SYMBOL_GPL(kvm_hv_get_assist_page);
747 
stimer_prepare_msg(struct kvm_vcpu_hv_stimer * stimer)748 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
749 {
750 	struct hv_message *msg = &stimer->msg;
751 	struct hv_timer_message_payload *payload =
752 			(struct hv_timer_message_payload *)&msg->u.payload;
753 
754 	memset(&msg->header, 0, sizeof(msg->header));
755 	msg->header.message_type = HVMSG_TIMER_EXPIRED;
756 	msg->header.payload_size = sizeof(*payload);
757 
758 	payload->timer_index = stimer->index;
759 	payload->expiration_time = 0;
760 	payload->delivery_time = 0;
761 }
762 
stimer_init(struct kvm_vcpu_hv_stimer * stimer,int timer_index)763 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
764 {
765 	memset(stimer, 0, sizeof(*stimer));
766 	stimer->index = timer_index;
767 	hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
768 	stimer->timer.function = stimer_timer_callback;
769 	stimer_prepare_msg(stimer);
770 }
771 
kvm_hv_vcpu_init(struct kvm_vcpu * vcpu)772 void kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
773 {
774 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
775 	int i;
776 
777 	synic_init(&hv_vcpu->synic);
778 
779 	bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
780 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
781 		stimer_init(&hv_vcpu->stimer[i], i);
782 }
783 
kvm_hv_vcpu_postcreate(struct kvm_vcpu * vcpu)784 void kvm_hv_vcpu_postcreate(struct kvm_vcpu *vcpu)
785 {
786 	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
787 
788 	hv_vcpu->vp_index = kvm_vcpu_get_idx(vcpu);
789 }
790 
kvm_hv_activate_synic(struct kvm_vcpu * vcpu,bool dont_zero_synic_pages)791 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
792 {
793 	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
794 
795 	/*
796 	 * Hyper-V SynIC auto EOI SINT's are
797 	 * not compatible with APICV, so deactivate APICV
798 	 */
799 	kvm_vcpu_deactivate_apicv(vcpu);
800 	synic->active = true;
801 	synic->dont_zero_synic_pages = dont_zero_synic_pages;
802 	return 0;
803 }
804 
kvm_hv_msr_partition_wide(u32 msr)805 static bool kvm_hv_msr_partition_wide(u32 msr)
806 {
807 	bool r = false;
808 
809 	switch (msr) {
810 	case HV_X64_MSR_GUEST_OS_ID:
811 	case HV_X64_MSR_HYPERCALL:
812 	case HV_X64_MSR_REFERENCE_TSC:
813 	case HV_X64_MSR_TIME_REF_COUNT:
814 	case HV_X64_MSR_CRASH_CTL:
815 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
816 	case HV_X64_MSR_RESET:
817 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
818 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
819 	case HV_X64_MSR_TSC_EMULATION_STATUS:
820 		r = true;
821 		break;
822 	}
823 
824 	return r;
825 }
826 
kvm_hv_msr_get_crash_data(struct kvm_vcpu * vcpu,u32 index,u64 * pdata)827 static int kvm_hv_msr_get_crash_data(struct kvm_vcpu *vcpu,
828 				     u32 index, u64 *pdata)
829 {
830 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
831 	size_t size = ARRAY_SIZE(hv->hv_crash_param);
832 
833 	if (WARN_ON_ONCE(index >= size))
834 		return -EINVAL;
835 
836 	*pdata = hv->hv_crash_param[array_index_nospec(index, size)];
837 	return 0;
838 }
839 
kvm_hv_msr_get_crash_ctl(struct kvm_vcpu * vcpu,u64 * pdata)840 static int kvm_hv_msr_get_crash_ctl(struct kvm_vcpu *vcpu, u64 *pdata)
841 {
842 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
843 
844 	*pdata = hv->hv_crash_ctl;
845 	return 0;
846 }
847 
kvm_hv_msr_set_crash_ctl(struct kvm_vcpu * vcpu,u64 data,bool host)848 static int kvm_hv_msr_set_crash_ctl(struct kvm_vcpu *vcpu, u64 data, bool host)
849 {
850 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
851 
852 	if (host)
853 		hv->hv_crash_ctl = data & HV_CRASH_CTL_CRASH_NOTIFY;
854 
855 	if (!host && (data & HV_CRASH_CTL_CRASH_NOTIFY)) {
856 
857 		vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
858 			  hv->hv_crash_param[0],
859 			  hv->hv_crash_param[1],
860 			  hv->hv_crash_param[2],
861 			  hv->hv_crash_param[3],
862 			  hv->hv_crash_param[4]);
863 
864 		/* Send notification about crash to user space */
865 		kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
866 	}
867 
868 	return 0;
869 }
870 
kvm_hv_msr_set_crash_data(struct kvm_vcpu * vcpu,u32 index,u64 data)871 static int kvm_hv_msr_set_crash_data(struct kvm_vcpu *vcpu,
872 				     u32 index, u64 data)
873 {
874 	struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
875 	size_t size = ARRAY_SIZE(hv->hv_crash_param);
876 
877 	if (WARN_ON_ONCE(index >= size))
878 		return -EINVAL;
879 
880 	hv->hv_crash_param[array_index_nospec(index, size)] = data;
881 	return 0;
882 }
883 
884 /*
885  * The kvmclock and Hyper-V TSC page use similar formulas, and converting
886  * between them is possible:
887  *
888  * kvmclock formula:
889  *    nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
890  *           + system_time
891  *
892  * Hyper-V formula:
893  *    nsec/100 = ticks * scale / 2^64 + offset
894  *
895  * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
896  * By dividing the kvmclock formula by 100 and equating what's left we get:
897  *    ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
898  *            scale / 2^64 =         tsc_to_system_mul * 2^(tsc_shift-32) / 100
899  *            scale        =         tsc_to_system_mul * 2^(32+tsc_shift) / 100
900  *
901  * Now expand the kvmclock formula and divide by 100:
902  *    nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
903  *           - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
904  *           + system_time
905  *    nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
906  *               - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
907  *               + system_time / 100
908  *
909  * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
910  *    nsec/100 = ticks * scale / 2^64
911  *               - tsc_timestamp * scale / 2^64
912  *               + system_time / 100
913  *
914  * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
915  *    offset = system_time / 100 - tsc_timestamp * scale / 2^64
916  *
917  * These two equivalencies are implemented in this function.
918  */
compute_tsc_page_parameters(struct pvclock_vcpu_time_info * hv_clock,HV_REFERENCE_TSC_PAGE * tsc_ref)919 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
920 					HV_REFERENCE_TSC_PAGE *tsc_ref)
921 {
922 	u64 max_mul;
923 
924 	if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
925 		return false;
926 
927 	/*
928 	 * check if scale would overflow, if so we use the time ref counter
929 	 *    tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
930 	 *    tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
931 	 *    tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
932 	 */
933 	max_mul = 100ull << (32 - hv_clock->tsc_shift);
934 	if (hv_clock->tsc_to_system_mul >= max_mul)
935 		return false;
936 
937 	/*
938 	 * Otherwise compute the scale and offset according to the formulas
939 	 * derived above.
940 	 */
941 	tsc_ref->tsc_scale =
942 		mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
943 				hv_clock->tsc_to_system_mul,
944 				100);
945 
946 	tsc_ref->tsc_offset = hv_clock->system_time;
947 	do_div(tsc_ref->tsc_offset, 100);
948 	tsc_ref->tsc_offset -=
949 		mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
950 	return true;
951 }
952 
kvm_hv_setup_tsc_page(struct kvm * kvm,struct pvclock_vcpu_time_info * hv_clock)953 void kvm_hv_setup_tsc_page(struct kvm *kvm,
954 			   struct pvclock_vcpu_time_info *hv_clock)
955 {
956 	struct kvm_hv *hv = &kvm->arch.hyperv;
957 	u32 tsc_seq;
958 	u64 gfn;
959 
960 	BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
961 	BUILD_BUG_ON(offsetof(HV_REFERENCE_TSC_PAGE, tsc_sequence) != 0);
962 
963 	if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
964 		return;
965 
966 	mutex_lock(&kvm->arch.hyperv.hv_lock);
967 	if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
968 		goto out_unlock;
969 
970 	gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
971 	/*
972 	 * Because the TSC parameters only vary when there is a
973 	 * change in the master clock, do not bother with caching.
974 	 */
975 	if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
976 				    &tsc_seq, sizeof(tsc_seq))))
977 		goto out_unlock;
978 
979 	/*
980 	 * While we're computing and writing the parameters, force the
981 	 * guest to use the time reference count MSR.
982 	 */
983 	hv->tsc_ref.tsc_sequence = 0;
984 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
985 			    &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
986 		goto out_unlock;
987 
988 	if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
989 		goto out_unlock;
990 
991 	/* Ensure sequence is zero before writing the rest of the struct.  */
992 	smp_wmb();
993 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
994 		goto out_unlock;
995 
996 	/*
997 	 * Now switch to the TSC page mechanism by writing the sequence.
998 	 */
999 	tsc_seq++;
1000 	if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
1001 		tsc_seq = 1;
1002 
1003 	/* Write the struct entirely before the non-zero sequence.  */
1004 	smp_wmb();
1005 
1006 	hv->tsc_ref.tsc_sequence = tsc_seq;
1007 	kvm_write_guest(kvm, gfn_to_gpa(gfn),
1008 			&hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence));
1009 out_unlock:
1010 	mutex_unlock(&kvm->arch.hyperv.hv_lock);
1011 }
1012 
kvm_hv_set_msr_pw(struct kvm_vcpu * vcpu,u32 msr,u64 data,bool host)1013 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
1014 			     bool host)
1015 {
1016 	struct kvm *kvm = vcpu->kvm;
1017 	struct kvm_hv *hv = &kvm->arch.hyperv;
1018 
1019 	switch (msr) {
1020 	case HV_X64_MSR_GUEST_OS_ID:
1021 		hv->hv_guest_os_id = data;
1022 		/* setting guest os id to zero disables hypercall page */
1023 		if (!hv->hv_guest_os_id)
1024 			hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1025 		break;
1026 	case HV_X64_MSR_HYPERCALL: {
1027 		u64 gfn;
1028 		unsigned long addr;
1029 		u8 instructions[4];
1030 
1031 		/* if guest os id is not set hypercall should remain disabled */
1032 		if (!hv->hv_guest_os_id)
1033 			break;
1034 		if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1035 			hv->hv_hypercall = data;
1036 			break;
1037 		}
1038 		gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1039 		addr = gfn_to_hva(kvm, gfn);
1040 		if (kvm_is_error_hva(addr))
1041 			return 1;
1042 		kvm_x86_ops->patch_hypercall(vcpu, instructions);
1043 		((unsigned char *)instructions)[3] = 0xc3; /* ret */
1044 		if (__copy_to_user((void __user *)addr, instructions, 4))
1045 			return 1;
1046 		hv->hv_hypercall = data;
1047 		mark_page_dirty(kvm, gfn);
1048 		break;
1049 	}
1050 	case HV_X64_MSR_REFERENCE_TSC:
1051 		hv->hv_tsc_page = data;
1052 		if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE)
1053 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1054 		break;
1055 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1056 		return kvm_hv_msr_set_crash_data(vcpu,
1057 						 msr - HV_X64_MSR_CRASH_P0,
1058 						 data);
1059 	case HV_X64_MSR_CRASH_CTL:
1060 		return kvm_hv_msr_set_crash_ctl(vcpu, data, host);
1061 	case HV_X64_MSR_RESET:
1062 		if (data == 1) {
1063 			vcpu_debug(vcpu, "hyper-v reset requested\n");
1064 			kvm_make_request(KVM_REQ_HV_RESET, vcpu);
1065 		}
1066 		break;
1067 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1068 		hv->hv_reenlightenment_control = data;
1069 		break;
1070 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1071 		hv->hv_tsc_emulation_control = data;
1072 		break;
1073 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1074 		hv->hv_tsc_emulation_status = data;
1075 		break;
1076 	case HV_X64_MSR_TIME_REF_COUNT:
1077 		/* read-only, but still ignore it if host-initiated */
1078 		if (!host)
1079 			return 1;
1080 		break;
1081 	default:
1082 		vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
1083 			    msr, data);
1084 		return 1;
1085 	}
1086 	return 0;
1087 }
1088 
1089 /* Calculate cpu time spent by current task in 100ns units */
current_task_runtime_100ns(void)1090 static u64 current_task_runtime_100ns(void)
1091 {
1092 	u64 utime, stime;
1093 
1094 	task_cputime_adjusted(current, &utime, &stime);
1095 
1096 	return div_u64(utime + stime, 100);
1097 }
1098 
kvm_hv_set_msr(struct kvm_vcpu * vcpu,u32 msr,u64 data,bool host)1099 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1100 {
1101 	struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
1102 
1103 	switch (msr) {
1104 	case HV_X64_MSR_VP_INDEX: {
1105 		struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
1106 		int vcpu_idx = kvm_vcpu_get_idx(vcpu);
1107 		u32 new_vp_index = (u32)data;
1108 
1109 		if (!host || new_vp_index >= KVM_MAX_VCPUS)
1110 			return 1;
1111 
1112 		if (new_vp_index == hv_vcpu->vp_index)
1113 			return 0;
1114 
1115 		/*
1116 		 * The VP index is initialized to vcpu_index by
1117 		 * kvm_hv_vcpu_postcreate so they initially match.  Now the
1118 		 * VP index is changing, adjust num_mismatched_vp_indexes if
1119 		 * it now matches or no longer matches vcpu_idx.
1120 		 */
1121 		if (hv_vcpu->vp_index == vcpu_idx)
1122 			atomic_inc(&hv->num_mismatched_vp_indexes);
1123 		else if (new_vp_index == vcpu_idx)
1124 			atomic_dec(&hv->num_mismatched_vp_indexes);
1125 
1126 		hv_vcpu->vp_index = new_vp_index;
1127 		break;
1128 	}
1129 	case HV_X64_MSR_VP_ASSIST_PAGE: {
1130 		u64 gfn;
1131 		unsigned long addr;
1132 
1133 		if (!(data & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE)) {
1134 			hv_vcpu->hv_vapic = data;
1135 			if (kvm_lapic_enable_pv_eoi(vcpu, 0, 0))
1136 				return 1;
1137 			break;
1138 		}
1139 		gfn = data >> HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT;
1140 		addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1141 		if (kvm_is_error_hva(addr))
1142 			return 1;
1143 
1144 		/*
1145 		 * Clear apic_assist portion of f(struct hv_vp_assist_page
1146 		 * only, there can be valuable data in the rest which needs
1147 		 * to be preserved e.g. on migration.
1148 		 */
1149 		if (__clear_user((void __user *)addr, sizeof(u32)))
1150 			return 1;
1151 		hv_vcpu->hv_vapic = data;
1152 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
1153 		if (kvm_lapic_enable_pv_eoi(vcpu,
1154 					    gfn_to_gpa(gfn) | KVM_MSR_ENABLED,
1155 					    sizeof(struct hv_vp_assist_page)))
1156 			return 1;
1157 		break;
1158 	}
1159 	case HV_X64_MSR_EOI:
1160 		return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1161 	case HV_X64_MSR_ICR:
1162 		return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1163 	case HV_X64_MSR_TPR:
1164 		return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1165 	case HV_X64_MSR_VP_RUNTIME:
1166 		if (!host)
1167 			return 1;
1168 		hv_vcpu->runtime_offset = data - current_task_runtime_100ns();
1169 		break;
1170 	case HV_X64_MSR_SCONTROL:
1171 	case HV_X64_MSR_SVERSION:
1172 	case HV_X64_MSR_SIEFP:
1173 	case HV_X64_MSR_SIMP:
1174 	case HV_X64_MSR_EOM:
1175 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1176 		return synic_set_msr(vcpu_to_synic(vcpu), msr, data, host);
1177 	case HV_X64_MSR_STIMER0_CONFIG:
1178 	case HV_X64_MSR_STIMER1_CONFIG:
1179 	case HV_X64_MSR_STIMER2_CONFIG:
1180 	case HV_X64_MSR_STIMER3_CONFIG: {
1181 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1182 
1183 		return stimer_set_config(vcpu_to_stimer(vcpu, timer_index),
1184 					 data, host);
1185 	}
1186 	case HV_X64_MSR_STIMER0_COUNT:
1187 	case HV_X64_MSR_STIMER1_COUNT:
1188 	case HV_X64_MSR_STIMER2_COUNT:
1189 	case HV_X64_MSR_STIMER3_COUNT: {
1190 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1191 
1192 		return stimer_set_count(vcpu_to_stimer(vcpu, timer_index),
1193 					data, host);
1194 	}
1195 	case HV_X64_MSR_TSC_FREQUENCY:
1196 	case HV_X64_MSR_APIC_FREQUENCY:
1197 		/* read-only, but still ignore it if host-initiated */
1198 		if (!host)
1199 			return 1;
1200 		break;
1201 	default:
1202 		vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
1203 			    msr, data);
1204 		return 1;
1205 	}
1206 
1207 	return 0;
1208 }
1209 
kvm_hv_get_msr_pw(struct kvm_vcpu * vcpu,u32 msr,u64 * pdata)1210 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1211 {
1212 	u64 data = 0;
1213 	struct kvm *kvm = vcpu->kvm;
1214 	struct kvm_hv *hv = &kvm->arch.hyperv;
1215 
1216 	switch (msr) {
1217 	case HV_X64_MSR_GUEST_OS_ID:
1218 		data = hv->hv_guest_os_id;
1219 		break;
1220 	case HV_X64_MSR_HYPERCALL:
1221 		data = hv->hv_hypercall;
1222 		break;
1223 	case HV_X64_MSR_TIME_REF_COUNT:
1224 		data = get_time_ref_counter(kvm);
1225 		break;
1226 	case HV_X64_MSR_REFERENCE_TSC:
1227 		data = hv->hv_tsc_page;
1228 		break;
1229 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1230 		return kvm_hv_msr_get_crash_data(vcpu,
1231 						 msr - HV_X64_MSR_CRASH_P0,
1232 						 pdata);
1233 	case HV_X64_MSR_CRASH_CTL:
1234 		return kvm_hv_msr_get_crash_ctl(vcpu, pdata);
1235 	case HV_X64_MSR_RESET:
1236 		data = 0;
1237 		break;
1238 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1239 		data = hv->hv_reenlightenment_control;
1240 		break;
1241 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1242 		data = hv->hv_tsc_emulation_control;
1243 		break;
1244 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1245 		data = hv->hv_tsc_emulation_status;
1246 		break;
1247 	default:
1248 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1249 		return 1;
1250 	}
1251 
1252 	*pdata = data;
1253 	return 0;
1254 }
1255 
kvm_hv_get_msr(struct kvm_vcpu * vcpu,u32 msr,u64 * pdata,bool host)1256 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1257 			  bool host)
1258 {
1259 	u64 data = 0;
1260 	struct kvm_vcpu_hv *hv_vcpu = &vcpu->arch.hyperv;
1261 
1262 	switch (msr) {
1263 	case HV_X64_MSR_VP_INDEX:
1264 		data = hv_vcpu->vp_index;
1265 		break;
1266 	case HV_X64_MSR_EOI:
1267 		return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1268 	case HV_X64_MSR_ICR:
1269 		return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1270 	case HV_X64_MSR_TPR:
1271 		return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1272 	case HV_X64_MSR_VP_ASSIST_PAGE:
1273 		data = hv_vcpu->hv_vapic;
1274 		break;
1275 	case HV_X64_MSR_VP_RUNTIME:
1276 		data = current_task_runtime_100ns() + hv_vcpu->runtime_offset;
1277 		break;
1278 	case HV_X64_MSR_SCONTROL:
1279 	case HV_X64_MSR_SVERSION:
1280 	case HV_X64_MSR_SIEFP:
1281 	case HV_X64_MSR_SIMP:
1282 	case HV_X64_MSR_EOM:
1283 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1284 		return synic_get_msr(vcpu_to_synic(vcpu), msr, pdata, host);
1285 	case HV_X64_MSR_STIMER0_CONFIG:
1286 	case HV_X64_MSR_STIMER1_CONFIG:
1287 	case HV_X64_MSR_STIMER2_CONFIG:
1288 	case HV_X64_MSR_STIMER3_CONFIG: {
1289 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1290 
1291 		return stimer_get_config(vcpu_to_stimer(vcpu, timer_index),
1292 					 pdata);
1293 	}
1294 	case HV_X64_MSR_STIMER0_COUNT:
1295 	case HV_X64_MSR_STIMER1_COUNT:
1296 	case HV_X64_MSR_STIMER2_COUNT:
1297 	case HV_X64_MSR_STIMER3_COUNT: {
1298 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1299 
1300 		return stimer_get_count(vcpu_to_stimer(vcpu, timer_index),
1301 					pdata);
1302 	}
1303 	case HV_X64_MSR_TSC_FREQUENCY:
1304 		data = (u64)vcpu->arch.virtual_tsc_khz * 1000;
1305 		break;
1306 	case HV_X64_MSR_APIC_FREQUENCY:
1307 		data = APIC_BUS_FREQUENCY;
1308 		break;
1309 	default:
1310 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1311 		return 1;
1312 	}
1313 	*pdata = data;
1314 	return 0;
1315 }
1316 
kvm_hv_set_msr_common(struct kvm_vcpu * vcpu,u32 msr,u64 data,bool host)1317 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1318 {
1319 	if (kvm_hv_msr_partition_wide(msr)) {
1320 		int r;
1321 
1322 		mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1323 		r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1324 		mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1325 		return r;
1326 	} else
1327 		return kvm_hv_set_msr(vcpu, msr, data, host);
1328 }
1329 
kvm_hv_get_msr_common(struct kvm_vcpu * vcpu,u32 msr,u64 * pdata,bool host)1330 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
1331 {
1332 	if (kvm_hv_msr_partition_wide(msr)) {
1333 		int r;
1334 
1335 		mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
1336 		r = kvm_hv_get_msr_pw(vcpu, msr, pdata);
1337 		mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
1338 		return r;
1339 	} else
1340 		return kvm_hv_get_msr(vcpu, msr, pdata, host);
1341 }
1342 
sparse_set_to_vcpu_mask(struct kvm * kvm,u64 * sparse_banks,u64 valid_bank_mask,u64 * vp_bitmap,unsigned long * vcpu_bitmap)1343 static __always_inline unsigned long *sparse_set_to_vcpu_mask(
1344 	struct kvm *kvm, u64 *sparse_banks, u64 valid_bank_mask,
1345 	u64 *vp_bitmap, unsigned long *vcpu_bitmap)
1346 {
1347 	struct kvm_hv *hv = &kvm->arch.hyperv;
1348 	struct kvm_vcpu *vcpu;
1349 	int i, bank, sbank = 0;
1350 
1351 	memset(vp_bitmap, 0,
1352 	       KVM_HV_MAX_SPARSE_VCPU_SET_BITS * sizeof(*vp_bitmap));
1353 	for_each_set_bit(bank, (unsigned long *)&valid_bank_mask,
1354 			 KVM_HV_MAX_SPARSE_VCPU_SET_BITS)
1355 		vp_bitmap[bank] = sparse_banks[sbank++];
1356 
1357 	if (likely(!atomic_read(&hv->num_mismatched_vp_indexes))) {
1358 		/* for all vcpus vp_index == vcpu_idx */
1359 		return (unsigned long *)vp_bitmap;
1360 	}
1361 
1362 	bitmap_zero(vcpu_bitmap, KVM_MAX_VCPUS);
1363 	kvm_for_each_vcpu(i, vcpu, kvm) {
1364 		if (test_bit(vcpu_to_hv_vcpu(vcpu)->vp_index,
1365 			     (unsigned long *)vp_bitmap))
1366 			__set_bit(i, vcpu_bitmap);
1367 	}
1368 	return vcpu_bitmap;
1369 }
1370 
kvm_hv_flush_tlb(struct kvm_vcpu * current_vcpu,u64 ingpa,u16 rep_cnt,bool ex)1371 static u64 kvm_hv_flush_tlb(struct kvm_vcpu *current_vcpu, u64 ingpa,
1372 			    u16 rep_cnt, bool ex)
1373 {
1374 	struct kvm *kvm = current_vcpu->kvm;
1375 	struct kvm_vcpu_hv *hv_vcpu = &current_vcpu->arch.hyperv;
1376 	struct hv_tlb_flush_ex flush_ex;
1377 	struct hv_tlb_flush flush;
1378 	u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1379 	DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1380 	unsigned long *vcpu_mask;
1381 	u64 valid_bank_mask;
1382 	u64 sparse_banks[64];
1383 	int sparse_banks_len;
1384 	bool all_cpus;
1385 
1386 	if (!ex) {
1387 		if (unlikely(kvm_read_guest(kvm, ingpa, &flush, sizeof(flush))))
1388 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1389 
1390 		trace_kvm_hv_flush_tlb(flush.processor_mask,
1391 				       flush.address_space, flush.flags);
1392 
1393 		valid_bank_mask = BIT_ULL(0);
1394 		sparse_banks[0] = flush.processor_mask;
1395 
1396 		/*
1397 		 * Work around possible WS2012 bug: it sends hypercalls
1398 		 * with processor_mask = 0x0 and HV_FLUSH_ALL_PROCESSORS clear,
1399 		 * while also expecting us to flush something and crashing if
1400 		 * we don't. Let's treat processor_mask == 0 same as
1401 		 * HV_FLUSH_ALL_PROCESSORS.
1402 		 */
1403 		all_cpus = (flush.flags & HV_FLUSH_ALL_PROCESSORS) ||
1404 			flush.processor_mask == 0;
1405 	} else {
1406 		if (unlikely(kvm_read_guest(kvm, ingpa, &flush_ex,
1407 					    sizeof(flush_ex))))
1408 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1409 
1410 		trace_kvm_hv_flush_tlb_ex(flush_ex.hv_vp_set.valid_bank_mask,
1411 					  flush_ex.hv_vp_set.format,
1412 					  flush_ex.address_space,
1413 					  flush_ex.flags);
1414 
1415 		valid_bank_mask = flush_ex.hv_vp_set.valid_bank_mask;
1416 		all_cpus = flush_ex.hv_vp_set.format !=
1417 			HV_GENERIC_SET_SPARSE_4K;
1418 
1419 		sparse_banks_len =
1420 			bitmap_weight((unsigned long *)&valid_bank_mask, 64) *
1421 			sizeof(sparse_banks[0]);
1422 
1423 		if (!sparse_banks_len && !all_cpus)
1424 			goto ret_success;
1425 
1426 		if (!all_cpus &&
1427 		    kvm_read_guest(kvm,
1428 				   ingpa + offsetof(struct hv_tlb_flush_ex,
1429 						    hv_vp_set.bank_contents),
1430 				   sparse_banks,
1431 				   sparse_banks_len))
1432 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1433 	}
1434 
1435 	cpumask_clear(&hv_vcpu->tlb_flush);
1436 
1437 	vcpu_mask = all_cpus ? NULL :
1438 		sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1439 					vp_bitmap, vcpu_bitmap);
1440 
1441 	/*
1442 	 * vcpu->arch.cr3 may not be up-to-date for running vCPUs so we can't
1443 	 * analyze it here, flush TLB regardless of the specified address space.
1444 	 */
1445 	kvm_make_vcpus_request_mask(kvm,
1446 				    KVM_REQ_TLB_FLUSH | KVM_REQUEST_NO_WAKEUP,
1447 				    vcpu_mask, &hv_vcpu->tlb_flush);
1448 
1449 ret_success:
1450 	/* We always do full TLB flush, set rep_done = rep_cnt. */
1451 	return (u64)HV_STATUS_SUCCESS |
1452 		((u64)rep_cnt << HV_HYPERCALL_REP_COMP_OFFSET);
1453 }
1454 
kvm_send_ipi_to_many(struct kvm * kvm,u32 vector,unsigned long * vcpu_bitmap)1455 static void kvm_send_ipi_to_many(struct kvm *kvm, u32 vector,
1456 				 unsigned long *vcpu_bitmap)
1457 {
1458 	struct kvm_lapic_irq irq = {
1459 		.delivery_mode = APIC_DM_FIXED,
1460 		.vector = vector
1461 	};
1462 	struct kvm_vcpu *vcpu;
1463 	int i;
1464 
1465 	kvm_for_each_vcpu(i, vcpu, kvm) {
1466 		if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
1467 			continue;
1468 
1469 		/* We fail only when APIC is disabled */
1470 		kvm_apic_set_irq(vcpu, &irq, NULL);
1471 	}
1472 }
1473 
kvm_hv_send_ipi(struct kvm_vcpu * current_vcpu,u64 ingpa,u64 outgpa,bool ex,bool fast)1474 static u64 kvm_hv_send_ipi(struct kvm_vcpu *current_vcpu, u64 ingpa, u64 outgpa,
1475 			   bool ex, bool fast)
1476 {
1477 	struct kvm *kvm = current_vcpu->kvm;
1478 	struct hv_send_ipi_ex send_ipi_ex;
1479 	struct hv_send_ipi send_ipi;
1480 	u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1481 	DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
1482 	unsigned long *vcpu_mask;
1483 	unsigned long valid_bank_mask;
1484 	u64 sparse_banks[64];
1485 	int sparse_banks_len;
1486 	u32 vector;
1487 	bool all_cpus;
1488 
1489 	if (!ex) {
1490 		if (!fast) {
1491 			if (unlikely(kvm_read_guest(kvm, ingpa, &send_ipi,
1492 						    sizeof(send_ipi))))
1493 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
1494 			sparse_banks[0] = send_ipi.cpu_mask;
1495 			vector = send_ipi.vector;
1496 		} else {
1497 			/* 'reserved' part of hv_send_ipi should be 0 */
1498 			if (unlikely(ingpa >> 32 != 0))
1499 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
1500 			sparse_banks[0] = outgpa;
1501 			vector = (u32)ingpa;
1502 		}
1503 		all_cpus = false;
1504 		valid_bank_mask = BIT_ULL(0);
1505 
1506 		trace_kvm_hv_send_ipi(vector, sparse_banks[0]);
1507 	} else {
1508 		if (unlikely(kvm_read_guest(kvm, ingpa, &send_ipi_ex,
1509 					    sizeof(send_ipi_ex))))
1510 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1511 
1512 		trace_kvm_hv_send_ipi_ex(send_ipi_ex.vector,
1513 					 send_ipi_ex.vp_set.format,
1514 					 send_ipi_ex.vp_set.valid_bank_mask);
1515 
1516 		vector = send_ipi_ex.vector;
1517 		valid_bank_mask = send_ipi_ex.vp_set.valid_bank_mask;
1518 		sparse_banks_len = bitmap_weight(&valid_bank_mask, 64) *
1519 			sizeof(sparse_banks[0]);
1520 
1521 		all_cpus = send_ipi_ex.vp_set.format == HV_GENERIC_SET_ALL;
1522 
1523 		if (all_cpus)
1524 			goto check_and_send_ipi;
1525 
1526 		if (!sparse_banks_len)
1527 			goto ret_success;
1528 
1529 		if (kvm_read_guest(kvm,
1530 				   ingpa + offsetof(struct hv_send_ipi_ex,
1531 						    vp_set.bank_contents),
1532 				   sparse_banks,
1533 				   sparse_banks_len))
1534 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1535 	}
1536 
1537 check_and_send_ipi:
1538 	if ((vector < HV_IPI_LOW_VECTOR) || (vector > HV_IPI_HIGH_VECTOR))
1539 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
1540 
1541 	vcpu_mask = all_cpus ? NULL :
1542 		sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask,
1543 					vp_bitmap, vcpu_bitmap);
1544 
1545 	kvm_send_ipi_to_many(kvm, vector, vcpu_mask);
1546 
1547 ret_success:
1548 	return HV_STATUS_SUCCESS;
1549 }
1550 
kvm_hv_hypercall_enabled(struct kvm * kvm)1551 bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1552 {
1553 	return READ_ONCE(kvm->arch.hyperv.hv_hypercall) & HV_X64_MSR_HYPERCALL_ENABLE;
1554 }
1555 
kvm_hv_hypercall_set_result(struct kvm_vcpu * vcpu,u64 result)1556 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
1557 {
1558 	bool longmode;
1559 
1560 	longmode = is_64_bit_mode(vcpu);
1561 	if (longmode)
1562 		kvm_rax_write(vcpu, result);
1563 	else {
1564 		kvm_rdx_write(vcpu, result >> 32);
1565 		kvm_rax_write(vcpu, result & 0xffffffff);
1566 	}
1567 }
1568 
kvm_hv_hypercall_complete(struct kvm_vcpu * vcpu,u64 result)1569 static int kvm_hv_hypercall_complete(struct kvm_vcpu *vcpu, u64 result)
1570 {
1571 	kvm_hv_hypercall_set_result(vcpu, result);
1572 	++vcpu->stat.hypercalls;
1573 	return kvm_skip_emulated_instruction(vcpu);
1574 }
1575 
kvm_hv_hypercall_complete_userspace(struct kvm_vcpu * vcpu)1576 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
1577 {
1578 	return kvm_hv_hypercall_complete(vcpu, vcpu->run->hyperv.u.hcall.result);
1579 }
1580 
kvm_hvcall_signal_event(struct kvm_vcpu * vcpu,bool fast,u64 param)1581 static u16 kvm_hvcall_signal_event(struct kvm_vcpu *vcpu, bool fast, u64 param)
1582 {
1583 	struct eventfd_ctx *eventfd;
1584 
1585 	if (unlikely(!fast)) {
1586 		int ret;
1587 		gpa_t gpa = param;
1588 
1589 		if ((gpa & (__alignof__(param) - 1)) ||
1590 		    offset_in_page(gpa) + sizeof(param) > PAGE_SIZE)
1591 			return HV_STATUS_INVALID_ALIGNMENT;
1592 
1593 		ret = kvm_vcpu_read_guest(vcpu, gpa, &param, sizeof(param));
1594 		if (ret < 0)
1595 			return HV_STATUS_INVALID_ALIGNMENT;
1596 	}
1597 
1598 	/*
1599 	 * Per spec, bits 32-47 contain the extra "flag number".  However, we
1600 	 * have no use for it, and in all known usecases it is zero, so just
1601 	 * report lookup failure if it isn't.
1602 	 */
1603 	if (param & 0xffff00000000ULL)
1604 		return HV_STATUS_INVALID_PORT_ID;
1605 	/* remaining bits are reserved-zero */
1606 	if (param & ~KVM_HYPERV_CONN_ID_MASK)
1607 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
1608 
1609 	/* the eventfd is protected by vcpu->kvm->srcu, but conn_to_evt isn't */
1610 	rcu_read_lock();
1611 	eventfd = idr_find(&vcpu->kvm->arch.hyperv.conn_to_evt, param);
1612 	rcu_read_unlock();
1613 	if (!eventfd)
1614 		return HV_STATUS_INVALID_PORT_ID;
1615 
1616 	eventfd_signal(eventfd, 1);
1617 	return HV_STATUS_SUCCESS;
1618 }
1619 
kvm_hv_hypercall(struct kvm_vcpu * vcpu)1620 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
1621 {
1622 	u64 param, ingpa, outgpa, ret = HV_STATUS_SUCCESS;
1623 	uint16_t code, rep_idx, rep_cnt;
1624 	bool fast, rep;
1625 
1626 	/*
1627 	 * hypercall generates UD from non zero cpl and real mode
1628 	 * per HYPER-V spec
1629 	 */
1630 	if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
1631 		kvm_queue_exception(vcpu, UD_VECTOR);
1632 		return 1;
1633 	}
1634 
1635 #ifdef CONFIG_X86_64
1636 	if (is_64_bit_mode(vcpu)) {
1637 		param = kvm_rcx_read(vcpu);
1638 		ingpa = kvm_rdx_read(vcpu);
1639 		outgpa = kvm_r8_read(vcpu);
1640 	} else
1641 #endif
1642 	{
1643 		param = ((u64)kvm_rdx_read(vcpu) << 32) |
1644 			(kvm_rax_read(vcpu) & 0xffffffff);
1645 		ingpa = ((u64)kvm_rbx_read(vcpu) << 32) |
1646 			(kvm_rcx_read(vcpu) & 0xffffffff);
1647 		outgpa = ((u64)kvm_rdi_read(vcpu) << 32) |
1648 			(kvm_rsi_read(vcpu) & 0xffffffff);
1649 	}
1650 
1651 	code = param & 0xffff;
1652 	fast = !!(param & HV_HYPERCALL_FAST_BIT);
1653 	rep_cnt = (param >> HV_HYPERCALL_REP_COMP_OFFSET) & 0xfff;
1654 	rep_idx = (param >> HV_HYPERCALL_REP_START_OFFSET) & 0xfff;
1655 	rep = !!(rep_cnt || rep_idx);
1656 
1657 	trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
1658 
1659 	switch (code) {
1660 	case HVCALL_NOTIFY_LONG_SPIN_WAIT:
1661 		if (unlikely(rep)) {
1662 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1663 			break;
1664 		}
1665 		kvm_vcpu_on_spin(vcpu, true);
1666 		break;
1667 	case HVCALL_SIGNAL_EVENT:
1668 		if (unlikely(rep)) {
1669 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1670 			break;
1671 		}
1672 		ret = kvm_hvcall_signal_event(vcpu, fast, ingpa);
1673 		if (ret != HV_STATUS_INVALID_PORT_ID)
1674 			break;
1675 		/* fall through - maybe userspace knows this conn_id. */
1676 	case HVCALL_POST_MESSAGE:
1677 		/* don't bother userspace if it has no way to handle it */
1678 		if (unlikely(rep || !vcpu_to_synic(vcpu)->active)) {
1679 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1680 			break;
1681 		}
1682 		vcpu->run->exit_reason = KVM_EXIT_HYPERV;
1683 		vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
1684 		vcpu->run->hyperv.u.hcall.input = param;
1685 		vcpu->run->hyperv.u.hcall.params[0] = ingpa;
1686 		vcpu->run->hyperv.u.hcall.params[1] = outgpa;
1687 		vcpu->arch.complete_userspace_io =
1688 				kvm_hv_hypercall_complete_userspace;
1689 		return 0;
1690 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
1691 		if (unlikely(fast || !rep_cnt || rep_idx)) {
1692 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1693 			break;
1694 		}
1695 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false);
1696 		break;
1697 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
1698 		if (unlikely(fast || rep)) {
1699 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1700 			break;
1701 		}
1702 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, false);
1703 		break;
1704 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
1705 		if (unlikely(fast || !rep_cnt || rep_idx)) {
1706 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1707 			break;
1708 		}
1709 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
1710 		break;
1711 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
1712 		if (unlikely(fast || rep)) {
1713 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1714 			break;
1715 		}
1716 		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
1717 		break;
1718 	case HVCALL_SEND_IPI:
1719 		if (unlikely(rep)) {
1720 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1721 			break;
1722 		}
1723 		ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, false, fast);
1724 		break;
1725 	case HVCALL_SEND_IPI_EX:
1726 		if (unlikely(fast || rep)) {
1727 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
1728 			break;
1729 		}
1730 		ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, true, false);
1731 		break;
1732 	default:
1733 		ret = HV_STATUS_INVALID_HYPERCALL_CODE;
1734 		break;
1735 	}
1736 
1737 	return kvm_hv_hypercall_complete(vcpu, ret);
1738 }
1739 
kvm_hv_init_vm(struct kvm * kvm)1740 void kvm_hv_init_vm(struct kvm *kvm)
1741 {
1742 	mutex_init(&kvm->arch.hyperv.hv_lock);
1743 	idr_init(&kvm->arch.hyperv.conn_to_evt);
1744 }
1745 
kvm_hv_destroy_vm(struct kvm * kvm)1746 void kvm_hv_destroy_vm(struct kvm *kvm)
1747 {
1748 	struct eventfd_ctx *eventfd;
1749 	int i;
1750 
1751 	idr_for_each_entry(&kvm->arch.hyperv.conn_to_evt, eventfd, i)
1752 		eventfd_ctx_put(eventfd);
1753 	idr_destroy(&kvm->arch.hyperv.conn_to_evt);
1754 }
1755 
kvm_hv_eventfd_assign(struct kvm * kvm,u32 conn_id,int fd)1756 static int kvm_hv_eventfd_assign(struct kvm *kvm, u32 conn_id, int fd)
1757 {
1758 	struct kvm_hv *hv = &kvm->arch.hyperv;
1759 	struct eventfd_ctx *eventfd;
1760 	int ret;
1761 
1762 	eventfd = eventfd_ctx_fdget(fd);
1763 	if (IS_ERR(eventfd))
1764 		return PTR_ERR(eventfd);
1765 
1766 	mutex_lock(&hv->hv_lock);
1767 	ret = idr_alloc(&hv->conn_to_evt, eventfd, conn_id, conn_id + 1,
1768 			GFP_KERNEL_ACCOUNT);
1769 	mutex_unlock(&hv->hv_lock);
1770 
1771 	if (ret >= 0)
1772 		return 0;
1773 
1774 	if (ret == -ENOSPC)
1775 		ret = -EEXIST;
1776 	eventfd_ctx_put(eventfd);
1777 	return ret;
1778 }
1779 
kvm_hv_eventfd_deassign(struct kvm * kvm,u32 conn_id)1780 static int kvm_hv_eventfd_deassign(struct kvm *kvm, u32 conn_id)
1781 {
1782 	struct kvm_hv *hv = &kvm->arch.hyperv;
1783 	struct eventfd_ctx *eventfd;
1784 
1785 	mutex_lock(&hv->hv_lock);
1786 	eventfd = idr_remove(&hv->conn_to_evt, conn_id);
1787 	mutex_unlock(&hv->hv_lock);
1788 
1789 	if (!eventfd)
1790 		return -ENOENT;
1791 
1792 	synchronize_srcu(&kvm->srcu);
1793 	eventfd_ctx_put(eventfd);
1794 	return 0;
1795 }
1796 
kvm_vm_ioctl_hv_eventfd(struct kvm * kvm,struct kvm_hyperv_eventfd * args)1797 int kvm_vm_ioctl_hv_eventfd(struct kvm *kvm, struct kvm_hyperv_eventfd *args)
1798 {
1799 	if ((args->flags & ~KVM_HYPERV_EVENTFD_DEASSIGN) ||
1800 	    (args->conn_id & ~KVM_HYPERV_CONN_ID_MASK))
1801 		return -EINVAL;
1802 
1803 	if (args->flags == KVM_HYPERV_EVENTFD_DEASSIGN)
1804 		return kvm_hv_eventfd_deassign(kvm, args->conn_id);
1805 	return kvm_hv_eventfd_assign(kvm, args->conn_id, args->fd);
1806 }
1807 
kvm_vcpu_ioctl_get_hv_cpuid(struct kvm_vcpu * vcpu,struct kvm_cpuid2 * cpuid,struct kvm_cpuid_entry2 __user * entries)1808 int kvm_vcpu_ioctl_get_hv_cpuid(struct kvm_vcpu *vcpu, struct kvm_cpuid2 *cpuid,
1809 				struct kvm_cpuid_entry2 __user *entries)
1810 {
1811 	uint16_t evmcs_ver = 0;
1812 	struct kvm_cpuid_entry2 cpuid_entries[] = {
1813 		{ .function = HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS },
1814 		{ .function = HYPERV_CPUID_INTERFACE },
1815 		{ .function = HYPERV_CPUID_VERSION },
1816 		{ .function = HYPERV_CPUID_FEATURES },
1817 		{ .function = HYPERV_CPUID_ENLIGHTMENT_INFO },
1818 		{ .function = HYPERV_CPUID_IMPLEMENT_LIMITS },
1819 		{ .function = HYPERV_CPUID_NESTED_FEATURES },
1820 	};
1821 	int i, nent = ARRAY_SIZE(cpuid_entries);
1822 
1823 	if (kvm_x86_ops->nested_get_evmcs_version)
1824 		evmcs_ver = kvm_x86_ops->nested_get_evmcs_version(vcpu);
1825 
1826 	/* Skip NESTED_FEATURES if eVMCS is not supported */
1827 	if (!evmcs_ver)
1828 		--nent;
1829 
1830 	if (cpuid->nent < nent)
1831 		return -E2BIG;
1832 
1833 	if (cpuid->nent > nent)
1834 		cpuid->nent = nent;
1835 
1836 	for (i = 0; i < nent; i++) {
1837 		struct kvm_cpuid_entry2 *ent = &cpuid_entries[i];
1838 		u32 signature[3];
1839 
1840 		switch (ent->function) {
1841 		case HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS:
1842 			memcpy(signature, "Linux KVM Hv", 12);
1843 
1844 			ent->eax = HYPERV_CPUID_NESTED_FEATURES;
1845 			ent->ebx = signature[0];
1846 			ent->ecx = signature[1];
1847 			ent->edx = signature[2];
1848 			break;
1849 
1850 		case HYPERV_CPUID_INTERFACE:
1851 			memcpy(signature, "Hv#1\0\0\0\0\0\0\0\0", 12);
1852 			ent->eax = signature[0];
1853 			break;
1854 
1855 		case HYPERV_CPUID_VERSION:
1856 			/*
1857 			 * We implement some Hyper-V 2016 functions so let's use
1858 			 * this version.
1859 			 */
1860 			ent->eax = 0x00003839;
1861 			ent->ebx = 0x000A0000;
1862 			break;
1863 
1864 		case HYPERV_CPUID_FEATURES:
1865 			ent->eax |= HV_X64_MSR_VP_RUNTIME_AVAILABLE;
1866 			ent->eax |= HV_MSR_TIME_REF_COUNT_AVAILABLE;
1867 			ent->eax |= HV_X64_MSR_SYNIC_AVAILABLE;
1868 			ent->eax |= HV_MSR_SYNTIMER_AVAILABLE;
1869 			ent->eax |= HV_X64_MSR_APIC_ACCESS_AVAILABLE;
1870 			ent->eax |= HV_X64_MSR_HYPERCALL_AVAILABLE;
1871 			ent->eax |= HV_X64_MSR_VP_INDEX_AVAILABLE;
1872 			ent->eax |= HV_X64_MSR_RESET_AVAILABLE;
1873 			ent->eax |= HV_MSR_REFERENCE_TSC_AVAILABLE;
1874 			ent->eax |= HV_X64_ACCESS_FREQUENCY_MSRS;
1875 			ent->eax |= HV_X64_ACCESS_REENLIGHTENMENT;
1876 
1877 			ent->ebx |= HV_X64_POST_MESSAGES;
1878 			ent->ebx |= HV_X64_SIGNAL_EVENTS;
1879 
1880 			ent->edx |= HV_FEATURE_FREQUENCY_MSRS_AVAILABLE;
1881 			ent->edx |= HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
1882 
1883 			/*
1884 			 * Direct Synthetic timers only make sense with in-kernel
1885 			 * LAPIC
1886 			 */
1887 			if (lapic_in_kernel(vcpu))
1888 				ent->edx |= HV_STIMER_DIRECT_MODE_AVAILABLE;
1889 
1890 			break;
1891 
1892 		case HYPERV_CPUID_ENLIGHTMENT_INFO:
1893 			ent->eax |= HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
1894 			ent->eax |= HV_X64_APIC_ACCESS_RECOMMENDED;
1895 			ent->eax |= HV_X64_RELAXED_TIMING_RECOMMENDED;
1896 			ent->eax |= HV_X64_CLUSTER_IPI_RECOMMENDED;
1897 			ent->eax |= HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED;
1898 			if (evmcs_ver)
1899 				ent->eax |= HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
1900 			if (!cpu_smt_possible())
1901 				ent->eax |= HV_X64_NO_NONARCH_CORESHARING;
1902 			/*
1903 			 * Default number of spinlock retry attempts, matches
1904 			 * HyperV 2016.
1905 			 */
1906 			ent->ebx = 0x00000FFF;
1907 
1908 			break;
1909 
1910 		case HYPERV_CPUID_IMPLEMENT_LIMITS:
1911 			/* Maximum number of virtual processors */
1912 			ent->eax = KVM_MAX_VCPUS;
1913 			/*
1914 			 * Maximum number of logical processors, matches
1915 			 * HyperV 2016.
1916 			 */
1917 			ent->ebx = 64;
1918 
1919 			break;
1920 
1921 		case HYPERV_CPUID_NESTED_FEATURES:
1922 			ent->eax = evmcs_ver;
1923 
1924 			break;
1925 
1926 		default:
1927 			break;
1928 		}
1929 	}
1930 
1931 	if (copy_to_user(entries, cpuid_entries,
1932 			 nent * sizeof(struct kvm_cpuid_entry2)))
1933 		return -EFAULT;
1934 
1935 	return 0;
1936 }
1937