xref: /openbmc/linux/arch/x86/kvm/hyperv.c (revision e0c63803)
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 
22 #include "x86.h"
23 #include "lapic.h"
24 #include "ioapic.h"
25 #include "cpuid.h"
26 #include "hyperv.h"
27 #include "mmu.h"
28 #include "xen.h"
29 
30 #include <linux/cpu.h>
31 #include <linux/kvm_host.h>
32 #include <linux/highmem.h>
33 #include <linux/sched/cputime.h>
34 #include <linux/spinlock.h>
35 #include <linux/eventfd.h>
36 
37 #include <asm/apicdef.h>
38 #include <asm/mshyperv.h>
39 #include <trace/events/kvm.h>
40 
41 #include "trace.h"
42 #include "irq.h"
43 #include "fpu.h"
44 
45 #define KVM_HV_MAX_SPARSE_VCPU_SET_BITS DIV_ROUND_UP(KVM_MAX_VCPUS, HV_VCPUS_PER_SPARSE_BANK)
46 
47 /*
48  * As per Hyper-V TLFS, extended hypercalls start from 0x8001
49  * (HvExtCallQueryCapabilities). Response of this hypercalls is a 64 bit value
50  * where each bit tells which extended hypercall is available besides
51  * HvExtCallQueryCapabilities.
52  *
53  * 0x8001 - First extended hypercall, HvExtCallQueryCapabilities, no bit
54  * assigned.
55  *
56  * 0x8002 - Bit 0
57  * 0x8003 - Bit 1
58  * ..
59  * 0x8041 - Bit 63
60  *
61  * Therefore, HV_EXT_CALL_MAX = 0x8001 + 64
62  */
63 #define HV_EXT_CALL_MAX (HV_EXT_CALL_QUERY_CAPABILITIES + 64)
64 
65 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
66 				bool vcpu_kick);
67 
synic_read_sint(struct kvm_vcpu_hv_synic * synic,int sint)68 static inline u64 synic_read_sint(struct kvm_vcpu_hv_synic *synic, int sint)
69 {
70 	return atomic64_read(&synic->sint[sint]);
71 }
72 
synic_get_sint_vector(u64 sint_value)73 static inline int synic_get_sint_vector(u64 sint_value)
74 {
75 	if (sint_value & HV_SYNIC_SINT_MASKED)
76 		return -1;
77 	return sint_value & HV_SYNIC_SINT_VECTOR_MASK;
78 }
79 
synic_has_vector_connected(struct kvm_vcpu_hv_synic * synic,int vector)80 static bool synic_has_vector_connected(struct kvm_vcpu_hv_synic *synic,
81 				      int vector)
82 {
83 	int i;
84 
85 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
86 		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
87 			return true;
88 	}
89 	return false;
90 }
91 
synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic * synic,int vector)92 static bool synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic *synic,
93 				     int vector)
94 {
95 	int i;
96 	u64 sint_value;
97 
98 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
99 		sint_value = synic_read_sint(synic, i);
100 		if (synic_get_sint_vector(sint_value) == vector &&
101 		    sint_value & HV_SYNIC_SINT_AUTO_EOI)
102 			return true;
103 	}
104 	return false;
105 }
106 
synic_update_vector(struct kvm_vcpu_hv_synic * synic,int vector)107 static void synic_update_vector(struct kvm_vcpu_hv_synic *synic,
108 				int vector)
109 {
110 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
111 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
112 	bool auto_eoi_old, auto_eoi_new;
113 
114 	if (vector < HV_SYNIC_FIRST_VALID_VECTOR)
115 		return;
116 
117 	if (synic_has_vector_connected(synic, vector))
118 		__set_bit(vector, synic->vec_bitmap);
119 	else
120 		__clear_bit(vector, synic->vec_bitmap);
121 
122 	auto_eoi_old = !bitmap_empty(synic->auto_eoi_bitmap, 256);
123 
124 	if (synic_has_vector_auto_eoi(synic, vector))
125 		__set_bit(vector, synic->auto_eoi_bitmap);
126 	else
127 		__clear_bit(vector, synic->auto_eoi_bitmap);
128 
129 	auto_eoi_new = !bitmap_empty(synic->auto_eoi_bitmap, 256);
130 
131 	if (auto_eoi_old == auto_eoi_new)
132 		return;
133 
134 	if (!enable_apicv)
135 		return;
136 
137 	down_write(&vcpu->kvm->arch.apicv_update_lock);
138 
139 	if (auto_eoi_new)
140 		hv->synic_auto_eoi_used++;
141 	else
142 		hv->synic_auto_eoi_used--;
143 
144 	/*
145 	 * Inhibit APICv if any vCPU is using SynIC's AutoEOI, which relies on
146 	 * the hypervisor to manually inject IRQs.
147 	 */
148 	__kvm_set_or_clear_apicv_inhibit(vcpu->kvm,
149 					 APICV_INHIBIT_REASON_HYPERV,
150 					 !!hv->synic_auto_eoi_used);
151 
152 	up_write(&vcpu->kvm->arch.apicv_update_lock);
153 }
154 
synic_set_sint(struct kvm_vcpu_hv_synic * synic,int sint,u64 data,bool host)155 static int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint,
156 			  u64 data, bool host)
157 {
158 	int vector, old_vector;
159 	bool masked;
160 
161 	vector = data & HV_SYNIC_SINT_VECTOR_MASK;
162 	masked = data & HV_SYNIC_SINT_MASKED;
163 
164 	/*
165 	 * Valid vectors are 16-255, however, nested Hyper-V attempts to write
166 	 * default '0x10000' value on boot and this should not #GP. We need to
167 	 * allow zero-initing the register from host as well.
168 	 */
169 	if (vector < HV_SYNIC_FIRST_VALID_VECTOR && !host && !masked)
170 		return 1;
171 	/*
172 	 * Guest may configure multiple SINTs to use the same vector, so
173 	 * we maintain a bitmap of vectors handled by synic, and a
174 	 * bitmap of vectors with auto-eoi behavior.  The bitmaps are
175 	 * updated here, and atomically queried on fast paths.
176 	 */
177 	old_vector = synic_read_sint(synic, sint) & HV_SYNIC_SINT_VECTOR_MASK;
178 
179 	atomic64_set(&synic->sint[sint], data);
180 
181 	synic_update_vector(synic, old_vector);
182 
183 	synic_update_vector(synic, vector);
184 
185 	/* Load SynIC vectors into EOI exit bitmap */
186 	kvm_make_request(KVM_REQ_SCAN_IOAPIC, hv_synic_to_vcpu(synic));
187 	return 0;
188 }
189 
get_vcpu_by_vpidx(struct kvm * kvm,u32 vpidx)190 static struct kvm_vcpu *get_vcpu_by_vpidx(struct kvm *kvm, u32 vpidx)
191 {
192 	struct kvm_vcpu *vcpu = NULL;
193 	unsigned long i;
194 
195 	if (vpidx >= KVM_MAX_VCPUS)
196 		return NULL;
197 
198 	vcpu = kvm_get_vcpu(kvm, vpidx);
199 	if (vcpu && kvm_hv_get_vpindex(vcpu) == vpidx)
200 		return vcpu;
201 	kvm_for_each_vcpu(i, vcpu, kvm)
202 		if (kvm_hv_get_vpindex(vcpu) == vpidx)
203 			return vcpu;
204 	return NULL;
205 }
206 
synic_get(struct kvm * kvm,u32 vpidx)207 static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vpidx)
208 {
209 	struct kvm_vcpu *vcpu;
210 	struct kvm_vcpu_hv_synic *synic;
211 
212 	vcpu = get_vcpu_by_vpidx(kvm, vpidx);
213 	if (!vcpu || !to_hv_vcpu(vcpu))
214 		return NULL;
215 	synic = to_hv_synic(vcpu);
216 	return (synic->active) ? synic : NULL;
217 }
218 
kvm_hv_notify_acked_sint(struct kvm_vcpu * vcpu,u32 sint)219 static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
220 {
221 	struct kvm *kvm = vcpu->kvm;
222 	struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
223 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
224 	struct kvm_vcpu_hv_stimer *stimer;
225 	int gsi, idx;
226 
227 	trace_kvm_hv_notify_acked_sint(vcpu->vcpu_id, sint);
228 
229 	/* Try to deliver pending Hyper-V SynIC timers messages */
230 	for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
231 		stimer = &hv_vcpu->stimer[idx];
232 		if (stimer->msg_pending && stimer->config.enable &&
233 		    !stimer->config.direct_mode &&
234 		    stimer->config.sintx == sint)
235 			stimer_mark_pending(stimer, false);
236 	}
237 
238 	idx = srcu_read_lock(&kvm->irq_srcu);
239 	gsi = atomic_read(&synic->sint_to_gsi[sint]);
240 	if (gsi != -1)
241 		kvm_notify_acked_gsi(kvm, gsi);
242 	srcu_read_unlock(&kvm->irq_srcu, idx);
243 }
244 
synic_exit(struct kvm_vcpu_hv_synic * synic,u32 msr)245 static void synic_exit(struct kvm_vcpu_hv_synic *synic, u32 msr)
246 {
247 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
248 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
249 
250 	hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNIC;
251 	hv_vcpu->exit.u.synic.msr = msr;
252 	hv_vcpu->exit.u.synic.control = synic->control;
253 	hv_vcpu->exit.u.synic.evt_page = synic->evt_page;
254 	hv_vcpu->exit.u.synic.msg_page = synic->msg_page;
255 
256 	kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
257 }
258 
synic_set_msr(struct kvm_vcpu_hv_synic * synic,u32 msr,u64 data,bool host)259 static int synic_set_msr(struct kvm_vcpu_hv_synic *synic,
260 			 u32 msr, u64 data, bool host)
261 {
262 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
263 	int ret;
264 
265 	if (!synic->active && (!host || data))
266 		return 1;
267 
268 	trace_kvm_hv_synic_set_msr(vcpu->vcpu_id, msr, data, host);
269 
270 	ret = 0;
271 	switch (msr) {
272 	case HV_X64_MSR_SCONTROL:
273 		synic->control = data;
274 		if (!host)
275 			synic_exit(synic, msr);
276 		break;
277 	case HV_X64_MSR_SVERSION:
278 		if (!host) {
279 			ret = 1;
280 			break;
281 		}
282 		synic->version = data;
283 		break;
284 	case HV_X64_MSR_SIEFP:
285 		if ((data & HV_SYNIC_SIEFP_ENABLE) && !host &&
286 		    !synic->dont_zero_synic_pages)
287 			if (kvm_clear_guest(vcpu->kvm,
288 					    data & PAGE_MASK, PAGE_SIZE)) {
289 				ret = 1;
290 				break;
291 			}
292 		synic->evt_page = data;
293 		if (!host)
294 			synic_exit(synic, msr);
295 		break;
296 	case HV_X64_MSR_SIMP:
297 		if ((data & HV_SYNIC_SIMP_ENABLE) && !host &&
298 		    !synic->dont_zero_synic_pages)
299 			if (kvm_clear_guest(vcpu->kvm,
300 					    data & PAGE_MASK, PAGE_SIZE)) {
301 				ret = 1;
302 				break;
303 			}
304 		synic->msg_page = data;
305 		if (!host)
306 			synic_exit(synic, msr);
307 		break;
308 	case HV_X64_MSR_EOM: {
309 		int i;
310 
311 		if (!synic->active)
312 			break;
313 
314 		for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
315 			kvm_hv_notify_acked_sint(vcpu, i);
316 		break;
317 	}
318 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
319 		ret = synic_set_sint(synic, msr - HV_X64_MSR_SINT0, data, host);
320 		break;
321 	default:
322 		ret = 1;
323 		break;
324 	}
325 	return ret;
326 }
327 
kvm_hv_is_syndbg_enabled(struct kvm_vcpu * vcpu)328 static bool kvm_hv_is_syndbg_enabled(struct kvm_vcpu *vcpu)
329 {
330 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
331 
332 	return hv_vcpu->cpuid_cache.syndbg_cap_eax &
333 		HV_X64_SYNDBG_CAP_ALLOW_KERNEL_DEBUGGING;
334 }
335 
kvm_hv_syndbg_complete_userspace(struct kvm_vcpu * vcpu)336 static int kvm_hv_syndbg_complete_userspace(struct kvm_vcpu *vcpu)
337 {
338 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
339 
340 	if (vcpu->run->hyperv.u.syndbg.msr == HV_X64_MSR_SYNDBG_CONTROL)
341 		hv->hv_syndbg.control.status =
342 			vcpu->run->hyperv.u.syndbg.status;
343 	return 1;
344 }
345 
syndbg_exit(struct kvm_vcpu * vcpu,u32 msr)346 static void syndbg_exit(struct kvm_vcpu *vcpu, u32 msr)
347 {
348 	struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
349 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
350 
351 	hv_vcpu->exit.type = KVM_EXIT_HYPERV_SYNDBG;
352 	hv_vcpu->exit.u.syndbg.msr = msr;
353 	hv_vcpu->exit.u.syndbg.control = syndbg->control.control;
354 	hv_vcpu->exit.u.syndbg.send_page = syndbg->control.send_page;
355 	hv_vcpu->exit.u.syndbg.recv_page = syndbg->control.recv_page;
356 	hv_vcpu->exit.u.syndbg.pending_page = syndbg->control.pending_page;
357 	vcpu->arch.complete_userspace_io =
358 			kvm_hv_syndbg_complete_userspace;
359 
360 	kvm_make_request(KVM_REQ_HV_EXIT, vcpu);
361 }
362 
syndbg_set_msr(struct kvm_vcpu * vcpu,u32 msr,u64 data,bool host)363 static int syndbg_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
364 {
365 	struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
366 
367 	if (!kvm_hv_is_syndbg_enabled(vcpu) && !host)
368 		return 1;
369 
370 	trace_kvm_hv_syndbg_set_msr(vcpu->vcpu_id,
371 				    to_hv_vcpu(vcpu)->vp_index, msr, data);
372 	switch (msr) {
373 	case HV_X64_MSR_SYNDBG_CONTROL:
374 		syndbg->control.control = data;
375 		if (!host)
376 			syndbg_exit(vcpu, msr);
377 		break;
378 	case HV_X64_MSR_SYNDBG_STATUS:
379 		syndbg->control.status = data;
380 		break;
381 	case HV_X64_MSR_SYNDBG_SEND_BUFFER:
382 		syndbg->control.send_page = data;
383 		break;
384 	case HV_X64_MSR_SYNDBG_RECV_BUFFER:
385 		syndbg->control.recv_page = data;
386 		break;
387 	case HV_X64_MSR_SYNDBG_PENDING_BUFFER:
388 		syndbg->control.pending_page = data;
389 		if (!host)
390 			syndbg_exit(vcpu, msr);
391 		break;
392 	case HV_X64_MSR_SYNDBG_OPTIONS:
393 		syndbg->options = data;
394 		break;
395 	default:
396 		break;
397 	}
398 
399 	return 0;
400 }
401 
syndbg_get_msr(struct kvm_vcpu * vcpu,u32 msr,u64 * pdata,bool host)402 static int syndbg_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
403 {
404 	struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
405 
406 	if (!kvm_hv_is_syndbg_enabled(vcpu) && !host)
407 		return 1;
408 
409 	switch (msr) {
410 	case HV_X64_MSR_SYNDBG_CONTROL:
411 		*pdata = syndbg->control.control;
412 		break;
413 	case HV_X64_MSR_SYNDBG_STATUS:
414 		*pdata = syndbg->control.status;
415 		break;
416 	case HV_X64_MSR_SYNDBG_SEND_BUFFER:
417 		*pdata = syndbg->control.send_page;
418 		break;
419 	case HV_X64_MSR_SYNDBG_RECV_BUFFER:
420 		*pdata = syndbg->control.recv_page;
421 		break;
422 	case HV_X64_MSR_SYNDBG_PENDING_BUFFER:
423 		*pdata = syndbg->control.pending_page;
424 		break;
425 	case HV_X64_MSR_SYNDBG_OPTIONS:
426 		*pdata = syndbg->options;
427 		break;
428 	default:
429 		break;
430 	}
431 
432 	trace_kvm_hv_syndbg_get_msr(vcpu->vcpu_id, kvm_hv_get_vpindex(vcpu), msr, *pdata);
433 
434 	return 0;
435 }
436 
synic_get_msr(struct kvm_vcpu_hv_synic * synic,u32 msr,u64 * pdata,bool host)437 static int synic_get_msr(struct kvm_vcpu_hv_synic *synic, u32 msr, u64 *pdata,
438 			 bool host)
439 {
440 	int ret;
441 
442 	if (!synic->active && !host)
443 		return 1;
444 
445 	ret = 0;
446 	switch (msr) {
447 	case HV_X64_MSR_SCONTROL:
448 		*pdata = synic->control;
449 		break;
450 	case HV_X64_MSR_SVERSION:
451 		*pdata = synic->version;
452 		break;
453 	case HV_X64_MSR_SIEFP:
454 		*pdata = synic->evt_page;
455 		break;
456 	case HV_X64_MSR_SIMP:
457 		*pdata = synic->msg_page;
458 		break;
459 	case HV_X64_MSR_EOM:
460 		*pdata = 0;
461 		break;
462 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
463 		*pdata = atomic64_read(&synic->sint[msr - HV_X64_MSR_SINT0]);
464 		break;
465 	default:
466 		ret = 1;
467 		break;
468 	}
469 	return ret;
470 }
471 
synic_set_irq(struct kvm_vcpu_hv_synic * synic,u32 sint)472 static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
473 {
474 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
475 	struct kvm_lapic_irq irq;
476 	int ret, vector;
477 
478 	if (KVM_BUG_ON(!lapic_in_kernel(vcpu), vcpu->kvm))
479 		return -EINVAL;
480 
481 	if (sint >= ARRAY_SIZE(synic->sint))
482 		return -EINVAL;
483 
484 	vector = synic_get_sint_vector(synic_read_sint(synic, sint));
485 	if (vector < 0)
486 		return -ENOENT;
487 
488 	memset(&irq, 0, sizeof(irq));
489 	irq.shorthand = APIC_DEST_SELF;
490 	irq.dest_mode = APIC_DEST_PHYSICAL;
491 	irq.delivery_mode = APIC_DM_FIXED;
492 	irq.vector = vector;
493 	irq.level = 1;
494 
495 	ret = kvm_irq_delivery_to_apic(vcpu->kvm, vcpu->arch.apic, &irq, NULL);
496 	trace_kvm_hv_synic_set_irq(vcpu->vcpu_id, sint, irq.vector, ret);
497 	return ret;
498 }
499 
kvm_hv_synic_set_irq(struct kvm * kvm,u32 vpidx,u32 sint)500 int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vpidx, u32 sint)
501 {
502 	struct kvm_vcpu_hv_synic *synic;
503 
504 	synic = synic_get(kvm, vpidx);
505 	if (!synic)
506 		return -EINVAL;
507 
508 	return synic_set_irq(synic, sint);
509 }
510 
kvm_hv_synic_send_eoi(struct kvm_vcpu * vcpu,int vector)511 void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
512 {
513 	struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
514 	int i;
515 
516 	trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
517 
518 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
519 		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
520 			kvm_hv_notify_acked_sint(vcpu, i);
521 }
522 
kvm_hv_set_sint_gsi(struct kvm * kvm,u32 vpidx,u32 sint,int gsi)523 static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi)
524 {
525 	struct kvm_vcpu_hv_synic *synic;
526 
527 	synic = synic_get(kvm, vpidx);
528 	if (!synic)
529 		return -EINVAL;
530 
531 	if (sint >= ARRAY_SIZE(synic->sint_to_gsi))
532 		return -EINVAL;
533 
534 	atomic_set(&synic->sint_to_gsi[sint], gsi);
535 	return 0;
536 }
537 
kvm_hv_irq_routing_update(struct kvm * kvm)538 void kvm_hv_irq_routing_update(struct kvm *kvm)
539 {
540 	struct kvm_irq_routing_table *irq_rt;
541 	struct kvm_kernel_irq_routing_entry *e;
542 	u32 gsi;
543 
544 	irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
545 					lockdep_is_held(&kvm->irq_lock));
546 
547 	for (gsi = 0; gsi < irq_rt->nr_rt_entries; gsi++) {
548 		hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
549 			if (e->type == KVM_IRQ_ROUTING_HV_SINT)
550 				kvm_hv_set_sint_gsi(kvm, e->hv_sint.vcpu,
551 						    e->hv_sint.sint, gsi);
552 		}
553 	}
554 }
555 
synic_init(struct kvm_vcpu_hv_synic * synic)556 static void synic_init(struct kvm_vcpu_hv_synic *synic)
557 {
558 	int i;
559 
560 	memset(synic, 0, sizeof(*synic));
561 	synic->version = HV_SYNIC_VERSION_1;
562 	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
563 		atomic64_set(&synic->sint[i], HV_SYNIC_SINT_MASKED);
564 		atomic_set(&synic->sint_to_gsi[i], -1);
565 	}
566 }
567 
get_time_ref_counter(struct kvm * kvm)568 static u64 get_time_ref_counter(struct kvm *kvm)
569 {
570 	struct kvm_hv *hv = to_kvm_hv(kvm);
571 	struct kvm_vcpu *vcpu;
572 	u64 tsc;
573 
574 	/*
575 	 * Fall back to get_kvmclock_ns() when TSC page hasn't been set up,
576 	 * is broken, disabled or being updated.
577 	 */
578 	if (hv->hv_tsc_page_status != HV_TSC_PAGE_SET)
579 		return div_u64(get_kvmclock_ns(kvm), 100);
580 
581 	vcpu = kvm_get_vcpu(kvm, 0);
582 	tsc = kvm_read_l1_tsc(vcpu, rdtsc());
583 	return mul_u64_u64_shr(tsc, hv->tsc_ref.tsc_scale, 64)
584 		+ hv->tsc_ref.tsc_offset;
585 }
586 
stimer_mark_pending(struct kvm_vcpu_hv_stimer * stimer,bool vcpu_kick)587 static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
588 				bool vcpu_kick)
589 {
590 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
591 
592 	set_bit(stimer->index,
593 		to_hv_vcpu(vcpu)->stimer_pending_bitmap);
594 	kvm_make_request(KVM_REQ_HV_STIMER, vcpu);
595 	if (vcpu_kick)
596 		kvm_vcpu_kick(vcpu);
597 }
598 
stimer_cleanup(struct kvm_vcpu_hv_stimer * stimer)599 static void stimer_cleanup(struct kvm_vcpu_hv_stimer *stimer)
600 {
601 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
602 
603 	trace_kvm_hv_stimer_cleanup(hv_stimer_to_vcpu(stimer)->vcpu_id,
604 				    stimer->index);
605 
606 	hrtimer_cancel(&stimer->timer);
607 	clear_bit(stimer->index,
608 		  to_hv_vcpu(vcpu)->stimer_pending_bitmap);
609 	stimer->msg_pending = false;
610 	stimer->exp_time = 0;
611 }
612 
stimer_timer_callback(struct hrtimer * timer)613 static enum hrtimer_restart stimer_timer_callback(struct hrtimer *timer)
614 {
615 	struct kvm_vcpu_hv_stimer *stimer;
616 
617 	stimer = container_of(timer, struct kvm_vcpu_hv_stimer, timer);
618 	trace_kvm_hv_stimer_callback(hv_stimer_to_vcpu(stimer)->vcpu_id,
619 				     stimer->index);
620 	stimer_mark_pending(stimer, true);
621 
622 	return HRTIMER_NORESTART;
623 }
624 
625 /*
626  * stimer_start() assumptions:
627  * a) stimer->count is not equal to 0
628  * b) stimer->config has HV_STIMER_ENABLE flag
629  */
stimer_start(struct kvm_vcpu_hv_stimer * stimer)630 static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
631 {
632 	u64 time_now;
633 	ktime_t ktime_now;
634 
635 	time_now = get_time_ref_counter(hv_stimer_to_vcpu(stimer)->kvm);
636 	ktime_now = ktime_get();
637 
638 	if (stimer->config.periodic) {
639 		if (stimer->exp_time) {
640 			if (time_now >= stimer->exp_time) {
641 				u64 remainder;
642 
643 				div64_u64_rem(time_now - stimer->exp_time,
644 					      stimer->count, &remainder);
645 				stimer->exp_time =
646 					time_now + (stimer->count - remainder);
647 			}
648 		} else
649 			stimer->exp_time = time_now + stimer->count;
650 
651 		trace_kvm_hv_stimer_start_periodic(
652 					hv_stimer_to_vcpu(stimer)->vcpu_id,
653 					stimer->index,
654 					time_now, stimer->exp_time);
655 
656 		hrtimer_start(&stimer->timer,
657 			      ktime_add_ns(ktime_now,
658 					   100 * (stimer->exp_time - time_now)),
659 			      HRTIMER_MODE_ABS);
660 		return 0;
661 	}
662 	stimer->exp_time = stimer->count;
663 	if (time_now >= stimer->count) {
664 		/*
665 		 * Expire timer according to Hypervisor Top-Level Functional
666 		 * specification v4(15.3.1):
667 		 * "If a one shot is enabled and the specified count is in
668 		 * the past, it will expire immediately."
669 		 */
670 		stimer_mark_pending(stimer, false);
671 		return 0;
672 	}
673 
674 	trace_kvm_hv_stimer_start_one_shot(hv_stimer_to_vcpu(stimer)->vcpu_id,
675 					   stimer->index,
676 					   time_now, stimer->count);
677 
678 	hrtimer_start(&stimer->timer,
679 		      ktime_add_ns(ktime_now, 100 * (stimer->count - time_now)),
680 		      HRTIMER_MODE_ABS);
681 	return 0;
682 }
683 
stimer_set_config(struct kvm_vcpu_hv_stimer * stimer,u64 config,bool host)684 static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
685 			     bool host)
686 {
687 	union hv_stimer_config new_config = {.as_uint64 = config},
688 		old_config = {.as_uint64 = stimer->config.as_uint64};
689 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
690 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
691 	struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
692 
693 	if (!synic->active && (!host || config))
694 		return 1;
695 
696 	if (unlikely(!host && hv_vcpu->enforce_cpuid && new_config.direct_mode &&
697 		     !(hv_vcpu->cpuid_cache.features_edx &
698 		       HV_STIMER_DIRECT_MODE_AVAILABLE)))
699 		return 1;
700 
701 	trace_kvm_hv_stimer_set_config(hv_stimer_to_vcpu(stimer)->vcpu_id,
702 				       stimer->index, config, host);
703 
704 	stimer_cleanup(stimer);
705 	if (old_config.enable &&
706 	    !new_config.direct_mode && new_config.sintx == 0)
707 		new_config.enable = 0;
708 	stimer->config.as_uint64 = new_config.as_uint64;
709 
710 	if (stimer->config.enable)
711 		stimer_mark_pending(stimer, false);
712 
713 	return 0;
714 }
715 
stimer_set_count(struct kvm_vcpu_hv_stimer * stimer,u64 count,bool host)716 static int stimer_set_count(struct kvm_vcpu_hv_stimer *stimer, u64 count,
717 			    bool host)
718 {
719 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
720 	struct kvm_vcpu_hv_synic *synic = to_hv_synic(vcpu);
721 
722 	if (!synic->active && (!host || count))
723 		return 1;
724 
725 	trace_kvm_hv_stimer_set_count(hv_stimer_to_vcpu(stimer)->vcpu_id,
726 				      stimer->index, count, host);
727 
728 	stimer_cleanup(stimer);
729 	stimer->count = count;
730 	if (!host) {
731 		if (stimer->count == 0)
732 			stimer->config.enable = 0;
733 		else if (stimer->config.auto_enable)
734 			stimer->config.enable = 1;
735 	}
736 
737 	if (stimer->config.enable)
738 		stimer_mark_pending(stimer, false);
739 
740 	return 0;
741 }
742 
stimer_get_config(struct kvm_vcpu_hv_stimer * stimer,u64 * pconfig)743 static int stimer_get_config(struct kvm_vcpu_hv_stimer *stimer, u64 *pconfig)
744 {
745 	*pconfig = stimer->config.as_uint64;
746 	return 0;
747 }
748 
stimer_get_count(struct kvm_vcpu_hv_stimer * stimer,u64 * pcount)749 static int stimer_get_count(struct kvm_vcpu_hv_stimer *stimer, u64 *pcount)
750 {
751 	*pcount = stimer->count;
752 	return 0;
753 }
754 
synic_deliver_msg(struct kvm_vcpu_hv_synic * synic,u32 sint,struct hv_message * src_msg,bool no_retry)755 static int synic_deliver_msg(struct kvm_vcpu_hv_synic *synic, u32 sint,
756 			     struct hv_message *src_msg, bool no_retry)
757 {
758 	struct kvm_vcpu *vcpu = hv_synic_to_vcpu(synic);
759 	int msg_off = offsetof(struct hv_message_page, sint_message[sint]);
760 	gfn_t msg_page_gfn;
761 	struct hv_message_header hv_hdr;
762 	int r;
763 
764 	if (!(synic->msg_page & HV_SYNIC_SIMP_ENABLE))
765 		return -ENOENT;
766 
767 	msg_page_gfn = synic->msg_page >> PAGE_SHIFT;
768 
769 	/*
770 	 * Strictly following the spec-mandated ordering would assume setting
771 	 * .msg_pending before checking .message_type.  However, this function
772 	 * is only called in vcpu context so the entire update is atomic from
773 	 * guest POV and thus the exact order here doesn't matter.
774 	 */
775 	r = kvm_vcpu_read_guest_page(vcpu, msg_page_gfn, &hv_hdr.message_type,
776 				     msg_off + offsetof(struct hv_message,
777 							header.message_type),
778 				     sizeof(hv_hdr.message_type));
779 	if (r < 0)
780 		return r;
781 
782 	if (hv_hdr.message_type != HVMSG_NONE) {
783 		if (no_retry)
784 			return 0;
785 
786 		hv_hdr.message_flags.msg_pending = 1;
787 		r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn,
788 					      &hv_hdr.message_flags,
789 					      msg_off +
790 					      offsetof(struct hv_message,
791 						       header.message_flags),
792 					      sizeof(hv_hdr.message_flags));
793 		if (r < 0)
794 			return r;
795 		return -EAGAIN;
796 	}
797 
798 	r = kvm_vcpu_write_guest_page(vcpu, msg_page_gfn, src_msg, msg_off,
799 				      sizeof(src_msg->header) +
800 				      src_msg->header.payload_size);
801 	if (r < 0)
802 		return r;
803 
804 	r = synic_set_irq(synic, sint);
805 	if (r < 0)
806 		return r;
807 	if (r == 0)
808 		return -EFAULT;
809 	return 0;
810 }
811 
stimer_send_msg(struct kvm_vcpu_hv_stimer * stimer)812 static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
813 {
814 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
815 	struct hv_message *msg = &stimer->msg;
816 	struct hv_timer_message_payload *payload =
817 			(struct hv_timer_message_payload *)&msg->u.payload;
818 
819 	/*
820 	 * To avoid piling up periodic ticks, don't retry message
821 	 * delivery for them (within "lazy" lost ticks policy).
822 	 */
823 	bool no_retry = stimer->config.periodic;
824 
825 	payload->expiration_time = stimer->exp_time;
826 	payload->delivery_time = get_time_ref_counter(vcpu->kvm);
827 	return synic_deliver_msg(to_hv_synic(vcpu),
828 				 stimer->config.sintx, msg,
829 				 no_retry);
830 }
831 
stimer_notify_direct(struct kvm_vcpu_hv_stimer * stimer)832 static int stimer_notify_direct(struct kvm_vcpu_hv_stimer *stimer)
833 {
834 	struct kvm_vcpu *vcpu = hv_stimer_to_vcpu(stimer);
835 	struct kvm_lapic_irq irq = {
836 		.delivery_mode = APIC_DM_FIXED,
837 		.vector = stimer->config.apic_vector
838 	};
839 
840 	if (lapic_in_kernel(vcpu))
841 		return !kvm_apic_set_irq(vcpu, &irq, NULL);
842 	return 0;
843 }
844 
stimer_expiration(struct kvm_vcpu_hv_stimer * stimer)845 static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
846 {
847 	int r, direct = stimer->config.direct_mode;
848 
849 	stimer->msg_pending = true;
850 	if (!direct)
851 		r = stimer_send_msg(stimer);
852 	else
853 		r = stimer_notify_direct(stimer);
854 	trace_kvm_hv_stimer_expiration(hv_stimer_to_vcpu(stimer)->vcpu_id,
855 				       stimer->index, direct, r);
856 	if (!r) {
857 		stimer->msg_pending = false;
858 		if (!(stimer->config.periodic))
859 			stimer->config.enable = 0;
860 	}
861 }
862 
kvm_hv_process_stimers(struct kvm_vcpu * vcpu)863 void kvm_hv_process_stimers(struct kvm_vcpu *vcpu)
864 {
865 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
866 	struct kvm_vcpu_hv_stimer *stimer;
867 	u64 time_now, exp_time;
868 	int i;
869 
870 	if (!hv_vcpu)
871 		return;
872 
873 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
874 		if (test_and_clear_bit(i, hv_vcpu->stimer_pending_bitmap)) {
875 			stimer = &hv_vcpu->stimer[i];
876 			if (stimer->config.enable) {
877 				exp_time = stimer->exp_time;
878 
879 				if (exp_time) {
880 					time_now =
881 						get_time_ref_counter(vcpu->kvm);
882 					if (time_now >= exp_time)
883 						stimer_expiration(stimer);
884 				}
885 
886 				if ((stimer->config.enable) &&
887 				    stimer->count) {
888 					if (!stimer->msg_pending)
889 						stimer_start(stimer);
890 				} else
891 					stimer_cleanup(stimer);
892 			}
893 		}
894 }
895 
kvm_hv_vcpu_uninit(struct kvm_vcpu * vcpu)896 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu)
897 {
898 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
899 	int i;
900 
901 	if (!hv_vcpu)
902 		return;
903 
904 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
905 		stimer_cleanup(&hv_vcpu->stimer[i]);
906 
907 	kfree(hv_vcpu);
908 	vcpu->arch.hyperv = NULL;
909 }
910 
kvm_hv_assist_page_enabled(struct kvm_vcpu * vcpu)911 bool kvm_hv_assist_page_enabled(struct kvm_vcpu *vcpu)
912 {
913 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
914 
915 	if (!hv_vcpu)
916 		return false;
917 
918 	if (!(hv_vcpu->hv_vapic & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE))
919 		return false;
920 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
921 }
922 EXPORT_SYMBOL_GPL(kvm_hv_assist_page_enabled);
923 
kvm_hv_get_assist_page(struct kvm_vcpu * vcpu)924 int kvm_hv_get_assist_page(struct kvm_vcpu *vcpu)
925 {
926 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
927 
928 	if (!hv_vcpu || !kvm_hv_assist_page_enabled(vcpu))
929 		return -EFAULT;
930 
931 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data,
932 				     &hv_vcpu->vp_assist_page, sizeof(struct hv_vp_assist_page));
933 }
934 EXPORT_SYMBOL_GPL(kvm_hv_get_assist_page);
935 
stimer_prepare_msg(struct kvm_vcpu_hv_stimer * stimer)936 static void stimer_prepare_msg(struct kvm_vcpu_hv_stimer *stimer)
937 {
938 	struct hv_message *msg = &stimer->msg;
939 	struct hv_timer_message_payload *payload =
940 			(struct hv_timer_message_payload *)&msg->u.payload;
941 
942 	memset(&msg->header, 0, sizeof(msg->header));
943 	msg->header.message_type = HVMSG_TIMER_EXPIRED;
944 	msg->header.payload_size = sizeof(*payload);
945 
946 	payload->timer_index = stimer->index;
947 	payload->expiration_time = 0;
948 	payload->delivery_time = 0;
949 }
950 
stimer_init(struct kvm_vcpu_hv_stimer * stimer,int timer_index)951 static void stimer_init(struct kvm_vcpu_hv_stimer *stimer, int timer_index)
952 {
953 	memset(stimer, 0, sizeof(*stimer));
954 	stimer->index = timer_index;
955 	hrtimer_init(&stimer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
956 	stimer->timer.function = stimer_timer_callback;
957 	stimer_prepare_msg(stimer);
958 }
959 
kvm_hv_vcpu_init(struct kvm_vcpu * vcpu)960 int kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
961 {
962 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
963 	int i;
964 
965 	if (hv_vcpu)
966 		return 0;
967 
968 	hv_vcpu = kzalloc(sizeof(struct kvm_vcpu_hv), GFP_KERNEL_ACCOUNT);
969 	if (!hv_vcpu)
970 		return -ENOMEM;
971 
972 	vcpu->arch.hyperv = hv_vcpu;
973 	hv_vcpu->vcpu = vcpu;
974 
975 	synic_init(&hv_vcpu->synic);
976 
977 	bitmap_zero(hv_vcpu->stimer_pending_bitmap, HV_SYNIC_STIMER_COUNT);
978 	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++)
979 		stimer_init(&hv_vcpu->stimer[i], i);
980 
981 	hv_vcpu->vp_index = vcpu->vcpu_idx;
982 
983 	for (i = 0; i < HV_NR_TLB_FLUSH_FIFOS; i++) {
984 		INIT_KFIFO(hv_vcpu->tlb_flush_fifo[i].entries);
985 		spin_lock_init(&hv_vcpu->tlb_flush_fifo[i].write_lock);
986 	}
987 
988 	return 0;
989 }
990 
kvm_hv_activate_synic(struct kvm_vcpu * vcpu,bool dont_zero_synic_pages)991 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
992 {
993 	struct kvm_vcpu_hv_synic *synic;
994 	int r;
995 
996 	r = kvm_hv_vcpu_init(vcpu);
997 	if (r)
998 		return r;
999 
1000 	synic = to_hv_synic(vcpu);
1001 
1002 	synic->active = true;
1003 	synic->dont_zero_synic_pages = dont_zero_synic_pages;
1004 	synic->control = HV_SYNIC_CONTROL_ENABLE;
1005 	return 0;
1006 }
1007 
kvm_hv_msr_partition_wide(u32 msr)1008 static bool kvm_hv_msr_partition_wide(u32 msr)
1009 {
1010 	bool r = false;
1011 
1012 	switch (msr) {
1013 	case HV_X64_MSR_GUEST_OS_ID:
1014 	case HV_X64_MSR_HYPERCALL:
1015 	case HV_X64_MSR_REFERENCE_TSC:
1016 	case HV_X64_MSR_TIME_REF_COUNT:
1017 	case HV_X64_MSR_CRASH_CTL:
1018 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1019 	case HV_X64_MSR_RESET:
1020 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1021 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1022 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1023 	case HV_X64_MSR_TSC_INVARIANT_CONTROL:
1024 	case HV_X64_MSR_SYNDBG_OPTIONS:
1025 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1026 		r = true;
1027 		break;
1028 	}
1029 
1030 	return r;
1031 }
1032 
kvm_hv_msr_get_crash_data(struct kvm * kvm,u32 index,u64 * pdata)1033 static int kvm_hv_msr_get_crash_data(struct kvm *kvm, u32 index, u64 *pdata)
1034 {
1035 	struct kvm_hv *hv = to_kvm_hv(kvm);
1036 	size_t size = ARRAY_SIZE(hv->hv_crash_param);
1037 
1038 	if (WARN_ON_ONCE(index >= size))
1039 		return -EINVAL;
1040 
1041 	*pdata = hv->hv_crash_param[array_index_nospec(index, size)];
1042 	return 0;
1043 }
1044 
kvm_hv_msr_get_crash_ctl(struct kvm * kvm,u64 * pdata)1045 static int kvm_hv_msr_get_crash_ctl(struct kvm *kvm, u64 *pdata)
1046 {
1047 	struct kvm_hv *hv = to_kvm_hv(kvm);
1048 
1049 	*pdata = hv->hv_crash_ctl;
1050 	return 0;
1051 }
1052 
kvm_hv_msr_set_crash_ctl(struct kvm * kvm,u64 data)1053 static int kvm_hv_msr_set_crash_ctl(struct kvm *kvm, u64 data)
1054 {
1055 	struct kvm_hv *hv = to_kvm_hv(kvm);
1056 
1057 	hv->hv_crash_ctl = data & HV_CRASH_CTL_CRASH_NOTIFY;
1058 
1059 	return 0;
1060 }
1061 
kvm_hv_msr_set_crash_data(struct kvm * kvm,u32 index,u64 data)1062 static int kvm_hv_msr_set_crash_data(struct kvm *kvm, u32 index, u64 data)
1063 {
1064 	struct kvm_hv *hv = to_kvm_hv(kvm);
1065 	size_t size = ARRAY_SIZE(hv->hv_crash_param);
1066 
1067 	if (WARN_ON_ONCE(index >= size))
1068 		return -EINVAL;
1069 
1070 	hv->hv_crash_param[array_index_nospec(index, size)] = data;
1071 	return 0;
1072 }
1073 
1074 /*
1075  * The kvmclock and Hyper-V TSC page use similar formulas, and converting
1076  * between them is possible:
1077  *
1078  * kvmclock formula:
1079  *    nsec = (ticks - tsc_timestamp) * tsc_to_system_mul * 2^(tsc_shift-32)
1080  *           + system_time
1081  *
1082  * Hyper-V formula:
1083  *    nsec/100 = ticks * scale / 2^64 + offset
1084  *
1085  * When tsc_timestamp = system_time = 0, offset is zero in the Hyper-V formula.
1086  * By dividing the kvmclock formula by 100 and equating what's left we get:
1087  *    ticks * scale / 2^64 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1088  *            scale / 2^64 =         tsc_to_system_mul * 2^(tsc_shift-32) / 100
1089  *            scale        =         tsc_to_system_mul * 2^(32+tsc_shift) / 100
1090  *
1091  * Now expand the kvmclock formula and divide by 100:
1092  *    nsec = ticks * tsc_to_system_mul * 2^(tsc_shift-32)
1093  *           - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32)
1094  *           + system_time
1095  *    nsec/100 = ticks * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1096  *               - tsc_timestamp * tsc_to_system_mul * 2^(tsc_shift-32) / 100
1097  *               + system_time / 100
1098  *
1099  * Replace tsc_to_system_mul * 2^(tsc_shift-32) / 100 by scale / 2^64:
1100  *    nsec/100 = ticks * scale / 2^64
1101  *               - tsc_timestamp * scale / 2^64
1102  *               + system_time / 100
1103  *
1104  * Equate with the Hyper-V formula so that ticks * scale / 2^64 cancels out:
1105  *    offset = system_time / 100 - tsc_timestamp * scale / 2^64
1106  *
1107  * These two equivalencies are implemented in this function.
1108  */
compute_tsc_page_parameters(struct pvclock_vcpu_time_info * hv_clock,struct ms_hyperv_tsc_page * tsc_ref)1109 static bool compute_tsc_page_parameters(struct pvclock_vcpu_time_info *hv_clock,
1110 					struct ms_hyperv_tsc_page *tsc_ref)
1111 {
1112 	u64 max_mul;
1113 
1114 	if (!(hv_clock->flags & PVCLOCK_TSC_STABLE_BIT))
1115 		return false;
1116 
1117 	/*
1118 	 * check if scale would overflow, if so we use the time ref counter
1119 	 *    tsc_to_system_mul * 2^(tsc_shift+32) / 100 >= 2^64
1120 	 *    tsc_to_system_mul / 100 >= 2^(32-tsc_shift)
1121 	 *    tsc_to_system_mul >= 100 * 2^(32-tsc_shift)
1122 	 */
1123 	max_mul = 100ull << (32 - hv_clock->tsc_shift);
1124 	if (hv_clock->tsc_to_system_mul >= max_mul)
1125 		return false;
1126 
1127 	/*
1128 	 * Otherwise compute the scale and offset according to the formulas
1129 	 * derived above.
1130 	 */
1131 	tsc_ref->tsc_scale =
1132 		mul_u64_u32_div(1ULL << (32 + hv_clock->tsc_shift),
1133 				hv_clock->tsc_to_system_mul,
1134 				100);
1135 
1136 	tsc_ref->tsc_offset = hv_clock->system_time;
1137 	do_div(tsc_ref->tsc_offset, 100);
1138 	tsc_ref->tsc_offset -=
1139 		mul_u64_u64_shr(hv_clock->tsc_timestamp, tsc_ref->tsc_scale, 64);
1140 	return true;
1141 }
1142 
1143 /*
1144  * Don't touch TSC page values if the guest has opted for TSC emulation after
1145  * migration. KVM doesn't fully support reenlightenment notifications and TSC
1146  * access emulation and Hyper-V is known to expect the values in TSC page to
1147  * stay constant before TSC access emulation is disabled from guest side
1148  * (HV_X64_MSR_TSC_EMULATION_STATUS). KVM userspace is expected to preserve TSC
1149  * frequency and guest visible TSC value across migration (and prevent it when
1150  * TSC scaling is unsupported).
1151  */
tsc_page_update_unsafe(struct kvm_hv * hv)1152 static inline bool tsc_page_update_unsafe(struct kvm_hv *hv)
1153 {
1154 	return (hv->hv_tsc_page_status != HV_TSC_PAGE_GUEST_CHANGED) &&
1155 		hv->hv_tsc_emulation_control;
1156 }
1157 
kvm_hv_setup_tsc_page(struct kvm * kvm,struct pvclock_vcpu_time_info * hv_clock)1158 void kvm_hv_setup_tsc_page(struct kvm *kvm,
1159 			   struct pvclock_vcpu_time_info *hv_clock)
1160 {
1161 	struct kvm_hv *hv = to_kvm_hv(kvm);
1162 	u32 tsc_seq;
1163 	u64 gfn;
1164 
1165 	BUILD_BUG_ON(sizeof(tsc_seq) != sizeof(hv->tsc_ref.tsc_sequence));
1166 	BUILD_BUG_ON(offsetof(struct ms_hyperv_tsc_page, tsc_sequence) != 0);
1167 
1168 	mutex_lock(&hv->hv_lock);
1169 
1170 	if (hv->hv_tsc_page_status == HV_TSC_PAGE_BROKEN ||
1171 	    hv->hv_tsc_page_status == HV_TSC_PAGE_SET ||
1172 	    hv->hv_tsc_page_status == HV_TSC_PAGE_UNSET)
1173 		goto out_unlock;
1174 
1175 	if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1176 		goto out_unlock;
1177 
1178 	gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
1179 	/*
1180 	 * Because the TSC parameters only vary when there is a
1181 	 * change in the master clock, do not bother with caching.
1182 	 */
1183 	if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
1184 				    &tsc_seq, sizeof(tsc_seq))))
1185 		goto out_err;
1186 
1187 	if (tsc_seq && tsc_page_update_unsafe(hv)) {
1188 		if (kvm_read_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
1189 			goto out_err;
1190 
1191 		hv->hv_tsc_page_status = HV_TSC_PAGE_SET;
1192 		goto out_unlock;
1193 	}
1194 
1195 	/*
1196 	 * While we're computing and writing the parameters, force the
1197 	 * guest to use the time reference count MSR.
1198 	 */
1199 	hv->tsc_ref.tsc_sequence = 0;
1200 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
1201 			    &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
1202 		goto out_err;
1203 
1204 	if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
1205 		goto out_err;
1206 
1207 	/* Ensure sequence is zero before writing the rest of the struct.  */
1208 	smp_wmb();
1209 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
1210 		goto out_err;
1211 
1212 	/*
1213 	 * Now switch to the TSC page mechanism by writing the sequence.
1214 	 */
1215 	tsc_seq++;
1216 	if (tsc_seq == 0xFFFFFFFF || tsc_seq == 0)
1217 		tsc_seq = 1;
1218 
1219 	/* Write the struct entirely before the non-zero sequence.  */
1220 	smp_wmb();
1221 
1222 	hv->tsc_ref.tsc_sequence = tsc_seq;
1223 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
1224 			    &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
1225 		goto out_err;
1226 
1227 	hv->hv_tsc_page_status = HV_TSC_PAGE_SET;
1228 	goto out_unlock;
1229 
1230 out_err:
1231 	hv->hv_tsc_page_status = HV_TSC_PAGE_BROKEN;
1232 out_unlock:
1233 	mutex_unlock(&hv->hv_lock);
1234 }
1235 
kvm_hv_request_tsc_page_update(struct kvm * kvm)1236 void kvm_hv_request_tsc_page_update(struct kvm *kvm)
1237 {
1238 	struct kvm_hv *hv = to_kvm_hv(kvm);
1239 
1240 	mutex_lock(&hv->hv_lock);
1241 
1242 	if (hv->hv_tsc_page_status == HV_TSC_PAGE_SET &&
1243 	    !tsc_page_update_unsafe(hv))
1244 		hv->hv_tsc_page_status = HV_TSC_PAGE_HOST_CHANGED;
1245 
1246 	mutex_unlock(&hv->hv_lock);
1247 }
1248 
hv_check_msr_access(struct kvm_vcpu_hv * hv_vcpu,u32 msr)1249 static bool hv_check_msr_access(struct kvm_vcpu_hv *hv_vcpu, u32 msr)
1250 {
1251 	if (!hv_vcpu->enforce_cpuid)
1252 		return true;
1253 
1254 	switch (msr) {
1255 	case HV_X64_MSR_GUEST_OS_ID:
1256 	case HV_X64_MSR_HYPERCALL:
1257 		return hv_vcpu->cpuid_cache.features_eax &
1258 			HV_MSR_HYPERCALL_AVAILABLE;
1259 	case HV_X64_MSR_VP_RUNTIME:
1260 		return hv_vcpu->cpuid_cache.features_eax &
1261 			HV_MSR_VP_RUNTIME_AVAILABLE;
1262 	case HV_X64_MSR_TIME_REF_COUNT:
1263 		return hv_vcpu->cpuid_cache.features_eax &
1264 			HV_MSR_TIME_REF_COUNT_AVAILABLE;
1265 	case HV_X64_MSR_VP_INDEX:
1266 		return hv_vcpu->cpuid_cache.features_eax &
1267 			HV_MSR_VP_INDEX_AVAILABLE;
1268 	case HV_X64_MSR_RESET:
1269 		return hv_vcpu->cpuid_cache.features_eax &
1270 			HV_MSR_RESET_AVAILABLE;
1271 	case HV_X64_MSR_REFERENCE_TSC:
1272 		return hv_vcpu->cpuid_cache.features_eax &
1273 			HV_MSR_REFERENCE_TSC_AVAILABLE;
1274 	case HV_X64_MSR_SCONTROL:
1275 	case HV_X64_MSR_SVERSION:
1276 	case HV_X64_MSR_SIEFP:
1277 	case HV_X64_MSR_SIMP:
1278 	case HV_X64_MSR_EOM:
1279 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1280 		return hv_vcpu->cpuid_cache.features_eax &
1281 			HV_MSR_SYNIC_AVAILABLE;
1282 	case HV_X64_MSR_STIMER0_CONFIG:
1283 	case HV_X64_MSR_STIMER1_CONFIG:
1284 	case HV_X64_MSR_STIMER2_CONFIG:
1285 	case HV_X64_MSR_STIMER3_CONFIG:
1286 	case HV_X64_MSR_STIMER0_COUNT:
1287 	case HV_X64_MSR_STIMER1_COUNT:
1288 	case HV_X64_MSR_STIMER2_COUNT:
1289 	case HV_X64_MSR_STIMER3_COUNT:
1290 		return hv_vcpu->cpuid_cache.features_eax &
1291 			HV_MSR_SYNTIMER_AVAILABLE;
1292 	case HV_X64_MSR_EOI:
1293 	case HV_X64_MSR_ICR:
1294 	case HV_X64_MSR_TPR:
1295 	case HV_X64_MSR_VP_ASSIST_PAGE:
1296 		return hv_vcpu->cpuid_cache.features_eax &
1297 			HV_MSR_APIC_ACCESS_AVAILABLE;
1298 	case HV_X64_MSR_TSC_FREQUENCY:
1299 	case HV_X64_MSR_APIC_FREQUENCY:
1300 		return hv_vcpu->cpuid_cache.features_eax &
1301 			HV_ACCESS_FREQUENCY_MSRS;
1302 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1303 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1304 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1305 		return hv_vcpu->cpuid_cache.features_eax &
1306 			HV_ACCESS_REENLIGHTENMENT;
1307 	case HV_X64_MSR_TSC_INVARIANT_CONTROL:
1308 		return hv_vcpu->cpuid_cache.features_eax &
1309 			HV_ACCESS_TSC_INVARIANT;
1310 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1311 	case HV_X64_MSR_CRASH_CTL:
1312 		return hv_vcpu->cpuid_cache.features_edx &
1313 			HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
1314 	case HV_X64_MSR_SYNDBG_OPTIONS:
1315 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1316 		return hv_vcpu->cpuid_cache.features_edx &
1317 			HV_FEATURE_DEBUG_MSRS_AVAILABLE;
1318 	default:
1319 		break;
1320 	}
1321 
1322 	return false;
1323 }
1324 
kvm_hv_set_msr_pw(struct kvm_vcpu * vcpu,u32 msr,u64 data,bool host)1325 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
1326 			     bool host)
1327 {
1328 	struct kvm *kvm = vcpu->kvm;
1329 	struct kvm_hv *hv = to_kvm_hv(kvm);
1330 
1331 	if (unlikely(!host && !hv_check_msr_access(to_hv_vcpu(vcpu), msr)))
1332 		return 1;
1333 
1334 	switch (msr) {
1335 	case HV_X64_MSR_GUEST_OS_ID:
1336 		hv->hv_guest_os_id = data;
1337 		/* setting guest os id to zero disables hypercall page */
1338 		if (!hv->hv_guest_os_id)
1339 			hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1340 		break;
1341 	case HV_X64_MSR_HYPERCALL: {
1342 		u8 instructions[9];
1343 		int i = 0;
1344 		u64 addr;
1345 
1346 		/* if guest os id is not set hypercall should remain disabled */
1347 		if (!hv->hv_guest_os_id)
1348 			break;
1349 		if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1350 			hv->hv_hypercall = data;
1351 			break;
1352 		}
1353 
1354 		/*
1355 		 * If Xen and Hyper-V hypercalls are both enabled, disambiguate
1356 		 * the same way Xen itself does, by setting the bit 31 of EAX
1357 		 * which is RsvdZ in the 32-bit Hyper-V hypercall ABI and just
1358 		 * going to be clobbered on 64-bit.
1359 		 */
1360 		if (kvm_xen_hypercall_enabled(kvm)) {
1361 			/* orl $0x80000000, %eax */
1362 			instructions[i++] = 0x0d;
1363 			instructions[i++] = 0x00;
1364 			instructions[i++] = 0x00;
1365 			instructions[i++] = 0x00;
1366 			instructions[i++] = 0x80;
1367 		}
1368 
1369 		/* vmcall/vmmcall */
1370 		static_call(kvm_x86_patch_hypercall)(vcpu, instructions + i);
1371 		i += 3;
1372 
1373 		/* ret */
1374 		((unsigned char *)instructions)[i++] = 0xc3;
1375 
1376 		addr = data & HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_MASK;
1377 		if (kvm_vcpu_write_guest(vcpu, addr, instructions, i))
1378 			return 1;
1379 		hv->hv_hypercall = data;
1380 		break;
1381 	}
1382 	case HV_X64_MSR_REFERENCE_TSC:
1383 		hv->hv_tsc_page = data;
1384 		if (hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE) {
1385 			if (!host)
1386 				hv->hv_tsc_page_status = HV_TSC_PAGE_GUEST_CHANGED;
1387 			else
1388 				hv->hv_tsc_page_status = HV_TSC_PAGE_HOST_CHANGED;
1389 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1390 		} else {
1391 			hv->hv_tsc_page_status = HV_TSC_PAGE_UNSET;
1392 		}
1393 		break;
1394 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1395 		return kvm_hv_msr_set_crash_data(kvm,
1396 						 msr - HV_X64_MSR_CRASH_P0,
1397 						 data);
1398 	case HV_X64_MSR_CRASH_CTL:
1399 		if (host)
1400 			return kvm_hv_msr_set_crash_ctl(kvm, data);
1401 
1402 		if (data & HV_CRASH_CTL_CRASH_NOTIFY) {
1403 			vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
1404 				   hv->hv_crash_param[0],
1405 				   hv->hv_crash_param[1],
1406 				   hv->hv_crash_param[2],
1407 				   hv->hv_crash_param[3],
1408 				   hv->hv_crash_param[4]);
1409 
1410 			/* Send notification about crash to user space */
1411 			kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
1412 		}
1413 		break;
1414 	case HV_X64_MSR_RESET:
1415 		if (data == 1) {
1416 			vcpu_debug(vcpu, "hyper-v reset requested\n");
1417 			kvm_make_request(KVM_REQ_HV_RESET, vcpu);
1418 		}
1419 		break;
1420 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1421 		hv->hv_reenlightenment_control = data;
1422 		break;
1423 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1424 		hv->hv_tsc_emulation_control = data;
1425 		break;
1426 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1427 		if (data && !host)
1428 			return 1;
1429 
1430 		hv->hv_tsc_emulation_status = data;
1431 		break;
1432 	case HV_X64_MSR_TIME_REF_COUNT:
1433 		/* read-only, but still ignore it if host-initiated */
1434 		if (!host)
1435 			return 1;
1436 		break;
1437 	case HV_X64_MSR_TSC_INVARIANT_CONTROL:
1438 		/* Only bit 0 is supported */
1439 		if (data & ~HV_EXPOSE_INVARIANT_TSC)
1440 			return 1;
1441 
1442 		/* The feature can't be disabled from the guest */
1443 		if (!host && hv->hv_invtsc_control && !data)
1444 			return 1;
1445 
1446 		hv->hv_invtsc_control = data;
1447 		break;
1448 	case HV_X64_MSR_SYNDBG_OPTIONS:
1449 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1450 		return syndbg_set_msr(vcpu, msr, data, host);
1451 	default:
1452 		kvm_pr_unimpl_wrmsr(vcpu, msr, data);
1453 		return 1;
1454 	}
1455 	return 0;
1456 }
1457 
1458 /* Calculate cpu time spent by current task in 100ns units */
current_task_runtime_100ns(void)1459 static u64 current_task_runtime_100ns(void)
1460 {
1461 	u64 utime, stime;
1462 
1463 	task_cputime_adjusted(current, &utime, &stime);
1464 
1465 	return div_u64(utime + stime, 100);
1466 }
1467 
kvm_hv_set_msr(struct kvm_vcpu * vcpu,u32 msr,u64 data,bool host)1468 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1469 {
1470 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
1471 
1472 	if (unlikely(!host && !hv_check_msr_access(hv_vcpu, msr)))
1473 		return 1;
1474 
1475 	switch (msr) {
1476 	case HV_X64_MSR_VP_INDEX: {
1477 		struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1478 		u32 new_vp_index = (u32)data;
1479 
1480 		if (!host || new_vp_index >= KVM_MAX_VCPUS)
1481 			return 1;
1482 
1483 		if (new_vp_index == hv_vcpu->vp_index)
1484 			return 0;
1485 
1486 		/*
1487 		 * The VP index is initialized to vcpu_index by
1488 		 * kvm_hv_vcpu_postcreate so they initially match.  Now the
1489 		 * VP index is changing, adjust num_mismatched_vp_indexes if
1490 		 * it now matches or no longer matches vcpu_idx.
1491 		 */
1492 		if (hv_vcpu->vp_index == vcpu->vcpu_idx)
1493 			atomic_inc(&hv->num_mismatched_vp_indexes);
1494 		else if (new_vp_index == vcpu->vcpu_idx)
1495 			atomic_dec(&hv->num_mismatched_vp_indexes);
1496 
1497 		hv_vcpu->vp_index = new_vp_index;
1498 		break;
1499 	}
1500 	case HV_X64_MSR_VP_ASSIST_PAGE: {
1501 		u64 gfn;
1502 		unsigned long addr;
1503 
1504 		if (!(data & HV_X64_MSR_VP_ASSIST_PAGE_ENABLE)) {
1505 			hv_vcpu->hv_vapic = data;
1506 			if (kvm_lapic_set_pv_eoi(vcpu, 0, 0))
1507 				return 1;
1508 			break;
1509 		}
1510 		gfn = data >> HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT;
1511 		addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1512 		if (kvm_is_error_hva(addr))
1513 			return 1;
1514 
1515 		/*
1516 		 * Clear apic_assist portion of struct hv_vp_assist_page
1517 		 * only, there can be valuable data in the rest which needs
1518 		 * to be preserved e.g. on migration.
1519 		 */
1520 		if (__put_user(0, (u32 __user *)addr))
1521 			return 1;
1522 		hv_vcpu->hv_vapic = data;
1523 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
1524 		if (kvm_lapic_set_pv_eoi(vcpu,
1525 					    gfn_to_gpa(gfn) | KVM_MSR_ENABLED,
1526 					    sizeof(struct hv_vp_assist_page)))
1527 			return 1;
1528 		break;
1529 	}
1530 	case HV_X64_MSR_EOI:
1531 		return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1532 	case HV_X64_MSR_ICR:
1533 		return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1534 	case HV_X64_MSR_TPR:
1535 		return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1536 	case HV_X64_MSR_VP_RUNTIME:
1537 		if (!host)
1538 			return 1;
1539 		hv_vcpu->runtime_offset = data - current_task_runtime_100ns();
1540 		break;
1541 	case HV_X64_MSR_SCONTROL:
1542 	case HV_X64_MSR_SVERSION:
1543 	case HV_X64_MSR_SIEFP:
1544 	case HV_X64_MSR_SIMP:
1545 	case HV_X64_MSR_EOM:
1546 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1547 		return synic_set_msr(to_hv_synic(vcpu), msr, data, host);
1548 	case HV_X64_MSR_STIMER0_CONFIG:
1549 	case HV_X64_MSR_STIMER1_CONFIG:
1550 	case HV_X64_MSR_STIMER2_CONFIG:
1551 	case HV_X64_MSR_STIMER3_CONFIG: {
1552 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1553 
1554 		return stimer_set_config(to_hv_stimer(vcpu, timer_index),
1555 					 data, host);
1556 	}
1557 	case HV_X64_MSR_STIMER0_COUNT:
1558 	case HV_X64_MSR_STIMER1_COUNT:
1559 	case HV_X64_MSR_STIMER2_COUNT:
1560 	case HV_X64_MSR_STIMER3_COUNT: {
1561 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1562 
1563 		return stimer_set_count(to_hv_stimer(vcpu, timer_index),
1564 					data, host);
1565 	}
1566 	case HV_X64_MSR_TSC_FREQUENCY:
1567 	case HV_X64_MSR_APIC_FREQUENCY:
1568 		/* read-only, but still ignore it if host-initiated */
1569 		if (!host)
1570 			return 1;
1571 		break;
1572 	default:
1573 		kvm_pr_unimpl_wrmsr(vcpu, msr, data);
1574 		return 1;
1575 	}
1576 
1577 	return 0;
1578 }
1579 
kvm_hv_get_msr_pw(struct kvm_vcpu * vcpu,u32 msr,u64 * pdata,bool host)1580 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1581 			     bool host)
1582 {
1583 	u64 data = 0;
1584 	struct kvm *kvm = vcpu->kvm;
1585 	struct kvm_hv *hv = to_kvm_hv(kvm);
1586 
1587 	if (unlikely(!host && !hv_check_msr_access(to_hv_vcpu(vcpu), msr)))
1588 		return 1;
1589 
1590 	switch (msr) {
1591 	case HV_X64_MSR_GUEST_OS_ID:
1592 		data = hv->hv_guest_os_id;
1593 		break;
1594 	case HV_X64_MSR_HYPERCALL:
1595 		data = hv->hv_hypercall;
1596 		break;
1597 	case HV_X64_MSR_TIME_REF_COUNT:
1598 		data = get_time_ref_counter(kvm);
1599 		break;
1600 	case HV_X64_MSR_REFERENCE_TSC:
1601 		data = hv->hv_tsc_page;
1602 		break;
1603 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
1604 		return kvm_hv_msr_get_crash_data(kvm,
1605 						 msr - HV_X64_MSR_CRASH_P0,
1606 						 pdata);
1607 	case HV_X64_MSR_CRASH_CTL:
1608 		return kvm_hv_msr_get_crash_ctl(kvm, pdata);
1609 	case HV_X64_MSR_RESET:
1610 		data = 0;
1611 		break;
1612 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
1613 		data = hv->hv_reenlightenment_control;
1614 		break;
1615 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
1616 		data = hv->hv_tsc_emulation_control;
1617 		break;
1618 	case HV_X64_MSR_TSC_EMULATION_STATUS:
1619 		data = hv->hv_tsc_emulation_status;
1620 		break;
1621 	case HV_X64_MSR_TSC_INVARIANT_CONTROL:
1622 		data = hv->hv_invtsc_control;
1623 		break;
1624 	case HV_X64_MSR_SYNDBG_OPTIONS:
1625 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
1626 		return syndbg_get_msr(vcpu, msr, pdata, host);
1627 	default:
1628 		kvm_pr_unimpl_rdmsr(vcpu, msr);
1629 		return 1;
1630 	}
1631 
1632 	*pdata = data;
1633 	return 0;
1634 }
1635 
kvm_hv_get_msr(struct kvm_vcpu * vcpu,u32 msr,u64 * pdata,bool host)1636 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata,
1637 			  bool host)
1638 {
1639 	u64 data = 0;
1640 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
1641 
1642 	if (unlikely(!host && !hv_check_msr_access(hv_vcpu, msr)))
1643 		return 1;
1644 
1645 	switch (msr) {
1646 	case HV_X64_MSR_VP_INDEX:
1647 		data = hv_vcpu->vp_index;
1648 		break;
1649 	case HV_X64_MSR_EOI:
1650 		return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1651 	case HV_X64_MSR_ICR:
1652 		return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1653 	case HV_X64_MSR_TPR:
1654 		return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1655 	case HV_X64_MSR_VP_ASSIST_PAGE:
1656 		data = hv_vcpu->hv_vapic;
1657 		break;
1658 	case HV_X64_MSR_VP_RUNTIME:
1659 		data = current_task_runtime_100ns() + hv_vcpu->runtime_offset;
1660 		break;
1661 	case HV_X64_MSR_SCONTROL:
1662 	case HV_X64_MSR_SVERSION:
1663 	case HV_X64_MSR_SIEFP:
1664 	case HV_X64_MSR_SIMP:
1665 	case HV_X64_MSR_EOM:
1666 	case HV_X64_MSR_SINT0 ... HV_X64_MSR_SINT15:
1667 		return synic_get_msr(to_hv_synic(vcpu), msr, pdata, host);
1668 	case HV_X64_MSR_STIMER0_CONFIG:
1669 	case HV_X64_MSR_STIMER1_CONFIG:
1670 	case HV_X64_MSR_STIMER2_CONFIG:
1671 	case HV_X64_MSR_STIMER3_CONFIG: {
1672 		int timer_index = (msr - HV_X64_MSR_STIMER0_CONFIG)/2;
1673 
1674 		return stimer_get_config(to_hv_stimer(vcpu, timer_index),
1675 					 pdata);
1676 	}
1677 	case HV_X64_MSR_STIMER0_COUNT:
1678 	case HV_X64_MSR_STIMER1_COUNT:
1679 	case HV_X64_MSR_STIMER2_COUNT:
1680 	case HV_X64_MSR_STIMER3_COUNT: {
1681 		int timer_index = (msr - HV_X64_MSR_STIMER0_COUNT)/2;
1682 
1683 		return stimer_get_count(to_hv_stimer(vcpu, timer_index),
1684 					pdata);
1685 	}
1686 	case HV_X64_MSR_TSC_FREQUENCY:
1687 		data = (u64)vcpu->arch.virtual_tsc_khz * 1000;
1688 		break;
1689 	case HV_X64_MSR_APIC_FREQUENCY:
1690 		data = APIC_BUS_FREQUENCY;
1691 		break;
1692 	default:
1693 		kvm_pr_unimpl_rdmsr(vcpu, msr);
1694 		return 1;
1695 	}
1696 	*pdata = data;
1697 	return 0;
1698 }
1699 
kvm_hv_set_msr_common(struct kvm_vcpu * vcpu,u32 msr,u64 data,bool host)1700 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
1701 {
1702 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1703 
1704 	if (!host && !vcpu->arch.hyperv_enabled)
1705 		return 1;
1706 
1707 	if (kvm_hv_vcpu_init(vcpu))
1708 		return 1;
1709 
1710 	if (kvm_hv_msr_partition_wide(msr)) {
1711 		int r;
1712 
1713 		mutex_lock(&hv->hv_lock);
1714 		r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
1715 		mutex_unlock(&hv->hv_lock);
1716 		return r;
1717 	} else
1718 		return kvm_hv_set_msr(vcpu, msr, data, host);
1719 }
1720 
kvm_hv_get_msr_common(struct kvm_vcpu * vcpu,u32 msr,u64 * pdata,bool host)1721 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
1722 {
1723 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
1724 
1725 	if (!host && !vcpu->arch.hyperv_enabled)
1726 		return 1;
1727 
1728 	if (kvm_hv_vcpu_init(vcpu))
1729 		return 1;
1730 
1731 	if (kvm_hv_msr_partition_wide(msr)) {
1732 		int r;
1733 
1734 		mutex_lock(&hv->hv_lock);
1735 		r = kvm_hv_get_msr_pw(vcpu, msr, pdata, host);
1736 		mutex_unlock(&hv->hv_lock);
1737 		return r;
1738 	} else
1739 		return kvm_hv_get_msr(vcpu, msr, pdata, host);
1740 }
1741 
sparse_set_to_vcpu_mask(struct kvm * kvm,u64 * sparse_banks,u64 valid_bank_mask,unsigned long * vcpu_mask)1742 static void sparse_set_to_vcpu_mask(struct kvm *kvm, u64 *sparse_banks,
1743 				    u64 valid_bank_mask, unsigned long *vcpu_mask)
1744 {
1745 	struct kvm_hv *hv = to_kvm_hv(kvm);
1746 	bool has_mismatch = atomic_read(&hv->num_mismatched_vp_indexes);
1747 	u64 vp_bitmap[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
1748 	struct kvm_vcpu *vcpu;
1749 	int bank, sbank = 0;
1750 	unsigned long i;
1751 	u64 *bitmap;
1752 
1753 	BUILD_BUG_ON(sizeof(vp_bitmap) >
1754 		     sizeof(*vcpu_mask) * BITS_TO_LONGS(KVM_MAX_VCPUS));
1755 
1756 	/*
1757 	 * If vp_index == vcpu_idx for all vCPUs, fill vcpu_mask directly, else
1758 	 * fill a temporary buffer and manually test each vCPU's VP index.
1759 	 */
1760 	if (likely(!has_mismatch))
1761 		bitmap = (u64 *)vcpu_mask;
1762 	else
1763 		bitmap = vp_bitmap;
1764 
1765 	/*
1766 	 * Each set of 64 VPs is packed into sparse_banks, with valid_bank_mask
1767 	 * having a '1' for each bank that exists in sparse_banks.  Sets must
1768 	 * be in ascending order, i.e. bank0..bankN.
1769 	 */
1770 	memset(bitmap, 0, sizeof(vp_bitmap));
1771 	for_each_set_bit(bank, (unsigned long *)&valid_bank_mask,
1772 			 KVM_HV_MAX_SPARSE_VCPU_SET_BITS)
1773 		bitmap[bank] = sparse_banks[sbank++];
1774 
1775 	if (likely(!has_mismatch))
1776 		return;
1777 
1778 	bitmap_zero(vcpu_mask, KVM_MAX_VCPUS);
1779 	kvm_for_each_vcpu(i, vcpu, kvm) {
1780 		if (test_bit(kvm_hv_get_vpindex(vcpu), (unsigned long *)vp_bitmap))
1781 			__set_bit(i, vcpu_mask);
1782 	}
1783 }
1784 
hv_is_vp_in_sparse_set(u32 vp_id,u64 valid_bank_mask,u64 sparse_banks[])1785 static bool hv_is_vp_in_sparse_set(u32 vp_id, u64 valid_bank_mask, u64 sparse_banks[])
1786 {
1787 	int valid_bit_nr = vp_id / HV_VCPUS_PER_SPARSE_BANK;
1788 	unsigned long sbank;
1789 
1790 	if (!test_bit(valid_bit_nr, (unsigned long *)&valid_bank_mask))
1791 		return false;
1792 
1793 	/*
1794 	 * The index into the sparse bank is the number of preceding bits in
1795 	 * the valid mask.  Optimize for VMs with <64 vCPUs by skipping the
1796 	 * fancy math if there can't possibly be preceding bits.
1797 	 */
1798 	if (valid_bit_nr)
1799 		sbank = hweight64(valid_bank_mask & GENMASK_ULL(valid_bit_nr - 1, 0));
1800 	else
1801 		sbank = 0;
1802 
1803 	return test_bit(vp_id % HV_VCPUS_PER_SPARSE_BANK,
1804 			(unsigned long *)&sparse_banks[sbank]);
1805 }
1806 
1807 struct kvm_hv_hcall {
1808 	/* Hypercall input data */
1809 	u64 param;
1810 	u64 ingpa;
1811 	u64 outgpa;
1812 	u16 code;
1813 	u16 var_cnt;
1814 	u16 rep_cnt;
1815 	u16 rep_idx;
1816 	bool fast;
1817 	bool rep;
1818 	sse128_t xmm[HV_HYPERCALL_MAX_XMM_REGISTERS];
1819 
1820 	/*
1821 	 * Current read offset when KVM reads hypercall input data gradually,
1822 	 * either offset in bytes from 'ingpa' for regular hypercalls or the
1823 	 * number of already consumed 'XMM halves' for 'fast' hypercalls.
1824 	 */
1825 	union {
1826 		gpa_t data_offset;
1827 		int consumed_xmm_halves;
1828 	};
1829 };
1830 
1831 
kvm_hv_get_hc_data(struct kvm * kvm,struct kvm_hv_hcall * hc,u16 orig_cnt,u16 cnt_cap,u64 * data)1832 static int kvm_hv_get_hc_data(struct kvm *kvm, struct kvm_hv_hcall *hc,
1833 			      u16 orig_cnt, u16 cnt_cap, u64 *data)
1834 {
1835 	/*
1836 	 * Preserve the original count when ignoring entries via a "cap", KVM
1837 	 * still needs to validate the guest input (though the non-XMM path
1838 	 * punts on the checks).
1839 	 */
1840 	u16 cnt = min(orig_cnt, cnt_cap);
1841 	int i, j;
1842 
1843 	if (hc->fast) {
1844 		/*
1845 		 * Each XMM holds two sparse banks, but do not count halves that
1846 		 * have already been consumed for hypercall parameters.
1847 		 */
1848 		if (orig_cnt > 2 * HV_HYPERCALL_MAX_XMM_REGISTERS - hc->consumed_xmm_halves)
1849 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1850 
1851 		for (i = 0; i < cnt; i++) {
1852 			j = i + hc->consumed_xmm_halves;
1853 			if (j % 2)
1854 				data[i] = sse128_hi(hc->xmm[j / 2]);
1855 			else
1856 				data[i] = sse128_lo(hc->xmm[j / 2]);
1857 		}
1858 		return 0;
1859 	}
1860 
1861 	return kvm_read_guest(kvm, hc->ingpa + hc->data_offset, data,
1862 			      cnt * sizeof(*data));
1863 }
1864 
kvm_get_sparse_vp_set(struct kvm * kvm,struct kvm_hv_hcall * hc,u64 * sparse_banks)1865 static u64 kvm_get_sparse_vp_set(struct kvm *kvm, struct kvm_hv_hcall *hc,
1866 				 u64 *sparse_banks)
1867 {
1868 	if (hc->var_cnt > HV_MAX_SPARSE_VCPU_BANKS)
1869 		return -EINVAL;
1870 
1871 	/* Cap var_cnt to ignore banks that cannot contain a legal VP index. */
1872 	return kvm_hv_get_hc_data(kvm, hc, hc->var_cnt, KVM_HV_MAX_SPARSE_VCPU_SET_BITS,
1873 				  sparse_banks);
1874 }
1875 
kvm_hv_get_tlb_flush_entries(struct kvm * kvm,struct kvm_hv_hcall * hc,u64 entries[])1876 static int kvm_hv_get_tlb_flush_entries(struct kvm *kvm, struct kvm_hv_hcall *hc, u64 entries[])
1877 {
1878 	return kvm_hv_get_hc_data(kvm, hc, hc->rep_cnt, hc->rep_cnt, entries);
1879 }
1880 
hv_tlb_flush_enqueue(struct kvm_vcpu * vcpu,struct kvm_vcpu_hv_tlb_flush_fifo * tlb_flush_fifo,u64 * entries,int count)1881 static void hv_tlb_flush_enqueue(struct kvm_vcpu *vcpu,
1882 				 struct kvm_vcpu_hv_tlb_flush_fifo *tlb_flush_fifo,
1883 				 u64 *entries, int count)
1884 {
1885 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
1886 	u64 flush_all_entry = KVM_HV_TLB_FLUSHALL_ENTRY;
1887 
1888 	if (!hv_vcpu)
1889 		return;
1890 
1891 	spin_lock(&tlb_flush_fifo->write_lock);
1892 
1893 	/*
1894 	 * All entries should fit on the fifo leaving one free for 'flush all'
1895 	 * entry in case another request comes in. In case there's not enough
1896 	 * space, just put 'flush all' entry there.
1897 	 */
1898 	if (count && entries && count < kfifo_avail(&tlb_flush_fifo->entries)) {
1899 		WARN_ON(kfifo_in(&tlb_flush_fifo->entries, entries, count) != count);
1900 		goto out_unlock;
1901 	}
1902 
1903 	/*
1904 	 * Note: full fifo always contains 'flush all' entry, no need to check the
1905 	 * return value.
1906 	 */
1907 	kfifo_in(&tlb_flush_fifo->entries, &flush_all_entry, 1);
1908 
1909 out_unlock:
1910 	spin_unlock(&tlb_flush_fifo->write_lock);
1911 }
1912 
kvm_hv_vcpu_flush_tlb(struct kvm_vcpu * vcpu)1913 int kvm_hv_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
1914 {
1915 	struct kvm_vcpu_hv_tlb_flush_fifo *tlb_flush_fifo;
1916 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
1917 	u64 entries[KVM_HV_TLB_FLUSH_FIFO_SIZE];
1918 	int i, j, count;
1919 	gva_t gva;
1920 
1921 	if (!tdp_enabled || !hv_vcpu)
1922 		return -EINVAL;
1923 
1924 	tlb_flush_fifo = kvm_hv_get_tlb_flush_fifo(vcpu, is_guest_mode(vcpu));
1925 
1926 	count = kfifo_out(&tlb_flush_fifo->entries, entries, KVM_HV_TLB_FLUSH_FIFO_SIZE);
1927 
1928 	for (i = 0; i < count; i++) {
1929 		if (entries[i] == KVM_HV_TLB_FLUSHALL_ENTRY)
1930 			goto out_flush_all;
1931 
1932 		/*
1933 		 * Lower 12 bits of 'address' encode the number of additional
1934 		 * pages to flush.
1935 		 */
1936 		gva = entries[i] & PAGE_MASK;
1937 		for (j = 0; j < (entries[i] & ~PAGE_MASK) + 1; j++)
1938 			static_call(kvm_x86_flush_tlb_gva)(vcpu, gva + j * PAGE_SIZE);
1939 
1940 		++vcpu->stat.tlb_flush;
1941 	}
1942 	return 0;
1943 
1944 out_flush_all:
1945 	kfifo_reset_out(&tlb_flush_fifo->entries);
1946 
1947 	/* Fall back to full flush. */
1948 	return -ENOSPC;
1949 }
1950 
kvm_hv_flush_tlb(struct kvm_vcpu * vcpu,struct kvm_hv_hcall * hc)1951 static u64 kvm_hv_flush_tlb(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc)
1952 {
1953 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
1954 	u64 *sparse_banks = hv_vcpu->sparse_banks;
1955 	struct kvm *kvm = vcpu->kvm;
1956 	struct hv_tlb_flush_ex flush_ex;
1957 	struct hv_tlb_flush flush;
1958 	DECLARE_BITMAP(vcpu_mask, KVM_MAX_VCPUS);
1959 	struct kvm_vcpu_hv_tlb_flush_fifo *tlb_flush_fifo;
1960 	/*
1961 	 * Normally, there can be no more than 'KVM_HV_TLB_FLUSH_FIFO_SIZE'
1962 	 * entries on the TLB flush fifo. The last entry, however, needs to be
1963 	 * always left free for 'flush all' entry which gets placed when
1964 	 * there is not enough space to put all the requested entries.
1965 	 */
1966 	u64 __tlb_flush_entries[KVM_HV_TLB_FLUSH_FIFO_SIZE - 1];
1967 	u64 *tlb_flush_entries;
1968 	u64 valid_bank_mask;
1969 	struct kvm_vcpu *v;
1970 	unsigned long i;
1971 	bool all_cpus;
1972 
1973 	/*
1974 	 * The Hyper-V TLFS doesn't allow more than HV_MAX_SPARSE_VCPU_BANKS
1975 	 * sparse banks. Fail the build if KVM's max allowed number of
1976 	 * vCPUs (>4096) exceeds this limit.
1977 	 */
1978 	BUILD_BUG_ON(KVM_HV_MAX_SPARSE_VCPU_SET_BITS > HV_MAX_SPARSE_VCPU_BANKS);
1979 
1980 	/*
1981 	 * 'Slow' hypercall's first parameter is the address in guest's memory
1982 	 * where hypercall parameters are placed. This is either a GPA or a
1983 	 * nested GPA when KVM is handling the call from L2 ('direct' TLB
1984 	 * flush).  Translate the address here so the memory can be uniformly
1985 	 * read with kvm_read_guest().
1986 	 */
1987 	if (!hc->fast && is_guest_mode(vcpu)) {
1988 		hc->ingpa = translate_nested_gpa(vcpu, hc->ingpa, 0, NULL);
1989 		if (unlikely(hc->ingpa == INVALID_GPA))
1990 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
1991 	}
1992 
1993 	if (hc->code == HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST ||
1994 	    hc->code == HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE) {
1995 		if (hc->fast) {
1996 			flush.address_space = hc->ingpa;
1997 			flush.flags = hc->outgpa;
1998 			flush.processor_mask = sse128_lo(hc->xmm[0]);
1999 			hc->consumed_xmm_halves = 1;
2000 		} else {
2001 			if (unlikely(kvm_read_guest(kvm, hc->ingpa,
2002 						    &flush, sizeof(flush))))
2003 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
2004 			hc->data_offset = sizeof(flush);
2005 		}
2006 
2007 		trace_kvm_hv_flush_tlb(flush.processor_mask,
2008 				       flush.address_space, flush.flags,
2009 				       is_guest_mode(vcpu));
2010 
2011 		valid_bank_mask = BIT_ULL(0);
2012 		sparse_banks[0] = flush.processor_mask;
2013 
2014 		/*
2015 		 * Work around possible WS2012 bug: it sends hypercalls
2016 		 * with processor_mask = 0x0 and HV_FLUSH_ALL_PROCESSORS clear,
2017 		 * while also expecting us to flush something and crashing if
2018 		 * we don't. Let's treat processor_mask == 0 same as
2019 		 * HV_FLUSH_ALL_PROCESSORS.
2020 		 */
2021 		all_cpus = (flush.flags & HV_FLUSH_ALL_PROCESSORS) ||
2022 			flush.processor_mask == 0;
2023 	} else {
2024 		if (hc->fast) {
2025 			flush_ex.address_space = hc->ingpa;
2026 			flush_ex.flags = hc->outgpa;
2027 			memcpy(&flush_ex.hv_vp_set,
2028 			       &hc->xmm[0], sizeof(hc->xmm[0]));
2029 			hc->consumed_xmm_halves = 2;
2030 		} else {
2031 			if (unlikely(kvm_read_guest(kvm, hc->ingpa, &flush_ex,
2032 						    sizeof(flush_ex))))
2033 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
2034 			hc->data_offset = sizeof(flush_ex);
2035 		}
2036 
2037 		trace_kvm_hv_flush_tlb_ex(flush_ex.hv_vp_set.valid_bank_mask,
2038 					  flush_ex.hv_vp_set.format,
2039 					  flush_ex.address_space,
2040 					  flush_ex.flags, is_guest_mode(vcpu));
2041 
2042 		valid_bank_mask = flush_ex.hv_vp_set.valid_bank_mask;
2043 		all_cpus = flush_ex.hv_vp_set.format !=
2044 			HV_GENERIC_SET_SPARSE_4K;
2045 
2046 		if (hc->var_cnt != hweight64(valid_bank_mask))
2047 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
2048 
2049 		if (!all_cpus) {
2050 			if (!hc->var_cnt)
2051 				goto ret_success;
2052 
2053 			if (kvm_get_sparse_vp_set(kvm, hc, sparse_banks))
2054 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
2055 		}
2056 
2057 		/*
2058 		 * Hyper-V TLFS doesn't explicitly forbid non-empty sparse vCPU
2059 		 * banks (and, thus, non-zero 'var_cnt') for the 'all vCPUs'
2060 		 * case (HV_GENERIC_SET_ALL).  Always adjust data_offset and
2061 		 * consumed_xmm_halves to make sure TLB flush entries are read
2062 		 * from the correct offset.
2063 		 */
2064 		if (hc->fast)
2065 			hc->consumed_xmm_halves += hc->var_cnt;
2066 		else
2067 			hc->data_offset += hc->var_cnt * sizeof(sparse_banks[0]);
2068 	}
2069 
2070 	if (hc->code == HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE ||
2071 	    hc->code == HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX ||
2072 	    hc->rep_cnt > ARRAY_SIZE(__tlb_flush_entries)) {
2073 		tlb_flush_entries = NULL;
2074 	} else {
2075 		if (kvm_hv_get_tlb_flush_entries(kvm, hc, __tlb_flush_entries))
2076 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
2077 		tlb_flush_entries = __tlb_flush_entries;
2078 	}
2079 
2080 	/*
2081 	 * vcpu->arch.cr3 may not be up-to-date for running vCPUs so we can't
2082 	 * analyze it here, flush TLB regardless of the specified address space.
2083 	 */
2084 	if (all_cpus && !is_guest_mode(vcpu)) {
2085 		kvm_for_each_vcpu(i, v, kvm) {
2086 			tlb_flush_fifo = kvm_hv_get_tlb_flush_fifo(v, false);
2087 			hv_tlb_flush_enqueue(v, tlb_flush_fifo,
2088 					     tlb_flush_entries, hc->rep_cnt);
2089 		}
2090 
2091 		kvm_make_all_cpus_request(kvm, KVM_REQ_HV_TLB_FLUSH);
2092 	} else if (!is_guest_mode(vcpu)) {
2093 		sparse_set_to_vcpu_mask(kvm, sparse_banks, valid_bank_mask, vcpu_mask);
2094 
2095 		for_each_set_bit(i, vcpu_mask, KVM_MAX_VCPUS) {
2096 			v = kvm_get_vcpu(kvm, i);
2097 			if (!v)
2098 				continue;
2099 			tlb_flush_fifo = kvm_hv_get_tlb_flush_fifo(v, false);
2100 			hv_tlb_flush_enqueue(v, tlb_flush_fifo,
2101 					     tlb_flush_entries, hc->rep_cnt);
2102 		}
2103 
2104 		kvm_make_vcpus_request_mask(kvm, KVM_REQ_HV_TLB_FLUSH, vcpu_mask);
2105 	} else {
2106 		struct kvm_vcpu_hv *hv_v;
2107 
2108 		bitmap_zero(vcpu_mask, KVM_MAX_VCPUS);
2109 
2110 		kvm_for_each_vcpu(i, v, kvm) {
2111 			hv_v = to_hv_vcpu(v);
2112 
2113 			/*
2114 			 * The following check races with nested vCPUs entering/exiting
2115 			 * and/or migrating between L1's vCPUs, however the only case when
2116 			 * KVM *must* flush the TLB is when the target L2 vCPU keeps
2117 			 * running on the same L1 vCPU from the moment of the request until
2118 			 * kvm_hv_flush_tlb() returns. TLB is fully flushed in all other
2119 			 * cases, e.g. when the target L2 vCPU migrates to a different L1
2120 			 * vCPU or when the corresponding L1 vCPU temporary switches to a
2121 			 * different L2 vCPU while the request is being processed.
2122 			 */
2123 			if (!hv_v || hv_v->nested.vm_id != hv_vcpu->nested.vm_id)
2124 				continue;
2125 
2126 			if (!all_cpus &&
2127 			    !hv_is_vp_in_sparse_set(hv_v->nested.vp_id, valid_bank_mask,
2128 						    sparse_banks))
2129 				continue;
2130 
2131 			__set_bit(i, vcpu_mask);
2132 			tlb_flush_fifo = kvm_hv_get_tlb_flush_fifo(v, true);
2133 			hv_tlb_flush_enqueue(v, tlb_flush_fifo,
2134 					     tlb_flush_entries, hc->rep_cnt);
2135 		}
2136 
2137 		kvm_make_vcpus_request_mask(kvm, KVM_REQ_HV_TLB_FLUSH, vcpu_mask);
2138 	}
2139 
2140 ret_success:
2141 	/* We always do full TLB flush, set 'Reps completed' = 'Rep Count' */
2142 	return (u64)HV_STATUS_SUCCESS |
2143 		((u64)hc->rep_cnt << HV_HYPERCALL_REP_COMP_OFFSET);
2144 }
2145 
kvm_hv_send_ipi_to_many(struct kvm * kvm,u32 vector,u64 * sparse_banks,u64 valid_bank_mask)2146 static void kvm_hv_send_ipi_to_many(struct kvm *kvm, u32 vector,
2147 				    u64 *sparse_banks, u64 valid_bank_mask)
2148 {
2149 	struct kvm_lapic_irq irq = {
2150 		.delivery_mode = APIC_DM_FIXED,
2151 		.vector = vector
2152 	};
2153 	struct kvm_vcpu *vcpu;
2154 	unsigned long i;
2155 
2156 	kvm_for_each_vcpu(i, vcpu, kvm) {
2157 		if (sparse_banks &&
2158 		    !hv_is_vp_in_sparse_set(kvm_hv_get_vpindex(vcpu),
2159 					    valid_bank_mask, sparse_banks))
2160 			continue;
2161 
2162 		/* We fail only when APIC is disabled */
2163 		kvm_apic_set_irq(vcpu, &irq, NULL);
2164 	}
2165 }
2166 
kvm_hv_send_ipi(struct kvm_vcpu * vcpu,struct kvm_hv_hcall * hc)2167 static u64 kvm_hv_send_ipi(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc)
2168 {
2169 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
2170 	u64 *sparse_banks = hv_vcpu->sparse_banks;
2171 	struct kvm *kvm = vcpu->kvm;
2172 	struct hv_send_ipi_ex send_ipi_ex;
2173 	struct hv_send_ipi send_ipi;
2174 	u64 valid_bank_mask;
2175 	u32 vector;
2176 	bool all_cpus;
2177 
2178 	if (hc->code == HVCALL_SEND_IPI) {
2179 		if (!hc->fast) {
2180 			if (unlikely(kvm_read_guest(kvm, hc->ingpa, &send_ipi,
2181 						    sizeof(send_ipi))))
2182 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
2183 			sparse_banks[0] = send_ipi.cpu_mask;
2184 			vector = send_ipi.vector;
2185 		} else {
2186 			/* 'reserved' part of hv_send_ipi should be 0 */
2187 			if (unlikely(hc->ingpa >> 32 != 0))
2188 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
2189 			sparse_banks[0] = hc->outgpa;
2190 			vector = (u32)hc->ingpa;
2191 		}
2192 		all_cpus = false;
2193 		valid_bank_mask = BIT_ULL(0);
2194 
2195 		trace_kvm_hv_send_ipi(vector, sparse_banks[0]);
2196 	} else {
2197 		if (!hc->fast) {
2198 			if (unlikely(kvm_read_guest(kvm, hc->ingpa, &send_ipi_ex,
2199 						    sizeof(send_ipi_ex))))
2200 				return HV_STATUS_INVALID_HYPERCALL_INPUT;
2201 		} else {
2202 			send_ipi_ex.vector = (u32)hc->ingpa;
2203 			send_ipi_ex.vp_set.format = hc->outgpa;
2204 			send_ipi_ex.vp_set.valid_bank_mask = sse128_lo(hc->xmm[0]);
2205 		}
2206 
2207 		trace_kvm_hv_send_ipi_ex(send_ipi_ex.vector,
2208 					 send_ipi_ex.vp_set.format,
2209 					 send_ipi_ex.vp_set.valid_bank_mask);
2210 
2211 		vector = send_ipi_ex.vector;
2212 		valid_bank_mask = send_ipi_ex.vp_set.valid_bank_mask;
2213 		all_cpus = send_ipi_ex.vp_set.format == HV_GENERIC_SET_ALL;
2214 
2215 		if (hc->var_cnt != hweight64(valid_bank_mask))
2216 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
2217 
2218 		if (all_cpus)
2219 			goto check_and_send_ipi;
2220 
2221 		if (!hc->var_cnt)
2222 			goto ret_success;
2223 
2224 		if (!hc->fast)
2225 			hc->data_offset = offsetof(struct hv_send_ipi_ex,
2226 						   vp_set.bank_contents);
2227 		else
2228 			hc->consumed_xmm_halves = 1;
2229 
2230 		if (kvm_get_sparse_vp_set(kvm, hc, sparse_banks))
2231 			return HV_STATUS_INVALID_HYPERCALL_INPUT;
2232 	}
2233 
2234 check_and_send_ipi:
2235 	if ((vector < HV_IPI_LOW_VECTOR) || (vector > HV_IPI_HIGH_VECTOR))
2236 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
2237 
2238 	if (all_cpus)
2239 		kvm_hv_send_ipi_to_many(kvm, vector, NULL, 0);
2240 	else
2241 		kvm_hv_send_ipi_to_many(kvm, vector, sparse_banks, valid_bank_mask);
2242 
2243 ret_success:
2244 	return HV_STATUS_SUCCESS;
2245 }
2246 
kvm_hv_set_cpuid(struct kvm_vcpu * vcpu,bool hyperv_enabled)2247 void kvm_hv_set_cpuid(struct kvm_vcpu *vcpu, bool hyperv_enabled)
2248 {
2249 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
2250 	struct kvm_cpuid_entry2 *entry;
2251 
2252 	vcpu->arch.hyperv_enabled = hyperv_enabled;
2253 
2254 	if (!hv_vcpu) {
2255 		/*
2256 		 * KVM should have already allocated kvm_vcpu_hv if Hyper-V is
2257 		 * enabled in CPUID.
2258 		 */
2259 		WARN_ON_ONCE(vcpu->arch.hyperv_enabled);
2260 		return;
2261 	}
2262 
2263 	memset(&hv_vcpu->cpuid_cache, 0, sizeof(hv_vcpu->cpuid_cache));
2264 
2265 	if (!vcpu->arch.hyperv_enabled)
2266 		return;
2267 
2268 	entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_FEATURES);
2269 	if (entry) {
2270 		hv_vcpu->cpuid_cache.features_eax = entry->eax;
2271 		hv_vcpu->cpuid_cache.features_ebx = entry->ebx;
2272 		hv_vcpu->cpuid_cache.features_edx = entry->edx;
2273 	}
2274 
2275 	entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_ENLIGHTMENT_INFO);
2276 	if (entry) {
2277 		hv_vcpu->cpuid_cache.enlightenments_eax = entry->eax;
2278 		hv_vcpu->cpuid_cache.enlightenments_ebx = entry->ebx;
2279 	}
2280 
2281 	entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES);
2282 	if (entry)
2283 		hv_vcpu->cpuid_cache.syndbg_cap_eax = entry->eax;
2284 
2285 	entry = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_NESTED_FEATURES);
2286 	if (entry) {
2287 		hv_vcpu->cpuid_cache.nested_eax = entry->eax;
2288 		hv_vcpu->cpuid_cache.nested_ebx = entry->ebx;
2289 	}
2290 }
2291 
kvm_hv_set_enforce_cpuid(struct kvm_vcpu * vcpu,bool enforce)2292 int kvm_hv_set_enforce_cpuid(struct kvm_vcpu *vcpu, bool enforce)
2293 {
2294 	struct kvm_vcpu_hv *hv_vcpu;
2295 	int ret = 0;
2296 
2297 	if (!to_hv_vcpu(vcpu)) {
2298 		if (enforce) {
2299 			ret = kvm_hv_vcpu_init(vcpu);
2300 			if (ret)
2301 				return ret;
2302 		} else {
2303 			return 0;
2304 		}
2305 	}
2306 
2307 	hv_vcpu = to_hv_vcpu(vcpu);
2308 	hv_vcpu->enforce_cpuid = enforce;
2309 
2310 	return ret;
2311 }
2312 
kvm_hv_hypercall_set_result(struct kvm_vcpu * vcpu,u64 result)2313 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
2314 {
2315 	bool longmode;
2316 
2317 	longmode = is_64_bit_hypercall(vcpu);
2318 	if (longmode)
2319 		kvm_rax_write(vcpu, result);
2320 	else {
2321 		kvm_rdx_write(vcpu, result >> 32);
2322 		kvm_rax_write(vcpu, result & 0xffffffff);
2323 	}
2324 }
2325 
kvm_hv_hypercall_complete(struct kvm_vcpu * vcpu,u64 result)2326 static int kvm_hv_hypercall_complete(struct kvm_vcpu *vcpu, u64 result)
2327 {
2328 	u32 tlb_lock_count = 0;
2329 	int ret;
2330 
2331 	if (hv_result_success(result) && is_guest_mode(vcpu) &&
2332 	    kvm_hv_is_tlb_flush_hcall(vcpu) &&
2333 	    kvm_read_guest(vcpu->kvm, to_hv_vcpu(vcpu)->nested.pa_page_gpa,
2334 			   &tlb_lock_count, sizeof(tlb_lock_count)))
2335 		result = HV_STATUS_INVALID_HYPERCALL_INPUT;
2336 
2337 	trace_kvm_hv_hypercall_done(result);
2338 	kvm_hv_hypercall_set_result(vcpu, result);
2339 	++vcpu->stat.hypercalls;
2340 
2341 	ret = kvm_skip_emulated_instruction(vcpu);
2342 
2343 	if (tlb_lock_count)
2344 		kvm_x86_ops.nested_ops->hv_inject_synthetic_vmexit_post_tlb_flush(vcpu);
2345 
2346 	return ret;
2347 }
2348 
kvm_hv_hypercall_complete_userspace(struct kvm_vcpu * vcpu)2349 static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
2350 {
2351 	return kvm_hv_hypercall_complete(vcpu, vcpu->run->hyperv.u.hcall.result);
2352 }
2353 
kvm_hvcall_signal_event(struct kvm_vcpu * vcpu,struct kvm_hv_hcall * hc)2354 static u16 kvm_hvcall_signal_event(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc)
2355 {
2356 	struct kvm_hv *hv = to_kvm_hv(vcpu->kvm);
2357 	struct eventfd_ctx *eventfd;
2358 
2359 	if (unlikely(!hc->fast)) {
2360 		int ret;
2361 		gpa_t gpa = hc->ingpa;
2362 
2363 		if ((gpa & (__alignof__(hc->ingpa) - 1)) ||
2364 		    offset_in_page(gpa) + sizeof(hc->ingpa) > PAGE_SIZE)
2365 			return HV_STATUS_INVALID_ALIGNMENT;
2366 
2367 		ret = kvm_vcpu_read_guest(vcpu, gpa,
2368 					  &hc->ingpa, sizeof(hc->ingpa));
2369 		if (ret < 0)
2370 			return HV_STATUS_INVALID_ALIGNMENT;
2371 	}
2372 
2373 	/*
2374 	 * Per spec, bits 32-47 contain the extra "flag number".  However, we
2375 	 * have no use for it, and in all known usecases it is zero, so just
2376 	 * report lookup failure if it isn't.
2377 	 */
2378 	if (hc->ingpa & 0xffff00000000ULL)
2379 		return HV_STATUS_INVALID_PORT_ID;
2380 	/* remaining bits are reserved-zero */
2381 	if (hc->ingpa & ~KVM_HYPERV_CONN_ID_MASK)
2382 		return HV_STATUS_INVALID_HYPERCALL_INPUT;
2383 
2384 	/* the eventfd is protected by vcpu->kvm->srcu, but conn_to_evt isn't */
2385 	rcu_read_lock();
2386 	eventfd = idr_find(&hv->conn_to_evt, hc->ingpa);
2387 	rcu_read_unlock();
2388 	if (!eventfd)
2389 		return HV_STATUS_INVALID_PORT_ID;
2390 
2391 	eventfd_signal(eventfd, 1);
2392 	return HV_STATUS_SUCCESS;
2393 }
2394 
is_xmm_fast_hypercall(struct kvm_hv_hcall * hc)2395 static bool is_xmm_fast_hypercall(struct kvm_hv_hcall *hc)
2396 {
2397 	switch (hc->code) {
2398 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
2399 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
2400 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
2401 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
2402 	case HVCALL_SEND_IPI_EX:
2403 		return true;
2404 	}
2405 
2406 	return false;
2407 }
2408 
kvm_hv_hypercall_read_xmm(struct kvm_hv_hcall * hc)2409 static void kvm_hv_hypercall_read_xmm(struct kvm_hv_hcall *hc)
2410 {
2411 	int reg;
2412 
2413 	kvm_fpu_get();
2414 	for (reg = 0; reg < HV_HYPERCALL_MAX_XMM_REGISTERS; reg++)
2415 		_kvm_read_sse_reg(reg, &hc->xmm[reg]);
2416 	kvm_fpu_put();
2417 }
2418 
hv_check_hypercall_access(struct kvm_vcpu_hv * hv_vcpu,u16 code)2419 static bool hv_check_hypercall_access(struct kvm_vcpu_hv *hv_vcpu, u16 code)
2420 {
2421 	if (!hv_vcpu->enforce_cpuid)
2422 		return true;
2423 
2424 	switch (code) {
2425 	case HVCALL_NOTIFY_LONG_SPIN_WAIT:
2426 		return hv_vcpu->cpuid_cache.enlightenments_ebx &&
2427 			hv_vcpu->cpuid_cache.enlightenments_ebx != U32_MAX;
2428 	case HVCALL_POST_MESSAGE:
2429 		return hv_vcpu->cpuid_cache.features_ebx & HV_POST_MESSAGES;
2430 	case HVCALL_SIGNAL_EVENT:
2431 		return hv_vcpu->cpuid_cache.features_ebx & HV_SIGNAL_EVENTS;
2432 	case HVCALL_POST_DEBUG_DATA:
2433 	case HVCALL_RETRIEVE_DEBUG_DATA:
2434 	case HVCALL_RESET_DEBUG_SESSION:
2435 		/*
2436 		 * Return 'true' when SynDBG is disabled so the resulting code
2437 		 * will be HV_STATUS_INVALID_HYPERCALL_CODE.
2438 		 */
2439 		return !kvm_hv_is_syndbg_enabled(hv_vcpu->vcpu) ||
2440 			hv_vcpu->cpuid_cache.features_ebx & HV_DEBUGGING;
2441 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
2442 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
2443 		if (!(hv_vcpu->cpuid_cache.enlightenments_eax &
2444 		      HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED))
2445 			return false;
2446 		fallthrough;
2447 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
2448 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
2449 		return hv_vcpu->cpuid_cache.enlightenments_eax &
2450 			HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
2451 	case HVCALL_SEND_IPI_EX:
2452 		if (!(hv_vcpu->cpuid_cache.enlightenments_eax &
2453 		      HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED))
2454 			return false;
2455 		fallthrough;
2456 	case HVCALL_SEND_IPI:
2457 		return hv_vcpu->cpuid_cache.enlightenments_eax &
2458 			HV_X64_CLUSTER_IPI_RECOMMENDED;
2459 	case HV_EXT_CALL_QUERY_CAPABILITIES ... HV_EXT_CALL_MAX:
2460 		return hv_vcpu->cpuid_cache.features_ebx &
2461 			HV_ENABLE_EXTENDED_HYPERCALLS;
2462 	default:
2463 		break;
2464 	}
2465 
2466 	return true;
2467 }
2468 
kvm_hv_hypercall(struct kvm_vcpu * vcpu)2469 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
2470 {
2471 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
2472 	struct kvm_hv_hcall hc;
2473 	u64 ret = HV_STATUS_SUCCESS;
2474 
2475 	/*
2476 	 * hypercall generates UD from non zero cpl and real mode
2477 	 * per HYPER-V spec
2478 	 */
2479 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0 || !is_protmode(vcpu)) {
2480 		kvm_queue_exception(vcpu, UD_VECTOR);
2481 		return 1;
2482 	}
2483 
2484 #ifdef CONFIG_X86_64
2485 	if (is_64_bit_hypercall(vcpu)) {
2486 		hc.param = kvm_rcx_read(vcpu);
2487 		hc.ingpa = kvm_rdx_read(vcpu);
2488 		hc.outgpa = kvm_r8_read(vcpu);
2489 	} else
2490 #endif
2491 	{
2492 		hc.param = ((u64)kvm_rdx_read(vcpu) << 32) |
2493 			    (kvm_rax_read(vcpu) & 0xffffffff);
2494 		hc.ingpa = ((u64)kvm_rbx_read(vcpu) << 32) |
2495 			    (kvm_rcx_read(vcpu) & 0xffffffff);
2496 		hc.outgpa = ((u64)kvm_rdi_read(vcpu) << 32) |
2497 			     (kvm_rsi_read(vcpu) & 0xffffffff);
2498 	}
2499 
2500 	hc.code = hc.param & 0xffff;
2501 	hc.var_cnt = (hc.param & HV_HYPERCALL_VARHEAD_MASK) >> HV_HYPERCALL_VARHEAD_OFFSET;
2502 	hc.fast = !!(hc.param & HV_HYPERCALL_FAST_BIT);
2503 	hc.rep_cnt = (hc.param >> HV_HYPERCALL_REP_COMP_OFFSET) & 0xfff;
2504 	hc.rep_idx = (hc.param >> HV_HYPERCALL_REP_START_OFFSET) & 0xfff;
2505 	hc.rep = !!(hc.rep_cnt || hc.rep_idx);
2506 
2507 	trace_kvm_hv_hypercall(hc.code, hc.fast, hc.var_cnt, hc.rep_cnt,
2508 			       hc.rep_idx, hc.ingpa, hc.outgpa);
2509 
2510 	if (unlikely(!hv_check_hypercall_access(hv_vcpu, hc.code))) {
2511 		ret = HV_STATUS_ACCESS_DENIED;
2512 		goto hypercall_complete;
2513 	}
2514 
2515 	if (unlikely(hc.param & HV_HYPERCALL_RSVD_MASK)) {
2516 		ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2517 		goto hypercall_complete;
2518 	}
2519 
2520 	if (hc.fast && is_xmm_fast_hypercall(&hc)) {
2521 		if (unlikely(hv_vcpu->enforce_cpuid &&
2522 			     !(hv_vcpu->cpuid_cache.features_edx &
2523 			       HV_X64_HYPERCALL_XMM_INPUT_AVAILABLE))) {
2524 			kvm_queue_exception(vcpu, UD_VECTOR);
2525 			return 1;
2526 		}
2527 
2528 		kvm_hv_hypercall_read_xmm(&hc);
2529 	}
2530 
2531 	switch (hc.code) {
2532 	case HVCALL_NOTIFY_LONG_SPIN_WAIT:
2533 		if (unlikely(hc.rep || hc.var_cnt)) {
2534 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2535 			break;
2536 		}
2537 		kvm_vcpu_on_spin(vcpu, true);
2538 		break;
2539 	case HVCALL_SIGNAL_EVENT:
2540 		if (unlikely(hc.rep || hc.var_cnt)) {
2541 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2542 			break;
2543 		}
2544 		ret = kvm_hvcall_signal_event(vcpu, &hc);
2545 		if (ret != HV_STATUS_INVALID_PORT_ID)
2546 			break;
2547 		fallthrough;	/* maybe userspace knows this conn_id */
2548 	case HVCALL_POST_MESSAGE:
2549 		/* don't bother userspace if it has no way to handle it */
2550 		if (unlikely(hc.rep || hc.var_cnt || !to_hv_synic(vcpu)->active)) {
2551 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2552 			break;
2553 		}
2554 		goto hypercall_userspace_exit;
2555 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST:
2556 		if (unlikely(hc.var_cnt)) {
2557 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2558 			break;
2559 		}
2560 		fallthrough;
2561 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST_EX:
2562 		if (unlikely(!hc.rep_cnt || hc.rep_idx)) {
2563 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2564 			break;
2565 		}
2566 		ret = kvm_hv_flush_tlb(vcpu, &hc);
2567 		break;
2568 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
2569 		if (unlikely(hc.var_cnt)) {
2570 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2571 			break;
2572 		}
2573 		fallthrough;
2574 	case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX:
2575 		if (unlikely(hc.rep)) {
2576 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2577 			break;
2578 		}
2579 		ret = kvm_hv_flush_tlb(vcpu, &hc);
2580 		break;
2581 	case HVCALL_SEND_IPI:
2582 		if (unlikely(hc.var_cnt)) {
2583 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2584 			break;
2585 		}
2586 		fallthrough;
2587 	case HVCALL_SEND_IPI_EX:
2588 		if (unlikely(hc.rep)) {
2589 			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
2590 			break;
2591 		}
2592 		ret = kvm_hv_send_ipi(vcpu, &hc);
2593 		break;
2594 	case HVCALL_POST_DEBUG_DATA:
2595 	case HVCALL_RETRIEVE_DEBUG_DATA:
2596 		if (unlikely(hc.fast)) {
2597 			ret = HV_STATUS_INVALID_PARAMETER;
2598 			break;
2599 		}
2600 		fallthrough;
2601 	case HVCALL_RESET_DEBUG_SESSION: {
2602 		struct kvm_hv_syndbg *syndbg = to_hv_syndbg(vcpu);
2603 
2604 		if (!kvm_hv_is_syndbg_enabled(vcpu)) {
2605 			ret = HV_STATUS_INVALID_HYPERCALL_CODE;
2606 			break;
2607 		}
2608 
2609 		if (!(syndbg->options & HV_X64_SYNDBG_OPTION_USE_HCALLS)) {
2610 			ret = HV_STATUS_OPERATION_DENIED;
2611 			break;
2612 		}
2613 		goto hypercall_userspace_exit;
2614 	}
2615 	case HV_EXT_CALL_QUERY_CAPABILITIES ... HV_EXT_CALL_MAX:
2616 		if (unlikely(hc.fast)) {
2617 			ret = HV_STATUS_INVALID_PARAMETER;
2618 			break;
2619 		}
2620 		goto hypercall_userspace_exit;
2621 	default:
2622 		ret = HV_STATUS_INVALID_HYPERCALL_CODE;
2623 		break;
2624 	}
2625 
2626 hypercall_complete:
2627 	return kvm_hv_hypercall_complete(vcpu, ret);
2628 
2629 hypercall_userspace_exit:
2630 	vcpu->run->exit_reason = KVM_EXIT_HYPERV;
2631 	vcpu->run->hyperv.type = KVM_EXIT_HYPERV_HCALL;
2632 	vcpu->run->hyperv.u.hcall.input = hc.param;
2633 	vcpu->run->hyperv.u.hcall.params[0] = hc.ingpa;
2634 	vcpu->run->hyperv.u.hcall.params[1] = hc.outgpa;
2635 	vcpu->arch.complete_userspace_io = kvm_hv_hypercall_complete_userspace;
2636 	return 0;
2637 }
2638 
kvm_hv_init_vm(struct kvm * kvm)2639 void kvm_hv_init_vm(struct kvm *kvm)
2640 {
2641 	struct kvm_hv *hv = to_kvm_hv(kvm);
2642 
2643 	mutex_init(&hv->hv_lock);
2644 	idr_init(&hv->conn_to_evt);
2645 }
2646 
kvm_hv_destroy_vm(struct kvm * kvm)2647 void kvm_hv_destroy_vm(struct kvm *kvm)
2648 {
2649 	struct kvm_hv *hv = to_kvm_hv(kvm);
2650 	struct eventfd_ctx *eventfd;
2651 	int i;
2652 
2653 	idr_for_each_entry(&hv->conn_to_evt, eventfd, i)
2654 		eventfd_ctx_put(eventfd);
2655 	idr_destroy(&hv->conn_to_evt);
2656 }
2657 
kvm_hv_eventfd_assign(struct kvm * kvm,u32 conn_id,int fd)2658 static int kvm_hv_eventfd_assign(struct kvm *kvm, u32 conn_id, int fd)
2659 {
2660 	struct kvm_hv *hv = to_kvm_hv(kvm);
2661 	struct eventfd_ctx *eventfd;
2662 	int ret;
2663 
2664 	eventfd = eventfd_ctx_fdget(fd);
2665 	if (IS_ERR(eventfd))
2666 		return PTR_ERR(eventfd);
2667 
2668 	mutex_lock(&hv->hv_lock);
2669 	ret = idr_alloc(&hv->conn_to_evt, eventfd, conn_id, conn_id + 1,
2670 			GFP_KERNEL_ACCOUNT);
2671 	mutex_unlock(&hv->hv_lock);
2672 
2673 	if (ret >= 0)
2674 		return 0;
2675 
2676 	if (ret == -ENOSPC)
2677 		ret = -EEXIST;
2678 	eventfd_ctx_put(eventfd);
2679 	return ret;
2680 }
2681 
kvm_hv_eventfd_deassign(struct kvm * kvm,u32 conn_id)2682 static int kvm_hv_eventfd_deassign(struct kvm *kvm, u32 conn_id)
2683 {
2684 	struct kvm_hv *hv = to_kvm_hv(kvm);
2685 	struct eventfd_ctx *eventfd;
2686 
2687 	mutex_lock(&hv->hv_lock);
2688 	eventfd = idr_remove(&hv->conn_to_evt, conn_id);
2689 	mutex_unlock(&hv->hv_lock);
2690 
2691 	if (!eventfd)
2692 		return -ENOENT;
2693 
2694 	synchronize_srcu(&kvm->srcu);
2695 	eventfd_ctx_put(eventfd);
2696 	return 0;
2697 }
2698 
kvm_vm_ioctl_hv_eventfd(struct kvm * kvm,struct kvm_hyperv_eventfd * args)2699 int kvm_vm_ioctl_hv_eventfd(struct kvm *kvm, struct kvm_hyperv_eventfd *args)
2700 {
2701 	if ((args->flags & ~KVM_HYPERV_EVENTFD_DEASSIGN) ||
2702 	    (args->conn_id & ~KVM_HYPERV_CONN_ID_MASK))
2703 		return -EINVAL;
2704 
2705 	if (args->flags == KVM_HYPERV_EVENTFD_DEASSIGN)
2706 		return kvm_hv_eventfd_deassign(kvm, args->conn_id);
2707 	return kvm_hv_eventfd_assign(kvm, args->conn_id, args->fd);
2708 }
2709 
kvm_get_hv_cpuid(struct kvm_vcpu * vcpu,struct kvm_cpuid2 * cpuid,struct kvm_cpuid_entry2 __user * entries)2710 int kvm_get_hv_cpuid(struct kvm_vcpu *vcpu, struct kvm_cpuid2 *cpuid,
2711 		     struct kvm_cpuid_entry2 __user *entries)
2712 {
2713 	uint16_t evmcs_ver = 0;
2714 	struct kvm_cpuid_entry2 cpuid_entries[] = {
2715 		{ .function = HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS },
2716 		{ .function = HYPERV_CPUID_INTERFACE },
2717 		{ .function = HYPERV_CPUID_VERSION },
2718 		{ .function = HYPERV_CPUID_FEATURES },
2719 		{ .function = HYPERV_CPUID_ENLIGHTMENT_INFO },
2720 		{ .function = HYPERV_CPUID_IMPLEMENT_LIMITS },
2721 		{ .function = HYPERV_CPUID_SYNDBG_VENDOR_AND_MAX_FUNCTIONS },
2722 		{ .function = HYPERV_CPUID_SYNDBG_INTERFACE },
2723 		{ .function = HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES	},
2724 		{ .function = HYPERV_CPUID_NESTED_FEATURES },
2725 	};
2726 	int i, nent = ARRAY_SIZE(cpuid_entries);
2727 
2728 	if (kvm_x86_ops.nested_ops->get_evmcs_version)
2729 		evmcs_ver = kvm_x86_ops.nested_ops->get_evmcs_version(vcpu);
2730 
2731 	if (cpuid->nent < nent)
2732 		return -E2BIG;
2733 
2734 	if (cpuid->nent > nent)
2735 		cpuid->nent = nent;
2736 
2737 	for (i = 0; i < nent; i++) {
2738 		struct kvm_cpuid_entry2 *ent = &cpuid_entries[i];
2739 		u32 signature[3];
2740 
2741 		switch (ent->function) {
2742 		case HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS:
2743 			memcpy(signature, "Linux KVM Hv", 12);
2744 
2745 			ent->eax = HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES;
2746 			ent->ebx = signature[0];
2747 			ent->ecx = signature[1];
2748 			ent->edx = signature[2];
2749 			break;
2750 
2751 		case HYPERV_CPUID_INTERFACE:
2752 			ent->eax = HYPERV_CPUID_SIGNATURE_EAX;
2753 			break;
2754 
2755 		case HYPERV_CPUID_VERSION:
2756 			/*
2757 			 * We implement some Hyper-V 2016 functions so let's use
2758 			 * this version.
2759 			 */
2760 			ent->eax = 0x00003839;
2761 			ent->ebx = 0x000A0000;
2762 			break;
2763 
2764 		case HYPERV_CPUID_FEATURES:
2765 			ent->eax |= HV_MSR_VP_RUNTIME_AVAILABLE;
2766 			ent->eax |= HV_MSR_TIME_REF_COUNT_AVAILABLE;
2767 			ent->eax |= HV_MSR_SYNIC_AVAILABLE;
2768 			ent->eax |= HV_MSR_SYNTIMER_AVAILABLE;
2769 			ent->eax |= HV_MSR_APIC_ACCESS_AVAILABLE;
2770 			ent->eax |= HV_MSR_HYPERCALL_AVAILABLE;
2771 			ent->eax |= HV_MSR_VP_INDEX_AVAILABLE;
2772 			ent->eax |= HV_MSR_RESET_AVAILABLE;
2773 			ent->eax |= HV_MSR_REFERENCE_TSC_AVAILABLE;
2774 			ent->eax |= HV_ACCESS_FREQUENCY_MSRS;
2775 			ent->eax |= HV_ACCESS_REENLIGHTENMENT;
2776 			ent->eax |= HV_ACCESS_TSC_INVARIANT;
2777 
2778 			ent->ebx |= HV_POST_MESSAGES;
2779 			ent->ebx |= HV_SIGNAL_EVENTS;
2780 			ent->ebx |= HV_ENABLE_EXTENDED_HYPERCALLS;
2781 
2782 			ent->edx |= HV_X64_HYPERCALL_XMM_INPUT_AVAILABLE;
2783 			ent->edx |= HV_FEATURE_FREQUENCY_MSRS_AVAILABLE;
2784 			ent->edx |= HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
2785 
2786 			ent->ebx |= HV_DEBUGGING;
2787 			ent->edx |= HV_X64_GUEST_DEBUGGING_AVAILABLE;
2788 			ent->edx |= HV_FEATURE_DEBUG_MSRS_AVAILABLE;
2789 			ent->edx |= HV_FEATURE_EXT_GVA_RANGES_FLUSH;
2790 
2791 			/*
2792 			 * Direct Synthetic timers only make sense with in-kernel
2793 			 * LAPIC
2794 			 */
2795 			if (!vcpu || lapic_in_kernel(vcpu))
2796 				ent->edx |= HV_STIMER_DIRECT_MODE_AVAILABLE;
2797 
2798 			break;
2799 
2800 		case HYPERV_CPUID_ENLIGHTMENT_INFO:
2801 			ent->eax |= HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
2802 			ent->eax |= HV_X64_APIC_ACCESS_RECOMMENDED;
2803 			ent->eax |= HV_X64_RELAXED_TIMING_RECOMMENDED;
2804 			ent->eax |= HV_X64_CLUSTER_IPI_RECOMMENDED;
2805 			ent->eax |= HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED;
2806 			if (evmcs_ver)
2807 				ent->eax |= HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
2808 			if (!cpu_smt_possible())
2809 				ent->eax |= HV_X64_NO_NONARCH_CORESHARING;
2810 
2811 			ent->eax |= HV_DEPRECATING_AEOI_RECOMMENDED;
2812 			/*
2813 			 * Default number of spinlock retry attempts, matches
2814 			 * HyperV 2016.
2815 			 */
2816 			ent->ebx = 0x00000FFF;
2817 
2818 			break;
2819 
2820 		case HYPERV_CPUID_IMPLEMENT_LIMITS:
2821 			/* Maximum number of virtual processors */
2822 			ent->eax = KVM_MAX_VCPUS;
2823 			/*
2824 			 * Maximum number of logical processors, matches
2825 			 * HyperV 2016.
2826 			 */
2827 			ent->ebx = 64;
2828 
2829 			break;
2830 
2831 		case HYPERV_CPUID_NESTED_FEATURES:
2832 			ent->eax = evmcs_ver;
2833 			ent->eax |= HV_X64_NESTED_DIRECT_FLUSH;
2834 			ent->eax |= HV_X64_NESTED_MSR_BITMAP;
2835 			ent->ebx |= HV_X64_NESTED_EVMCS1_PERF_GLOBAL_CTRL;
2836 			break;
2837 
2838 		case HYPERV_CPUID_SYNDBG_VENDOR_AND_MAX_FUNCTIONS:
2839 			memcpy(signature, "Linux KVM Hv", 12);
2840 
2841 			ent->eax = 0;
2842 			ent->ebx = signature[0];
2843 			ent->ecx = signature[1];
2844 			ent->edx = signature[2];
2845 			break;
2846 
2847 		case HYPERV_CPUID_SYNDBG_INTERFACE:
2848 			memcpy(signature, "VS#1\0\0\0\0\0\0\0\0", 12);
2849 			ent->eax = signature[0];
2850 			break;
2851 
2852 		case HYPERV_CPUID_SYNDBG_PLATFORM_CAPABILITIES:
2853 			ent->eax |= HV_X64_SYNDBG_CAP_ALLOW_KERNEL_DEBUGGING;
2854 			break;
2855 
2856 		default:
2857 			break;
2858 		}
2859 	}
2860 
2861 	if (copy_to_user(entries, cpuid_entries,
2862 			 nent * sizeof(struct kvm_cpuid_entry2)))
2863 		return -EFAULT;
2864 
2865 	return 0;
2866 }
2867