xref: /linux/arch/x86/kvm/x86.c (revision 58b3bafff8257c6946df5d6aeb215b8ac839ed2a)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * derived from drivers/kvm/kvm_main.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  *
12  * Authors:
13  *   Avi Kivity   <avi@qumranet.com>
14  *   Yaniv Kamay  <yaniv@qumranet.com>
15  *   Amit Shah    <amit.shah@qumranet.com>
16  *   Ben-Ami Yassour <benami@il.ibm.com>
17  */
18 
19 #include <linux/kvm_host.h>
20 #include "irq.h"
21 #include "mmu.h"
22 #include "i8254.h"
23 #include "tss.h"
24 #include "kvm_cache_regs.h"
25 #include "x86.h"
26 #include "cpuid.h"
27 #include "pmu.h"
28 #include "hyperv.h"
29 
30 #include <linux/clocksource.h>
31 #include <linux/interrupt.h>
32 #include <linux/kvm.h>
33 #include <linux/fs.h>
34 #include <linux/vmalloc.h>
35 #include <linux/export.h>
36 #include <linux/moduleparam.h>
37 #include <linux/mman.h>
38 #include <linux/highmem.h>
39 #include <linux/iommu.h>
40 #include <linux/intel-iommu.h>
41 #include <linux/cpufreq.h>
42 #include <linux/user-return-notifier.h>
43 #include <linux/srcu.h>
44 #include <linux/slab.h>
45 #include <linux/perf_event.h>
46 #include <linux/uaccess.h>
47 #include <linux/hash.h>
48 #include <linux/pci.h>
49 #include <linux/timekeeper_internal.h>
50 #include <linux/pvclock_gtod.h>
51 #include <linux/kvm_irqfd.h>
52 #include <linux/irqbypass.h>
53 #include <linux/sched/stat.h>
54 #include <linux/sched/isolation.h>
55 #include <linux/mem_encrypt.h>
56 
57 #include <trace/events/kvm.h>
58 
59 #include <asm/debugreg.h>
60 #include <asm/msr.h>
61 #include <asm/desc.h>
62 #include <asm/mce.h>
63 #include <linux/kernel_stat.h>
64 #include <asm/fpu/internal.h> /* Ugh! */
65 #include <asm/pvclock.h>
66 #include <asm/div64.h>
67 #include <asm/irq_remapping.h>
68 #include <asm/mshyperv.h>
69 #include <asm/hypervisor.h>
70 #include <asm/intel_pt.h>
71 #include <asm/emulate_prefix.h>
72 #include <clocksource/hyperv_timer.h>
73 
74 #define CREATE_TRACE_POINTS
75 #include "trace.h"
76 
77 #define MAX_IO_MSRS 256
78 #define KVM_MAX_MCE_BANKS 32
79 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
80 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
81 
82 #define emul_to_vcpu(ctxt) \
83 	container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
84 
85 /* EFER defaults:
86  * - enable syscall per default because its emulated by KVM
87  * - enable LME and LMA per default on 64 bit KVM
88  */
89 #ifdef CONFIG_X86_64
90 static
91 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
92 #else
93 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
94 #endif
95 
96 #define VM_STAT(x, ...) offsetof(struct kvm, stat.x), KVM_STAT_VM, ## __VA_ARGS__
97 #define VCPU_STAT(x, ...) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU, ## __VA_ARGS__
98 
99 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
100                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
101 
102 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
103 static void process_nmi(struct kvm_vcpu *vcpu);
104 static void enter_smm(struct kvm_vcpu *vcpu);
105 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
106 static void store_regs(struct kvm_vcpu *vcpu);
107 static int sync_regs(struct kvm_vcpu *vcpu);
108 
109 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
110 EXPORT_SYMBOL_GPL(kvm_x86_ops);
111 
112 static bool __read_mostly ignore_msrs = 0;
113 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
114 
115 static bool __read_mostly report_ignored_msrs = true;
116 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
117 
118 unsigned int min_timer_period_us = 200;
119 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
120 
121 static bool __read_mostly kvmclock_periodic_sync = true;
122 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
123 
124 bool __read_mostly kvm_has_tsc_control;
125 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
126 u32  __read_mostly kvm_max_guest_tsc_khz;
127 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
128 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
129 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
130 u64  __read_mostly kvm_max_tsc_scaling_ratio;
131 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
132 u64 __read_mostly kvm_default_tsc_scaling_ratio;
133 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
134 
135 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
136 static u32 __read_mostly tsc_tolerance_ppm = 250;
137 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
138 
139 /*
140  * lapic timer advance (tscdeadline mode only) in nanoseconds.  '-1' enables
141  * adaptive tuning starting from default advancment of 1000ns.  '0' disables
142  * advancement entirely.  Any other value is used as-is and disables adaptive
143  * tuning, i.e. allows priveleged userspace to set an exact advancement time.
144  */
145 static int __read_mostly lapic_timer_advance_ns = -1;
146 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
147 
148 static bool __read_mostly vector_hashing = true;
149 module_param(vector_hashing, bool, S_IRUGO);
150 
151 bool __read_mostly enable_vmware_backdoor = false;
152 module_param(enable_vmware_backdoor, bool, S_IRUGO);
153 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
154 
155 static bool __read_mostly force_emulation_prefix = false;
156 module_param(force_emulation_prefix, bool, S_IRUGO);
157 
158 int __read_mostly pi_inject_timer = -1;
159 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
160 
161 #define KVM_NR_SHARED_MSRS 16
162 
163 struct kvm_shared_msrs_global {
164 	int nr;
165 	u32 msrs[KVM_NR_SHARED_MSRS];
166 };
167 
168 struct kvm_shared_msrs {
169 	struct user_return_notifier urn;
170 	bool registered;
171 	struct kvm_shared_msr_values {
172 		u64 host;
173 		u64 curr;
174 	} values[KVM_NR_SHARED_MSRS];
175 };
176 
177 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
178 static struct kvm_shared_msrs __percpu *shared_msrs;
179 
180 static u64 __read_mostly host_xss;
181 
182 struct kvm_stats_debugfs_item debugfs_entries[] = {
183 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
184 	{ "pf_guest", VCPU_STAT(pf_guest) },
185 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
186 	{ "invlpg", VCPU_STAT(invlpg) },
187 	{ "exits", VCPU_STAT(exits) },
188 	{ "io_exits", VCPU_STAT(io_exits) },
189 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
190 	{ "signal_exits", VCPU_STAT(signal_exits) },
191 	{ "irq_window", VCPU_STAT(irq_window_exits) },
192 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
193 	{ "halt_exits", VCPU_STAT(halt_exits) },
194 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
195 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
196 	{ "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
197 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
198 	{ "hypercalls", VCPU_STAT(hypercalls) },
199 	{ "request_irq", VCPU_STAT(request_irq_exits) },
200 	{ "irq_exits", VCPU_STAT(irq_exits) },
201 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
202 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
203 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
204 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
205 	{ "irq_injections", VCPU_STAT(irq_injections) },
206 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
207 	{ "req_event", VCPU_STAT(req_event) },
208 	{ "l1d_flush", VCPU_STAT(l1d_flush) },
209 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
210 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
211 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
212 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
213 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
214 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
215 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
216 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
217 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
218 	{ "largepages", VM_STAT(lpages, .mode = 0444) },
219 	{ "nx_largepages_splitted", VM_STAT(nx_lpage_splits, .mode = 0444) },
220 	{ "max_mmu_page_hash_collisions",
221 		VM_STAT(max_mmu_page_hash_collisions) },
222 	{ NULL }
223 };
224 
225 u64 __read_mostly host_xcr0;
226 
227 struct kmem_cache *x86_fpu_cache;
228 EXPORT_SYMBOL_GPL(x86_fpu_cache);
229 
230 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
231 
232 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
233 {
234 	int i;
235 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
236 		vcpu->arch.apf.gfns[i] = ~0;
237 }
238 
239 static void kvm_on_user_return(struct user_return_notifier *urn)
240 {
241 	unsigned slot;
242 	struct kvm_shared_msrs *locals
243 		= container_of(urn, struct kvm_shared_msrs, urn);
244 	struct kvm_shared_msr_values *values;
245 	unsigned long flags;
246 
247 	/*
248 	 * Disabling irqs at this point since the following code could be
249 	 * interrupted and executed through kvm_arch_hardware_disable()
250 	 */
251 	local_irq_save(flags);
252 	if (locals->registered) {
253 		locals->registered = false;
254 		user_return_notifier_unregister(urn);
255 	}
256 	local_irq_restore(flags);
257 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
258 		values = &locals->values[slot];
259 		if (values->host != values->curr) {
260 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
261 			values->curr = values->host;
262 		}
263 	}
264 }
265 
266 void kvm_define_shared_msr(unsigned slot, u32 msr)
267 {
268 	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
269 	shared_msrs_global.msrs[slot] = msr;
270 	if (slot >= shared_msrs_global.nr)
271 		shared_msrs_global.nr = slot + 1;
272 }
273 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
274 
275 static void kvm_shared_msr_cpu_online(void)
276 {
277 	unsigned int cpu = smp_processor_id();
278 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
279 	u64 value;
280 	int i;
281 
282 	for (i = 0; i < shared_msrs_global.nr; ++i) {
283 		rdmsrl_safe(shared_msrs_global.msrs[i], &value);
284 		smsr->values[i].host = value;
285 		smsr->values[i].curr = value;
286 	}
287 }
288 
289 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
290 {
291 	unsigned int cpu = smp_processor_id();
292 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
293 	int err;
294 
295 	value = (value & mask) | (smsr->values[slot].host & ~mask);
296 	if (value == smsr->values[slot].curr)
297 		return 0;
298 	err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
299 	if (err)
300 		return 1;
301 
302 	smsr->values[slot].curr = value;
303 	if (!smsr->registered) {
304 		smsr->urn.on_user_return = kvm_on_user_return;
305 		user_return_notifier_register(&smsr->urn);
306 		smsr->registered = true;
307 	}
308 	return 0;
309 }
310 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
311 
312 static void drop_user_return_notifiers(void)
313 {
314 	unsigned int cpu = smp_processor_id();
315 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
316 
317 	if (smsr->registered)
318 		kvm_on_user_return(&smsr->urn);
319 }
320 
321 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
322 {
323 	return vcpu->arch.apic_base;
324 }
325 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
326 
327 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
328 {
329 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
330 }
331 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
332 
333 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
334 {
335 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
336 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
337 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
338 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
339 
340 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
341 		return 1;
342 	if (!msr_info->host_initiated) {
343 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
344 			return 1;
345 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
346 			return 1;
347 	}
348 
349 	kvm_lapic_set_base(vcpu, msr_info->data);
350 	return 0;
351 }
352 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
353 
354 asmlinkage __visible void kvm_spurious_fault(void)
355 {
356 	/* Fault while not rebooting.  We want the trace. */
357 	BUG_ON(!kvm_rebooting);
358 }
359 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
360 
361 #define EXCPT_BENIGN		0
362 #define EXCPT_CONTRIBUTORY	1
363 #define EXCPT_PF		2
364 
365 static int exception_class(int vector)
366 {
367 	switch (vector) {
368 	case PF_VECTOR:
369 		return EXCPT_PF;
370 	case DE_VECTOR:
371 	case TS_VECTOR:
372 	case NP_VECTOR:
373 	case SS_VECTOR:
374 	case GP_VECTOR:
375 		return EXCPT_CONTRIBUTORY;
376 	default:
377 		break;
378 	}
379 	return EXCPT_BENIGN;
380 }
381 
382 #define EXCPT_FAULT		0
383 #define EXCPT_TRAP		1
384 #define EXCPT_ABORT		2
385 #define EXCPT_INTERRUPT		3
386 
387 static int exception_type(int vector)
388 {
389 	unsigned int mask;
390 
391 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
392 		return EXCPT_INTERRUPT;
393 
394 	mask = 1 << vector;
395 
396 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
397 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
398 		return EXCPT_TRAP;
399 
400 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
401 		return EXCPT_ABORT;
402 
403 	/* Reserved exceptions will result in fault */
404 	return EXCPT_FAULT;
405 }
406 
407 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
408 {
409 	unsigned nr = vcpu->arch.exception.nr;
410 	bool has_payload = vcpu->arch.exception.has_payload;
411 	unsigned long payload = vcpu->arch.exception.payload;
412 
413 	if (!has_payload)
414 		return;
415 
416 	switch (nr) {
417 	case DB_VECTOR:
418 		/*
419 		 * "Certain debug exceptions may clear bit 0-3.  The
420 		 * remaining contents of the DR6 register are never
421 		 * cleared by the processor".
422 		 */
423 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
424 		/*
425 		 * DR6.RTM is set by all #DB exceptions that don't clear it.
426 		 */
427 		vcpu->arch.dr6 |= DR6_RTM;
428 		vcpu->arch.dr6 |= payload;
429 		/*
430 		 * Bit 16 should be set in the payload whenever the #DB
431 		 * exception should clear DR6.RTM. This makes the payload
432 		 * compatible with the pending debug exceptions under VMX.
433 		 * Though not currently documented in the SDM, this also
434 		 * makes the payload compatible with the exit qualification
435 		 * for #DB exceptions under VMX.
436 		 */
437 		vcpu->arch.dr6 ^= payload & DR6_RTM;
438 		break;
439 	case PF_VECTOR:
440 		vcpu->arch.cr2 = payload;
441 		break;
442 	}
443 
444 	vcpu->arch.exception.has_payload = false;
445 	vcpu->arch.exception.payload = 0;
446 }
447 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
448 
449 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
450 		unsigned nr, bool has_error, u32 error_code,
451 	        bool has_payload, unsigned long payload, bool reinject)
452 {
453 	u32 prev_nr;
454 	int class1, class2;
455 
456 	kvm_make_request(KVM_REQ_EVENT, vcpu);
457 
458 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
459 	queue:
460 		if (has_error && !is_protmode(vcpu))
461 			has_error = false;
462 		if (reinject) {
463 			/*
464 			 * On vmentry, vcpu->arch.exception.pending is only
465 			 * true if an event injection was blocked by
466 			 * nested_run_pending.  In that case, however,
467 			 * vcpu_enter_guest requests an immediate exit,
468 			 * and the guest shouldn't proceed far enough to
469 			 * need reinjection.
470 			 */
471 			WARN_ON_ONCE(vcpu->arch.exception.pending);
472 			vcpu->arch.exception.injected = true;
473 			if (WARN_ON_ONCE(has_payload)) {
474 				/*
475 				 * A reinjected event has already
476 				 * delivered its payload.
477 				 */
478 				has_payload = false;
479 				payload = 0;
480 			}
481 		} else {
482 			vcpu->arch.exception.pending = true;
483 			vcpu->arch.exception.injected = false;
484 		}
485 		vcpu->arch.exception.has_error_code = has_error;
486 		vcpu->arch.exception.nr = nr;
487 		vcpu->arch.exception.error_code = error_code;
488 		vcpu->arch.exception.has_payload = has_payload;
489 		vcpu->arch.exception.payload = payload;
490 		/*
491 		 * In guest mode, payload delivery should be deferred,
492 		 * so that the L1 hypervisor can intercept #PF before
493 		 * CR2 is modified (or intercept #DB before DR6 is
494 		 * modified under nVMX).  However, for ABI
495 		 * compatibility with KVM_GET_VCPU_EVENTS and
496 		 * KVM_SET_VCPU_EVENTS, we can't delay payload
497 		 * delivery unless userspace has enabled this
498 		 * functionality via the per-VM capability,
499 		 * KVM_CAP_EXCEPTION_PAYLOAD.
500 		 */
501 		if (!vcpu->kvm->arch.exception_payload_enabled ||
502 		    !is_guest_mode(vcpu))
503 			kvm_deliver_exception_payload(vcpu);
504 		return;
505 	}
506 
507 	/* to check exception */
508 	prev_nr = vcpu->arch.exception.nr;
509 	if (prev_nr == DF_VECTOR) {
510 		/* triple fault -> shutdown */
511 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
512 		return;
513 	}
514 	class1 = exception_class(prev_nr);
515 	class2 = exception_class(nr);
516 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
517 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
518 		/*
519 		 * Generate double fault per SDM Table 5-5.  Set
520 		 * exception.pending = true so that the double fault
521 		 * can trigger a nested vmexit.
522 		 */
523 		vcpu->arch.exception.pending = true;
524 		vcpu->arch.exception.injected = false;
525 		vcpu->arch.exception.has_error_code = true;
526 		vcpu->arch.exception.nr = DF_VECTOR;
527 		vcpu->arch.exception.error_code = 0;
528 		vcpu->arch.exception.has_payload = false;
529 		vcpu->arch.exception.payload = 0;
530 	} else
531 		/* replace previous exception with a new one in a hope
532 		   that instruction re-execution will regenerate lost
533 		   exception */
534 		goto queue;
535 }
536 
537 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
538 {
539 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
540 }
541 EXPORT_SYMBOL_GPL(kvm_queue_exception);
542 
543 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
544 {
545 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
546 }
547 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
548 
549 static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
550 				  unsigned long payload)
551 {
552 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
553 }
554 
555 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
556 				    u32 error_code, unsigned long payload)
557 {
558 	kvm_multiple_exception(vcpu, nr, true, error_code,
559 			       true, payload, false);
560 }
561 
562 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
563 {
564 	if (err)
565 		kvm_inject_gp(vcpu, 0);
566 	else
567 		return kvm_skip_emulated_instruction(vcpu);
568 
569 	return 1;
570 }
571 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
572 
573 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
574 {
575 	++vcpu->stat.pf_guest;
576 	vcpu->arch.exception.nested_apf =
577 		is_guest_mode(vcpu) && fault->async_page_fault;
578 	if (vcpu->arch.exception.nested_apf) {
579 		vcpu->arch.apf.nested_apf_token = fault->address;
580 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
581 	} else {
582 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
583 					fault->address);
584 	}
585 }
586 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
587 
588 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
589 {
590 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
591 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
592 	else
593 		vcpu->arch.mmu->inject_page_fault(vcpu, fault);
594 
595 	return fault->nested_page_fault;
596 }
597 
598 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
599 {
600 	atomic_inc(&vcpu->arch.nmi_queued);
601 	kvm_make_request(KVM_REQ_NMI, vcpu);
602 }
603 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
604 
605 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
606 {
607 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
608 }
609 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
610 
611 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
612 {
613 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
614 }
615 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
616 
617 /*
618  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
619  * a #GP and return false.
620  */
621 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
622 {
623 	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
624 		return true;
625 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
626 	return false;
627 }
628 EXPORT_SYMBOL_GPL(kvm_require_cpl);
629 
630 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
631 {
632 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
633 		return true;
634 
635 	kvm_queue_exception(vcpu, UD_VECTOR);
636 	return false;
637 }
638 EXPORT_SYMBOL_GPL(kvm_require_dr);
639 
640 /*
641  * This function will be used to read from the physical memory of the currently
642  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
643  * can read from guest physical or from the guest's guest physical memory.
644  */
645 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
646 			    gfn_t ngfn, void *data, int offset, int len,
647 			    u32 access)
648 {
649 	struct x86_exception exception;
650 	gfn_t real_gfn;
651 	gpa_t ngpa;
652 
653 	ngpa     = gfn_to_gpa(ngfn);
654 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
655 	if (real_gfn == UNMAPPED_GVA)
656 		return -EFAULT;
657 
658 	real_gfn = gpa_to_gfn(real_gfn);
659 
660 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
661 }
662 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
663 
664 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
665 			       void *data, int offset, int len, u32 access)
666 {
667 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
668 				       data, offset, len, access);
669 }
670 
671 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
672 {
673 	return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
674 	       rsvd_bits(1, 2);
675 }
676 
677 /*
678  * Load the pae pdptrs.  Return 1 if they are all valid, 0 otherwise.
679  */
680 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
681 {
682 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
683 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
684 	int i;
685 	int ret;
686 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
687 
688 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
689 				      offset * sizeof(u64), sizeof(pdpte),
690 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
691 	if (ret < 0) {
692 		ret = 0;
693 		goto out;
694 	}
695 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
696 		if ((pdpte[i] & PT_PRESENT_MASK) &&
697 		    (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
698 			ret = 0;
699 			goto out;
700 		}
701 	}
702 	ret = 1;
703 
704 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
705 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
706 
707 out:
708 
709 	return ret;
710 }
711 EXPORT_SYMBOL_GPL(load_pdptrs);
712 
713 bool pdptrs_changed(struct kvm_vcpu *vcpu)
714 {
715 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
716 	int offset;
717 	gfn_t gfn;
718 	int r;
719 
720 	if (!is_pae_paging(vcpu))
721 		return false;
722 
723 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
724 		return true;
725 
726 	gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
727 	offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
728 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
729 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
730 	if (r < 0)
731 		return true;
732 
733 	return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
734 }
735 EXPORT_SYMBOL_GPL(pdptrs_changed);
736 
737 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
738 {
739 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
740 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
741 
742 	cr0 |= X86_CR0_ET;
743 
744 #ifdef CONFIG_X86_64
745 	if (cr0 & 0xffffffff00000000UL)
746 		return 1;
747 #endif
748 
749 	cr0 &= ~CR0_RESERVED_BITS;
750 
751 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
752 		return 1;
753 
754 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
755 		return 1;
756 
757 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
758 #ifdef CONFIG_X86_64
759 		if ((vcpu->arch.efer & EFER_LME)) {
760 			int cs_db, cs_l;
761 
762 			if (!is_pae(vcpu))
763 				return 1;
764 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
765 			if (cs_l)
766 				return 1;
767 		} else
768 #endif
769 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
770 						 kvm_read_cr3(vcpu)))
771 			return 1;
772 	}
773 
774 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
775 		return 1;
776 
777 	kvm_x86_ops->set_cr0(vcpu, cr0);
778 
779 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
780 		kvm_clear_async_pf_completion_queue(vcpu);
781 		kvm_async_pf_hash_reset(vcpu);
782 	}
783 
784 	if ((cr0 ^ old_cr0) & update_bits)
785 		kvm_mmu_reset_context(vcpu);
786 
787 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
788 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
789 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
790 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
791 
792 	return 0;
793 }
794 EXPORT_SYMBOL_GPL(kvm_set_cr0);
795 
796 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
797 {
798 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
799 }
800 EXPORT_SYMBOL_GPL(kvm_lmsw);
801 
802 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
803 {
804 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
805 
806 		if (vcpu->arch.xcr0 != host_xcr0)
807 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
808 
809 		if (vcpu->arch.xsaves_enabled &&
810 		    vcpu->arch.ia32_xss != host_xss)
811 			wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
812 	}
813 }
814 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
815 
816 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
817 {
818 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
819 
820 		if (vcpu->arch.xcr0 != host_xcr0)
821 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
822 
823 		if (vcpu->arch.xsaves_enabled &&
824 		    vcpu->arch.ia32_xss != host_xss)
825 			wrmsrl(MSR_IA32_XSS, host_xss);
826 	}
827 
828 }
829 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
830 
831 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
832 {
833 	u64 xcr0 = xcr;
834 	u64 old_xcr0 = vcpu->arch.xcr0;
835 	u64 valid_bits;
836 
837 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
838 	if (index != XCR_XFEATURE_ENABLED_MASK)
839 		return 1;
840 	if (!(xcr0 & XFEATURE_MASK_FP))
841 		return 1;
842 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
843 		return 1;
844 
845 	/*
846 	 * Do not allow the guest to set bits that we do not support
847 	 * saving.  However, xcr0 bit 0 is always set, even if the
848 	 * emulated CPU does not support XSAVE (see fx_init).
849 	 */
850 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
851 	if (xcr0 & ~valid_bits)
852 		return 1;
853 
854 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
855 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
856 		return 1;
857 
858 	if (xcr0 & XFEATURE_MASK_AVX512) {
859 		if (!(xcr0 & XFEATURE_MASK_YMM))
860 			return 1;
861 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
862 			return 1;
863 	}
864 	vcpu->arch.xcr0 = xcr0;
865 
866 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
867 		kvm_update_cpuid(vcpu);
868 	return 0;
869 }
870 
871 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
872 {
873 	if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
874 	    __kvm_set_xcr(vcpu, index, xcr)) {
875 		kvm_inject_gp(vcpu, 0);
876 		return 1;
877 	}
878 	return 0;
879 }
880 EXPORT_SYMBOL_GPL(kvm_set_xcr);
881 
882 static int kvm_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
883 {
884 	if (cr4 & CR4_RESERVED_BITS)
885 		return -EINVAL;
886 
887 	if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
888 		return -EINVAL;
889 
890 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
891 		return -EINVAL;
892 
893 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
894 		return -EINVAL;
895 
896 	if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
897 		return -EINVAL;
898 
899 	if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
900 		return -EINVAL;
901 
902 	if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
903 		return -EINVAL;
904 
905 	if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
906 		return -EINVAL;
907 
908 	return 0;
909 }
910 
911 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
912 {
913 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
914 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
915 				   X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
916 
917 	if (kvm_valid_cr4(vcpu, cr4))
918 		return 1;
919 
920 	if (is_long_mode(vcpu)) {
921 		if (!(cr4 & X86_CR4_PAE))
922 			return 1;
923 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
924 		   && ((cr4 ^ old_cr4) & pdptr_bits)
925 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
926 				   kvm_read_cr3(vcpu)))
927 		return 1;
928 
929 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
930 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
931 			return 1;
932 
933 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
934 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
935 			return 1;
936 	}
937 
938 	if (kvm_x86_ops->set_cr4(vcpu, cr4))
939 		return 1;
940 
941 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
942 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
943 		kvm_mmu_reset_context(vcpu);
944 
945 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
946 		kvm_update_cpuid(vcpu);
947 
948 	return 0;
949 }
950 EXPORT_SYMBOL_GPL(kvm_set_cr4);
951 
952 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
953 {
954 	bool skip_tlb_flush = false;
955 #ifdef CONFIG_X86_64
956 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
957 
958 	if (pcid_enabled) {
959 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
960 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
961 	}
962 #endif
963 
964 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
965 		if (!skip_tlb_flush) {
966 			kvm_mmu_sync_roots(vcpu);
967 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
968 		}
969 		return 0;
970 	}
971 
972 	if (is_long_mode(vcpu) &&
973 	    (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
974 		return 1;
975 	else if (is_pae_paging(vcpu) &&
976 		 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
977 		return 1;
978 
979 	kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
980 	vcpu->arch.cr3 = cr3;
981 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
982 
983 	return 0;
984 }
985 EXPORT_SYMBOL_GPL(kvm_set_cr3);
986 
987 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
988 {
989 	if (cr8 & CR8_RESERVED_BITS)
990 		return 1;
991 	if (lapic_in_kernel(vcpu))
992 		kvm_lapic_set_tpr(vcpu, cr8);
993 	else
994 		vcpu->arch.cr8 = cr8;
995 	return 0;
996 }
997 EXPORT_SYMBOL_GPL(kvm_set_cr8);
998 
999 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1000 {
1001 	if (lapic_in_kernel(vcpu))
1002 		return kvm_lapic_get_cr8(vcpu);
1003 	else
1004 		return vcpu->arch.cr8;
1005 }
1006 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1007 
1008 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1009 {
1010 	int i;
1011 
1012 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1013 		for (i = 0; i < KVM_NR_DB_REGS; i++)
1014 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1015 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1016 	}
1017 }
1018 
1019 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1020 {
1021 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1022 		kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
1023 }
1024 
1025 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1026 {
1027 	unsigned long dr7;
1028 
1029 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1030 		dr7 = vcpu->arch.guest_debug_dr7;
1031 	else
1032 		dr7 = vcpu->arch.dr7;
1033 	kvm_x86_ops->set_dr7(vcpu, dr7);
1034 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1035 	if (dr7 & DR7_BP_EN_MASK)
1036 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1037 }
1038 
1039 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1040 {
1041 	u64 fixed = DR6_FIXED_1;
1042 
1043 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1044 		fixed |= DR6_RTM;
1045 	return fixed;
1046 }
1047 
1048 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1049 {
1050 	switch (dr) {
1051 	case 0 ... 3:
1052 		vcpu->arch.db[dr] = val;
1053 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1054 			vcpu->arch.eff_db[dr] = val;
1055 		break;
1056 	case 4:
1057 		/* fall through */
1058 	case 6:
1059 		if (val & 0xffffffff00000000ULL)
1060 			return -1; /* #GP */
1061 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1062 		kvm_update_dr6(vcpu);
1063 		break;
1064 	case 5:
1065 		/* fall through */
1066 	default: /* 7 */
1067 		if (val & 0xffffffff00000000ULL)
1068 			return -1; /* #GP */
1069 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1070 		kvm_update_dr7(vcpu);
1071 		break;
1072 	}
1073 
1074 	return 0;
1075 }
1076 
1077 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1078 {
1079 	if (__kvm_set_dr(vcpu, dr, val)) {
1080 		kvm_inject_gp(vcpu, 0);
1081 		return 1;
1082 	}
1083 	return 0;
1084 }
1085 EXPORT_SYMBOL_GPL(kvm_set_dr);
1086 
1087 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1088 {
1089 	switch (dr) {
1090 	case 0 ... 3:
1091 		*val = vcpu->arch.db[dr];
1092 		break;
1093 	case 4:
1094 		/* fall through */
1095 	case 6:
1096 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1097 			*val = vcpu->arch.dr6;
1098 		else
1099 			*val = kvm_x86_ops->get_dr6(vcpu);
1100 		break;
1101 	case 5:
1102 		/* fall through */
1103 	default: /* 7 */
1104 		*val = vcpu->arch.dr7;
1105 		break;
1106 	}
1107 	return 0;
1108 }
1109 EXPORT_SYMBOL_GPL(kvm_get_dr);
1110 
1111 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1112 {
1113 	u32 ecx = kvm_rcx_read(vcpu);
1114 	u64 data;
1115 	int err;
1116 
1117 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1118 	if (err)
1119 		return err;
1120 	kvm_rax_write(vcpu, (u32)data);
1121 	kvm_rdx_write(vcpu, data >> 32);
1122 	return err;
1123 }
1124 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1125 
1126 /*
1127  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1128  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1129  *
1130  * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1131  * extract the supported MSRs from the related const lists.
1132  * msrs_to_save is selected from the msrs_to_save_all to reflect the
1133  * capabilities of the host cpu. This capabilities test skips MSRs that are
1134  * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1135  * may depend on host virtualization features rather than host cpu features.
1136  */
1137 
1138 static const u32 msrs_to_save_all[] = {
1139 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1140 	MSR_STAR,
1141 #ifdef CONFIG_X86_64
1142 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1143 #endif
1144 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1145 	MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1146 	MSR_IA32_SPEC_CTRL,
1147 	MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1148 	MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1149 	MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1150 	MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1151 	MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1152 	MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1153 	MSR_IA32_UMWAIT_CONTROL,
1154 
1155 	MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1156 	MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1157 	MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1158 	MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1159 	MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1160 	MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1161 	MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1162 	MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1163 	MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1164 	MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1165 	MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1166 	MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1167 	MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1168 	MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1169 	MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1170 	MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1171 	MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1172 	MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1173 	MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1174 	MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1175 	MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1176 	MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1177 };
1178 
1179 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1180 static unsigned num_msrs_to_save;
1181 
1182 static const u32 emulated_msrs_all[] = {
1183 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1184 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1185 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1186 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1187 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1188 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1189 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1190 	HV_X64_MSR_RESET,
1191 	HV_X64_MSR_VP_INDEX,
1192 	HV_X64_MSR_VP_RUNTIME,
1193 	HV_X64_MSR_SCONTROL,
1194 	HV_X64_MSR_STIMER0_CONFIG,
1195 	HV_X64_MSR_VP_ASSIST_PAGE,
1196 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1197 	HV_X64_MSR_TSC_EMULATION_STATUS,
1198 
1199 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1200 	MSR_KVM_PV_EOI_EN,
1201 
1202 	MSR_IA32_TSC_ADJUST,
1203 	MSR_IA32_TSCDEADLINE,
1204 	MSR_IA32_ARCH_CAPABILITIES,
1205 	MSR_IA32_MISC_ENABLE,
1206 	MSR_IA32_MCG_STATUS,
1207 	MSR_IA32_MCG_CTL,
1208 	MSR_IA32_MCG_EXT_CTL,
1209 	MSR_IA32_SMBASE,
1210 	MSR_SMI_COUNT,
1211 	MSR_PLATFORM_INFO,
1212 	MSR_MISC_FEATURES_ENABLES,
1213 	MSR_AMD64_VIRT_SPEC_CTRL,
1214 	MSR_IA32_POWER_CTL,
1215 
1216 	/*
1217 	 * The following list leaves out MSRs whose values are determined
1218 	 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1219 	 * We always support the "true" VMX control MSRs, even if the host
1220 	 * processor does not, so I am putting these registers here rather
1221 	 * than in msrs_to_save_all.
1222 	 */
1223 	MSR_IA32_VMX_BASIC,
1224 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1225 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1226 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1227 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1228 	MSR_IA32_VMX_MISC,
1229 	MSR_IA32_VMX_CR0_FIXED0,
1230 	MSR_IA32_VMX_CR4_FIXED0,
1231 	MSR_IA32_VMX_VMCS_ENUM,
1232 	MSR_IA32_VMX_PROCBASED_CTLS2,
1233 	MSR_IA32_VMX_EPT_VPID_CAP,
1234 	MSR_IA32_VMX_VMFUNC,
1235 
1236 	MSR_K7_HWCR,
1237 	MSR_KVM_POLL_CONTROL,
1238 };
1239 
1240 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1241 static unsigned num_emulated_msrs;
1242 
1243 /*
1244  * List of msr numbers which are used to expose MSR-based features that
1245  * can be used by a hypervisor to validate requested CPU features.
1246  */
1247 static const u32 msr_based_features_all[] = {
1248 	MSR_IA32_VMX_BASIC,
1249 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1250 	MSR_IA32_VMX_PINBASED_CTLS,
1251 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1252 	MSR_IA32_VMX_PROCBASED_CTLS,
1253 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1254 	MSR_IA32_VMX_EXIT_CTLS,
1255 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1256 	MSR_IA32_VMX_ENTRY_CTLS,
1257 	MSR_IA32_VMX_MISC,
1258 	MSR_IA32_VMX_CR0_FIXED0,
1259 	MSR_IA32_VMX_CR0_FIXED1,
1260 	MSR_IA32_VMX_CR4_FIXED0,
1261 	MSR_IA32_VMX_CR4_FIXED1,
1262 	MSR_IA32_VMX_VMCS_ENUM,
1263 	MSR_IA32_VMX_PROCBASED_CTLS2,
1264 	MSR_IA32_VMX_EPT_VPID_CAP,
1265 	MSR_IA32_VMX_VMFUNC,
1266 
1267 	MSR_F10H_DECFG,
1268 	MSR_IA32_UCODE_REV,
1269 	MSR_IA32_ARCH_CAPABILITIES,
1270 };
1271 
1272 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1273 static unsigned int num_msr_based_features;
1274 
1275 static u64 kvm_get_arch_capabilities(void)
1276 {
1277 	u64 data = 0;
1278 
1279 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1280 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1281 
1282 	/*
1283 	 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1284 	 * the nested hypervisor runs with NX huge pages.  If it is not,
1285 	 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1286 	 * L1 guests, so it need not worry about its own (L2) guests.
1287 	 */
1288 	data |= ARCH_CAP_PSCHANGE_MC_NO;
1289 
1290 	/*
1291 	 * If we're doing cache flushes (either "always" or "cond")
1292 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1293 	 * If an outer hypervisor is doing the cache flush for us
1294 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1295 	 * capability to the guest too, and if EPT is disabled we're not
1296 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1297 	 * require a nested hypervisor to do a flush of its own.
1298 	 */
1299 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1300 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1301 
1302 	if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1303 		data |= ARCH_CAP_RDCL_NO;
1304 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1305 		data |= ARCH_CAP_SSB_NO;
1306 	if (!boot_cpu_has_bug(X86_BUG_MDS))
1307 		data |= ARCH_CAP_MDS_NO;
1308 
1309 	/*
1310 	 * On TAA affected systems:
1311 	 *      - nothing to do if TSX is disabled on the host.
1312 	 *      - we emulate TSX_CTRL if present on the host.
1313 	 *	  This lets the guest use VERW to clear CPU buffers.
1314 	 */
1315 	if (!boot_cpu_has(X86_FEATURE_RTM))
1316 		data &= ~(ARCH_CAP_TAA_NO | ARCH_CAP_TSX_CTRL_MSR);
1317 	else if (!boot_cpu_has_bug(X86_BUG_TAA))
1318 		data |= ARCH_CAP_TAA_NO;
1319 
1320 	return data;
1321 }
1322 
1323 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1324 {
1325 	switch (msr->index) {
1326 	case MSR_IA32_ARCH_CAPABILITIES:
1327 		msr->data = kvm_get_arch_capabilities();
1328 		break;
1329 	case MSR_IA32_UCODE_REV:
1330 		rdmsrl_safe(msr->index, &msr->data);
1331 		break;
1332 	default:
1333 		if (kvm_x86_ops->get_msr_feature(msr))
1334 			return 1;
1335 	}
1336 	return 0;
1337 }
1338 
1339 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1340 {
1341 	struct kvm_msr_entry msr;
1342 	int r;
1343 
1344 	msr.index = index;
1345 	r = kvm_get_msr_feature(&msr);
1346 	if (r)
1347 		return r;
1348 
1349 	*data = msr.data;
1350 
1351 	return 0;
1352 }
1353 
1354 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1355 {
1356 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1357 		return false;
1358 
1359 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1360 		return false;
1361 
1362 	if (efer & (EFER_LME | EFER_LMA) &&
1363 	    !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1364 		return false;
1365 
1366 	if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1367 		return false;
1368 
1369 	return true;
1370 
1371 }
1372 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1373 {
1374 	if (efer & efer_reserved_bits)
1375 		return false;
1376 
1377 	return __kvm_valid_efer(vcpu, efer);
1378 }
1379 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1380 
1381 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1382 {
1383 	u64 old_efer = vcpu->arch.efer;
1384 	u64 efer = msr_info->data;
1385 
1386 	if (efer & efer_reserved_bits)
1387 		return 1;
1388 
1389 	if (!msr_info->host_initiated) {
1390 		if (!__kvm_valid_efer(vcpu, efer))
1391 			return 1;
1392 
1393 		if (is_paging(vcpu) &&
1394 		    (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1395 			return 1;
1396 	}
1397 
1398 	efer &= ~EFER_LMA;
1399 	efer |= vcpu->arch.efer & EFER_LMA;
1400 
1401 	kvm_x86_ops->set_efer(vcpu, efer);
1402 
1403 	/* Update reserved bits */
1404 	if ((efer ^ old_efer) & EFER_NX)
1405 		kvm_mmu_reset_context(vcpu);
1406 
1407 	return 0;
1408 }
1409 
1410 void kvm_enable_efer_bits(u64 mask)
1411 {
1412        efer_reserved_bits &= ~mask;
1413 }
1414 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1415 
1416 /*
1417  * Write @data into the MSR specified by @index.  Select MSR specific fault
1418  * checks are bypassed if @host_initiated is %true.
1419  * Returns 0 on success, non-0 otherwise.
1420  * Assumes vcpu_load() was already called.
1421  */
1422 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1423 			 bool host_initiated)
1424 {
1425 	struct msr_data msr;
1426 
1427 	switch (index) {
1428 	case MSR_FS_BASE:
1429 	case MSR_GS_BASE:
1430 	case MSR_KERNEL_GS_BASE:
1431 	case MSR_CSTAR:
1432 	case MSR_LSTAR:
1433 		if (is_noncanonical_address(data, vcpu))
1434 			return 1;
1435 		break;
1436 	case MSR_IA32_SYSENTER_EIP:
1437 	case MSR_IA32_SYSENTER_ESP:
1438 		/*
1439 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1440 		 * non-canonical address is written on Intel but not on
1441 		 * AMD (which ignores the top 32-bits, because it does
1442 		 * not implement 64-bit SYSENTER).
1443 		 *
1444 		 * 64-bit code should hence be able to write a non-canonical
1445 		 * value on AMD.  Making the address canonical ensures that
1446 		 * vmentry does not fail on Intel after writing a non-canonical
1447 		 * value, and that something deterministic happens if the guest
1448 		 * invokes 64-bit SYSENTER.
1449 		 */
1450 		data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1451 	}
1452 
1453 	msr.data = data;
1454 	msr.index = index;
1455 	msr.host_initiated = host_initiated;
1456 
1457 	return kvm_x86_ops->set_msr(vcpu, &msr);
1458 }
1459 
1460 /*
1461  * Read the MSR specified by @index into @data.  Select MSR specific fault
1462  * checks are bypassed if @host_initiated is %true.
1463  * Returns 0 on success, non-0 otherwise.
1464  * Assumes vcpu_load() was already called.
1465  */
1466 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1467 		  bool host_initiated)
1468 {
1469 	struct msr_data msr;
1470 	int ret;
1471 
1472 	msr.index = index;
1473 	msr.host_initiated = host_initiated;
1474 
1475 	ret = kvm_x86_ops->get_msr(vcpu, &msr);
1476 	if (!ret)
1477 		*data = msr.data;
1478 	return ret;
1479 }
1480 
1481 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1482 {
1483 	return __kvm_get_msr(vcpu, index, data, false);
1484 }
1485 EXPORT_SYMBOL_GPL(kvm_get_msr);
1486 
1487 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1488 {
1489 	return __kvm_set_msr(vcpu, index, data, false);
1490 }
1491 EXPORT_SYMBOL_GPL(kvm_set_msr);
1492 
1493 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1494 {
1495 	u32 ecx = kvm_rcx_read(vcpu);
1496 	u64 data;
1497 
1498 	if (kvm_get_msr(vcpu, ecx, &data)) {
1499 		trace_kvm_msr_read_ex(ecx);
1500 		kvm_inject_gp(vcpu, 0);
1501 		return 1;
1502 	}
1503 
1504 	trace_kvm_msr_read(ecx, data);
1505 
1506 	kvm_rax_write(vcpu, data & -1u);
1507 	kvm_rdx_write(vcpu, (data >> 32) & -1u);
1508 	return kvm_skip_emulated_instruction(vcpu);
1509 }
1510 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1511 
1512 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1513 {
1514 	u32 ecx = kvm_rcx_read(vcpu);
1515 	u64 data = kvm_read_edx_eax(vcpu);
1516 
1517 	if (kvm_set_msr(vcpu, ecx, data)) {
1518 		trace_kvm_msr_write_ex(ecx, data);
1519 		kvm_inject_gp(vcpu, 0);
1520 		return 1;
1521 	}
1522 
1523 	trace_kvm_msr_write(ecx, data);
1524 	return kvm_skip_emulated_instruction(vcpu);
1525 }
1526 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1527 
1528 /*
1529  * Adapt set_msr() to msr_io()'s calling convention
1530  */
1531 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1532 {
1533 	return __kvm_get_msr(vcpu, index, data, true);
1534 }
1535 
1536 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1537 {
1538 	return __kvm_set_msr(vcpu, index, *data, true);
1539 }
1540 
1541 #ifdef CONFIG_X86_64
1542 struct pvclock_clock {
1543 	int vclock_mode;
1544 	u64 cycle_last;
1545 	u64 mask;
1546 	u32 mult;
1547 	u32 shift;
1548 };
1549 
1550 struct pvclock_gtod_data {
1551 	seqcount_t	seq;
1552 
1553 	struct pvclock_clock clock; /* extract of a clocksource struct */
1554 	struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1555 
1556 	u64		boot_ns_raw;
1557 	u64		boot_ns;
1558 	u64		nsec_base;
1559 	u64		wall_time_sec;
1560 	u64		monotonic_raw_nsec;
1561 };
1562 
1563 static struct pvclock_gtod_data pvclock_gtod_data;
1564 
1565 static void update_pvclock_gtod(struct timekeeper *tk)
1566 {
1567 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1568 	u64 boot_ns, boot_ns_raw;
1569 
1570 	boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1571 	boot_ns_raw = ktime_to_ns(ktime_add(tk->tkr_raw.base, tk->offs_boot));
1572 
1573 	write_seqcount_begin(&vdata->seq);
1574 
1575 	/* copy pvclock gtod data */
1576 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->archdata.vclock_mode;
1577 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1578 	vdata->clock.mask		= tk->tkr_mono.mask;
1579 	vdata->clock.mult		= tk->tkr_mono.mult;
1580 	vdata->clock.shift		= tk->tkr_mono.shift;
1581 
1582 	vdata->raw_clock.vclock_mode	= tk->tkr_raw.clock->archdata.vclock_mode;
1583 	vdata->raw_clock.cycle_last	= tk->tkr_raw.cycle_last;
1584 	vdata->raw_clock.mask		= tk->tkr_raw.mask;
1585 	vdata->raw_clock.mult		= tk->tkr_raw.mult;
1586 	vdata->raw_clock.shift		= tk->tkr_raw.shift;
1587 
1588 	vdata->boot_ns			= boot_ns;
1589 	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
1590 
1591 	vdata->wall_time_sec            = tk->xtime_sec;
1592 
1593 	vdata->boot_ns_raw		= boot_ns_raw;
1594 	vdata->monotonic_raw_nsec	= tk->tkr_raw.xtime_nsec;
1595 
1596 	write_seqcount_end(&vdata->seq);
1597 }
1598 #endif
1599 
1600 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1601 {
1602 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1603 	kvm_vcpu_kick(vcpu);
1604 }
1605 
1606 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1607 {
1608 	int version;
1609 	int r;
1610 	struct pvclock_wall_clock wc;
1611 	struct timespec64 boot;
1612 
1613 	if (!wall_clock)
1614 		return;
1615 
1616 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1617 	if (r)
1618 		return;
1619 
1620 	if (version & 1)
1621 		++version;  /* first time write, random junk */
1622 
1623 	++version;
1624 
1625 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1626 		return;
1627 
1628 	/*
1629 	 * The guest calculates current wall clock time by adding
1630 	 * system time (updated by kvm_guest_time_update below) to the
1631 	 * wall clock specified here.  guest system time equals host
1632 	 * system time for us, thus we must fill in host boot time here.
1633 	 */
1634 	getboottime64(&boot);
1635 
1636 	if (kvm->arch.kvmclock_offset) {
1637 		struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1638 		boot = timespec64_sub(boot, ts);
1639 	}
1640 	wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1641 	wc.nsec = boot.tv_nsec;
1642 	wc.version = version;
1643 
1644 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1645 
1646 	version++;
1647 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1648 }
1649 
1650 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1651 {
1652 	do_shl32_div32(dividend, divisor);
1653 	return dividend;
1654 }
1655 
1656 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1657 			       s8 *pshift, u32 *pmultiplier)
1658 {
1659 	uint64_t scaled64;
1660 	int32_t  shift = 0;
1661 	uint64_t tps64;
1662 	uint32_t tps32;
1663 
1664 	tps64 = base_hz;
1665 	scaled64 = scaled_hz;
1666 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1667 		tps64 >>= 1;
1668 		shift--;
1669 	}
1670 
1671 	tps32 = (uint32_t)tps64;
1672 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1673 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1674 			scaled64 >>= 1;
1675 		else
1676 			tps32 <<= 1;
1677 		shift++;
1678 	}
1679 
1680 	*pshift = shift;
1681 	*pmultiplier = div_frac(scaled64, tps32);
1682 }
1683 
1684 #ifdef CONFIG_X86_64
1685 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1686 #endif
1687 
1688 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1689 static unsigned long max_tsc_khz;
1690 
1691 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1692 {
1693 	u64 v = (u64)khz * (1000000 + ppm);
1694 	do_div(v, 1000000);
1695 	return v;
1696 }
1697 
1698 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1699 {
1700 	u64 ratio;
1701 
1702 	/* Guest TSC same frequency as host TSC? */
1703 	if (!scale) {
1704 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1705 		return 0;
1706 	}
1707 
1708 	/* TSC scaling supported? */
1709 	if (!kvm_has_tsc_control) {
1710 		if (user_tsc_khz > tsc_khz) {
1711 			vcpu->arch.tsc_catchup = 1;
1712 			vcpu->arch.tsc_always_catchup = 1;
1713 			return 0;
1714 		} else {
1715 			pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
1716 			return -1;
1717 		}
1718 	}
1719 
1720 	/* TSC scaling required  - calculate ratio */
1721 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1722 				user_tsc_khz, tsc_khz);
1723 
1724 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1725 		pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1726 			            user_tsc_khz);
1727 		return -1;
1728 	}
1729 
1730 	vcpu->arch.tsc_scaling_ratio = ratio;
1731 	return 0;
1732 }
1733 
1734 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1735 {
1736 	u32 thresh_lo, thresh_hi;
1737 	int use_scaling = 0;
1738 
1739 	/* tsc_khz can be zero if TSC calibration fails */
1740 	if (user_tsc_khz == 0) {
1741 		/* set tsc_scaling_ratio to a safe value */
1742 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1743 		return -1;
1744 	}
1745 
1746 	/* Compute a scale to convert nanoseconds in TSC cycles */
1747 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1748 			   &vcpu->arch.virtual_tsc_shift,
1749 			   &vcpu->arch.virtual_tsc_mult);
1750 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1751 
1752 	/*
1753 	 * Compute the variation in TSC rate which is acceptable
1754 	 * within the range of tolerance and decide if the
1755 	 * rate being applied is within that bounds of the hardware
1756 	 * rate.  If so, no scaling or compensation need be done.
1757 	 */
1758 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1759 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1760 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1761 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1762 		use_scaling = 1;
1763 	}
1764 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1765 }
1766 
1767 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1768 {
1769 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1770 				      vcpu->arch.virtual_tsc_mult,
1771 				      vcpu->arch.virtual_tsc_shift);
1772 	tsc += vcpu->arch.this_tsc_write;
1773 	return tsc;
1774 }
1775 
1776 static inline int gtod_is_based_on_tsc(int mode)
1777 {
1778 	return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1779 }
1780 
1781 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1782 {
1783 #ifdef CONFIG_X86_64
1784 	bool vcpus_matched;
1785 	struct kvm_arch *ka = &vcpu->kvm->arch;
1786 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1787 
1788 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1789 			 atomic_read(&vcpu->kvm->online_vcpus));
1790 
1791 	/*
1792 	 * Once the masterclock is enabled, always perform request in
1793 	 * order to update it.
1794 	 *
1795 	 * In order to enable masterclock, the host clocksource must be TSC
1796 	 * and the vcpus need to have matched TSCs.  When that happens,
1797 	 * perform request to enable masterclock.
1798 	 */
1799 	if (ka->use_master_clock ||
1800 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1801 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1802 
1803 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1804 			    atomic_read(&vcpu->kvm->online_vcpus),
1805 		            ka->use_master_clock, gtod->clock.vclock_mode);
1806 #endif
1807 }
1808 
1809 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1810 {
1811 	u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1812 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1813 }
1814 
1815 /*
1816  * Multiply tsc by a fixed point number represented by ratio.
1817  *
1818  * The most significant 64-N bits (mult) of ratio represent the
1819  * integral part of the fixed point number; the remaining N bits
1820  * (frac) represent the fractional part, ie. ratio represents a fixed
1821  * point number (mult + frac * 2^(-N)).
1822  *
1823  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1824  */
1825 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1826 {
1827 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1828 }
1829 
1830 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1831 {
1832 	u64 _tsc = tsc;
1833 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
1834 
1835 	if (ratio != kvm_default_tsc_scaling_ratio)
1836 		_tsc = __scale_tsc(ratio, tsc);
1837 
1838 	return _tsc;
1839 }
1840 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1841 
1842 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1843 {
1844 	u64 tsc;
1845 
1846 	tsc = kvm_scale_tsc(vcpu, rdtsc());
1847 
1848 	return target_tsc - tsc;
1849 }
1850 
1851 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1852 {
1853 	u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1854 
1855 	return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1856 }
1857 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1858 
1859 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1860 {
1861 	vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1862 }
1863 
1864 static inline bool kvm_check_tsc_unstable(void)
1865 {
1866 #ifdef CONFIG_X86_64
1867 	/*
1868 	 * TSC is marked unstable when we're running on Hyper-V,
1869 	 * 'TSC page' clocksource is good.
1870 	 */
1871 	if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1872 		return false;
1873 #endif
1874 	return check_tsc_unstable();
1875 }
1876 
1877 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1878 {
1879 	struct kvm *kvm = vcpu->kvm;
1880 	u64 offset, ns, elapsed;
1881 	unsigned long flags;
1882 	bool matched;
1883 	bool already_matched;
1884 	u64 data = msr->data;
1885 	bool synchronizing = false;
1886 
1887 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1888 	offset = kvm_compute_tsc_offset(vcpu, data);
1889 	ns = ktime_get_boottime_ns();
1890 	elapsed = ns - kvm->arch.last_tsc_nsec;
1891 
1892 	if (vcpu->arch.virtual_tsc_khz) {
1893 		if (data == 0 && msr->host_initiated) {
1894 			/*
1895 			 * detection of vcpu initialization -- need to sync
1896 			 * with other vCPUs. This particularly helps to keep
1897 			 * kvm_clock stable after CPU hotplug
1898 			 */
1899 			synchronizing = true;
1900 		} else {
1901 			u64 tsc_exp = kvm->arch.last_tsc_write +
1902 						nsec_to_cycles(vcpu, elapsed);
1903 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1904 			/*
1905 			 * Special case: TSC write with a small delta (1 second)
1906 			 * of virtual cycle time against real time is
1907 			 * interpreted as an attempt to synchronize the CPU.
1908 			 */
1909 			synchronizing = data < tsc_exp + tsc_hz &&
1910 					data + tsc_hz > tsc_exp;
1911 		}
1912 	}
1913 
1914 	/*
1915 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1916 	 * TSC, we add elapsed time in this computation.  We could let the
1917 	 * compensation code attempt to catch up if we fall behind, but
1918 	 * it's better to try to match offsets from the beginning.
1919          */
1920 	if (synchronizing &&
1921 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1922 		if (!kvm_check_tsc_unstable()) {
1923 			offset = kvm->arch.cur_tsc_offset;
1924 		} else {
1925 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1926 			data += delta;
1927 			offset = kvm_compute_tsc_offset(vcpu, data);
1928 		}
1929 		matched = true;
1930 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1931 	} else {
1932 		/*
1933 		 * We split periods of matched TSC writes into generations.
1934 		 * For each generation, we track the original measured
1935 		 * nanosecond time, offset, and write, so if TSCs are in
1936 		 * sync, we can match exact offset, and if not, we can match
1937 		 * exact software computation in compute_guest_tsc()
1938 		 *
1939 		 * These values are tracked in kvm->arch.cur_xxx variables.
1940 		 */
1941 		kvm->arch.cur_tsc_generation++;
1942 		kvm->arch.cur_tsc_nsec = ns;
1943 		kvm->arch.cur_tsc_write = data;
1944 		kvm->arch.cur_tsc_offset = offset;
1945 		matched = false;
1946 	}
1947 
1948 	/*
1949 	 * We also track th most recent recorded KHZ, write and time to
1950 	 * allow the matching interval to be extended at each write.
1951 	 */
1952 	kvm->arch.last_tsc_nsec = ns;
1953 	kvm->arch.last_tsc_write = data;
1954 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1955 
1956 	vcpu->arch.last_guest_tsc = data;
1957 
1958 	/* Keep track of which generation this VCPU has synchronized to */
1959 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1960 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1961 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1962 
1963 	if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1964 		update_ia32_tsc_adjust_msr(vcpu, offset);
1965 
1966 	kvm_vcpu_write_tsc_offset(vcpu, offset);
1967 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1968 
1969 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1970 	if (!matched) {
1971 		kvm->arch.nr_vcpus_matched_tsc = 0;
1972 	} else if (!already_matched) {
1973 		kvm->arch.nr_vcpus_matched_tsc++;
1974 	}
1975 
1976 	kvm_track_tsc_matching(vcpu);
1977 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1978 }
1979 
1980 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1981 
1982 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1983 					   s64 adjustment)
1984 {
1985 	u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1986 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
1987 }
1988 
1989 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1990 {
1991 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1992 		WARN_ON(adjustment < 0);
1993 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1994 	adjust_tsc_offset_guest(vcpu, adjustment);
1995 }
1996 
1997 #ifdef CONFIG_X86_64
1998 
1999 static u64 read_tsc(void)
2000 {
2001 	u64 ret = (u64)rdtsc_ordered();
2002 	u64 last = pvclock_gtod_data.clock.cycle_last;
2003 
2004 	if (likely(ret >= last))
2005 		return ret;
2006 
2007 	/*
2008 	 * GCC likes to generate cmov here, but this branch is extremely
2009 	 * predictable (it's just a function of time and the likely is
2010 	 * very likely) and there's a data dependence, so force GCC
2011 	 * to generate a branch instead.  I don't barrier() because
2012 	 * we don't actually need a barrier, and if this function
2013 	 * ever gets inlined it will generate worse code.
2014 	 */
2015 	asm volatile ("");
2016 	return last;
2017 }
2018 
2019 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2020 			  int *mode)
2021 {
2022 	long v;
2023 	u64 tsc_pg_val;
2024 
2025 	switch (clock->vclock_mode) {
2026 	case VCLOCK_HVCLOCK:
2027 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2028 						  tsc_timestamp);
2029 		if (tsc_pg_val != U64_MAX) {
2030 			/* TSC page valid */
2031 			*mode = VCLOCK_HVCLOCK;
2032 			v = (tsc_pg_val - clock->cycle_last) &
2033 				clock->mask;
2034 		} else {
2035 			/* TSC page invalid */
2036 			*mode = VCLOCK_NONE;
2037 		}
2038 		break;
2039 	case VCLOCK_TSC:
2040 		*mode = VCLOCK_TSC;
2041 		*tsc_timestamp = read_tsc();
2042 		v = (*tsc_timestamp - clock->cycle_last) &
2043 			clock->mask;
2044 		break;
2045 	default:
2046 		*mode = VCLOCK_NONE;
2047 	}
2048 
2049 	if (*mode == VCLOCK_NONE)
2050 		*tsc_timestamp = v = 0;
2051 
2052 	return v * clock->mult;
2053 }
2054 
2055 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2056 {
2057 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2058 	unsigned long seq;
2059 	int mode;
2060 	u64 ns;
2061 
2062 	do {
2063 		seq = read_seqcount_begin(&gtod->seq);
2064 		ns = gtod->monotonic_raw_nsec;
2065 		ns += vgettsc(&gtod->raw_clock, tsc_timestamp, &mode);
2066 		ns >>= gtod->clock.shift;
2067 		ns += gtod->boot_ns_raw;
2068 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2069 	*t = ns;
2070 
2071 	return mode;
2072 }
2073 
2074 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2075 {
2076 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2077 	unsigned long seq;
2078 	int mode;
2079 	u64 ns;
2080 
2081 	do {
2082 		seq = read_seqcount_begin(&gtod->seq);
2083 		ts->tv_sec = gtod->wall_time_sec;
2084 		ns = gtod->nsec_base;
2085 		ns += vgettsc(&gtod->clock, tsc_timestamp, &mode);
2086 		ns >>= gtod->clock.shift;
2087 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2088 
2089 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2090 	ts->tv_nsec = ns;
2091 
2092 	return mode;
2093 }
2094 
2095 /* returns true if host is using TSC based clocksource */
2096 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2097 {
2098 	/* checked again under seqlock below */
2099 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2100 		return false;
2101 
2102 	return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2103 						      tsc_timestamp));
2104 }
2105 
2106 /* returns true if host is using TSC based clocksource */
2107 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2108 					   u64 *tsc_timestamp)
2109 {
2110 	/* checked again under seqlock below */
2111 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2112 		return false;
2113 
2114 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2115 }
2116 #endif
2117 
2118 /*
2119  *
2120  * Assuming a stable TSC across physical CPUS, and a stable TSC
2121  * across virtual CPUs, the following condition is possible.
2122  * Each numbered line represents an event visible to both
2123  * CPUs at the next numbered event.
2124  *
2125  * "timespecX" represents host monotonic time. "tscX" represents
2126  * RDTSC value.
2127  *
2128  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
2129  *
2130  * 1.  read timespec0,tsc0
2131  * 2.					| timespec1 = timespec0 + N
2132  * 					| tsc1 = tsc0 + M
2133  * 3. transition to guest		| transition to guest
2134  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2135  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
2136  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2137  *
2138  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2139  *
2140  * 	- ret0 < ret1
2141  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2142  *		...
2143  *	- 0 < N - M => M < N
2144  *
2145  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2146  * always the case (the difference between two distinct xtime instances
2147  * might be smaller then the difference between corresponding TSC reads,
2148  * when updating guest vcpus pvclock areas).
2149  *
2150  * To avoid that problem, do not allow visibility of distinct
2151  * system_timestamp/tsc_timestamp values simultaneously: use a master
2152  * copy of host monotonic time values. Update that master copy
2153  * in lockstep.
2154  *
2155  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2156  *
2157  */
2158 
2159 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2160 {
2161 #ifdef CONFIG_X86_64
2162 	struct kvm_arch *ka = &kvm->arch;
2163 	int vclock_mode;
2164 	bool host_tsc_clocksource, vcpus_matched;
2165 
2166 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2167 			atomic_read(&kvm->online_vcpus));
2168 
2169 	/*
2170 	 * If the host uses TSC clock, then passthrough TSC as stable
2171 	 * to the guest.
2172 	 */
2173 	host_tsc_clocksource = kvm_get_time_and_clockread(
2174 					&ka->master_kernel_ns,
2175 					&ka->master_cycle_now);
2176 
2177 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2178 				&& !ka->backwards_tsc_observed
2179 				&& !ka->boot_vcpu_runs_old_kvmclock;
2180 
2181 	if (ka->use_master_clock)
2182 		atomic_set(&kvm_guest_has_master_clock, 1);
2183 
2184 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2185 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2186 					vcpus_matched);
2187 #endif
2188 }
2189 
2190 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2191 {
2192 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2193 }
2194 
2195 static void kvm_gen_update_masterclock(struct kvm *kvm)
2196 {
2197 #ifdef CONFIG_X86_64
2198 	int i;
2199 	struct kvm_vcpu *vcpu;
2200 	struct kvm_arch *ka = &kvm->arch;
2201 
2202 	spin_lock(&ka->pvclock_gtod_sync_lock);
2203 	kvm_make_mclock_inprogress_request(kvm);
2204 	/* no guest entries from this point */
2205 	pvclock_update_vm_gtod_copy(kvm);
2206 
2207 	kvm_for_each_vcpu(i, vcpu, kvm)
2208 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2209 
2210 	/* guest entries allowed */
2211 	kvm_for_each_vcpu(i, vcpu, kvm)
2212 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2213 
2214 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2215 #endif
2216 }
2217 
2218 u64 get_kvmclock_ns(struct kvm *kvm)
2219 {
2220 	struct kvm_arch *ka = &kvm->arch;
2221 	struct pvclock_vcpu_time_info hv_clock;
2222 	u64 ret;
2223 
2224 	spin_lock(&ka->pvclock_gtod_sync_lock);
2225 	if (!ka->use_master_clock) {
2226 		spin_unlock(&ka->pvclock_gtod_sync_lock);
2227 		return ktime_get_boottime_ns() + ka->kvmclock_offset;
2228 	}
2229 
2230 	hv_clock.tsc_timestamp = ka->master_cycle_now;
2231 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2232 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2233 
2234 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2235 	get_cpu();
2236 
2237 	if (__this_cpu_read(cpu_tsc_khz)) {
2238 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2239 				   &hv_clock.tsc_shift,
2240 				   &hv_clock.tsc_to_system_mul);
2241 		ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2242 	} else
2243 		ret = ktime_get_boottime_ns() + ka->kvmclock_offset;
2244 
2245 	put_cpu();
2246 
2247 	return ret;
2248 }
2249 
2250 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2251 {
2252 	struct kvm_vcpu_arch *vcpu = &v->arch;
2253 	struct pvclock_vcpu_time_info guest_hv_clock;
2254 
2255 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2256 		&guest_hv_clock, sizeof(guest_hv_clock))))
2257 		return;
2258 
2259 	/* This VCPU is paused, but it's legal for a guest to read another
2260 	 * VCPU's kvmclock, so we really have to follow the specification where
2261 	 * it says that version is odd if data is being modified, and even after
2262 	 * it is consistent.
2263 	 *
2264 	 * Version field updates must be kept separate.  This is because
2265 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2266 	 * writes within a string instruction are weakly ordered.  So there
2267 	 * are three writes overall.
2268 	 *
2269 	 * As a small optimization, only write the version field in the first
2270 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2271 	 * version field is the first in the struct.
2272 	 */
2273 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2274 
2275 	if (guest_hv_clock.version & 1)
2276 		++guest_hv_clock.version;  /* first time write, random junk */
2277 
2278 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2279 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2280 				&vcpu->hv_clock,
2281 				sizeof(vcpu->hv_clock.version));
2282 
2283 	smp_wmb();
2284 
2285 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2286 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2287 
2288 	if (vcpu->pvclock_set_guest_stopped_request) {
2289 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2290 		vcpu->pvclock_set_guest_stopped_request = false;
2291 	}
2292 
2293 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2294 
2295 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2296 				&vcpu->hv_clock,
2297 				sizeof(vcpu->hv_clock));
2298 
2299 	smp_wmb();
2300 
2301 	vcpu->hv_clock.version++;
2302 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2303 				&vcpu->hv_clock,
2304 				sizeof(vcpu->hv_clock.version));
2305 }
2306 
2307 static int kvm_guest_time_update(struct kvm_vcpu *v)
2308 {
2309 	unsigned long flags, tgt_tsc_khz;
2310 	struct kvm_vcpu_arch *vcpu = &v->arch;
2311 	struct kvm_arch *ka = &v->kvm->arch;
2312 	s64 kernel_ns;
2313 	u64 tsc_timestamp, host_tsc;
2314 	u8 pvclock_flags;
2315 	bool use_master_clock;
2316 
2317 	kernel_ns = 0;
2318 	host_tsc = 0;
2319 
2320 	/*
2321 	 * If the host uses TSC clock, then passthrough TSC as stable
2322 	 * to the guest.
2323 	 */
2324 	spin_lock(&ka->pvclock_gtod_sync_lock);
2325 	use_master_clock = ka->use_master_clock;
2326 	if (use_master_clock) {
2327 		host_tsc = ka->master_cycle_now;
2328 		kernel_ns = ka->master_kernel_ns;
2329 	}
2330 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2331 
2332 	/* Keep irq disabled to prevent changes to the clock */
2333 	local_irq_save(flags);
2334 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2335 	if (unlikely(tgt_tsc_khz == 0)) {
2336 		local_irq_restore(flags);
2337 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2338 		return 1;
2339 	}
2340 	if (!use_master_clock) {
2341 		host_tsc = rdtsc();
2342 		kernel_ns = ktime_get_boottime_ns();
2343 	}
2344 
2345 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2346 
2347 	/*
2348 	 * We may have to catch up the TSC to match elapsed wall clock
2349 	 * time for two reasons, even if kvmclock is used.
2350 	 *   1) CPU could have been running below the maximum TSC rate
2351 	 *   2) Broken TSC compensation resets the base at each VCPU
2352 	 *      entry to avoid unknown leaps of TSC even when running
2353 	 *      again on the same CPU.  This may cause apparent elapsed
2354 	 *      time to disappear, and the guest to stand still or run
2355 	 *	very slowly.
2356 	 */
2357 	if (vcpu->tsc_catchup) {
2358 		u64 tsc = compute_guest_tsc(v, kernel_ns);
2359 		if (tsc > tsc_timestamp) {
2360 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2361 			tsc_timestamp = tsc;
2362 		}
2363 	}
2364 
2365 	local_irq_restore(flags);
2366 
2367 	/* With all the info we got, fill in the values */
2368 
2369 	if (kvm_has_tsc_control)
2370 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2371 
2372 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2373 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2374 				   &vcpu->hv_clock.tsc_shift,
2375 				   &vcpu->hv_clock.tsc_to_system_mul);
2376 		vcpu->hw_tsc_khz = tgt_tsc_khz;
2377 	}
2378 
2379 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2380 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2381 	vcpu->last_guest_tsc = tsc_timestamp;
2382 
2383 	/* If the host uses TSC clocksource, then it is stable */
2384 	pvclock_flags = 0;
2385 	if (use_master_clock)
2386 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2387 
2388 	vcpu->hv_clock.flags = pvclock_flags;
2389 
2390 	if (vcpu->pv_time_enabled)
2391 		kvm_setup_pvclock_page(v);
2392 	if (v == kvm_get_vcpu(v->kvm, 0))
2393 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2394 	return 0;
2395 }
2396 
2397 /*
2398  * kvmclock updates which are isolated to a given vcpu, such as
2399  * vcpu->cpu migration, should not allow system_timestamp from
2400  * the rest of the vcpus to remain static. Otherwise ntp frequency
2401  * correction applies to one vcpu's system_timestamp but not
2402  * the others.
2403  *
2404  * So in those cases, request a kvmclock update for all vcpus.
2405  * We need to rate-limit these requests though, as they can
2406  * considerably slow guests that have a large number of vcpus.
2407  * The time for a remote vcpu to update its kvmclock is bound
2408  * by the delay we use to rate-limit the updates.
2409  */
2410 
2411 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2412 
2413 static void kvmclock_update_fn(struct work_struct *work)
2414 {
2415 	int i;
2416 	struct delayed_work *dwork = to_delayed_work(work);
2417 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2418 					   kvmclock_update_work);
2419 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2420 	struct kvm_vcpu *vcpu;
2421 
2422 	kvm_for_each_vcpu(i, vcpu, kvm) {
2423 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2424 		kvm_vcpu_kick(vcpu);
2425 	}
2426 }
2427 
2428 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2429 {
2430 	struct kvm *kvm = v->kvm;
2431 
2432 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2433 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2434 					KVMCLOCK_UPDATE_DELAY);
2435 }
2436 
2437 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2438 
2439 static void kvmclock_sync_fn(struct work_struct *work)
2440 {
2441 	struct delayed_work *dwork = to_delayed_work(work);
2442 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2443 					   kvmclock_sync_work);
2444 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2445 
2446 	if (!kvmclock_periodic_sync)
2447 		return;
2448 
2449 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2450 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2451 					KVMCLOCK_SYNC_PERIOD);
2452 }
2453 
2454 /*
2455  * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2456  */
2457 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2458 {
2459 	/* McStatusWrEn enabled? */
2460 	if (guest_cpuid_is_amd(vcpu))
2461 		return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2462 
2463 	return false;
2464 }
2465 
2466 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2467 {
2468 	u64 mcg_cap = vcpu->arch.mcg_cap;
2469 	unsigned bank_num = mcg_cap & 0xff;
2470 	u32 msr = msr_info->index;
2471 	u64 data = msr_info->data;
2472 
2473 	switch (msr) {
2474 	case MSR_IA32_MCG_STATUS:
2475 		vcpu->arch.mcg_status = data;
2476 		break;
2477 	case MSR_IA32_MCG_CTL:
2478 		if (!(mcg_cap & MCG_CTL_P) &&
2479 		    (data || !msr_info->host_initiated))
2480 			return 1;
2481 		if (data != 0 && data != ~(u64)0)
2482 			return 1;
2483 		vcpu->arch.mcg_ctl = data;
2484 		break;
2485 	default:
2486 		if (msr >= MSR_IA32_MC0_CTL &&
2487 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2488 			u32 offset = msr - MSR_IA32_MC0_CTL;
2489 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2490 			 * some Linux kernels though clear bit 10 in bank 4 to
2491 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2492 			 * this to avoid an uncatched #GP in the guest
2493 			 */
2494 			if ((offset & 0x3) == 0 &&
2495 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2496 				return -1;
2497 
2498 			/* MCi_STATUS */
2499 			if (!msr_info->host_initiated &&
2500 			    (offset & 0x3) == 1 && data != 0) {
2501 				if (!can_set_mci_status(vcpu))
2502 					return -1;
2503 			}
2504 
2505 			vcpu->arch.mce_banks[offset] = data;
2506 			break;
2507 		}
2508 		return 1;
2509 	}
2510 	return 0;
2511 }
2512 
2513 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2514 {
2515 	struct kvm *kvm = vcpu->kvm;
2516 	int lm = is_long_mode(vcpu);
2517 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2518 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2519 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2520 		: kvm->arch.xen_hvm_config.blob_size_32;
2521 	u32 page_num = data & ~PAGE_MASK;
2522 	u64 page_addr = data & PAGE_MASK;
2523 	u8 *page;
2524 	int r;
2525 
2526 	r = -E2BIG;
2527 	if (page_num >= blob_size)
2528 		goto out;
2529 	r = -ENOMEM;
2530 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2531 	if (IS_ERR(page)) {
2532 		r = PTR_ERR(page);
2533 		goto out;
2534 	}
2535 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2536 		goto out_free;
2537 	r = 0;
2538 out_free:
2539 	kfree(page);
2540 out:
2541 	return r;
2542 }
2543 
2544 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2545 {
2546 	gpa_t gpa = data & ~0x3f;
2547 
2548 	/* Bits 3:5 are reserved, Should be zero */
2549 	if (data & 0x38)
2550 		return 1;
2551 
2552 	vcpu->arch.apf.msr_val = data;
2553 
2554 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
2555 		kvm_clear_async_pf_completion_queue(vcpu);
2556 		kvm_async_pf_hash_reset(vcpu);
2557 		return 0;
2558 	}
2559 
2560 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2561 					sizeof(u32)))
2562 		return 1;
2563 
2564 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2565 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2566 	kvm_async_pf_wakeup_all(vcpu);
2567 	return 0;
2568 }
2569 
2570 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2571 {
2572 	vcpu->arch.pv_time_enabled = false;
2573 	vcpu->arch.time = 0;
2574 }
2575 
2576 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2577 {
2578 	++vcpu->stat.tlb_flush;
2579 	kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2580 }
2581 
2582 static void record_steal_time(struct kvm_vcpu *vcpu)
2583 {
2584 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2585 		return;
2586 
2587 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2588 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2589 		return;
2590 
2591 	/*
2592 	 * Doing a TLB flush here, on the guest's behalf, can avoid
2593 	 * expensive IPIs.
2594 	 */
2595 	trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2596 		vcpu->arch.st.steal.preempted & KVM_VCPU_FLUSH_TLB);
2597 	if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2598 		kvm_vcpu_flush_tlb(vcpu, false);
2599 
2600 	if (vcpu->arch.st.steal.version & 1)
2601 		vcpu->arch.st.steal.version += 1;  /* first time write, random junk */
2602 
2603 	vcpu->arch.st.steal.version += 1;
2604 
2605 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2606 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2607 
2608 	smp_wmb();
2609 
2610 	vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2611 		vcpu->arch.st.last_steal;
2612 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2613 
2614 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2615 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2616 
2617 	smp_wmb();
2618 
2619 	vcpu->arch.st.steal.version += 1;
2620 
2621 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2622 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2623 }
2624 
2625 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2626 {
2627 	bool pr = false;
2628 	u32 msr = msr_info->index;
2629 	u64 data = msr_info->data;
2630 
2631 	switch (msr) {
2632 	case MSR_AMD64_NB_CFG:
2633 	case MSR_IA32_UCODE_WRITE:
2634 	case MSR_VM_HSAVE_PA:
2635 	case MSR_AMD64_PATCH_LOADER:
2636 	case MSR_AMD64_BU_CFG2:
2637 	case MSR_AMD64_DC_CFG:
2638 	case MSR_F15H_EX_CFG:
2639 		break;
2640 
2641 	case MSR_IA32_UCODE_REV:
2642 		if (msr_info->host_initiated)
2643 			vcpu->arch.microcode_version = data;
2644 		break;
2645 	case MSR_IA32_ARCH_CAPABILITIES:
2646 		if (!msr_info->host_initiated)
2647 			return 1;
2648 		vcpu->arch.arch_capabilities = data;
2649 		break;
2650 	case MSR_EFER:
2651 		return set_efer(vcpu, msr_info);
2652 	case MSR_K7_HWCR:
2653 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2654 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2655 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2656 
2657 		/* Handle McStatusWrEn */
2658 		if (data == BIT_ULL(18)) {
2659 			vcpu->arch.msr_hwcr = data;
2660 		} else if (data != 0) {
2661 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2662 				    data);
2663 			return 1;
2664 		}
2665 		break;
2666 	case MSR_FAM10H_MMIO_CONF_BASE:
2667 		if (data != 0) {
2668 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2669 				    "0x%llx\n", data);
2670 			return 1;
2671 		}
2672 		break;
2673 	case MSR_IA32_DEBUGCTLMSR:
2674 		if (!data) {
2675 			/* We support the non-activated case already */
2676 			break;
2677 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2678 			/* Values other than LBR and BTF are vendor-specific,
2679 			   thus reserved and should throw a #GP */
2680 			return 1;
2681 		}
2682 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2683 			    __func__, data);
2684 		break;
2685 	case 0x200 ... 0x2ff:
2686 		return kvm_mtrr_set_msr(vcpu, msr, data);
2687 	case MSR_IA32_APICBASE:
2688 		return kvm_set_apic_base(vcpu, msr_info);
2689 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2690 		return kvm_x2apic_msr_write(vcpu, msr, data);
2691 	case MSR_IA32_TSCDEADLINE:
2692 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2693 		break;
2694 	case MSR_IA32_TSC_ADJUST:
2695 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2696 			if (!msr_info->host_initiated) {
2697 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2698 				adjust_tsc_offset_guest(vcpu, adj);
2699 			}
2700 			vcpu->arch.ia32_tsc_adjust_msr = data;
2701 		}
2702 		break;
2703 	case MSR_IA32_MISC_ENABLE:
2704 		if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
2705 		    ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
2706 			if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
2707 				return 1;
2708 			vcpu->arch.ia32_misc_enable_msr = data;
2709 			kvm_update_cpuid(vcpu);
2710 		} else {
2711 			vcpu->arch.ia32_misc_enable_msr = data;
2712 		}
2713 		break;
2714 	case MSR_IA32_SMBASE:
2715 		if (!msr_info->host_initiated)
2716 			return 1;
2717 		vcpu->arch.smbase = data;
2718 		break;
2719 	case MSR_IA32_POWER_CTL:
2720 		vcpu->arch.msr_ia32_power_ctl = data;
2721 		break;
2722 	case MSR_IA32_TSC:
2723 		kvm_write_tsc(vcpu, msr_info);
2724 		break;
2725 	case MSR_IA32_XSS:
2726 		if (!msr_info->host_initiated &&
2727 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
2728 			return 1;
2729 		/*
2730 		 * We do support PT if kvm_x86_ops->pt_supported(), but we do
2731 		 * not support IA32_XSS[bit 8]. Guests will have to use
2732 		 * RDMSR/WRMSR rather than XSAVES/XRSTORS to save/restore PT
2733 		 * MSRs.
2734 		 */
2735 		if (data != 0)
2736 			return 1;
2737 		vcpu->arch.ia32_xss = data;
2738 		break;
2739 	case MSR_SMI_COUNT:
2740 		if (!msr_info->host_initiated)
2741 			return 1;
2742 		vcpu->arch.smi_count = data;
2743 		break;
2744 	case MSR_KVM_WALL_CLOCK_NEW:
2745 	case MSR_KVM_WALL_CLOCK:
2746 		vcpu->kvm->arch.wall_clock = data;
2747 		kvm_write_wall_clock(vcpu->kvm, data);
2748 		break;
2749 	case MSR_KVM_SYSTEM_TIME_NEW:
2750 	case MSR_KVM_SYSTEM_TIME: {
2751 		struct kvm_arch *ka = &vcpu->kvm->arch;
2752 
2753 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2754 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2755 
2756 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2757 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2758 
2759 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2760 		}
2761 
2762 		vcpu->arch.time = data;
2763 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2764 
2765 		/* we verify if the enable bit is set... */
2766 		vcpu->arch.pv_time_enabled = false;
2767 		if (!(data & 1))
2768 			break;
2769 
2770 		if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2771 		     &vcpu->arch.pv_time, data & ~1ULL,
2772 		     sizeof(struct pvclock_vcpu_time_info)))
2773 			vcpu->arch.pv_time_enabled = true;
2774 
2775 		break;
2776 	}
2777 	case MSR_KVM_ASYNC_PF_EN:
2778 		if (kvm_pv_enable_async_pf(vcpu, data))
2779 			return 1;
2780 		break;
2781 	case MSR_KVM_STEAL_TIME:
2782 
2783 		if (unlikely(!sched_info_on()))
2784 			return 1;
2785 
2786 		if (data & KVM_STEAL_RESERVED_MASK)
2787 			return 1;
2788 
2789 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2790 						data & KVM_STEAL_VALID_BITS,
2791 						sizeof(struct kvm_steal_time)))
2792 			return 1;
2793 
2794 		vcpu->arch.st.msr_val = data;
2795 
2796 		if (!(data & KVM_MSR_ENABLED))
2797 			break;
2798 
2799 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2800 
2801 		break;
2802 	case MSR_KVM_PV_EOI_EN:
2803 		if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2804 			return 1;
2805 		break;
2806 
2807 	case MSR_KVM_POLL_CONTROL:
2808 		/* only enable bit supported */
2809 		if (data & (-1ULL << 1))
2810 			return 1;
2811 
2812 		vcpu->arch.msr_kvm_poll_control = data;
2813 		break;
2814 
2815 	case MSR_IA32_MCG_CTL:
2816 	case MSR_IA32_MCG_STATUS:
2817 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2818 		return set_msr_mce(vcpu, msr_info);
2819 
2820 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2821 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2822 		pr = true; /* fall through */
2823 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2824 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2825 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2826 			return kvm_pmu_set_msr(vcpu, msr_info);
2827 
2828 		if (pr || data != 0)
2829 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2830 				    "0x%x data 0x%llx\n", msr, data);
2831 		break;
2832 	case MSR_K7_CLK_CTL:
2833 		/*
2834 		 * Ignore all writes to this no longer documented MSR.
2835 		 * Writes are only relevant for old K7 processors,
2836 		 * all pre-dating SVM, but a recommended workaround from
2837 		 * AMD for these chips. It is possible to specify the
2838 		 * affected processor models on the command line, hence
2839 		 * the need to ignore the workaround.
2840 		 */
2841 		break;
2842 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2843 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2844 	case HV_X64_MSR_CRASH_CTL:
2845 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2846 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2847 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
2848 	case HV_X64_MSR_TSC_EMULATION_STATUS:
2849 		return kvm_hv_set_msr_common(vcpu, msr, data,
2850 					     msr_info->host_initiated);
2851 	case MSR_IA32_BBL_CR_CTL3:
2852 		/* Drop writes to this legacy MSR -- see rdmsr
2853 		 * counterpart for further detail.
2854 		 */
2855 		if (report_ignored_msrs)
2856 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2857 				msr, data);
2858 		break;
2859 	case MSR_AMD64_OSVW_ID_LENGTH:
2860 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2861 			return 1;
2862 		vcpu->arch.osvw.length = data;
2863 		break;
2864 	case MSR_AMD64_OSVW_STATUS:
2865 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2866 			return 1;
2867 		vcpu->arch.osvw.status = data;
2868 		break;
2869 	case MSR_PLATFORM_INFO:
2870 		if (!msr_info->host_initiated ||
2871 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2872 		     cpuid_fault_enabled(vcpu)))
2873 			return 1;
2874 		vcpu->arch.msr_platform_info = data;
2875 		break;
2876 	case MSR_MISC_FEATURES_ENABLES:
2877 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2878 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2879 		     !supports_cpuid_fault(vcpu)))
2880 			return 1;
2881 		vcpu->arch.msr_misc_features_enables = data;
2882 		break;
2883 	default:
2884 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2885 			return xen_hvm_config(vcpu, data);
2886 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2887 			return kvm_pmu_set_msr(vcpu, msr_info);
2888 		if (!ignore_msrs) {
2889 			vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2890 				    msr, data);
2891 			return 1;
2892 		} else {
2893 			if (report_ignored_msrs)
2894 				vcpu_unimpl(vcpu,
2895 					"ignored wrmsr: 0x%x data 0x%llx\n",
2896 					msr, data);
2897 			break;
2898 		}
2899 	}
2900 	return 0;
2901 }
2902 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2903 
2904 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2905 {
2906 	u64 data;
2907 	u64 mcg_cap = vcpu->arch.mcg_cap;
2908 	unsigned bank_num = mcg_cap & 0xff;
2909 
2910 	switch (msr) {
2911 	case MSR_IA32_P5_MC_ADDR:
2912 	case MSR_IA32_P5_MC_TYPE:
2913 		data = 0;
2914 		break;
2915 	case MSR_IA32_MCG_CAP:
2916 		data = vcpu->arch.mcg_cap;
2917 		break;
2918 	case MSR_IA32_MCG_CTL:
2919 		if (!(mcg_cap & MCG_CTL_P) && !host)
2920 			return 1;
2921 		data = vcpu->arch.mcg_ctl;
2922 		break;
2923 	case MSR_IA32_MCG_STATUS:
2924 		data = vcpu->arch.mcg_status;
2925 		break;
2926 	default:
2927 		if (msr >= MSR_IA32_MC0_CTL &&
2928 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2929 			u32 offset = msr - MSR_IA32_MC0_CTL;
2930 			data = vcpu->arch.mce_banks[offset];
2931 			break;
2932 		}
2933 		return 1;
2934 	}
2935 	*pdata = data;
2936 	return 0;
2937 }
2938 
2939 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2940 {
2941 	switch (msr_info->index) {
2942 	case MSR_IA32_PLATFORM_ID:
2943 	case MSR_IA32_EBL_CR_POWERON:
2944 	case MSR_IA32_DEBUGCTLMSR:
2945 	case MSR_IA32_LASTBRANCHFROMIP:
2946 	case MSR_IA32_LASTBRANCHTOIP:
2947 	case MSR_IA32_LASTINTFROMIP:
2948 	case MSR_IA32_LASTINTTOIP:
2949 	case MSR_K8_SYSCFG:
2950 	case MSR_K8_TSEG_ADDR:
2951 	case MSR_K8_TSEG_MASK:
2952 	case MSR_VM_HSAVE_PA:
2953 	case MSR_K8_INT_PENDING_MSG:
2954 	case MSR_AMD64_NB_CFG:
2955 	case MSR_FAM10H_MMIO_CONF_BASE:
2956 	case MSR_AMD64_BU_CFG2:
2957 	case MSR_IA32_PERF_CTL:
2958 	case MSR_AMD64_DC_CFG:
2959 	case MSR_F15H_EX_CFG:
2960 		msr_info->data = 0;
2961 		break;
2962 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2963 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2964 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2965 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2966 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2967 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2968 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2969 		msr_info->data = 0;
2970 		break;
2971 	case MSR_IA32_UCODE_REV:
2972 		msr_info->data = vcpu->arch.microcode_version;
2973 		break;
2974 	case MSR_IA32_ARCH_CAPABILITIES:
2975 		if (!msr_info->host_initiated &&
2976 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
2977 			return 1;
2978 		msr_info->data = vcpu->arch.arch_capabilities;
2979 		break;
2980 	case MSR_IA32_POWER_CTL:
2981 		msr_info->data = vcpu->arch.msr_ia32_power_ctl;
2982 		break;
2983 	case MSR_IA32_TSC:
2984 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2985 		break;
2986 	case MSR_MTRRcap:
2987 	case 0x200 ... 0x2ff:
2988 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2989 	case 0xcd: /* fsb frequency */
2990 		msr_info->data = 3;
2991 		break;
2992 		/*
2993 		 * MSR_EBC_FREQUENCY_ID
2994 		 * Conservative value valid for even the basic CPU models.
2995 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2996 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2997 		 * and 266MHz for model 3, or 4. Set Core Clock
2998 		 * Frequency to System Bus Frequency Ratio to 1 (bits
2999 		 * 31:24) even though these are only valid for CPU
3000 		 * models > 2, however guests may end up dividing or
3001 		 * multiplying by zero otherwise.
3002 		 */
3003 	case MSR_EBC_FREQUENCY_ID:
3004 		msr_info->data = 1 << 24;
3005 		break;
3006 	case MSR_IA32_APICBASE:
3007 		msr_info->data = kvm_get_apic_base(vcpu);
3008 		break;
3009 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
3010 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3011 		break;
3012 	case MSR_IA32_TSCDEADLINE:
3013 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3014 		break;
3015 	case MSR_IA32_TSC_ADJUST:
3016 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3017 		break;
3018 	case MSR_IA32_MISC_ENABLE:
3019 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3020 		break;
3021 	case MSR_IA32_SMBASE:
3022 		if (!msr_info->host_initiated)
3023 			return 1;
3024 		msr_info->data = vcpu->arch.smbase;
3025 		break;
3026 	case MSR_SMI_COUNT:
3027 		msr_info->data = vcpu->arch.smi_count;
3028 		break;
3029 	case MSR_IA32_PERF_STATUS:
3030 		/* TSC increment by tick */
3031 		msr_info->data = 1000ULL;
3032 		/* CPU multiplier */
3033 		msr_info->data |= (((uint64_t)4ULL) << 40);
3034 		break;
3035 	case MSR_EFER:
3036 		msr_info->data = vcpu->arch.efer;
3037 		break;
3038 	case MSR_KVM_WALL_CLOCK:
3039 	case MSR_KVM_WALL_CLOCK_NEW:
3040 		msr_info->data = vcpu->kvm->arch.wall_clock;
3041 		break;
3042 	case MSR_KVM_SYSTEM_TIME:
3043 	case MSR_KVM_SYSTEM_TIME_NEW:
3044 		msr_info->data = vcpu->arch.time;
3045 		break;
3046 	case MSR_KVM_ASYNC_PF_EN:
3047 		msr_info->data = vcpu->arch.apf.msr_val;
3048 		break;
3049 	case MSR_KVM_STEAL_TIME:
3050 		msr_info->data = vcpu->arch.st.msr_val;
3051 		break;
3052 	case MSR_KVM_PV_EOI_EN:
3053 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
3054 		break;
3055 	case MSR_KVM_POLL_CONTROL:
3056 		msr_info->data = vcpu->arch.msr_kvm_poll_control;
3057 		break;
3058 	case MSR_IA32_P5_MC_ADDR:
3059 	case MSR_IA32_P5_MC_TYPE:
3060 	case MSR_IA32_MCG_CAP:
3061 	case MSR_IA32_MCG_CTL:
3062 	case MSR_IA32_MCG_STATUS:
3063 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3064 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3065 				   msr_info->host_initiated);
3066 	case MSR_IA32_XSS:
3067 		if (!msr_info->host_initiated &&
3068 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3069 			return 1;
3070 		msr_info->data = vcpu->arch.ia32_xss;
3071 		break;
3072 	case MSR_K7_CLK_CTL:
3073 		/*
3074 		 * Provide expected ramp-up count for K7. All other
3075 		 * are set to zero, indicating minimum divisors for
3076 		 * every field.
3077 		 *
3078 		 * This prevents guest kernels on AMD host with CPU
3079 		 * type 6, model 8 and higher from exploding due to
3080 		 * the rdmsr failing.
3081 		 */
3082 		msr_info->data = 0x20000000;
3083 		break;
3084 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3085 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3086 	case HV_X64_MSR_CRASH_CTL:
3087 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3088 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3089 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3090 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3091 		return kvm_hv_get_msr_common(vcpu,
3092 					     msr_info->index, &msr_info->data,
3093 					     msr_info->host_initiated);
3094 		break;
3095 	case MSR_IA32_BBL_CR_CTL3:
3096 		/* This legacy MSR exists but isn't fully documented in current
3097 		 * silicon.  It is however accessed by winxp in very narrow
3098 		 * scenarios where it sets bit #19, itself documented as
3099 		 * a "reserved" bit.  Best effort attempt to source coherent
3100 		 * read data here should the balance of the register be
3101 		 * interpreted by the guest:
3102 		 *
3103 		 * L2 cache control register 3: 64GB range, 256KB size,
3104 		 * enabled, latency 0x1, configured
3105 		 */
3106 		msr_info->data = 0xbe702111;
3107 		break;
3108 	case MSR_AMD64_OSVW_ID_LENGTH:
3109 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3110 			return 1;
3111 		msr_info->data = vcpu->arch.osvw.length;
3112 		break;
3113 	case MSR_AMD64_OSVW_STATUS:
3114 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3115 			return 1;
3116 		msr_info->data = vcpu->arch.osvw.status;
3117 		break;
3118 	case MSR_PLATFORM_INFO:
3119 		if (!msr_info->host_initiated &&
3120 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3121 			return 1;
3122 		msr_info->data = vcpu->arch.msr_platform_info;
3123 		break;
3124 	case MSR_MISC_FEATURES_ENABLES:
3125 		msr_info->data = vcpu->arch.msr_misc_features_enables;
3126 		break;
3127 	case MSR_K7_HWCR:
3128 		msr_info->data = vcpu->arch.msr_hwcr;
3129 		break;
3130 	default:
3131 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3132 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3133 		if (!ignore_msrs) {
3134 			vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
3135 					       msr_info->index);
3136 			return 1;
3137 		} else {
3138 			if (report_ignored_msrs)
3139 				vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
3140 					msr_info->index);
3141 			msr_info->data = 0;
3142 		}
3143 		break;
3144 	}
3145 	return 0;
3146 }
3147 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3148 
3149 /*
3150  * Read or write a bunch of msrs. All parameters are kernel addresses.
3151  *
3152  * @return number of msrs set successfully.
3153  */
3154 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3155 		    struct kvm_msr_entry *entries,
3156 		    int (*do_msr)(struct kvm_vcpu *vcpu,
3157 				  unsigned index, u64 *data))
3158 {
3159 	int i;
3160 
3161 	for (i = 0; i < msrs->nmsrs; ++i)
3162 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
3163 			break;
3164 
3165 	return i;
3166 }
3167 
3168 /*
3169  * Read or write a bunch of msrs. Parameters are user addresses.
3170  *
3171  * @return number of msrs set successfully.
3172  */
3173 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3174 		  int (*do_msr)(struct kvm_vcpu *vcpu,
3175 				unsigned index, u64 *data),
3176 		  int writeback)
3177 {
3178 	struct kvm_msrs msrs;
3179 	struct kvm_msr_entry *entries;
3180 	int r, n;
3181 	unsigned size;
3182 
3183 	r = -EFAULT;
3184 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3185 		goto out;
3186 
3187 	r = -E2BIG;
3188 	if (msrs.nmsrs >= MAX_IO_MSRS)
3189 		goto out;
3190 
3191 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3192 	entries = memdup_user(user_msrs->entries, size);
3193 	if (IS_ERR(entries)) {
3194 		r = PTR_ERR(entries);
3195 		goto out;
3196 	}
3197 
3198 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3199 	if (r < 0)
3200 		goto out_free;
3201 
3202 	r = -EFAULT;
3203 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
3204 		goto out_free;
3205 
3206 	r = n;
3207 
3208 out_free:
3209 	kfree(entries);
3210 out:
3211 	return r;
3212 }
3213 
3214 static inline bool kvm_can_mwait_in_guest(void)
3215 {
3216 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
3217 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
3218 		boot_cpu_has(X86_FEATURE_ARAT);
3219 }
3220 
3221 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3222 {
3223 	int r = 0;
3224 
3225 	switch (ext) {
3226 	case KVM_CAP_IRQCHIP:
3227 	case KVM_CAP_HLT:
3228 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3229 	case KVM_CAP_SET_TSS_ADDR:
3230 	case KVM_CAP_EXT_CPUID:
3231 	case KVM_CAP_EXT_EMUL_CPUID:
3232 	case KVM_CAP_CLOCKSOURCE:
3233 	case KVM_CAP_PIT:
3234 	case KVM_CAP_NOP_IO_DELAY:
3235 	case KVM_CAP_MP_STATE:
3236 	case KVM_CAP_SYNC_MMU:
3237 	case KVM_CAP_USER_NMI:
3238 	case KVM_CAP_REINJECT_CONTROL:
3239 	case KVM_CAP_IRQ_INJECT_STATUS:
3240 	case KVM_CAP_IOEVENTFD:
3241 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
3242 	case KVM_CAP_PIT2:
3243 	case KVM_CAP_PIT_STATE2:
3244 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3245 	case KVM_CAP_XEN_HVM:
3246 	case KVM_CAP_VCPU_EVENTS:
3247 	case KVM_CAP_HYPERV:
3248 	case KVM_CAP_HYPERV_VAPIC:
3249 	case KVM_CAP_HYPERV_SPIN:
3250 	case KVM_CAP_HYPERV_SYNIC:
3251 	case KVM_CAP_HYPERV_SYNIC2:
3252 	case KVM_CAP_HYPERV_VP_INDEX:
3253 	case KVM_CAP_HYPERV_EVENTFD:
3254 	case KVM_CAP_HYPERV_TLBFLUSH:
3255 	case KVM_CAP_HYPERV_SEND_IPI:
3256 	case KVM_CAP_HYPERV_CPUID:
3257 	case KVM_CAP_PCI_SEGMENT:
3258 	case KVM_CAP_DEBUGREGS:
3259 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
3260 	case KVM_CAP_XSAVE:
3261 	case KVM_CAP_ASYNC_PF:
3262 	case KVM_CAP_GET_TSC_KHZ:
3263 	case KVM_CAP_KVMCLOCK_CTRL:
3264 	case KVM_CAP_READONLY_MEM:
3265 	case KVM_CAP_HYPERV_TIME:
3266 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3267 	case KVM_CAP_TSC_DEADLINE_TIMER:
3268 	case KVM_CAP_DISABLE_QUIRKS:
3269 	case KVM_CAP_SET_BOOT_CPU_ID:
3270  	case KVM_CAP_SPLIT_IRQCHIP:
3271 	case KVM_CAP_IMMEDIATE_EXIT:
3272 	case KVM_CAP_PMU_EVENT_FILTER:
3273 	case KVM_CAP_GET_MSR_FEATURES:
3274 	case KVM_CAP_MSR_PLATFORM_INFO:
3275 	case KVM_CAP_EXCEPTION_PAYLOAD:
3276 		r = 1;
3277 		break;
3278 	case KVM_CAP_SYNC_REGS:
3279 		r = KVM_SYNC_X86_VALID_FIELDS;
3280 		break;
3281 	case KVM_CAP_ADJUST_CLOCK:
3282 		r = KVM_CLOCK_TSC_STABLE;
3283 		break;
3284 	case KVM_CAP_X86_DISABLE_EXITS:
3285 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3286 		      KVM_X86_DISABLE_EXITS_CSTATE;
3287 		if(kvm_can_mwait_in_guest())
3288 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
3289 		break;
3290 	case KVM_CAP_X86_SMM:
3291 		/* SMBASE is usually relocated above 1M on modern chipsets,
3292 		 * and SMM handlers might indeed rely on 4G segment limits,
3293 		 * so do not report SMM to be available if real mode is
3294 		 * emulated via vm86 mode.  Still, do not go to great lengths
3295 		 * to avoid userspace's usage of the feature, because it is a
3296 		 * fringe case that is not enabled except via specific settings
3297 		 * of the module parameters.
3298 		 */
3299 		r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3300 		break;
3301 	case KVM_CAP_VAPIC:
3302 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3303 		break;
3304 	case KVM_CAP_NR_VCPUS:
3305 		r = KVM_SOFT_MAX_VCPUS;
3306 		break;
3307 	case KVM_CAP_MAX_VCPUS:
3308 		r = KVM_MAX_VCPUS;
3309 		break;
3310 	case KVM_CAP_MAX_VCPU_ID:
3311 		r = KVM_MAX_VCPU_ID;
3312 		break;
3313 	case KVM_CAP_PV_MMU:	/* obsolete */
3314 		r = 0;
3315 		break;
3316 	case KVM_CAP_MCE:
3317 		r = KVM_MAX_MCE_BANKS;
3318 		break;
3319 	case KVM_CAP_XCRS:
3320 		r = boot_cpu_has(X86_FEATURE_XSAVE);
3321 		break;
3322 	case KVM_CAP_TSC_CONTROL:
3323 		r = kvm_has_tsc_control;
3324 		break;
3325 	case KVM_CAP_X2APIC_API:
3326 		r = KVM_X2APIC_API_VALID_FLAGS;
3327 		break;
3328 	case KVM_CAP_NESTED_STATE:
3329 		r = kvm_x86_ops->get_nested_state ?
3330 			kvm_x86_ops->get_nested_state(NULL, NULL, 0) : 0;
3331 		break;
3332 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3333 		r = kvm_x86_ops->enable_direct_tlbflush != NULL;
3334 		break;
3335 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3336 		r = kvm_x86_ops->nested_enable_evmcs != NULL;
3337 		break;
3338 	default:
3339 		break;
3340 	}
3341 	return r;
3342 
3343 }
3344 
3345 long kvm_arch_dev_ioctl(struct file *filp,
3346 			unsigned int ioctl, unsigned long arg)
3347 {
3348 	void __user *argp = (void __user *)arg;
3349 	long r;
3350 
3351 	switch (ioctl) {
3352 	case KVM_GET_MSR_INDEX_LIST: {
3353 		struct kvm_msr_list __user *user_msr_list = argp;
3354 		struct kvm_msr_list msr_list;
3355 		unsigned n;
3356 
3357 		r = -EFAULT;
3358 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3359 			goto out;
3360 		n = msr_list.nmsrs;
3361 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3362 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3363 			goto out;
3364 		r = -E2BIG;
3365 		if (n < msr_list.nmsrs)
3366 			goto out;
3367 		r = -EFAULT;
3368 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3369 				 num_msrs_to_save * sizeof(u32)))
3370 			goto out;
3371 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3372 				 &emulated_msrs,
3373 				 num_emulated_msrs * sizeof(u32)))
3374 			goto out;
3375 		r = 0;
3376 		break;
3377 	}
3378 	case KVM_GET_SUPPORTED_CPUID:
3379 	case KVM_GET_EMULATED_CPUID: {
3380 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3381 		struct kvm_cpuid2 cpuid;
3382 
3383 		r = -EFAULT;
3384 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3385 			goto out;
3386 
3387 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3388 					    ioctl);
3389 		if (r)
3390 			goto out;
3391 
3392 		r = -EFAULT;
3393 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3394 			goto out;
3395 		r = 0;
3396 		break;
3397 	}
3398 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3399 		r = -EFAULT;
3400 		if (copy_to_user(argp, &kvm_mce_cap_supported,
3401 				 sizeof(kvm_mce_cap_supported)))
3402 			goto out;
3403 		r = 0;
3404 		break;
3405 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3406 		struct kvm_msr_list __user *user_msr_list = argp;
3407 		struct kvm_msr_list msr_list;
3408 		unsigned int n;
3409 
3410 		r = -EFAULT;
3411 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3412 			goto out;
3413 		n = msr_list.nmsrs;
3414 		msr_list.nmsrs = num_msr_based_features;
3415 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3416 			goto out;
3417 		r = -E2BIG;
3418 		if (n < msr_list.nmsrs)
3419 			goto out;
3420 		r = -EFAULT;
3421 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
3422 				 num_msr_based_features * sizeof(u32)))
3423 			goto out;
3424 		r = 0;
3425 		break;
3426 	}
3427 	case KVM_GET_MSRS:
3428 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
3429 		break;
3430 	}
3431 	default:
3432 		r = -EINVAL;
3433 	}
3434 out:
3435 	return r;
3436 }
3437 
3438 static void wbinvd_ipi(void *garbage)
3439 {
3440 	wbinvd();
3441 }
3442 
3443 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3444 {
3445 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3446 }
3447 
3448 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3449 {
3450 	/* Address WBINVD may be executed by guest */
3451 	if (need_emulate_wbinvd(vcpu)) {
3452 		if (kvm_x86_ops->has_wbinvd_exit())
3453 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3454 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3455 			smp_call_function_single(vcpu->cpu,
3456 					wbinvd_ipi, NULL, 1);
3457 	}
3458 
3459 	kvm_x86_ops->vcpu_load(vcpu, cpu);
3460 
3461 	fpregs_assert_state_consistent();
3462 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
3463 		switch_fpu_return();
3464 
3465 	/* Apply any externally detected TSC adjustments (due to suspend) */
3466 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3467 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3468 		vcpu->arch.tsc_offset_adjustment = 0;
3469 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3470 	}
3471 
3472 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3473 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3474 				rdtsc() - vcpu->arch.last_host_tsc;
3475 		if (tsc_delta < 0)
3476 			mark_tsc_unstable("KVM discovered backwards TSC");
3477 
3478 		if (kvm_check_tsc_unstable()) {
3479 			u64 offset = kvm_compute_tsc_offset(vcpu,
3480 						vcpu->arch.last_guest_tsc);
3481 			kvm_vcpu_write_tsc_offset(vcpu, offset);
3482 			vcpu->arch.tsc_catchup = 1;
3483 		}
3484 
3485 		if (kvm_lapic_hv_timer_in_use(vcpu))
3486 			kvm_lapic_restart_hv_timer(vcpu);
3487 
3488 		/*
3489 		 * On a host with synchronized TSC, there is no need to update
3490 		 * kvmclock on vcpu->cpu migration
3491 		 */
3492 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3493 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3494 		if (vcpu->cpu != cpu)
3495 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3496 		vcpu->cpu = cpu;
3497 	}
3498 
3499 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3500 }
3501 
3502 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3503 {
3504 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3505 		return;
3506 
3507 	vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3508 
3509 	kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3510 			&vcpu->arch.st.steal.preempted,
3511 			offsetof(struct kvm_steal_time, preempted),
3512 			sizeof(vcpu->arch.st.steal.preempted));
3513 }
3514 
3515 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3516 {
3517 	int idx;
3518 
3519 	if (vcpu->preempted)
3520 		vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3521 
3522 	/*
3523 	 * Disable page faults because we're in atomic context here.
3524 	 * kvm_write_guest_offset_cached() would call might_fault()
3525 	 * that relies on pagefault_disable() to tell if there's a
3526 	 * bug. NOTE: the write to guest memory may not go through if
3527 	 * during postcopy live migration or if there's heavy guest
3528 	 * paging.
3529 	 */
3530 	pagefault_disable();
3531 	/*
3532 	 * kvm_memslots() will be called by
3533 	 * kvm_write_guest_offset_cached() so take the srcu lock.
3534 	 */
3535 	idx = srcu_read_lock(&vcpu->kvm->srcu);
3536 	kvm_steal_time_set_preempted(vcpu);
3537 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
3538 	pagefault_enable();
3539 	kvm_x86_ops->vcpu_put(vcpu);
3540 	vcpu->arch.last_host_tsc = rdtsc();
3541 	/*
3542 	 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3543 	 * on every vmexit, but if not, we might have a stale dr6 from the
3544 	 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3545 	 */
3546 	set_debugreg(0, 6);
3547 }
3548 
3549 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3550 				    struct kvm_lapic_state *s)
3551 {
3552 	if (vcpu->arch.apicv_active)
3553 		kvm_x86_ops->sync_pir_to_irr(vcpu);
3554 
3555 	return kvm_apic_get_state(vcpu, s);
3556 }
3557 
3558 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3559 				    struct kvm_lapic_state *s)
3560 {
3561 	int r;
3562 
3563 	r = kvm_apic_set_state(vcpu, s);
3564 	if (r)
3565 		return r;
3566 	update_cr8_intercept(vcpu);
3567 
3568 	return 0;
3569 }
3570 
3571 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3572 {
3573 	return (!lapic_in_kernel(vcpu) ||
3574 		kvm_apic_accept_pic_intr(vcpu));
3575 }
3576 
3577 /*
3578  * if userspace requested an interrupt window, check that the
3579  * interrupt window is open.
3580  *
3581  * No need to exit to userspace if we already have an interrupt queued.
3582  */
3583 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3584 {
3585 	return kvm_arch_interrupt_allowed(vcpu) &&
3586 		!kvm_cpu_has_interrupt(vcpu) &&
3587 		!kvm_event_needs_reinjection(vcpu) &&
3588 		kvm_cpu_accept_dm_intr(vcpu);
3589 }
3590 
3591 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3592 				    struct kvm_interrupt *irq)
3593 {
3594 	if (irq->irq >= KVM_NR_INTERRUPTS)
3595 		return -EINVAL;
3596 
3597 	if (!irqchip_in_kernel(vcpu->kvm)) {
3598 		kvm_queue_interrupt(vcpu, irq->irq, false);
3599 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3600 		return 0;
3601 	}
3602 
3603 	/*
3604 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3605 	 * fail for in-kernel 8259.
3606 	 */
3607 	if (pic_in_kernel(vcpu->kvm))
3608 		return -ENXIO;
3609 
3610 	if (vcpu->arch.pending_external_vector != -1)
3611 		return -EEXIST;
3612 
3613 	vcpu->arch.pending_external_vector = irq->irq;
3614 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3615 	return 0;
3616 }
3617 
3618 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3619 {
3620 	kvm_inject_nmi(vcpu);
3621 
3622 	return 0;
3623 }
3624 
3625 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3626 {
3627 	kvm_make_request(KVM_REQ_SMI, vcpu);
3628 
3629 	return 0;
3630 }
3631 
3632 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3633 					   struct kvm_tpr_access_ctl *tac)
3634 {
3635 	if (tac->flags)
3636 		return -EINVAL;
3637 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
3638 	return 0;
3639 }
3640 
3641 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3642 					u64 mcg_cap)
3643 {
3644 	int r;
3645 	unsigned bank_num = mcg_cap & 0xff, bank;
3646 
3647 	r = -EINVAL;
3648 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3649 		goto out;
3650 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3651 		goto out;
3652 	r = 0;
3653 	vcpu->arch.mcg_cap = mcg_cap;
3654 	/* Init IA32_MCG_CTL to all 1s */
3655 	if (mcg_cap & MCG_CTL_P)
3656 		vcpu->arch.mcg_ctl = ~(u64)0;
3657 	/* Init IA32_MCi_CTL to all 1s */
3658 	for (bank = 0; bank < bank_num; bank++)
3659 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3660 
3661 	kvm_x86_ops->setup_mce(vcpu);
3662 out:
3663 	return r;
3664 }
3665 
3666 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3667 				      struct kvm_x86_mce *mce)
3668 {
3669 	u64 mcg_cap = vcpu->arch.mcg_cap;
3670 	unsigned bank_num = mcg_cap & 0xff;
3671 	u64 *banks = vcpu->arch.mce_banks;
3672 
3673 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3674 		return -EINVAL;
3675 	/*
3676 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3677 	 * reporting is disabled
3678 	 */
3679 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3680 	    vcpu->arch.mcg_ctl != ~(u64)0)
3681 		return 0;
3682 	banks += 4 * mce->bank;
3683 	/*
3684 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3685 	 * reporting is disabled for the bank
3686 	 */
3687 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3688 		return 0;
3689 	if (mce->status & MCI_STATUS_UC) {
3690 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3691 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3692 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3693 			return 0;
3694 		}
3695 		if (banks[1] & MCI_STATUS_VAL)
3696 			mce->status |= MCI_STATUS_OVER;
3697 		banks[2] = mce->addr;
3698 		banks[3] = mce->misc;
3699 		vcpu->arch.mcg_status = mce->mcg_status;
3700 		banks[1] = mce->status;
3701 		kvm_queue_exception(vcpu, MC_VECTOR);
3702 	} else if (!(banks[1] & MCI_STATUS_VAL)
3703 		   || !(banks[1] & MCI_STATUS_UC)) {
3704 		if (banks[1] & MCI_STATUS_VAL)
3705 			mce->status |= MCI_STATUS_OVER;
3706 		banks[2] = mce->addr;
3707 		banks[3] = mce->misc;
3708 		banks[1] = mce->status;
3709 	} else
3710 		banks[1] |= MCI_STATUS_OVER;
3711 	return 0;
3712 }
3713 
3714 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3715 					       struct kvm_vcpu_events *events)
3716 {
3717 	process_nmi(vcpu);
3718 
3719 	/*
3720 	 * The API doesn't provide the instruction length for software
3721 	 * exceptions, so don't report them. As long as the guest RIP
3722 	 * isn't advanced, we should expect to encounter the exception
3723 	 * again.
3724 	 */
3725 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3726 		events->exception.injected = 0;
3727 		events->exception.pending = 0;
3728 	} else {
3729 		events->exception.injected = vcpu->arch.exception.injected;
3730 		events->exception.pending = vcpu->arch.exception.pending;
3731 		/*
3732 		 * For ABI compatibility, deliberately conflate
3733 		 * pending and injected exceptions when
3734 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3735 		 */
3736 		if (!vcpu->kvm->arch.exception_payload_enabled)
3737 			events->exception.injected |=
3738 				vcpu->arch.exception.pending;
3739 	}
3740 	events->exception.nr = vcpu->arch.exception.nr;
3741 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3742 	events->exception.error_code = vcpu->arch.exception.error_code;
3743 	events->exception_has_payload = vcpu->arch.exception.has_payload;
3744 	events->exception_payload = vcpu->arch.exception.payload;
3745 
3746 	events->interrupt.injected =
3747 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3748 	events->interrupt.nr = vcpu->arch.interrupt.nr;
3749 	events->interrupt.soft = 0;
3750 	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3751 
3752 	events->nmi.injected = vcpu->arch.nmi_injected;
3753 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
3754 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3755 	events->nmi.pad = 0;
3756 
3757 	events->sipi_vector = 0; /* never valid when reporting to user space */
3758 
3759 	events->smi.smm = is_smm(vcpu);
3760 	events->smi.pending = vcpu->arch.smi_pending;
3761 	events->smi.smm_inside_nmi =
3762 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3763 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3764 
3765 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3766 			 | KVM_VCPUEVENT_VALID_SHADOW
3767 			 | KVM_VCPUEVENT_VALID_SMM);
3768 	if (vcpu->kvm->arch.exception_payload_enabled)
3769 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3770 
3771 	memset(&events->reserved, 0, sizeof(events->reserved));
3772 }
3773 
3774 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
3775 
3776 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3777 					      struct kvm_vcpu_events *events)
3778 {
3779 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3780 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3781 			      | KVM_VCPUEVENT_VALID_SHADOW
3782 			      | KVM_VCPUEVENT_VALID_SMM
3783 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
3784 		return -EINVAL;
3785 
3786 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3787 		if (!vcpu->kvm->arch.exception_payload_enabled)
3788 			return -EINVAL;
3789 		if (events->exception.pending)
3790 			events->exception.injected = 0;
3791 		else
3792 			events->exception_has_payload = 0;
3793 	} else {
3794 		events->exception.pending = 0;
3795 		events->exception_has_payload = 0;
3796 	}
3797 
3798 	if ((events->exception.injected || events->exception.pending) &&
3799 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3800 		return -EINVAL;
3801 
3802 	/* INITs are latched while in SMM */
3803 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3804 	    (events->smi.smm || events->smi.pending) &&
3805 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3806 		return -EINVAL;
3807 
3808 	process_nmi(vcpu);
3809 	vcpu->arch.exception.injected = events->exception.injected;
3810 	vcpu->arch.exception.pending = events->exception.pending;
3811 	vcpu->arch.exception.nr = events->exception.nr;
3812 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3813 	vcpu->arch.exception.error_code = events->exception.error_code;
3814 	vcpu->arch.exception.has_payload = events->exception_has_payload;
3815 	vcpu->arch.exception.payload = events->exception_payload;
3816 
3817 	vcpu->arch.interrupt.injected = events->interrupt.injected;
3818 	vcpu->arch.interrupt.nr = events->interrupt.nr;
3819 	vcpu->arch.interrupt.soft = events->interrupt.soft;
3820 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3821 		kvm_x86_ops->set_interrupt_shadow(vcpu,
3822 						  events->interrupt.shadow);
3823 
3824 	vcpu->arch.nmi_injected = events->nmi.injected;
3825 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3826 		vcpu->arch.nmi_pending = events->nmi.pending;
3827 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3828 
3829 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3830 	    lapic_in_kernel(vcpu))
3831 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
3832 
3833 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3834 		if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
3835 			if (events->smi.smm)
3836 				vcpu->arch.hflags |= HF_SMM_MASK;
3837 			else
3838 				vcpu->arch.hflags &= ~HF_SMM_MASK;
3839 			kvm_smm_changed(vcpu);
3840 		}
3841 
3842 		vcpu->arch.smi_pending = events->smi.pending;
3843 
3844 		if (events->smi.smm) {
3845 			if (events->smi.smm_inside_nmi)
3846 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3847 			else
3848 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3849 		}
3850 
3851 		if (lapic_in_kernel(vcpu)) {
3852 			if (events->smi.latched_init)
3853 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3854 			else
3855 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3856 		}
3857 	}
3858 
3859 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3860 
3861 	return 0;
3862 }
3863 
3864 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3865 					     struct kvm_debugregs *dbgregs)
3866 {
3867 	unsigned long val;
3868 
3869 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3870 	kvm_get_dr(vcpu, 6, &val);
3871 	dbgregs->dr6 = val;
3872 	dbgregs->dr7 = vcpu->arch.dr7;
3873 	dbgregs->flags = 0;
3874 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3875 }
3876 
3877 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3878 					    struct kvm_debugregs *dbgregs)
3879 {
3880 	if (dbgregs->flags)
3881 		return -EINVAL;
3882 
3883 	if (dbgregs->dr6 & ~0xffffffffull)
3884 		return -EINVAL;
3885 	if (dbgregs->dr7 & ~0xffffffffull)
3886 		return -EINVAL;
3887 
3888 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3889 	kvm_update_dr0123(vcpu);
3890 	vcpu->arch.dr6 = dbgregs->dr6;
3891 	kvm_update_dr6(vcpu);
3892 	vcpu->arch.dr7 = dbgregs->dr7;
3893 	kvm_update_dr7(vcpu);
3894 
3895 	return 0;
3896 }
3897 
3898 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3899 
3900 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3901 {
3902 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3903 	u64 xstate_bv = xsave->header.xfeatures;
3904 	u64 valid;
3905 
3906 	/*
3907 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3908 	 * leaves 0 and 1 in the loop below.
3909 	 */
3910 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3911 
3912 	/* Set XSTATE_BV */
3913 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3914 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3915 
3916 	/*
3917 	 * Copy each region from the possibly compacted offset to the
3918 	 * non-compacted offset.
3919 	 */
3920 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3921 	while (valid) {
3922 		u64 xfeature_mask = valid & -valid;
3923 		int xfeature_nr = fls64(xfeature_mask) - 1;
3924 		void *src = get_xsave_addr(xsave, xfeature_nr);
3925 
3926 		if (src) {
3927 			u32 size, offset, ecx, edx;
3928 			cpuid_count(XSTATE_CPUID, xfeature_nr,
3929 				    &size, &offset, &ecx, &edx);
3930 			if (xfeature_nr == XFEATURE_PKRU)
3931 				memcpy(dest + offset, &vcpu->arch.pkru,
3932 				       sizeof(vcpu->arch.pkru));
3933 			else
3934 				memcpy(dest + offset, src, size);
3935 
3936 		}
3937 
3938 		valid -= xfeature_mask;
3939 	}
3940 }
3941 
3942 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3943 {
3944 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3945 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3946 	u64 valid;
3947 
3948 	/*
3949 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3950 	 * leaves 0 and 1 in the loop below.
3951 	 */
3952 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
3953 
3954 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
3955 	xsave->header.xfeatures = xstate_bv;
3956 	if (boot_cpu_has(X86_FEATURE_XSAVES))
3957 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3958 
3959 	/*
3960 	 * Copy each region from the non-compacted offset to the
3961 	 * possibly compacted offset.
3962 	 */
3963 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3964 	while (valid) {
3965 		u64 xfeature_mask = valid & -valid;
3966 		int xfeature_nr = fls64(xfeature_mask) - 1;
3967 		void *dest = get_xsave_addr(xsave, xfeature_nr);
3968 
3969 		if (dest) {
3970 			u32 size, offset, ecx, edx;
3971 			cpuid_count(XSTATE_CPUID, xfeature_nr,
3972 				    &size, &offset, &ecx, &edx);
3973 			if (xfeature_nr == XFEATURE_PKRU)
3974 				memcpy(&vcpu->arch.pkru, src + offset,
3975 				       sizeof(vcpu->arch.pkru));
3976 			else
3977 				memcpy(dest, src + offset, size);
3978 		}
3979 
3980 		valid -= xfeature_mask;
3981 	}
3982 }
3983 
3984 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3985 					 struct kvm_xsave *guest_xsave)
3986 {
3987 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3988 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3989 		fill_xsave((u8 *) guest_xsave->region, vcpu);
3990 	} else {
3991 		memcpy(guest_xsave->region,
3992 			&vcpu->arch.guest_fpu->state.fxsave,
3993 			sizeof(struct fxregs_state));
3994 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3995 			XFEATURE_MASK_FPSSE;
3996 	}
3997 }
3998 
3999 #define XSAVE_MXCSR_OFFSET 24
4000 
4001 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4002 					struct kvm_xsave *guest_xsave)
4003 {
4004 	u64 xstate_bv =
4005 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4006 	u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4007 
4008 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4009 		/*
4010 		 * Here we allow setting states that are not present in
4011 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
4012 		 * with old userspace.
4013 		 */
4014 		if (xstate_bv & ~kvm_supported_xcr0() ||
4015 			mxcsr & ~mxcsr_feature_mask)
4016 			return -EINVAL;
4017 		load_xsave(vcpu, (u8 *)guest_xsave->region);
4018 	} else {
4019 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4020 			mxcsr & ~mxcsr_feature_mask)
4021 			return -EINVAL;
4022 		memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4023 			guest_xsave->region, sizeof(struct fxregs_state));
4024 	}
4025 	return 0;
4026 }
4027 
4028 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4029 					struct kvm_xcrs *guest_xcrs)
4030 {
4031 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4032 		guest_xcrs->nr_xcrs = 0;
4033 		return;
4034 	}
4035 
4036 	guest_xcrs->nr_xcrs = 1;
4037 	guest_xcrs->flags = 0;
4038 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4039 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4040 }
4041 
4042 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4043 				       struct kvm_xcrs *guest_xcrs)
4044 {
4045 	int i, r = 0;
4046 
4047 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
4048 		return -EINVAL;
4049 
4050 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4051 		return -EINVAL;
4052 
4053 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4054 		/* Only support XCR0 currently */
4055 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4056 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4057 				guest_xcrs->xcrs[i].value);
4058 			break;
4059 		}
4060 	if (r)
4061 		r = -EINVAL;
4062 	return r;
4063 }
4064 
4065 /*
4066  * kvm_set_guest_paused() indicates to the guest kernel that it has been
4067  * stopped by the hypervisor.  This function will be called from the host only.
4068  * EINVAL is returned when the host attempts to set the flag for a guest that
4069  * does not support pv clocks.
4070  */
4071 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4072 {
4073 	if (!vcpu->arch.pv_time_enabled)
4074 		return -EINVAL;
4075 	vcpu->arch.pvclock_set_guest_stopped_request = true;
4076 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4077 	return 0;
4078 }
4079 
4080 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4081 				     struct kvm_enable_cap *cap)
4082 {
4083 	int r;
4084 	uint16_t vmcs_version;
4085 	void __user *user_ptr;
4086 
4087 	if (cap->flags)
4088 		return -EINVAL;
4089 
4090 	switch (cap->cap) {
4091 	case KVM_CAP_HYPERV_SYNIC2:
4092 		if (cap->args[0])
4093 			return -EINVAL;
4094 		/* fall through */
4095 
4096 	case KVM_CAP_HYPERV_SYNIC:
4097 		if (!irqchip_in_kernel(vcpu->kvm))
4098 			return -EINVAL;
4099 		return kvm_hv_activate_synic(vcpu, cap->cap ==
4100 					     KVM_CAP_HYPERV_SYNIC2);
4101 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4102 		if (!kvm_x86_ops->nested_enable_evmcs)
4103 			return -ENOTTY;
4104 		r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
4105 		if (!r) {
4106 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
4107 			if (copy_to_user(user_ptr, &vmcs_version,
4108 					 sizeof(vmcs_version)))
4109 				r = -EFAULT;
4110 		}
4111 		return r;
4112 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4113 		if (!kvm_x86_ops->enable_direct_tlbflush)
4114 			return -ENOTTY;
4115 
4116 		return kvm_x86_ops->enable_direct_tlbflush(vcpu);
4117 
4118 	default:
4119 		return -EINVAL;
4120 	}
4121 }
4122 
4123 long kvm_arch_vcpu_ioctl(struct file *filp,
4124 			 unsigned int ioctl, unsigned long arg)
4125 {
4126 	struct kvm_vcpu *vcpu = filp->private_data;
4127 	void __user *argp = (void __user *)arg;
4128 	int r;
4129 	union {
4130 		struct kvm_lapic_state *lapic;
4131 		struct kvm_xsave *xsave;
4132 		struct kvm_xcrs *xcrs;
4133 		void *buffer;
4134 	} u;
4135 
4136 	vcpu_load(vcpu);
4137 
4138 	u.buffer = NULL;
4139 	switch (ioctl) {
4140 	case KVM_GET_LAPIC: {
4141 		r = -EINVAL;
4142 		if (!lapic_in_kernel(vcpu))
4143 			goto out;
4144 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4145 				GFP_KERNEL_ACCOUNT);
4146 
4147 		r = -ENOMEM;
4148 		if (!u.lapic)
4149 			goto out;
4150 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4151 		if (r)
4152 			goto out;
4153 		r = -EFAULT;
4154 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4155 			goto out;
4156 		r = 0;
4157 		break;
4158 	}
4159 	case KVM_SET_LAPIC: {
4160 		r = -EINVAL;
4161 		if (!lapic_in_kernel(vcpu))
4162 			goto out;
4163 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
4164 		if (IS_ERR(u.lapic)) {
4165 			r = PTR_ERR(u.lapic);
4166 			goto out_nofree;
4167 		}
4168 
4169 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4170 		break;
4171 	}
4172 	case KVM_INTERRUPT: {
4173 		struct kvm_interrupt irq;
4174 
4175 		r = -EFAULT;
4176 		if (copy_from_user(&irq, argp, sizeof(irq)))
4177 			goto out;
4178 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4179 		break;
4180 	}
4181 	case KVM_NMI: {
4182 		r = kvm_vcpu_ioctl_nmi(vcpu);
4183 		break;
4184 	}
4185 	case KVM_SMI: {
4186 		r = kvm_vcpu_ioctl_smi(vcpu);
4187 		break;
4188 	}
4189 	case KVM_SET_CPUID: {
4190 		struct kvm_cpuid __user *cpuid_arg = argp;
4191 		struct kvm_cpuid cpuid;
4192 
4193 		r = -EFAULT;
4194 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4195 			goto out;
4196 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4197 		break;
4198 	}
4199 	case KVM_SET_CPUID2: {
4200 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4201 		struct kvm_cpuid2 cpuid;
4202 
4203 		r = -EFAULT;
4204 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4205 			goto out;
4206 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4207 					      cpuid_arg->entries);
4208 		break;
4209 	}
4210 	case KVM_GET_CPUID2: {
4211 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4212 		struct kvm_cpuid2 cpuid;
4213 
4214 		r = -EFAULT;
4215 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4216 			goto out;
4217 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4218 					      cpuid_arg->entries);
4219 		if (r)
4220 			goto out;
4221 		r = -EFAULT;
4222 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4223 			goto out;
4224 		r = 0;
4225 		break;
4226 	}
4227 	case KVM_GET_MSRS: {
4228 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4229 		r = msr_io(vcpu, argp, do_get_msr, 1);
4230 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4231 		break;
4232 	}
4233 	case KVM_SET_MSRS: {
4234 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4235 		r = msr_io(vcpu, argp, do_set_msr, 0);
4236 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4237 		break;
4238 	}
4239 	case KVM_TPR_ACCESS_REPORTING: {
4240 		struct kvm_tpr_access_ctl tac;
4241 
4242 		r = -EFAULT;
4243 		if (copy_from_user(&tac, argp, sizeof(tac)))
4244 			goto out;
4245 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4246 		if (r)
4247 			goto out;
4248 		r = -EFAULT;
4249 		if (copy_to_user(argp, &tac, sizeof(tac)))
4250 			goto out;
4251 		r = 0;
4252 		break;
4253 	};
4254 	case KVM_SET_VAPIC_ADDR: {
4255 		struct kvm_vapic_addr va;
4256 		int idx;
4257 
4258 		r = -EINVAL;
4259 		if (!lapic_in_kernel(vcpu))
4260 			goto out;
4261 		r = -EFAULT;
4262 		if (copy_from_user(&va, argp, sizeof(va)))
4263 			goto out;
4264 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4265 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4266 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4267 		break;
4268 	}
4269 	case KVM_X86_SETUP_MCE: {
4270 		u64 mcg_cap;
4271 
4272 		r = -EFAULT;
4273 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4274 			goto out;
4275 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4276 		break;
4277 	}
4278 	case KVM_X86_SET_MCE: {
4279 		struct kvm_x86_mce mce;
4280 
4281 		r = -EFAULT;
4282 		if (copy_from_user(&mce, argp, sizeof(mce)))
4283 			goto out;
4284 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4285 		break;
4286 	}
4287 	case KVM_GET_VCPU_EVENTS: {
4288 		struct kvm_vcpu_events events;
4289 
4290 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4291 
4292 		r = -EFAULT;
4293 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4294 			break;
4295 		r = 0;
4296 		break;
4297 	}
4298 	case KVM_SET_VCPU_EVENTS: {
4299 		struct kvm_vcpu_events events;
4300 
4301 		r = -EFAULT;
4302 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4303 			break;
4304 
4305 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4306 		break;
4307 	}
4308 	case KVM_GET_DEBUGREGS: {
4309 		struct kvm_debugregs dbgregs;
4310 
4311 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4312 
4313 		r = -EFAULT;
4314 		if (copy_to_user(argp, &dbgregs,
4315 				 sizeof(struct kvm_debugregs)))
4316 			break;
4317 		r = 0;
4318 		break;
4319 	}
4320 	case KVM_SET_DEBUGREGS: {
4321 		struct kvm_debugregs dbgregs;
4322 
4323 		r = -EFAULT;
4324 		if (copy_from_user(&dbgregs, argp,
4325 				   sizeof(struct kvm_debugregs)))
4326 			break;
4327 
4328 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4329 		break;
4330 	}
4331 	case KVM_GET_XSAVE: {
4332 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4333 		r = -ENOMEM;
4334 		if (!u.xsave)
4335 			break;
4336 
4337 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4338 
4339 		r = -EFAULT;
4340 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4341 			break;
4342 		r = 0;
4343 		break;
4344 	}
4345 	case KVM_SET_XSAVE: {
4346 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
4347 		if (IS_ERR(u.xsave)) {
4348 			r = PTR_ERR(u.xsave);
4349 			goto out_nofree;
4350 		}
4351 
4352 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4353 		break;
4354 	}
4355 	case KVM_GET_XCRS: {
4356 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4357 		r = -ENOMEM;
4358 		if (!u.xcrs)
4359 			break;
4360 
4361 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4362 
4363 		r = -EFAULT;
4364 		if (copy_to_user(argp, u.xcrs,
4365 				 sizeof(struct kvm_xcrs)))
4366 			break;
4367 		r = 0;
4368 		break;
4369 	}
4370 	case KVM_SET_XCRS: {
4371 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4372 		if (IS_ERR(u.xcrs)) {
4373 			r = PTR_ERR(u.xcrs);
4374 			goto out_nofree;
4375 		}
4376 
4377 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4378 		break;
4379 	}
4380 	case KVM_SET_TSC_KHZ: {
4381 		u32 user_tsc_khz;
4382 
4383 		r = -EINVAL;
4384 		user_tsc_khz = (u32)arg;
4385 
4386 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4387 			goto out;
4388 
4389 		if (user_tsc_khz == 0)
4390 			user_tsc_khz = tsc_khz;
4391 
4392 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4393 			r = 0;
4394 
4395 		goto out;
4396 	}
4397 	case KVM_GET_TSC_KHZ: {
4398 		r = vcpu->arch.virtual_tsc_khz;
4399 		goto out;
4400 	}
4401 	case KVM_KVMCLOCK_CTRL: {
4402 		r = kvm_set_guest_paused(vcpu);
4403 		goto out;
4404 	}
4405 	case KVM_ENABLE_CAP: {
4406 		struct kvm_enable_cap cap;
4407 
4408 		r = -EFAULT;
4409 		if (copy_from_user(&cap, argp, sizeof(cap)))
4410 			goto out;
4411 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4412 		break;
4413 	}
4414 	case KVM_GET_NESTED_STATE: {
4415 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4416 		u32 user_data_size;
4417 
4418 		r = -EINVAL;
4419 		if (!kvm_x86_ops->get_nested_state)
4420 			break;
4421 
4422 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4423 		r = -EFAULT;
4424 		if (get_user(user_data_size, &user_kvm_nested_state->size))
4425 			break;
4426 
4427 		r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4428 						  user_data_size);
4429 		if (r < 0)
4430 			break;
4431 
4432 		if (r > user_data_size) {
4433 			if (put_user(r, &user_kvm_nested_state->size))
4434 				r = -EFAULT;
4435 			else
4436 				r = -E2BIG;
4437 			break;
4438 		}
4439 
4440 		r = 0;
4441 		break;
4442 	}
4443 	case KVM_SET_NESTED_STATE: {
4444 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4445 		struct kvm_nested_state kvm_state;
4446 		int idx;
4447 
4448 		r = -EINVAL;
4449 		if (!kvm_x86_ops->set_nested_state)
4450 			break;
4451 
4452 		r = -EFAULT;
4453 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4454 			break;
4455 
4456 		r = -EINVAL;
4457 		if (kvm_state.size < sizeof(kvm_state))
4458 			break;
4459 
4460 		if (kvm_state.flags &
4461 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4462 		      | KVM_STATE_NESTED_EVMCS))
4463 			break;
4464 
4465 		/* nested_run_pending implies guest_mode.  */
4466 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4467 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4468 			break;
4469 
4470 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4471 		r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4472 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4473 		break;
4474 	}
4475 	case KVM_GET_SUPPORTED_HV_CPUID: {
4476 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4477 		struct kvm_cpuid2 cpuid;
4478 
4479 		r = -EFAULT;
4480 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4481 			goto out;
4482 
4483 		r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4484 						cpuid_arg->entries);
4485 		if (r)
4486 			goto out;
4487 
4488 		r = -EFAULT;
4489 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4490 			goto out;
4491 		r = 0;
4492 		break;
4493 	}
4494 	default:
4495 		r = -EINVAL;
4496 	}
4497 out:
4498 	kfree(u.buffer);
4499 out_nofree:
4500 	vcpu_put(vcpu);
4501 	return r;
4502 }
4503 
4504 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4505 {
4506 	return VM_FAULT_SIGBUS;
4507 }
4508 
4509 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4510 {
4511 	int ret;
4512 
4513 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
4514 		return -EINVAL;
4515 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4516 	return ret;
4517 }
4518 
4519 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4520 					      u64 ident_addr)
4521 {
4522 	return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4523 }
4524 
4525 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4526 					 unsigned long kvm_nr_mmu_pages)
4527 {
4528 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4529 		return -EINVAL;
4530 
4531 	mutex_lock(&kvm->slots_lock);
4532 
4533 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4534 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4535 
4536 	mutex_unlock(&kvm->slots_lock);
4537 	return 0;
4538 }
4539 
4540 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4541 {
4542 	return kvm->arch.n_max_mmu_pages;
4543 }
4544 
4545 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4546 {
4547 	struct kvm_pic *pic = kvm->arch.vpic;
4548 	int r;
4549 
4550 	r = 0;
4551 	switch (chip->chip_id) {
4552 	case KVM_IRQCHIP_PIC_MASTER:
4553 		memcpy(&chip->chip.pic, &pic->pics[0],
4554 			sizeof(struct kvm_pic_state));
4555 		break;
4556 	case KVM_IRQCHIP_PIC_SLAVE:
4557 		memcpy(&chip->chip.pic, &pic->pics[1],
4558 			sizeof(struct kvm_pic_state));
4559 		break;
4560 	case KVM_IRQCHIP_IOAPIC:
4561 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
4562 		break;
4563 	default:
4564 		r = -EINVAL;
4565 		break;
4566 	}
4567 	return r;
4568 }
4569 
4570 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4571 {
4572 	struct kvm_pic *pic = kvm->arch.vpic;
4573 	int r;
4574 
4575 	r = 0;
4576 	switch (chip->chip_id) {
4577 	case KVM_IRQCHIP_PIC_MASTER:
4578 		spin_lock(&pic->lock);
4579 		memcpy(&pic->pics[0], &chip->chip.pic,
4580 			sizeof(struct kvm_pic_state));
4581 		spin_unlock(&pic->lock);
4582 		break;
4583 	case KVM_IRQCHIP_PIC_SLAVE:
4584 		spin_lock(&pic->lock);
4585 		memcpy(&pic->pics[1], &chip->chip.pic,
4586 			sizeof(struct kvm_pic_state));
4587 		spin_unlock(&pic->lock);
4588 		break;
4589 	case KVM_IRQCHIP_IOAPIC:
4590 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
4591 		break;
4592 	default:
4593 		r = -EINVAL;
4594 		break;
4595 	}
4596 	kvm_pic_update_irq(pic);
4597 	return r;
4598 }
4599 
4600 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4601 {
4602 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4603 
4604 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4605 
4606 	mutex_lock(&kps->lock);
4607 	memcpy(ps, &kps->channels, sizeof(*ps));
4608 	mutex_unlock(&kps->lock);
4609 	return 0;
4610 }
4611 
4612 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4613 {
4614 	int i;
4615 	struct kvm_pit *pit = kvm->arch.vpit;
4616 
4617 	mutex_lock(&pit->pit_state.lock);
4618 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4619 	for (i = 0; i < 3; i++)
4620 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4621 	mutex_unlock(&pit->pit_state.lock);
4622 	return 0;
4623 }
4624 
4625 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4626 {
4627 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
4628 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4629 		sizeof(ps->channels));
4630 	ps->flags = kvm->arch.vpit->pit_state.flags;
4631 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4632 	memset(&ps->reserved, 0, sizeof(ps->reserved));
4633 	return 0;
4634 }
4635 
4636 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4637 {
4638 	int start = 0;
4639 	int i;
4640 	u32 prev_legacy, cur_legacy;
4641 	struct kvm_pit *pit = kvm->arch.vpit;
4642 
4643 	mutex_lock(&pit->pit_state.lock);
4644 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4645 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4646 	if (!prev_legacy && cur_legacy)
4647 		start = 1;
4648 	memcpy(&pit->pit_state.channels, &ps->channels,
4649 	       sizeof(pit->pit_state.channels));
4650 	pit->pit_state.flags = ps->flags;
4651 	for (i = 0; i < 3; i++)
4652 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4653 				   start && i == 0);
4654 	mutex_unlock(&pit->pit_state.lock);
4655 	return 0;
4656 }
4657 
4658 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4659 				 struct kvm_reinject_control *control)
4660 {
4661 	struct kvm_pit *pit = kvm->arch.vpit;
4662 
4663 	if (!pit)
4664 		return -ENXIO;
4665 
4666 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
4667 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4668 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
4669 	 */
4670 	mutex_lock(&pit->pit_state.lock);
4671 	kvm_pit_set_reinject(pit, control->pit_reinject);
4672 	mutex_unlock(&pit->pit_state.lock);
4673 
4674 	return 0;
4675 }
4676 
4677 /**
4678  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4679  * @kvm: kvm instance
4680  * @log: slot id and address to which we copy the log
4681  *
4682  * Steps 1-4 below provide general overview of dirty page logging. See
4683  * kvm_get_dirty_log_protect() function description for additional details.
4684  *
4685  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4686  * always flush the TLB (step 4) even if previous step failed  and the dirty
4687  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4688  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4689  * writes will be marked dirty for next log read.
4690  *
4691  *   1. Take a snapshot of the bit and clear it if needed.
4692  *   2. Write protect the corresponding page.
4693  *   3. Copy the snapshot to the userspace.
4694  *   4. Flush TLB's if needed.
4695  */
4696 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4697 {
4698 	bool flush = false;
4699 	int r;
4700 
4701 	mutex_lock(&kvm->slots_lock);
4702 
4703 	/*
4704 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4705 	 */
4706 	if (kvm_x86_ops->flush_log_dirty)
4707 		kvm_x86_ops->flush_log_dirty(kvm);
4708 
4709 	r = kvm_get_dirty_log_protect(kvm, log, &flush);
4710 
4711 	/*
4712 	 * All the TLBs can be flushed out of mmu lock, see the comments in
4713 	 * kvm_mmu_slot_remove_write_access().
4714 	 */
4715 	lockdep_assert_held(&kvm->slots_lock);
4716 	if (flush)
4717 		kvm_flush_remote_tlbs(kvm);
4718 
4719 	mutex_unlock(&kvm->slots_lock);
4720 	return r;
4721 }
4722 
4723 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
4724 {
4725 	bool flush = false;
4726 	int r;
4727 
4728 	mutex_lock(&kvm->slots_lock);
4729 
4730 	/*
4731 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4732 	 */
4733 	if (kvm_x86_ops->flush_log_dirty)
4734 		kvm_x86_ops->flush_log_dirty(kvm);
4735 
4736 	r = kvm_clear_dirty_log_protect(kvm, log, &flush);
4737 
4738 	/*
4739 	 * All the TLBs can be flushed out of mmu lock, see the comments in
4740 	 * kvm_mmu_slot_remove_write_access().
4741 	 */
4742 	lockdep_assert_held(&kvm->slots_lock);
4743 	if (flush)
4744 		kvm_flush_remote_tlbs(kvm);
4745 
4746 	mutex_unlock(&kvm->slots_lock);
4747 	return r;
4748 }
4749 
4750 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4751 			bool line_status)
4752 {
4753 	if (!irqchip_in_kernel(kvm))
4754 		return -ENXIO;
4755 
4756 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4757 					irq_event->irq, irq_event->level,
4758 					line_status);
4759 	return 0;
4760 }
4761 
4762 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4763 			    struct kvm_enable_cap *cap)
4764 {
4765 	int r;
4766 
4767 	if (cap->flags)
4768 		return -EINVAL;
4769 
4770 	switch (cap->cap) {
4771 	case KVM_CAP_DISABLE_QUIRKS:
4772 		kvm->arch.disabled_quirks = cap->args[0];
4773 		r = 0;
4774 		break;
4775 	case KVM_CAP_SPLIT_IRQCHIP: {
4776 		mutex_lock(&kvm->lock);
4777 		r = -EINVAL;
4778 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4779 			goto split_irqchip_unlock;
4780 		r = -EEXIST;
4781 		if (irqchip_in_kernel(kvm))
4782 			goto split_irqchip_unlock;
4783 		if (kvm->created_vcpus)
4784 			goto split_irqchip_unlock;
4785 		r = kvm_setup_empty_irq_routing(kvm);
4786 		if (r)
4787 			goto split_irqchip_unlock;
4788 		/* Pairs with irqchip_in_kernel. */
4789 		smp_wmb();
4790 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4791 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4792 		r = 0;
4793 split_irqchip_unlock:
4794 		mutex_unlock(&kvm->lock);
4795 		break;
4796 	}
4797 	case KVM_CAP_X2APIC_API:
4798 		r = -EINVAL;
4799 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4800 			break;
4801 
4802 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4803 			kvm->arch.x2apic_format = true;
4804 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4805 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
4806 
4807 		r = 0;
4808 		break;
4809 	case KVM_CAP_X86_DISABLE_EXITS:
4810 		r = -EINVAL;
4811 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4812 			break;
4813 
4814 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4815 			kvm_can_mwait_in_guest())
4816 			kvm->arch.mwait_in_guest = true;
4817 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4818 			kvm->arch.hlt_in_guest = true;
4819 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4820 			kvm->arch.pause_in_guest = true;
4821 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
4822 			kvm->arch.cstate_in_guest = true;
4823 		r = 0;
4824 		break;
4825 	case KVM_CAP_MSR_PLATFORM_INFO:
4826 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4827 		r = 0;
4828 		break;
4829 	case KVM_CAP_EXCEPTION_PAYLOAD:
4830 		kvm->arch.exception_payload_enabled = cap->args[0];
4831 		r = 0;
4832 		break;
4833 	default:
4834 		r = -EINVAL;
4835 		break;
4836 	}
4837 	return r;
4838 }
4839 
4840 long kvm_arch_vm_ioctl(struct file *filp,
4841 		       unsigned int ioctl, unsigned long arg)
4842 {
4843 	struct kvm *kvm = filp->private_data;
4844 	void __user *argp = (void __user *)arg;
4845 	int r = -ENOTTY;
4846 	/*
4847 	 * This union makes it completely explicit to gcc-3.x
4848 	 * that these two variables' stack usage should be
4849 	 * combined, not added together.
4850 	 */
4851 	union {
4852 		struct kvm_pit_state ps;
4853 		struct kvm_pit_state2 ps2;
4854 		struct kvm_pit_config pit_config;
4855 	} u;
4856 
4857 	switch (ioctl) {
4858 	case KVM_SET_TSS_ADDR:
4859 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4860 		break;
4861 	case KVM_SET_IDENTITY_MAP_ADDR: {
4862 		u64 ident_addr;
4863 
4864 		mutex_lock(&kvm->lock);
4865 		r = -EINVAL;
4866 		if (kvm->created_vcpus)
4867 			goto set_identity_unlock;
4868 		r = -EFAULT;
4869 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4870 			goto set_identity_unlock;
4871 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4872 set_identity_unlock:
4873 		mutex_unlock(&kvm->lock);
4874 		break;
4875 	}
4876 	case KVM_SET_NR_MMU_PAGES:
4877 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4878 		break;
4879 	case KVM_GET_NR_MMU_PAGES:
4880 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4881 		break;
4882 	case KVM_CREATE_IRQCHIP: {
4883 		mutex_lock(&kvm->lock);
4884 
4885 		r = -EEXIST;
4886 		if (irqchip_in_kernel(kvm))
4887 			goto create_irqchip_unlock;
4888 
4889 		r = -EINVAL;
4890 		if (kvm->created_vcpus)
4891 			goto create_irqchip_unlock;
4892 
4893 		r = kvm_pic_init(kvm);
4894 		if (r)
4895 			goto create_irqchip_unlock;
4896 
4897 		r = kvm_ioapic_init(kvm);
4898 		if (r) {
4899 			kvm_pic_destroy(kvm);
4900 			goto create_irqchip_unlock;
4901 		}
4902 
4903 		r = kvm_setup_default_irq_routing(kvm);
4904 		if (r) {
4905 			kvm_ioapic_destroy(kvm);
4906 			kvm_pic_destroy(kvm);
4907 			goto create_irqchip_unlock;
4908 		}
4909 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4910 		smp_wmb();
4911 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4912 	create_irqchip_unlock:
4913 		mutex_unlock(&kvm->lock);
4914 		break;
4915 	}
4916 	case KVM_CREATE_PIT:
4917 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4918 		goto create_pit;
4919 	case KVM_CREATE_PIT2:
4920 		r = -EFAULT;
4921 		if (copy_from_user(&u.pit_config, argp,
4922 				   sizeof(struct kvm_pit_config)))
4923 			goto out;
4924 	create_pit:
4925 		mutex_lock(&kvm->lock);
4926 		r = -EEXIST;
4927 		if (kvm->arch.vpit)
4928 			goto create_pit_unlock;
4929 		r = -ENOMEM;
4930 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4931 		if (kvm->arch.vpit)
4932 			r = 0;
4933 	create_pit_unlock:
4934 		mutex_unlock(&kvm->lock);
4935 		break;
4936 	case KVM_GET_IRQCHIP: {
4937 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4938 		struct kvm_irqchip *chip;
4939 
4940 		chip = memdup_user(argp, sizeof(*chip));
4941 		if (IS_ERR(chip)) {
4942 			r = PTR_ERR(chip);
4943 			goto out;
4944 		}
4945 
4946 		r = -ENXIO;
4947 		if (!irqchip_kernel(kvm))
4948 			goto get_irqchip_out;
4949 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4950 		if (r)
4951 			goto get_irqchip_out;
4952 		r = -EFAULT;
4953 		if (copy_to_user(argp, chip, sizeof(*chip)))
4954 			goto get_irqchip_out;
4955 		r = 0;
4956 	get_irqchip_out:
4957 		kfree(chip);
4958 		break;
4959 	}
4960 	case KVM_SET_IRQCHIP: {
4961 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4962 		struct kvm_irqchip *chip;
4963 
4964 		chip = memdup_user(argp, sizeof(*chip));
4965 		if (IS_ERR(chip)) {
4966 			r = PTR_ERR(chip);
4967 			goto out;
4968 		}
4969 
4970 		r = -ENXIO;
4971 		if (!irqchip_kernel(kvm))
4972 			goto set_irqchip_out;
4973 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4974 	set_irqchip_out:
4975 		kfree(chip);
4976 		break;
4977 	}
4978 	case KVM_GET_PIT: {
4979 		r = -EFAULT;
4980 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4981 			goto out;
4982 		r = -ENXIO;
4983 		if (!kvm->arch.vpit)
4984 			goto out;
4985 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4986 		if (r)
4987 			goto out;
4988 		r = -EFAULT;
4989 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4990 			goto out;
4991 		r = 0;
4992 		break;
4993 	}
4994 	case KVM_SET_PIT: {
4995 		r = -EFAULT;
4996 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
4997 			goto out;
4998 		r = -ENXIO;
4999 		if (!kvm->arch.vpit)
5000 			goto out;
5001 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5002 		break;
5003 	}
5004 	case KVM_GET_PIT2: {
5005 		r = -ENXIO;
5006 		if (!kvm->arch.vpit)
5007 			goto out;
5008 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5009 		if (r)
5010 			goto out;
5011 		r = -EFAULT;
5012 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5013 			goto out;
5014 		r = 0;
5015 		break;
5016 	}
5017 	case KVM_SET_PIT2: {
5018 		r = -EFAULT;
5019 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5020 			goto out;
5021 		r = -ENXIO;
5022 		if (!kvm->arch.vpit)
5023 			goto out;
5024 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5025 		break;
5026 	}
5027 	case KVM_REINJECT_CONTROL: {
5028 		struct kvm_reinject_control control;
5029 		r =  -EFAULT;
5030 		if (copy_from_user(&control, argp, sizeof(control)))
5031 			goto out;
5032 		r = kvm_vm_ioctl_reinject(kvm, &control);
5033 		break;
5034 	}
5035 	case KVM_SET_BOOT_CPU_ID:
5036 		r = 0;
5037 		mutex_lock(&kvm->lock);
5038 		if (kvm->created_vcpus)
5039 			r = -EBUSY;
5040 		else
5041 			kvm->arch.bsp_vcpu_id = arg;
5042 		mutex_unlock(&kvm->lock);
5043 		break;
5044 	case KVM_XEN_HVM_CONFIG: {
5045 		struct kvm_xen_hvm_config xhc;
5046 		r = -EFAULT;
5047 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
5048 			goto out;
5049 		r = -EINVAL;
5050 		if (xhc.flags)
5051 			goto out;
5052 		memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5053 		r = 0;
5054 		break;
5055 	}
5056 	case KVM_SET_CLOCK: {
5057 		struct kvm_clock_data user_ns;
5058 		u64 now_ns;
5059 
5060 		r = -EFAULT;
5061 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5062 			goto out;
5063 
5064 		r = -EINVAL;
5065 		if (user_ns.flags)
5066 			goto out;
5067 
5068 		r = 0;
5069 		/*
5070 		 * TODO: userspace has to take care of races with VCPU_RUN, so
5071 		 * kvm_gen_update_masterclock() can be cut down to locked
5072 		 * pvclock_update_vm_gtod_copy().
5073 		 */
5074 		kvm_gen_update_masterclock(kvm);
5075 		now_ns = get_kvmclock_ns(kvm);
5076 		kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5077 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5078 		break;
5079 	}
5080 	case KVM_GET_CLOCK: {
5081 		struct kvm_clock_data user_ns;
5082 		u64 now_ns;
5083 
5084 		now_ns = get_kvmclock_ns(kvm);
5085 		user_ns.clock = now_ns;
5086 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5087 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5088 
5089 		r = -EFAULT;
5090 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5091 			goto out;
5092 		r = 0;
5093 		break;
5094 	}
5095 	case KVM_MEMORY_ENCRYPT_OP: {
5096 		r = -ENOTTY;
5097 		if (kvm_x86_ops->mem_enc_op)
5098 			r = kvm_x86_ops->mem_enc_op(kvm, argp);
5099 		break;
5100 	}
5101 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
5102 		struct kvm_enc_region region;
5103 
5104 		r = -EFAULT;
5105 		if (copy_from_user(&region, argp, sizeof(region)))
5106 			goto out;
5107 
5108 		r = -ENOTTY;
5109 		if (kvm_x86_ops->mem_enc_reg_region)
5110 			r = kvm_x86_ops->mem_enc_reg_region(kvm, &region);
5111 		break;
5112 	}
5113 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5114 		struct kvm_enc_region region;
5115 
5116 		r = -EFAULT;
5117 		if (copy_from_user(&region, argp, sizeof(region)))
5118 			goto out;
5119 
5120 		r = -ENOTTY;
5121 		if (kvm_x86_ops->mem_enc_unreg_region)
5122 			r = kvm_x86_ops->mem_enc_unreg_region(kvm, &region);
5123 		break;
5124 	}
5125 	case KVM_HYPERV_EVENTFD: {
5126 		struct kvm_hyperv_eventfd hvevfd;
5127 
5128 		r = -EFAULT;
5129 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5130 			goto out;
5131 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5132 		break;
5133 	}
5134 	case KVM_SET_PMU_EVENT_FILTER:
5135 		r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5136 		break;
5137 	default:
5138 		r = -ENOTTY;
5139 	}
5140 out:
5141 	return r;
5142 }
5143 
5144 static void kvm_init_msr_list(void)
5145 {
5146 	struct x86_pmu_capability x86_pmu;
5147 	u32 dummy[2];
5148 	unsigned i;
5149 
5150 	BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5151 			 "Please update the fixed PMCs in msrs_to_saved_all[]");
5152 
5153 	perf_get_x86_pmu_capability(&x86_pmu);
5154 
5155 	num_msrs_to_save = 0;
5156 	num_emulated_msrs = 0;
5157 	num_msr_based_features = 0;
5158 
5159 	for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5160 		if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5161 			continue;
5162 
5163 		/*
5164 		 * Even MSRs that are valid in the host may not be exposed
5165 		 * to the guests in some cases.
5166 		 */
5167 		switch (msrs_to_save_all[i]) {
5168 		case MSR_IA32_BNDCFGS:
5169 			if (!kvm_mpx_supported())
5170 				continue;
5171 			break;
5172 		case MSR_TSC_AUX:
5173 			if (!kvm_x86_ops->rdtscp_supported())
5174 				continue;
5175 			break;
5176 		case MSR_IA32_RTIT_CTL:
5177 		case MSR_IA32_RTIT_STATUS:
5178 			if (!kvm_x86_ops->pt_supported())
5179 				continue;
5180 			break;
5181 		case MSR_IA32_RTIT_CR3_MATCH:
5182 			if (!kvm_x86_ops->pt_supported() ||
5183 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5184 				continue;
5185 			break;
5186 		case MSR_IA32_RTIT_OUTPUT_BASE:
5187 		case MSR_IA32_RTIT_OUTPUT_MASK:
5188 			if (!kvm_x86_ops->pt_supported() ||
5189 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5190 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5191 				continue;
5192 			break;
5193 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
5194 			if (!kvm_x86_ops->pt_supported() ||
5195 				msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5196 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5197 				continue;
5198 			break;
5199 		case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5200 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5201 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5202 				continue;
5203 			break;
5204 		case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5205 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5206 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5207 				continue;
5208 		}
5209 		default:
5210 			break;
5211 		}
5212 
5213 		msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5214 	}
5215 
5216 	for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5217 		if (!kvm_x86_ops->has_emulated_msr(emulated_msrs_all[i]))
5218 			continue;
5219 
5220 		emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5221 	}
5222 
5223 	for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5224 		struct kvm_msr_entry msr;
5225 
5226 		msr.index = msr_based_features_all[i];
5227 		if (kvm_get_msr_feature(&msr))
5228 			continue;
5229 
5230 		msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
5231 	}
5232 }
5233 
5234 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5235 			   const void *v)
5236 {
5237 	int handled = 0;
5238 	int n;
5239 
5240 	do {
5241 		n = min(len, 8);
5242 		if (!(lapic_in_kernel(vcpu) &&
5243 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5244 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5245 			break;
5246 		handled += n;
5247 		addr += n;
5248 		len -= n;
5249 		v += n;
5250 	} while (len);
5251 
5252 	return handled;
5253 }
5254 
5255 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5256 {
5257 	int handled = 0;
5258 	int n;
5259 
5260 	do {
5261 		n = min(len, 8);
5262 		if (!(lapic_in_kernel(vcpu) &&
5263 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5264 					 addr, n, v))
5265 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5266 			break;
5267 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5268 		handled += n;
5269 		addr += n;
5270 		len -= n;
5271 		v += n;
5272 	} while (len);
5273 
5274 	return handled;
5275 }
5276 
5277 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5278 			struct kvm_segment *var, int seg)
5279 {
5280 	kvm_x86_ops->set_segment(vcpu, var, seg);
5281 }
5282 
5283 void kvm_get_segment(struct kvm_vcpu *vcpu,
5284 		     struct kvm_segment *var, int seg)
5285 {
5286 	kvm_x86_ops->get_segment(vcpu, var, seg);
5287 }
5288 
5289 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5290 			   struct x86_exception *exception)
5291 {
5292 	gpa_t t_gpa;
5293 
5294 	BUG_ON(!mmu_is_nested(vcpu));
5295 
5296 	/* NPT walks are always user-walks */
5297 	access |= PFERR_USER_MASK;
5298 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5299 
5300 	return t_gpa;
5301 }
5302 
5303 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5304 			      struct x86_exception *exception)
5305 {
5306 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5307 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5308 }
5309 
5310  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5311 				struct x86_exception *exception)
5312 {
5313 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5314 	access |= PFERR_FETCH_MASK;
5315 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5316 }
5317 
5318 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5319 			       struct x86_exception *exception)
5320 {
5321 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5322 	access |= PFERR_WRITE_MASK;
5323 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5324 }
5325 
5326 /* uses this to access any guest's mapped memory without checking CPL */
5327 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5328 				struct x86_exception *exception)
5329 {
5330 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5331 }
5332 
5333 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5334 				      struct kvm_vcpu *vcpu, u32 access,
5335 				      struct x86_exception *exception)
5336 {
5337 	void *data = val;
5338 	int r = X86EMUL_CONTINUE;
5339 
5340 	while (bytes) {
5341 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5342 							    exception);
5343 		unsigned offset = addr & (PAGE_SIZE-1);
5344 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5345 		int ret;
5346 
5347 		if (gpa == UNMAPPED_GVA)
5348 			return X86EMUL_PROPAGATE_FAULT;
5349 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5350 					       offset, toread);
5351 		if (ret < 0) {
5352 			r = X86EMUL_IO_NEEDED;
5353 			goto out;
5354 		}
5355 
5356 		bytes -= toread;
5357 		data += toread;
5358 		addr += toread;
5359 	}
5360 out:
5361 	return r;
5362 }
5363 
5364 /* used for instruction fetching */
5365 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5366 				gva_t addr, void *val, unsigned int bytes,
5367 				struct x86_exception *exception)
5368 {
5369 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5370 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5371 	unsigned offset;
5372 	int ret;
5373 
5374 	/* Inline kvm_read_guest_virt_helper for speed.  */
5375 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5376 						    exception);
5377 	if (unlikely(gpa == UNMAPPED_GVA))
5378 		return X86EMUL_PROPAGATE_FAULT;
5379 
5380 	offset = addr & (PAGE_SIZE-1);
5381 	if (WARN_ON(offset + bytes > PAGE_SIZE))
5382 		bytes = (unsigned)PAGE_SIZE - offset;
5383 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5384 				       offset, bytes);
5385 	if (unlikely(ret < 0))
5386 		return X86EMUL_IO_NEEDED;
5387 
5388 	return X86EMUL_CONTINUE;
5389 }
5390 
5391 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5392 			       gva_t addr, void *val, unsigned int bytes,
5393 			       struct x86_exception *exception)
5394 {
5395 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5396 
5397 	/*
5398 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5399 	 * is returned, but our callers are not ready for that and they blindly
5400 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
5401 	 * uninitialized kernel stack memory into cr2 and error code.
5402 	 */
5403 	memset(exception, 0, sizeof(*exception));
5404 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5405 					  exception);
5406 }
5407 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5408 
5409 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5410 			     gva_t addr, void *val, unsigned int bytes,
5411 			     struct x86_exception *exception, bool system)
5412 {
5413 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5414 	u32 access = 0;
5415 
5416 	if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5417 		access |= PFERR_USER_MASK;
5418 
5419 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5420 }
5421 
5422 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5423 		unsigned long addr, void *val, unsigned int bytes)
5424 {
5425 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5426 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5427 
5428 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5429 }
5430 
5431 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5432 				      struct kvm_vcpu *vcpu, u32 access,
5433 				      struct x86_exception *exception)
5434 {
5435 	void *data = val;
5436 	int r = X86EMUL_CONTINUE;
5437 
5438 	while (bytes) {
5439 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5440 							     access,
5441 							     exception);
5442 		unsigned offset = addr & (PAGE_SIZE-1);
5443 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5444 		int ret;
5445 
5446 		if (gpa == UNMAPPED_GVA)
5447 			return X86EMUL_PROPAGATE_FAULT;
5448 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5449 		if (ret < 0) {
5450 			r = X86EMUL_IO_NEEDED;
5451 			goto out;
5452 		}
5453 
5454 		bytes -= towrite;
5455 		data += towrite;
5456 		addr += towrite;
5457 	}
5458 out:
5459 	return r;
5460 }
5461 
5462 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5463 			      unsigned int bytes, struct x86_exception *exception,
5464 			      bool system)
5465 {
5466 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5467 	u32 access = PFERR_WRITE_MASK;
5468 
5469 	if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5470 		access |= PFERR_USER_MASK;
5471 
5472 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5473 					   access, exception);
5474 }
5475 
5476 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5477 				unsigned int bytes, struct x86_exception *exception)
5478 {
5479 	/* kvm_write_guest_virt_system can pull in tons of pages. */
5480 	vcpu->arch.l1tf_flush_l1d = true;
5481 
5482 	/*
5483 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5484 	 * is returned, but our callers are not ready for that and they blindly
5485 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
5486 	 * uninitialized kernel stack memory into cr2 and error code.
5487 	 */
5488 	memset(exception, 0, sizeof(*exception));
5489 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5490 					   PFERR_WRITE_MASK, exception);
5491 }
5492 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5493 
5494 int handle_ud(struct kvm_vcpu *vcpu)
5495 {
5496 	static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
5497 	int emul_type = EMULTYPE_TRAP_UD;
5498 	char sig[5]; /* ud2; .ascii "kvm" */
5499 	struct x86_exception e;
5500 
5501 	if (force_emulation_prefix &&
5502 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5503 				sig, sizeof(sig), &e) == 0 &&
5504 	    memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
5505 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5506 		emul_type = EMULTYPE_TRAP_UD_FORCED;
5507 	}
5508 
5509 	return kvm_emulate_instruction(vcpu, emul_type);
5510 }
5511 EXPORT_SYMBOL_GPL(handle_ud);
5512 
5513 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5514 			    gpa_t gpa, bool write)
5515 {
5516 	/* For APIC access vmexit */
5517 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5518 		return 1;
5519 
5520 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5521 		trace_vcpu_match_mmio(gva, gpa, write, true);
5522 		return 1;
5523 	}
5524 
5525 	return 0;
5526 }
5527 
5528 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5529 				gpa_t *gpa, struct x86_exception *exception,
5530 				bool write)
5531 {
5532 	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5533 		| (write ? PFERR_WRITE_MASK : 0);
5534 
5535 	/*
5536 	 * currently PKRU is only applied to ept enabled guest so
5537 	 * there is no pkey in EPT page table for L1 guest or EPT
5538 	 * shadow page table for L2 guest.
5539 	 */
5540 	if (vcpu_match_mmio_gva(vcpu, gva)
5541 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5542 				 vcpu->arch.mmio_access, 0, access)) {
5543 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5544 					(gva & (PAGE_SIZE - 1));
5545 		trace_vcpu_match_mmio(gva, *gpa, write, false);
5546 		return 1;
5547 	}
5548 
5549 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5550 
5551 	if (*gpa == UNMAPPED_GVA)
5552 		return -1;
5553 
5554 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5555 }
5556 
5557 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5558 			const void *val, int bytes)
5559 {
5560 	int ret;
5561 
5562 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5563 	if (ret < 0)
5564 		return 0;
5565 	kvm_page_track_write(vcpu, gpa, val, bytes);
5566 	return 1;
5567 }
5568 
5569 struct read_write_emulator_ops {
5570 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5571 				  int bytes);
5572 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5573 				  void *val, int bytes);
5574 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5575 			       int bytes, void *val);
5576 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5577 				    void *val, int bytes);
5578 	bool write;
5579 };
5580 
5581 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5582 {
5583 	if (vcpu->mmio_read_completed) {
5584 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5585 			       vcpu->mmio_fragments[0].gpa, val);
5586 		vcpu->mmio_read_completed = 0;
5587 		return 1;
5588 	}
5589 
5590 	return 0;
5591 }
5592 
5593 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5594 			void *val, int bytes)
5595 {
5596 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5597 }
5598 
5599 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5600 			 void *val, int bytes)
5601 {
5602 	return emulator_write_phys(vcpu, gpa, val, bytes);
5603 }
5604 
5605 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5606 {
5607 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5608 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
5609 }
5610 
5611 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5612 			  void *val, int bytes)
5613 {
5614 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5615 	return X86EMUL_IO_NEEDED;
5616 }
5617 
5618 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5619 			   void *val, int bytes)
5620 {
5621 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5622 
5623 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5624 	return X86EMUL_CONTINUE;
5625 }
5626 
5627 static const struct read_write_emulator_ops read_emultor = {
5628 	.read_write_prepare = read_prepare,
5629 	.read_write_emulate = read_emulate,
5630 	.read_write_mmio = vcpu_mmio_read,
5631 	.read_write_exit_mmio = read_exit_mmio,
5632 };
5633 
5634 static const struct read_write_emulator_ops write_emultor = {
5635 	.read_write_emulate = write_emulate,
5636 	.read_write_mmio = write_mmio,
5637 	.read_write_exit_mmio = write_exit_mmio,
5638 	.write = true,
5639 };
5640 
5641 static int emulator_read_write_onepage(unsigned long addr, void *val,
5642 				       unsigned int bytes,
5643 				       struct x86_exception *exception,
5644 				       struct kvm_vcpu *vcpu,
5645 				       const struct read_write_emulator_ops *ops)
5646 {
5647 	gpa_t gpa;
5648 	int handled, ret;
5649 	bool write = ops->write;
5650 	struct kvm_mmio_fragment *frag;
5651 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5652 
5653 	/*
5654 	 * If the exit was due to a NPF we may already have a GPA.
5655 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5656 	 * Note, this cannot be used on string operations since string
5657 	 * operation using rep will only have the initial GPA from the NPF
5658 	 * occurred.
5659 	 */
5660 	if (vcpu->arch.gpa_available &&
5661 	    emulator_can_use_gpa(ctxt) &&
5662 	    (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5663 		gpa = vcpu->arch.gpa_val;
5664 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5665 	} else {
5666 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5667 		if (ret < 0)
5668 			return X86EMUL_PROPAGATE_FAULT;
5669 	}
5670 
5671 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5672 		return X86EMUL_CONTINUE;
5673 
5674 	/*
5675 	 * Is this MMIO handled locally?
5676 	 */
5677 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5678 	if (handled == bytes)
5679 		return X86EMUL_CONTINUE;
5680 
5681 	gpa += handled;
5682 	bytes -= handled;
5683 	val += handled;
5684 
5685 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5686 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5687 	frag->gpa = gpa;
5688 	frag->data = val;
5689 	frag->len = bytes;
5690 	return X86EMUL_CONTINUE;
5691 }
5692 
5693 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5694 			unsigned long addr,
5695 			void *val, unsigned int bytes,
5696 			struct x86_exception *exception,
5697 			const struct read_write_emulator_ops *ops)
5698 {
5699 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5700 	gpa_t gpa;
5701 	int rc;
5702 
5703 	if (ops->read_write_prepare &&
5704 		  ops->read_write_prepare(vcpu, val, bytes))
5705 		return X86EMUL_CONTINUE;
5706 
5707 	vcpu->mmio_nr_fragments = 0;
5708 
5709 	/* Crossing a page boundary? */
5710 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5711 		int now;
5712 
5713 		now = -addr & ~PAGE_MASK;
5714 		rc = emulator_read_write_onepage(addr, val, now, exception,
5715 						 vcpu, ops);
5716 
5717 		if (rc != X86EMUL_CONTINUE)
5718 			return rc;
5719 		addr += now;
5720 		if (ctxt->mode != X86EMUL_MODE_PROT64)
5721 			addr = (u32)addr;
5722 		val += now;
5723 		bytes -= now;
5724 	}
5725 
5726 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
5727 					 vcpu, ops);
5728 	if (rc != X86EMUL_CONTINUE)
5729 		return rc;
5730 
5731 	if (!vcpu->mmio_nr_fragments)
5732 		return rc;
5733 
5734 	gpa = vcpu->mmio_fragments[0].gpa;
5735 
5736 	vcpu->mmio_needed = 1;
5737 	vcpu->mmio_cur_fragment = 0;
5738 
5739 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5740 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5741 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
5742 	vcpu->run->mmio.phys_addr = gpa;
5743 
5744 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5745 }
5746 
5747 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5748 				  unsigned long addr,
5749 				  void *val,
5750 				  unsigned int bytes,
5751 				  struct x86_exception *exception)
5752 {
5753 	return emulator_read_write(ctxt, addr, val, bytes,
5754 				   exception, &read_emultor);
5755 }
5756 
5757 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5758 			    unsigned long addr,
5759 			    const void *val,
5760 			    unsigned int bytes,
5761 			    struct x86_exception *exception)
5762 {
5763 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
5764 				   exception, &write_emultor);
5765 }
5766 
5767 #define CMPXCHG_TYPE(t, ptr, old, new) \
5768 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5769 
5770 #ifdef CONFIG_X86_64
5771 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5772 #else
5773 #  define CMPXCHG64(ptr, old, new) \
5774 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5775 #endif
5776 
5777 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5778 				     unsigned long addr,
5779 				     const void *old,
5780 				     const void *new,
5781 				     unsigned int bytes,
5782 				     struct x86_exception *exception)
5783 {
5784 	struct kvm_host_map map;
5785 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5786 	gpa_t gpa;
5787 	char *kaddr;
5788 	bool exchanged;
5789 
5790 	/* guests cmpxchg8b have to be emulated atomically */
5791 	if (bytes > 8 || (bytes & (bytes - 1)))
5792 		goto emul_write;
5793 
5794 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5795 
5796 	if (gpa == UNMAPPED_GVA ||
5797 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5798 		goto emul_write;
5799 
5800 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5801 		goto emul_write;
5802 
5803 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
5804 		goto emul_write;
5805 
5806 	kaddr = map.hva + offset_in_page(gpa);
5807 
5808 	switch (bytes) {
5809 	case 1:
5810 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5811 		break;
5812 	case 2:
5813 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5814 		break;
5815 	case 4:
5816 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5817 		break;
5818 	case 8:
5819 		exchanged = CMPXCHG64(kaddr, old, new);
5820 		break;
5821 	default:
5822 		BUG();
5823 	}
5824 
5825 	kvm_vcpu_unmap(vcpu, &map, true);
5826 
5827 	if (!exchanged)
5828 		return X86EMUL_CMPXCHG_FAILED;
5829 
5830 	kvm_page_track_write(vcpu, gpa, new, bytes);
5831 
5832 	return X86EMUL_CONTINUE;
5833 
5834 emul_write:
5835 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5836 
5837 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5838 }
5839 
5840 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5841 {
5842 	int r = 0, i;
5843 
5844 	for (i = 0; i < vcpu->arch.pio.count; i++) {
5845 		if (vcpu->arch.pio.in)
5846 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5847 					    vcpu->arch.pio.size, pd);
5848 		else
5849 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5850 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
5851 					     pd);
5852 		if (r)
5853 			break;
5854 		pd += vcpu->arch.pio.size;
5855 	}
5856 	return r;
5857 }
5858 
5859 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5860 			       unsigned short port, void *val,
5861 			       unsigned int count, bool in)
5862 {
5863 	vcpu->arch.pio.port = port;
5864 	vcpu->arch.pio.in = in;
5865 	vcpu->arch.pio.count  = count;
5866 	vcpu->arch.pio.size = size;
5867 
5868 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5869 		vcpu->arch.pio.count = 0;
5870 		return 1;
5871 	}
5872 
5873 	vcpu->run->exit_reason = KVM_EXIT_IO;
5874 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5875 	vcpu->run->io.size = size;
5876 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5877 	vcpu->run->io.count = count;
5878 	vcpu->run->io.port = port;
5879 
5880 	return 0;
5881 }
5882 
5883 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5884 				    int size, unsigned short port, void *val,
5885 				    unsigned int count)
5886 {
5887 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5888 	int ret;
5889 
5890 	if (vcpu->arch.pio.count)
5891 		goto data_avail;
5892 
5893 	memset(vcpu->arch.pio_data, 0, size * count);
5894 
5895 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5896 	if (ret) {
5897 data_avail:
5898 		memcpy(val, vcpu->arch.pio_data, size * count);
5899 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5900 		vcpu->arch.pio.count = 0;
5901 		return 1;
5902 	}
5903 
5904 	return 0;
5905 }
5906 
5907 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5908 				     int size, unsigned short port,
5909 				     const void *val, unsigned int count)
5910 {
5911 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5912 
5913 	memcpy(vcpu->arch.pio_data, val, size * count);
5914 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5915 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5916 }
5917 
5918 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5919 {
5920 	return kvm_x86_ops->get_segment_base(vcpu, seg);
5921 }
5922 
5923 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5924 {
5925 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5926 }
5927 
5928 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5929 {
5930 	if (!need_emulate_wbinvd(vcpu))
5931 		return X86EMUL_CONTINUE;
5932 
5933 	if (kvm_x86_ops->has_wbinvd_exit()) {
5934 		int cpu = get_cpu();
5935 
5936 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5937 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5938 				wbinvd_ipi, NULL, 1);
5939 		put_cpu();
5940 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5941 	} else
5942 		wbinvd();
5943 	return X86EMUL_CONTINUE;
5944 }
5945 
5946 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5947 {
5948 	kvm_emulate_wbinvd_noskip(vcpu);
5949 	return kvm_skip_emulated_instruction(vcpu);
5950 }
5951 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5952 
5953 
5954 
5955 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5956 {
5957 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5958 }
5959 
5960 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5961 			   unsigned long *dest)
5962 {
5963 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5964 }
5965 
5966 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5967 			   unsigned long value)
5968 {
5969 
5970 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5971 }
5972 
5973 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5974 {
5975 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5976 }
5977 
5978 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5979 {
5980 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5981 	unsigned long value;
5982 
5983 	switch (cr) {
5984 	case 0:
5985 		value = kvm_read_cr0(vcpu);
5986 		break;
5987 	case 2:
5988 		value = vcpu->arch.cr2;
5989 		break;
5990 	case 3:
5991 		value = kvm_read_cr3(vcpu);
5992 		break;
5993 	case 4:
5994 		value = kvm_read_cr4(vcpu);
5995 		break;
5996 	case 8:
5997 		value = kvm_get_cr8(vcpu);
5998 		break;
5999 	default:
6000 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
6001 		return 0;
6002 	}
6003 
6004 	return value;
6005 }
6006 
6007 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6008 {
6009 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6010 	int res = 0;
6011 
6012 	switch (cr) {
6013 	case 0:
6014 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6015 		break;
6016 	case 2:
6017 		vcpu->arch.cr2 = val;
6018 		break;
6019 	case 3:
6020 		res = kvm_set_cr3(vcpu, val);
6021 		break;
6022 	case 4:
6023 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6024 		break;
6025 	case 8:
6026 		res = kvm_set_cr8(vcpu, val);
6027 		break;
6028 	default:
6029 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
6030 		res = -1;
6031 	}
6032 
6033 	return res;
6034 }
6035 
6036 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6037 {
6038 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
6039 }
6040 
6041 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6042 {
6043 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
6044 }
6045 
6046 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6047 {
6048 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
6049 }
6050 
6051 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6052 {
6053 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
6054 }
6055 
6056 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6057 {
6058 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
6059 }
6060 
6061 static unsigned long emulator_get_cached_segment_base(
6062 	struct x86_emulate_ctxt *ctxt, int seg)
6063 {
6064 	return get_segment_base(emul_to_vcpu(ctxt), seg);
6065 }
6066 
6067 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6068 				 struct desc_struct *desc, u32 *base3,
6069 				 int seg)
6070 {
6071 	struct kvm_segment var;
6072 
6073 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6074 	*selector = var.selector;
6075 
6076 	if (var.unusable) {
6077 		memset(desc, 0, sizeof(*desc));
6078 		if (base3)
6079 			*base3 = 0;
6080 		return false;
6081 	}
6082 
6083 	if (var.g)
6084 		var.limit >>= 12;
6085 	set_desc_limit(desc, var.limit);
6086 	set_desc_base(desc, (unsigned long)var.base);
6087 #ifdef CONFIG_X86_64
6088 	if (base3)
6089 		*base3 = var.base >> 32;
6090 #endif
6091 	desc->type = var.type;
6092 	desc->s = var.s;
6093 	desc->dpl = var.dpl;
6094 	desc->p = var.present;
6095 	desc->avl = var.avl;
6096 	desc->l = var.l;
6097 	desc->d = var.db;
6098 	desc->g = var.g;
6099 
6100 	return true;
6101 }
6102 
6103 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6104 				 struct desc_struct *desc, u32 base3,
6105 				 int seg)
6106 {
6107 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6108 	struct kvm_segment var;
6109 
6110 	var.selector = selector;
6111 	var.base = get_desc_base(desc);
6112 #ifdef CONFIG_X86_64
6113 	var.base |= ((u64)base3) << 32;
6114 #endif
6115 	var.limit = get_desc_limit(desc);
6116 	if (desc->g)
6117 		var.limit = (var.limit << 12) | 0xfff;
6118 	var.type = desc->type;
6119 	var.dpl = desc->dpl;
6120 	var.db = desc->d;
6121 	var.s = desc->s;
6122 	var.l = desc->l;
6123 	var.g = desc->g;
6124 	var.avl = desc->avl;
6125 	var.present = desc->p;
6126 	var.unusable = !var.present;
6127 	var.padding = 0;
6128 
6129 	kvm_set_segment(vcpu, &var, seg);
6130 	return;
6131 }
6132 
6133 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6134 			    u32 msr_index, u64 *pdata)
6135 {
6136 	return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
6137 }
6138 
6139 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6140 			    u32 msr_index, u64 data)
6141 {
6142 	return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
6143 }
6144 
6145 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6146 {
6147 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6148 
6149 	return vcpu->arch.smbase;
6150 }
6151 
6152 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6153 {
6154 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6155 
6156 	vcpu->arch.smbase = smbase;
6157 }
6158 
6159 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6160 			      u32 pmc)
6161 {
6162 	return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6163 }
6164 
6165 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6166 			     u32 pmc, u64 *pdata)
6167 {
6168 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6169 }
6170 
6171 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6172 {
6173 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
6174 }
6175 
6176 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6177 			      struct x86_instruction_info *info,
6178 			      enum x86_intercept_stage stage)
6179 {
6180 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
6181 }
6182 
6183 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6184 			u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
6185 {
6186 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
6187 }
6188 
6189 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6190 {
6191 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
6192 }
6193 
6194 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6195 {
6196 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6197 }
6198 
6199 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6200 {
6201 	kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
6202 }
6203 
6204 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6205 {
6206 	return emul_to_vcpu(ctxt)->arch.hflags;
6207 }
6208 
6209 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6210 {
6211 	emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6212 }
6213 
6214 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6215 				  const char *smstate)
6216 {
6217 	return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6218 }
6219 
6220 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6221 {
6222 	kvm_smm_changed(emul_to_vcpu(ctxt));
6223 }
6224 
6225 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6226 {
6227 	return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6228 }
6229 
6230 static const struct x86_emulate_ops emulate_ops = {
6231 	.read_gpr            = emulator_read_gpr,
6232 	.write_gpr           = emulator_write_gpr,
6233 	.read_std            = emulator_read_std,
6234 	.write_std           = emulator_write_std,
6235 	.read_phys           = kvm_read_guest_phys_system,
6236 	.fetch               = kvm_fetch_guest_virt,
6237 	.read_emulated       = emulator_read_emulated,
6238 	.write_emulated      = emulator_write_emulated,
6239 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
6240 	.invlpg              = emulator_invlpg,
6241 	.pio_in_emulated     = emulator_pio_in_emulated,
6242 	.pio_out_emulated    = emulator_pio_out_emulated,
6243 	.get_segment         = emulator_get_segment,
6244 	.set_segment         = emulator_set_segment,
6245 	.get_cached_segment_base = emulator_get_cached_segment_base,
6246 	.get_gdt             = emulator_get_gdt,
6247 	.get_idt	     = emulator_get_idt,
6248 	.set_gdt             = emulator_set_gdt,
6249 	.set_idt	     = emulator_set_idt,
6250 	.get_cr              = emulator_get_cr,
6251 	.set_cr              = emulator_set_cr,
6252 	.cpl                 = emulator_get_cpl,
6253 	.get_dr              = emulator_get_dr,
6254 	.set_dr              = emulator_set_dr,
6255 	.get_smbase          = emulator_get_smbase,
6256 	.set_smbase          = emulator_set_smbase,
6257 	.set_msr             = emulator_set_msr,
6258 	.get_msr             = emulator_get_msr,
6259 	.check_pmc	     = emulator_check_pmc,
6260 	.read_pmc            = emulator_read_pmc,
6261 	.halt                = emulator_halt,
6262 	.wbinvd              = emulator_wbinvd,
6263 	.fix_hypercall       = emulator_fix_hypercall,
6264 	.intercept           = emulator_intercept,
6265 	.get_cpuid           = emulator_get_cpuid,
6266 	.set_nmi_mask        = emulator_set_nmi_mask,
6267 	.get_hflags          = emulator_get_hflags,
6268 	.set_hflags          = emulator_set_hflags,
6269 	.pre_leave_smm       = emulator_pre_leave_smm,
6270 	.post_leave_smm      = emulator_post_leave_smm,
6271 	.set_xcr             = emulator_set_xcr,
6272 };
6273 
6274 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6275 {
6276 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
6277 	/*
6278 	 * an sti; sti; sequence only disable interrupts for the first
6279 	 * instruction. So, if the last instruction, be it emulated or
6280 	 * not, left the system with the INT_STI flag enabled, it
6281 	 * means that the last instruction is an sti. We should not
6282 	 * leave the flag on in this case. The same goes for mov ss
6283 	 */
6284 	if (int_shadow & mask)
6285 		mask = 0;
6286 	if (unlikely(int_shadow || mask)) {
6287 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6288 		if (!mask)
6289 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6290 	}
6291 }
6292 
6293 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6294 {
6295 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6296 	if (ctxt->exception.vector == PF_VECTOR)
6297 		return kvm_propagate_fault(vcpu, &ctxt->exception);
6298 
6299 	if (ctxt->exception.error_code_valid)
6300 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6301 				      ctxt->exception.error_code);
6302 	else
6303 		kvm_queue_exception(vcpu, ctxt->exception.vector);
6304 	return false;
6305 }
6306 
6307 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6308 {
6309 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6310 	int cs_db, cs_l;
6311 
6312 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6313 
6314 	ctxt->eflags = kvm_get_rflags(vcpu);
6315 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6316 
6317 	ctxt->eip = kvm_rip_read(vcpu);
6318 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
6319 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
6320 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
6321 		     cs_db				? X86EMUL_MODE_PROT32 :
6322 							  X86EMUL_MODE_PROT16;
6323 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6324 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6325 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6326 
6327 	init_decode_cache(ctxt);
6328 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6329 }
6330 
6331 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6332 {
6333 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6334 	int ret;
6335 
6336 	init_emulate_ctxt(vcpu);
6337 
6338 	ctxt->op_bytes = 2;
6339 	ctxt->ad_bytes = 2;
6340 	ctxt->_eip = ctxt->eip + inc_eip;
6341 	ret = emulate_int_real(ctxt, irq);
6342 
6343 	if (ret != X86EMUL_CONTINUE) {
6344 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6345 	} else {
6346 		ctxt->eip = ctxt->_eip;
6347 		kvm_rip_write(vcpu, ctxt->eip);
6348 		kvm_set_rflags(vcpu, ctxt->eflags);
6349 	}
6350 }
6351 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6352 
6353 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6354 {
6355 	++vcpu->stat.insn_emulation_fail;
6356 	trace_kvm_emulate_insn_failed(vcpu);
6357 
6358 	if (emulation_type & EMULTYPE_VMWARE_GP) {
6359 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6360 		return 1;
6361 	}
6362 
6363 	if (emulation_type & EMULTYPE_SKIP) {
6364 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6365 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6366 		vcpu->run->internal.ndata = 0;
6367 		return 0;
6368 	}
6369 
6370 	kvm_queue_exception(vcpu, UD_VECTOR);
6371 
6372 	if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6373 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6374 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6375 		vcpu->run->internal.ndata = 0;
6376 		return 0;
6377 	}
6378 
6379 	return 1;
6380 }
6381 
6382 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
6383 				  bool write_fault_to_shadow_pgtable,
6384 				  int emulation_type)
6385 {
6386 	gpa_t gpa = cr2;
6387 	kvm_pfn_t pfn;
6388 
6389 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6390 		return false;
6391 
6392 	if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6393 		return false;
6394 
6395 	if (!vcpu->arch.mmu->direct_map) {
6396 		/*
6397 		 * Write permission should be allowed since only
6398 		 * write access need to be emulated.
6399 		 */
6400 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6401 
6402 		/*
6403 		 * If the mapping is invalid in guest, let cpu retry
6404 		 * it to generate fault.
6405 		 */
6406 		if (gpa == UNMAPPED_GVA)
6407 			return true;
6408 	}
6409 
6410 	/*
6411 	 * Do not retry the unhandleable instruction if it faults on the
6412 	 * readonly host memory, otherwise it will goto a infinite loop:
6413 	 * retry instruction -> write #PF -> emulation fail -> retry
6414 	 * instruction -> ...
6415 	 */
6416 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6417 
6418 	/*
6419 	 * If the instruction failed on the error pfn, it can not be fixed,
6420 	 * report the error to userspace.
6421 	 */
6422 	if (is_error_noslot_pfn(pfn))
6423 		return false;
6424 
6425 	kvm_release_pfn_clean(pfn);
6426 
6427 	/* The instructions are well-emulated on direct mmu. */
6428 	if (vcpu->arch.mmu->direct_map) {
6429 		unsigned int indirect_shadow_pages;
6430 
6431 		spin_lock(&vcpu->kvm->mmu_lock);
6432 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6433 		spin_unlock(&vcpu->kvm->mmu_lock);
6434 
6435 		if (indirect_shadow_pages)
6436 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6437 
6438 		return true;
6439 	}
6440 
6441 	/*
6442 	 * if emulation was due to access to shadowed page table
6443 	 * and it failed try to unshadow page and re-enter the
6444 	 * guest to let CPU execute the instruction.
6445 	 */
6446 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6447 
6448 	/*
6449 	 * If the access faults on its page table, it can not
6450 	 * be fixed by unprotecting shadow page and it should
6451 	 * be reported to userspace.
6452 	 */
6453 	return !write_fault_to_shadow_pgtable;
6454 }
6455 
6456 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6457 			      unsigned long cr2,  int emulation_type)
6458 {
6459 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6460 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
6461 
6462 	last_retry_eip = vcpu->arch.last_retry_eip;
6463 	last_retry_addr = vcpu->arch.last_retry_addr;
6464 
6465 	/*
6466 	 * If the emulation is caused by #PF and it is non-page_table
6467 	 * writing instruction, it means the VM-EXIT is caused by shadow
6468 	 * page protected, we can zap the shadow page and retry this
6469 	 * instruction directly.
6470 	 *
6471 	 * Note: if the guest uses a non-page-table modifying instruction
6472 	 * on the PDE that points to the instruction, then we will unmap
6473 	 * the instruction and go to an infinite loop. So, we cache the
6474 	 * last retried eip and the last fault address, if we meet the eip
6475 	 * and the address again, we can break out of the potential infinite
6476 	 * loop.
6477 	 */
6478 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6479 
6480 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6481 		return false;
6482 
6483 	if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6484 		return false;
6485 
6486 	if (x86_page_table_writing_insn(ctxt))
6487 		return false;
6488 
6489 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
6490 		return false;
6491 
6492 	vcpu->arch.last_retry_eip = ctxt->eip;
6493 	vcpu->arch.last_retry_addr = cr2;
6494 
6495 	if (!vcpu->arch.mmu->direct_map)
6496 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6497 
6498 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6499 
6500 	return true;
6501 }
6502 
6503 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6504 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6505 
6506 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6507 {
6508 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6509 		/* This is a good place to trace that we are exiting SMM.  */
6510 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6511 
6512 		/* Process a latched INIT or SMI, if any.  */
6513 		kvm_make_request(KVM_REQ_EVENT, vcpu);
6514 	}
6515 
6516 	kvm_mmu_reset_context(vcpu);
6517 }
6518 
6519 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6520 				unsigned long *db)
6521 {
6522 	u32 dr6 = 0;
6523 	int i;
6524 	u32 enable, rwlen;
6525 
6526 	enable = dr7;
6527 	rwlen = dr7 >> 16;
6528 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6529 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6530 			dr6 |= (1 << i);
6531 	return dr6;
6532 }
6533 
6534 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
6535 {
6536 	struct kvm_run *kvm_run = vcpu->run;
6537 
6538 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6539 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6540 		kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6541 		kvm_run->debug.arch.exception = DB_VECTOR;
6542 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
6543 		return 0;
6544 	}
6545 	kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6546 	return 1;
6547 }
6548 
6549 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6550 {
6551 	unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6552 	int r;
6553 
6554 	r = kvm_x86_ops->skip_emulated_instruction(vcpu);
6555 	if (unlikely(!r))
6556 		return 0;
6557 
6558 	/*
6559 	 * rflags is the old, "raw" value of the flags.  The new value has
6560 	 * not been saved yet.
6561 	 *
6562 	 * This is correct even for TF set by the guest, because "the
6563 	 * processor will not generate this exception after the instruction
6564 	 * that sets the TF flag".
6565 	 */
6566 	if (unlikely(rflags & X86_EFLAGS_TF))
6567 		r = kvm_vcpu_do_singlestep(vcpu);
6568 	return r;
6569 }
6570 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6571 
6572 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6573 {
6574 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6575 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6576 		struct kvm_run *kvm_run = vcpu->run;
6577 		unsigned long eip = kvm_get_linear_rip(vcpu);
6578 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6579 					   vcpu->arch.guest_debug_dr7,
6580 					   vcpu->arch.eff_db);
6581 
6582 		if (dr6 != 0) {
6583 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6584 			kvm_run->debug.arch.pc = eip;
6585 			kvm_run->debug.arch.exception = DB_VECTOR;
6586 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
6587 			*r = 0;
6588 			return true;
6589 		}
6590 	}
6591 
6592 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6593 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6594 		unsigned long eip = kvm_get_linear_rip(vcpu);
6595 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6596 					   vcpu->arch.dr7,
6597 					   vcpu->arch.db);
6598 
6599 		if (dr6 != 0) {
6600 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
6601 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
6602 			kvm_queue_exception(vcpu, DB_VECTOR);
6603 			*r = 1;
6604 			return true;
6605 		}
6606 	}
6607 
6608 	return false;
6609 }
6610 
6611 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6612 {
6613 	switch (ctxt->opcode_len) {
6614 	case 1:
6615 		switch (ctxt->b) {
6616 		case 0xe4:	/* IN */
6617 		case 0xe5:
6618 		case 0xec:
6619 		case 0xed:
6620 		case 0xe6:	/* OUT */
6621 		case 0xe7:
6622 		case 0xee:
6623 		case 0xef:
6624 		case 0x6c:	/* INS */
6625 		case 0x6d:
6626 		case 0x6e:	/* OUTS */
6627 		case 0x6f:
6628 			return true;
6629 		}
6630 		break;
6631 	case 2:
6632 		switch (ctxt->b) {
6633 		case 0x33:	/* RDPMC */
6634 			return true;
6635 		}
6636 		break;
6637 	}
6638 
6639 	return false;
6640 }
6641 
6642 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6643 			    unsigned long cr2,
6644 			    int emulation_type,
6645 			    void *insn,
6646 			    int insn_len)
6647 {
6648 	int r;
6649 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6650 	bool writeback = true;
6651 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6652 
6653 	vcpu->arch.l1tf_flush_l1d = true;
6654 
6655 	/*
6656 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
6657 	 * never reused.
6658 	 */
6659 	vcpu->arch.write_fault_to_shadow_pgtable = false;
6660 	kvm_clear_exception_queue(vcpu);
6661 
6662 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6663 		init_emulate_ctxt(vcpu);
6664 
6665 		/*
6666 		 * We will reenter on the same instruction since
6667 		 * we do not set complete_userspace_io.  This does not
6668 		 * handle watchpoints yet, those would be handled in
6669 		 * the emulate_ops.
6670 		 */
6671 		if (!(emulation_type & EMULTYPE_SKIP) &&
6672 		    kvm_vcpu_check_breakpoint(vcpu, &r))
6673 			return r;
6674 
6675 		ctxt->interruptibility = 0;
6676 		ctxt->have_exception = false;
6677 		ctxt->exception.vector = -1;
6678 		ctxt->perm_ok = false;
6679 
6680 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6681 
6682 		r = x86_decode_insn(ctxt, insn, insn_len);
6683 
6684 		trace_kvm_emulate_insn_start(vcpu);
6685 		++vcpu->stat.insn_emulation;
6686 		if (r != EMULATION_OK)  {
6687 			if ((emulation_type & EMULTYPE_TRAP_UD) ||
6688 			    (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
6689 				kvm_queue_exception(vcpu, UD_VECTOR);
6690 				return 1;
6691 			}
6692 			if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6693 						emulation_type))
6694 				return 1;
6695 			if (ctxt->have_exception) {
6696 				/*
6697 				 * #UD should result in just EMULATION_FAILED, and trap-like
6698 				 * exception should not be encountered during decode.
6699 				 */
6700 				WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
6701 					     exception_type(ctxt->exception.vector) == EXCPT_TRAP);
6702 				inject_emulated_exception(vcpu);
6703 				return 1;
6704 			}
6705 			return handle_emulation_failure(vcpu, emulation_type);
6706 		}
6707 	}
6708 
6709 	if ((emulation_type & EMULTYPE_VMWARE_GP) &&
6710 	    !is_vmware_backdoor_opcode(ctxt)) {
6711 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6712 		return 1;
6713 	}
6714 
6715 	/*
6716 	 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
6717 	 * for kvm_skip_emulated_instruction().  The caller is responsible for
6718 	 * updating interruptibility state and injecting single-step #DBs.
6719 	 */
6720 	if (emulation_type & EMULTYPE_SKIP) {
6721 		kvm_rip_write(vcpu, ctxt->_eip);
6722 		if (ctxt->eflags & X86_EFLAGS_RF)
6723 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6724 		return 1;
6725 	}
6726 
6727 	if (retry_instruction(ctxt, cr2, emulation_type))
6728 		return 1;
6729 
6730 	/* this is needed for vmware backdoor interface to work since it
6731 	   changes registers values  during IO operation */
6732 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6733 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6734 		emulator_invalidate_register_cache(ctxt);
6735 	}
6736 
6737 restart:
6738 	/* Save the faulting GPA (cr2) in the address field */
6739 	ctxt->exception.address = cr2;
6740 
6741 	r = x86_emulate_insn(ctxt);
6742 
6743 	if (r == EMULATION_INTERCEPTED)
6744 		return 1;
6745 
6746 	if (r == EMULATION_FAILED) {
6747 		if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6748 					emulation_type))
6749 			return 1;
6750 
6751 		return handle_emulation_failure(vcpu, emulation_type);
6752 	}
6753 
6754 	if (ctxt->have_exception) {
6755 		r = 1;
6756 		if (inject_emulated_exception(vcpu))
6757 			return r;
6758 	} else if (vcpu->arch.pio.count) {
6759 		if (!vcpu->arch.pio.in) {
6760 			/* FIXME: return into emulator if single-stepping.  */
6761 			vcpu->arch.pio.count = 0;
6762 		} else {
6763 			writeback = false;
6764 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
6765 		}
6766 		r = 0;
6767 	} else if (vcpu->mmio_needed) {
6768 		++vcpu->stat.mmio_exits;
6769 
6770 		if (!vcpu->mmio_is_write)
6771 			writeback = false;
6772 		r = 0;
6773 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6774 	} else if (r == EMULATION_RESTART)
6775 		goto restart;
6776 	else
6777 		r = 1;
6778 
6779 	if (writeback) {
6780 		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6781 		toggle_interruptibility(vcpu, ctxt->interruptibility);
6782 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6783 		if (!ctxt->have_exception ||
6784 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
6785 			kvm_rip_write(vcpu, ctxt->eip);
6786 			if (r && ctxt->tf)
6787 				r = kvm_vcpu_do_singlestep(vcpu);
6788 			__kvm_set_rflags(vcpu, ctxt->eflags);
6789 		}
6790 
6791 		/*
6792 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6793 		 * do nothing, and it will be requested again as soon as
6794 		 * the shadow expires.  But we still need to check here,
6795 		 * because POPF has no interrupt shadow.
6796 		 */
6797 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6798 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6799 	} else
6800 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6801 
6802 	return r;
6803 }
6804 
6805 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6806 {
6807 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6808 }
6809 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6810 
6811 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6812 					void *insn, int insn_len)
6813 {
6814 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6815 }
6816 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6817 
6818 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
6819 {
6820 	vcpu->arch.pio.count = 0;
6821 	return 1;
6822 }
6823 
6824 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
6825 {
6826 	vcpu->arch.pio.count = 0;
6827 
6828 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
6829 		return 1;
6830 
6831 	return kvm_skip_emulated_instruction(vcpu);
6832 }
6833 
6834 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6835 			    unsigned short port)
6836 {
6837 	unsigned long val = kvm_rax_read(vcpu);
6838 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6839 					    size, port, &val, 1);
6840 	if (ret)
6841 		return ret;
6842 
6843 	/*
6844 	 * Workaround userspace that relies on old KVM behavior of %rip being
6845 	 * incremented prior to exiting to userspace to handle "OUT 0x7e".
6846 	 */
6847 	if (port == 0x7e &&
6848 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
6849 		vcpu->arch.complete_userspace_io =
6850 			complete_fast_pio_out_port_0x7e;
6851 		kvm_skip_emulated_instruction(vcpu);
6852 	} else {
6853 		vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6854 		vcpu->arch.complete_userspace_io = complete_fast_pio_out;
6855 	}
6856 	return 0;
6857 }
6858 
6859 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6860 {
6861 	unsigned long val;
6862 
6863 	/* We should only ever be called with arch.pio.count equal to 1 */
6864 	BUG_ON(vcpu->arch.pio.count != 1);
6865 
6866 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
6867 		vcpu->arch.pio.count = 0;
6868 		return 1;
6869 	}
6870 
6871 	/* For size less than 4 we merge, else we zero extend */
6872 	val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
6873 
6874 	/*
6875 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6876 	 * the copy and tracing
6877 	 */
6878 	emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6879 				 vcpu->arch.pio.port, &val, 1);
6880 	kvm_rax_write(vcpu, val);
6881 
6882 	return kvm_skip_emulated_instruction(vcpu);
6883 }
6884 
6885 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6886 			   unsigned short port)
6887 {
6888 	unsigned long val;
6889 	int ret;
6890 
6891 	/* For size less than 4 we merge, else we zero extend */
6892 	val = (size < 4) ? kvm_rax_read(vcpu) : 0;
6893 
6894 	ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6895 				       &val, 1);
6896 	if (ret) {
6897 		kvm_rax_write(vcpu, val);
6898 		return ret;
6899 	}
6900 
6901 	vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6902 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6903 
6904 	return 0;
6905 }
6906 
6907 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6908 {
6909 	int ret;
6910 
6911 	if (in)
6912 		ret = kvm_fast_pio_in(vcpu, size, port);
6913 	else
6914 		ret = kvm_fast_pio_out(vcpu, size, port);
6915 	return ret && kvm_skip_emulated_instruction(vcpu);
6916 }
6917 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6918 
6919 static int kvmclock_cpu_down_prep(unsigned int cpu)
6920 {
6921 	__this_cpu_write(cpu_tsc_khz, 0);
6922 	return 0;
6923 }
6924 
6925 static void tsc_khz_changed(void *data)
6926 {
6927 	struct cpufreq_freqs *freq = data;
6928 	unsigned long khz = 0;
6929 
6930 	if (data)
6931 		khz = freq->new;
6932 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6933 		khz = cpufreq_quick_get(raw_smp_processor_id());
6934 	if (!khz)
6935 		khz = tsc_khz;
6936 	__this_cpu_write(cpu_tsc_khz, khz);
6937 }
6938 
6939 #ifdef CONFIG_X86_64
6940 static void kvm_hyperv_tsc_notifier(void)
6941 {
6942 	struct kvm *kvm;
6943 	struct kvm_vcpu *vcpu;
6944 	int cpu;
6945 
6946 	mutex_lock(&kvm_lock);
6947 	list_for_each_entry(kvm, &vm_list, vm_list)
6948 		kvm_make_mclock_inprogress_request(kvm);
6949 
6950 	hyperv_stop_tsc_emulation();
6951 
6952 	/* TSC frequency always matches when on Hyper-V */
6953 	for_each_present_cpu(cpu)
6954 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6955 	kvm_max_guest_tsc_khz = tsc_khz;
6956 
6957 	list_for_each_entry(kvm, &vm_list, vm_list) {
6958 		struct kvm_arch *ka = &kvm->arch;
6959 
6960 		spin_lock(&ka->pvclock_gtod_sync_lock);
6961 
6962 		pvclock_update_vm_gtod_copy(kvm);
6963 
6964 		kvm_for_each_vcpu(cpu, vcpu, kvm)
6965 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6966 
6967 		kvm_for_each_vcpu(cpu, vcpu, kvm)
6968 			kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6969 
6970 		spin_unlock(&ka->pvclock_gtod_sync_lock);
6971 	}
6972 	mutex_unlock(&kvm_lock);
6973 }
6974 #endif
6975 
6976 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
6977 {
6978 	struct kvm *kvm;
6979 	struct kvm_vcpu *vcpu;
6980 	int i, send_ipi = 0;
6981 
6982 	/*
6983 	 * We allow guests to temporarily run on slowing clocks,
6984 	 * provided we notify them after, or to run on accelerating
6985 	 * clocks, provided we notify them before.  Thus time never
6986 	 * goes backwards.
6987 	 *
6988 	 * However, we have a problem.  We can't atomically update
6989 	 * the frequency of a given CPU from this function; it is
6990 	 * merely a notifier, which can be called from any CPU.
6991 	 * Changing the TSC frequency at arbitrary points in time
6992 	 * requires a recomputation of local variables related to
6993 	 * the TSC for each VCPU.  We must flag these local variables
6994 	 * to be updated and be sure the update takes place with the
6995 	 * new frequency before any guests proceed.
6996 	 *
6997 	 * Unfortunately, the combination of hotplug CPU and frequency
6998 	 * change creates an intractable locking scenario; the order
6999 	 * of when these callouts happen is undefined with respect to
7000 	 * CPU hotplug, and they can race with each other.  As such,
7001 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7002 	 * undefined; you can actually have a CPU frequency change take
7003 	 * place in between the computation of X and the setting of the
7004 	 * variable.  To protect against this problem, all updates of
7005 	 * the per_cpu tsc_khz variable are done in an interrupt
7006 	 * protected IPI, and all callers wishing to update the value
7007 	 * must wait for a synchronous IPI to complete (which is trivial
7008 	 * if the caller is on the CPU already).  This establishes the
7009 	 * necessary total order on variable updates.
7010 	 *
7011 	 * Note that because a guest time update may take place
7012 	 * anytime after the setting of the VCPU's request bit, the
7013 	 * correct TSC value must be set before the request.  However,
7014 	 * to ensure the update actually makes it to any guest which
7015 	 * starts running in hardware virtualization between the set
7016 	 * and the acquisition of the spinlock, we must also ping the
7017 	 * CPU after setting the request bit.
7018 	 *
7019 	 */
7020 
7021 	smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7022 
7023 	mutex_lock(&kvm_lock);
7024 	list_for_each_entry(kvm, &vm_list, vm_list) {
7025 		kvm_for_each_vcpu(i, vcpu, kvm) {
7026 			if (vcpu->cpu != cpu)
7027 				continue;
7028 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7029 			if (vcpu->cpu != raw_smp_processor_id())
7030 				send_ipi = 1;
7031 		}
7032 	}
7033 	mutex_unlock(&kvm_lock);
7034 
7035 	if (freq->old < freq->new && send_ipi) {
7036 		/*
7037 		 * We upscale the frequency.  Must make the guest
7038 		 * doesn't see old kvmclock values while running with
7039 		 * the new frequency, otherwise we risk the guest sees
7040 		 * time go backwards.
7041 		 *
7042 		 * In case we update the frequency for another cpu
7043 		 * (which might be in guest context) send an interrupt
7044 		 * to kick the cpu out of guest context.  Next time
7045 		 * guest context is entered kvmclock will be updated,
7046 		 * so the guest will not see stale values.
7047 		 */
7048 		smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7049 	}
7050 }
7051 
7052 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7053 				     void *data)
7054 {
7055 	struct cpufreq_freqs *freq = data;
7056 	int cpu;
7057 
7058 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7059 		return 0;
7060 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7061 		return 0;
7062 
7063 	for_each_cpu(cpu, freq->policy->cpus)
7064 		__kvmclock_cpufreq_notifier(freq, cpu);
7065 
7066 	return 0;
7067 }
7068 
7069 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7070 	.notifier_call  = kvmclock_cpufreq_notifier
7071 };
7072 
7073 static int kvmclock_cpu_online(unsigned int cpu)
7074 {
7075 	tsc_khz_changed(NULL);
7076 	return 0;
7077 }
7078 
7079 static void kvm_timer_init(void)
7080 {
7081 	max_tsc_khz = tsc_khz;
7082 
7083 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7084 #ifdef CONFIG_CPU_FREQ
7085 		struct cpufreq_policy policy;
7086 		int cpu;
7087 
7088 		memset(&policy, 0, sizeof(policy));
7089 		cpu = get_cpu();
7090 		cpufreq_get_policy(&policy, cpu);
7091 		if (policy.cpuinfo.max_freq)
7092 			max_tsc_khz = policy.cpuinfo.max_freq;
7093 		put_cpu();
7094 #endif
7095 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7096 					  CPUFREQ_TRANSITION_NOTIFIER);
7097 	}
7098 
7099 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7100 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
7101 }
7102 
7103 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7104 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7105 
7106 int kvm_is_in_guest(void)
7107 {
7108 	return __this_cpu_read(current_vcpu) != NULL;
7109 }
7110 
7111 static int kvm_is_user_mode(void)
7112 {
7113 	int user_mode = 3;
7114 
7115 	if (__this_cpu_read(current_vcpu))
7116 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
7117 
7118 	return user_mode != 0;
7119 }
7120 
7121 static unsigned long kvm_get_guest_ip(void)
7122 {
7123 	unsigned long ip = 0;
7124 
7125 	if (__this_cpu_read(current_vcpu))
7126 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7127 
7128 	return ip;
7129 }
7130 
7131 static void kvm_handle_intel_pt_intr(void)
7132 {
7133 	struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7134 
7135 	kvm_make_request(KVM_REQ_PMI, vcpu);
7136 	__set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7137 			(unsigned long *)&vcpu->arch.pmu.global_status);
7138 }
7139 
7140 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7141 	.is_in_guest		= kvm_is_in_guest,
7142 	.is_user_mode		= kvm_is_user_mode,
7143 	.get_guest_ip		= kvm_get_guest_ip,
7144 	.handle_intel_pt_intr	= kvm_handle_intel_pt_intr,
7145 };
7146 
7147 #ifdef CONFIG_X86_64
7148 static void pvclock_gtod_update_fn(struct work_struct *work)
7149 {
7150 	struct kvm *kvm;
7151 
7152 	struct kvm_vcpu *vcpu;
7153 	int i;
7154 
7155 	mutex_lock(&kvm_lock);
7156 	list_for_each_entry(kvm, &vm_list, vm_list)
7157 		kvm_for_each_vcpu(i, vcpu, kvm)
7158 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7159 	atomic_set(&kvm_guest_has_master_clock, 0);
7160 	mutex_unlock(&kvm_lock);
7161 }
7162 
7163 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7164 
7165 /*
7166  * Notification about pvclock gtod data update.
7167  */
7168 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7169 			       void *priv)
7170 {
7171 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7172 	struct timekeeper *tk = priv;
7173 
7174 	update_pvclock_gtod(tk);
7175 
7176 	/* disable master clock if host does not trust, or does not
7177 	 * use, TSC based clocksource.
7178 	 */
7179 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7180 	    atomic_read(&kvm_guest_has_master_clock) != 0)
7181 		queue_work(system_long_wq, &pvclock_gtod_work);
7182 
7183 	return 0;
7184 }
7185 
7186 static struct notifier_block pvclock_gtod_notifier = {
7187 	.notifier_call = pvclock_gtod_notify,
7188 };
7189 #endif
7190 
7191 int kvm_arch_init(void *opaque)
7192 {
7193 	int r;
7194 	struct kvm_x86_ops *ops = opaque;
7195 
7196 	if (kvm_x86_ops) {
7197 		printk(KERN_ERR "kvm: already loaded the other module\n");
7198 		r = -EEXIST;
7199 		goto out;
7200 	}
7201 
7202 	if (!ops->cpu_has_kvm_support()) {
7203 		printk(KERN_ERR "kvm: no hardware support\n");
7204 		r = -EOPNOTSUPP;
7205 		goto out;
7206 	}
7207 	if (ops->disabled_by_bios()) {
7208 		printk(KERN_ERR "kvm: disabled by bios\n");
7209 		r = -EOPNOTSUPP;
7210 		goto out;
7211 	}
7212 
7213 	/*
7214 	 * KVM explicitly assumes that the guest has an FPU and
7215 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7216 	 * vCPU's FPU state as a fxregs_state struct.
7217 	 */
7218 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7219 		printk(KERN_ERR "kvm: inadequate fpu\n");
7220 		r = -EOPNOTSUPP;
7221 		goto out;
7222 	}
7223 
7224 	r = -ENOMEM;
7225 	x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7226 					  __alignof__(struct fpu), SLAB_ACCOUNT,
7227 					  NULL);
7228 	if (!x86_fpu_cache) {
7229 		printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7230 		goto out;
7231 	}
7232 
7233 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
7234 	if (!shared_msrs) {
7235 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
7236 		goto out_free_x86_fpu_cache;
7237 	}
7238 
7239 	r = kvm_mmu_module_init();
7240 	if (r)
7241 		goto out_free_percpu;
7242 
7243 	kvm_x86_ops = ops;
7244 
7245 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7246 			PT_DIRTY_MASK, PT64_NX_MASK, 0,
7247 			PT_PRESENT_MASK, 0, sme_me_mask);
7248 	kvm_timer_init();
7249 
7250 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
7251 
7252 	if (boot_cpu_has(X86_FEATURE_XSAVE))
7253 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7254 
7255 	kvm_lapic_init();
7256 	if (pi_inject_timer == -1)
7257 		pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7258 #ifdef CONFIG_X86_64
7259 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7260 
7261 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7262 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7263 #endif
7264 
7265 	return 0;
7266 
7267 out_free_percpu:
7268 	free_percpu(shared_msrs);
7269 out_free_x86_fpu_cache:
7270 	kmem_cache_destroy(x86_fpu_cache);
7271 out:
7272 	return r;
7273 }
7274 
7275 void kvm_arch_exit(void)
7276 {
7277 #ifdef CONFIG_X86_64
7278 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7279 		clear_hv_tscchange_cb();
7280 #endif
7281 	kvm_lapic_exit();
7282 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7283 
7284 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7285 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7286 					    CPUFREQ_TRANSITION_NOTIFIER);
7287 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7288 #ifdef CONFIG_X86_64
7289 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7290 #endif
7291 	kvm_x86_ops = NULL;
7292 	kvm_mmu_module_exit();
7293 	free_percpu(shared_msrs);
7294 	kmem_cache_destroy(x86_fpu_cache);
7295 }
7296 
7297 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7298 {
7299 	++vcpu->stat.halt_exits;
7300 	if (lapic_in_kernel(vcpu)) {
7301 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7302 		return 1;
7303 	} else {
7304 		vcpu->run->exit_reason = KVM_EXIT_HLT;
7305 		return 0;
7306 	}
7307 }
7308 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7309 
7310 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7311 {
7312 	int ret = kvm_skip_emulated_instruction(vcpu);
7313 	/*
7314 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7315 	 * KVM_EXIT_DEBUG here.
7316 	 */
7317 	return kvm_vcpu_halt(vcpu) && ret;
7318 }
7319 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7320 
7321 #ifdef CONFIG_X86_64
7322 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7323 			        unsigned long clock_type)
7324 {
7325 	struct kvm_clock_pairing clock_pairing;
7326 	struct timespec64 ts;
7327 	u64 cycle;
7328 	int ret;
7329 
7330 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7331 		return -KVM_EOPNOTSUPP;
7332 
7333 	if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7334 		return -KVM_EOPNOTSUPP;
7335 
7336 	clock_pairing.sec = ts.tv_sec;
7337 	clock_pairing.nsec = ts.tv_nsec;
7338 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7339 	clock_pairing.flags = 0;
7340 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7341 
7342 	ret = 0;
7343 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7344 			    sizeof(struct kvm_clock_pairing)))
7345 		ret = -KVM_EFAULT;
7346 
7347 	return ret;
7348 }
7349 #endif
7350 
7351 /*
7352  * kvm_pv_kick_cpu_op:  Kick a vcpu.
7353  *
7354  * @apicid - apicid of vcpu to be kicked.
7355  */
7356 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7357 {
7358 	struct kvm_lapic_irq lapic_irq;
7359 
7360 	lapic_irq.shorthand = 0;
7361 	lapic_irq.dest_mode = 0;
7362 	lapic_irq.level = 0;
7363 	lapic_irq.dest_id = apicid;
7364 	lapic_irq.msi_redir_hint = false;
7365 
7366 	lapic_irq.delivery_mode = APIC_DM_REMRD;
7367 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7368 }
7369 
7370 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
7371 {
7372 	if (!lapic_in_kernel(vcpu)) {
7373 		WARN_ON_ONCE(vcpu->arch.apicv_active);
7374 		return;
7375 	}
7376 	if (!vcpu->arch.apicv_active)
7377 		return;
7378 
7379 	vcpu->arch.apicv_active = false;
7380 	kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
7381 }
7382 
7383 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
7384 {
7385 	struct kvm_vcpu *target = NULL;
7386 	struct kvm_apic_map *map;
7387 
7388 	rcu_read_lock();
7389 	map = rcu_dereference(kvm->arch.apic_map);
7390 
7391 	if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
7392 		target = map->phys_map[dest_id]->vcpu;
7393 
7394 	rcu_read_unlock();
7395 
7396 	if (target && READ_ONCE(target->ready))
7397 		kvm_vcpu_yield_to(target);
7398 }
7399 
7400 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7401 {
7402 	unsigned long nr, a0, a1, a2, a3, ret;
7403 	int op_64_bit;
7404 
7405 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
7406 		return kvm_hv_hypercall(vcpu);
7407 
7408 	nr = kvm_rax_read(vcpu);
7409 	a0 = kvm_rbx_read(vcpu);
7410 	a1 = kvm_rcx_read(vcpu);
7411 	a2 = kvm_rdx_read(vcpu);
7412 	a3 = kvm_rsi_read(vcpu);
7413 
7414 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
7415 
7416 	op_64_bit = is_64_bit_mode(vcpu);
7417 	if (!op_64_bit) {
7418 		nr &= 0xFFFFFFFF;
7419 		a0 &= 0xFFFFFFFF;
7420 		a1 &= 0xFFFFFFFF;
7421 		a2 &= 0xFFFFFFFF;
7422 		a3 &= 0xFFFFFFFF;
7423 	}
7424 
7425 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
7426 		ret = -KVM_EPERM;
7427 		goto out;
7428 	}
7429 
7430 	switch (nr) {
7431 	case KVM_HC_VAPIC_POLL_IRQ:
7432 		ret = 0;
7433 		break;
7434 	case KVM_HC_KICK_CPU:
7435 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7436 		kvm_sched_yield(vcpu->kvm, a1);
7437 		ret = 0;
7438 		break;
7439 #ifdef CONFIG_X86_64
7440 	case KVM_HC_CLOCK_PAIRING:
7441 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7442 		break;
7443 #endif
7444 	case KVM_HC_SEND_IPI:
7445 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7446 		break;
7447 	case KVM_HC_SCHED_YIELD:
7448 		kvm_sched_yield(vcpu->kvm, a0);
7449 		ret = 0;
7450 		break;
7451 	default:
7452 		ret = -KVM_ENOSYS;
7453 		break;
7454 	}
7455 out:
7456 	if (!op_64_bit)
7457 		ret = (u32)ret;
7458 	kvm_rax_write(vcpu, ret);
7459 
7460 	++vcpu->stat.hypercalls;
7461 	return kvm_skip_emulated_instruction(vcpu);
7462 }
7463 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7464 
7465 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7466 {
7467 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7468 	char instruction[3];
7469 	unsigned long rip = kvm_rip_read(vcpu);
7470 
7471 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
7472 
7473 	return emulator_write_emulated(ctxt, rip, instruction, 3,
7474 		&ctxt->exception);
7475 }
7476 
7477 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7478 {
7479 	return vcpu->run->request_interrupt_window &&
7480 		likely(!pic_in_kernel(vcpu->kvm));
7481 }
7482 
7483 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7484 {
7485 	struct kvm_run *kvm_run = vcpu->run;
7486 
7487 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7488 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7489 	kvm_run->cr8 = kvm_get_cr8(vcpu);
7490 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
7491 	kvm_run->ready_for_interrupt_injection =
7492 		pic_in_kernel(vcpu->kvm) ||
7493 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
7494 }
7495 
7496 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7497 {
7498 	int max_irr, tpr;
7499 
7500 	if (!kvm_x86_ops->update_cr8_intercept)
7501 		return;
7502 
7503 	if (!lapic_in_kernel(vcpu))
7504 		return;
7505 
7506 	if (vcpu->arch.apicv_active)
7507 		return;
7508 
7509 	if (!vcpu->arch.apic->vapic_addr)
7510 		max_irr = kvm_lapic_find_highest_irr(vcpu);
7511 	else
7512 		max_irr = -1;
7513 
7514 	if (max_irr != -1)
7515 		max_irr >>= 4;
7516 
7517 	tpr = kvm_lapic_get_cr8(vcpu);
7518 
7519 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7520 }
7521 
7522 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7523 {
7524 	int r;
7525 
7526 	/* try to reinject previous events if any */
7527 
7528 	if (vcpu->arch.exception.injected)
7529 		kvm_x86_ops->queue_exception(vcpu);
7530 	/*
7531 	 * Do not inject an NMI or interrupt if there is a pending
7532 	 * exception.  Exceptions and interrupts are recognized at
7533 	 * instruction boundaries, i.e. the start of an instruction.
7534 	 * Trap-like exceptions, e.g. #DB, have higher priority than
7535 	 * NMIs and interrupts, i.e. traps are recognized before an
7536 	 * NMI/interrupt that's pending on the same instruction.
7537 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7538 	 * priority, but are only generated (pended) during instruction
7539 	 * execution, i.e. a pending fault-like exception means the
7540 	 * fault occurred on the *previous* instruction and must be
7541 	 * serviced prior to recognizing any new events in order to
7542 	 * fully complete the previous instruction.
7543 	 */
7544 	else if (!vcpu->arch.exception.pending) {
7545 		if (vcpu->arch.nmi_injected)
7546 			kvm_x86_ops->set_nmi(vcpu);
7547 		else if (vcpu->arch.interrupt.injected)
7548 			kvm_x86_ops->set_irq(vcpu);
7549 	}
7550 
7551 	/*
7552 	 * Call check_nested_events() even if we reinjected a previous event
7553 	 * in order for caller to determine if it should require immediate-exit
7554 	 * from L2 to L1 due to pending L1 events which require exit
7555 	 * from L2 to L1.
7556 	 */
7557 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7558 		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7559 		if (r != 0)
7560 			return r;
7561 	}
7562 
7563 	/* try to inject new event if pending */
7564 	if (vcpu->arch.exception.pending) {
7565 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
7566 					vcpu->arch.exception.has_error_code,
7567 					vcpu->arch.exception.error_code);
7568 
7569 		WARN_ON_ONCE(vcpu->arch.exception.injected);
7570 		vcpu->arch.exception.pending = false;
7571 		vcpu->arch.exception.injected = true;
7572 
7573 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7574 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7575 					     X86_EFLAGS_RF);
7576 
7577 		if (vcpu->arch.exception.nr == DB_VECTOR) {
7578 			/*
7579 			 * This code assumes that nSVM doesn't use
7580 			 * check_nested_events(). If it does, the
7581 			 * DR6/DR7 changes should happen before L1
7582 			 * gets a #VMEXIT for an intercepted #DB in
7583 			 * L2.  (Under VMX, on the other hand, the
7584 			 * DR6/DR7 changes should not happen in the
7585 			 * event of a VM-exit to L1 for an intercepted
7586 			 * #DB in L2.)
7587 			 */
7588 			kvm_deliver_exception_payload(vcpu);
7589 			if (vcpu->arch.dr7 & DR7_GD) {
7590 				vcpu->arch.dr7 &= ~DR7_GD;
7591 				kvm_update_dr7(vcpu);
7592 			}
7593 		}
7594 
7595 		kvm_x86_ops->queue_exception(vcpu);
7596 	}
7597 
7598 	/* Don't consider new event if we re-injected an event */
7599 	if (kvm_event_needs_reinjection(vcpu))
7600 		return 0;
7601 
7602 	if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7603 	    kvm_x86_ops->smi_allowed(vcpu)) {
7604 		vcpu->arch.smi_pending = false;
7605 		++vcpu->arch.smi_count;
7606 		enter_smm(vcpu);
7607 	} else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7608 		--vcpu->arch.nmi_pending;
7609 		vcpu->arch.nmi_injected = true;
7610 		kvm_x86_ops->set_nmi(vcpu);
7611 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
7612 		/*
7613 		 * Because interrupts can be injected asynchronously, we are
7614 		 * calling check_nested_events again here to avoid a race condition.
7615 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7616 		 * proposal and current concerns.  Perhaps we should be setting
7617 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
7618 		 */
7619 		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7620 			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7621 			if (r != 0)
7622 				return r;
7623 		}
7624 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7625 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7626 					    false);
7627 			kvm_x86_ops->set_irq(vcpu);
7628 		}
7629 	}
7630 
7631 	return 0;
7632 }
7633 
7634 static void process_nmi(struct kvm_vcpu *vcpu)
7635 {
7636 	unsigned limit = 2;
7637 
7638 	/*
7639 	 * x86 is limited to one NMI running, and one NMI pending after it.
7640 	 * If an NMI is already in progress, limit further NMIs to just one.
7641 	 * Otherwise, allow two (and we'll inject the first one immediately).
7642 	 */
7643 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7644 		limit = 1;
7645 
7646 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7647 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7648 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7649 }
7650 
7651 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7652 {
7653 	u32 flags = 0;
7654 	flags |= seg->g       << 23;
7655 	flags |= seg->db      << 22;
7656 	flags |= seg->l       << 21;
7657 	flags |= seg->avl     << 20;
7658 	flags |= seg->present << 15;
7659 	flags |= seg->dpl     << 13;
7660 	flags |= seg->s       << 12;
7661 	flags |= seg->type    << 8;
7662 	return flags;
7663 }
7664 
7665 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7666 {
7667 	struct kvm_segment seg;
7668 	int offset;
7669 
7670 	kvm_get_segment(vcpu, &seg, n);
7671 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7672 
7673 	if (n < 3)
7674 		offset = 0x7f84 + n * 12;
7675 	else
7676 		offset = 0x7f2c + (n - 3) * 12;
7677 
7678 	put_smstate(u32, buf, offset + 8, seg.base);
7679 	put_smstate(u32, buf, offset + 4, seg.limit);
7680 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7681 }
7682 
7683 #ifdef CONFIG_X86_64
7684 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7685 {
7686 	struct kvm_segment seg;
7687 	int offset;
7688 	u16 flags;
7689 
7690 	kvm_get_segment(vcpu, &seg, n);
7691 	offset = 0x7e00 + n * 16;
7692 
7693 	flags = enter_smm_get_segment_flags(&seg) >> 8;
7694 	put_smstate(u16, buf, offset, seg.selector);
7695 	put_smstate(u16, buf, offset + 2, flags);
7696 	put_smstate(u32, buf, offset + 4, seg.limit);
7697 	put_smstate(u64, buf, offset + 8, seg.base);
7698 }
7699 #endif
7700 
7701 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7702 {
7703 	struct desc_ptr dt;
7704 	struct kvm_segment seg;
7705 	unsigned long val;
7706 	int i;
7707 
7708 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7709 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7710 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7711 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7712 
7713 	for (i = 0; i < 8; i++)
7714 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7715 
7716 	kvm_get_dr(vcpu, 6, &val);
7717 	put_smstate(u32, buf, 0x7fcc, (u32)val);
7718 	kvm_get_dr(vcpu, 7, &val);
7719 	put_smstate(u32, buf, 0x7fc8, (u32)val);
7720 
7721 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7722 	put_smstate(u32, buf, 0x7fc4, seg.selector);
7723 	put_smstate(u32, buf, 0x7f64, seg.base);
7724 	put_smstate(u32, buf, 0x7f60, seg.limit);
7725 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7726 
7727 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7728 	put_smstate(u32, buf, 0x7fc0, seg.selector);
7729 	put_smstate(u32, buf, 0x7f80, seg.base);
7730 	put_smstate(u32, buf, 0x7f7c, seg.limit);
7731 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7732 
7733 	kvm_x86_ops->get_gdt(vcpu, &dt);
7734 	put_smstate(u32, buf, 0x7f74, dt.address);
7735 	put_smstate(u32, buf, 0x7f70, dt.size);
7736 
7737 	kvm_x86_ops->get_idt(vcpu, &dt);
7738 	put_smstate(u32, buf, 0x7f58, dt.address);
7739 	put_smstate(u32, buf, 0x7f54, dt.size);
7740 
7741 	for (i = 0; i < 6; i++)
7742 		enter_smm_save_seg_32(vcpu, buf, i);
7743 
7744 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7745 
7746 	/* revision id */
7747 	put_smstate(u32, buf, 0x7efc, 0x00020000);
7748 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7749 }
7750 
7751 #ifdef CONFIG_X86_64
7752 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7753 {
7754 	struct desc_ptr dt;
7755 	struct kvm_segment seg;
7756 	unsigned long val;
7757 	int i;
7758 
7759 	for (i = 0; i < 16; i++)
7760 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7761 
7762 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7763 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7764 
7765 	kvm_get_dr(vcpu, 6, &val);
7766 	put_smstate(u64, buf, 0x7f68, val);
7767 	kvm_get_dr(vcpu, 7, &val);
7768 	put_smstate(u64, buf, 0x7f60, val);
7769 
7770 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7771 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7772 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7773 
7774 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7775 
7776 	/* revision id */
7777 	put_smstate(u32, buf, 0x7efc, 0x00020064);
7778 
7779 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7780 
7781 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7782 	put_smstate(u16, buf, 0x7e90, seg.selector);
7783 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7784 	put_smstate(u32, buf, 0x7e94, seg.limit);
7785 	put_smstate(u64, buf, 0x7e98, seg.base);
7786 
7787 	kvm_x86_ops->get_idt(vcpu, &dt);
7788 	put_smstate(u32, buf, 0x7e84, dt.size);
7789 	put_smstate(u64, buf, 0x7e88, dt.address);
7790 
7791 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7792 	put_smstate(u16, buf, 0x7e70, seg.selector);
7793 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7794 	put_smstate(u32, buf, 0x7e74, seg.limit);
7795 	put_smstate(u64, buf, 0x7e78, seg.base);
7796 
7797 	kvm_x86_ops->get_gdt(vcpu, &dt);
7798 	put_smstate(u32, buf, 0x7e64, dt.size);
7799 	put_smstate(u64, buf, 0x7e68, dt.address);
7800 
7801 	for (i = 0; i < 6; i++)
7802 		enter_smm_save_seg_64(vcpu, buf, i);
7803 }
7804 #endif
7805 
7806 static void enter_smm(struct kvm_vcpu *vcpu)
7807 {
7808 	struct kvm_segment cs, ds;
7809 	struct desc_ptr dt;
7810 	char buf[512];
7811 	u32 cr0;
7812 
7813 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7814 	memset(buf, 0, 512);
7815 #ifdef CONFIG_X86_64
7816 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7817 		enter_smm_save_state_64(vcpu, buf);
7818 	else
7819 #endif
7820 		enter_smm_save_state_32(vcpu, buf);
7821 
7822 	/*
7823 	 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7824 	 * vCPU state (e.g. leave guest mode) after we've saved the state into
7825 	 * the SMM state-save area.
7826 	 */
7827 	kvm_x86_ops->pre_enter_smm(vcpu, buf);
7828 
7829 	vcpu->arch.hflags |= HF_SMM_MASK;
7830 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7831 
7832 	if (kvm_x86_ops->get_nmi_mask(vcpu))
7833 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7834 	else
7835 		kvm_x86_ops->set_nmi_mask(vcpu, true);
7836 
7837 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7838 	kvm_rip_write(vcpu, 0x8000);
7839 
7840 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7841 	kvm_x86_ops->set_cr0(vcpu, cr0);
7842 	vcpu->arch.cr0 = cr0;
7843 
7844 	kvm_x86_ops->set_cr4(vcpu, 0);
7845 
7846 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
7847 	dt.address = dt.size = 0;
7848 	kvm_x86_ops->set_idt(vcpu, &dt);
7849 
7850 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7851 
7852 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7853 	cs.base = vcpu->arch.smbase;
7854 
7855 	ds.selector = 0;
7856 	ds.base = 0;
7857 
7858 	cs.limit    = ds.limit = 0xffffffff;
7859 	cs.type     = ds.type = 0x3;
7860 	cs.dpl      = ds.dpl = 0;
7861 	cs.db       = ds.db = 0;
7862 	cs.s        = ds.s = 1;
7863 	cs.l        = ds.l = 0;
7864 	cs.g        = ds.g = 1;
7865 	cs.avl      = ds.avl = 0;
7866 	cs.present  = ds.present = 1;
7867 	cs.unusable = ds.unusable = 0;
7868 	cs.padding  = ds.padding = 0;
7869 
7870 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7871 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7872 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7873 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7874 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7875 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7876 
7877 #ifdef CONFIG_X86_64
7878 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7879 		kvm_x86_ops->set_efer(vcpu, 0);
7880 #endif
7881 
7882 	kvm_update_cpuid(vcpu);
7883 	kvm_mmu_reset_context(vcpu);
7884 }
7885 
7886 static void process_smi(struct kvm_vcpu *vcpu)
7887 {
7888 	vcpu->arch.smi_pending = true;
7889 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7890 }
7891 
7892 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
7893 				       unsigned long *vcpu_bitmap)
7894 {
7895 	cpumask_var_t cpus;
7896 
7897 	zalloc_cpumask_var(&cpus, GFP_ATOMIC);
7898 
7899 	kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
7900 				    vcpu_bitmap, cpus);
7901 
7902 	free_cpumask_var(cpus);
7903 }
7904 
7905 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7906 {
7907 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7908 }
7909 
7910 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7911 {
7912 	if (!kvm_apic_present(vcpu))
7913 		return;
7914 
7915 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7916 
7917 	if (irqchip_split(vcpu->kvm))
7918 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7919 	else {
7920 		if (vcpu->arch.apicv_active)
7921 			kvm_x86_ops->sync_pir_to_irr(vcpu);
7922 		if (ioapic_in_kernel(vcpu->kvm))
7923 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7924 	}
7925 
7926 	if (is_guest_mode(vcpu))
7927 		vcpu->arch.load_eoi_exitmap_pending = true;
7928 	else
7929 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7930 }
7931 
7932 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7933 {
7934 	u64 eoi_exit_bitmap[4];
7935 
7936 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7937 		return;
7938 
7939 	bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7940 		  vcpu_to_synic(vcpu)->vec_bitmap, 256);
7941 	kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7942 }
7943 
7944 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7945 		unsigned long start, unsigned long end,
7946 		bool blockable)
7947 {
7948 	unsigned long apic_address;
7949 
7950 	/*
7951 	 * The physical address of apic access page is stored in the VMCS.
7952 	 * Update it when it becomes invalid.
7953 	 */
7954 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7955 	if (start <= apic_address && apic_address < end)
7956 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7957 
7958 	return 0;
7959 }
7960 
7961 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7962 {
7963 	struct page *page = NULL;
7964 
7965 	if (!lapic_in_kernel(vcpu))
7966 		return;
7967 
7968 	if (!kvm_x86_ops->set_apic_access_page_addr)
7969 		return;
7970 
7971 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7972 	if (is_error_page(page))
7973 		return;
7974 	kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7975 
7976 	/*
7977 	 * Do not pin apic access page in memory, the MMU notifier
7978 	 * will call us again if it is migrated or swapped out.
7979 	 */
7980 	put_page(page);
7981 }
7982 
7983 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
7984 {
7985 	smp_send_reschedule(vcpu->cpu);
7986 }
7987 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
7988 
7989 /*
7990  * Returns 1 to let vcpu_run() continue the guest execution loop without
7991  * exiting to the userspace.  Otherwise, the value will be returned to the
7992  * userspace.
7993  */
7994 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7995 {
7996 	int r;
7997 	bool req_int_win =
7998 		dm_request_for_irq_injection(vcpu) &&
7999 		kvm_cpu_accept_dm_intr(vcpu);
8000 
8001 	bool req_immediate_exit = false;
8002 
8003 	if (kvm_request_pending(vcpu)) {
8004 		if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu)) {
8005 			if (unlikely(!kvm_x86_ops->get_vmcs12_pages(vcpu))) {
8006 				r = 0;
8007 				goto out;
8008 			}
8009 		}
8010 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
8011 			kvm_mmu_unload(vcpu);
8012 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
8013 			__kvm_migrate_timers(vcpu);
8014 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
8015 			kvm_gen_update_masterclock(vcpu->kvm);
8016 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
8017 			kvm_gen_kvmclock_update(vcpu);
8018 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
8019 			r = kvm_guest_time_update(vcpu);
8020 			if (unlikely(r))
8021 				goto out;
8022 		}
8023 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
8024 			kvm_mmu_sync_roots(vcpu);
8025 		if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
8026 			kvm_mmu_load_cr3(vcpu);
8027 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
8028 			kvm_vcpu_flush_tlb(vcpu, true);
8029 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
8030 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
8031 			r = 0;
8032 			goto out;
8033 		}
8034 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8035 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
8036 			vcpu->mmio_needed = 0;
8037 			r = 0;
8038 			goto out;
8039 		}
8040 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
8041 			/* Page is swapped out. Do synthetic halt */
8042 			vcpu->arch.apf.halted = true;
8043 			r = 1;
8044 			goto out;
8045 		}
8046 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8047 			record_steal_time(vcpu);
8048 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
8049 			process_smi(vcpu);
8050 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
8051 			process_nmi(vcpu);
8052 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
8053 			kvm_pmu_handle_event(vcpu);
8054 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
8055 			kvm_pmu_deliver_pmi(vcpu);
8056 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8057 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8058 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
8059 				     vcpu->arch.ioapic_handled_vectors)) {
8060 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8061 				vcpu->run->eoi.vector =
8062 						vcpu->arch.pending_ioapic_eoi;
8063 				r = 0;
8064 				goto out;
8065 			}
8066 		}
8067 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8068 			vcpu_scan_ioapic(vcpu);
8069 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8070 			vcpu_load_eoi_exitmap(vcpu);
8071 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8072 			kvm_vcpu_reload_apic_access_page(vcpu);
8073 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8074 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8075 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8076 			r = 0;
8077 			goto out;
8078 		}
8079 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8080 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8081 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8082 			r = 0;
8083 			goto out;
8084 		}
8085 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8086 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8087 			vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8088 			r = 0;
8089 			goto out;
8090 		}
8091 
8092 		/*
8093 		 * KVM_REQ_HV_STIMER has to be processed after
8094 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8095 		 * depend on the guest clock being up-to-date
8096 		 */
8097 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8098 			kvm_hv_process_stimers(vcpu);
8099 	}
8100 
8101 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8102 		++vcpu->stat.req_event;
8103 		kvm_apic_accept_events(vcpu);
8104 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8105 			r = 1;
8106 			goto out;
8107 		}
8108 
8109 		if (inject_pending_event(vcpu, req_int_win) != 0)
8110 			req_immediate_exit = true;
8111 		else {
8112 			/* Enable SMI/NMI/IRQ window open exits if needed.
8113 			 *
8114 			 * SMIs have three cases:
8115 			 * 1) They can be nested, and then there is nothing to
8116 			 *    do here because RSM will cause a vmexit anyway.
8117 			 * 2) There is an ISA-specific reason why SMI cannot be
8118 			 *    injected, and the moment when this changes can be
8119 			 *    intercepted.
8120 			 * 3) Or the SMI can be pending because
8121 			 *    inject_pending_event has completed the injection
8122 			 *    of an IRQ or NMI from the previous vmexit, and
8123 			 *    then we request an immediate exit to inject the
8124 			 *    SMI.
8125 			 */
8126 			if (vcpu->arch.smi_pending && !is_smm(vcpu))
8127 				if (!kvm_x86_ops->enable_smi_window(vcpu))
8128 					req_immediate_exit = true;
8129 			if (vcpu->arch.nmi_pending)
8130 				kvm_x86_ops->enable_nmi_window(vcpu);
8131 			if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
8132 				kvm_x86_ops->enable_irq_window(vcpu);
8133 			WARN_ON(vcpu->arch.exception.pending);
8134 		}
8135 
8136 		if (kvm_lapic_enabled(vcpu)) {
8137 			update_cr8_intercept(vcpu);
8138 			kvm_lapic_sync_to_vapic(vcpu);
8139 		}
8140 	}
8141 
8142 	r = kvm_mmu_reload(vcpu);
8143 	if (unlikely(r)) {
8144 		goto cancel_injection;
8145 	}
8146 
8147 	preempt_disable();
8148 
8149 	kvm_x86_ops->prepare_guest_switch(vcpu);
8150 
8151 	/*
8152 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
8153 	 * IPI are then delayed after guest entry, which ensures that they
8154 	 * result in virtual interrupt delivery.
8155 	 */
8156 	local_irq_disable();
8157 	vcpu->mode = IN_GUEST_MODE;
8158 
8159 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8160 
8161 	/*
8162 	 * 1) We should set ->mode before checking ->requests.  Please see
8163 	 * the comment in kvm_vcpu_exiting_guest_mode().
8164 	 *
8165 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
8166 	 * pairs with the memory barrier implicit in pi_test_and_set_on
8167 	 * (see vmx_deliver_posted_interrupt).
8168 	 *
8169 	 * 3) This also orders the write to mode from any reads to the page
8170 	 * tables done while the VCPU is running.  Please see the comment
8171 	 * in kvm_flush_remote_tlbs.
8172 	 */
8173 	smp_mb__after_srcu_read_unlock();
8174 
8175 	/*
8176 	 * This handles the case where a posted interrupt was
8177 	 * notified with kvm_vcpu_kick.
8178 	 */
8179 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8180 		kvm_x86_ops->sync_pir_to_irr(vcpu);
8181 
8182 	if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
8183 	    || need_resched() || signal_pending(current)) {
8184 		vcpu->mode = OUTSIDE_GUEST_MODE;
8185 		smp_wmb();
8186 		local_irq_enable();
8187 		preempt_enable();
8188 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8189 		r = 1;
8190 		goto cancel_injection;
8191 	}
8192 
8193 	if (req_immediate_exit) {
8194 		kvm_make_request(KVM_REQ_EVENT, vcpu);
8195 		kvm_x86_ops->request_immediate_exit(vcpu);
8196 	}
8197 
8198 	trace_kvm_entry(vcpu->vcpu_id);
8199 	guest_enter_irqoff();
8200 
8201 	/* The preempt notifier should have taken care of the FPU already.  */
8202 	WARN_ON_ONCE(test_thread_flag(TIF_NEED_FPU_LOAD));
8203 
8204 	if (unlikely(vcpu->arch.switch_db_regs)) {
8205 		set_debugreg(0, 7);
8206 		set_debugreg(vcpu->arch.eff_db[0], 0);
8207 		set_debugreg(vcpu->arch.eff_db[1], 1);
8208 		set_debugreg(vcpu->arch.eff_db[2], 2);
8209 		set_debugreg(vcpu->arch.eff_db[3], 3);
8210 		set_debugreg(vcpu->arch.dr6, 6);
8211 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8212 	}
8213 
8214 	kvm_x86_ops->run(vcpu);
8215 
8216 	/*
8217 	 * Do this here before restoring debug registers on the host.  And
8218 	 * since we do this before handling the vmexit, a DR access vmexit
8219 	 * can (a) read the correct value of the debug registers, (b) set
8220 	 * KVM_DEBUGREG_WONT_EXIT again.
8221 	 */
8222 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8223 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8224 		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
8225 		kvm_update_dr0123(vcpu);
8226 		kvm_update_dr6(vcpu);
8227 		kvm_update_dr7(vcpu);
8228 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8229 	}
8230 
8231 	/*
8232 	 * If the guest has used debug registers, at least dr7
8233 	 * will be disabled while returning to the host.
8234 	 * If we don't have active breakpoints in the host, we don't
8235 	 * care about the messed up debug address registers. But if
8236 	 * we have some of them active, restore the old state.
8237 	 */
8238 	if (hw_breakpoint_active())
8239 		hw_breakpoint_restore();
8240 
8241 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8242 
8243 	vcpu->mode = OUTSIDE_GUEST_MODE;
8244 	smp_wmb();
8245 
8246 	kvm_x86_ops->handle_exit_irqoff(vcpu);
8247 
8248 	/*
8249 	 * Consume any pending interrupts, including the possible source of
8250 	 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8251 	 * An instruction is required after local_irq_enable() to fully unblock
8252 	 * interrupts on processors that implement an interrupt shadow, the
8253 	 * stat.exits increment will do nicely.
8254 	 */
8255 	kvm_before_interrupt(vcpu);
8256 	local_irq_enable();
8257 	++vcpu->stat.exits;
8258 	local_irq_disable();
8259 	kvm_after_interrupt(vcpu);
8260 
8261 	guest_exit_irqoff();
8262 	if (lapic_in_kernel(vcpu)) {
8263 		s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8264 		if (delta != S64_MIN) {
8265 			trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8266 			vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8267 		}
8268 	}
8269 
8270 	local_irq_enable();
8271 	preempt_enable();
8272 
8273 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8274 
8275 	/*
8276 	 * Profile KVM exit RIPs:
8277 	 */
8278 	if (unlikely(prof_on == KVM_PROFILING)) {
8279 		unsigned long rip = kvm_rip_read(vcpu);
8280 		profile_hit(KVM_PROFILING, (void *)rip);
8281 	}
8282 
8283 	if (unlikely(vcpu->arch.tsc_always_catchup))
8284 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8285 
8286 	if (vcpu->arch.apic_attention)
8287 		kvm_lapic_sync_from_vapic(vcpu);
8288 
8289 	vcpu->arch.gpa_available = false;
8290 	r = kvm_x86_ops->handle_exit(vcpu);
8291 	return r;
8292 
8293 cancel_injection:
8294 	kvm_x86_ops->cancel_injection(vcpu);
8295 	if (unlikely(vcpu->arch.apic_attention))
8296 		kvm_lapic_sync_from_vapic(vcpu);
8297 out:
8298 	return r;
8299 }
8300 
8301 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8302 {
8303 	if (!kvm_arch_vcpu_runnable(vcpu) &&
8304 	    (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
8305 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8306 		kvm_vcpu_block(vcpu);
8307 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8308 
8309 		if (kvm_x86_ops->post_block)
8310 			kvm_x86_ops->post_block(vcpu);
8311 
8312 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8313 			return 1;
8314 	}
8315 
8316 	kvm_apic_accept_events(vcpu);
8317 	switch(vcpu->arch.mp_state) {
8318 	case KVM_MP_STATE_HALTED:
8319 		vcpu->arch.pv.pv_unhalted = false;
8320 		vcpu->arch.mp_state =
8321 			KVM_MP_STATE_RUNNABLE;
8322 		/* fall through */
8323 	case KVM_MP_STATE_RUNNABLE:
8324 		vcpu->arch.apf.halted = false;
8325 		break;
8326 	case KVM_MP_STATE_INIT_RECEIVED:
8327 		break;
8328 	default:
8329 		return -EINTR;
8330 		break;
8331 	}
8332 	return 1;
8333 }
8334 
8335 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8336 {
8337 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8338 		kvm_x86_ops->check_nested_events(vcpu, false);
8339 
8340 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8341 		!vcpu->arch.apf.halted);
8342 }
8343 
8344 static int vcpu_run(struct kvm_vcpu *vcpu)
8345 {
8346 	int r;
8347 	struct kvm *kvm = vcpu->kvm;
8348 
8349 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8350 	vcpu->arch.l1tf_flush_l1d = true;
8351 
8352 	for (;;) {
8353 		if (kvm_vcpu_running(vcpu)) {
8354 			r = vcpu_enter_guest(vcpu);
8355 		} else {
8356 			r = vcpu_block(kvm, vcpu);
8357 		}
8358 
8359 		if (r <= 0)
8360 			break;
8361 
8362 		kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8363 		if (kvm_cpu_has_pending_timer(vcpu))
8364 			kvm_inject_pending_timer_irqs(vcpu);
8365 
8366 		if (dm_request_for_irq_injection(vcpu) &&
8367 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8368 			r = 0;
8369 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8370 			++vcpu->stat.request_irq_exits;
8371 			break;
8372 		}
8373 
8374 		kvm_check_async_pf_completion(vcpu);
8375 
8376 		if (signal_pending(current)) {
8377 			r = -EINTR;
8378 			vcpu->run->exit_reason = KVM_EXIT_INTR;
8379 			++vcpu->stat.signal_exits;
8380 			break;
8381 		}
8382 		if (need_resched()) {
8383 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8384 			cond_resched();
8385 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8386 		}
8387 	}
8388 
8389 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8390 
8391 	return r;
8392 }
8393 
8394 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8395 {
8396 	int r;
8397 
8398 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8399 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8400 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8401 	return r;
8402 }
8403 
8404 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8405 {
8406 	BUG_ON(!vcpu->arch.pio.count);
8407 
8408 	return complete_emulated_io(vcpu);
8409 }
8410 
8411 /*
8412  * Implements the following, as a state machine:
8413  *
8414  * read:
8415  *   for each fragment
8416  *     for each mmio piece in the fragment
8417  *       write gpa, len
8418  *       exit
8419  *       copy data
8420  *   execute insn
8421  *
8422  * write:
8423  *   for each fragment
8424  *     for each mmio piece in the fragment
8425  *       write gpa, len
8426  *       copy data
8427  *       exit
8428  */
8429 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8430 {
8431 	struct kvm_run *run = vcpu->run;
8432 	struct kvm_mmio_fragment *frag;
8433 	unsigned len;
8434 
8435 	BUG_ON(!vcpu->mmio_needed);
8436 
8437 	/* Complete previous fragment */
8438 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8439 	len = min(8u, frag->len);
8440 	if (!vcpu->mmio_is_write)
8441 		memcpy(frag->data, run->mmio.data, len);
8442 
8443 	if (frag->len <= 8) {
8444 		/* Switch to the next fragment. */
8445 		frag++;
8446 		vcpu->mmio_cur_fragment++;
8447 	} else {
8448 		/* Go forward to the next mmio piece. */
8449 		frag->data += len;
8450 		frag->gpa += len;
8451 		frag->len -= len;
8452 	}
8453 
8454 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8455 		vcpu->mmio_needed = 0;
8456 
8457 		/* FIXME: return into emulator if single-stepping.  */
8458 		if (vcpu->mmio_is_write)
8459 			return 1;
8460 		vcpu->mmio_read_completed = 1;
8461 		return complete_emulated_io(vcpu);
8462 	}
8463 
8464 	run->exit_reason = KVM_EXIT_MMIO;
8465 	run->mmio.phys_addr = frag->gpa;
8466 	if (vcpu->mmio_is_write)
8467 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8468 	run->mmio.len = min(8u, frag->len);
8469 	run->mmio.is_write = vcpu->mmio_is_write;
8470 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8471 	return 0;
8472 }
8473 
8474 /* Swap (qemu) user FPU context for the guest FPU context. */
8475 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8476 {
8477 	fpregs_lock();
8478 
8479 	copy_fpregs_to_fpstate(vcpu->arch.user_fpu);
8480 	/* PKRU is separately restored in kvm_x86_ops->run.  */
8481 	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8482 				~XFEATURE_MASK_PKRU);
8483 
8484 	fpregs_mark_activate();
8485 	fpregs_unlock();
8486 
8487 	trace_kvm_fpu(1);
8488 }
8489 
8490 /* When vcpu_run ends, restore user space FPU context. */
8491 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8492 {
8493 	fpregs_lock();
8494 
8495 	copy_fpregs_to_fpstate(vcpu->arch.guest_fpu);
8496 	copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
8497 
8498 	fpregs_mark_activate();
8499 	fpregs_unlock();
8500 
8501 	++vcpu->stat.fpu_reload;
8502 	trace_kvm_fpu(0);
8503 }
8504 
8505 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8506 {
8507 	int r;
8508 
8509 	vcpu_load(vcpu);
8510 	kvm_sigset_activate(vcpu);
8511 	kvm_load_guest_fpu(vcpu);
8512 
8513 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8514 		if (kvm_run->immediate_exit) {
8515 			r = -EINTR;
8516 			goto out;
8517 		}
8518 		kvm_vcpu_block(vcpu);
8519 		kvm_apic_accept_events(vcpu);
8520 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8521 		r = -EAGAIN;
8522 		if (signal_pending(current)) {
8523 			r = -EINTR;
8524 			vcpu->run->exit_reason = KVM_EXIT_INTR;
8525 			++vcpu->stat.signal_exits;
8526 		}
8527 		goto out;
8528 	}
8529 
8530 	if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8531 		r = -EINVAL;
8532 		goto out;
8533 	}
8534 
8535 	if (vcpu->run->kvm_dirty_regs) {
8536 		r = sync_regs(vcpu);
8537 		if (r != 0)
8538 			goto out;
8539 	}
8540 
8541 	/* re-sync apic's tpr */
8542 	if (!lapic_in_kernel(vcpu)) {
8543 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8544 			r = -EINVAL;
8545 			goto out;
8546 		}
8547 	}
8548 
8549 	if (unlikely(vcpu->arch.complete_userspace_io)) {
8550 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8551 		vcpu->arch.complete_userspace_io = NULL;
8552 		r = cui(vcpu);
8553 		if (r <= 0)
8554 			goto out;
8555 	} else
8556 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8557 
8558 	if (kvm_run->immediate_exit)
8559 		r = -EINTR;
8560 	else
8561 		r = vcpu_run(vcpu);
8562 
8563 out:
8564 	kvm_put_guest_fpu(vcpu);
8565 	if (vcpu->run->kvm_valid_regs)
8566 		store_regs(vcpu);
8567 	post_kvm_run_save(vcpu);
8568 	kvm_sigset_deactivate(vcpu);
8569 
8570 	vcpu_put(vcpu);
8571 	return r;
8572 }
8573 
8574 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8575 {
8576 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8577 		/*
8578 		 * We are here if userspace calls get_regs() in the middle of
8579 		 * instruction emulation. Registers state needs to be copied
8580 		 * back from emulation context to vcpu. Userspace shouldn't do
8581 		 * that usually, but some bad designed PV devices (vmware
8582 		 * backdoor interface) need this to work
8583 		 */
8584 		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8585 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8586 	}
8587 	regs->rax = kvm_rax_read(vcpu);
8588 	regs->rbx = kvm_rbx_read(vcpu);
8589 	regs->rcx = kvm_rcx_read(vcpu);
8590 	regs->rdx = kvm_rdx_read(vcpu);
8591 	regs->rsi = kvm_rsi_read(vcpu);
8592 	regs->rdi = kvm_rdi_read(vcpu);
8593 	regs->rsp = kvm_rsp_read(vcpu);
8594 	regs->rbp = kvm_rbp_read(vcpu);
8595 #ifdef CONFIG_X86_64
8596 	regs->r8 = kvm_r8_read(vcpu);
8597 	regs->r9 = kvm_r9_read(vcpu);
8598 	regs->r10 = kvm_r10_read(vcpu);
8599 	regs->r11 = kvm_r11_read(vcpu);
8600 	regs->r12 = kvm_r12_read(vcpu);
8601 	regs->r13 = kvm_r13_read(vcpu);
8602 	regs->r14 = kvm_r14_read(vcpu);
8603 	regs->r15 = kvm_r15_read(vcpu);
8604 #endif
8605 
8606 	regs->rip = kvm_rip_read(vcpu);
8607 	regs->rflags = kvm_get_rflags(vcpu);
8608 }
8609 
8610 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8611 {
8612 	vcpu_load(vcpu);
8613 	__get_regs(vcpu, regs);
8614 	vcpu_put(vcpu);
8615 	return 0;
8616 }
8617 
8618 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8619 {
8620 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8621 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8622 
8623 	kvm_rax_write(vcpu, regs->rax);
8624 	kvm_rbx_write(vcpu, regs->rbx);
8625 	kvm_rcx_write(vcpu, regs->rcx);
8626 	kvm_rdx_write(vcpu, regs->rdx);
8627 	kvm_rsi_write(vcpu, regs->rsi);
8628 	kvm_rdi_write(vcpu, regs->rdi);
8629 	kvm_rsp_write(vcpu, regs->rsp);
8630 	kvm_rbp_write(vcpu, regs->rbp);
8631 #ifdef CONFIG_X86_64
8632 	kvm_r8_write(vcpu, regs->r8);
8633 	kvm_r9_write(vcpu, regs->r9);
8634 	kvm_r10_write(vcpu, regs->r10);
8635 	kvm_r11_write(vcpu, regs->r11);
8636 	kvm_r12_write(vcpu, regs->r12);
8637 	kvm_r13_write(vcpu, regs->r13);
8638 	kvm_r14_write(vcpu, regs->r14);
8639 	kvm_r15_write(vcpu, regs->r15);
8640 #endif
8641 
8642 	kvm_rip_write(vcpu, regs->rip);
8643 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8644 
8645 	vcpu->arch.exception.pending = false;
8646 
8647 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8648 }
8649 
8650 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8651 {
8652 	vcpu_load(vcpu);
8653 	__set_regs(vcpu, regs);
8654 	vcpu_put(vcpu);
8655 	return 0;
8656 }
8657 
8658 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8659 {
8660 	struct kvm_segment cs;
8661 
8662 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8663 	*db = cs.db;
8664 	*l = cs.l;
8665 }
8666 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8667 
8668 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8669 {
8670 	struct desc_ptr dt;
8671 
8672 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8673 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8674 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8675 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8676 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8677 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8678 
8679 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8680 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8681 
8682 	kvm_x86_ops->get_idt(vcpu, &dt);
8683 	sregs->idt.limit = dt.size;
8684 	sregs->idt.base = dt.address;
8685 	kvm_x86_ops->get_gdt(vcpu, &dt);
8686 	sregs->gdt.limit = dt.size;
8687 	sregs->gdt.base = dt.address;
8688 
8689 	sregs->cr0 = kvm_read_cr0(vcpu);
8690 	sregs->cr2 = vcpu->arch.cr2;
8691 	sregs->cr3 = kvm_read_cr3(vcpu);
8692 	sregs->cr4 = kvm_read_cr4(vcpu);
8693 	sregs->cr8 = kvm_get_cr8(vcpu);
8694 	sregs->efer = vcpu->arch.efer;
8695 	sregs->apic_base = kvm_get_apic_base(vcpu);
8696 
8697 	memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8698 
8699 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8700 		set_bit(vcpu->arch.interrupt.nr,
8701 			(unsigned long *)sregs->interrupt_bitmap);
8702 }
8703 
8704 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8705 				  struct kvm_sregs *sregs)
8706 {
8707 	vcpu_load(vcpu);
8708 	__get_sregs(vcpu, sregs);
8709 	vcpu_put(vcpu);
8710 	return 0;
8711 }
8712 
8713 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8714 				    struct kvm_mp_state *mp_state)
8715 {
8716 	vcpu_load(vcpu);
8717 
8718 	kvm_apic_accept_events(vcpu);
8719 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8720 					vcpu->arch.pv.pv_unhalted)
8721 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8722 	else
8723 		mp_state->mp_state = vcpu->arch.mp_state;
8724 
8725 	vcpu_put(vcpu);
8726 	return 0;
8727 }
8728 
8729 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8730 				    struct kvm_mp_state *mp_state)
8731 {
8732 	int ret = -EINVAL;
8733 
8734 	vcpu_load(vcpu);
8735 
8736 	if (!lapic_in_kernel(vcpu) &&
8737 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8738 		goto out;
8739 
8740 	/*
8741 	 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
8742 	 * INIT state; latched init should be reported using
8743 	 * KVM_SET_VCPU_EVENTS, so reject it here.
8744 	 */
8745 	if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
8746 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8747 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8748 		goto out;
8749 
8750 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8751 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8752 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8753 	} else
8754 		vcpu->arch.mp_state = mp_state->mp_state;
8755 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8756 
8757 	ret = 0;
8758 out:
8759 	vcpu_put(vcpu);
8760 	return ret;
8761 }
8762 
8763 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8764 		    int reason, bool has_error_code, u32 error_code)
8765 {
8766 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8767 	int ret;
8768 
8769 	init_emulate_ctxt(vcpu);
8770 
8771 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8772 				   has_error_code, error_code);
8773 	if (ret) {
8774 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8775 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
8776 		vcpu->run->internal.ndata = 0;
8777 		return 0;
8778 	}
8779 
8780 	kvm_rip_write(vcpu, ctxt->eip);
8781 	kvm_set_rflags(vcpu, ctxt->eflags);
8782 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8783 	return 1;
8784 }
8785 EXPORT_SYMBOL_GPL(kvm_task_switch);
8786 
8787 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8788 {
8789 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8790 		/*
8791 		 * When EFER.LME and CR0.PG are set, the processor is in
8792 		 * 64-bit mode (though maybe in a 32-bit code segment).
8793 		 * CR4.PAE and EFER.LMA must be set.
8794 		 */
8795 		if (!(sregs->cr4 & X86_CR4_PAE)
8796 		    || !(sregs->efer & EFER_LMA))
8797 			return -EINVAL;
8798 	} else {
8799 		/*
8800 		 * Not in 64-bit mode: EFER.LMA is clear and the code
8801 		 * segment cannot be 64-bit.
8802 		 */
8803 		if (sregs->efer & EFER_LMA || sregs->cs.l)
8804 			return -EINVAL;
8805 	}
8806 
8807 	return kvm_valid_cr4(vcpu, sregs->cr4);
8808 }
8809 
8810 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8811 {
8812 	struct msr_data apic_base_msr;
8813 	int mmu_reset_needed = 0;
8814 	int cpuid_update_needed = 0;
8815 	int pending_vec, max_bits, idx;
8816 	struct desc_ptr dt;
8817 	int ret = -EINVAL;
8818 
8819 	if (kvm_valid_sregs(vcpu, sregs))
8820 		goto out;
8821 
8822 	apic_base_msr.data = sregs->apic_base;
8823 	apic_base_msr.host_initiated = true;
8824 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
8825 		goto out;
8826 
8827 	dt.size = sregs->idt.limit;
8828 	dt.address = sregs->idt.base;
8829 	kvm_x86_ops->set_idt(vcpu, &dt);
8830 	dt.size = sregs->gdt.limit;
8831 	dt.address = sregs->gdt.base;
8832 	kvm_x86_ops->set_gdt(vcpu, &dt);
8833 
8834 	vcpu->arch.cr2 = sregs->cr2;
8835 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8836 	vcpu->arch.cr3 = sregs->cr3;
8837 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
8838 
8839 	kvm_set_cr8(vcpu, sregs->cr8);
8840 
8841 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8842 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
8843 
8844 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8845 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8846 	vcpu->arch.cr0 = sregs->cr0;
8847 
8848 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8849 	cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8850 				(X86_CR4_OSXSAVE | X86_CR4_PKE));
8851 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8852 	if (cpuid_update_needed)
8853 		kvm_update_cpuid(vcpu);
8854 
8855 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8856 	if (is_pae_paging(vcpu)) {
8857 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8858 		mmu_reset_needed = 1;
8859 	}
8860 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8861 
8862 	if (mmu_reset_needed)
8863 		kvm_mmu_reset_context(vcpu);
8864 
8865 	max_bits = KVM_NR_INTERRUPTS;
8866 	pending_vec = find_first_bit(
8867 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
8868 	if (pending_vec < max_bits) {
8869 		kvm_queue_interrupt(vcpu, pending_vec, false);
8870 		pr_debug("Set back pending irq %d\n", pending_vec);
8871 	}
8872 
8873 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8874 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8875 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8876 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8877 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8878 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8879 
8880 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8881 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8882 
8883 	update_cr8_intercept(vcpu);
8884 
8885 	/* Older userspace won't unhalt the vcpu on reset. */
8886 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8887 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8888 	    !is_protmode(vcpu))
8889 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8890 
8891 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8892 
8893 	ret = 0;
8894 out:
8895 	return ret;
8896 }
8897 
8898 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8899 				  struct kvm_sregs *sregs)
8900 {
8901 	int ret;
8902 
8903 	vcpu_load(vcpu);
8904 	ret = __set_sregs(vcpu, sregs);
8905 	vcpu_put(vcpu);
8906 	return ret;
8907 }
8908 
8909 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8910 					struct kvm_guest_debug *dbg)
8911 {
8912 	unsigned long rflags;
8913 	int i, r;
8914 
8915 	vcpu_load(vcpu);
8916 
8917 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8918 		r = -EBUSY;
8919 		if (vcpu->arch.exception.pending)
8920 			goto out;
8921 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8922 			kvm_queue_exception(vcpu, DB_VECTOR);
8923 		else
8924 			kvm_queue_exception(vcpu, BP_VECTOR);
8925 	}
8926 
8927 	/*
8928 	 * Read rflags as long as potentially injected trace flags are still
8929 	 * filtered out.
8930 	 */
8931 	rflags = kvm_get_rflags(vcpu);
8932 
8933 	vcpu->guest_debug = dbg->control;
8934 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8935 		vcpu->guest_debug = 0;
8936 
8937 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8938 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
8939 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8940 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8941 	} else {
8942 		for (i = 0; i < KVM_NR_DB_REGS; i++)
8943 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8944 	}
8945 	kvm_update_dr7(vcpu);
8946 
8947 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8948 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8949 			get_segment_base(vcpu, VCPU_SREG_CS);
8950 
8951 	/*
8952 	 * Trigger an rflags update that will inject or remove the trace
8953 	 * flags.
8954 	 */
8955 	kvm_set_rflags(vcpu, rflags);
8956 
8957 	kvm_x86_ops->update_bp_intercept(vcpu);
8958 
8959 	r = 0;
8960 
8961 out:
8962 	vcpu_put(vcpu);
8963 	return r;
8964 }
8965 
8966 /*
8967  * Translate a guest virtual address to a guest physical address.
8968  */
8969 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8970 				    struct kvm_translation *tr)
8971 {
8972 	unsigned long vaddr = tr->linear_address;
8973 	gpa_t gpa;
8974 	int idx;
8975 
8976 	vcpu_load(vcpu);
8977 
8978 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8979 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8980 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8981 	tr->physical_address = gpa;
8982 	tr->valid = gpa != UNMAPPED_GVA;
8983 	tr->writeable = 1;
8984 	tr->usermode = 0;
8985 
8986 	vcpu_put(vcpu);
8987 	return 0;
8988 }
8989 
8990 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8991 {
8992 	struct fxregs_state *fxsave;
8993 
8994 	vcpu_load(vcpu);
8995 
8996 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8997 	memcpy(fpu->fpr, fxsave->st_space, 128);
8998 	fpu->fcw = fxsave->cwd;
8999 	fpu->fsw = fxsave->swd;
9000 	fpu->ftwx = fxsave->twd;
9001 	fpu->last_opcode = fxsave->fop;
9002 	fpu->last_ip = fxsave->rip;
9003 	fpu->last_dp = fxsave->rdp;
9004 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
9005 
9006 	vcpu_put(vcpu);
9007 	return 0;
9008 }
9009 
9010 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9011 {
9012 	struct fxregs_state *fxsave;
9013 
9014 	vcpu_load(vcpu);
9015 
9016 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9017 
9018 	memcpy(fxsave->st_space, fpu->fpr, 128);
9019 	fxsave->cwd = fpu->fcw;
9020 	fxsave->swd = fpu->fsw;
9021 	fxsave->twd = fpu->ftwx;
9022 	fxsave->fop = fpu->last_opcode;
9023 	fxsave->rip = fpu->last_ip;
9024 	fxsave->rdp = fpu->last_dp;
9025 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
9026 
9027 	vcpu_put(vcpu);
9028 	return 0;
9029 }
9030 
9031 static void store_regs(struct kvm_vcpu *vcpu)
9032 {
9033 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
9034 
9035 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
9036 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
9037 
9038 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
9039 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
9040 
9041 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
9042 		kvm_vcpu_ioctl_x86_get_vcpu_events(
9043 				vcpu, &vcpu->run->s.regs.events);
9044 }
9045 
9046 static int sync_regs(struct kvm_vcpu *vcpu)
9047 {
9048 	if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
9049 		return -EINVAL;
9050 
9051 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
9052 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
9053 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
9054 	}
9055 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
9056 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
9057 			return -EINVAL;
9058 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
9059 	}
9060 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
9061 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9062 				vcpu, &vcpu->run->s.regs.events))
9063 			return -EINVAL;
9064 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9065 	}
9066 
9067 	return 0;
9068 }
9069 
9070 static void fx_init(struct kvm_vcpu *vcpu)
9071 {
9072 	fpstate_init(&vcpu->arch.guest_fpu->state);
9073 	if (boot_cpu_has(X86_FEATURE_XSAVES))
9074 		vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9075 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
9076 
9077 	/*
9078 	 * Ensure guest xcr0 is valid for loading
9079 	 */
9080 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9081 
9082 	vcpu->arch.cr0 |= X86_CR0_ET;
9083 }
9084 
9085 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
9086 {
9087 	void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
9088 
9089 	kvmclock_reset(vcpu);
9090 
9091 	kvm_x86_ops->vcpu_free(vcpu);
9092 	free_cpumask_var(wbinvd_dirty_mask);
9093 }
9094 
9095 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
9096 						unsigned int id)
9097 {
9098 	struct kvm_vcpu *vcpu;
9099 
9100 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9101 		printk_once(KERN_WARNING
9102 		"kvm: SMP vm created on host with unstable TSC; "
9103 		"guest TSC will not be reliable\n");
9104 
9105 	vcpu = kvm_x86_ops->vcpu_create(kvm, id);
9106 
9107 	return vcpu;
9108 }
9109 
9110 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
9111 {
9112 	vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
9113 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
9114 	kvm_vcpu_mtrr_init(vcpu);
9115 	vcpu_load(vcpu);
9116 	kvm_vcpu_reset(vcpu, false);
9117 	kvm_init_mmu(vcpu, false);
9118 	vcpu_put(vcpu);
9119 	return 0;
9120 }
9121 
9122 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
9123 {
9124 	struct msr_data msr;
9125 	struct kvm *kvm = vcpu->kvm;
9126 
9127 	kvm_hv_vcpu_postcreate(vcpu);
9128 
9129 	if (mutex_lock_killable(&vcpu->mutex))
9130 		return;
9131 	vcpu_load(vcpu);
9132 	msr.data = 0x0;
9133 	msr.index = MSR_IA32_TSC;
9134 	msr.host_initiated = true;
9135 	kvm_write_tsc(vcpu, &msr);
9136 	vcpu_put(vcpu);
9137 
9138 	/* poll control enabled by default */
9139 	vcpu->arch.msr_kvm_poll_control = 1;
9140 
9141 	mutex_unlock(&vcpu->mutex);
9142 
9143 	if (!kvmclock_periodic_sync)
9144 		return;
9145 
9146 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9147 					KVMCLOCK_SYNC_PERIOD);
9148 }
9149 
9150 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9151 {
9152 	vcpu->arch.apf.msr_val = 0;
9153 
9154 	vcpu_load(vcpu);
9155 	kvm_mmu_unload(vcpu);
9156 	vcpu_put(vcpu);
9157 
9158 	kvm_x86_ops->vcpu_free(vcpu);
9159 }
9160 
9161 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
9162 {
9163 	kvm_lapic_reset(vcpu, init_event);
9164 
9165 	vcpu->arch.hflags = 0;
9166 
9167 	vcpu->arch.smi_pending = 0;
9168 	vcpu->arch.smi_count = 0;
9169 	atomic_set(&vcpu->arch.nmi_queued, 0);
9170 	vcpu->arch.nmi_pending = 0;
9171 	vcpu->arch.nmi_injected = false;
9172 	kvm_clear_interrupt_queue(vcpu);
9173 	kvm_clear_exception_queue(vcpu);
9174 	vcpu->arch.exception.pending = false;
9175 
9176 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
9177 	kvm_update_dr0123(vcpu);
9178 	vcpu->arch.dr6 = DR6_INIT;
9179 	kvm_update_dr6(vcpu);
9180 	vcpu->arch.dr7 = DR7_FIXED_1;
9181 	kvm_update_dr7(vcpu);
9182 
9183 	vcpu->arch.cr2 = 0;
9184 
9185 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9186 	vcpu->arch.apf.msr_val = 0;
9187 	vcpu->arch.st.msr_val = 0;
9188 
9189 	kvmclock_reset(vcpu);
9190 
9191 	kvm_clear_async_pf_completion_queue(vcpu);
9192 	kvm_async_pf_hash_reset(vcpu);
9193 	vcpu->arch.apf.halted = false;
9194 
9195 	if (kvm_mpx_supported()) {
9196 		void *mpx_state_buffer;
9197 
9198 		/*
9199 		 * To avoid have the INIT path from kvm_apic_has_events() that be
9200 		 * called with loaded FPU and does not let userspace fix the state.
9201 		 */
9202 		if (init_event)
9203 			kvm_put_guest_fpu(vcpu);
9204 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9205 					XFEATURE_BNDREGS);
9206 		if (mpx_state_buffer)
9207 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
9208 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9209 					XFEATURE_BNDCSR);
9210 		if (mpx_state_buffer)
9211 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
9212 		if (init_event)
9213 			kvm_load_guest_fpu(vcpu);
9214 	}
9215 
9216 	if (!init_event) {
9217 		kvm_pmu_reset(vcpu);
9218 		vcpu->arch.smbase = 0x30000;
9219 
9220 		vcpu->arch.msr_misc_features_enables = 0;
9221 
9222 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9223 	}
9224 
9225 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
9226 	vcpu->arch.regs_avail = ~0;
9227 	vcpu->arch.regs_dirty = ~0;
9228 
9229 	vcpu->arch.ia32_xss = 0;
9230 
9231 	kvm_x86_ops->vcpu_reset(vcpu, init_event);
9232 }
9233 
9234 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
9235 {
9236 	struct kvm_segment cs;
9237 
9238 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9239 	cs.selector = vector << 8;
9240 	cs.base = vector << 12;
9241 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9242 	kvm_rip_write(vcpu, 0);
9243 }
9244 
9245 int kvm_arch_hardware_enable(void)
9246 {
9247 	struct kvm *kvm;
9248 	struct kvm_vcpu *vcpu;
9249 	int i;
9250 	int ret;
9251 	u64 local_tsc;
9252 	u64 max_tsc = 0;
9253 	bool stable, backwards_tsc = false;
9254 
9255 	kvm_shared_msr_cpu_online();
9256 	ret = kvm_x86_ops->hardware_enable();
9257 	if (ret != 0)
9258 		return ret;
9259 
9260 	local_tsc = rdtsc();
9261 	stable = !kvm_check_tsc_unstable();
9262 	list_for_each_entry(kvm, &vm_list, vm_list) {
9263 		kvm_for_each_vcpu(i, vcpu, kvm) {
9264 			if (!stable && vcpu->cpu == smp_processor_id())
9265 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9266 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
9267 				backwards_tsc = true;
9268 				if (vcpu->arch.last_host_tsc > max_tsc)
9269 					max_tsc = vcpu->arch.last_host_tsc;
9270 			}
9271 		}
9272 	}
9273 
9274 	/*
9275 	 * Sometimes, even reliable TSCs go backwards.  This happens on
9276 	 * platforms that reset TSC during suspend or hibernate actions, but
9277 	 * maintain synchronization.  We must compensate.  Fortunately, we can
9278 	 * detect that condition here, which happens early in CPU bringup,
9279 	 * before any KVM threads can be running.  Unfortunately, we can't
9280 	 * bring the TSCs fully up to date with real time, as we aren't yet far
9281 	 * enough into CPU bringup that we know how much real time has actually
9282 	 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
9283 	 * variables that haven't been updated yet.
9284 	 *
9285 	 * So we simply find the maximum observed TSC above, then record the
9286 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
9287 	 * the adjustment will be applied.  Note that we accumulate
9288 	 * adjustments, in case multiple suspend cycles happen before some VCPU
9289 	 * gets a chance to run again.  In the event that no KVM threads get a
9290 	 * chance to run, we will miss the entire elapsed period, as we'll have
9291 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
9292 	 * loose cycle time.  This isn't too big a deal, since the loss will be
9293 	 * uniform across all VCPUs (not to mention the scenario is extremely
9294 	 * unlikely). It is possible that a second hibernate recovery happens
9295 	 * much faster than a first, causing the observed TSC here to be
9296 	 * smaller; this would require additional padding adjustment, which is
9297 	 * why we set last_host_tsc to the local tsc observed here.
9298 	 *
9299 	 * N.B. - this code below runs only on platforms with reliable TSC,
9300 	 * as that is the only way backwards_tsc is set above.  Also note
9301 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
9302 	 * have the same delta_cyc adjustment applied if backwards_tsc
9303 	 * is detected.  Note further, this adjustment is only done once,
9304 	 * as we reset last_host_tsc on all VCPUs to stop this from being
9305 	 * called multiple times (one for each physical CPU bringup).
9306 	 *
9307 	 * Platforms with unreliable TSCs don't have to deal with this, they
9308 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
9309 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
9310 	 * guarantee that they stay in perfect synchronization.
9311 	 */
9312 	if (backwards_tsc) {
9313 		u64 delta_cyc = max_tsc - local_tsc;
9314 		list_for_each_entry(kvm, &vm_list, vm_list) {
9315 			kvm->arch.backwards_tsc_observed = true;
9316 			kvm_for_each_vcpu(i, vcpu, kvm) {
9317 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
9318 				vcpu->arch.last_host_tsc = local_tsc;
9319 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9320 			}
9321 
9322 			/*
9323 			 * We have to disable TSC offset matching.. if you were
9324 			 * booting a VM while issuing an S4 host suspend....
9325 			 * you may have some problem.  Solving this issue is
9326 			 * left as an exercise to the reader.
9327 			 */
9328 			kvm->arch.last_tsc_nsec = 0;
9329 			kvm->arch.last_tsc_write = 0;
9330 		}
9331 
9332 	}
9333 	return 0;
9334 }
9335 
9336 void kvm_arch_hardware_disable(void)
9337 {
9338 	kvm_x86_ops->hardware_disable();
9339 	drop_user_return_notifiers();
9340 }
9341 
9342 int kvm_arch_hardware_setup(void)
9343 {
9344 	int r;
9345 
9346 	r = kvm_x86_ops->hardware_setup();
9347 	if (r != 0)
9348 		return r;
9349 
9350 	if (kvm_has_tsc_control) {
9351 		/*
9352 		 * Make sure the user can only configure tsc_khz values that
9353 		 * fit into a signed integer.
9354 		 * A min value is not calculated because it will always
9355 		 * be 1 on all machines.
9356 		 */
9357 		u64 max = min(0x7fffffffULL,
9358 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9359 		kvm_max_guest_tsc_khz = max;
9360 
9361 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9362 	}
9363 
9364 	if (boot_cpu_has(X86_FEATURE_XSAVES))
9365 		rdmsrl(MSR_IA32_XSS, host_xss);
9366 
9367 	kvm_init_msr_list();
9368 	return 0;
9369 }
9370 
9371 void kvm_arch_hardware_unsetup(void)
9372 {
9373 	kvm_x86_ops->hardware_unsetup();
9374 }
9375 
9376 int kvm_arch_check_processor_compat(void)
9377 {
9378 	return kvm_x86_ops->check_processor_compatibility();
9379 }
9380 
9381 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9382 {
9383 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9384 }
9385 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9386 
9387 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9388 {
9389 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9390 }
9391 
9392 struct static_key kvm_no_apic_vcpu __read_mostly;
9393 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9394 
9395 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
9396 {
9397 	struct page *page;
9398 	int r;
9399 
9400 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
9401 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9402 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9403 	else
9404 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9405 
9406 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9407 	if (!page) {
9408 		r = -ENOMEM;
9409 		goto fail;
9410 	}
9411 	vcpu->arch.pio_data = page_address(page);
9412 
9413 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
9414 
9415 	r = kvm_mmu_create(vcpu);
9416 	if (r < 0)
9417 		goto fail_free_pio_data;
9418 
9419 	if (irqchip_in_kernel(vcpu->kvm)) {
9420 		vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu->kvm);
9421 		r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9422 		if (r < 0)
9423 			goto fail_mmu_destroy;
9424 	} else
9425 		static_key_slow_inc(&kvm_no_apic_vcpu);
9426 
9427 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9428 				       GFP_KERNEL_ACCOUNT);
9429 	if (!vcpu->arch.mce_banks) {
9430 		r = -ENOMEM;
9431 		goto fail_free_lapic;
9432 	}
9433 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9434 
9435 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9436 				GFP_KERNEL_ACCOUNT)) {
9437 		r = -ENOMEM;
9438 		goto fail_free_mce_banks;
9439 	}
9440 
9441 	fx_init(vcpu);
9442 
9443 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9444 
9445 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9446 
9447 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9448 
9449 	kvm_async_pf_hash_reset(vcpu);
9450 	kvm_pmu_init(vcpu);
9451 
9452 	vcpu->arch.pending_external_vector = -1;
9453 	vcpu->arch.preempted_in_kernel = false;
9454 
9455 	kvm_hv_vcpu_init(vcpu);
9456 
9457 	return 0;
9458 
9459 fail_free_mce_banks:
9460 	kfree(vcpu->arch.mce_banks);
9461 fail_free_lapic:
9462 	kvm_free_lapic(vcpu);
9463 fail_mmu_destroy:
9464 	kvm_mmu_destroy(vcpu);
9465 fail_free_pio_data:
9466 	free_page((unsigned long)vcpu->arch.pio_data);
9467 fail:
9468 	return r;
9469 }
9470 
9471 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
9472 {
9473 	int idx;
9474 
9475 	kvm_hv_vcpu_uninit(vcpu);
9476 	kvm_pmu_destroy(vcpu);
9477 	kfree(vcpu->arch.mce_banks);
9478 	kvm_free_lapic(vcpu);
9479 	idx = srcu_read_lock(&vcpu->kvm->srcu);
9480 	kvm_mmu_destroy(vcpu);
9481 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
9482 	free_page((unsigned long)vcpu->arch.pio_data);
9483 	if (!lapic_in_kernel(vcpu))
9484 		static_key_slow_dec(&kvm_no_apic_vcpu);
9485 }
9486 
9487 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9488 {
9489 	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
9490 
9491 	vcpu->arch.l1tf_flush_l1d = true;
9492 	if (pmu->version && unlikely(pmu->event_count)) {
9493 		pmu->need_cleanup = true;
9494 		kvm_make_request(KVM_REQ_PMU, vcpu);
9495 	}
9496 	kvm_x86_ops->sched_in(vcpu, cpu);
9497 }
9498 
9499 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9500 {
9501 	if (type)
9502 		return -EINVAL;
9503 
9504 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9505 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9506 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9507 	INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
9508 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9509 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9510 
9511 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9512 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9513 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9514 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9515 		&kvm->arch.irq_sources_bitmap);
9516 
9517 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9518 	mutex_init(&kvm->arch.apic_map_lock);
9519 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9520 
9521 	kvm->arch.kvmclock_offset = -ktime_get_boottime_ns();
9522 	pvclock_update_vm_gtod_copy(kvm);
9523 
9524 	kvm->arch.guest_can_read_msr_platform_info = true;
9525 
9526 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9527 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9528 
9529 	kvm_hv_init_vm(kvm);
9530 	kvm_page_track_init(kvm);
9531 	kvm_mmu_init_vm(kvm);
9532 
9533 	return kvm_x86_ops->vm_init(kvm);
9534 }
9535 
9536 int kvm_arch_post_init_vm(struct kvm *kvm)
9537 {
9538 	return kvm_mmu_post_init_vm(kvm);
9539 }
9540 
9541 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9542 {
9543 	vcpu_load(vcpu);
9544 	kvm_mmu_unload(vcpu);
9545 	vcpu_put(vcpu);
9546 }
9547 
9548 static void kvm_free_vcpus(struct kvm *kvm)
9549 {
9550 	unsigned int i;
9551 	struct kvm_vcpu *vcpu;
9552 
9553 	/*
9554 	 * Unpin any mmu pages first.
9555 	 */
9556 	kvm_for_each_vcpu(i, vcpu, kvm) {
9557 		kvm_clear_async_pf_completion_queue(vcpu);
9558 		kvm_unload_vcpu_mmu(vcpu);
9559 	}
9560 	kvm_for_each_vcpu(i, vcpu, kvm)
9561 		kvm_arch_vcpu_free(vcpu);
9562 
9563 	mutex_lock(&kvm->lock);
9564 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9565 		kvm->vcpus[i] = NULL;
9566 
9567 	atomic_set(&kvm->online_vcpus, 0);
9568 	mutex_unlock(&kvm->lock);
9569 }
9570 
9571 void kvm_arch_sync_events(struct kvm *kvm)
9572 {
9573 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9574 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9575 	kvm_free_pit(kvm);
9576 }
9577 
9578 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9579 {
9580 	int i, r;
9581 	unsigned long hva;
9582 	struct kvm_memslots *slots = kvm_memslots(kvm);
9583 	struct kvm_memory_slot *slot, old;
9584 
9585 	/* Called with kvm->slots_lock held.  */
9586 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9587 		return -EINVAL;
9588 
9589 	slot = id_to_memslot(slots, id);
9590 	if (size) {
9591 		if (slot->npages)
9592 			return -EEXIST;
9593 
9594 		/*
9595 		 * MAP_SHARED to prevent internal slot pages from being moved
9596 		 * by fork()/COW.
9597 		 */
9598 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9599 			      MAP_SHARED | MAP_ANONYMOUS, 0);
9600 		if (IS_ERR((void *)hva))
9601 			return PTR_ERR((void *)hva);
9602 	} else {
9603 		if (!slot->npages)
9604 			return 0;
9605 
9606 		hva = 0;
9607 	}
9608 
9609 	old = *slot;
9610 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9611 		struct kvm_userspace_memory_region m;
9612 
9613 		m.slot = id | (i << 16);
9614 		m.flags = 0;
9615 		m.guest_phys_addr = gpa;
9616 		m.userspace_addr = hva;
9617 		m.memory_size = size;
9618 		r = __kvm_set_memory_region(kvm, &m);
9619 		if (r < 0)
9620 			return r;
9621 	}
9622 
9623 	if (!size)
9624 		vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9625 
9626 	return 0;
9627 }
9628 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9629 
9630 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9631 {
9632 	int r;
9633 
9634 	mutex_lock(&kvm->slots_lock);
9635 	r = __x86_set_memory_region(kvm, id, gpa, size);
9636 	mutex_unlock(&kvm->slots_lock);
9637 
9638 	return r;
9639 }
9640 EXPORT_SYMBOL_GPL(x86_set_memory_region);
9641 
9642 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
9643 {
9644 	kvm_mmu_pre_destroy_vm(kvm);
9645 }
9646 
9647 void kvm_arch_destroy_vm(struct kvm *kvm)
9648 {
9649 	if (current->mm == kvm->mm) {
9650 		/*
9651 		 * Free memory regions allocated on behalf of userspace,
9652 		 * unless the the memory map has changed due to process exit
9653 		 * or fd copying.
9654 		 */
9655 		x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9656 		x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9657 		x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9658 	}
9659 	if (kvm_x86_ops->vm_destroy)
9660 		kvm_x86_ops->vm_destroy(kvm);
9661 	kvm_pic_destroy(kvm);
9662 	kvm_ioapic_destroy(kvm);
9663 	kvm_free_vcpus(kvm);
9664 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9665 	kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
9666 	kvm_mmu_uninit_vm(kvm);
9667 	kvm_page_track_cleanup(kvm);
9668 	kvm_hv_destroy_vm(kvm);
9669 }
9670 
9671 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9672 			   struct kvm_memory_slot *dont)
9673 {
9674 	int i;
9675 
9676 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9677 		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9678 			kvfree(free->arch.rmap[i]);
9679 			free->arch.rmap[i] = NULL;
9680 		}
9681 		if (i == 0)
9682 			continue;
9683 
9684 		if (!dont || free->arch.lpage_info[i - 1] !=
9685 			     dont->arch.lpage_info[i - 1]) {
9686 			kvfree(free->arch.lpage_info[i - 1]);
9687 			free->arch.lpage_info[i - 1] = NULL;
9688 		}
9689 	}
9690 
9691 	kvm_page_track_free_memslot(free, dont);
9692 }
9693 
9694 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9695 			    unsigned long npages)
9696 {
9697 	int i;
9698 
9699 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9700 		struct kvm_lpage_info *linfo;
9701 		unsigned long ugfn;
9702 		int lpages;
9703 		int level = i + 1;
9704 
9705 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
9706 				      slot->base_gfn, level) + 1;
9707 
9708 		slot->arch.rmap[i] =
9709 			kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9710 				 GFP_KERNEL_ACCOUNT);
9711 		if (!slot->arch.rmap[i])
9712 			goto out_free;
9713 		if (i == 0)
9714 			continue;
9715 
9716 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
9717 		if (!linfo)
9718 			goto out_free;
9719 
9720 		slot->arch.lpage_info[i - 1] = linfo;
9721 
9722 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9723 			linfo[0].disallow_lpage = 1;
9724 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9725 			linfo[lpages - 1].disallow_lpage = 1;
9726 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
9727 		/*
9728 		 * If the gfn and userspace address are not aligned wrt each
9729 		 * other, or if explicitly asked to, disable large page
9730 		 * support for this slot
9731 		 */
9732 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9733 		    !kvm_largepages_enabled()) {
9734 			unsigned long j;
9735 
9736 			for (j = 0; j < lpages; ++j)
9737 				linfo[j].disallow_lpage = 1;
9738 		}
9739 	}
9740 
9741 	if (kvm_page_track_create_memslot(slot, npages))
9742 		goto out_free;
9743 
9744 	return 0;
9745 
9746 out_free:
9747 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9748 		kvfree(slot->arch.rmap[i]);
9749 		slot->arch.rmap[i] = NULL;
9750 		if (i == 0)
9751 			continue;
9752 
9753 		kvfree(slot->arch.lpage_info[i - 1]);
9754 		slot->arch.lpage_info[i - 1] = NULL;
9755 	}
9756 	return -ENOMEM;
9757 }
9758 
9759 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
9760 {
9761 	/*
9762 	 * memslots->generation has been incremented.
9763 	 * mmio generation may have reached its maximum value.
9764 	 */
9765 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
9766 }
9767 
9768 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9769 				struct kvm_memory_slot *memslot,
9770 				const struct kvm_userspace_memory_region *mem,
9771 				enum kvm_mr_change change)
9772 {
9773 	return 0;
9774 }
9775 
9776 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9777 				     struct kvm_memory_slot *new)
9778 {
9779 	/* Still write protect RO slot */
9780 	if (new->flags & KVM_MEM_READONLY) {
9781 		kvm_mmu_slot_remove_write_access(kvm, new);
9782 		return;
9783 	}
9784 
9785 	/*
9786 	 * Call kvm_x86_ops dirty logging hooks when they are valid.
9787 	 *
9788 	 * kvm_x86_ops->slot_disable_log_dirty is called when:
9789 	 *
9790 	 *  - KVM_MR_CREATE with dirty logging is disabled
9791 	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9792 	 *
9793 	 * The reason is, in case of PML, we need to set D-bit for any slots
9794 	 * with dirty logging disabled in order to eliminate unnecessary GPA
9795 	 * logging in PML buffer (and potential PML buffer full VMEXT). This
9796 	 * guarantees leaving PML enabled during guest's lifetime won't have
9797 	 * any additional overhead from PML when guest is running with dirty
9798 	 * logging disabled for memory slots.
9799 	 *
9800 	 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9801 	 * to dirty logging mode.
9802 	 *
9803 	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9804 	 *
9805 	 * In case of write protect:
9806 	 *
9807 	 * Write protect all pages for dirty logging.
9808 	 *
9809 	 * All the sptes including the large sptes which point to this
9810 	 * slot are set to readonly. We can not create any new large
9811 	 * spte on this slot until the end of the logging.
9812 	 *
9813 	 * See the comments in fast_page_fault().
9814 	 */
9815 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9816 		if (kvm_x86_ops->slot_enable_log_dirty)
9817 			kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9818 		else
9819 			kvm_mmu_slot_remove_write_access(kvm, new);
9820 	} else {
9821 		if (kvm_x86_ops->slot_disable_log_dirty)
9822 			kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9823 	}
9824 }
9825 
9826 void kvm_arch_commit_memory_region(struct kvm *kvm,
9827 				const struct kvm_userspace_memory_region *mem,
9828 				const struct kvm_memory_slot *old,
9829 				const struct kvm_memory_slot *new,
9830 				enum kvm_mr_change change)
9831 {
9832 	if (!kvm->arch.n_requested_mmu_pages)
9833 		kvm_mmu_change_mmu_pages(kvm,
9834 				kvm_mmu_calculate_default_mmu_pages(kvm));
9835 
9836 	/*
9837 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
9838 	 * sptes have to be split.  If live migration is successful, the guest
9839 	 * in the source machine will be destroyed and large sptes will be
9840 	 * created in the destination. However, if the guest continues to run
9841 	 * in the source machine (for example if live migration fails), small
9842 	 * sptes will remain around and cause bad performance.
9843 	 *
9844 	 * Scan sptes if dirty logging has been stopped, dropping those
9845 	 * which can be collapsed into a single large-page spte.  Later
9846 	 * page faults will create the large-page sptes.
9847 	 *
9848 	 * There is no need to do this in any of the following cases:
9849 	 * CREATE:	No dirty mappings will already exist.
9850 	 * MOVE/DELETE:	The old mappings will already have been cleaned up by
9851 	 *		kvm_arch_flush_shadow_memslot()
9852 	 */
9853 	if (change == KVM_MR_FLAGS_ONLY &&
9854 		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9855 		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9856 		kvm_mmu_zap_collapsible_sptes(kvm, new);
9857 
9858 	/*
9859 	 * Set up write protection and/or dirty logging for the new slot.
9860 	 *
9861 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9862 	 * been zapped so no dirty logging staff is needed for old slot. For
9863 	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9864 	 * new and it's also covered when dealing with the new slot.
9865 	 *
9866 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
9867 	 */
9868 	if (change != KVM_MR_DELETE)
9869 		kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9870 }
9871 
9872 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9873 {
9874 	kvm_mmu_zap_all(kvm);
9875 }
9876 
9877 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9878 				   struct kvm_memory_slot *slot)
9879 {
9880 	kvm_page_track_flush_slot(kvm, slot);
9881 }
9882 
9883 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9884 {
9885 	return (is_guest_mode(vcpu) &&
9886 			kvm_x86_ops->guest_apic_has_interrupt &&
9887 			kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9888 }
9889 
9890 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9891 {
9892 	if (!list_empty_careful(&vcpu->async_pf.done))
9893 		return true;
9894 
9895 	if (kvm_apic_has_events(vcpu))
9896 		return true;
9897 
9898 	if (vcpu->arch.pv.pv_unhalted)
9899 		return true;
9900 
9901 	if (vcpu->arch.exception.pending)
9902 		return true;
9903 
9904 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9905 	    (vcpu->arch.nmi_pending &&
9906 	     kvm_x86_ops->nmi_allowed(vcpu)))
9907 		return true;
9908 
9909 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9910 	    (vcpu->arch.smi_pending && !is_smm(vcpu)))
9911 		return true;
9912 
9913 	if (kvm_arch_interrupt_allowed(vcpu) &&
9914 	    (kvm_cpu_has_interrupt(vcpu) ||
9915 	    kvm_guest_apic_has_interrupt(vcpu)))
9916 		return true;
9917 
9918 	if (kvm_hv_has_stimer_pending(vcpu))
9919 		return true;
9920 
9921 	return false;
9922 }
9923 
9924 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9925 {
9926 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9927 }
9928 
9929 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
9930 {
9931 	if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
9932 		return true;
9933 
9934 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9935 		kvm_test_request(KVM_REQ_SMI, vcpu) ||
9936 		 kvm_test_request(KVM_REQ_EVENT, vcpu))
9937 		return true;
9938 
9939 	if (vcpu->arch.apicv_active && kvm_x86_ops->dy_apicv_has_pending_interrupt(vcpu))
9940 		return true;
9941 
9942 	return false;
9943 }
9944 
9945 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9946 {
9947 	return vcpu->arch.preempted_in_kernel;
9948 }
9949 
9950 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9951 {
9952 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9953 }
9954 
9955 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9956 {
9957 	return kvm_x86_ops->interrupt_allowed(vcpu);
9958 }
9959 
9960 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9961 {
9962 	if (is_64_bit_mode(vcpu))
9963 		return kvm_rip_read(vcpu);
9964 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9965 		     kvm_rip_read(vcpu));
9966 }
9967 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9968 
9969 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9970 {
9971 	return kvm_get_linear_rip(vcpu) == linear_rip;
9972 }
9973 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9974 
9975 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9976 {
9977 	unsigned long rflags;
9978 
9979 	rflags = kvm_x86_ops->get_rflags(vcpu);
9980 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9981 		rflags &= ~X86_EFLAGS_TF;
9982 	return rflags;
9983 }
9984 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9985 
9986 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9987 {
9988 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9989 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9990 		rflags |= X86_EFLAGS_TF;
9991 	kvm_x86_ops->set_rflags(vcpu, rflags);
9992 }
9993 
9994 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9995 {
9996 	__kvm_set_rflags(vcpu, rflags);
9997 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9998 }
9999 EXPORT_SYMBOL_GPL(kvm_set_rflags);
10000 
10001 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
10002 {
10003 	int r;
10004 
10005 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
10006 	      work->wakeup_all)
10007 		return;
10008 
10009 	r = kvm_mmu_reload(vcpu);
10010 	if (unlikely(r))
10011 		return;
10012 
10013 	if (!vcpu->arch.mmu->direct_map &&
10014 	      work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
10015 		return;
10016 
10017 	vcpu->arch.mmu->page_fault(vcpu, work->gva, 0, true);
10018 }
10019 
10020 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
10021 {
10022 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
10023 }
10024 
10025 static inline u32 kvm_async_pf_next_probe(u32 key)
10026 {
10027 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
10028 }
10029 
10030 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10031 {
10032 	u32 key = kvm_async_pf_hash_fn(gfn);
10033 
10034 	while (vcpu->arch.apf.gfns[key] != ~0)
10035 		key = kvm_async_pf_next_probe(key);
10036 
10037 	vcpu->arch.apf.gfns[key] = gfn;
10038 }
10039 
10040 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
10041 {
10042 	int i;
10043 	u32 key = kvm_async_pf_hash_fn(gfn);
10044 
10045 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
10046 		     (vcpu->arch.apf.gfns[key] != gfn &&
10047 		      vcpu->arch.apf.gfns[key] != ~0); i++)
10048 		key = kvm_async_pf_next_probe(key);
10049 
10050 	return key;
10051 }
10052 
10053 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10054 {
10055 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
10056 }
10057 
10058 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10059 {
10060 	u32 i, j, k;
10061 
10062 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
10063 	while (true) {
10064 		vcpu->arch.apf.gfns[i] = ~0;
10065 		do {
10066 			j = kvm_async_pf_next_probe(j);
10067 			if (vcpu->arch.apf.gfns[j] == ~0)
10068 				return;
10069 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
10070 			/*
10071 			 * k lies cyclically in ]i,j]
10072 			 * |    i.k.j |
10073 			 * |....j i.k.| or  |.k..j i...|
10074 			 */
10075 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
10076 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
10077 		i = j;
10078 	}
10079 }
10080 
10081 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
10082 {
10083 
10084 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
10085 				      sizeof(val));
10086 }
10087 
10088 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
10089 {
10090 
10091 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
10092 				      sizeof(u32));
10093 }
10094 
10095 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
10096 {
10097 	if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
10098 		return false;
10099 
10100 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
10101 	    (vcpu->arch.apf.send_user_only &&
10102 	     kvm_x86_ops->get_cpl(vcpu) == 0))
10103 		return false;
10104 
10105 	return true;
10106 }
10107 
10108 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
10109 {
10110 	if (unlikely(!lapic_in_kernel(vcpu) ||
10111 		     kvm_event_needs_reinjection(vcpu) ||
10112 		     vcpu->arch.exception.pending))
10113 		return false;
10114 
10115 	if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
10116 		return false;
10117 
10118 	/*
10119 	 * If interrupts are off we cannot even use an artificial
10120 	 * halt state.
10121 	 */
10122 	return kvm_x86_ops->interrupt_allowed(vcpu);
10123 }
10124 
10125 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
10126 				     struct kvm_async_pf *work)
10127 {
10128 	struct x86_exception fault;
10129 
10130 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
10131 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
10132 
10133 	if (kvm_can_deliver_async_pf(vcpu) &&
10134 	    !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
10135 		fault.vector = PF_VECTOR;
10136 		fault.error_code_valid = true;
10137 		fault.error_code = 0;
10138 		fault.nested_page_fault = false;
10139 		fault.address = work->arch.token;
10140 		fault.async_page_fault = true;
10141 		kvm_inject_page_fault(vcpu, &fault);
10142 	} else {
10143 		/*
10144 		 * It is not possible to deliver a paravirtualized asynchronous
10145 		 * page fault, but putting the guest in an artificial halt state
10146 		 * can be beneficial nevertheless: if an interrupt arrives, we
10147 		 * can deliver it timely and perhaps the guest will schedule
10148 		 * another process.  When the instruction that triggered a page
10149 		 * fault is retried, hopefully the page will be ready in the host.
10150 		 */
10151 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
10152 	}
10153 }
10154 
10155 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
10156 				 struct kvm_async_pf *work)
10157 {
10158 	struct x86_exception fault;
10159 	u32 val;
10160 
10161 	if (work->wakeup_all)
10162 		work->arch.token = ~0; /* broadcast wakeup */
10163 	else
10164 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10165 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
10166 
10167 	if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
10168 	    !apf_get_user(vcpu, &val)) {
10169 		if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
10170 		    vcpu->arch.exception.pending &&
10171 		    vcpu->arch.exception.nr == PF_VECTOR &&
10172 		    !apf_put_user(vcpu, 0)) {
10173 			vcpu->arch.exception.injected = false;
10174 			vcpu->arch.exception.pending = false;
10175 			vcpu->arch.exception.nr = 0;
10176 			vcpu->arch.exception.has_error_code = false;
10177 			vcpu->arch.exception.error_code = 0;
10178 			vcpu->arch.exception.has_payload = false;
10179 			vcpu->arch.exception.payload = 0;
10180 		} else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
10181 			fault.vector = PF_VECTOR;
10182 			fault.error_code_valid = true;
10183 			fault.error_code = 0;
10184 			fault.nested_page_fault = false;
10185 			fault.address = work->arch.token;
10186 			fault.async_page_fault = true;
10187 			kvm_inject_page_fault(vcpu, &fault);
10188 		}
10189 	}
10190 	vcpu->arch.apf.halted = false;
10191 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10192 }
10193 
10194 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
10195 {
10196 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
10197 		return true;
10198 	else
10199 		return kvm_can_do_async_pf(vcpu);
10200 }
10201 
10202 void kvm_arch_start_assignment(struct kvm *kvm)
10203 {
10204 	atomic_inc(&kvm->arch.assigned_device_count);
10205 }
10206 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
10207 
10208 void kvm_arch_end_assignment(struct kvm *kvm)
10209 {
10210 	atomic_dec(&kvm->arch.assigned_device_count);
10211 }
10212 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
10213 
10214 bool kvm_arch_has_assigned_device(struct kvm *kvm)
10215 {
10216 	return atomic_read(&kvm->arch.assigned_device_count);
10217 }
10218 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
10219 
10220 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
10221 {
10222 	atomic_inc(&kvm->arch.noncoherent_dma_count);
10223 }
10224 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
10225 
10226 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
10227 {
10228 	atomic_dec(&kvm->arch.noncoherent_dma_count);
10229 }
10230 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
10231 
10232 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
10233 {
10234 	return atomic_read(&kvm->arch.noncoherent_dma_count);
10235 }
10236 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
10237 
10238 bool kvm_arch_has_irq_bypass(void)
10239 {
10240 	return true;
10241 }
10242 
10243 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
10244 				      struct irq_bypass_producer *prod)
10245 {
10246 	struct kvm_kernel_irqfd *irqfd =
10247 		container_of(cons, struct kvm_kernel_irqfd, consumer);
10248 
10249 	irqfd->producer = prod;
10250 
10251 	return kvm_x86_ops->update_pi_irte(irqfd->kvm,
10252 					   prod->irq, irqfd->gsi, 1);
10253 }
10254 
10255 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
10256 				      struct irq_bypass_producer *prod)
10257 {
10258 	int ret;
10259 	struct kvm_kernel_irqfd *irqfd =
10260 		container_of(cons, struct kvm_kernel_irqfd, consumer);
10261 
10262 	WARN_ON(irqfd->producer != prod);
10263 	irqfd->producer = NULL;
10264 
10265 	/*
10266 	 * When producer of consumer is unregistered, we change back to
10267 	 * remapped mode, so we can re-use the current implementation
10268 	 * when the irq is masked/disabled or the consumer side (KVM
10269 	 * int this case doesn't want to receive the interrupts.
10270 	*/
10271 	ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
10272 	if (ret)
10273 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
10274 		       " fails: %d\n", irqfd->consumer.token, ret);
10275 }
10276 
10277 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
10278 				   uint32_t guest_irq, bool set)
10279 {
10280 	return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
10281 }
10282 
10283 bool kvm_vector_hashing_enabled(void)
10284 {
10285 	return vector_hashing;
10286 }
10287 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
10288 
10289 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
10290 {
10291 	return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
10292 }
10293 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
10294 
10295 
10296 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
10297 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
10298 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
10299 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
10300 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
10301 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
10302 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
10303 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
10304 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
10305 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
10306 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
10307 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
10308 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
10309 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
10310 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
10311 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
10312 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
10313 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
10314 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
10315 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
10316