xref: /linux/arch/x86/kvm/x86.c (revision 3da3cc1b5f47115b16b5ffeeb4bf09ec331b0164)
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 "ioapic.h"
22 #include "mmu.h"
23 #include "i8254.h"
24 #include "tss.h"
25 #include "kvm_cache_regs.h"
26 #include "kvm_emulate.h"
27 #include "x86.h"
28 #include "cpuid.h"
29 #include "pmu.h"
30 #include "hyperv.h"
31 #include "lapic.h"
32 
33 #include <linux/clocksource.h>
34 #include <linux/interrupt.h>
35 #include <linux/kvm.h>
36 #include <linux/fs.h>
37 #include <linux/vmalloc.h>
38 #include <linux/export.h>
39 #include <linux/moduleparam.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <linux/sched/stat.h>
57 #include <linux/sched/isolation.h>
58 #include <linux/mem_encrypt.h>
59 #include <linux/entry-kvm.h>
60 
61 #include <trace/events/kvm.h>
62 
63 #include <asm/debugreg.h>
64 #include <asm/msr.h>
65 #include <asm/desc.h>
66 #include <asm/mce.h>
67 #include <linux/kernel_stat.h>
68 #include <asm/fpu/internal.h> /* Ugh! */
69 #include <asm/pvclock.h>
70 #include <asm/div64.h>
71 #include <asm/irq_remapping.h>
72 #include <asm/mshyperv.h>
73 #include <asm/hypervisor.h>
74 #include <asm/tlbflush.h>
75 #include <asm/intel_pt.h>
76 #include <asm/emulate_prefix.h>
77 #include <clocksource/hyperv_timer.h>
78 
79 #define CREATE_TRACE_POINTS
80 #include "trace.h"
81 
82 #define MAX_IO_MSRS 256
83 #define KVM_MAX_MCE_BANKS 32
84 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
85 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
86 
87 #define emul_to_vcpu(ctxt) \
88 	((struct kvm_vcpu *)(ctxt)->vcpu)
89 
90 /* EFER defaults:
91  * - enable syscall per default because its emulated by KVM
92  * - enable LME and LMA per default on 64 bit KVM
93  */
94 #ifdef CONFIG_X86_64
95 static
96 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
97 #else
98 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
99 #endif
100 
101 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
102 
103 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
104                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
105 
106 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
107 static void process_nmi(struct kvm_vcpu *vcpu);
108 static void process_smi(struct kvm_vcpu *vcpu);
109 static void enter_smm(struct kvm_vcpu *vcpu);
110 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
111 static void store_regs(struct kvm_vcpu *vcpu);
112 static int sync_regs(struct kvm_vcpu *vcpu);
113 
114 struct kvm_x86_ops kvm_x86_ops __read_mostly;
115 EXPORT_SYMBOL_GPL(kvm_x86_ops);
116 
117 static bool __read_mostly ignore_msrs = 0;
118 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
119 
120 static bool __read_mostly report_ignored_msrs = true;
121 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
122 
123 unsigned int min_timer_period_us = 200;
124 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
125 
126 static bool __read_mostly kvmclock_periodic_sync = true;
127 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
128 
129 bool __read_mostly kvm_has_tsc_control;
130 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
131 u32  __read_mostly kvm_max_guest_tsc_khz;
132 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
133 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
134 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
135 u64  __read_mostly kvm_max_tsc_scaling_ratio;
136 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
137 u64 __read_mostly kvm_default_tsc_scaling_ratio;
138 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
139 
140 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
141 static u32 __read_mostly tsc_tolerance_ppm = 250;
142 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
143 
144 /*
145  * lapic timer advance (tscdeadline mode only) in nanoseconds.  '-1' enables
146  * adaptive tuning starting from default advancment of 1000ns.  '0' disables
147  * advancement entirely.  Any other value is used as-is and disables adaptive
148  * tuning, i.e. allows priveleged userspace to set an exact advancement time.
149  */
150 static int __read_mostly lapic_timer_advance_ns = -1;
151 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
152 
153 static bool __read_mostly vector_hashing = true;
154 module_param(vector_hashing, bool, S_IRUGO);
155 
156 bool __read_mostly enable_vmware_backdoor = false;
157 module_param(enable_vmware_backdoor, bool, S_IRUGO);
158 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
159 
160 static bool __read_mostly force_emulation_prefix = false;
161 module_param(force_emulation_prefix, bool, S_IRUGO);
162 
163 int __read_mostly pi_inject_timer = -1;
164 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
165 
166 /*
167  * Restoring the host value for MSRs that are only consumed when running in
168  * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
169  * returns to userspace, i.e. the kernel can run with the guest's value.
170  */
171 #define KVM_MAX_NR_USER_RETURN_MSRS 16
172 
173 struct kvm_user_return_msrs_global {
174 	int nr;
175 	u32 msrs[KVM_MAX_NR_USER_RETURN_MSRS];
176 };
177 
178 struct kvm_user_return_msrs {
179 	struct user_return_notifier urn;
180 	bool registered;
181 	struct kvm_user_return_msr_values {
182 		u64 host;
183 		u64 curr;
184 	} values[KVM_MAX_NR_USER_RETURN_MSRS];
185 };
186 
187 static struct kvm_user_return_msrs_global __read_mostly user_return_msrs_global;
188 static struct kvm_user_return_msrs __percpu *user_return_msrs;
189 
190 #define KVM_SUPPORTED_XCR0     (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
191 				| XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
192 				| XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
193 				| XFEATURE_MASK_PKRU)
194 
195 u64 __read_mostly host_efer;
196 EXPORT_SYMBOL_GPL(host_efer);
197 
198 bool __read_mostly allow_smaller_maxphyaddr = 0;
199 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
200 
201 u64 __read_mostly host_xss;
202 EXPORT_SYMBOL_GPL(host_xss);
203 u64 __read_mostly supported_xss;
204 EXPORT_SYMBOL_GPL(supported_xss);
205 
206 struct kvm_stats_debugfs_item debugfs_entries[] = {
207 	VCPU_STAT("pf_fixed", pf_fixed),
208 	VCPU_STAT("pf_guest", pf_guest),
209 	VCPU_STAT("tlb_flush", tlb_flush),
210 	VCPU_STAT("invlpg", invlpg),
211 	VCPU_STAT("exits", exits),
212 	VCPU_STAT("io_exits", io_exits),
213 	VCPU_STAT("mmio_exits", mmio_exits),
214 	VCPU_STAT("signal_exits", signal_exits),
215 	VCPU_STAT("irq_window", irq_window_exits),
216 	VCPU_STAT("nmi_window", nmi_window_exits),
217 	VCPU_STAT("halt_exits", halt_exits),
218 	VCPU_STAT("halt_successful_poll", halt_successful_poll),
219 	VCPU_STAT("halt_attempted_poll", halt_attempted_poll),
220 	VCPU_STAT("halt_poll_invalid", halt_poll_invalid),
221 	VCPU_STAT("halt_wakeup", halt_wakeup),
222 	VCPU_STAT("hypercalls", hypercalls),
223 	VCPU_STAT("request_irq", request_irq_exits),
224 	VCPU_STAT("irq_exits", irq_exits),
225 	VCPU_STAT("host_state_reload", host_state_reload),
226 	VCPU_STAT("fpu_reload", fpu_reload),
227 	VCPU_STAT("insn_emulation", insn_emulation),
228 	VCPU_STAT("insn_emulation_fail", insn_emulation_fail),
229 	VCPU_STAT("irq_injections", irq_injections),
230 	VCPU_STAT("nmi_injections", nmi_injections),
231 	VCPU_STAT("req_event", req_event),
232 	VCPU_STAT("l1d_flush", l1d_flush),
233 	VCPU_STAT("halt_poll_success_ns", halt_poll_success_ns),
234 	VCPU_STAT("halt_poll_fail_ns", halt_poll_fail_ns),
235 	VM_STAT("mmu_shadow_zapped", mmu_shadow_zapped),
236 	VM_STAT("mmu_pte_write", mmu_pte_write),
237 	VM_STAT("mmu_pte_updated", mmu_pte_updated),
238 	VM_STAT("mmu_pde_zapped", mmu_pde_zapped),
239 	VM_STAT("mmu_flooded", mmu_flooded),
240 	VM_STAT("mmu_recycled", mmu_recycled),
241 	VM_STAT("mmu_cache_miss", mmu_cache_miss),
242 	VM_STAT("mmu_unsync", mmu_unsync),
243 	VM_STAT("remote_tlb_flush", remote_tlb_flush),
244 	VM_STAT("largepages", lpages, .mode = 0444),
245 	VM_STAT("nx_largepages_splitted", nx_lpage_splits, .mode = 0444),
246 	VM_STAT("max_mmu_page_hash_collisions", max_mmu_page_hash_collisions),
247 	{ NULL }
248 };
249 
250 u64 __read_mostly host_xcr0;
251 u64 __read_mostly supported_xcr0;
252 EXPORT_SYMBOL_GPL(supported_xcr0);
253 
254 static struct kmem_cache *x86_fpu_cache;
255 
256 static struct kmem_cache *x86_emulator_cache;
257 
258 /*
259  * When called, it means the previous get/set msr reached an invalid msr.
260  * Return true if we want to ignore/silent this failed msr access.
261  */
262 static bool kvm_msr_ignored_check(struct kvm_vcpu *vcpu, u32 msr,
263 				  u64 data, bool write)
264 {
265 	const char *op = write ? "wrmsr" : "rdmsr";
266 
267 	if (ignore_msrs) {
268 		if (report_ignored_msrs)
269 			kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
270 				      op, msr, data);
271 		/* Mask the error */
272 		return true;
273 	} else {
274 		kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
275 				      op, msr, data);
276 		return false;
277 	}
278 }
279 
280 static struct kmem_cache *kvm_alloc_emulator_cache(void)
281 {
282 	unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
283 	unsigned int size = sizeof(struct x86_emulate_ctxt);
284 
285 	return kmem_cache_create_usercopy("x86_emulator", size,
286 					  __alignof__(struct x86_emulate_ctxt),
287 					  SLAB_ACCOUNT, useroffset,
288 					  size - useroffset, NULL);
289 }
290 
291 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
292 
293 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
294 {
295 	int i;
296 	for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
297 		vcpu->arch.apf.gfns[i] = ~0;
298 }
299 
300 static void kvm_on_user_return(struct user_return_notifier *urn)
301 {
302 	unsigned slot;
303 	struct kvm_user_return_msrs *msrs
304 		= container_of(urn, struct kvm_user_return_msrs, urn);
305 	struct kvm_user_return_msr_values *values;
306 	unsigned long flags;
307 
308 	/*
309 	 * Disabling irqs at this point since the following code could be
310 	 * interrupted and executed through kvm_arch_hardware_disable()
311 	 */
312 	local_irq_save(flags);
313 	if (msrs->registered) {
314 		msrs->registered = false;
315 		user_return_notifier_unregister(urn);
316 	}
317 	local_irq_restore(flags);
318 	for (slot = 0; slot < user_return_msrs_global.nr; ++slot) {
319 		values = &msrs->values[slot];
320 		if (values->host != values->curr) {
321 			wrmsrl(user_return_msrs_global.msrs[slot], values->host);
322 			values->curr = values->host;
323 		}
324 	}
325 }
326 
327 void kvm_define_user_return_msr(unsigned slot, u32 msr)
328 {
329 	BUG_ON(slot >= KVM_MAX_NR_USER_RETURN_MSRS);
330 	user_return_msrs_global.msrs[slot] = msr;
331 	if (slot >= user_return_msrs_global.nr)
332 		user_return_msrs_global.nr = slot + 1;
333 }
334 EXPORT_SYMBOL_GPL(kvm_define_user_return_msr);
335 
336 static void kvm_user_return_msr_cpu_online(void)
337 {
338 	unsigned int cpu = smp_processor_id();
339 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
340 	u64 value;
341 	int i;
342 
343 	for (i = 0; i < user_return_msrs_global.nr; ++i) {
344 		rdmsrl_safe(user_return_msrs_global.msrs[i], &value);
345 		msrs->values[i].host = value;
346 		msrs->values[i].curr = value;
347 	}
348 }
349 
350 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
351 {
352 	unsigned int cpu = smp_processor_id();
353 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
354 	int err;
355 
356 	value = (value & mask) | (msrs->values[slot].host & ~mask);
357 	if (value == msrs->values[slot].curr)
358 		return 0;
359 	err = wrmsrl_safe(user_return_msrs_global.msrs[slot], value);
360 	if (err)
361 		return 1;
362 
363 	msrs->values[slot].curr = value;
364 	if (!msrs->registered) {
365 		msrs->urn.on_user_return = kvm_on_user_return;
366 		user_return_notifier_register(&msrs->urn);
367 		msrs->registered = true;
368 	}
369 	return 0;
370 }
371 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
372 
373 static void drop_user_return_notifiers(void)
374 {
375 	unsigned int cpu = smp_processor_id();
376 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
377 
378 	if (msrs->registered)
379 		kvm_on_user_return(&msrs->urn);
380 }
381 
382 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
383 {
384 	return vcpu->arch.apic_base;
385 }
386 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
387 
388 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
389 {
390 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
391 }
392 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
393 
394 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
395 {
396 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
397 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
398 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
399 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
400 
401 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
402 		return 1;
403 	if (!msr_info->host_initiated) {
404 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
405 			return 1;
406 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
407 			return 1;
408 	}
409 
410 	kvm_lapic_set_base(vcpu, msr_info->data);
411 	kvm_recalculate_apic_map(vcpu->kvm);
412 	return 0;
413 }
414 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
415 
416 asmlinkage __visible noinstr void kvm_spurious_fault(void)
417 {
418 	/* Fault while not rebooting.  We want the trace. */
419 	BUG_ON(!kvm_rebooting);
420 }
421 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
422 
423 #define EXCPT_BENIGN		0
424 #define EXCPT_CONTRIBUTORY	1
425 #define EXCPT_PF		2
426 
427 static int exception_class(int vector)
428 {
429 	switch (vector) {
430 	case PF_VECTOR:
431 		return EXCPT_PF;
432 	case DE_VECTOR:
433 	case TS_VECTOR:
434 	case NP_VECTOR:
435 	case SS_VECTOR:
436 	case GP_VECTOR:
437 		return EXCPT_CONTRIBUTORY;
438 	default:
439 		break;
440 	}
441 	return EXCPT_BENIGN;
442 }
443 
444 #define EXCPT_FAULT		0
445 #define EXCPT_TRAP		1
446 #define EXCPT_ABORT		2
447 #define EXCPT_INTERRUPT		3
448 
449 static int exception_type(int vector)
450 {
451 	unsigned int mask;
452 
453 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
454 		return EXCPT_INTERRUPT;
455 
456 	mask = 1 << vector;
457 
458 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
459 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
460 		return EXCPT_TRAP;
461 
462 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
463 		return EXCPT_ABORT;
464 
465 	/* Reserved exceptions will result in fault */
466 	return EXCPT_FAULT;
467 }
468 
469 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
470 {
471 	unsigned nr = vcpu->arch.exception.nr;
472 	bool has_payload = vcpu->arch.exception.has_payload;
473 	unsigned long payload = vcpu->arch.exception.payload;
474 
475 	if (!has_payload)
476 		return;
477 
478 	switch (nr) {
479 	case DB_VECTOR:
480 		/*
481 		 * "Certain debug exceptions may clear bit 0-3.  The
482 		 * remaining contents of the DR6 register are never
483 		 * cleared by the processor".
484 		 */
485 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
486 		/*
487 		 * DR6.RTM is set by all #DB exceptions that don't clear it.
488 		 */
489 		vcpu->arch.dr6 |= DR6_RTM;
490 		vcpu->arch.dr6 |= payload;
491 		/*
492 		 * Bit 16 should be set in the payload whenever the #DB
493 		 * exception should clear DR6.RTM. This makes the payload
494 		 * compatible with the pending debug exceptions under VMX.
495 		 * Though not currently documented in the SDM, this also
496 		 * makes the payload compatible with the exit qualification
497 		 * for #DB exceptions under VMX.
498 		 */
499 		vcpu->arch.dr6 ^= payload & DR6_RTM;
500 
501 		/*
502 		 * The #DB payload is defined as compatible with the 'pending
503 		 * debug exceptions' field under VMX, not DR6. While bit 12 is
504 		 * defined in the 'pending debug exceptions' field (enabled
505 		 * breakpoint), it is reserved and must be zero in DR6.
506 		 */
507 		vcpu->arch.dr6 &= ~BIT(12);
508 		break;
509 	case PF_VECTOR:
510 		vcpu->arch.cr2 = payload;
511 		break;
512 	}
513 
514 	vcpu->arch.exception.has_payload = false;
515 	vcpu->arch.exception.payload = 0;
516 }
517 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
518 
519 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
520 		unsigned nr, bool has_error, u32 error_code,
521 	        bool has_payload, unsigned long payload, bool reinject)
522 {
523 	u32 prev_nr;
524 	int class1, class2;
525 
526 	kvm_make_request(KVM_REQ_EVENT, vcpu);
527 
528 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
529 	queue:
530 		if (has_error && !is_protmode(vcpu))
531 			has_error = false;
532 		if (reinject) {
533 			/*
534 			 * On vmentry, vcpu->arch.exception.pending is only
535 			 * true if an event injection was blocked by
536 			 * nested_run_pending.  In that case, however,
537 			 * vcpu_enter_guest requests an immediate exit,
538 			 * and the guest shouldn't proceed far enough to
539 			 * need reinjection.
540 			 */
541 			WARN_ON_ONCE(vcpu->arch.exception.pending);
542 			vcpu->arch.exception.injected = true;
543 			if (WARN_ON_ONCE(has_payload)) {
544 				/*
545 				 * A reinjected event has already
546 				 * delivered its payload.
547 				 */
548 				has_payload = false;
549 				payload = 0;
550 			}
551 		} else {
552 			vcpu->arch.exception.pending = true;
553 			vcpu->arch.exception.injected = false;
554 		}
555 		vcpu->arch.exception.has_error_code = has_error;
556 		vcpu->arch.exception.nr = nr;
557 		vcpu->arch.exception.error_code = error_code;
558 		vcpu->arch.exception.has_payload = has_payload;
559 		vcpu->arch.exception.payload = payload;
560 		if (!is_guest_mode(vcpu))
561 			kvm_deliver_exception_payload(vcpu);
562 		return;
563 	}
564 
565 	/* to check exception */
566 	prev_nr = vcpu->arch.exception.nr;
567 	if (prev_nr == DF_VECTOR) {
568 		/* triple fault -> shutdown */
569 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
570 		return;
571 	}
572 	class1 = exception_class(prev_nr);
573 	class2 = exception_class(nr);
574 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
575 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
576 		/*
577 		 * Generate double fault per SDM Table 5-5.  Set
578 		 * exception.pending = true so that the double fault
579 		 * can trigger a nested vmexit.
580 		 */
581 		vcpu->arch.exception.pending = true;
582 		vcpu->arch.exception.injected = false;
583 		vcpu->arch.exception.has_error_code = true;
584 		vcpu->arch.exception.nr = DF_VECTOR;
585 		vcpu->arch.exception.error_code = 0;
586 		vcpu->arch.exception.has_payload = false;
587 		vcpu->arch.exception.payload = 0;
588 	} else
589 		/* replace previous exception with a new one in a hope
590 		   that instruction re-execution will regenerate lost
591 		   exception */
592 		goto queue;
593 }
594 
595 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
596 {
597 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
598 }
599 EXPORT_SYMBOL_GPL(kvm_queue_exception);
600 
601 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
602 {
603 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
604 }
605 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
606 
607 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
608 			   unsigned long payload)
609 {
610 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
611 }
612 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
613 
614 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
615 				    u32 error_code, unsigned long payload)
616 {
617 	kvm_multiple_exception(vcpu, nr, true, error_code,
618 			       true, payload, false);
619 }
620 
621 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
622 {
623 	if (err)
624 		kvm_inject_gp(vcpu, 0);
625 	else
626 		return kvm_skip_emulated_instruction(vcpu);
627 
628 	return 1;
629 }
630 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
631 
632 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
633 {
634 	++vcpu->stat.pf_guest;
635 	vcpu->arch.exception.nested_apf =
636 		is_guest_mode(vcpu) && fault->async_page_fault;
637 	if (vcpu->arch.exception.nested_apf) {
638 		vcpu->arch.apf.nested_apf_token = fault->address;
639 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
640 	} else {
641 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
642 					fault->address);
643 	}
644 }
645 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
646 
647 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
648 				    struct x86_exception *fault)
649 {
650 	struct kvm_mmu *fault_mmu;
651 	WARN_ON_ONCE(fault->vector != PF_VECTOR);
652 
653 	fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
654 					       vcpu->arch.walk_mmu;
655 
656 	/*
657 	 * Invalidate the TLB entry for the faulting address, if it exists,
658 	 * else the access will fault indefinitely (and to emulate hardware).
659 	 */
660 	if ((fault->error_code & PFERR_PRESENT_MASK) &&
661 	    !(fault->error_code & PFERR_RSVD_MASK))
662 		kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
663 				       fault_mmu->root_hpa);
664 
665 	fault_mmu->inject_page_fault(vcpu, fault);
666 	return fault->nested_page_fault;
667 }
668 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
669 
670 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
671 {
672 	atomic_inc(&vcpu->arch.nmi_queued);
673 	kvm_make_request(KVM_REQ_NMI, vcpu);
674 }
675 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
676 
677 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
678 {
679 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
680 }
681 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
682 
683 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
684 {
685 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
686 }
687 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
688 
689 /*
690  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
691  * a #GP and return false.
692  */
693 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
694 {
695 	if (kvm_x86_ops.get_cpl(vcpu) <= required_cpl)
696 		return true;
697 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
698 	return false;
699 }
700 EXPORT_SYMBOL_GPL(kvm_require_cpl);
701 
702 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
703 {
704 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
705 		return true;
706 
707 	kvm_queue_exception(vcpu, UD_VECTOR);
708 	return false;
709 }
710 EXPORT_SYMBOL_GPL(kvm_require_dr);
711 
712 /*
713  * This function will be used to read from the physical memory of the currently
714  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
715  * can read from guest physical or from the guest's guest physical memory.
716  */
717 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
718 			    gfn_t ngfn, void *data, int offset, int len,
719 			    u32 access)
720 {
721 	struct x86_exception exception;
722 	gfn_t real_gfn;
723 	gpa_t ngpa;
724 
725 	ngpa     = gfn_to_gpa(ngfn);
726 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
727 	if (real_gfn == UNMAPPED_GVA)
728 		return -EFAULT;
729 
730 	real_gfn = gpa_to_gfn(real_gfn);
731 
732 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
733 }
734 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
735 
736 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
737 			       void *data, int offset, int len, u32 access)
738 {
739 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
740 				       data, offset, len, access);
741 }
742 
743 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
744 {
745 	return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
746 	       rsvd_bits(1, 2);
747 }
748 
749 /*
750  * Load the pae pdptrs.  Return 1 if they are all valid, 0 otherwise.
751  */
752 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
753 {
754 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
755 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
756 	int i;
757 	int ret;
758 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
759 
760 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
761 				      offset * sizeof(u64), sizeof(pdpte),
762 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
763 	if (ret < 0) {
764 		ret = 0;
765 		goto out;
766 	}
767 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
768 		if ((pdpte[i] & PT_PRESENT_MASK) &&
769 		    (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
770 			ret = 0;
771 			goto out;
772 		}
773 	}
774 	ret = 1;
775 
776 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
777 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
778 
779 out:
780 
781 	return ret;
782 }
783 EXPORT_SYMBOL_GPL(load_pdptrs);
784 
785 bool pdptrs_changed(struct kvm_vcpu *vcpu)
786 {
787 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
788 	int offset;
789 	gfn_t gfn;
790 	int r;
791 
792 	if (!is_pae_paging(vcpu))
793 		return false;
794 
795 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
796 		return true;
797 
798 	gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
799 	offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
800 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
801 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
802 	if (r < 0)
803 		return true;
804 
805 	return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
806 }
807 EXPORT_SYMBOL_GPL(pdptrs_changed);
808 
809 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
810 {
811 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
812 
813 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
814 		kvm_clear_async_pf_completion_queue(vcpu);
815 		kvm_async_pf_hash_reset(vcpu);
816 	}
817 
818 	if ((cr0 ^ old_cr0) & update_bits)
819 		kvm_mmu_reset_context(vcpu);
820 
821 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
822 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
823 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
824 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
825 }
826 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
827 
828 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
829 {
830 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
831 	unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
832 
833 	cr0 |= X86_CR0_ET;
834 
835 #ifdef CONFIG_X86_64
836 	if (cr0 & 0xffffffff00000000UL)
837 		return 1;
838 #endif
839 
840 	cr0 &= ~CR0_RESERVED_BITS;
841 
842 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
843 		return 1;
844 
845 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
846 		return 1;
847 
848 #ifdef CONFIG_X86_64
849 	if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
850 	    (cr0 & X86_CR0_PG)) {
851 		int cs_db, cs_l;
852 
853 		if (!is_pae(vcpu))
854 			return 1;
855 		kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
856 		if (cs_l)
857 			return 1;
858 	}
859 #endif
860 	if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
861 	    is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
862 	    !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
863 		return 1;
864 
865 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
866 		return 1;
867 
868 	kvm_x86_ops.set_cr0(vcpu, cr0);
869 
870 	kvm_post_set_cr0(vcpu, old_cr0, cr0);
871 
872 	return 0;
873 }
874 EXPORT_SYMBOL_GPL(kvm_set_cr0);
875 
876 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
877 {
878 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
879 }
880 EXPORT_SYMBOL_GPL(kvm_lmsw);
881 
882 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
883 {
884 	if (vcpu->arch.guest_state_protected)
885 		return;
886 
887 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
888 
889 		if (vcpu->arch.xcr0 != host_xcr0)
890 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
891 
892 		if (vcpu->arch.xsaves_enabled &&
893 		    vcpu->arch.ia32_xss != host_xss)
894 			wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
895 	}
896 
897 	if (static_cpu_has(X86_FEATURE_PKU) &&
898 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
899 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
900 	    vcpu->arch.pkru != vcpu->arch.host_pkru)
901 		__write_pkru(vcpu->arch.pkru);
902 }
903 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
904 
905 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
906 {
907 	if (vcpu->arch.guest_state_protected)
908 		return;
909 
910 	if (static_cpu_has(X86_FEATURE_PKU) &&
911 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
912 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
913 		vcpu->arch.pkru = rdpkru();
914 		if (vcpu->arch.pkru != vcpu->arch.host_pkru)
915 			__write_pkru(vcpu->arch.host_pkru);
916 	}
917 
918 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
919 
920 		if (vcpu->arch.xcr0 != host_xcr0)
921 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
922 
923 		if (vcpu->arch.xsaves_enabled &&
924 		    vcpu->arch.ia32_xss != host_xss)
925 			wrmsrl(MSR_IA32_XSS, host_xss);
926 	}
927 
928 }
929 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
930 
931 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
932 {
933 	u64 xcr0 = xcr;
934 	u64 old_xcr0 = vcpu->arch.xcr0;
935 	u64 valid_bits;
936 
937 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
938 	if (index != XCR_XFEATURE_ENABLED_MASK)
939 		return 1;
940 	if (!(xcr0 & XFEATURE_MASK_FP))
941 		return 1;
942 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
943 		return 1;
944 
945 	/*
946 	 * Do not allow the guest to set bits that we do not support
947 	 * saving.  However, xcr0 bit 0 is always set, even if the
948 	 * emulated CPU does not support XSAVE (see fx_init).
949 	 */
950 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
951 	if (xcr0 & ~valid_bits)
952 		return 1;
953 
954 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
955 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
956 		return 1;
957 
958 	if (xcr0 & XFEATURE_MASK_AVX512) {
959 		if (!(xcr0 & XFEATURE_MASK_YMM))
960 			return 1;
961 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
962 			return 1;
963 	}
964 	vcpu->arch.xcr0 = xcr0;
965 
966 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
967 		kvm_update_cpuid_runtime(vcpu);
968 	return 0;
969 }
970 
971 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
972 {
973 	if (kvm_x86_ops.get_cpl(vcpu) != 0 ||
974 	    __kvm_set_xcr(vcpu, index, xcr)) {
975 		kvm_inject_gp(vcpu, 0);
976 		return 1;
977 	}
978 	return 0;
979 }
980 EXPORT_SYMBOL_GPL(kvm_set_xcr);
981 
982 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
983 {
984 	if (cr4 & cr4_reserved_bits)
985 		return false;
986 
987 	if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
988 		return false;
989 
990 	return kvm_x86_ops.is_valid_cr4(vcpu, cr4);
991 }
992 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
993 
994 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
995 {
996 	unsigned long mmu_role_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
997 				      X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
998 
999 	if (((cr4 ^ old_cr4) & mmu_role_bits) ||
1000 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1001 		kvm_mmu_reset_context(vcpu);
1002 }
1003 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1004 
1005 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1006 {
1007 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
1008 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1009 				   X86_CR4_SMEP;
1010 
1011 	if (!kvm_is_valid_cr4(vcpu, cr4))
1012 		return 1;
1013 
1014 	if (is_long_mode(vcpu)) {
1015 		if (!(cr4 & X86_CR4_PAE))
1016 			return 1;
1017 		if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1018 			return 1;
1019 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1020 		   && ((cr4 ^ old_cr4) & pdptr_bits)
1021 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1022 				   kvm_read_cr3(vcpu)))
1023 		return 1;
1024 
1025 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1026 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1027 			return 1;
1028 
1029 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1030 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1031 			return 1;
1032 	}
1033 
1034 	kvm_x86_ops.set_cr4(vcpu, cr4);
1035 
1036 	kvm_post_set_cr4(vcpu, old_cr4, cr4);
1037 
1038 	return 0;
1039 }
1040 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1041 
1042 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1043 {
1044 	bool skip_tlb_flush = false;
1045 #ifdef CONFIG_X86_64
1046 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1047 
1048 	if (pcid_enabled) {
1049 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1050 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
1051 	}
1052 #endif
1053 
1054 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
1055 		if (!skip_tlb_flush) {
1056 			kvm_mmu_sync_roots(vcpu);
1057 			kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1058 		}
1059 		return 0;
1060 	}
1061 
1062 	if (is_long_mode(vcpu) &&
1063 	    (cr3 & vcpu->arch.cr3_lm_rsvd_bits))
1064 		return 1;
1065 	else if (is_pae_paging(vcpu) &&
1066 		 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1067 		return 1;
1068 
1069 	kvm_mmu_new_pgd(vcpu, cr3, skip_tlb_flush, skip_tlb_flush);
1070 	vcpu->arch.cr3 = cr3;
1071 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1072 
1073 	return 0;
1074 }
1075 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1076 
1077 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1078 {
1079 	if (cr8 & CR8_RESERVED_BITS)
1080 		return 1;
1081 	if (lapic_in_kernel(vcpu))
1082 		kvm_lapic_set_tpr(vcpu, cr8);
1083 	else
1084 		vcpu->arch.cr8 = cr8;
1085 	return 0;
1086 }
1087 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1088 
1089 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1090 {
1091 	if (lapic_in_kernel(vcpu))
1092 		return kvm_lapic_get_cr8(vcpu);
1093 	else
1094 		return vcpu->arch.cr8;
1095 }
1096 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1097 
1098 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1099 {
1100 	int i;
1101 
1102 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1103 		for (i = 0; i < KVM_NR_DB_REGS; i++)
1104 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1105 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1106 	}
1107 }
1108 
1109 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1110 {
1111 	unsigned long dr7;
1112 
1113 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1114 		dr7 = vcpu->arch.guest_debug_dr7;
1115 	else
1116 		dr7 = vcpu->arch.dr7;
1117 	kvm_x86_ops.set_dr7(vcpu, dr7);
1118 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1119 	if (dr7 & DR7_BP_EN_MASK)
1120 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1121 }
1122 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1123 
1124 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1125 {
1126 	u64 fixed = DR6_FIXED_1;
1127 
1128 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1129 		fixed |= DR6_RTM;
1130 	return fixed;
1131 }
1132 
1133 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1134 {
1135 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1136 
1137 	switch (dr) {
1138 	case 0 ... 3:
1139 		vcpu->arch.db[array_index_nospec(dr, size)] = val;
1140 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1141 			vcpu->arch.eff_db[dr] = val;
1142 		break;
1143 	case 4:
1144 	case 6:
1145 		if (!kvm_dr6_valid(val))
1146 			return -1; /* #GP */
1147 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1148 		break;
1149 	case 5:
1150 	default: /* 7 */
1151 		if (!kvm_dr7_valid(val))
1152 			return -1; /* #GP */
1153 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1154 		kvm_update_dr7(vcpu);
1155 		break;
1156 	}
1157 
1158 	return 0;
1159 }
1160 
1161 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1162 {
1163 	if (__kvm_set_dr(vcpu, dr, val)) {
1164 		kvm_inject_gp(vcpu, 0);
1165 		return 1;
1166 	}
1167 	return 0;
1168 }
1169 EXPORT_SYMBOL_GPL(kvm_set_dr);
1170 
1171 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1172 {
1173 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1174 
1175 	switch (dr) {
1176 	case 0 ... 3:
1177 		*val = vcpu->arch.db[array_index_nospec(dr, size)];
1178 		break;
1179 	case 4:
1180 	case 6:
1181 		*val = vcpu->arch.dr6;
1182 		break;
1183 	case 5:
1184 	default: /* 7 */
1185 		*val = vcpu->arch.dr7;
1186 		break;
1187 	}
1188 	return 0;
1189 }
1190 EXPORT_SYMBOL_GPL(kvm_get_dr);
1191 
1192 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1193 {
1194 	u32 ecx = kvm_rcx_read(vcpu);
1195 	u64 data;
1196 	int err;
1197 
1198 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1199 	if (err)
1200 		return err;
1201 	kvm_rax_write(vcpu, (u32)data);
1202 	kvm_rdx_write(vcpu, data >> 32);
1203 	return err;
1204 }
1205 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1206 
1207 /*
1208  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1209  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1210  *
1211  * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1212  * extract the supported MSRs from the related const lists.
1213  * msrs_to_save is selected from the msrs_to_save_all to reflect the
1214  * capabilities of the host cpu. This capabilities test skips MSRs that are
1215  * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1216  * may depend on host virtualization features rather than host cpu features.
1217  */
1218 
1219 static const u32 msrs_to_save_all[] = {
1220 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1221 	MSR_STAR,
1222 #ifdef CONFIG_X86_64
1223 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1224 #endif
1225 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1226 	MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1227 	MSR_IA32_SPEC_CTRL,
1228 	MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1229 	MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1230 	MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1231 	MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1232 	MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1233 	MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1234 	MSR_IA32_UMWAIT_CONTROL,
1235 
1236 	MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1237 	MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1238 	MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1239 	MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1240 	MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1241 	MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1242 	MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1243 	MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1244 	MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1245 	MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1246 	MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1247 	MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1248 	MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1249 	MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1250 	MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1251 	MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1252 	MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1253 	MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1254 	MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1255 	MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1256 	MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1257 	MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1258 };
1259 
1260 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1261 static unsigned num_msrs_to_save;
1262 
1263 static const u32 emulated_msrs_all[] = {
1264 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1265 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1266 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1267 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1268 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1269 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1270 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1271 	HV_X64_MSR_RESET,
1272 	HV_X64_MSR_VP_INDEX,
1273 	HV_X64_MSR_VP_RUNTIME,
1274 	HV_X64_MSR_SCONTROL,
1275 	HV_X64_MSR_STIMER0_CONFIG,
1276 	HV_X64_MSR_VP_ASSIST_PAGE,
1277 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1278 	HV_X64_MSR_TSC_EMULATION_STATUS,
1279 	HV_X64_MSR_SYNDBG_OPTIONS,
1280 	HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1281 	HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1282 	HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1283 
1284 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1285 	MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1286 
1287 	MSR_IA32_TSC_ADJUST,
1288 	MSR_IA32_TSCDEADLINE,
1289 	MSR_IA32_ARCH_CAPABILITIES,
1290 	MSR_IA32_PERF_CAPABILITIES,
1291 	MSR_IA32_MISC_ENABLE,
1292 	MSR_IA32_MCG_STATUS,
1293 	MSR_IA32_MCG_CTL,
1294 	MSR_IA32_MCG_EXT_CTL,
1295 	MSR_IA32_SMBASE,
1296 	MSR_SMI_COUNT,
1297 	MSR_PLATFORM_INFO,
1298 	MSR_MISC_FEATURES_ENABLES,
1299 	MSR_AMD64_VIRT_SPEC_CTRL,
1300 	MSR_IA32_POWER_CTL,
1301 	MSR_IA32_UCODE_REV,
1302 
1303 	/*
1304 	 * The following list leaves out MSRs whose values are determined
1305 	 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1306 	 * We always support the "true" VMX control MSRs, even if the host
1307 	 * processor does not, so I am putting these registers here rather
1308 	 * than in msrs_to_save_all.
1309 	 */
1310 	MSR_IA32_VMX_BASIC,
1311 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1312 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1313 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1314 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1315 	MSR_IA32_VMX_MISC,
1316 	MSR_IA32_VMX_CR0_FIXED0,
1317 	MSR_IA32_VMX_CR4_FIXED0,
1318 	MSR_IA32_VMX_VMCS_ENUM,
1319 	MSR_IA32_VMX_PROCBASED_CTLS2,
1320 	MSR_IA32_VMX_EPT_VPID_CAP,
1321 	MSR_IA32_VMX_VMFUNC,
1322 
1323 	MSR_K7_HWCR,
1324 	MSR_KVM_POLL_CONTROL,
1325 };
1326 
1327 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1328 static unsigned num_emulated_msrs;
1329 
1330 /*
1331  * List of msr numbers which are used to expose MSR-based features that
1332  * can be used by a hypervisor to validate requested CPU features.
1333  */
1334 static const u32 msr_based_features_all[] = {
1335 	MSR_IA32_VMX_BASIC,
1336 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1337 	MSR_IA32_VMX_PINBASED_CTLS,
1338 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1339 	MSR_IA32_VMX_PROCBASED_CTLS,
1340 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1341 	MSR_IA32_VMX_EXIT_CTLS,
1342 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1343 	MSR_IA32_VMX_ENTRY_CTLS,
1344 	MSR_IA32_VMX_MISC,
1345 	MSR_IA32_VMX_CR0_FIXED0,
1346 	MSR_IA32_VMX_CR0_FIXED1,
1347 	MSR_IA32_VMX_CR4_FIXED0,
1348 	MSR_IA32_VMX_CR4_FIXED1,
1349 	MSR_IA32_VMX_VMCS_ENUM,
1350 	MSR_IA32_VMX_PROCBASED_CTLS2,
1351 	MSR_IA32_VMX_EPT_VPID_CAP,
1352 	MSR_IA32_VMX_VMFUNC,
1353 
1354 	MSR_F10H_DECFG,
1355 	MSR_IA32_UCODE_REV,
1356 	MSR_IA32_ARCH_CAPABILITIES,
1357 	MSR_IA32_PERF_CAPABILITIES,
1358 };
1359 
1360 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1361 static unsigned int num_msr_based_features;
1362 
1363 static u64 kvm_get_arch_capabilities(void)
1364 {
1365 	u64 data = 0;
1366 
1367 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1368 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1369 
1370 	/*
1371 	 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1372 	 * the nested hypervisor runs with NX huge pages.  If it is not,
1373 	 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1374 	 * L1 guests, so it need not worry about its own (L2) guests.
1375 	 */
1376 	data |= ARCH_CAP_PSCHANGE_MC_NO;
1377 
1378 	/*
1379 	 * If we're doing cache flushes (either "always" or "cond")
1380 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1381 	 * If an outer hypervisor is doing the cache flush for us
1382 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1383 	 * capability to the guest too, and if EPT is disabled we're not
1384 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1385 	 * require a nested hypervisor to do a flush of its own.
1386 	 */
1387 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1388 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1389 
1390 	if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1391 		data |= ARCH_CAP_RDCL_NO;
1392 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1393 		data |= ARCH_CAP_SSB_NO;
1394 	if (!boot_cpu_has_bug(X86_BUG_MDS))
1395 		data |= ARCH_CAP_MDS_NO;
1396 
1397 	if (!boot_cpu_has(X86_FEATURE_RTM)) {
1398 		/*
1399 		 * If RTM=0 because the kernel has disabled TSX, the host might
1400 		 * have TAA_NO or TSX_CTRL.  Clear TAA_NO (the guest sees RTM=0
1401 		 * and therefore knows that there cannot be TAA) but keep
1402 		 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1403 		 * and we want to allow migrating those guests to tsx=off hosts.
1404 		 */
1405 		data &= ~ARCH_CAP_TAA_NO;
1406 	} else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1407 		data |= ARCH_CAP_TAA_NO;
1408 	} else {
1409 		/*
1410 		 * Nothing to do here; we emulate TSX_CTRL if present on the
1411 		 * host so the guest can choose between disabling TSX or
1412 		 * using VERW to clear CPU buffers.
1413 		 */
1414 	}
1415 
1416 	return data;
1417 }
1418 
1419 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1420 {
1421 	switch (msr->index) {
1422 	case MSR_IA32_ARCH_CAPABILITIES:
1423 		msr->data = kvm_get_arch_capabilities();
1424 		break;
1425 	case MSR_IA32_UCODE_REV:
1426 		rdmsrl_safe(msr->index, &msr->data);
1427 		break;
1428 	default:
1429 		return kvm_x86_ops.get_msr_feature(msr);
1430 	}
1431 	return 0;
1432 }
1433 
1434 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1435 {
1436 	struct kvm_msr_entry msr;
1437 	int r;
1438 
1439 	msr.index = index;
1440 	r = kvm_get_msr_feature(&msr);
1441 
1442 	if (r == KVM_MSR_RET_INVALID) {
1443 		/* Unconditionally clear the output for simplicity */
1444 		*data = 0;
1445 		if (kvm_msr_ignored_check(vcpu, index, 0, false))
1446 			r = 0;
1447 	}
1448 
1449 	if (r)
1450 		return r;
1451 
1452 	*data = msr.data;
1453 
1454 	return 0;
1455 }
1456 
1457 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1458 {
1459 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1460 		return false;
1461 
1462 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1463 		return false;
1464 
1465 	if (efer & (EFER_LME | EFER_LMA) &&
1466 	    !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1467 		return false;
1468 
1469 	if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1470 		return false;
1471 
1472 	return true;
1473 
1474 }
1475 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1476 {
1477 	if (efer & efer_reserved_bits)
1478 		return false;
1479 
1480 	return __kvm_valid_efer(vcpu, efer);
1481 }
1482 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1483 
1484 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1485 {
1486 	u64 old_efer = vcpu->arch.efer;
1487 	u64 efer = msr_info->data;
1488 	int r;
1489 
1490 	if (efer & efer_reserved_bits)
1491 		return 1;
1492 
1493 	if (!msr_info->host_initiated) {
1494 		if (!__kvm_valid_efer(vcpu, efer))
1495 			return 1;
1496 
1497 		if (is_paging(vcpu) &&
1498 		    (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1499 			return 1;
1500 	}
1501 
1502 	efer &= ~EFER_LMA;
1503 	efer |= vcpu->arch.efer & EFER_LMA;
1504 
1505 	r = kvm_x86_ops.set_efer(vcpu, efer);
1506 	if (r) {
1507 		WARN_ON(r > 0);
1508 		return r;
1509 	}
1510 
1511 	/* Update reserved bits */
1512 	if ((efer ^ old_efer) & EFER_NX)
1513 		kvm_mmu_reset_context(vcpu);
1514 
1515 	return 0;
1516 }
1517 
1518 void kvm_enable_efer_bits(u64 mask)
1519 {
1520        efer_reserved_bits &= ~mask;
1521 }
1522 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1523 
1524 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1525 {
1526 	struct kvm *kvm = vcpu->kvm;
1527 	struct msr_bitmap_range *ranges = kvm->arch.msr_filter.ranges;
1528 	u32 count = kvm->arch.msr_filter.count;
1529 	u32 i;
1530 	bool r = kvm->arch.msr_filter.default_allow;
1531 	int idx;
1532 
1533 	/* MSR filtering not set up or x2APIC enabled, allow everything */
1534 	if (!count || (index >= 0x800 && index <= 0x8ff))
1535 		return true;
1536 
1537 	/* Prevent collision with set_msr_filter */
1538 	idx = srcu_read_lock(&kvm->srcu);
1539 
1540 	for (i = 0; i < count; i++) {
1541 		u32 start = ranges[i].base;
1542 		u32 end = start + ranges[i].nmsrs;
1543 		u32 flags = ranges[i].flags;
1544 		unsigned long *bitmap = ranges[i].bitmap;
1545 
1546 		if ((index >= start) && (index < end) && (flags & type)) {
1547 			r = !!test_bit(index - start, bitmap);
1548 			break;
1549 		}
1550 	}
1551 
1552 	srcu_read_unlock(&kvm->srcu, idx);
1553 
1554 	return r;
1555 }
1556 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1557 
1558 /*
1559  * Write @data into the MSR specified by @index.  Select MSR specific fault
1560  * checks are bypassed if @host_initiated is %true.
1561  * Returns 0 on success, non-0 otherwise.
1562  * Assumes vcpu_load() was already called.
1563  */
1564 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1565 			 bool host_initiated)
1566 {
1567 	struct msr_data msr;
1568 
1569 	if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1570 		return KVM_MSR_RET_FILTERED;
1571 
1572 	switch (index) {
1573 	case MSR_FS_BASE:
1574 	case MSR_GS_BASE:
1575 	case MSR_KERNEL_GS_BASE:
1576 	case MSR_CSTAR:
1577 	case MSR_LSTAR:
1578 		if (is_noncanonical_address(data, vcpu))
1579 			return 1;
1580 		break;
1581 	case MSR_IA32_SYSENTER_EIP:
1582 	case MSR_IA32_SYSENTER_ESP:
1583 		/*
1584 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1585 		 * non-canonical address is written on Intel but not on
1586 		 * AMD (which ignores the top 32-bits, because it does
1587 		 * not implement 64-bit SYSENTER).
1588 		 *
1589 		 * 64-bit code should hence be able to write a non-canonical
1590 		 * value on AMD.  Making the address canonical ensures that
1591 		 * vmentry does not fail on Intel after writing a non-canonical
1592 		 * value, and that something deterministic happens if the guest
1593 		 * invokes 64-bit SYSENTER.
1594 		 */
1595 		data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1596 	}
1597 
1598 	msr.data = data;
1599 	msr.index = index;
1600 	msr.host_initiated = host_initiated;
1601 
1602 	return kvm_x86_ops.set_msr(vcpu, &msr);
1603 }
1604 
1605 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1606 				     u32 index, u64 data, bool host_initiated)
1607 {
1608 	int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1609 
1610 	if (ret == KVM_MSR_RET_INVALID)
1611 		if (kvm_msr_ignored_check(vcpu, index, data, true))
1612 			ret = 0;
1613 
1614 	return ret;
1615 }
1616 
1617 /*
1618  * Read the MSR specified by @index into @data.  Select MSR specific fault
1619  * checks are bypassed if @host_initiated is %true.
1620  * Returns 0 on success, non-0 otherwise.
1621  * Assumes vcpu_load() was already called.
1622  */
1623 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1624 		  bool host_initiated)
1625 {
1626 	struct msr_data msr;
1627 	int ret;
1628 
1629 	if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1630 		return KVM_MSR_RET_FILTERED;
1631 
1632 	msr.index = index;
1633 	msr.host_initiated = host_initiated;
1634 
1635 	ret = kvm_x86_ops.get_msr(vcpu, &msr);
1636 	if (!ret)
1637 		*data = msr.data;
1638 	return ret;
1639 }
1640 
1641 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1642 				     u32 index, u64 *data, bool host_initiated)
1643 {
1644 	int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1645 
1646 	if (ret == KVM_MSR_RET_INVALID) {
1647 		/* Unconditionally clear *data for simplicity */
1648 		*data = 0;
1649 		if (kvm_msr_ignored_check(vcpu, index, 0, false))
1650 			ret = 0;
1651 	}
1652 
1653 	return ret;
1654 }
1655 
1656 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1657 {
1658 	return kvm_get_msr_ignored_check(vcpu, index, data, false);
1659 }
1660 EXPORT_SYMBOL_GPL(kvm_get_msr);
1661 
1662 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1663 {
1664 	return kvm_set_msr_ignored_check(vcpu, index, data, false);
1665 }
1666 EXPORT_SYMBOL_GPL(kvm_set_msr);
1667 
1668 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1669 {
1670 	int err = vcpu->run->msr.error;
1671 	if (!err) {
1672 		kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1673 		kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1674 	}
1675 
1676 	return kvm_x86_ops.complete_emulated_msr(vcpu, err);
1677 }
1678 
1679 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1680 {
1681 	return kvm_x86_ops.complete_emulated_msr(vcpu, vcpu->run->msr.error);
1682 }
1683 
1684 static u64 kvm_msr_reason(int r)
1685 {
1686 	switch (r) {
1687 	case KVM_MSR_RET_INVALID:
1688 		return KVM_MSR_EXIT_REASON_UNKNOWN;
1689 	case KVM_MSR_RET_FILTERED:
1690 		return KVM_MSR_EXIT_REASON_FILTER;
1691 	default:
1692 		return KVM_MSR_EXIT_REASON_INVAL;
1693 	}
1694 }
1695 
1696 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1697 			      u32 exit_reason, u64 data,
1698 			      int (*completion)(struct kvm_vcpu *vcpu),
1699 			      int r)
1700 {
1701 	u64 msr_reason = kvm_msr_reason(r);
1702 
1703 	/* Check if the user wanted to know about this MSR fault */
1704 	if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1705 		return 0;
1706 
1707 	vcpu->run->exit_reason = exit_reason;
1708 	vcpu->run->msr.error = 0;
1709 	memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1710 	vcpu->run->msr.reason = msr_reason;
1711 	vcpu->run->msr.index = index;
1712 	vcpu->run->msr.data = data;
1713 	vcpu->arch.complete_userspace_io = completion;
1714 
1715 	return 1;
1716 }
1717 
1718 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1719 {
1720 	return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1721 				   complete_emulated_rdmsr, r);
1722 }
1723 
1724 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1725 {
1726 	return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1727 				   complete_emulated_wrmsr, r);
1728 }
1729 
1730 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1731 {
1732 	u32 ecx = kvm_rcx_read(vcpu);
1733 	u64 data;
1734 	int r;
1735 
1736 	r = kvm_get_msr(vcpu, ecx, &data);
1737 
1738 	/* MSR read failed? See if we should ask user space */
1739 	if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1740 		/* Bounce to user space */
1741 		return 0;
1742 	}
1743 
1744 	if (!r) {
1745 		trace_kvm_msr_read(ecx, data);
1746 
1747 		kvm_rax_write(vcpu, data & -1u);
1748 		kvm_rdx_write(vcpu, (data >> 32) & -1u);
1749 	} else {
1750 		trace_kvm_msr_read_ex(ecx);
1751 	}
1752 
1753 	return kvm_x86_ops.complete_emulated_msr(vcpu, r);
1754 }
1755 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1756 
1757 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1758 {
1759 	u32 ecx = kvm_rcx_read(vcpu);
1760 	u64 data = kvm_read_edx_eax(vcpu);
1761 	int r;
1762 
1763 	r = kvm_set_msr(vcpu, ecx, data);
1764 
1765 	/* MSR write failed? See if we should ask user space */
1766 	if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1767 		/* Bounce to user space */
1768 		return 0;
1769 
1770 	/* Signal all other negative errors to userspace */
1771 	if (r < 0)
1772 		return r;
1773 
1774 	if (!r)
1775 		trace_kvm_msr_write(ecx, data);
1776 	else
1777 		trace_kvm_msr_write_ex(ecx, data);
1778 
1779 	return kvm_x86_ops.complete_emulated_msr(vcpu, r);
1780 }
1781 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1782 
1783 bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1784 {
1785 	return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1786 		xfer_to_guest_mode_work_pending();
1787 }
1788 EXPORT_SYMBOL_GPL(kvm_vcpu_exit_request);
1789 
1790 /*
1791  * The fast path for frequent and performance sensitive wrmsr emulation,
1792  * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1793  * the latency of virtual IPI by avoiding the expensive bits of transitioning
1794  * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1795  * other cases which must be called after interrupts are enabled on the host.
1796  */
1797 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1798 {
1799 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1800 		return 1;
1801 
1802 	if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1803 		((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1804 		((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1805 		((u32)(data >> 32) != X2APIC_BROADCAST)) {
1806 
1807 		data &= ~(1 << 12);
1808 		kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1809 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1810 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1811 		trace_kvm_apic_write(APIC_ICR, (u32)data);
1812 		return 0;
1813 	}
1814 
1815 	return 1;
1816 }
1817 
1818 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1819 {
1820 	if (!kvm_can_use_hv_timer(vcpu))
1821 		return 1;
1822 
1823 	kvm_set_lapic_tscdeadline_msr(vcpu, data);
1824 	return 0;
1825 }
1826 
1827 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1828 {
1829 	u32 msr = kvm_rcx_read(vcpu);
1830 	u64 data;
1831 	fastpath_t ret = EXIT_FASTPATH_NONE;
1832 
1833 	switch (msr) {
1834 	case APIC_BASE_MSR + (APIC_ICR >> 4):
1835 		data = kvm_read_edx_eax(vcpu);
1836 		if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1837 			kvm_skip_emulated_instruction(vcpu);
1838 			ret = EXIT_FASTPATH_EXIT_HANDLED;
1839 		}
1840 		break;
1841 	case MSR_IA32_TSCDEADLINE:
1842 		data = kvm_read_edx_eax(vcpu);
1843 		if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1844 			kvm_skip_emulated_instruction(vcpu);
1845 			ret = EXIT_FASTPATH_REENTER_GUEST;
1846 		}
1847 		break;
1848 	default:
1849 		break;
1850 	}
1851 
1852 	if (ret != EXIT_FASTPATH_NONE)
1853 		trace_kvm_msr_write(msr, data);
1854 
1855 	return ret;
1856 }
1857 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1858 
1859 /*
1860  * Adapt set_msr() to msr_io()'s calling convention
1861  */
1862 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1863 {
1864 	return kvm_get_msr_ignored_check(vcpu, index, data, true);
1865 }
1866 
1867 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1868 {
1869 	return kvm_set_msr_ignored_check(vcpu, index, *data, true);
1870 }
1871 
1872 #ifdef CONFIG_X86_64
1873 struct pvclock_clock {
1874 	int vclock_mode;
1875 	u64 cycle_last;
1876 	u64 mask;
1877 	u32 mult;
1878 	u32 shift;
1879 	u64 base_cycles;
1880 	u64 offset;
1881 };
1882 
1883 struct pvclock_gtod_data {
1884 	seqcount_t	seq;
1885 
1886 	struct pvclock_clock clock; /* extract of a clocksource struct */
1887 	struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1888 
1889 	ktime_t		offs_boot;
1890 	u64		wall_time_sec;
1891 };
1892 
1893 static struct pvclock_gtod_data pvclock_gtod_data;
1894 
1895 static void update_pvclock_gtod(struct timekeeper *tk)
1896 {
1897 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1898 
1899 	write_seqcount_begin(&vdata->seq);
1900 
1901 	/* copy pvclock gtod data */
1902 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->vdso_clock_mode;
1903 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1904 	vdata->clock.mask		= tk->tkr_mono.mask;
1905 	vdata->clock.mult		= tk->tkr_mono.mult;
1906 	vdata->clock.shift		= tk->tkr_mono.shift;
1907 	vdata->clock.base_cycles	= tk->tkr_mono.xtime_nsec;
1908 	vdata->clock.offset		= tk->tkr_mono.base;
1909 
1910 	vdata->raw_clock.vclock_mode	= tk->tkr_raw.clock->vdso_clock_mode;
1911 	vdata->raw_clock.cycle_last	= tk->tkr_raw.cycle_last;
1912 	vdata->raw_clock.mask		= tk->tkr_raw.mask;
1913 	vdata->raw_clock.mult		= tk->tkr_raw.mult;
1914 	vdata->raw_clock.shift		= tk->tkr_raw.shift;
1915 	vdata->raw_clock.base_cycles	= tk->tkr_raw.xtime_nsec;
1916 	vdata->raw_clock.offset		= tk->tkr_raw.base;
1917 
1918 	vdata->wall_time_sec            = tk->xtime_sec;
1919 
1920 	vdata->offs_boot		= tk->offs_boot;
1921 
1922 	write_seqcount_end(&vdata->seq);
1923 }
1924 
1925 static s64 get_kvmclock_base_ns(void)
1926 {
1927 	/* Count up from boot time, but with the frequency of the raw clock.  */
1928 	return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
1929 }
1930 #else
1931 static s64 get_kvmclock_base_ns(void)
1932 {
1933 	/* Master clock not used, so we can just use CLOCK_BOOTTIME.  */
1934 	return ktime_get_boottime_ns();
1935 }
1936 #endif
1937 
1938 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1939 {
1940 	int version;
1941 	int r;
1942 	struct pvclock_wall_clock wc;
1943 	u64 wall_nsec;
1944 
1945 	kvm->arch.wall_clock = wall_clock;
1946 
1947 	if (!wall_clock)
1948 		return;
1949 
1950 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1951 	if (r)
1952 		return;
1953 
1954 	if (version & 1)
1955 		++version;  /* first time write, random junk */
1956 
1957 	++version;
1958 
1959 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1960 		return;
1961 
1962 	/*
1963 	 * The guest calculates current wall clock time by adding
1964 	 * system time (updated by kvm_guest_time_update below) to the
1965 	 * wall clock specified here.  We do the reverse here.
1966 	 */
1967 	wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
1968 
1969 	wc.nsec = do_div(wall_nsec, 1000000000);
1970 	wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
1971 	wc.version = version;
1972 
1973 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1974 
1975 	version++;
1976 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1977 }
1978 
1979 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
1980 				  bool old_msr, bool host_initiated)
1981 {
1982 	struct kvm_arch *ka = &vcpu->kvm->arch;
1983 
1984 	if (vcpu->vcpu_id == 0 && !host_initiated) {
1985 		if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
1986 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1987 
1988 		ka->boot_vcpu_runs_old_kvmclock = old_msr;
1989 	}
1990 
1991 	vcpu->arch.time = system_time;
1992 	kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
1993 
1994 	/* we verify if the enable bit is set... */
1995 	vcpu->arch.pv_time_enabled = false;
1996 	if (!(system_time & 1))
1997 		return;
1998 
1999 	if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2000 				       &vcpu->arch.pv_time, system_time & ~1ULL,
2001 				       sizeof(struct pvclock_vcpu_time_info)))
2002 		vcpu->arch.pv_time_enabled = true;
2003 
2004 	return;
2005 }
2006 
2007 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2008 {
2009 	do_shl32_div32(dividend, divisor);
2010 	return dividend;
2011 }
2012 
2013 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2014 			       s8 *pshift, u32 *pmultiplier)
2015 {
2016 	uint64_t scaled64;
2017 	int32_t  shift = 0;
2018 	uint64_t tps64;
2019 	uint32_t tps32;
2020 
2021 	tps64 = base_hz;
2022 	scaled64 = scaled_hz;
2023 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2024 		tps64 >>= 1;
2025 		shift--;
2026 	}
2027 
2028 	tps32 = (uint32_t)tps64;
2029 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2030 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2031 			scaled64 >>= 1;
2032 		else
2033 			tps32 <<= 1;
2034 		shift++;
2035 	}
2036 
2037 	*pshift = shift;
2038 	*pmultiplier = div_frac(scaled64, tps32);
2039 }
2040 
2041 #ifdef CONFIG_X86_64
2042 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2043 #endif
2044 
2045 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2046 static unsigned long max_tsc_khz;
2047 
2048 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2049 {
2050 	u64 v = (u64)khz * (1000000 + ppm);
2051 	do_div(v, 1000000);
2052 	return v;
2053 }
2054 
2055 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2056 {
2057 	u64 ratio;
2058 
2059 	/* Guest TSC same frequency as host TSC? */
2060 	if (!scale) {
2061 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2062 		return 0;
2063 	}
2064 
2065 	/* TSC scaling supported? */
2066 	if (!kvm_has_tsc_control) {
2067 		if (user_tsc_khz > tsc_khz) {
2068 			vcpu->arch.tsc_catchup = 1;
2069 			vcpu->arch.tsc_always_catchup = 1;
2070 			return 0;
2071 		} else {
2072 			pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2073 			return -1;
2074 		}
2075 	}
2076 
2077 	/* TSC scaling required  - calculate ratio */
2078 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2079 				user_tsc_khz, tsc_khz);
2080 
2081 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2082 		pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2083 			            user_tsc_khz);
2084 		return -1;
2085 	}
2086 
2087 	vcpu->arch.tsc_scaling_ratio = ratio;
2088 	return 0;
2089 }
2090 
2091 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2092 {
2093 	u32 thresh_lo, thresh_hi;
2094 	int use_scaling = 0;
2095 
2096 	/* tsc_khz can be zero if TSC calibration fails */
2097 	if (user_tsc_khz == 0) {
2098 		/* set tsc_scaling_ratio to a safe value */
2099 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2100 		return -1;
2101 	}
2102 
2103 	/* Compute a scale to convert nanoseconds in TSC cycles */
2104 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2105 			   &vcpu->arch.virtual_tsc_shift,
2106 			   &vcpu->arch.virtual_tsc_mult);
2107 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2108 
2109 	/*
2110 	 * Compute the variation in TSC rate which is acceptable
2111 	 * within the range of tolerance and decide if the
2112 	 * rate being applied is within that bounds of the hardware
2113 	 * rate.  If so, no scaling or compensation need be done.
2114 	 */
2115 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2116 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2117 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2118 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2119 		use_scaling = 1;
2120 	}
2121 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2122 }
2123 
2124 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2125 {
2126 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2127 				      vcpu->arch.virtual_tsc_mult,
2128 				      vcpu->arch.virtual_tsc_shift);
2129 	tsc += vcpu->arch.this_tsc_write;
2130 	return tsc;
2131 }
2132 
2133 static inline int gtod_is_based_on_tsc(int mode)
2134 {
2135 	return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2136 }
2137 
2138 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2139 {
2140 #ifdef CONFIG_X86_64
2141 	bool vcpus_matched;
2142 	struct kvm_arch *ka = &vcpu->kvm->arch;
2143 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2144 
2145 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2146 			 atomic_read(&vcpu->kvm->online_vcpus));
2147 
2148 	/*
2149 	 * Once the masterclock is enabled, always perform request in
2150 	 * order to update it.
2151 	 *
2152 	 * In order to enable masterclock, the host clocksource must be TSC
2153 	 * and the vcpus need to have matched TSCs.  When that happens,
2154 	 * perform request to enable masterclock.
2155 	 */
2156 	if (ka->use_master_clock ||
2157 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2158 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2159 
2160 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2161 			    atomic_read(&vcpu->kvm->online_vcpus),
2162 		            ka->use_master_clock, gtod->clock.vclock_mode);
2163 #endif
2164 }
2165 
2166 /*
2167  * Multiply tsc by a fixed point number represented by ratio.
2168  *
2169  * The most significant 64-N bits (mult) of ratio represent the
2170  * integral part of the fixed point number; the remaining N bits
2171  * (frac) represent the fractional part, ie. ratio represents a fixed
2172  * point number (mult + frac * 2^(-N)).
2173  *
2174  * N equals to kvm_tsc_scaling_ratio_frac_bits.
2175  */
2176 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2177 {
2178 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2179 }
2180 
2181 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
2182 {
2183 	u64 _tsc = tsc;
2184 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
2185 
2186 	if (ratio != kvm_default_tsc_scaling_ratio)
2187 		_tsc = __scale_tsc(ratio, tsc);
2188 
2189 	return _tsc;
2190 }
2191 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2192 
2193 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2194 {
2195 	u64 tsc;
2196 
2197 	tsc = kvm_scale_tsc(vcpu, rdtsc());
2198 
2199 	return target_tsc - tsc;
2200 }
2201 
2202 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2203 {
2204 	return vcpu->arch.l1_tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
2205 }
2206 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2207 
2208 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2209 {
2210 	vcpu->arch.l1_tsc_offset = offset;
2211 	vcpu->arch.tsc_offset = kvm_x86_ops.write_l1_tsc_offset(vcpu, offset);
2212 }
2213 
2214 static inline bool kvm_check_tsc_unstable(void)
2215 {
2216 #ifdef CONFIG_X86_64
2217 	/*
2218 	 * TSC is marked unstable when we're running on Hyper-V,
2219 	 * 'TSC page' clocksource is good.
2220 	 */
2221 	if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2222 		return false;
2223 #endif
2224 	return check_tsc_unstable();
2225 }
2226 
2227 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2228 {
2229 	struct kvm *kvm = vcpu->kvm;
2230 	u64 offset, ns, elapsed;
2231 	unsigned long flags;
2232 	bool matched;
2233 	bool already_matched;
2234 	bool synchronizing = false;
2235 
2236 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2237 	offset = kvm_compute_tsc_offset(vcpu, data);
2238 	ns = get_kvmclock_base_ns();
2239 	elapsed = ns - kvm->arch.last_tsc_nsec;
2240 
2241 	if (vcpu->arch.virtual_tsc_khz) {
2242 		if (data == 0) {
2243 			/*
2244 			 * detection of vcpu initialization -- need to sync
2245 			 * with other vCPUs. This particularly helps to keep
2246 			 * kvm_clock stable after CPU hotplug
2247 			 */
2248 			synchronizing = true;
2249 		} else {
2250 			u64 tsc_exp = kvm->arch.last_tsc_write +
2251 						nsec_to_cycles(vcpu, elapsed);
2252 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2253 			/*
2254 			 * Special case: TSC write with a small delta (1 second)
2255 			 * of virtual cycle time against real time is
2256 			 * interpreted as an attempt to synchronize the CPU.
2257 			 */
2258 			synchronizing = data < tsc_exp + tsc_hz &&
2259 					data + tsc_hz > tsc_exp;
2260 		}
2261 	}
2262 
2263 	/*
2264 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
2265 	 * TSC, we add elapsed time in this computation.  We could let the
2266 	 * compensation code attempt to catch up if we fall behind, but
2267 	 * it's better to try to match offsets from the beginning.
2268          */
2269 	if (synchronizing &&
2270 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2271 		if (!kvm_check_tsc_unstable()) {
2272 			offset = kvm->arch.cur_tsc_offset;
2273 		} else {
2274 			u64 delta = nsec_to_cycles(vcpu, elapsed);
2275 			data += delta;
2276 			offset = kvm_compute_tsc_offset(vcpu, data);
2277 		}
2278 		matched = true;
2279 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2280 	} else {
2281 		/*
2282 		 * We split periods of matched TSC writes into generations.
2283 		 * For each generation, we track the original measured
2284 		 * nanosecond time, offset, and write, so if TSCs are in
2285 		 * sync, we can match exact offset, and if not, we can match
2286 		 * exact software computation in compute_guest_tsc()
2287 		 *
2288 		 * These values are tracked in kvm->arch.cur_xxx variables.
2289 		 */
2290 		kvm->arch.cur_tsc_generation++;
2291 		kvm->arch.cur_tsc_nsec = ns;
2292 		kvm->arch.cur_tsc_write = data;
2293 		kvm->arch.cur_tsc_offset = offset;
2294 		matched = false;
2295 	}
2296 
2297 	/*
2298 	 * We also track th most recent recorded KHZ, write and time to
2299 	 * allow the matching interval to be extended at each write.
2300 	 */
2301 	kvm->arch.last_tsc_nsec = ns;
2302 	kvm->arch.last_tsc_write = data;
2303 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2304 
2305 	vcpu->arch.last_guest_tsc = data;
2306 
2307 	/* Keep track of which generation this VCPU has synchronized to */
2308 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2309 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2310 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2311 
2312 	kvm_vcpu_write_tsc_offset(vcpu, offset);
2313 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2314 
2315 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
2316 	if (!matched) {
2317 		kvm->arch.nr_vcpus_matched_tsc = 0;
2318 	} else if (!already_matched) {
2319 		kvm->arch.nr_vcpus_matched_tsc++;
2320 	}
2321 
2322 	kvm_track_tsc_matching(vcpu);
2323 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
2324 }
2325 
2326 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2327 					   s64 adjustment)
2328 {
2329 	u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2330 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2331 }
2332 
2333 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2334 {
2335 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2336 		WARN_ON(adjustment < 0);
2337 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2338 	adjust_tsc_offset_guest(vcpu, adjustment);
2339 }
2340 
2341 #ifdef CONFIG_X86_64
2342 
2343 static u64 read_tsc(void)
2344 {
2345 	u64 ret = (u64)rdtsc_ordered();
2346 	u64 last = pvclock_gtod_data.clock.cycle_last;
2347 
2348 	if (likely(ret >= last))
2349 		return ret;
2350 
2351 	/*
2352 	 * GCC likes to generate cmov here, but this branch is extremely
2353 	 * predictable (it's just a function of time and the likely is
2354 	 * very likely) and there's a data dependence, so force GCC
2355 	 * to generate a branch instead.  I don't barrier() because
2356 	 * we don't actually need a barrier, and if this function
2357 	 * ever gets inlined it will generate worse code.
2358 	 */
2359 	asm volatile ("");
2360 	return last;
2361 }
2362 
2363 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2364 			  int *mode)
2365 {
2366 	long v;
2367 	u64 tsc_pg_val;
2368 
2369 	switch (clock->vclock_mode) {
2370 	case VDSO_CLOCKMODE_HVCLOCK:
2371 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2372 						  tsc_timestamp);
2373 		if (tsc_pg_val != U64_MAX) {
2374 			/* TSC page valid */
2375 			*mode = VDSO_CLOCKMODE_HVCLOCK;
2376 			v = (tsc_pg_val - clock->cycle_last) &
2377 				clock->mask;
2378 		} else {
2379 			/* TSC page invalid */
2380 			*mode = VDSO_CLOCKMODE_NONE;
2381 		}
2382 		break;
2383 	case VDSO_CLOCKMODE_TSC:
2384 		*mode = VDSO_CLOCKMODE_TSC;
2385 		*tsc_timestamp = read_tsc();
2386 		v = (*tsc_timestamp - clock->cycle_last) &
2387 			clock->mask;
2388 		break;
2389 	default:
2390 		*mode = VDSO_CLOCKMODE_NONE;
2391 	}
2392 
2393 	if (*mode == VDSO_CLOCKMODE_NONE)
2394 		*tsc_timestamp = v = 0;
2395 
2396 	return v * clock->mult;
2397 }
2398 
2399 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2400 {
2401 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2402 	unsigned long seq;
2403 	int mode;
2404 	u64 ns;
2405 
2406 	do {
2407 		seq = read_seqcount_begin(&gtod->seq);
2408 		ns = gtod->raw_clock.base_cycles;
2409 		ns += vgettsc(&gtod->raw_clock, tsc_timestamp, &mode);
2410 		ns >>= gtod->raw_clock.shift;
2411 		ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2412 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2413 	*t = ns;
2414 
2415 	return mode;
2416 }
2417 
2418 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2419 {
2420 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2421 	unsigned long seq;
2422 	int mode;
2423 	u64 ns;
2424 
2425 	do {
2426 		seq = read_seqcount_begin(&gtod->seq);
2427 		ts->tv_sec = gtod->wall_time_sec;
2428 		ns = gtod->clock.base_cycles;
2429 		ns += vgettsc(&gtod->clock, tsc_timestamp, &mode);
2430 		ns >>= gtod->clock.shift;
2431 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2432 
2433 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2434 	ts->tv_nsec = ns;
2435 
2436 	return mode;
2437 }
2438 
2439 /* returns true if host is using TSC based clocksource */
2440 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2441 {
2442 	/* checked again under seqlock below */
2443 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2444 		return false;
2445 
2446 	return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2447 						      tsc_timestamp));
2448 }
2449 
2450 /* returns true if host is using TSC based clocksource */
2451 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2452 					   u64 *tsc_timestamp)
2453 {
2454 	/* checked again under seqlock below */
2455 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2456 		return false;
2457 
2458 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2459 }
2460 #endif
2461 
2462 /*
2463  *
2464  * Assuming a stable TSC across physical CPUS, and a stable TSC
2465  * across virtual CPUs, the following condition is possible.
2466  * Each numbered line represents an event visible to both
2467  * CPUs at the next numbered event.
2468  *
2469  * "timespecX" represents host monotonic time. "tscX" represents
2470  * RDTSC value.
2471  *
2472  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
2473  *
2474  * 1.  read timespec0,tsc0
2475  * 2.					| timespec1 = timespec0 + N
2476  * 					| tsc1 = tsc0 + M
2477  * 3. transition to guest		| transition to guest
2478  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2479  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
2480  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2481  *
2482  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2483  *
2484  * 	- ret0 < ret1
2485  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2486  *		...
2487  *	- 0 < N - M => M < N
2488  *
2489  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2490  * always the case (the difference between two distinct xtime instances
2491  * might be smaller then the difference between corresponding TSC reads,
2492  * when updating guest vcpus pvclock areas).
2493  *
2494  * To avoid that problem, do not allow visibility of distinct
2495  * system_timestamp/tsc_timestamp values simultaneously: use a master
2496  * copy of host monotonic time values. Update that master copy
2497  * in lockstep.
2498  *
2499  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2500  *
2501  */
2502 
2503 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2504 {
2505 #ifdef CONFIG_X86_64
2506 	struct kvm_arch *ka = &kvm->arch;
2507 	int vclock_mode;
2508 	bool host_tsc_clocksource, vcpus_matched;
2509 
2510 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2511 			atomic_read(&kvm->online_vcpus));
2512 
2513 	/*
2514 	 * If the host uses TSC clock, then passthrough TSC as stable
2515 	 * to the guest.
2516 	 */
2517 	host_tsc_clocksource = kvm_get_time_and_clockread(
2518 					&ka->master_kernel_ns,
2519 					&ka->master_cycle_now);
2520 
2521 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2522 				&& !ka->backwards_tsc_observed
2523 				&& !ka->boot_vcpu_runs_old_kvmclock;
2524 
2525 	if (ka->use_master_clock)
2526 		atomic_set(&kvm_guest_has_master_clock, 1);
2527 
2528 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2529 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2530 					vcpus_matched);
2531 #endif
2532 }
2533 
2534 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2535 {
2536 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2537 }
2538 
2539 static void kvm_gen_update_masterclock(struct kvm *kvm)
2540 {
2541 #ifdef CONFIG_X86_64
2542 	int i;
2543 	struct kvm_vcpu *vcpu;
2544 	struct kvm_arch *ka = &kvm->arch;
2545 
2546 	spin_lock(&ka->pvclock_gtod_sync_lock);
2547 	kvm_make_mclock_inprogress_request(kvm);
2548 	/* no guest entries from this point */
2549 	pvclock_update_vm_gtod_copy(kvm);
2550 
2551 	kvm_for_each_vcpu(i, vcpu, kvm)
2552 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2553 
2554 	/* guest entries allowed */
2555 	kvm_for_each_vcpu(i, vcpu, kvm)
2556 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2557 
2558 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2559 #endif
2560 }
2561 
2562 u64 get_kvmclock_ns(struct kvm *kvm)
2563 {
2564 	struct kvm_arch *ka = &kvm->arch;
2565 	struct pvclock_vcpu_time_info hv_clock;
2566 	u64 ret;
2567 
2568 	spin_lock(&ka->pvclock_gtod_sync_lock);
2569 	if (!ka->use_master_clock) {
2570 		spin_unlock(&ka->pvclock_gtod_sync_lock);
2571 		return get_kvmclock_base_ns() + ka->kvmclock_offset;
2572 	}
2573 
2574 	hv_clock.tsc_timestamp = ka->master_cycle_now;
2575 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2576 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2577 
2578 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2579 	get_cpu();
2580 
2581 	if (__this_cpu_read(cpu_tsc_khz)) {
2582 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2583 				   &hv_clock.tsc_shift,
2584 				   &hv_clock.tsc_to_system_mul);
2585 		ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2586 	} else
2587 		ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2588 
2589 	put_cpu();
2590 
2591 	return ret;
2592 }
2593 
2594 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2595 {
2596 	struct kvm_vcpu_arch *vcpu = &v->arch;
2597 	struct pvclock_vcpu_time_info guest_hv_clock;
2598 
2599 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2600 		&guest_hv_clock, sizeof(guest_hv_clock))))
2601 		return;
2602 
2603 	/* This VCPU is paused, but it's legal for a guest to read another
2604 	 * VCPU's kvmclock, so we really have to follow the specification where
2605 	 * it says that version is odd if data is being modified, and even after
2606 	 * it is consistent.
2607 	 *
2608 	 * Version field updates must be kept separate.  This is because
2609 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2610 	 * writes within a string instruction are weakly ordered.  So there
2611 	 * are three writes overall.
2612 	 *
2613 	 * As a small optimization, only write the version field in the first
2614 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2615 	 * version field is the first in the struct.
2616 	 */
2617 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2618 
2619 	if (guest_hv_clock.version & 1)
2620 		++guest_hv_clock.version;  /* first time write, random junk */
2621 
2622 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2623 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2624 				&vcpu->hv_clock,
2625 				sizeof(vcpu->hv_clock.version));
2626 
2627 	smp_wmb();
2628 
2629 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2630 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2631 
2632 	if (vcpu->pvclock_set_guest_stopped_request) {
2633 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2634 		vcpu->pvclock_set_guest_stopped_request = false;
2635 	}
2636 
2637 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2638 
2639 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2640 				&vcpu->hv_clock,
2641 				sizeof(vcpu->hv_clock));
2642 
2643 	smp_wmb();
2644 
2645 	vcpu->hv_clock.version++;
2646 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2647 				&vcpu->hv_clock,
2648 				sizeof(vcpu->hv_clock.version));
2649 }
2650 
2651 static int kvm_guest_time_update(struct kvm_vcpu *v)
2652 {
2653 	unsigned long flags, tgt_tsc_khz;
2654 	struct kvm_vcpu_arch *vcpu = &v->arch;
2655 	struct kvm_arch *ka = &v->kvm->arch;
2656 	s64 kernel_ns;
2657 	u64 tsc_timestamp, host_tsc;
2658 	u8 pvclock_flags;
2659 	bool use_master_clock;
2660 
2661 	kernel_ns = 0;
2662 	host_tsc = 0;
2663 
2664 	/*
2665 	 * If the host uses TSC clock, then passthrough TSC as stable
2666 	 * to the guest.
2667 	 */
2668 	spin_lock(&ka->pvclock_gtod_sync_lock);
2669 	use_master_clock = ka->use_master_clock;
2670 	if (use_master_clock) {
2671 		host_tsc = ka->master_cycle_now;
2672 		kernel_ns = ka->master_kernel_ns;
2673 	}
2674 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2675 
2676 	/* Keep irq disabled to prevent changes to the clock */
2677 	local_irq_save(flags);
2678 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2679 	if (unlikely(tgt_tsc_khz == 0)) {
2680 		local_irq_restore(flags);
2681 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2682 		return 1;
2683 	}
2684 	if (!use_master_clock) {
2685 		host_tsc = rdtsc();
2686 		kernel_ns = get_kvmclock_base_ns();
2687 	}
2688 
2689 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2690 
2691 	/*
2692 	 * We may have to catch up the TSC to match elapsed wall clock
2693 	 * time for two reasons, even if kvmclock is used.
2694 	 *   1) CPU could have been running below the maximum TSC rate
2695 	 *   2) Broken TSC compensation resets the base at each VCPU
2696 	 *      entry to avoid unknown leaps of TSC even when running
2697 	 *      again on the same CPU.  This may cause apparent elapsed
2698 	 *      time to disappear, and the guest to stand still or run
2699 	 *	very slowly.
2700 	 */
2701 	if (vcpu->tsc_catchup) {
2702 		u64 tsc = compute_guest_tsc(v, kernel_ns);
2703 		if (tsc > tsc_timestamp) {
2704 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2705 			tsc_timestamp = tsc;
2706 		}
2707 	}
2708 
2709 	local_irq_restore(flags);
2710 
2711 	/* With all the info we got, fill in the values */
2712 
2713 	if (kvm_has_tsc_control)
2714 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2715 
2716 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2717 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2718 				   &vcpu->hv_clock.tsc_shift,
2719 				   &vcpu->hv_clock.tsc_to_system_mul);
2720 		vcpu->hw_tsc_khz = tgt_tsc_khz;
2721 	}
2722 
2723 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2724 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2725 	vcpu->last_guest_tsc = tsc_timestamp;
2726 
2727 	/* If the host uses TSC clocksource, then it is stable */
2728 	pvclock_flags = 0;
2729 	if (use_master_clock)
2730 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2731 
2732 	vcpu->hv_clock.flags = pvclock_flags;
2733 
2734 	if (vcpu->pv_time_enabled)
2735 		kvm_setup_pvclock_page(v);
2736 	if (v == kvm_get_vcpu(v->kvm, 0))
2737 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2738 	return 0;
2739 }
2740 
2741 /*
2742  * kvmclock updates which are isolated to a given vcpu, such as
2743  * vcpu->cpu migration, should not allow system_timestamp from
2744  * the rest of the vcpus to remain static. Otherwise ntp frequency
2745  * correction applies to one vcpu's system_timestamp but not
2746  * the others.
2747  *
2748  * So in those cases, request a kvmclock update for all vcpus.
2749  * We need to rate-limit these requests though, as they can
2750  * considerably slow guests that have a large number of vcpus.
2751  * The time for a remote vcpu to update its kvmclock is bound
2752  * by the delay we use to rate-limit the updates.
2753  */
2754 
2755 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2756 
2757 static void kvmclock_update_fn(struct work_struct *work)
2758 {
2759 	int i;
2760 	struct delayed_work *dwork = to_delayed_work(work);
2761 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2762 					   kvmclock_update_work);
2763 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2764 	struct kvm_vcpu *vcpu;
2765 
2766 	kvm_for_each_vcpu(i, vcpu, kvm) {
2767 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2768 		kvm_vcpu_kick(vcpu);
2769 	}
2770 }
2771 
2772 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2773 {
2774 	struct kvm *kvm = v->kvm;
2775 
2776 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2777 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2778 					KVMCLOCK_UPDATE_DELAY);
2779 }
2780 
2781 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2782 
2783 static void kvmclock_sync_fn(struct work_struct *work)
2784 {
2785 	struct delayed_work *dwork = to_delayed_work(work);
2786 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2787 					   kvmclock_sync_work);
2788 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2789 
2790 	if (!kvmclock_periodic_sync)
2791 		return;
2792 
2793 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2794 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2795 					KVMCLOCK_SYNC_PERIOD);
2796 }
2797 
2798 /*
2799  * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2800  */
2801 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2802 {
2803 	/* McStatusWrEn enabled? */
2804 	if (guest_cpuid_is_amd_or_hygon(vcpu))
2805 		return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2806 
2807 	return false;
2808 }
2809 
2810 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2811 {
2812 	u64 mcg_cap = vcpu->arch.mcg_cap;
2813 	unsigned bank_num = mcg_cap & 0xff;
2814 	u32 msr = msr_info->index;
2815 	u64 data = msr_info->data;
2816 
2817 	switch (msr) {
2818 	case MSR_IA32_MCG_STATUS:
2819 		vcpu->arch.mcg_status = data;
2820 		break;
2821 	case MSR_IA32_MCG_CTL:
2822 		if (!(mcg_cap & MCG_CTL_P) &&
2823 		    (data || !msr_info->host_initiated))
2824 			return 1;
2825 		if (data != 0 && data != ~(u64)0)
2826 			return 1;
2827 		vcpu->arch.mcg_ctl = data;
2828 		break;
2829 	default:
2830 		if (msr >= MSR_IA32_MC0_CTL &&
2831 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2832 			u32 offset = array_index_nospec(
2833 				msr - MSR_IA32_MC0_CTL,
2834 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2835 
2836 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2837 			 * some Linux kernels though clear bit 10 in bank 4 to
2838 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2839 			 * this to avoid an uncatched #GP in the guest
2840 			 */
2841 			if ((offset & 0x3) == 0 &&
2842 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2843 				return -1;
2844 
2845 			/* MCi_STATUS */
2846 			if (!msr_info->host_initiated &&
2847 			    (offset & 0x3) == 1 && data != 0) {
2848 				if (!can_set_mci_status(vcpu))
2849 					return -1;
2850 			}
2851 
2852 			vcpu->arch.mce_banks[offset] = data;
2853 			break;
2854 		}
2855 		return 1;
2856 	}
2857 	return 0;
2858 }
2859 
2860 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2861 {
2862 	struct kvm *kvm = vcpu->kvm;
2863 	int lm = is_long_mode(vcpu);
2864 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2865 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2866 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2867 		: kvm->arch.xen_hvm_config.blob_size_32;
2868 	u32 page_num = data & ~PAGE_MASK;
2869 	u64 page_addr = data & PAGE_MASK;
2870 	u8 *page;
2871 
2872 	if (page_num >= blob_size)
2873 		return 1;
2874 
2875 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2876 	if (IS_ERR(page))
2877 		return PTR_ERR(page);
2878 
2879 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE)) {
2880 		kfree(page);
2881 		return 1;
2882 	}
2883 	return 0;
2884 }
2885 
2886 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
2887 {
2888 	u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
2889 
2890 	return (vcpu->arch.apf.msr_en_val & mask) == mask;
2891 }
2892 
2893 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2894 {
2895 	gpa_t gpa = data & ~0x3f;
2896 
2897 	/* Bits 4:5 are reserved, Should be zero */
2898 	if (data & 0x30)
2899 		return 1;
2900 
2901 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
2902 	    (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
2903 		return 1;
2904 
2905 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
2906 	    (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
2907 		return 1;
2908 
2909 	if (!lapic_in_kernel(vcpu))
2910 		return data ? 1 : 0;
2911 
2912 	vcpu->arch.apf.msr_en_val = data;
2913 
2914 	if (!kvm_pv_async_pf_enabled(vcpu)) {
2915 		kvm_clear_async_pf_completion_queue(vcpu);
2916 		kvm_async_pf_hash_reset(vcpu);
2917 		return 0;
2918 	}
2919 
2920 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2921 					sizeof(u64)))
2922 		return 1;
2923 
2924 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2925 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2926 
2927 	kvm_async_pf_wakeup_all(vcpu);
2928 
2929 	return 0;
2930 }
2931 
2932 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
2933 {
2934 	/* Bits 8-63 are reserved */
2935 	if (data >> 8)
2936 		return 1;
2937 
2938 	if (!lapic_in_kernel(vcpu))
2939 		return 1;
2940 
2941 	vcpu->arch.apf.msr_int_val = data;
2942 
2943 	vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
2944 
2945 	return 0;
2946 }
2947 
2948 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2949 {
2950 	vcpu->arch.pv_time_enabled = false;
2951 	vcpu->arch.time = 0;
2952 }
2953 
2954 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
2955 {
2956 	++vcpu->stat.tlb_flush;
2957 	kvm_x86_ops.tlb_flush_all(vcpu);
2958 }
2959 
2960 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
2961 {
2962 	++vcpu->stat.tlb_flush;
2963 	kvm_x86_ops.tlb_flush_guest(vcpu);
2964 }
2965 
2966 static void record_steal_time(struct kvm_vcpu *vcpu)
2967 {
2968 	struct kvm_host_map map;
2969 	struct kvm_steal_time *st;
2970 
2971 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2972 		return;
2973 
2974 	/* -EAGAIN is returned in atomic context so we can just return. */
2975 	if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
2976 			&map, &vcpu->arch.st.cache, false))
2977 		return;
2978 
2979 	st = map.hva +
2980 		offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
2981 
2982 	/*
2983 	 * Doing a TLB flush here, on the guest's behalf, can avoid
2984 	 * expensive IPIs.
2985 	 */
2986 	if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
2987 		trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2988 				       st->preempted & KVM_VCPU_FLUSH_TLB);
2989 		if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
2990 			kvm_vcpu_flush_tlb_guest(vcpu);
2991 	}
2992 
2993 	vcpu->arch.st.preempted = 0;
2994 
2995 	if (st->version & 1)
2996 		st->version += 1;  /* first time write, random junk */
2997 
2998 	st->version += 1;
2999 
3000 	smp_wmb();
3001 
3002 	st->steal += current->sched_info.run_delay -
3003 		vcpu->arch.st.last_steal;
3004 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
3005 
3006 	smp_wmb();
3007 
3008 	st->version += 1;
3009 
3010 	kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
3011 }
3012 
3013 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3014 {
3015 	bool pr = false;
3016 	u32 msr = msr_info->index;
3017 	u64 data = msr_info->data;
3018 
3019 	switch (msr) {
3020 	case MSR_AMD64_NB_CFG:
3021 	case MSR_IA32_UCODE_WRITE:
3022 	case MSR_VM_HSAVE_PA:
3023 	case MSR_AMD64_PATCH_LOADER:
3024 	case MSR_AMD64_BU_CFG2:
3025 	case MSR_AMD64_DC_CFG:
3026 	case MSR_F15H_EX_CFG:
3027 		break;
3028 
3029 	case MSR_IA32_UCODE_REV:
3030 		if (msr_info->host_initiated)
3031 			vcpu->arch.microcode_version = data;
3032 		break;
3033 	case MSR_IA32_ARCH_CAPABILITIES:
3034 		if (!msr_info->host_initiated)
3035 			return 1;
3036 		vcpu->arch.arch_capabilities = data;
3037 		break;
3038 	case MSR_IA32_PERF_CAPABILITIES: {
3039 		struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3040 
3041 		if (!msr_info->host_initiated)
3042 			return 1;
3043 		if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3044 			return 1;
3045 		if (data & ~msr_ent.data)
3046 			return 1;
3047 
3048 		vcpu->arch.perf_capabilities = data;
3049 
3050 		return 0;
3051 		}
3052 	case MSR_EFER:
3053 		return set_efer(vcpu, msr_info);
3054 	case MSR_K7_HWCR:
3055 		data &= ~(u64)0x40;	/* ignore flush filter disable */
3056 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
3057 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
3058 
3059 		/* Handle McStatusWrEn */
3060 		if (data == BIT_ULL(18)) {
3061 			vcpu->arch.msr_hwcr = data;
3062 		} else if (data != 0) {
3063 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3064 				    data);
3065 			return 1;
3066 		}
3067 		break;
3068 	case MSR_FAM10H_MMIO_CONF_BASE:
3069 		if (data != 0) {
3070 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3071 				    "0x%llx\n", data);
3072 			return 1;
3073 		}
3074 		break;
3075 	case MSR_IA32_DEBUGCTLMSR:
3076 		if (!data) {
3077 			/* We support the non-activated case already */
3078 			break;
3079 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
3080 			/* Values other than LBR and BTF are vendor-specific,
3081 			   thus reserved and should throw a #GP */
3082 			return 1;
3083 		} else if (report_ignored_msrs)
3084 			vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
3085 				    __func__, data);
3086 		break;
3087 	case 0x200 ... 0x2ff:
3088 		return kvm_mtrr_set_msr(vcpu, msr, data);
3089 	case MSR_IA32_APICBASE:
3090 		return kvm_set_apic_base(vcpu, msr_info);
3091 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3092 		return kvm_x2apic_msr_write(vcpu, msr, data);
3093 	case MSR_IA32_TSCDEADLINE:
3094 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
3095 		break;
3096 	case MSR_IA32_TSC_ADJUST:
3097 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3098 			if (!msr_info->host_initiated) {
3099 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3100 				adjust_tsc_offset_guest(vcpu, adj);
3101 			}
3102 			vcpu->arch.ia32_tsc_adjust_msr = data;
3103 		}
3104 		break;
3105 	case MSR_IA32_MISC_ENABLE:
3106 		if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3107 		    ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3108 			if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3109 				return 1;
3110 			vcpu->arch.ia32_misc_enable_msr = data;
3111 			kvm_update_cpuid_runtime(vcpu);
3112 		} else {
3113 			vcpu->arch.ia32_misc_enable_msr = data;
3114 		}
3115 		break;
3116 	case MSR_IA32_SMBASE:
3117 		if (!msr_info->host_initiated)
3118 			return 1;
3119 		vcpu->arch.smbase = data;
3120 		break;
3121 	case MSR_IA32_POWER_CTL:
3122 		vcpu->arch.msr_ia32_power_ctl = data;
3123 		break;
3124 	case MSR_IA32_TSC:
3125 		if (msr_info->host_initiated) {
3126 			kvm_synchronize_tsc(vcpu, data);
3127 		} else {
3128 			u64 adj = kvm_compute_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3129 			adjust_tsc_offset_guest(vcpu, adj);
3130 			vcpu->arch.ia32_tsc_adjust_msr += adj;
3131 		}
3132 		break;
3133 	case MSR_IA32_XSS:
3134 		if (!msr_info->host_initiated &&
3135 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3136 			return 1;
3137 		/*
3138 		 * KVM supports exposing PT to the guest, but does not support
3139 		 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3140 		 * XSAVES/XRSTORS to save/restore PT MSRs.
3141 		 */
3142 		if (data & ~supported_xss)
3143 			return 1;
3144 		vcpu->arch.ia32_xss = data;
3145 		break;
3146 	case MSR_SMI_COUNT:
3147 		if (!msr_info->host_initiated)
3148 			return 1;
3149 		vcpu->arch.smi_count = data;
3150 		break;
3151 	case MSR_KVM_WALL_CLOCK_NEW:
3152 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3153 			return 1;
3154 
3155 		kvm_write_wall_clock(vcpu->kvm, data);
3156 		break;
3157 	case MSR_KVM_WALL_CLOCK:
3158 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3159 			return 1;
3160 
3161 		kvm_write_wall_clock(vcpu->kvm, data);
3162 		break;
3163 	case MSR_KVM_SYSTEM_TIME_NEW:
3164 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3165 			return 1;
3166 
3167 		kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3168 		break;
3169 	case MSR_KVM_SYSTEM_TIME:
3170 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3171 			return 1;
3172 
3173 		kvm_write_system_time(vcpu, data, true,  msr_info->host_initiated);
3174 		break;
3175 	case MSR_KVM_ASYNC_PF_EN:
3176 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3177 			return 1;
3178 
3179 		if (kvm_pv_enable_async_pf(vcpu, data))
3180 			return 1;
3181 		break;
3182 	case MSR_KVM_ASYNC_PF_INT:
3183 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3184 			return 1;
3185 
3186 		if (kvm_pv_enable_async_pf_int(vcpu, data))
3187 			return 1;
3188 		break;
3189 	case MSR_KVM_ASYNC_PF_ACK:
3190 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3191 			return 1;
3192 		if (data & 0x1) {
3193 			vcpu->arch.apf.pageready_pending = false;
3194 			kvm_check_async_pf_completion(vcpu);
3195 		}
3196 		break;
3197 	case MSR_KVM_STEAL_TIME:
3198 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3199 			return 1;
3200 
3201 		if (unlikely(!sched_info_on()))
3202 			return 1;
3203 
3204 		if (data & KVM_STEAL_RESERVED_MASK)
3205 			return 1;
3206 
3207 		vcpu->arch.st.msr_val = data;
3208 
3209 		if (!(data & KVM_MSR_ENABLED))
3210 			break;
3211 
3212 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3213 
3214 		break;
3215 	case MSR_KVM_PV_EOI_EN:
3216 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3217 			return 1;
3218 
3219 		if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3220 			return 1;
3221 		break;
3222 
3223 	case MSR_KVM_POLL_CONTROL:
3224 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3225 			return 1;
3226 
3227 		/* only enable bit supported */
3228 		if (data & (-1ULL << 1))
3229 			return 1;
3230 
3231 		vcpu->arch.msr_kvm_poll_control = data;
3232 		break;
3233 
3234 	case MSR_IA32_MCG_CTL:
3235 	case MSR_IA32_MCG_STATUS:
3236 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3237 		return set_msr_mce(vcpu, msr_info);
3238 
3239 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3240 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3241 		pr = true;
3242 		fallthrough;
3243 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3244 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3245 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3246 			return kvm_pmu_set_msr(vcpu, msr_info);
3247 
3248 		if (pr || data != 0)
3249 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3250 				    "0x%x data 0x%llx\n", msr, data);
3251 		break;
3252 	case MSR_K7_CLK_CTL:
3253 		/*
3254 		 * Ignore all writes to this no longer documented MSR.
3255 		 * Writes are only relevant for old K7 processors,
3256 		 * all pre-dating SVM, but a recommended workaround from
3257 		 * AMD for these chips. It is possible to specify the
3258 		 * affected processor models on the command line, hence
3259 		 * the need to ignore the workaround.
3260 		 */
3261 		break;
3262 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3263 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3264 	case HV_X64_MSR_SYNDBG_OPTIONS:
3265 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3266 	case HV_X64_MSR_CRASH_CTL:
3267 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3268 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3269 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3270 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3271 		return kvm_hv_set_msr_common(vcpu, msr, data,
3272 					     msr_info->host_initiated);
3273 	case MSR_IA32_BBL_CR_CTL3:
3274 		/* Drop writes to this legacy MSR -- see rdmsr
3275 		 * counterpart for further detail.
3276 		 */
3277 		if (report_ignored_msrs)
3278 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3279 				msr, data);
3280 		break;
3281 	case MSR_AMD64_OSVW_ID_LENGTH:
3282 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3283 			return 1;
3284 		vcpu->arch.osvw.length = data;
3285 		break;
3286 	case MSR_AMD64_OSVW_STATUS:
3287 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3288 			return 1;
3289 		vcpu->arch.osvw.status = data;
3290 		break;
3291 	case MSR_PLATFORM_INFO:
3292 		if (!msr_info->host_initiated ||
3293 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3294 		     cpuid_fault_enabled(vcpu)))
3295 			return 1;
3296 		vcpu->arch.msr_platform_info = data;
3297 		break;
3298 	case MSR_MISC_FEATURES_ENABLES:
3299 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3300 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3301 		     !supports_cpuid_fault(vcpu)))
3302 			return 1;
3303 		vcpu->arch.msr_misc_features_enables = data;
3304 		break;
3305 	default:
3306 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
3307 			return xen_hvm_config(vcpu, data);
3308 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3309 			return kvm_pmu_set_msr(vcpu, msr_info);
3310 		return KVM_MSR_RET_INVALID;
3311 	}
3312 	return 0;
3313 }
3314 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3315 
3316 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3317 {
3318 	u64 data;
3319 	u64 mcg_cap = vcpu->arch.mcg_cap;
3320 	unsigned bank_num = mcg_cap & 0xff;
3321 
3322 	switch (msr) {
3323 	case MSR_IA32_P5_MC_ADDR:
3324 	case MSR_IA32_P5_MC_TYPE:
3325 		data = 0;
3326 		break;
3327 	case MSR_IA32_MCG_CAP:
3328 		data = vcpu->arch.mcg_cap;
3329 		break;
3330 	case MSR_IA32_MCG_CTL:
3331 		if (!(mcg_cap & MCG_CTL_P) && !host)
3332 			return 1;
3333 		data = vcpu->arch.mcg_ctl;
3334 		break;
3335 	case MSR_IA32_MCG_STATUS:
3336 		data = vcpu->arch.mcg_status;
3337 		break;
3338 	default:
3339 		if (msr >= MSR_IA32_MC0_CTL &&
3340 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
3341 			u32 offset = array_index_nospec(
3342 				msr - MSR_IA32_MC0_CTL,
3343 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3344 
3345 			data = vcpu->arch.mce_banks[offset];
3346 			break;
3347 		}
3348 		return 1;
3349 	}
3350 	*pdata = data;
3351 	return 0;
3352 }
3353 
3354 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3355 {
3356 	switch (msr_info->index) {
3357 	case MSR_IA32_PLATFORM_ID:
3358 	case MSR_IA32_EBL_CR_POWERON:
3359 	case MSR_IA32_DEBUGCTLMSR:
3360 	case MSR_IA32_LASTBRANCHFROMIP:
3361 	case MSR_IA32_LASTBRANCHTOIP:
3362 	case MSR_IA32_LASTINTFROMIP:
3363 	case MSR_IA32_LASTINTTOIP:
3364 	case MSR_K8_SYSCFG:
3365 	case MSR_K8_TSEG_ADDR:
3366 	case MSR_K8_TSEG_MASK:
3367 	case MSR_VM_HSAVE_PA:
3368 	case MSR_K8_INT_PENDING_MSG:
3369 	case MSR_AMD64_NB_CFG:
3370 	case MSR_FAM10H_MMIO_CONF_BASE:
3371 	case MSR_AMD64_BU_CFG2:
3372 	case MSR_IA32_PERF_CTL:
3373 	case MSR_AMD64_DC_CFG:
3374 	case MSR_F15H_EX_CFG:
3375 	/*
3376 	 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3377 	 * limit) MSRs. Just return 0, as we do not want to expose the host
3378 	 * data here. Do not conditionalize this on CPUID, as KVM does not do
3379 	 * so for existing CPU-specific MSRs.
3380 	 */
3381 	case MSR_RAPL_POWER_UNIT:
3382 	case MSR_PP0_ENERGY_STATUS:	/* Power plane 0 (core) */
3383 	case MSR_PP1_ENERGY_STATUS:	/* Power plane 1 (graphics uncore) */
3384 	case MSR_PKG_ENERGY_STATUS:	/* Total package */
3385 	case MSR_DRAM_ENERGY_STATUS:	/* DRAM controller */
3386 		msr_info->data = 0;
3387 		break;
3388 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3389 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3390 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3391 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3392 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3393 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3394 			return kvm_pmu_get_msr(vcpu, msr_info);
3395 		msr_info->data = 0;
3396 		break;
3397 	case MSR_IA32_UCODE_REV:
3398 		msr_info->data = vcpu->arch.microcode_version;
3399 		break;
3400 	case MSR_IA32_ARCH_CAPABILITIES:
3401 		if (!msr_info->host_initiated &&
3402 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3403 			return 1;
3404 		msr_info->data = vcpu->arch.arch_capabilities;
3405 		break;
3406 	case MSR_IA32_PERF_CAPABILITIES:
3407 		if (!msr_info->host_initiated &&
3408 		    !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3409 			return 1;
3410 		msr_info->data = vcpu->arch.perf_capabilities;
3411 		break;
3412 	case MSR_IA32_POWER_CTL:
3413 		msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3414 		break;
3415 	case MSR_IA32_TSC: {
3416 		/*
3417 		 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3418 		 * even when not intercepted. AMD manual doesn't explicitly
3419 		 * state this but appears to behave the same.
3420 		 *
3421 		 * On userspace reads and writes, however, we unconditionally
3422 		 * return L1's TSC value to ensure backwards-compatible
3423 		 * behavior for migration.
3424 		 */
3425 		u64 tsc_offset = msr_info->host_initiated ? vcpu->arch.l1_tsc_offset :
3426 							    vcpu->arch.tsc_offset;
3427 
3428 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + tsc_offset;
3429 		break;
3430 	}
3431 	case MSR_MTRRcap:
3432 	case 0x200 ... 0x2ff:
3433 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3434 	case 0xcd: /* fsb frequency */
3435 		msr_info->data = 3;
3436 		break;
3437 		/*
3438 		 * MSR_EBC_FREQUENCY_ID
3439 		 * Conservative value valid for even the basic CPU models.
3440 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3441 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3442 		 * and 266MHz for model 3, or 4. Set Core Clock
3443 		 * Frequency to System Bus Frequency Ratio to 1 (bits
3444 		 * 31:24) even though these are only valid for CPU
3445 		 * models > 2, however guests may end up dividing or
3446 		 * multiplying by zero otherwise.
3447 		 */
3448 	case MSR_EBC_FREQUENCY_ID:
3449 		msr_info->data = 1 << 24;
3450 		break;
3451 	case MSR_IA32_APICBASE:
3452 		msr_info->data = kvm_get_apic_base(vcpu);
3453 		break;
3454 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3455 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3456 	case MSR_IA32_TSCDEADLINE:
3457 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3458 		break;
3459 	case MSR_IA32_TSC_ADJUST:
3460 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3461 		break;
3462 	case MSR_IA32_MISC_ENABLE:
3463 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3464 		break;
3465 	case MSR_IA32_SMBASE:
3466 		if (!msr_info->host_initiated)
3467 			return 1;
3468 		msr_info->data = vcpu->arch.smbase;
3469 		break;
3470 	case MSR_SMI_COUNT:
3471 		msr_info->data = vcpu->arch.smi_count;
3472 		break;
3473 	case MSR_IA32_PERF_STATUS:
3474 		/* TSC increment by tick */
3475 		msr_info->data = 1000ULL;
3476 		/* CPU multiplier */
3477 		msr_info->data |= (((uint64_t)4ULL) << 40);
3478 		break;
3479 	case MSR_EFER:
3480 		msr_info->data = vcpu->arch.efer;
3481 		break;
3482 	case MSR_KVM_WALL_CLOCK:
3483 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3484 			return 1;
3485 
3486 		msr_info->data = vcpu->kvm->arch.wall_clock;
3487 		break;
3488 	case MSR_KVM_WALL_CLOCK_NEW:
3489 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3490 			return 1;
3491 
3492 		msr_info->data = vcpu->kvm->arch.wall_clock;
3493 		break;
3494 	case MSR_KVM_SYSTEM_TIME:
3495 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3496 			return 1;
3497 
3498 		msr_info->data = vcpu->arch.time;
3499 		break;
3500 	case MSR_KVM_SYSTEM_TIME_NEW:
3501 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3502 			return 1;
3503 
3504 		msr_info->data = vcpu->arch.time;
3505 		break;
3506 	case MSR_KVM_ASYNC_PF_EN:
3507 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3508 			return 1;
3509 
3510 		msr_info->data = vcpu->arch.apf.msr_en_val;
3511 		break;
3512 	case MSR_KVM_ASYNC_PF_INT:
3513 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3514 			return 1;
3515 
3516 		msr_info->data = vcpu->arch.apf.msr_int_val;
3517 		break;
3518 	case MSR_KVM_ASYNC_PF_ACK:
3519 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3520 			return 1;
3521 
3522 		msr_info->data = 0;
3523 		break;
3524 	case MSR_KVM_STEAL_TIME:
3525 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3526 			return 1;
3527 
3528 		msr_info->data = vcpu->arch.st.msr_val;
3529 		break;
3530 	case MSR_KVM_PV_EOI_EN:
3531 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3532 			return 1;
3533 
3534 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
3535 		break;
3536 	case MSR_KVM_POLL_CONTROL:
3537 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3538 			return 1;
3539 
3540 		msr_info->data = vcpu->arch.msr_kvm_poll_control;
3541 		break;
3542 	case MSR_IA32_P5_MC_ADDR:
3543 	case MSR_IA32_P5_MC_TYPE:
3544 	case MSR_IA32_MCG_CAP:
3545 	case MSR_IA32_MCG_CTL:
3546 	case MSR_IA32_MCG_STATUS:
3547 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3548 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3549 				   msr_info->host_initiated);
3550 	case MSR_IA32_XSS:
3551 		if (!msr_info->host_initiated &&
3552 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3553 			return 1;
3554 		msr_info->data = vcpu->arch.ia32_xss;
3555 		break;
3556 	case MSR_K7_CLK_CTL:
3557 		/*
3558 		 * Provide expected ramp-up count for K7. All other
3559 		 * are set to zero, indicating minimum divisors for
3560 		 * every field.
3561 		 *
3562 		 * This prevents guest kernels on AMD host with CPU
3563 		 * type 6, model 8 and higher from exploding due to
3564 		 * the rdmsr failing.
3565 		 */
3566 		msr_info->data = 0x20000000;
3567 		break;
3568 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3569 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3570 	case HV_X64_MSR_SYNDBG_OPTIONS:
3571 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3572 	case HV_X64_MSR_CRASH_CTL:
3573 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3574 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3575 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3576 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3577 		return kvm_hv_get_msr_common(vcpu,
3578 					     msr_info->index, &msr_info->data,
3579 					     msr_info->host_initiated);
3580 	case MSR_IA32_BBL_CR_CTL3:
3581 		/* This legacy MSR exists but isn't fully documented in current
3582 		 * silicon.  It is however accessed by winxp in very narrow
3583 		 * scenarios where it sets bit #19, itself documented as
3584 		 * a "reserved" bit.  Best effort attempt to source coherent
3585 		 * read data here should the balance of the register be
3586 		 * interpreted by the guest:
3587 		 *
3588 		 * L2 cache control register 3: 64GB range, 256KB size,
3589 		 * enabled, latency 0x1, configured
3590 		 */
3591 		msr_info->data = 0xbe702111;
3592 		break;
3593 	case MSR_AMD64_OSVW_ID_LENGTH:
3594 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3595 			return 1;
3596 		msr_info->data = vcpu->arch.osvw.length;
3597 		break;
3598 	case MSR_AMD64_OSVW_STATUS:
3599 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3600 			return 1;
3601 		msr_info->data = vcpu->arch.osvw.status;
3602 		break;
3603 	case MSR_PLATFORM_INFO:
3604 		if (!msr_info->host_initiated &&
3605 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3606 			return 1;
3607 		msr_info->data = vcpu->arch.msr_platform_info;
3608 		break;
3609 	case MSR_MISC_FEATURES_ENABLES:
3610 		msr_info->data = vcpu->arch.msr_misc_features_enables;
3611 		break;
3612 	case MSR_K7_HWCR:
3613 		msr_info->data = vcpu->arch.msr_hwcr;
3614 		break;
3615 	default:
3616 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3617 			return kvm_pmu_get_msr(vcpu, msr_info);
3618 		return KVM_MSR_RET_INVALID;
3619 	}
3620 	return 0;
3621 }
3622 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3623 
3624 /*
3625  * Read or write a bunch of msrs. All parameters are kernel addresses.
3626  *
3627  * @return number of msrs set successfully.
3628  */
3629 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3630 		    struct kvm_msr_entry *entries,
3631 		    int (*do_msr)(struct kvm_vcpu *vcpu,
3632 				  unsigned index, u64 *data))
3633 {
3634 	int i;
3635 
3636 	for (i = 0; i < msrs->nmsrs; ++i)
3637 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
3638 			break;
3639 
3640 	return i;
3641 }
3642 
3643 /*
3644  * Read or write a bunch of msrs. Parameters are user addresses.
3645  *
3646  * @return number of msrs set successfully.
3647  */
3648 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3649 		  int (*do_msr)(struct kvm_vcpu *vcpu,
3650 				unsigned index, u64 *data),
3651 		  int writeback)
3652 {
3653 	struct kvm_msrs msrs;
3654 	struct kvm_msr_entry *entries;
3655 	int r, n;
3656 	unsigned size;
3657 
3658 	r = -EFAULT;
3659 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3660 		goto out;
3661 
3662 	r = -E2BIG;
3663 	if (msrs.nmsrs >= MAX_IO_MSRS)
3664 		goto out;
3665 
3666 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3667 	entries = memdup_user(user_msrs->entries, size);
3668 	if (IS_ERR(entries)) {
3669 		r = PTR_ERR(entries);
3670 		goto out;
3671 	}
3672 
3673 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3674 	if (r < 0)
3675 		goto out_free;
3676 
3677 	r = -EFAULT;
3678 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
3679 		goto out_free;
3680 
3681 	r = n;
3682 
3683 out_free:
3684 	kfree(entries);
3685 out:
3686 	return r;
3687 }
3688 
3689 static inline bool kvm_can_mwait_in_guest(void)
3690 {
3691 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
3692 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
3693 		boot_cpu_has(X86_FEATURE_ARAT);
3694 }
3695 
3696 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
3697 					    struct kvm_cpuid2 __user *cpuid_arg)
3698 {
3699 	struct kvm_cpuid2 cpuid;
3700 	int r;
3701 
3702 	r = -EFAULT;
3703 	if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3704 		return r;
3705 
3706 	r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3707 	if (r)
3708 		return r;
3709 
3710 	r = -EFAULT;
3711 	if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3712 		return r;
3713 
3714 	return 0;
3715 }
3716 
3717 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3718 {
3719 	int r = 0;
3720 
3721 	switch (ext) {
3722 	case KVM_CAP_IRQCHIP:
3723 	case KVM_CAP_HLT:
3724 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3725 	case KVM_CAP_SET_TSS_ADDR:
3726 	case KVM_CAP_EXT_CPUID:
3727 	case KVM_CAP_EXT_EMUL_CPUID:
3728 	case KVM_CAP_CLOCKSOURCE:
3729 	case KVM_CAP_PIT:
3730 	case KVM_CAP_NOP_IO_DELAY:
3731 	case KVM_CAP_MP_STATE:
3732 	case KVM_CAP_SYNC_MMU:
3733 	case KVM_CAP_USER_NMI:
3734 	case KVM_CAP_REINJECT_CONTROL:
3735 	case KVM_CAP_IRQ_INJECT_STATUS:
3736 	case KVM_CAP_IOEVENTFD:
3737 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
3738 	case KVM_CAP_PIT2:
3739 	case KVM_CAP_PIT_STATE2:
3740 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3741 	case KVM_CAP_XEN_HVM:
3742 	case KVM_CAP_VCPU_EVENTS:
3743 	case KVM_CAP_HYPERV:
3744 	case KVM_CAP_HYPERV_VAPIC:
3745 	case KVM_CAP_HYPERV_SPIN:
3746 	case KVM_CAP_HYPERV_SYNIC:
3747 	case KVM_CAP_HYPERV_SYNIC2:
3748 	case KVM_CAP_HYPERV_VP_INDEX:
3749 	case KVM_CAP_HYPERV_EVENTFD:
3750 	case KVM_CAP_HYPERV_TLBFLUSH:
3751 	case KVM_CAP_HYPERV_SEND_IPI:
3752 	case KVM_CAP_HYPERV_CPUID:
3753 	case KVM_CAP_SYS_HYPERV_CPUID:
3754 	case KVM_CAP_PCI_SEGMENT:
3755 	case KVM_CAP_DEBUGREGS:
3756 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
3757 	case KVM_CAP_XSAVE:
3758 	case KVM_CAP_ASYNC_PF:
3759 	case KVM_CAP_ASYNC_PF_INT:
3760 	case KVM_CAP_GET_TSC_KHZ:
3761 	case KVM_CAP_KVMCLOCK_CTRL:
3762 	case KVM_CAP_READONLY_MEM:
3763 	case KVM_CAP_HYPERV_TIME:
3764 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3765 	case KVM_CAP_TSC_DEADLINE_TIMER:
3766 	case KVM_CAP_DISABLE_QUIRKS:
3767 	case KVM_CAP_SET_BOOT_CPU_ID:
3768  	case KVM_CAP_SPLIT_IRQCHIP:
3769 	case KVM_CAP_IMMEDIATE_EXIT:
3770 	case KVM_CAP_PMU_EVENT_FILTER:
3771 	case KVM_CAP_GET_MSR_FEATURES:
3772 	case KVM_CAP_MSR_PLATFORM_INFO:
3773 	case KVM_CAP_EXCEPTION_PAYLOAD:
3774 	case KVM_CAP_SET_GUEST_DEBUG:
3775 	case KVM_CAP_LAST_CPU:
3776 	case KVM_CAP_X86_USER_SPACE_MSR:
3777 	case KVM_CAP_X86_MSR_FILTER:
3778 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
3779 		r = 1;
3780 		break;
3781 	case KVM_CAP_SYNC_REGS:
3782 		r = KVM_SYNC_X86_VALID_FIELDS;
3783 		break;
3784 	case KVM_CAP_ADJUST_CLOCK:
3785 		r = KVM_CLOCK_TSC_STABLE;
3786 		break;
3787 	case KVM_CAP_X86_DISABLE_EXITS:
3788 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3789 		      KVM_X86_DISABLE_EXITS_CSTATE;
3790 		if(kvm_can_mwait_in_guest())
3791 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
3792 		break;
3793 	case KVM_CAP_X86_SMM:
3794 		/* SMBASE is usually relocated above 1M on modern chipsets,
3795 		 * and SMM handlers might indeed rely on 4G segment limits,
3796 		 * so do not report SMM to be available if real mode is
3797 		 * emulated via vm86 mode.  Still, do not go to great lengths
3798 		 * to avoid userspace's usage of the feature, because it is a
3799 		 * fringe case that is not enabled except via specific settings
3800 		 * of the module parameters.
3801 		 */
3802 		r = kvm_x86_ops.has_emulated_msr(kvm, MSR_IA32_SMBASE);
3803 		break;
3804 	case KVM_CAP_VAPIC:
3805 		r = !kvm_x86_ops.cpu_has_accelerated_tpr();
3806 		break;
3807 	case KVM_CAP_NR_VCPUS:
3808 		r = KVM_SOFT_MAX_VCPUS;
3809 		break;
3810 	case KVM_CAP_MAX_VCPUS:
3811 		r = KVM_MAX_VCPUS;
3812 		break;
3813 	case KVM_CAP_MAX_VCPU_ID:
3814 		r = KVM_MAX_VCPU_ID;
3815 		break;
3816 	case KVM_CAP_PV_MMU:	/* obsolete */
3817 		r = 0;
3818 		break;
3819 	case KVM_CAP_MCE:
3820 		r = KVM_MAX_MCE_BANKS;
3821 		break;
3822 	case KVM_CAP_XCRS:
3823 		r = boot_cpu_has(X86_FEATURE_XSAVE);
3824 		break;
3825 	case KVM_CAP_TSC_CONTROL:
3826 		r = kvm_has_tsc_control;
3827 		break;
3828 	case KVM_CAP_X2APIC_API:
3829 		r = KVM_X2APIC_API_VALID_FLAGS;
3830 		break;
3831 	case KVM_CAP_NESTED_STATE:
3832 		r = kvm_x86_ops.nested_ops->get_state ?
3833 			kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
3834 		break;
3835 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3836 		r = kvm_x86_ops.enable_direct_tlbflush != NULL;
3837 		break;
3838 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3839 		r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
3840 		break;
3841 	case KVM_CAP_SMALLER_MAXPHYADDR:
3842 		r = (int) allow_smaller_maxphyaddr;
3843 		break;
3844 	case KVM_CAP_STEAL_TIME:
3845 		r = sched_info_on();
3846 		break;
3847 	default:
3848 		break;
3849 	}
3850 	return r;
3851 
3852 }
3853 
3854 long kvm_arch_dev_ioctl(struct file *filp,
3855 			unsigned int ioctl, unsigned long arg)
3856 {
3857 	void __user *argp = (void __user *)arg;
3858 	long r;
3859 
3860 	switch (ioctl) {
3861 	case KVM_GET_MSR_INDEX_LIST: {
3862 		struct kvm_msr_list __user *user_msr_list = argp;
3863 		struct kvm_msr_list msr_list;
3864 		unsigned n;
3865 
3866 		r = -EFAULT;
3867 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3868 			goto out;
3869 		n = msr_list.nmsrs;
3870 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3871 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3872 			goto out;
3873 		r = -E2BIG;
3874 		if (n < msr_list.nmsrs)
3875 			goto out;
3876 		r = -EFAULT;
3877 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3878 				 num_msrs_to_save * sizeof(u32)))
3879 			goto out;
3880 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3881 				 &emulated_msrs,
3882 				 num_emulated_msrs * sizeof(u32)))
3883 			goto out;
3884 		r = 0;
3885 		break;
3886 	}
3887 	case KVM_GET_SUPPORTED_CPUID:
3888 	case KVM_GET_EMULATED_CPUID: {
3889 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3890 		struct kvm_cpuid2 cpuid;
3891 
3892 		r = -EFAULT;
3893 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3894 			goto out;
3895 
3896 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3897 					    ioctl);
3898 		if (r)
3899 			goto out;
3900 
3901 		r = -EFAULT;
3902 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3903 			goto out;
3904 		r = 0;
3905 		break;
3906 	}
3907 	case KVM_X86_GET_MCE_CAP_SUPPORTED:
3908 		r = -EFAULT;
3909 		if (copy_to_user(argp, &kvm_mce_cap_supported,
3910 				 sizeof(kvm_mce_cap_supported)))
3911 			goto out;
3912 		r = 0;
3913 		break;
3914 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3915 		struct kvm_msr_list __user *user_msr_list = argp;
3916 		struct kvm_msr_list msr_list;
3917 		unsigned int n;
3918 
3919 		r = -EFAULT;
3920 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3921 			goto out;
3922 		n = msr_list.nmsrs;
3923 		msr_list.nmsrs = num_msr_based_features;
3924 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3925 			goto out;
3926 		r = -E2BIG;
3927 		if (n < msr_list.nmsrs)
3928 			goto out;
3929 		r = -EFAULT;
3930 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
3931 				 num_msr_based_features * sizeof(u32)))
3932 			goto out;
3933 		r = 0;
3934 		break;
3935 	}
3936 	case KVM_GET_MSRS:
3937 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
3938 		break;
3939 	case KVM_GET_SUPPORTED_HV_CPUID:
3940 		r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
3941 		break;
3942 	default:
3943 		r = -EINVAL;
3944 		break;
3945 	}
3946 out:
3947 	return r;
3948 }
3949 
3950 static void wbinvd_ipi(void *garbage)
3951 {
3952 	wbinvd();
3953 }
3954 
3955 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3956 {
3957 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3958 }
3959 
3960 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3961 {
3962 	/* Address WBINVD may be executed by guest */
3963 	if (need_emulate_wbinvd(vcpu)) {
3964 		if (kvm_x86_ops.has_wbinvd_exit())
3965 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3966 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3967 			smp_call_function_single(vcpu->cpu,
3968 					wbinvd_ipi, NULL, 1);
3969 	}
3970 
3971 	kvm_x86_ops.vcpu_load(vcpu, cpu);
3972 
3973 	/* Save host pkru register if supported */
3974 	vcpu->arch.host_pkru = read_pkru();
3975 
3976 	/* Apply any externally detected TSC adjustments (due to suspend) */
3977 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3978 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3979 		vcpu->arch.tsc_offset_adjustment = 0;
3980 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3981 	}
3982 
3983 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3984 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3985 				rdtsc() - vcpu->arch.last_host_tsc;
3986 		if (tsc_delta < 0)
3987 			mark_tsc_unstable("KVM discovered backwards TSC");
3988 
3989 		if (kvm_check_tsc_unstable()) {
3990 			u64 offset = kvm_compute_tsc_offset(vcpu,
3991 						vcpu->arch.last_guest_tsc);
3992 			kvm_vcpu_write_tsc_offset(vcpu, offset);
3993 			vcpu->arch.tsc_catchup = 1;
3994 		}
3995 
3996 		if (kvm_lapic_hv_timer_in_use(vcpu))
3997 			kvm_lapic_restart_hv_timer(vcpu);
3998 
3999 		/*
4000 		 * On a host with synchronized TSC, there is no need to update
4001 		 * kvmclock on vcpu->cpu migration
4002 		 */
4003 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4004 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4005 		if (vcpu->cpu != cpu)
4006 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4007 		vcpu->cpu = cpu;
4008 	}
4009 
4010 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4011 }
4012 
4013 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4014 {
4015 	struct kvm_host_map map;
4016 	struct kvm_steal_time *st;
4017 
4018 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4019 		return;
4020 
4021 	if (vcpu->arch.st.preempted)
4022 		return;
4023 
4024 	if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
4025 			&vcpu->arch.st.cache, true))
4026 		return;
4027 
4028 	st = map.hva +
4029 		offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
4030 
4031 	st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4032 
4033 	kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
4034 }
4035 
4036 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4037 {
4038 	int idx;
4039 
4040 	if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4041 		vcpu->arch.preempted_in_kernel = !kvm_x86_ops.get_cpl(vcpu);
4042 
4043 	/*
4044 	 * Disable page faults because we're in atomic context here.
4045 	 * kvm_write_guest_offset_cached() would call might_fault()
4046 	 * that relies on pagefault_disable() to tell if there's a
4047 	 * bug. NOTE: the write to guest memory may not go through if
4048 	 * during postcopy live migration or if there's heavy guest
4049 	 * paging.
4050 	 */
4051 	pagefault_disable();
4052 	/*
4053 	 * kvm_memslots() will be called by
4054 	 * kvm_write_guest_offset_cached() so take the srcu lock.
4055 	 */
4056 	idx = srcu_read_lock(&vcpu->kvm->srcu);
4057 	kvm_steal_time_set_preempted(vcpu);
4058 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
4059 	pagefault_enable();
4060 	kvm_x86_ops.vcpu_put(vcpu);
4061 	vcpu->arch.last_host_tsc = rdtsc();
4062 	/*
4063 	 * If userspace has set any breakpoints or watchpoints, dr6 is restored
4064 	 * on every vmexit, but if not, we might have a stale dr6 from the
4065 	 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
4066 	 */
4067 	set_debugreg(0, 6);
4068 }
4069 
4070 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4071 				    struct kvm_lapic_state *s)
4072 {
4073 	if (vcpu->arch.apicv_active)
4074 		kvm_x86_ops.sync_pir_to_irr(vcpu);
4075 
4076 	return kvm_apic_get_state(vcpu, s);
4077 }
4078 
4079 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4080 				    struct kvm_lapic_state *s)
4081 {
4082 	int r;
4083 
4084 	r = kvm_apic_set_state(vcpu, s);
4085 	if (r)
4086 		return r;
4087 	update_cr8_intercept(vcpu);
4088 
4089 	return 0;
4090 }
4091 
4092 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4093 {
4094 	/*
4095 	 * We can accept userspace's request for interrupt injection
4096 	 * as long as we have a place to store the interrupt number.
4097 	 * The actual injection will happen when the CPU is able to
4098 	 * deliver the interrupt.
4099 	 */
4100 	if (kvm_cpu_has_extint(vcpu))
4101 		return false;
4102 
4103 	/* Acknowledging ExtINT does not happen if LINT0 is masked.  */
4104 	return (!lapic_in_kernel(vcpu) ||
4105 		kvm_apic_accept_pic_intr(vcpu));
4106 }
4107 
4108 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4109 {
4110 	return kvm_arch_interrupt_allowed(vcpu) &&
4111 		kvm_cpu_accept_dm_intr(vcpu);
4112 }
4113 
4114 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4115 				    struct kvm_interrupt *irq)
4116 {
4117 	if (irq->irq >= KVM_NR_INTERRUPTS)
4118 		return -EINVAL;
4119 
4120 	if (!irqchip_in_kernel(vcpu->kvm)) {
4121 		kvm_queue_interrupt(vcpu, irq->irq, false);
4122 		kvm_make_request(KVM_REQ_EVENT, vcpu);
4123 		return 0;
4124 	}
4125 
4126 	/*
4127 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4128 	 * fail for in-kernel 8259.
4129 	 */
4130 	if (pic_in_kernel(vcpu->kvm))
4131 		return -ENXIO;
4132 
4133 	if (vcpu->arch.pending_external_vector != -1)
4134 		return -EEXIST;
4135 
4136 	vcpu->arch.pending_external_vector = irq->irq;
4137 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4138 	return 0;
4139 }
4140 
4141 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4142 {
4143 	kvm_inject_nmi(vcpu);
4144 
4145 	return 0;
4146 }
4147 
4148 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4149 {
4150 	kvm_make_request(KVM_REQ_SMI, vcpu);
4151 
4152 	return 0;
4153 }
4154 
4155 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4156 					   struct kvm_tpr_access_ctl *tac)
4157 {
4158 	if (tac->flags)
4159 		return -EINVAL;
4160 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
4161 	return 0;
4162 }
4163 
4164 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4165 					u64 mcg_cap)
4166 {
4167 	int r;
4168 	unsigned bank_num = mcg_cap & 0xff, bank;
4169 
4170 	r = -EINVAL;
4171 	if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4172 		goto out;
4173 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4174 		goto out;
4175 	r = 0;
4176 	vcpu->arch.mcg_cap = mcg_cap;
4177 	/* Init IA32_MCG_CTL to all 1s */
4178 	if (mcg_cap & MCG_CTL_P)
4179 		vcpu->arch.mcg_ctl = ~(u64)0;
4180 	/* Init IA32_MCi_CTL to all 1s */
4181 	for (bank = 0; bank < bank_num; bank++)
4182 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4183 
4184 	kvm_x86_ops.setup_mce(vcpu);
4185 out:
4186 	return r;
4187 }
4188 
4189 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4190 				      struct kvm_x86_mce *mce)
4191 {
4192 	u64 mcg_cap = vcpu->arch.mcg_cap;
4193 	unsigned bank_num = mcg_cap & 0xff;
4194 	u64 *banks = vcpu->arch.mce_banks;
4195 
4196 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4197 		return -EINVAL;
4198 	/*
4199 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4200 	 * reporting is disabled
4201 	 */
4202 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4203 	    vcpu->arch.mcg_ctl != ~(u64)0)
4204 		return 0;
4205 	banks += 4 * mce->bank;
4206 	/*
4207 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4208 	 * reporting is disabled for the bank
4209 	 */
4210 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4211 		return 0;
4212 	if (mce->status & MCI_STATUS_UC) {
4213 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4214 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4215 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4216 			return 0;
4217 		}
4218 		if (banks[1] & MCI_STATUS_VAL)
4219 			mce->status |= MCI_STATUS_OVER;
4220 		banks[2] = mce->addr;
4221 		banks[3] = mce->misc;
4222 		vcpu->arch.mcg_status = mce->mcg_status;
4223 		banks[1] = mce->status;
4224 		kvm_queue_exception(vcpu, MC_VECTOR);
4225 	} else if (!(banks[1] & MCI_STATUS_VAL)
4226 		   || !(banks[1] & MCI_STATUS_UC)) {
4227 		if (banks[1] & MCI_STATUS_VAL)
4228 			mce->status |= MCI_STATUS_OVER;
4229 		banks[2] = mce->addr;
4230 		banks[3] = mce->misc;
4231 		banks[1] = mce->status;
4232 	} else
4233 		banks[1] |= MCI_STATUS_OVER;
4234 	return 0;
4235 }
4236 
4237 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4238 					       struct kvm_vcpu_events *events)
4239 {
4240 	process_nmi(vcpu);
4241 
4242 	if (kvm_check_request(KVM_REQ_SMI, vcpu))
4243 		process_smi(vcpu);
4244 
4245 	/*
4246 	 * In guest mode, payload delivery should be deferred,
4247 	 * so that the L1 hypervisor can intercept #PF before
4248 	 * CR2 is modified (or intercept #DB before DR6 is
4249 	 * modified under nVMX). Unless the per-VM capability,
4250 	 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4251 	 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4252 	 * opportunistically defer the exception payload, deliver it if the
4253 	 * capability hasn't been requested before processing a
4254 	 * KVM_GET_VCPU_EVENTS.
4255 	 */
4256 	if (!vcpu->kvm->arch.exception_payload_enabled &&
4257 	    vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4258 		kvm_deliver_exception_payload(vcpu);
4259 
4260 	/*
4261 	 * The API doesn't provide the instruction length for software
4262 	 * exceptions, so don't report them. As long as the guest RIP
4263 	 * isn't advanced, we should expect to encounter the exception
4264 	 * again.
4265 	 */
4266 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4267 		events->exception.injected = 0;
4268 		events->exception.pending = 0;
4269 	} else {
4270 		events->exception.injected = vcpu->arch.exception.injected;
4271 		events->exception.pending = vcpu->arch.exception.pending;
4272 		/*
4273 		 * For ABI compatibility, deliberately conflate
4274 		 * pending and injected exceptions when
4275 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4276 		 */
4277 		if (!vcpu->kvm->arch.exception_payload_enabled)
4278 			events->exception.injected |=
4279 				vcpu->arch.exception.pending;
4280 	}
4281 	events->exception.nr = vcpu->arch.exception.nr;
4282 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4283 	events->exception.error_code = vcpu->arch.exception.error_code;
4284 	events->exception_has_payload = vcpu->arch.exception.has_payload;
4285 	events->exception_payload = vcpu->arch.exception.payload;
4286 
4287 	events->interrupt.injected =
4288 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4289 	events->interrupt.nr = vcpu->arch.interrupt.nr;
4290 	events->interrupt.soft = 0;
4291 	events->interrupt.shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
4292 
4293 	events->nmi.injected = vcpu->arch.nmi_injected;
4294 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
4295 	events->nmi.masked = kvm_x86_ops.get_nmi_mask(vcpu);
4296 	events->nmi.pad = 0;
4297 
4298 	events->sipi_vector = 0; /* never valid when reporting to user space */
4299 
4300 	events->smi.smm = is_smm(vcpu);
4301 	events->smi.pending = vcpu->arch.smi_pending;
4302 	events->smi.smm_inside_nmi =
4303 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4304 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4305 
4306 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4307 			 | KVM_VCPUEVENT_VALID_SHADOW
4308 			 | KVM_VCPUEVENT_VALID_SMM);
4309 	if (vcpu->kvm->arch.exception_payload_enabled)
4310 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4311 
4312 	memset(&events->reserved, 0, sizeof(events->reserved));
4313 }
4314 
4315 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
4316 
4317 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4318 					      struct kvm_vcpu_events *events)
4319 {
4320 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4321 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4322 			      | KVM_VCPUEVENT_VALID_SHADOW
4323 			      | KVM_VCPUEVENT_VALID_SMM
4324 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
4325 		return -EINVAL;
4326 
4327 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4328 		if (!vcpu->kvm->arch.exception_payload_enabled)
4329 			return -EINVAL;
4330 		if (events->exception.pending)
4331 			events->exception.injected = 0;
4332 		else
4333 			events->exception_has_payload = 0;
4334 	} else {
4335 		events->exception.pending = 0;
4336 		events->exception_has_payload = 0;
4337 	}
4338 
4339 	if ((events->exception.injected || events->exception.pending) &&
4340 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4341 		return -EINVAL;
4342 
4343 	/* INITs are latched while in SMM */
4344 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4345 	    (events->smi.smm || events->smi.pending) &&
4346 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4347 		return -EINVAL;
4348 
4349 	process_nmi(vcpu);
4350 	vcpu->arch.exception.injected = events->exception.injected;
4351 	vcpu->arch.exception.pending = events->exception.pending;
4352 	vcpu->arch.exception.nr = events->exception.nr;
4353 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4354 	vcpu->arch.exception.error_code = events->exception.error_code;
4355 	vcpu->arch.exception.has_payload = events->exception_has_payload;
4356 	vcpu->arch.exception.payload = events->exception_payload;
4357 
4358 	vcpu->arch.interrupt.injected = events->interrupt.injected;
4359 	vcpu->arch.interrupt.nr = events->interrupt.nr;
4360 	vcpu->arch.interrupt.soft = events->interrupt.soft;
4361 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4362 		kvm_x86_ops.set_interrupt_shadow(vcpu,
4363 						  events->interrupt.shadow);
4364 
4365 	vcpu->arch.nmi_injected = events->nmi.injected;
4366 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4367 		vcpu->arch.nmi_pending = events->nmi.pending;
4368 	kvm_x86_ops.set_nmi_mask(vcpu, events->nmi.masked);
4369 
4370 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4371 	    lapic_in_kernel(vcpu))
4372 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
4373 
4374 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4375 		if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
4376 			if (events->smi.smm)
4377 				vcpu->arch.hflags |= HF_SMM_MASK;
4378 			else
4379 				vcpu->arch.hflags &= ~HF_SMM_MASK;
4380 			kvm_smm_changed(vcpu);
4381 		}
4382 
4383 		vcpu->arch.smi_pending = events->smi.pending;
4384 
4385 		if (events->smi.smm) {
4386 			if (events->smi.smm_inside_nmi)
4387 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4388 			else
4389 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4390 		}
4391 
4392 		if (lapic_in_kernel(vcpu)) {
4393 			if (events->smi.latched_init)
4394 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4395 			else
4396 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4397 		}
4398 	}
4399 
4400 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4401 
4402 	return 0;
4403 }
4404 
4405 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4406 					     struct kvm_debugregs *dbgregs)
4407 {
4408 	unsigned long val;
4409 
4410 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4411 	kvm_get_dr(vcpu, 6, &val);
4412 	dbgregs->dr6 = val;
4413 	dbgregs->dr7 = vcpu->arch.dr7;
4414 	dbgregs->flags = 0;
4415 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4416 }
4417 
4418 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4419 					    struct kvm_debugregs *dbgregs)
4420 {
4421 	if (dbgregs->flags)
4422 		return -EINVAL;
4423 
4424 	if (dbgregs->dr6 & ~0xffffffffull)
4425 		return -EINVAL;
4426 	if (dbgregs->dr7 & ~0xffffffffull)
4427 		return -EINVAL;
4428 
4429 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4430 	kvm_update_dr0123(vcpu);
4431 	vcpu->arch.dr6 = dbgregs->dr6;
4432 	vcpu->arch.dr7 = dbgregs->dr7;
4433 	kvm_update_dr7(vcpu);
4434 
4435 	return 0;
4436 }
4437 
4438 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4439 
4440 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4441 {
4442 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4443 	u64 xstate_bv = xsave->header.xfeatures;
4444 	u64 valid;
4445 
4446 	/*
4447 	 * Copy legacy XSAVE area, to avoid complications with CPUID
4448 	 * leaves 0 and 1 in the loop below.
4449 	 */
4450 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4451 
4452 	/* Set XSTATE_BV */
4453 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4454 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4455 
4456 	/*
4457 	 * Copy each region from the possibly compacted offset to the
4458 	 * non-compacted offset.
4459 	 */
4460 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4461 	while (valid) {
4462 		u64 xfeature_mask = valid & -valid;
4463 		int xfeature_nr = fls64(xfeature_mask) - 1;
4464 		void *src = get_xsave_addr(xsave, xfeature_nr);
4465 
4466 		if (src) {
4467 			u32 size, offset, ecx, edx;
4468 			cpuid_count(XSTATE_CPUID, xfeature_nr,
4469 				    &size, &offset, &ecx, &edx);
4470 			if (xfeature_nr == XFEATURE_PKRU)
4471 				memcpy(dest + offset, &vcpu->arch.pkru,
4472 				       sizeof(vcpu->arch.pkru));
4473 			else
4474 				memcpy(dest + offset, src, size);
4475 
4476 		}
4477 
4478 		valid -= xfeature_mask;
4479 	}
4480 }
4481 
4482 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4483 {
4484 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4485 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4486 	u64 valid;
4487 
4488 	/*
4489 	 * Copy legacy XSAVE area, to avoid complications with CPUID
4490 	 * leaves 0 and 1 in the loop below.
4491 	 */
4492 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
4493 
4494 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
4495 	xsave->header.xfeatures = xstate_bv;
4496 	if (boot_cpu_has(X86_FEATURE_XSAVES))
4497 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4498 
4499 	/*
4500 	 * Copy each region from the non-compacted offset to the
4501 	 * possibly compacted offset.
4502 	 */
4503 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4504 	while (valid) {
4505 		u64 xfeature_mask = valid & -valid;
4506 		int xfeature_nr = fls64(xfeature_mask) - 1;
4507 		void *dest = get_xsave_addr(xsave, xfeature_nr);
4508 
4509 		if (dest) {
4510 			u32 size, offset, ecx, edx;
4511 			cpuid_count(XSTATE_CPUID, xfeature_nr,
4512 				    &size, &offset, &ecx, &edx);
4513 			if (xfeature_nr == XFEATURE_PKRU)
4514 				memcpy(&vcpu->arch.pkru, src + offset,
4515 				       sizeof(vcpu->arch.pkru));
4516 			else
4517 				memcpy(dest, src + offset, size);
4518 		}
4519 
4520 		valid -= xfeature_mask;
4521 	}
4522 }
4523 
4524 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4525 					 struct kvm_xsave *guest_xsave)
4526 {
4527 	if (!vcpu->arch.guest_fpu)
4528 		return;
4529 
4530 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4531 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4532 		fill_xsave((u8 *) guest_xsave->region, vcpu);
4533 	} else {
4534 		memcpy(guest_xsave->region,
4535 			&vcpu->arch.guest_fpu->state.fxsave,
4536 			sizeof(struct fxregs_state));
4537 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4538 			XFEATURE_MASK_FPSSE;
4539 	}
4540 }
4541 
4542 #define XSAVE_MXCSR_OFFSET 24
4543 
4544 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4545 					struct kvm_xsave *guest_xsave)
4546 {
4547 	u64 xstate_bv;
4548 	u32 mxcsr;
4549 
4550 	if (!vcpu->arch.guest_fpu)
4551 		return 0;
4552 
4553 	xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4554 	mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4555 
4556 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4557 		/*
4558 		 * Here we allow setting states that are not present in
4559 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
4560 		 * with old userspace.
4561 		 */
4562 		if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4563 			return -EINVAL;
4564 		load_xsave(vcpu, (u8 *)guest_xsave->region);
4565 	} else {
4566 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4567 			mxcsr & ~mxcsr_feature_mask)
4568 			return -EINVAL;
4569 		memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4570 			guest_xsave->region, sizeof(struct fxregs_state));
4571 	}
4572 	return 0;
4573 }
4574 
4575 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4576 					struct kvm_xcrs *guest_xcrs)
4577 {
4578 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4579 		guest_xcrs->nr_xcrs = 0;
4580 		return;
4581 	}
4582 
4583 	guest_xcrs->nr_xcrs = 1;
4584 	guest_xcrs->flags = 0;
4585 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4586 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4587 }
4588 
4589 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4590 				       struct kvm_xcrs *guest_xcrs)
4591 {
4592 	int i, r = 0;
4593 
4594 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
4595 		return -EINVAL;
4596 
4597 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4598 		return -EINVAL;
4599 
4600 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4601 		/* Only support XCR0 currently */
4602 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4603 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4604 				guest_xcrs->xcrs[i].value);
4605 			break;
4606 		}
4607 	if (r)
4608 		r = -EINVAL;
4609 	return r;
4610 }
4611 
4612 /*
4613  * kvm_set_guest_paused() indicates to the guest kernel that it has been
4614  * stopped by the hypervisor.  This function will be called from the host only.
4615  * EINVAL is returned when the host attempts to set the flag for a guest that
4616  * does not support pv clocks.
4617  */
4618 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4619 {
4620 	if (!vcpu->arch.pv_time_enabled)
4621 		return -EINVAL;
4622 	vcpu->arch.pvclock_set_guest_stopped_request = true;
4623 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4624 	return 0;
4625 }
4626 
4627 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4628 				     struct kvm_enable_cap *cap)
4629 {
4630 	int r;
4631 	uint16_t vmcs_version;
4632 	void __user *user_ptr;
4633 
4634 	if (cap->flags)
4635 		return -EINVAL;
4636 
4637 	switch (cap->cap) {
4638 	case KVM_CAP_HYPERV_SYNIC2:
4639 		if (cap->args[0])
4640 			return -EINVAL;
4641 		fallthrough;
4642 
4643 	case KVM_CAP_HYPERV_SYNIC:
4644 		if (!irqchip_in_kernel(vcpu->kvm))
4645 			return -EINVAL;
4646 		return kvm_hv_activate_synic(vcpu, cap->cap ==
4647 					     KVM_CAP_HYPERV_SYNIC2);
4648 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4649 		if (!kvm_x86_ops.nested_ops->enable_evmcs)
4650 			return -ENOTTY;
4651 		r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4652 		if (!r) {
4653 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
4654 			if (copy_to_user(user_ptr, &vmcs_version,
4655 					 sizeof(vmcs_version)))
4656 				r = -EFAULT;
4657 		}
4658 		return r;
4659 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4660 		if (!kvm_x86_ops.enable_direct_tlbflush)
4661 			return -ENOTTY;
4662 
4663 		return kvm_x86_ops.enable_direct_tlbflush(vcpu);
4664 
4665 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4666 		vcpu->arch.pv_cpuid.enforce = cap->args[0];
4667 		if (vcpu->arch.pv_cpuid.enforce)
4668 			kvm_update_pv_runtime(vcpu);
4669 
4670 		return 0;
4671 
4672 	default:
4673 		return -EINVAL;
4674 	}
4675 }
4676 
4677 long kvm_arch_vcpu_ioctl(struct file *filp,
4678 			 unsigned int ioctl, unsigned long arg)
4679 {
4680 	struct kvm_vcpu *vcpu = filp->private_data;
4681 	void __user *argp = (void __user *)arg;
4682 	int r;
4683 	union {
4684 		struct kvm_lapic_state *lapic;
4685 		struct kvm_xsave *xsave;
4686 		struct kvm_xcrs *xcrs;
4687 		void *buffer;
4688 	} u;
4689 
4690 	vcpu_load(vcpu);
4691 
4692 	u.buffer = NULL;
4693 	switch (ioctl) {
4694 	case KVM_GET_LAPIC: {
4695 		r = -EINVAL;
4696 		if (!lapic_in_kernel(vcpu))
4697 			goto out;
4698 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4699 				GFP_KERNEL_ACCOUNT);
4700 
4701 		r = -ENOMEM;
4702 		if (!u.lapic)
4703 			goto out;
4704 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4705 		if (r)
4706 			goto out;
4707 		r = -EFAULT;
4708 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4709 			goto out;
4710 		r = 0;
4711 		break;
4712 	}
4713 	case KVM_SET_LAPIC: {
4714 		r = -EINVAL;
4715 		if (!lapic_in_kernel(vcpu))
4716 			goto out;
4717 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
4718 		if (IS_ERR(u.lapic)) {
4719 			r = PTR_ERR(u.lapic);
4720 			goto out_nofree;
4721 		}
4722 
4723 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4724 		break;
4725 	}
4726 	case KVM_INTERRUPT: {
4727 		struct kvm_interrupt irq;
4728 
4729 		r = -EFAULT;
4730 		if (copy_from_user(&irq, argp, sizeof(irq)))
4731 			goto out;
4732 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4733 		break;
4734 	}
4735 	case KVM_NMI: {
4736 		r = kvm_vcpu_ioctl_nmi(vcpu);
4737 		break;
4738 	}
4739 	case KVM_SMI: {
4740 		r = kvm_vcpu_ioctl_smi(vcpu);
4741 		break;
4742 	}
4743 	case KVM_SET_CPUID: {
4744 		struct kvm_cpuid __user *cpuid_arg = argp;
4745 		struct kvm_cpuid cpuid;
4746 
4747 		r = -EFAULT;
4748 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4749 			goto out;
4750 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4751 		break;
4752 	}
4753 	case KVM_SET_CPUID2: {
4754 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4755 		struct kvm_cpuid2 cpuid;
4756 
4757 		r = -EFAULT;
4758 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4759 			goto out;
4760 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4761 					      cpuid_arg->entries);
4762 		break;
4763 	}
4764 	case KVM_GET_CPUID2: {
4765 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4766 		struct kvm_cpuid2 cpuid;
4767 
4768 		r = -EFAULT;
4769 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4770 			goto out;
4771 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4772 					      cpuid_arg->entries);
4773 		if (r)
4774 			goto out;
4775 		r = -EFAULT;
4776 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4777 			goto out;
4778 		r = 0;
4779 		break;
4780 	}
4781 	case KVM_GET_MSRS: {
4782 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4783 		r = msr_io(vcpu, argp, do_get_msr, 1);
4784 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4785 		break;
4786 	}
4787 	case KVM_SET_MSRS: {
4788 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4789 		r = msr_io(vcpu, argp, do_set_msr, 0);
4790 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4791 		break;
4792 	}
4793 	case KVM_TPR_ACCESS_REPORTING: {
4794 		struct kvm_tpr_access_ctl tac;
4795 
4796 		r = -EFAULT;
4797 		if (copy_from_user(&tac, argp, sizeof(tac)))
4798 			goto out;
4799 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4800 		if (r)
4801 			goto out;
4802 		r = -EFAULT;
4803 		if (copy_to_user(argp, &tac, sizeof(tac)))
4804 			goto out;
4805 		r = 0;
4806 		break;
4807 	};
4808 	case KVM_SET_VAPIC_ADDR: {
4809 		struct kvm_vapic_addr va;
4810 		int idx;
4811 
4812 		r = -EINVAL;
4813 		if (!lapic_in_kernel(vcpu))
4814 			goto out;
4815 		r = -EFAULT;
4816 		if (copy_from_user(&va, argp, sizeof(va)))
4817 			goto out;
4818 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4819 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4820 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4821 		break;
4822 	}
4823 	case KVM_X86_SETUP_MCE: {
4824 		u64 mcg_cap;
4825 
4826 		r = -EFAULT;
4827 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4828 			goto out;
4829 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4830 		break;
4831 	}
4832 	case KVM_X86_SET_MCE: {
4833 		struct kvm_x86_mce mce;
4834 
4835 		r = -EFAULT;
4836 		if (copy_from_user(&mce, argp, sizeof(mce)))
4837 			goto out;
4838 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4839 		break;
4840 	}
4841 	case KVM_GET_VCPU_EVENTS: {
4842 		struct kvm_vcpu_events events;
4843 
4844 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4845 
4846 		r = -EFAULT;
4847 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4848 			break;
4849 		r = 0;
4850 		break;
4851 	}
4852 	case KVM_SET_VCPU_EVENTS: {
4853 		struct kvm_vcpu_events events;
4854 
4855 		r = -EFAULT;
4856 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4857 			break;
4858 
4859 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4860 		break;
4861 	}
4862 	case KVM_GET_DEBUGREGS: {
4863 		struct kvm_debugregs dbgregs;
4864 
4865 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4866 
4867 		r = -EFAULT;
4868 		if (copy_to_user(argp, &dbgregs,
4869 				 sizeof(struct kvm_debugregs)))
4870 			break;
4871 		r = 0;
4872 		break;
4873 	}
4874 	case KVM_SET_DEBUGREGS: {
4875 		struct kvm_debugregs dbgregs;
4876 
4877 		r = -EFAULT;
4878 		if (copy_from_user(&dbgregs, argp,
4879 				   sizeof(struct kvm_debugregs)))
4880 			break;
4881 
4882 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4883 		break;
4884 	}
4885 	case KVM_GET_XSAVE: {
4886 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4887 		r = -ENOMEM;
4888 		if (!u.xsave)
4889 			break;
4890 
4891 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4892 
4893 		r = -EFAULT;
4894 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4895 			break;
4896 		r = 0;
4897 		break;
4898 	}
4899 	case KVM_SET_XSAVE: {
4900 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
4901 		if (IS_ERR(u.xsave)) {
4902 			r = PTR_ERR(u.xsave);
4903 			goto out_nofree;
4904 		}
4905 
4906 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4907 		break;
4908 	}
4909 	case KVM_GET_XCRS: {
4910 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4911 		r = -ENOMEM;
4912 		if (!u.xcrs)
4913 			break;
4914 
4915 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4916 
4917 		r = -EFAULT;
4918 		if (copy_to_user(argp, u.xcrs,
4919 				 sizeof(struct kvm_xcrs)))
4920 			break;
4921 		r = 0;
4922 		break;
4923 	}
4924 	case KVM_SET_XCRS: {
4925 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4926 		if (IS_ERR(u.xcrs)) {
4927 			r = PTR_ERR(u.xcrs);
4928 			goto out_nofree;
4929 		}
4930 
4931 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4932 		break;
4933 	}
4934 	case KVM_SET_TSC_KHZ: {
4935 		u32 user_tsc_khz;
4936 
4937 		r = -EINVAL;
4938 		user_tsc_khz = (u32)arg;
4939 
4940 		if (kvm_has_tsc_control &&
4941 		    user_tsc_khz >= kvm_max_guest_tsc_khz)
4942 			goto out;
4943 
4944 		if (user_tsc_khz == 0)
4945 			user_tsc_khz = tsc_khz;
4946 
4947 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4948 			r = 0;
4949 
4950 		goto out;
4951 	}
4952 	case KVM_GET_TSC_KHZ: {
4953 		r = vcpu->arch.virtual_tsc_khz;
4954 		goto out;
4955 	}
4956 	case KVM_KVMCLOCK_CTRL: {
4957 		r = kvm_set_guest_paused(vcpu);
4958 		goto out;
4959 	}
4960 	case KVM_ENABLE_CAP: {
4961 		struct kvm_enable_cap cap;
4962 
4963 		r = -EFAULT;
4964 		if (copy_from_user(&cap, argp, sizeof(cap)))
4965 			goto out;
4966 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4967 		break;
4968 	}
4969 	case KVM_GET_NESTED_STATE: {
4970 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4971 		u32 user_data_size;
4972 
4973 		r = -EINVAL;
4974 		if (!kvm_x86_ops.nested_ops->get_state)
4975 			break;
4976 
4977 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4978 		r = -EFAULT;
4979 		if (get_user(user_data_size, &user_kvm_nested_state->size))
4980 			break;
4981 
4982 		r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
4983 						     user_data_size);
4984 		if (r < 0)
4985 			break;
4986 
4987 		if (r > user_data_size) {
4988 			if (put_user(r, &user_kvm_nested_state->size))
4989 				r = -EFAULT;
4990 			else
4991 				r = -E2BIG;
4992 			break;
4993 		}
4994 
4995 		r = 0;
4996 		break;
4997 	}
4998 	case KVM_SET_NESTED_STATE: {
4999 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5000 		struct kvm_nested_state kvm_state;
5001 		int idx;
5002 
5003 		r = -EINVAL;
5004 		if (!kvm_x86_ops.nested_ops->set_state)
5005 			break;
5006 
5007 		r = -EFAULT;
5008 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5009 			break;
5010 
5011 		r = -EINVAL;
5012 		if (kvm_state.size < sizeof(kvm_state))
5013 			break;
5014 
5015 		if (kvm_state.flags &
5016 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5017 		      | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5018 		      | KVM_STATE_NESTED_GIF_SET))
5019 			break;
5020 
5021 		/* nested_run_pending implies guest_mode.  */
5022 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5023 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5024 			break;
5025 
5026 		idx = srcu_read_lock(&vcpu->kvm->srcu);
5027 		r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5028 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5029 		break;
5030 	}
5031 	case KVM_GET_SUPPORTED_HV_CPUID:
5032 		r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5033 		break;
5034 	default:
5035 		r = -EINVAL;
5036 	}
5037 out:
5038 	kfree(u.buffer);
5039 out_nofree:
5040 	vcpu_put(vcpu);
5041 	return r;
5042 }
5043 
5044 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5045 {
5046 	return VM_FAULT_SIGBUS;
5047 }
5048 
5049 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5050 {
5051 	int ret;
5052 
5053 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
5054 		return -EINVAL;
5055 	ret = kvm_x86_ops.set_tss_addr(kvm, addr);
5056 	return ret;
5057 }
5058 
5059 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5060 					      u64 ident_addr)
5061 {
5062 	return kvm_x86_ops.set_identity_map_addr(kvm, ident_addr);
5063 }
5064 
5065 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5066 					 unsigned long kvm_nr_mmu_pages)
5067 {
5068 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5069 		return -EINVAL;
5070 
5071 	mutex_lock(&kvm->slots_lock);
5072 
5073 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5074 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5075 
5076 	mutex_unlock(&kvm->slots_lock);
5077 	return 0;
5078 }
5079 
5080 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5081 {
5082 	return kvm->arch.n_max_mmu_pages;
5083 }
5084 
5085 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5086 {
5087 	struct kvm_pic *pic = kvm->arch.vpic;
5088 	int r;
5089 
5090 	r = 0;
5091 	switch (chip->chip_id) {
5092 	case KVM_IRQCHIP_PIC_MASTER:
5093 		memcpy(&chip->chip.pic, &pic->pics[0],
5094 			sizeof(struct kvm_pic_state));
5095 		break;
5096 	case KVM_IRQCHIP_PIC_SLAVE:
5097 		memcpy(&chip->chip.pic, &pic->pics[1],
5098 			sizeof(struct kvm_pic_state));
5099 		break;
5100 	case KVM_IRQCHIP_IOAPIC:
5101 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
5102 		break;
5103 	default:
5104 		r = -EINVAL;
5105 		break;
5106 	}
5107 	return r;
5108 }
5109 
5110 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5111 {
5112 	struct kvm_pic *pic = kvm->arch.vpic;
5113 	int r;
5114 
5115 	r = 0;
5116 	switch (chip->chip_id) {
5117 	case KVM_IRQCHIP_PIC_MASTER:
5118 		spin_lock(&pic->lock);
5119 		memcpy(&pic->pics[0], &chip->chip.pic,
5120 			sizeof(struct kvm_pic_state));
5121 		spin_unlock(&pic->lock);
5122 		break;
5123 	case KVM_IRQCHIP_PIC_SLAVE:
5124 		spin_lock(&pic->lock);
5125 		memcpy(&pic->pics[1], &chip->chip.pic,
5126 			sizeof(struct kvm_pic_state));
5127 		spin_unlock(&pic->lock);
5128 		break;
5129 	case KVM_IRQCHIP_IOAPIC:
5130 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
5131 		break;
5132 	default:
5133 		r = -EINVAL;
5134 		break;
5135 	}
5136 	kvm_pic_update_irq(pic);
5137 	return r;
5138 }
5139 
5140 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5141 {
5142 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5143 
5144 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5145 
5146 	mutex_lock(&kps->lock);
5147 	memcpy(ps, &kps->channels, sizeof(*ps));
5148 	mutex_unlock(&kps->lock);
5149 	return 0;
5150 }
5151 
5152 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5153 {
5154 	int i;
5155 	struct kvm_pit *pit = kvm->arch.vpit;
5156 
5157 	mutex_lock(&pit->pit_state.lock);
5158 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5159 	for (i = 0; i < 3; i++)
5160 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5161 	mutex_unlock(&pit->pit_state.lock);
5162 	return 0;
5163 }
5164 
5165 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5166 {
5167 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
5168 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5169 		sizeof(ps->channels));
5170 	ps->flags = kvm->arch.vpit->pit_state.flags;
5171 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5172 	memset(&ps->reserved, 0, sizeof(ps->reserved));
5173 	return 0;
5174 }
5175 
5176 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5177 {
5178 	int start = 0;
5179 	int i;
5180 	u32 prev_legacy, cur_legacy;
5181 	struct kvm_pit *pit = kvm->arch.vpit;
5182 
5183 	mutex_lock(&pit->pit_state.lock);
5184 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5185 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5186 	if (!prev_legacy && cur_legacy)
5187 		start = 1;
5188 	memcpy(&pit->pit_state.channels, &ps->channels,
5189 	       sizeof(pit->pit_state.channels));
5190 	pit->pit_state.flags = ps->flags;
5191 	for (i = 0; i < 3; i++)
5192 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5193 				   start && i == 0);
5194 	mutex_unlock(&pit->pit_state.lock);
5195 	return 0;
5196 }
5197 
5198 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5199 				 struct kvm_reinject_control *control)
5200 {
5201 	struct kvm_pit *pit = kvm->arch.vpit;
5202 
5203 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
5204 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5205 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
5206 	 */
5207 	mutex_lock(&pit->pit_state.lock);
5208 	kvm_pit_set_reinject(pit, control->pit_reinject);
5209 	mutex_unlock(&pit->pit_state.lock);
5210 
5211 	return 0;
5212 }
5213 
5214 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5215 {
5216 	/*
5217 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
5218 	 */
5219 	if (kvm_x86_ops.flush_log_dirty)
5220 		kvm_x86_ops.flush_log_dirty(kvm);
5221 }
5222 
5223 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5224 			bool line_status)
5225 {
5226 	if (!irqchip_in_kernel(kvm))
5227 		return -ENXIO;
5228 
5229 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5230 					irq_event->irq, irq_event->level,
5231 					line_status);
5232 	return 0;
5233 }
5234 
5235 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5236 			    struct kvm_enable_cap *cap)
5237 {
5238 	int r;
5239 
5240 	if (cap->flags)
5241 		return -EINVAL;
5242 
5243 	switch (cap->cap) {
5244 	case KVM_CAP_DISABLE_QUIRKS:
5245 		kvm->arch.disabled_quirks = cap->args[0];
5246 		r = 0;
5247 		break;
5248 	case KVM_CAP_SPLIT_IRQCHIP: {
5249 		mutex_lock(&kvm->lock);
5250 		r = -EINVAL;
5251 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5252 			goto split_irqchip_unlock;
5253 		r = -EEXIST;
5254 		if (irqchip_in_kernel(kvm))
5255 			goto split_irqchip_unlock;
5256 		if (kvm->created_vcpus)
5257 			goto split_irqchip_unlock;
5258 		r = kvm_setup_empty_irq_routing(kvm);
5259 		if (r)
5260 			goto split_irqchip_unlock;
5261 		/* Pairs with irqchip_in_kernel. */
5262 		smp_wmb();
5263 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5264 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5265 		r = 0;
5266 split_irqchip_unlock:
5267 		mutex_unlock(&kvm->lock);
5268 		break;
5269 	}
5270 	case KVM_CAP_X2APIC_API:
5271 		r = -EINVAL;
5272 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5273 			break;
5274 
5275 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5276 			kvm->arch.x2apic_format = true;
5277 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5278 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
5279 
5280 		r = 0;
5281 		break;
5282 	case KVM_CAP_X86_DISABLE_EXITS:
5283 		r = -EINVAL;
5284 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5285 			break;
5286 
5287 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5288 			kvm_can_mwait_in_guest())
5289 			kvm->arch.mwait_in_guest = true;
5290 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5291 			kvm->arch.hlt_in_guest = true;
5292 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5293 			kvm->arch.pause_in_guest = true;
5294 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5295 			kvm->arch.cstate_in_guest = true;
5296 		r = 0;
5297 		break;
5298 	case KVM_CAP_MSR_PLATFORM_INFO:
5299 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5300 		r = 0;
5301 		break;
5302 	case KVM_CAP_EXCEPTION_PAYLOAD:
5303 		kvm->arch.exception_payload_enabled = cap->args[0];
5304 		r = 0;
5305 		break;
5306 	case KVM_CAP_X86_USER_SPACE_MSR:
5307 		kvm->arch.user_space_msr_mask = cap->args[0];
5308 		r = 0;
5309 		break;
5310 	default:
5311 		r = -EINVAL;
5312 		break;
5313 	}
5314 	return r;
5315 }
5316 
5317 static void kvm_clear_msr_filter(struct kvm *kvm)
5318 {
5319 	u32 i;
5320 	u32 count = kvm->arch.msr_filter.count;
5321 	struct msr_bitmap_range ranges[16];
5322 
5323 	mutex_lock(&kvm->lock);
5324 	kvm->arch.msr_filter.count = 0;
5325 	memcpy(ranges, kvm->arch.msr_filter.ranges, count * sizeof(ranges[0]));
5326 	mutex_unlock(&kvm->lock);
5327 	synchronize_srcu(&kvm->srcu);
5328 
5329 	for (i = 0; i < count; i++)
5330 		kfree(ranges[i].bitmap);
5331 }
5332 
5333 static int kvm_add_msr_filter(struct kvm *kvm, struct kvm_msr_filter_range *user_range)
5334 {
5335 	struct msr_bitmap_range *ranges = kvm->arch.msr_filter.ranges;
5336 	struct msr_bitmap_range range;
5337 	unsigned long *bitmap = NULL;
5338 	size_t bitmap_size;
5339 	int r;
5340 
5341 	if (!user_range->nmsrs)
5342 		return 0;
5343 
5344 	bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5345 	if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5346 		return -EINVAL;
5347 
5348 	bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5349 	if (IS_ERR(bitmap))
5350 		return PTR_ERR(bitmap);
5351 
5352 	range = (struct msr_bitmap_range) {
5353 		.flags = user_range->flags,
5354 		.base = user_range->base,
5355 		.nmsrs = user_range->nmsrs,
5356 		.bitmap = bitmap,
5357 	};
5358 
5359 	if (range.flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE)) {
5360 		r = -EINVAL;
5361 		goto err;
5362 	}
5363 
5364 	if (!range.flags) {
5365 		r = -EINVAL;
5366 		goto err;
5367 	}
5368 
5369 	/* Everything ok, add this range identifier to our global pool */
5370 	ranges[kvm->arch.msr_filter.count] = range;
5371 	/* Make sure we filled the array before we tell anyone to walk it */
5372 	smp_wmb();
5373 	kvm->arch.msr_filter.count++;
5374 
5375 	return 0;
5376 err:
5377 	kfree(bitmap);
5378 	return r;
5379 }
5380 
5381 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5382 {
5383 	struct kvm_msr_filter __user *user_msr_filter = argp;
5384 	struct kvm_msr_filter filter;
5385 	bool default_allow;
5386 	int r = 0;
5387 	bool empty = true;
5388 	u32 i;
5389 
5390 	if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5391 		return -EFAULT;
5392 
5393 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5394 		empty &= !filter.ranges[i].nmsrs;
5395 
5396 	default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5397 	if (empty && !default_allow)
5398 		return -EINVAL;
5399 
5400 	kvm_clear_msr_filter(kvm);
5401 
5402 	kvm->arch.msr_filter.default_allow = default_allow;
5403 
5404 	/*
5405 	 * Protect from concurrent calls to this function that could trigger
5406 	 * a TOCTOU violation on kvm->arch.msr_filter.count.
5407 	 */
5408 	mutex_lock(&kvm->lock);
5409 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5410 		r = kvm_add_msr_filter(kvm, &filter.ranges[i]);
5411 		if (r)
5412 			break;
5413 	}
5414 
5415 	kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5416 	mutex_unlock(&kvm->lock);
5417 
5418 	return r;
5419 }
5420 
5421 long kvm_arch_vm_ioctl(struct file *filp,
5422 		       unsigned int ioctl, unsigned long arg)
5423 {
5424 	struct kvm *kvm = filp->private_data;
5425 	void __user *argp = (void __user *)arg;
5426 	int r = -ENOTTY;
5427 	/*
5428 	 * This union makes it completely explicit to gcc-3.x
5429 	 * that these two variables' stack usage should be
5430 	 * combined, not added together.
5431 	 */
5432 	union {
5433 		struct kvm_pit_state ps;
5434 		struct kvm_pit_state2 ps2;
5435 		struct kvm_pit_config pit_config;
5436 	} u;
5437 
5438 	switch (ioctl) {
5439 	case KVM_SET_TSS_ADDR:
5440 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5441 		break;
5442 	case KVM_SET_IDENTITY_MAP_ADDR: {
5443 		u64 ident_addr;
5444 
5445 		mutex_lock(&kvm->lock);
5446 		r = -EINVAL;
5447 		if (kvm->created_vcpus)
5448 			goto set_identity_unlock;
5449 		r = -EFAULT;
5450 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5451 			goto set_identity_unlock;
5452 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5453 set_identity_unlock:
5454 		mutex_unlock(&kvm->lock);
5455 		break;
5456 	}
5457 	case KVM_SET_NR_MMU_PAGES:
5458 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5459 		break;
5460 	case KVM_GET_NR_MMU_PAGES:
5461 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5462 		break;
5463 	case KVM_CREATE_IRQCHIP: {
5464 		mutex_lock(&kvm->lock);
5465 
5466 		r = -EEXIST;
5467 		if (irqchip_in_kernel(kvm))
5468 			goto create_irqchip_unlock;
5469 
5470 		r = -EINVAL;
5471 		if (kvm->created_vcpus)
5472 			goto create_irqchip_unlock;
5473 
5474 		r = kvm_pic_init(kvm);
5475 		if (r)
5476 			goto create_irqchip_unlock;
5477 
5478 		r = kvm_ioapic_init(kvm);
5479 		if (r) {
5480 			kvm_pic_destroy(kvm);
5481 			goto create_irqchip_unlock;
5482 		}
5483 
5484 		r = kvm_setup_default_irq_routing(kvm);
5485 		if (r) {
5486 			kvm_ioapic_destroy(kvm);
5487 			kvm_pic_destroy(kvm);
5488 			goto create_irqchip_unlock;
5489 		}
5490 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5491 		smp_wmb();
5492 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5493 	create_irqchip_unlock:
5494 		mutex_unlock(&kvm->lock);
5495 		break;
5496 	}
5497 	case KVM_CREATE_PIT:
5498 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5499 		goto create_pit;
5500 	case KVM_CREATE_PIT2:
5501 		r = -EFAULT;
5502 		if (copy_from_user(&u.pit_config, argp,
5503 				   sizeof(struct kvm_pit_config)))
5504 			goto out;
5505 	create_pit:
5506 		mutex_lock(&kvm->lock);
5507 		r = -EEXIST;
5508 		if (kvm->arch.vpit)
5509 			goto create_pit_unlock;
5510 		r = -ENOMEM;
5511 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5512 		if (kvm->arch.vpit)
5513 			r = 0;
5514 	create_pit_unlock:
5515 		mutex_unlock(&kvm->lock);
5516 		break;
5517 	case KVM_GET_IRQCHIP: {
5518 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5519 		struct kvm_irqchip *chip;
5520 
5521 		chip = memdup_user(argp, sizeof(*chip));
5522 		if (IS_ERR(chip)) {
5523 			r = PTR_ERR(chip);
5524 			goto out;
5525 		}
5526 
5527 		r = -ENXIO;
5528 		if (!irqchip_kernel(kvm))
5529 			goto get_irqchip_out;
5530 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5531 		if (r)
5532 			goto get_irqchip_out;
5533 		r = -EFAULT;
5534 		if (copy_to_user(argp, chip, sizeof(*chip)))
5535 			goto get_irqchip_out;
5536 		r = 0;
5537 	get_irqchip_out:
5538 		kfree(chip);
5539 		break;
5540 	}
5541 	case KVM_SET_IRQCHIP: {
5542 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5543 		struct kvm_irqchip *chip;
5544 
5545 		chip = memdup_user(argp, sizeof(*chip));
5546 		if (IS_ERR(chip)) {
5547 			r = PTR_ERR(chip);
5548 			goto out;
5549 		}
5550 
5551 		r = -ENXIO;
5552 		if (!irqchip_kernel(kvm))
5553 			goto set_irqchip_out;
5554 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5555 	set_irqchip_out:
5556 		kfree(chip);
5557 		break;
5558 	}
5559 	case KVM_GET_PIT: {
5560 		r = -EFAULT;
5561 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5562 			goto out;
5563 		r = -ENXIO;
5564 		if (!kvm->arch.vpit)
5565 			goto out;
5566 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5567 		if (r)
5568 			goto out;
5569 		r = -EFAULT;
5570 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5571 			goto out;
5572 		r = 0;
5573 		break;
5574 	}
5575 	case KVM_SET_PIT: {
5576 		r = -EFAULT;
5577 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5578 			goto out;
5579 		mutex_lock(&kvm->lock);
5580 		r = -ENXIO;
5581 		if (!kvm->arch.vpit)
5582 			goto set_pit_out;
5583 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5584 set_pit_out:
5585 		mutex_unlock(&kvm->lock);
5586 		break;
5587 	}
5588 	case KVM_GET_PIT2: {
5589 		r = -ENXIO;
5590 		if (!kvm->arch.vpit)
5591 			goto out;
5592 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5593 		if (r)
5594 			goto out;
5595 		r = -EFAULT;
5596 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5597 			goto out;
5598 		r = 0;
5599 		break;
5600 	}
5601 	case KVM_SET_PIT2: {
5602 		r = -EFAULT;
5603 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5604 			goto out;
5605 		mutex_lock(&kvm->lock);
5606 		r = -ENXIO;
5607 		if (!kvm->arch.vpit)
5608 			goto set_pit2_out;
5609 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5610 set_pit2_out:
5611 		mutex_unlock(&kvm->lock);
5612 		break;
5613 	}
5614 	case KVM_REINJECT_CONTROL: {
5615 		struct kvm_reinject_control control;
5616 		r =  -EFAULT;
5617 		if (copy_from_user(&control, argp, sizeof(control)))
5618 			goto out;
5619 		r = -ENXIO;
5620 		if (!kvm->arch.vpit)
5621 			goto out;
5622 		r = kvm_vm_ioctl_reinject(kvm, &control);
5623 		break;
5624 	}
5625 	case KVM_SET_BOOT_CPU_ID:
5626 		r = 0;
5627 		mutex_lock(&kvm->lock);
5628 		if (kvm->created_vcpus)
5629 			r = -EBUSY;
5630 		else
5631 			kvm->arch.bsp_vcpu_id = arg;
5632 		mutex_unlock(&kvm->lock);
5633 		break;
5634 	case KVM_XEN_HVM_CONFIG: {
5635 		struct kvm_xen_hvm_config xhc;
5636 		r = -EFAULT;
5637 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
5638 			goto out;
5639 		r = -EINVAL;
5640 		if (xhc.flags)
5641 			goto out;
5642 		memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5643 		r = 0;
5644 		break;
5645 	}
5646 	case KVM_SET_CLOCK: {
5647 		struct kvm_clock_data user_ns;
5648 		u64 now_ns;
5649 
5650 		r = -EFAULT;
5651 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5652 			goto out;
5653 
5654 		r = -EINVAL;
5655 		if (user_ns.flags)
5656 			goto out;
5657 
5658 		r = 0;
5659 		/*
5660 		 * TODO: userspace has to take care of races with VCPU_RUN, so
5661 		 * kvm_gen_update_masterclock() can be cut down to locked
5662 		 * pvclock_update_vm_gtod_copy().
5663 		 */
5664 		kvm_gen_update_masterclock(kvm);
5665 		now_ns = get_kvmclock_ns(kvm);
5666 		kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5667 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5668 		break;
5669 	}
5670 	case KVM_GET_CLOCK: {
5671 		struct kvm_clock_data user_ns;
5672 		u64 now_ns;
5673 
5674 		now_ns = get_kvmclock_ns(kvm);
5675 		user_ns.clock = now_ns;
5676 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5677 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5678 
5679 		r = -EFAULT;
5680 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5681 			goto out;
5682 		r = 0;
5683 		break;
5684 	}
5685 	case KVM_MEMORY_ENCRYPT_OP: {
5686 		r = -ENOTTY;
5687 		if (kvm_x86_ops.mem_enc_op)
5688 			r = kvm_x86_ops.mem_enc_op(kvm, argp);
5689 		break;
5690 	}
5691 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
5692 		struct kvm_enc_region region;
5693 
5694 		r = -EFAULT;
5695 		if (copy_from_user(&region, argp, sizeof(region)))
5696 			goto out;
5697 
5698 		r = -ENOTTY;
5699 		if (kvm_x86_ops.mem_enc_reg_region)
5700 			r = kvm_x86_ops.mem_enc_reg_region(kvm, &region);
5701 		break;
5702 	}
5703 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5704 		struct kvm_enc_region region;
5705 
5706 		r = -EFAULT;
5707 		if (copy_from_user(&region, argp, sizeof(region)))
5708 			goto out;
5709 
5710 		r = -ENOTTY;
5711 		if (kvm_x86_ops.mem_enc_unreg_region)
5712 			r = kvm_x86_ops.mem_enc_unreg_region(kvm, &region);
5713 		break;
5714 	}
5715 	case KVM_HYPERV_EVENTFD: {
5716 		struct kvm_hyperv_eventfd hvevfd;
5717 
5718 		r = -EFAULT;
5719 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5720 			goto out;
5721 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5722 		break;
5723 	}
5724 	case KVM_SET_PMU_EVENT_FILTER:
5725 		r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5726 		break;
5727 	case KVM_X86_SET_MSR_FILTER:
5728 		r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
5729 		break;
5730 	default:
5731 		r = -ENOTTY;
5732 	}
5733 out:
5734 	return r;
5735 }
5736 
5737 static void kvm_init_msr_list(void)
5738 {
5739 	struct x86_pmu_capability x86_pmu;
5740 	u32 dummy[2];
5741 	unsigned i;
5742 
5743 	BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5744 			 "Please update the fixed PMCs in msrs_to_saved_all[]");
5745 
5746 	perf_get_x86_pmu_capability(&x86_pmu);
5747 
5748 	num_msrs_to_save = 0;
5749 	num_emulated_msrs = 0;
5750 	num_msr_based_features = 0;
5751 
5752 	for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5753 		if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5754 			continue;
5755 
5756 		/*
5757 		 * Even MSRs that are valid in the host may not be exposed
5758 		 * to the guests in some cases.
5759 		 */
5760 		switch (msrs_to_save_all[i]) {
5761 		case MSR_IA32_BNDCFGS:
5762 			if (!kvm_mpx_supported())
5763 				continue;
5764 			break;
5765 		case MSR_TSC_AUX:
5766 			if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
5767 				continue;
5768 			break;
5769 		case MSR_IA32_UMWAIT_CONTROL:
5770 			if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
5771 				continue;
5772 			break;
5773 		case MSR_IA32_RTIT_CTL:
5774 		case MSR_IA32_RTIT_STATUS:
5775 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
5776 				continue;
5777 			break;
5778 		case MSR_IA32_RTIT_CR3_MATCH:
5779 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5780 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5781 				continue;
5782 			break;
5783 		case MSR_IA32_RTIT_OUTPUT_BASE:
5784 		case MSR_IA32_RTIT_OUTPUT_MASK:
5785 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5786 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5787 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5788 				continue;
5789 			break;
5790 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
5791 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5792 				msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5793 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5794 				continue;
5795 			break;
5796 		case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5797 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5798 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5799 				continue;
5800 			break;
5801 		case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5802 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5803 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5804 				continue;
5805 			break;
5806 		default:
5807 			break;
5808 		}
5809 
5810 		msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5811 	}
5812 
5813 	for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5814 		if (!kvm_x86_ops.has_emulated_msr(NULL, emulated_msrs_all[i]))
5815 			continue;
5816 
5817 		emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5818 	}
5819 
5820 	for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5821 		struct kvm_msr_entry msr;
5822 
5823 		msr.index = msr_based_features_all[i];
5824 		if (kvm_get_msr_feature(&msr))
5825 			continue;
5826 
5827 		msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
5828 	}
5829 }
5830 
5831 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5832 			   const void *v)
5833 {
5834 	int handled = 0;
5835 	int n;
5836 
5837 	do {
5838 		n = min(len, 8);
5839 		if (!(lapic_in_kernel(vcpu) &&
5840 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5841 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5842 			break;
5843 		handled += n;
5844 		addr += n;
5845 		len -= n;
5846 		v += n;
5847 	} while (len);
5848 
5849 	return handled;
5850 }
5851 
5852 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5853 {
5854 	int handled = 0;
5855 	int n;
5856 
5857 	do {
5858 		n = min(len, 8);
5859 		if (!(lapic_in_kernel(vcpu) &&
5860 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5861 					 addr, n, v))
5862 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5863 			break;
5864 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5865 		handled += n;
5866 		addr += n;
5867 		len -= n;
5868 		v += n;
5869 	} while (len);
5870 
5871 	return handled;
5872 }
5873 
5874 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5875 			struct kvm_segment *var, int seg)
5876 {
5877 	kvm_x86_ops.set_segment(vcpu, var, seg);
5878 }
5879 
5880 void kvm_get_segment(struct kvm_vcpu *vcpu,
5881 		     struct kvm_segment *var, int seg)
5882 {
5883 	kvm_x86_ops.get_segment(vcpu, var, seg);
5884 }
5885 
5886 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5887 			   struct x86_exception *exception)
5888 {
5889 	gpa_t t_gpa;
5890 
5891 	BUG_ON(!mmu_is_nested(vcpu));
5892 
5893 	/* NPT walks are always user-walks */
5894 	access |= PFERR_USER_MASK;
5895 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5896 
5897 	return t_gpa;
5898 }
5899 
5900 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5901 			      struct x86_exception *exception)
5902 {
5903 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5904 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5905 }
5906 
5907  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5908 				struct x86_exception *exception)
5909 {
5910 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5911 	access |= PFERR_FETCH_MASK;
5912 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5913 }
5914 
5915 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5916 			       struct x86_exception *exception)
5917 {
5918 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5919 	access |= PFERR_WRITE_MASK;
5920 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5921 }
5922 
5923 /* uses this to access any guest's mapped memory without checking CPL */
5924 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5925 				struct x86_exception *exception)
5926 {
5927 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5928 }
5929 
5930 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5931 				      struct kvm_vcpu *vcpu, u32 access,
5932 				      struct x86_exception *exception)
5933 {
5934 	void *data = val;
5935 	int r = X86EMUL_CONTINUE;
5936 
5937 	while (bytes) {
5938 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5939 							    exception);
5940 		unsigned offset = addr & (PAGE_SIZE-1);
5941 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5942 		int ret;
5943 
5944 		if (gpa == UNMAPPED_GVA)
5945 			return X86EMUL_PROPAGATE_FAULT;
5946 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5947 					       offset, toread);
5948 		if (ret < 0) {
5949 			r = X86EMUL_IO_NEEDED;
5950 			goto out;
5951 		}
5952 
5953 		bytes -= toread;
5954 		data += toread;
5955 		addr += toread;
5956 	}
5957 out:
5958 	return r;
5959 }
5960 
5961 /* used for instruction fetching */
5962 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5963 				gva_t addr, void *val, unsigned int bytes,
5964 				struct x86_exception *exception)
5965 {
5966 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5967 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5968 	unsigned offset;
5969 	int ret;
5970 
5971 	/* Inline kvm_read_guest_virt_helper for speed.  */
5972 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5973 						    exception);
5974 	if (unlikely(gpa == UNMAPPED_GVA))
5975 		return X86EMUL_PROPAGATE_FAULT;
5976 
5977 	offset = addr & (PAGE_SIZE-1);
5978 	if (WARN_ON(offset + bytes > PAGE_SIZE))
5979 		bytes = (unsigned)PAGE_SIZE - offset;
5980 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5981 				       offset, bytes);
5982 	if (unlikely(ret < 0))
5983 		return X86EMUL_IO_NEEDED;
5984 
5985 	return X86EMUL_CONTINUE;
5986 }
5987 
5988 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5989 			       gva_t addr, void *val, unsigned int bytes,
5990 			       struct x86_exception *exception)
5991 {
5992 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5993 
5994 	/*
5995 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5996 	 * is returned, but our callers are not ready for that and they blindly
5997 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
5998 	 * uninitialized kernel stack memory into cr2 and error code.
5999 	 */
6000 	memset(exception, 0, sizeof(*exception));
6001 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6002 					  exception);
6003 }
6004 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6005 
6006 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6007 			     gva_t addr, void *val, unsigned int bytes,
6008 			     struct x86_exception *exception, bool system)
6009 {
6010 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6011 	u32 access = 0;
6012 
6013 	if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
6014 		access |= PFERR_USER_MASK;
6015 
6016 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6017 }
6018 
6019 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6020 		unsigned long addr, void *val, unsigned int bytes)
6021 {
6022 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6023 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6024 
6025 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6026 }
6027 
6028 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6029 				      struct kvm_vcpu *vcpu, u32 access,
6030 				      struct x86_exception *exception)
6031 {
6032 	void *data = val;
6033 	int r = X86EMUL_CONTINUE;
6034 
6035 	while (bytes) {
6036 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6037 							     access,
6038 							     exception);
6039 		unsigned offset = addr & (PAGE_SIZE-1);
6040 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6041 		int ret;
6042 
6043 		if (gpa == UNMAPPED_GVA)
6044 			return X86EMUL_PROPAGATE_FAULT;
6045 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6046 		if (ret < 0) {
6047 			r = X86EMUL_IO_NEEDED;
6048 			goto out;
6049 		}
6050 
6051 		bytes -= towrite;
6052 		data += towrite;
6053 		addr += towrite;
6054 	}
6055 out:
6056 	return r;
6057 }
6058 
6059 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6060 			      unsigned int bytes, struct x86_exception *exception,
6061 			      bool system)
6062 {
6063 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6064 	u32 access = PFERR_WRITE_MASK;
6065 
6066 	if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
6067 		access |= PFERR_USER_MASK;
6068 
6069 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6070 					   access, exception);
6071 }
6072 
6073 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6074 				unsigned int bytes, struct x86_exception *exception)
6075 {
6076 	/* kvm_write_guest_virt_system can pull in tons of pages. */
6077 	vcpu->arch.l1tf_flush_l1d = true;
6078 
6079 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6080 					   PFERR_WRITE_MASK, exception);
6081 }
6082 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6083 
6084 int handle_ud(struct kvm_vcpu *vcpu)
6085 {
6086 	static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6087 	int emul_type = EMULTYPE_TRAP_UD;
6088 	char sig[5]; /* ud2; .ascii "kvm" */
6089 	struct x86_exception e;
6090 
6091 	if (unlikely(!kvm_x86_ops.can_emulate_instruction(vcpu, NULL, 0)))
6092 		return 1;
6093 
6094 	if (force_emulation_prefix &&
6095 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6096 				sig, sizeof(sig), &e) == 0 &&
6097 	    memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6098 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6099 		emul_type = EMULTYPE_TRAP_UD_FORCED;
6100 	}
6101 
6102 	return kvm_emulate_instruction(vcpu, emul_type);
6103 }
6104 EXPORT_SYMBOL_GPL(handle_ud);
6105 
6106 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6107 			    gpa_t gpa, bool write)
6108 {
6109 	/* For APIC access vmexit */
6110 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6111 		return 1;
6112 
6113 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6114 		trace_vcpu_match_mmio(gva, gpa, write, true);
6115 		return 1;
6116 	}
6117 
6118 	return 0;
6119 }
6120 
6121 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6122 				gpa_t *gpa, struct x86_exception *exception,
6123 				bool write)
6124 {
6125 	u32 access = ((kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
6126 		| (write ? PFERR_WRITE_MASK : 0);
6127 
6128 	/*
6129 	 * currently PKRU is only applied to ept enabled guest so
6130 	 * there is no pkey in EPT page table for L1 guest or EPT
6131 	 * shadow page table for L2 guest.
6132 	 */
6133 	if (vcpu_match_mmio_gva(vcpu, gva)
6134 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
6135 				 vcpu->arch.mmio_access, 0, access)) {
6136 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6137 					(gva & (PAGE_SIZE - 1));
6138 		trace_vcpu_match_mmio(gva, *gpa, write, false);
6139 		return 1;
6140 	}
6141 
6142 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6143 
6144 	if (*gpa == UNMAPPED_GVA)
6145 		return -1;
6146 
6147 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6148 }
6149 
6150 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6151 			const void *val, int bytes)
6152 {
6153 	int ret;
6154 
6155 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6156 	if (ret < 0)
6157 		return 0;
6158 	kvm_page_track_write(vcpu, gpa, val, bytes);
6159 	return 1;
6160 }
6161 
6162 struct read_write_emulator_ops {
6163 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6164 				  int bytes);
6165 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6166 				  void *val, int bytes);
6167 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6168 			       int bytes, void *val);
6169 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6170 				    void *val, int bytes);
6171 	bool write;
6172 };
6173 
6174 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6175 {
6176 	if (vcpu->mmio_read_completed) {
6177 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6178 			       vcpu->mmio_fragments[0].gpa, val);
6179 		vcpu->mmio_read_completed = 0;
6180 		return 1;
6181 	}
6182 
6183 	return 0;
6184 }
6185 
6186 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6187 			void *val, int bytes)
6188 {
6189 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6190 }
6191 
6192 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6193 			 void *val, int bytes)
6194 {
6195 	return emulator_write_phys(vcpu, gpa, val, bytes);
6196 }
6197 
6198 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6199 {
6200 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6201 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
6202 }
6203 
6204 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6205 			  void *val, int bytes)
6206 {
6207 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6208 	return X86EMUL_IO_NEEDED;
6209 }
6210 
6211 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6212 			   void *val, int bytes)
6213 {
6214 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6215 
6216 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6217 	return X86EMUL_CONTINUE;
6218 }
6219 
6220 static const struct read_write_emulator_ops read_emultor = {
6221 	.read_write_prepare = read_prepare,
6222 	.read_write_emulate = read_emulate,
6223 	.read_write_mmio = vcpu_mmio_read,
6224 	.read_write_exit_mmio = read_exit_mmio,
6225 };
6226 
6227 static const struct read_write_emulator_ops write_emultor = {
6228 	.read_write_emulate = write_emulate,
6229 	.read_write_mmio = write_mmio,
6230 	.read_write_exit_mmio = write_exit_mmio,
6231 	.write = true,
6232 };
6233 
6234 static int emulator_read_write_onepage(unsigned long addr, void *val,
6235 				       unsigned int bytes,
6236 				       struct x86_exception *exception,
6237 				       struct kvm_vcpu *vcpu,
6238 				       const struct read_write_emulator_ops *ops)
6239 {
6240 	gpa_t gpa;
6241 	int handled, ret;
6242 	bool write = ops->write;
6243 	struct kvm_mmio_fragment *frag;
6244 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6245 
6246 	/*
6247 	 * If the exit was due to a NPF we may already have a GPA.
6248 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6249 	 * Note, this cannot be used on string operations since string
6250 	 * operation using rep will only have the initial GPA from the NPF
6251 	 * occurred.
6252 	 */
6253 	if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6254 	    (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6255 		gpa = ctxt->gpa_val;
6256 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6257 	} else {
6258 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6259 		if (ret < 0)
6260 			return X86EMUL_PROPAGATE_FAULT;
6261 	}
6262 
6263 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6264 		return X86EMUL_CONTINUE;
6265 
6266 	/*
6267 	 * Is this MMIO handled locally?
6268 	 */
6269 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6270 	if (handled == bytes)
6271 		return X86EMUL_CONTINUE;
6272 
6273 	gpa += handled;
6274 	bytes -= handled;
6275 	val += handled;
6276 
6277 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6278 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6279 	frag->gpa = gpa;
6280 	frag->data = val;
6281 	frag->len = bytes;
6282 	return X86EMUL_CONTINUE;
6283 }
6284 
6285 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6286 			unsigned long addr,
6287 			void *val, unsigned int bytes,
6288 			struct x86_exception *exception,
6289 			const struct read_write_emulator_ops *ops)
6290 {
6291 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6292 	gpa_t gpa;
6293 	int rc;
6294 
6295 	if (ops->read_write_prepare &&
6296 		  ops->read_write_prepare(vcpu, val, bytes))
6297 		return X86EMUL_CONTINUE;
6298 
6299 	vcpu->mmio_nr_fragments = 0;
6300 
6301 	/* Crossing a page boundary? */
6302 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6303 		int now;
6304 
6305 		now = -addr & ~PAGE_MASK;
6306 		rc = emulator_read_write_onepage(addr, val, now, exception,
6307 						 vcpu, ops);
6308 
6309 		if (rc != X86EMUL_CONTINUE)
6310 			return rc;
6311 		addr += now;
6312 		if (ctxt->mode != X86EMUL_MODE_PROT64)
6313 			addr = (u32)addr;
6314 		val += now;
6315 		bytes -= now;
6316 	}
6317 
6318 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
6319 					 vcpu, ops);
6320 	if (rc != X86EMUL_CONTINUE)
6321 		return rc;
6322 
6323 	if (!vcpu->mmio_nr_fragments)
6324 		return rc;
6325 
6326 	gpa = vcpu->mmio_fragments[0].gpa;
6327 
6328 	vcpu->mmio_needed = 1;
6329 	vcpu->mmio_cur_fragment = 0;
6330 
6331 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6332 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6333 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
6334 	vcpu->run->mmio.phys_addr = gpa;
6335 
6336 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6337 }
6338 
6339 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6340 				  unsigned long addr,
6341 				  void *val,
6342 				  unsigned int bytes,
6343 				  struct x86_exception *exception)
6344 {
6345 	return emulator_read_write(ctxt, addr, val, bytes,
6346 				   exception, &read_emultor);
6347 }
6348 
6349 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6350 			    unsigned long addr,
6351 			    const void *val,
6352 			    unsigned int bytes,
6353 			    struct x86_exception *exception)
6354 {
6355 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
6356 				   exception, &write_emultor);
6357 }
6358 
6359 #define CMPXCHG_TYPE(t, ptr, old, new) \
6360 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6361 
6362 #ifdef CONFIG_X86_64
6363 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6364 #else
6365 #  define CMPXCHG64(ptr, old, new) \
6366 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6367 #endif
6368 
6369 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6370 				     unsigned long addr,
6371 				     const void *old,
6372 				     const void *new,
6373 				     unsigned int bytes,
6374 				     struct x86_exception *exception)
6375 {
6376 	struct kvm_host_map map;
6377 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6378 	u64 page_line_mask;
6379 	gpa_t gpa;
6380 	char *kaddr;
6381 	bool exchanged;
6382 
6383 	/* guests cmpxchg8b have to be emulated atomically */
6384 	if (bytes > 8 || (bytes & (bytes - 1)))
6385 		goto emul_write;
6386 
6387 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6388 
6389 	if (gpa == UNMAPPED_GVA ||
6390 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6391 		goto emul_write;
6392 
6393 	/*
6394 	 * Emulate the atomic as a straight write to avoid #AC if SLD is
6395 	 * enabled in the host and the access splits a cache line.
6396 	 */
6397 	if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6398 		page_line_mask = ~(cache_line_size() - 1);
6399 	else
6400 		page_line_mask = PAGE_MASK;
6401 
6402 	if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6403 		goto emul_write;
6404 
6405 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6406 		goto emul_write;
6407 
6408 	kaddr = map.hva + offset_in_page(gpa);
6409 
6410 	switch (bytes) {
6411 	case 1:
6412 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6413 		break;
6414 	case 2:
6415 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6416 		break;
6417 	case 4:
6418 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6419 		break;
6420 	case 8:
6421 		exchanged = CMPXCHG64(kaddr, old, new);
6422 		break;
6423 	default:
6424 		BUG();
6425 	}
6426 
6427 	kvm_vcpu_unmap(vcpu, &map, true);
6428 
6429 	if (!exchanged)
6430 		return X86EMUL_CMPXCHG_FAILED;
6431 
6432 	kvm_page_track_write(vcpu, gpa, new, bytes);
6433 
6434 	return X86EMUL_CONTINUE;
6435 
6436 emul_write:
6437 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6438 
6439 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6440 }
6441 
6442 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6443 {
6444 	int r = 0, i;
6445 
6446 	for (i = 0; i < vcpu->arch.pio.count; i++) {
6447 		if (vcpu->arch.pio.in)
6448 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6449 					    vcpu->arch.pio.size, pd);
6450 		else
6451 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6452 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
6453 					     pd);
6454 		if (r)
6455 			break;
6456 		pd += vcpu->arch.pio.size;
6457 	}
6458 	return r;
6459 }
6460 
6461 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6462 			       unsigned short port, void *val,
6463 			       unsigned int count, bool in)
6464 {
6465 	vcpu->arch.pio.port = port;
6466 	vcpu->arch.pio.in = in;
6467 	vcpu->arch.pio.count  = count;
6468 	vcpu->arch.pio.size = size;
6469 
6470 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6471 		vcpu->arch.pio.count = 0;
6472 		return 1;
6473 	}
6474 
6475 	vcpu->run->exit_reason = KVM_EXIT_IO;
6476 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6477 	vcpu->run->io.size = size;
6478 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6479 	vcpu->run->io.count = count;
6480 	vcpu->run->io.port = port;
6481 
6482 	return 0;
6483 }
6484 
6485 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6486 			   unsigned short port, void *val, unsigned int count)
6487 {
6488 	int ret;
6489 
6490 	if (vcpu->arch.pio.count)
6491 		goto data_avail;
6492 
6493 	memset(vcpu->arch.pio_data, 0, size * count);
6494 
6495 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6496 	if (ret) {
6497 data_avail:
6498 		memcpy(val, vcpu->arch.pio_data, size * count);
6499 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6500 		vcpu->arch.pio.count = 0;
6501 		return 1;
6502 	}
6503 
6504 	return 0;
6505 }
6506 
6507 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6508 				    int size, unsigned short port, void *val,
6509 				    unsigned int count)
6510 {
6511 	return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6512 
6513 }
6514 
6515 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6516 			    unsigned short port, const void *val,
6517 			    unsigned int count)
6518 {
6519 	memcpy(vcpu->arch.pio_data, val, size * count);
6520 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6521 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6522 }
6523 
6524 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6525 				     int size, unsigned short port,
6526 				     const void *val, unsigned int count)
6527 {
6528 	return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6529 }
6530 
6531 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6532 {
6533 	return kvm_x86_ops.get_segment_base(vcpu, seg);
6534 }
6535 
6536 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6537 {
6538 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6539 }
6540 
6541 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6542 {
6543 	if (!need_emulate_wbinvd(vcpu))
6544 		return X86EMUL_CONTINUE;
6545 
6546 	if (kvm_x86_ops.has_wbinvd_exit()) {
6547 		int cpu = get_cpu();
6548 
6549 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6550 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
6551 				wbinvd_ipi, NULL, 1);
6552 		put_cpu();
6553 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6554 	} else
6555 		wbinvd();
6556 	return X86EMUL_CONTINUE;
6557 }
6558 
6559 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6560 {
6561 	kvm_emulate_wbinvd_noskip(vcpu);
6562 	return kvm_skip_emulated_instruction(vcpu);
6563 }
6564 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6565 
6566 
6567 
6568 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6569 {
6570 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6571 }
6572 
6573 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6574 			   unsigned long *dest)
6575 {
6576 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6577 }
6578 
6579 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6580 			   unsigned long value)
6581 {
6582 
6583 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6584 }
6585 
6586 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6587 {
6588 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6589 }
6590 
6591 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6592 {
6593 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6594 	unsigned long value;
6595 
6596 	switch (cr) {
6597 	case 0:
6598 		value = kvm_read_cr0(vcpu);
6599 		break;
6600 	case 2:
6601 		value = vcpu->arch.cr2;
6602 		break;
6603 	case 3:
6604 		value = kvm_read_cr3(vcpu);
6605 		break;
6606 	case 4:
6607 		value = kvm_read_cr4(vcpu);
6608 		break;
6609 	case 8:
6610 		value = kvm_get_cr8(vcpu);
6611 		break;
6612 	default:
6613 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
6614 		return 0;
6615 	}
6616 
6617 	return value;
6618 }
6619 
6620 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6621 {
6622 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6623 	int res = 0;
6624 
6625 	switch (cr) {
6626 	case 0:
6627 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6628 		break;
6629 	case 2:
6630 		vcpu->arch.cr2 = val;
6631 		break;
6632 	case 3:
6633 		res = kvm_set_cr3(vcpu, val);
6634 		break;
6635 	case 4:
6636 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6637 		break;
6638 	case 8:
6639 		res = kvm_set_cr8(vcpu, val);
6640 		break;
6641 	default:
6642 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
6643 		res = -1;
6644 	}
6645 
6646 	return res;
6647 }
6648 
6649 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6650 {
6651 	return kvm_x86_ops.get_cpl(emul_to_vcpu(ctxt));
6652 }
6653 
6654 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6655 {
6656 	kvm_x86_ops.get_gdt(emul_to_vcpu(ctxt), dt);
6657 }
6658 
6659 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6660 {
6661 	kvm_x86_ops.get_idt(emul_to_vcpu(ctxt), dt);
6662 }
6663 
6664 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6665 {
6666 	kvm_x86_ops.set_gdt(emul_to_vcpu(ctxt), dt);
6667 }
6668 
6669 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6670 {
6671 	kvm_x86_ops.set_idt(emul_to_vcpu(ctxt), dt);
6672 }
6673 
6674 static unsigned long emulator_get_cached_segment_base(
6675 	struct x86_emulate_ctxt *ctxt, int seg)
6676 {
6677 	return get_segment_base(emul_to_vcpu(ctxt), seg);
6678 }
6679 
6680 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6681 				 struct desc_struct *desc, u32 *base3,
6682 				 int seg)
6683 {
6684 	struct kvm_segment var;
6685 
6686 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6687 	*selector = var.selector;
6688 
6689 	if (var.unusable) {
6690 		memset(desc, 0, sizeof(*desc));
6691 		if (base3)
6692 			*base3 = 0;
6693 		return false;
6694 	}
6695 
6696 	if (var.g)
6697 		var.limit >>= 12;
6698 	set_desc_limit(desc, var.limit);
6699 	set_desc_base(desc, (unsigned long)var.base);
6700 #ifdef CONFIG_X86_64
6701 	if (base3)
6702 		*base3 = var.base >> 32;
6703 #endif
6704 	desc->type = var.type;
6705 	desc->s = var.s;
6706 	desc->dpl = var.dpl;
6707 	desc->p = var.present;
6708 	desc->avl = var.avl;
6709 	desc->l = var.l;
6710 	desc->d = var.db;
6711 	desc->g = var.g;
6712 
6713 	return true;
6714 }
6715 
6716 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6717 				 struct desc_struct *desc, u32 base3,
6718 				 int seg)
6719 {
6720 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6721 	struct kvm_segment var;
6722 
6723 	var.selector = selector;
6724 	var.base = get_desc_base(desc);
6725 #ifdef CONFIG_X86_64
6726 	var.base |= ((u64)base3) << 32;
6727 #endif
6728 	var.limit = get_desc_limit(desc);
6729 	if (desc->g)
6730 		var.limit = (var.limit << 12) | 0xfff;
6731 	var.type = desc->type;
6732 	var.dpl = desc->dpl;
6733 	var.db = desc->d;
6734 	var.s = desc->s;
6735 	var.l = desc->l;
6736 	var.g = desc->g;
6737 	var.avl = desc->avl;
6738 	var.present = desc->p;
6739 	var.unusable = !var.present;
6740 	var.padding = 0;
6741 
6742 	kvm_set_segment(vcpu, &var, seg);
6743 	return;
6744 }
6745 
6746 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6747 			    u32 msr_index, u64 *pdata)
6748 {
6749 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6750 	int r;
6751 
6752 	r = kvm_get_msr(vcpu, msr_index, pdata);
6753 
6754 	if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
6755 		/* Bounce to user space */
6756 		return X86EMUL_IO_NEEDED;
6757 	}
6758 
6759 	return r;
6760 }
6761 
6762 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6763 			    u32 msr_index, u64 data)
6764 {
6765 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6766 	int r;
6767 
6768 	r = kvm_set_msr(vcpu, msr_index, data);
6769 
6770 	if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
6771 		/* Bounce to user space */
6772 		return X86EMUL_IO_NEEDED;
6773 	}
6774 
6775 	return r;
6776 }
6777 
6778 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6779 {
6780 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6781 
6782 	return vcpu->arch.smbase;
6783 }
6784 
6785 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6786 {
6787 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6788 
6789 	vcpu->arch.smbase = smbase;
6790 }
6791 
6792 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6793 			      u32 pmc)
6794 {
6795 	return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6796 }
6797 
6798 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6799 			     u32 pmc, u64 *pdata)
6800 {
6801 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6802 }
6803 
6804 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6805 {
6806 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
6807 }
6808 
6809 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6810 			      struct x86_instruction_info *info,
6811 			      enum x86_intercept_stage stage)
6812 {
6813 	return kvm_x86_ops.check_intercept(emul_to_vcpu(ctxt), info, stage,
6814 					    &ctxt->exception);
6815 }
6816 
6817 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6818 			      u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
6819 			      bool exact_only)
6820 {
6821 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
6822 }
6823 
6824 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
6825 {
6826 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
6827 }
6828 
6829 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
6830 {
6831 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
6832 }
6833 
6834 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
6835 {
6836 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
6837 }
6838 
6839 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6840 {
6841 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
6842 }
6843 
6844 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6845 {
6846 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6847 }
6848 
6849 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6850 {
6851 	kvm_x86_ops.set_nmi_mask(emul_to_vcpu(ctxt), masked);
6852 }
6853 
6854 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6855 {
6856 	return emul_to_vcpu(ctxt)->arch.hflags;
6857 }
6858 
6859 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6860 {
6861 	emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6862 }
6863 
6864 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6865 				  const char *smstate)
6866 {
6867 	return kvm_x86_ops.pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6868 }
6869 
6870 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6871 {
6872 	kvm_smm_changed(emul_to_vcpu(ctxt));
6873 }
6874 
6875 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6876 {
6877 	return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6878 }
6879 
6880 static const struct x86_emulate_ops emulate_ops = {
6881 	.read_gpr            = emulator_read_gpr,
6882 	.write_gpr           = emulator_write_gpr,
6883 	.read_std            = emulator_read_std,
6884 	.write_std           = emulator_write_std,
6885 	.read_phys           = kvm_read_guest_phys_system,
6886 	.fetch               = kvm_fetch_guest_virt,
6887 	.read_emulated       = emulator_read_emulated,
6888 	.write_emulated      = emulator_write_emulated,
6889 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
6890 	.invlpg              = emulator_invlpg,
6891 	.pio_in_emulated     = emulator_pio_in_emulated,
6892 	.pio_out_emulated    = emulator_pio_out_emulated,
6893 	.get_segment         = emulator_get_segment,
6894 	.set_segment         = emulator_set_segment,
6895 	.get_cached_segment_base = emulator_get_cached_segment_base,
6896 	.get_gdt             = emulator_get_gdt,
6897 	.get_idt	     = emulator_get_idt,
6898 	.set_gdt             = emulator_set_gdt,
6899 	.set_idt	     = emulator_set_idt,
6900 	.get_cr              = emulator_get_cr,
6901 	.set_cr              = emulator_set_cr,
6902 	.cpl                 = emulator_get_cpl,
6903 	.get_dr              = emulator_get_dr,
6904 	.set_dr              = emulator_set_dr,
6905 	.get_smbase          = emulator_get_smbase,
6906 	.set_smbase          = emulator_set_smbase,
6907 	.set_msr             = emulator_set_msr,
6908 	.get_msr             = emulator_get_msr,
6909 	.check_pmc	     = emulator_check_pmc,
6910 	.read_pmc            = emulator_read_pmc,
6911 	.halt                = emulator_halt,
6912 	.wbinvd              = emulator_wbinvd,
6913 	.fix_hypercall       = emulator_fix_hypercall,
6914 	.intercept           = emulator_intercept,
6915 	.get_cpuid           = emulator_get_cpuid,
6916 	.guest_has_long_mode = emulator_guest_has_long_mode,
6917 	.guest_has_movbe     = emulator_guest_has_movbe,
6918 	.guest_has_fxsr      = emulator_guest_has_fxsr,
6919 	.set_nmi_mask        = emulator_set_nmi_mask,
6920 	.get_hflags          = emulator_get_hflags,
6921 	.set_hflags          = emulator_set_hflags,
6922 	.pre_leave_smm       = emulator_pre_leave_smm,
6923 	.post_leave_smm      = emulator_post_leave_smm,
6924 	.set_xcr             = emulator_set_xcr,
6925 };
6926 
6927 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6928 {
6929 	u32 int_shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
6930 	/*
6931 	 * an sti; sti; sequence only disable interrupts for the first
6932 	 * instruction. So, if the last instruction, be it emulated or
6933 	 * not, left the system with the INT_STI flag enabled, it
6934 	 * means that the last instruction is an sti. We should not
6935 	 * leave the flag on in this case. The same goes for mov ss
6936 	 */
6937 	if (int_shadow & mask)
6938 		mask = 0;
6939 	if (unlikely(int_shadow || mask)) {
6940 		kvm_x86_ops.set_interrupt_shadow(vcpu, mask);
6941 		if (!mask)
6942 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6943 	}
6944 }
6945 
6946 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6947 {
6948 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6949 	if (ctxt->exception.vector == PF_VECTOR)
6950 		return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
6951 
6952 	if (ctxt->exception.error_code_valid)
6953 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6954 				      ctxt->exception.error_code);
6955 	else
6956 		kvm_queue_exception(vcpu, ctxt->exception.vector);
6957 	return false;
6958 }
6959 
6960 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
6961 {
6962 	struct x86_emulate_ctxt *ctxt;
6963 
6964 	ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
6965 	if (!ctxt) {
6966 		pr_err("kvm: failed to allocate vcpu's emulator\n");
6967 		return NULL;
6968 	}
6969 
6970 	ctxt->vcpu = vcpu;
6971 	ctxt->ops = &emulate_ops;
6972 	vcpu->arch.emulate_ctxt = ctxt;
6973 
6974 	return ctxt;
6975 }
6976 
6977 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6978 {
6979 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6980 	int cs_db, cs_l;
6981 
6982 	kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6983 
6984 	ctxt->gpa_available = false;
6985 	ctxt->eflags = kvm_get_rflags(vcpu);
6986 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6987 
6988 	ctxt->eip = kvm_rip_read(vcpu);
6989 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
6990 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
6991 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
6992 		     cs_db				? X86EMUL_MODE_PROT32 :
6993 							  X86EMUL_MODE_PROT16;
6994 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6995 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6996 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6997 
6998 	init_decode_cache(ctxt);
6999 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7000 }
7001 
7002 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7003 {
7004 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7005 	int ret;
7006 
7007 	init_emulate_ctxt(vcpu);
7008 
7009 	ctxt->op_bytes = 2;
7010 	ctxt->ad_bytes = 2;
7011 	ctxt->_eip = ctxt->eip + inc_eip;
7012 	ret = emulate_int_real(ctxt, irq);
7013 
7014 	if (ret != X86EMUL_CONTINUE) {
7015 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7016 	} else {
7017 		ctxt->eip = ctxt->_eip;
7018 		kvm_rip_write(vcpu, ctxt->eip);
7019 		kvm_set_rflags(vcpu, ctxt->eflags);
7020 	}
7021 }
7022 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7023 
7024 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7025 {
7026 	++vcpu->stat.insn_emulation_fail;
7027 	trace_kvm_emulate_insn_failed(vcpu);
7028 
7029 	if (emulation_type & EMULTYPE_VMWARE_GP) {
7030 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7031 		return 1;
7032 	}
7033 
7034 	if (emulation_type & EMULTYPE_SKIP) {
7035 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7036 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7037 		vcpu->run->internal.ndata = 0;
7038 		return 0;
7039 	}
7040 
7041 	kvm_queue_exception(vcpu, UD_VECTOR);
7042 
7043 	if (!is_guest_mode(vcpu) && kvm_x86_ops.get_cpl(vcpu) == 0) {
7044 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7045 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7046 		vcpu->run->internal.ndata = 0;
7047 		return 0;
7048 	}
7049 
7050 	return 1;
7051 }
7052 
7053 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7054 				  bool write_fault_to_shadow_pgtable,
7055 				  int emulation_type)
7056 {
7057 	gpa_t gpa = cr2_or_gpa;
7058 	kvm_pfn_t pfn;
7059 
7060 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7061 		return false;
7062 
7063 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7064 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7065 		return false;
7066 
7067 	if (!vcpu->arch.mmu->direct_map) {
7068 		/*
7069 		 * Write permission should be allowed since only
7070 		 * write access need to be emulated.
7071 		 */
7072 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7073 
7074 		/*
7075 		 * If the mapping is invalid in guest, let cpu retry
7076 		 * it to generate fault.
7077 		 */
7078 		if (gpa == UNMAPPED_GVA)
7079 			return true;
7080 	}
7081 
7082 	/*
7083 	 * Do not retry the unhandleable instruction if it faults on the
7084 	 * readonly host memory, otherwise it will goto a infinite loop:
7085 	 * retry instruction -> write #PF -> emulation fail -> retry
7086 	 * instruction -> ...
7087 	 */
7088 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7089 
7090 	/*
7091 	 * If the instruction failed on the error pfn, it can not be fixed,
7092 	 * report the error to userspace.
7093 	 */
7094 	if (is_error_noslot_pfn(pfn))
7095 		return false;
7096 
7097 	kvm_release_pfn_clean(pfn);
7098 
7099 	/* The instructions are well-emulated on direct mmu. */
7100 	if (vcpu->arch.mmu->direct_map) {
7101 		unsigned int indirect_shadow_pages;
7102 
7103 		spin_lock(&vcpu->kvm->mmu_lock);
7104 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7105 		spin_unlock(&vcpu->kvm->mmu_lock);
7106 
7107 		if (indirect_shadow_pages)
7108 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7109 
7110 		return true;
7111 	}
7112 
7113 	/*
7114 	 * if emulation was due to access to shadowed page table
7115 	 * and it failed try to unshadow page and re-enter the
7116 	 * guest to let CPU execute the instruction.
7117 	 */
7118 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7119 
7120 	/*
7121 	 * If the access faults on its page table, it can not
7122 	 * be fixed by unprotecting shadow page and it should
7123 	 * be reported to userspace.
7124 	 */
7125 	return !write_fault_to_shadow_pgtable;
7126 }
7127 
7128 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7129 			      gpa_t cr2_or_gpa,  int emulation_type)
7130 {
7131 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7132 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7133 
7134 	last_retry_eip = vcpu->arch.last_retry_eip;
7135 	last_retry_addr = vcpu->arch.last_retry_addr;
7136 
7137 	/*
7138 	 * If the emulation is caused by #PF and it is non-page_table
7139 	 * writing instruction, it means the VM-EXIT is caused by shadow
7140 	 * page protected, we can zap the shadow page and retry this
7141 	 * instruction directly.
7142 	 *
7143 	 * Note: if the guest uses a non-page-table modifying instruction
7144 	 * on the PDE that points to the instruction, then we will unmap
7145 	 * the instruction and go to an infinite loop. So, we cache the
7146 	 * last retried eip and the last fault address, if we meet the eip
7147 	 * and the address again, we can break out of the potential infinite
7148 	 * loop.
7149 	 */
7150 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7151 
7152 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7153 		return false;
7154 
7155 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7156 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7157 		return false;
7158 
7159 	if (x86_page_table_writing_insn(ctxt))
7160 		return false;
7161 
7162 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7163 		return false;
7164 
7165 	vcpu->arch.last_retry_eip = ctxt->eip;
7166 	vcpu->arch.last_retry_addr = cr2_or_gpa;
7167 
7168 	if (!vcpu->arch.mmu->direct_map)
7169 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7170 
7171 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7172 
7173 	return true;
7174 }
7175 
7176 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7177 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7178 
7179 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
7180 {
7181 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
7182 		/* This is a good place to trace that we are exiting SMM.  */
7183 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
7184 
7185 		/* Process a latched INIT or SMI, if any.  */
7186 		kvm_make_request(KVM_REQ_EVENT, vcpu);
7187 	}
7188 
7189 	kvm_mmu_reset_context(vcpu);
7190 }
7191 
7192 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7193 				unsigned long *db)
7194 {
7195 	u32 dr6 = 0;
7196 	int i;
7197 	u32 enable, rwlen;
7198 
7199 	enable = dr7;
7200 	rwlen = dr7 >> 16;
7201 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7202 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7203 			dr6 |= (1 << i);
7204 	return dr6;
7205 }
7206 
7207 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7208 {
7209 	struct kvm_run *kvm_run = vcpu->run;
7210 
7211 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7212 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
7213 		kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7214 		kvm_run->debug.arch.exception = DB_VECTOR;
7215 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
7216 		return 0;
7217 	}
7218 	kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7219 	return 1;
7220 }
7221 
7222 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7223 {
7224 	unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
7225 	int r;
7226 
7227 	r = kvm_x86_ops.skip_emulated_instruction(vcpu);
7228 	if (unlikely(!r))
7229 		return 0;
7230 
7231 	/*
7232 	 * rflags is the old, "raw" value of the flags.  The new value has
7233 	 * not been saved yet.
7234 	 *
7235 	 * This is correct even for TF set by the guest, because "the
7236 	 * processor will not generate this exception after the instruction
7237 	 * that sets the TF flag".
7238 	 */
7239 	if (unlikely(rflags & X86_EFLAGS_TF))
7240 		r = kvm_vcpu_do_singlestep(vcpu);
7241 	return r;
7242 }
7243 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7244 
7245 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7246 {
7247 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7248 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7249 		struct kvm_run *kvm_run = vcpu->run;
7250 		unsigned long eip = kvm_get_linear_rip(vcpu);
7251 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7252 					   vcpu->arch.guest_debug_dr7,
7253 					   vcpu->arch.eff_db);
7254 
7255 		if (dr6 != 0) {
7256 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
7257 			kvm_run->debug.arch.pc = eip;
7258 			kvm_run->debug.arch.exception = DB_VECTOR;
7259 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
7260 			*r = 0;
7261 			return true;
7262 		}
7263 	}
7264 
7265 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7266 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7267 		unsigned long eip = kvm_get_linear_rip(vcpu);
7268 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7269 					   vcpu->arch.dr7,
7270 					   vcpu->arch.db);
7271 
7272 		if (dr6 != 0) {
7273 			kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7274 			*r = 1;
7275 			return true;
7276 		}
7277 	}
7278 
7279 	return false;
7280 }
7281 
7282 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7283 {
7284 	switch (ctxt->opcode_len) {
7285 	case 1:
7286 		switch (ctxt->b) {
7287 		case 0xe4:	/* IN */
7288 		case 0xe5:
7289 		case 0xec:
7290 		case 0xed:
7291 		case 0xe6:	/* OUT */
7292 		case 0xe7:
7293 		case 0xee:
7294 		case 0xef:
7295 		case 0x6c:	/* INS */
7296 		case 0x6d:
7297 		case 0x6e:	/* OUTS */
7298 		case 0x6f:
7299 			return true;
7300 		}
7301 		break;
7302 	case 2:
7303 		switch (ctxt->b) {
7304 		case 0x33:	/* RDPMC */
7305 			return true;
7306 		}
7307 		break;
7308 	}
7309 
7310 	return false;
7311 }
7312 
7313 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7314 			    int emulation_type, void *insn, int insn_len)
7315 {
7316 	int r;
7317 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7318 	bool writeback = true;
7319 	bool write_fault_to_spt;
7320 
7321 	if (unlikely(!kvm_x86_ops.can_emulate_instruction(vcpu, insn, insn_len)))
7322 		return 1;
7323 
7324 	vcpu->arch.l1tf_flush_l1d = true;
7325 
7326 	/*
7327 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
7328 	 * never reused.
7329 	 */
7330 	write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7331 	vcpu->arch.write_fault_to_shadow_pgtable = false;
7332 	kvm_clear_exception_queue(vcpu);
7333 
7334 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7335 		init_emulate_ctxt(vcpu);
7336 
7337 		/*
7338 		 * We will reenter on the same instruction since
7339 		 * we do not set complete_userspace_io.  This does not
7340 		 * handle watchpoints yet, those would be handled in
7341 		 * the emulate_ops.
7342 		 */
7343 		if (!(emulation_type & EMULTYPE_SKIP) &&
7344 		    kvm_vcpu_check_breakpoint(vcpu, &r))
7345 			return r;
7346 
7347 		ctxt->interruptibility = 0;
7348 		ctxt->have_exception = false;
7349 		ctxt->exception.vector = -1;
7350 		ctxt->perm_ok = false;
7351 
7352 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
7353 
7354 		r = x86_decode_insn(ctxt, insn, insn_len);
7355 
7356 		trace_kvm_emulate_insn_start(vcpu);
7357 		++vcpu->stat.insn_emulation;
7358 		if (r != EMULATION_OK)  {
7359 			if ((emulation_type & EMULTYPE_TRAP_UD) ||
7360 			    (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7361 				kvm_queue_exception(vcpu, UD_VECTOR);
7362 				return 1;
7363 			}
7364 			if (reexecute_instruction(vcpu, cr2_or_gpa,
7365 						  write_fault_to_spt,
7366 						  emulation_type))
7367 				return 1;
7368 			if (ctxt->have_exception) {
7369 				/*
7370 				 * #UD should result in just EMULATION_FAILED, and trap-like
7371 				 * exception should not be encountered during decode.
7372 				 */
7373 				WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7374 					     exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7375 				inject_emulated_exception(vcpu);
7376 				return 1;
7377 			}
7378 			return handle_emulation_failure(vcpu, emulation_type);
7379 		}
7380 	}
7381 
7382 	if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7383 	    !is_vmware_backdoor_opcode(ctxt)) {
7384 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7385 		return 1;
7386 	}
7387 
7388 	/*
7389 	 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7390 	 * for kvm_skip_emulated_instruction().  The caller is responsible for
7391 	 * updating interruptibility state and injecting single-step #DBs.
7392 	 */
7393 	if (emulation_type & EMULTYPE_SKIP) {
7394 		kvm_rip_write(vcpu, ctxt->_eip);
7395 		if (ctxt->eflags & X86_EFLAGS_RF)
7396 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7397 		return 1;
7398 	}
7399 
7400 	if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7401 		return 1;
7402 
7403 	/* this is needed for vmware backdoor interface to work since it
7404 	   changes registers values  during IO operation */
7405 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7406 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7407 		emulator_invalidate_register_cache(ctxt);
7408 	}
7409 
7410 restart:
7411 	if (emulation_type & EMULTYPE_PF) {
7412 		/* Save the faulting GPA (cr2) in the address field */
7413 		ctxt->exception.address = cr2_or_gpa;
7414 
7415 		/* With shadow page tables, cr2 contains a GVA or nGPA. */
7416 		if (vcpu->arch.mmu->direct_map) {
7417 			ctxt->gpa_available = true;
7418 			ctxt->gpa_val = cr2_or_gpa;
7419 		}
7420 	} else {
7421 		/* Sanitize the address out of an abundance of paranoia. */
7422 		ctxt->exception.address = 0;
7423 	}
7424 
7425 	r = x86_emulate_insn(ctxt);
7426 
7427 	if (r == EMULATION_INTERCEPTED)
7428 		return 1;
7429 
7430 	if (r == EMULATION_FAILED) {
7431 		if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7432 					emulation_type))
7433 			return 1;
7434 
7435 		return handle_emulation_failure(vcpu, emulation_type);
7436 	}
7437 
7438 	if (ctxt->have_exception) {
7439 		r = 1;
7440 		if (inject_emulated_exception(vcpu))
7441 			return r;
7442 	} else if (vcpu->arch.pio.count) {
7443 		if (!vcpu->arch.pio.in) {
7444 			/* FIXME: return into emulator if single-stepping.  */
7445 			vcpu->arch.pio.count = 0;
7446 		} else {
7447 			writeback = false;
7448 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
7449 		}
7450 		r = 0;
7451 	} else if (vcpu->mmio_needed) {
7452 		++vcpu->stat.mmio_exits;
7453 
7454 		if (!vcpu->mmio_is_write)
7455 			writeback = false;
7456 		r = 0;
7457 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7458 	} else if (r == EMULATION_RESTART)
7459 		goto restart;
7460 	else
7461 		r = 1;
7462 
7463 	if (writeback) {
7464 		unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
7465 		toggle_interruptibility(vcpu, ctxt->interruptibility);
7466 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7467 		if (!ctxt->have_exception ||
7468 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7469 			kvm_rip_write(vcpu, ctxt->eip);
7470 			if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7471 				r = kvm_vcpu_do_singlestep(vcpu);
7472 			if (kvm_x86_ops.update_emulated_instruction)
7473 				kvm_x86_ops.update_emulated_instruction(vcpu);
7474 			__kvm_set_rflags(vcpu, ctxt->eflags);
7475 		}
7476 
7477 		/*
7478 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7479 		 * do nothing, and it will be requested again as soon as
7480 		 * the shadow expires.  But we still need to check here,
7481 		 * because POPF has no interrupt shadow.
7482 		 */
7483 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7484 			kvm_make_request(KVM_REQ_EVENT, vcpu);
7485 	} else
7486 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7487 
7488 	return r;
7489 }
7490 
7491 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7492 {
7493 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7494 }
7495 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7496 
7497 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7498 					void *insn, int insn_len)
7499 {
7500 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7501 }
7502 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7503 
7504 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7505 {
7506 	vcpu->arch.pio.count = 0;
7507 	return 1;
7508 }
7509 
7510 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7511 {
7512 	vcpu->arch.pio.count = 0;
7513 
7514 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7515 		return 1;
7516 
7517 	return kvm_skip_emulated_instruction(vcpu);
7518 }
7519 
7520 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7521 			    unsigned short port)
7522 {
7523 	unsigned long val = kvm_rax_read(vcpu);
7524 	int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7525 
7526 	if (ret)
7527 		return ret;
7528 
7529 	/*
7530 	 * Workaround userspace that relies on old KVM behavior of %rip being
7531 	 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7532 	 */
7533 	if (port == 0x7e &&
7534 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
7535 		vcpu->arch.complete_userspace_io =
7536 			complete_fast_pio_out_port_0x7e;
7537 		kvm_skip_emulated_instruction(vcpu);
7538 	} else {
7539 		vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7540 		vcpu->arch.complete_userspace_io = complete_fast_pio_out;
7541 	}
7542 	return 0;
7543 }
7544 
7545 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7546 {
7547 	unsigned long val;
7548 
7549 	/* We should only ever be called with arch.pio.count equal to 1 */
7550 	BUG_ON(vcpu->arch.pio.count != 1);
7551 
7552 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7553 		vcpu->arch.pio.count = 0;
7554 		return 1;
7555 	}
7556 
7557 	/* For size less than 4 we merge, else we zero extend */
7558 	val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7559 
7560 	/*
7561 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7562 	 * the copy and tracing
7563 	 */
7564 	emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7565 	kvm_rax_write(vcpu, val);
7566 
7567 	return kvm_skip_emulated_instruction(vcpu);
7568 }
7569 
7570 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7571 			   unsigned short port)
7572 {
7573 	unsigned long val;
7574 	int ret;
7575 
7576 	/* For size less than 4 we merge, else we zero extend */
7577 	val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7578 
7579 	ret = emulator_pio_in(vcpu, size, port, &val, 1);
7580 	if (ret) {
7581 		kvm_rax_write(vcpu, val);
7582 		return ret;
7583 	}
7584 
7585 	vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7586 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7587 
7588 	return 0;
7589 }
7590 
7591 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7592 {
7593 	int ret;
7594 
7595 	if (in)
7596 		ret = kvm_fast_pio_in(vcpu, size, port);
7597 	else
7598 		ret = kvm_fast_pio_out(vcpu, size, port);
7599 	return ret && kvm_skip_emulated_instruction(vcpu);
7600 }
7601 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7602 
7603 static int kvmclock_cpu_down_prep(unsigned int cpu)
7604 {
7605 	__this_cpu_write(cpu_tsc_khz, 0);
7606 	return 0;
7607 }
7608 
7609 static void tsc_khz_changed(void *data)
7610 {
7611 	struct cpufreq_freqs *freq = data;
7612 	unsigned long khz = 0;
7613 
7614 	if (data)
7615 		khz = freq->new;
7616 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7617 		khz = cpufreq_quick_get(raw_smp_processor_id());
7618 	if (!khz)
7619 		khz = tsc_khz;
7620 	__this_cpu_write(cpu_tsc_khz, khz);
7621 }
7622 
7623 #ifdef CONFIG_X86_64
7624 static void kvm_hyperv_tsc_notifier(void)
7625 {
7626 	struct kvm *kvm;
7627 	struct kvm_vcpu *vcpu;
7628 	int cpu;
7629 
7630 	mutex_lock(&kvm_lock);
7631 	list_for_each_entry(kvm, &vm_list, vm_list)
7632 		kvm_make_mclock_inprogress_request(kvm);
7633 
7634 	hyperv_stop_tsc_emulation();
7635 
7636 	/* TSC frequency always matches when on Hyper-V */
7637 	for_each_present_cpu(cpu)
7638 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7639 	kvm_max_guest_tsc_khz = tsc_khz;
7640 
7641 	list_for_each_entry(kvm, &vm_list, vm_list) {
7642 		struct kvm_arch *ka = &kvm->arch;
7643 
7644 		spin_lock(&ka->pvclock_gtod_sync_lock);
7645 
7646 		pvclock_update_vm_gtod_copy(kvm);
7647 
7648 		kvm_for_each_vcpu(cpu, vcpu, kvm)
7649 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7650 
7651 		kvm_for_each_vcpu(cpu, vcpu, kvm)
7652 			kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7653 
7654 		spin_unlock(&ka->pvclock_gtod_sync_lock);
7655 	}
7656 	mutex_unlock(&kvm_lock);
7657 }
7658 #endif
7659 
7660 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7661 {
7662 	struct kvm *kvm;
7663 	struct kvm_vcpu *vcpu;
7664 	int i, send_ipi = 0;
7665 
7666 	/*
7667 	 * We allow guests to temporarily run on slowing clocks,
7668 	 * provided we notify them after, or to run on accelerating
7669 	 * clocks, provided we notify them before.  Thus time never
7670 	 * goes backwards.
7671 	 *
7672 	 * However, we have a problem.  We can't atomically update
7673 	 * the frequency of a given CPU from this function; it is
7674 	 * merely a notifier, which can be called from any CPU.
7675 	 * Changing the TSC frequency at arbitrary points in time
7676 	 * requires a recomputation of local variables related to
7677 	 * the TSC for each VCPU.  We must flag these local variables
7678 	 * to be updated and be sure the update takes place with the
7679 	 * new frequency before any guests proceed.
7680 	 *
7681 	 * Unfortunately, the combination of hotplug CPU and frequency
7682 	 * change creates an intractable locking scenario; the order
7683 	 * of when these callouts happen is undefined with respect to
7684 	 * CPU hotplug, and they can race with each other.  As such,
7685 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7686 	 * undefined; you can actually have a CPU frequency change take
7687 	 * place in between the computation of X and the setting of the
7688 	 * variable.  To protect against this problem, all updates of
7689 	 * the per_cpu tsc_khz variable are done in an interrupt
7690 	 * protected IPI, and all callers wishing to update the value
7691 	 * must wait for a synchronous IPI to complete (which is trivial
7692 	 * if the caller is on the CPU already).  This establishes the
7693 	 * necessary total order on variable updates.
7694 	 *
7695 	 * Note that because a guest time update may take place
7696 	 * anytime after the setting of the VCPU's request bit, the
7697 	 * correct TSC value must be set before the request.  However,
7698 	 * to ensure the update actually makes it to any guest which
7699 	 * starts running in hardware virtualization between the set
7700 	 * and the acquisition of the spinlock, we must also ping the
7701 	 * CPU after setting the request bit.
7702 	 *
7703 	 */
7704 
7705 	smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7706 
7707 	mutex_lock(&kvm_lock);
7708 	list_for_each_entry(kvm, &vm_list, vm_list) {
7709 		kvm_for_each_vcpu(i, vcpu, kvm) {
7710 			if (vcpu->cpu != cpu)
7711 				continue;
7712 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7713 			if (vcpu->cpu != raw_smp_processor_id())
7714 				send_ipi = 1;
7715 		}
7716 	}
7717 	mutex_unlock(&kvm_lock);
7718 
7719 	if (freq->old < freq->new && send_ipi) {
7720 		/*
7721 		 * We upscale the frequency.  Must make the guest
7722 		 * doesn't see old kvmclock values while running with
7723 		 * the new frequency, otherwise we risk the guest sees
7724 		 * time go backwards.
7725 		 *
7726 		 * In case we update the frequency for another cpu
7727 		 * (which might be in guest context) send an interrupt
7728 		 * to kick the cpu out of guest context.  Next time
7729 		 * guest context is entered kvmclock will be updated,
7730 		 * so the guest will not see stale values.
7731 		 */
7732 		smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7733 	}
7734 }
7735 
7736 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7737 				     void *data)
7738 {
7739 	struct cpufreq_freqs *freq = data;
7740 	int cpu;
7741 
7742 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7743 		return 0;
7744 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7745 		return 0;
7746 
7747 	for_each_cpu(cpu, freq->policy->cpus)
7748 		__kvmclock_cpufreq_notifier(freq, cpu);
7749 
7750 	return 0;
7751 }
7752 
7753 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7754 	.notifier_call  = kvmclock_cpufreq_notifier
7755 };
7756 
7757 static int kvmclock_cpu_online(unsigned int cpu)
7758 {
7759 	tsc_khz_changed(NULL);
7760 	return 0;
7761 }
7762 
7763 static void kvm_timer_init(void)
7764 {
7765 	max_tsc_khz = tsc_khz;
7766 
7767 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7768 #ifdef CONFIG_CPU_FREQ
7769 		struct cpufreq_policy *policy;
7770 		int cpu;
7771 
7772 		cpu = get_cpu();
7773 		policy = cpufreq_cpu_get(cpu);
7774 		if (policy) {
7775 			if (policy->cpuinfo.max_freq)
7776 				max_tsc_khz = policy->cpuinfo.max_freq;
7777 			cpufreq_cpu_put(policy);
7778 		}
7779 		put_cpu();
7780 #endif
7781 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7782 					  CPUFREQ_TRANSITION_NOTIFIER);
7783 	}
7784 
7785 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7786 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
7787 }
7788 
7789 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7790 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7791 
7792 int kvm_is_in_guest(void)
7793 {
7794 	return __this_cpu_read(current_vcpu) != NULL;
7795 }
7796 
7797 static int kvm_is_user_mode(void)
7798 {
7799 	int user_mode = 3;
7800 
7801 	if (__this_cpu_read(current_vcpu))
7802 		user_mode = kvm_x86_ops.get_cpl(__this_cpu_read(current_vcpu));
7803 
7804 	return user_mode != 0;
7805 }
7806 
7807 static unsigned long kvm_get_guest_ip(void)
7808 {
7809 	unsigned long ip = 0;
7810 
7811 	if (__this_cpu_read(current_vcpu))
7812 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7813 
7814 	return ip;
7815 }
7816 
7817 static void kvm_handle_intel_pt_intr(void)
7818 {
7819 	struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7820 
7821 	kvm_make_request(KVM_REQ_PMI, vcpu);
7822 	__set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7823 			(unsigned long *)&vcpu->arch.pmu.global_status);
7824 }
7825 
7826 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7827 	.is_in_guest		= kvm_is_in_guest,
7828 	.is_user_mode		= kvm_is_user_mode,
7829 	.get_guest_ip		= kvm_get_guest_ip,
7830 	.handle_intel_pt_intr	= kvm_handle_intel_pt_intr,
7831 };
7832 
7833 #ifdef CONFIG_X86_64
7834 static void pvclock_gtod_update_fn(struct work_struct *work)
7835 {
7836 	struct kvm *kvm;
7837 
7838 	struct kvm_vcpu *vcpu;
7839 	int i;
7840 
7841 	mutex_lock(&kvm_lock);
7842 	list_for_each_entry(kvm, &vm_list, vm_list)
7843 		kvm_for_each_vcpu(i, vcpu, kvm)
7844 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7845 	atomic_set(&kvm_guest_has_master_clock, 0);
7846 	mutex_unlock(&kvm_lock);
7847 }
7848 
7849 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7850 
7851 /*
7852  * Notification about pvclock gtod data update.
7853  */
7854 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7855 			       void *priv)
7856 {
7857 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7858 	struct timekeeper *tk = priv;
7859 
7860 	update_pvclock_gtod(tk);
7861 
7862 	/* disable master clock if host does not trust, or does not
7863 	 * use, TSC based clocksource.
7864 	 */
7865 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7866 	    atomic_read(&kvm_guest_has_master_clock) != 0)
7867 		queue_work(system_long_wq, &pvclock_gtod_work);
7868 
7869 	return 0;
7870 }
7871 
7872 static struct notifier_block pvclock_gtod_notifier = {
7873 	.notifier_call = pvclock_gtod_notify,
7874 };
7875 #endif
7876 
7877 int kvm_arch_init(void *opaque)
7878 {
7879 	struct kvm_x86_init_ops *ops = opaque;
7880 	int r;
7881 
7882 	if (kvm_x86_ops.hardware_enable) {
7883 		printk(KERN_ERR "kvm: already loaded the other module\n");
7884 		r = -EEXIST;
7885 		goto out;
7886 	}
7887 
7888 	if (!ops->cpu_has_kvm_support()) {
7889 		pr_err_ratelimited("kvm: no hardware support\n");
7890 		r = -EOPNOTSUPP;
7891 		goto out;
7892 	}
7893 	if (ops->disabled_by_bios()) {
7894 		pr_err_ratelimited("kvm: disabled by bios\n");
7895 		r = -EOPNOTSUPP;
7896 		goto out;
7897 	}
7898 
7899 	/*
7900 	 * KVM explicitly assumes that the guest has an FPU and
7901 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7902 	 * vCPU's FPU state as a fxregs_state struct.
7903 	 */
7904 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7905 		printk(KERN_ERR "kvm: inadequate fpu\n");
7906 		r = -EOPNOTSUPP;
7907 		goto out;
7908 	}
7909 
7910 	r = -ENOMEM;
7911 	x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7912 					  __alignof__(struct fpu), SLAB_ACCOUNT,
7913 					  NULL);
7914 	if (!x86_fpu_cache) {
7915 		printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7916 		goto out;
7917 	}
7918 
7919 	x86_emulator_cache = kvm_alloc_emulator_cache();
7920 	if (!x86_emulator_cache) {
7921 		pr_err("kvm: failed to allocate cache for x86 emulator\n");
7922 		goto out_free_x86_fpu_cache;
7923 	}
7924 
7925 	user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
7926 	if (!user_return_msrs) {
7927 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
7928 		goto out_free_x86_emulator_cache;
7929 	}
7930 
7931 	r = kvm_mmu_module_init();
7932 	if (r)
7933 		goto out_free_percpu;
7934 
7935 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7936 			PT_DIRTY_MASK, PT64_NX_MASK, 0,
7937 			PT_PRESENT_MASK, 0, sme_me_mask);
7938 	kvm_timer_init();
7939 
7940 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
7941 
7942 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
7943 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7944 		supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
7945 	}
7946 
7947 	kvm_lapic_init();
7948 	if (pi_inject_timer == -1)
7949 		pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7950 #ifdef CONFIG_X86_64
7951 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7952 
7953 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7954 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7955 #endif
7956 
7957 	return 0;
7958 
7959 out_free_percpu:
7960 	free_percpu(user_return_msrs);
7961 out_free_x86_emulator_cache:
7962 	kmem_cache_destroy(x86_emulator_cache);
7963 out_free_x86_fpu_cache:
7964 	kmem_cache_destroy(x86_fpu_cache);
7965 out:
7966 	return r;
7967 }
7968 
7969 void kvm_arch_exit(void)
7970 {
7971 #ifdef CONFIG_X86_64
7972 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7973 		clear_hv_tscchange_cb();
7974 #endif
7975 	kvm_lapic_exit();
7976 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7977 
7978 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7979 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7980 					    CPUFREQ_TRANSITION_NOTIFIER);
7981 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7982 #ifdef CONFIG_X86_64
7983 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7984 #endif
7985 	kvm_x86_ops.hardware_enable = NULL;
7986 	kvm_mmu_module_exit();
7987 	free_percpu(user_return_msrs);
7988 	kmem_cache_destroy(x86_fpu_cache);
7989 }
7990 
7991 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
7992 {
7993 	++vcpu->stat.halt_exits;
7994 	if (lapic_in_kernel(vcpu)) {
7995 		vcpu->arch.mp_state = state;
7996 		return 1;
7997 	} else {
7998 		vcpu->run->exit_reason = reason;
7999 		return 0;
8000 	}
8001 }
8002 
8003 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8004 {
8005 	return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8006 }
8007 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8008 
8009 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8010 {
8011 	int ret = kvm_skip_emulated_instruction(vcpu);
8012 	/*
8013 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8014 	 * KVM_EXIT_DEBUG here.
8015 	 */
8016 	return kvm_vcpu_halt(vcpu) && ret;
8017 }
8018 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8019 
8020 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8021 {
8022 	int ret = kvm_skip_emulated_instruction(vcpu);
8023 
8024 	return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8025 }
8026 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8027 
8028 #ifdef CONFIG_X86_64
8029 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8030 			        unsigned long clock_type)
8031 {
8032 	struct kvm_clock_pairing clock_pairing;
8033 	struct timespec64 ts;
8034 	u64 cycle;
8035 	int ret;
8036 
8037 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8038 		return -KVM_EOPNOTSUPP;
8039 
8040 	if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
8041 		return -KVM_EOPNOTSUPP;
8042 
8043 	clock_pairing.sec = ts.tv_sec;
8044 	clock_pairing.nsec = ts.tv_nsec;
8045 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8046 	clock_pairing.flags = 0;
8047 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8048 
8049 	ret = 0;
8050 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8051 			    sizeof(struct kvm_clock_pairing)))
8052 		ret = -KVM_EFAULT;
8053 
8054 	return ret;
8055 }
8056 #endif
8057 
8058 /*
8059  * kvm_pv_kick_cpu_op:  Kick a vcpu.
8060  *
8061  * @apicid - apicid of vcpu to be kicked.
8062  */
8063 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8064 {
8065 	struct kvm_lapic_irq lapic_irq;
8066 
8067 	lapic_irq.shorthand = APIC_DEST_NOSHORT;
8068 	lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8069 	lapic_irq.level = 0;
8070 	lapic_irq.dest_id = apicid;
8071 	lapic_irq.msi_redir_hint = false;
8072 
8073 	lapic_irq.delivery_mode = APIC_DM_REMRD;
8074 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8075 }
8076 
8077 bool kvm_apicv_activated(struct kvm *kvm)
8078 {
8079 	return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8080 }
8081 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8082 
8083 void kvm_apicv_init(struct kvm *kvm, bool enable)
8084 {
8085 	if (enable)
8086 		clear_bit(APICV_INHIBIT_REASON_DISABLE,
8087 			  &kvm->arch.apicv_inhibit_reasons);
8088 	else
8089 		set_bit(APICV_INHIBIT_REASON_DISABLE,
8090 			&kvm->arch.apicv_inhibit_reasons);
8091 }
8092 EXPORT_SYMBOL_GPL(kvm_apicv_init);
8093 
8094 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
8095 {
8096 	struct kvm_vcpu *target = NULL;
8097 	struct kvm_apic_map *map;
8098 
8099 	rcu_read_lock();
8100 	map = rcu_dereference(kvm->arch.apic_map);
8101 
8102 	if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8103 		target = map->phys_map[dest_id]->vcpu;
8104 
8105 	rcu_read_unlock();
8106 
8107 	if (target && READ_ONCE(target->ready))
8108 		kvm_vcpu_yield_to(target);
8109 }
8110 
8111 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8112 {
8113 	unsigned long nr, a0, a1, a2, a3, ret;
8114 	int op_64_bit;
8115 
8116 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
8117 		return kvm_hv_hypercall(vcpu);
8118 
8119 	nr = kvm_rax_read(vcpu);
8120 	a0 = kvm_rbx_read(vcpu);
8121 	a1 = kvm_rcx_read(vcpu);
8122 	a2 = kvm_rdx_read(vcpu);
8123 	a3 = kvm_rsi_read(vcpu);
8124 
8125 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
8126 
8127 	op_64_bit = is_64_bit_mode(vcpu);
8128 	if (!op_64_bit) {
8129 		nr &= 0xFFFFFFFF;
8130 		a0 &= 0xFFFFFFFF;
8131 		a1 &= 0xFFFFFFFF;
8132 		a2 &= 0xFFFFFFFF;
8133 		a3 &= 0xFFFFFFFF;
8134 	}
8135 
8136 	if (kvm_x86_ops.get_cpl(vcpu) != 0) {
8137 		ret = -KVM_EPERM;
8138 		goto out;
8139 	}
8140 
8141 	ret = -KVM_ENOSYS;
8142 
8143 	switch (nr) {
8144 	case KVM_HC_VAPIC_POLL_IRQ:
8145 		ret = 0;
8146 		break;
8147 	case KVM_HC_KICK_CPU:
8148 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8149 			break;
8150 
8151 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8152 		kvm_sched_yield(vcpu->kvm, a1);
8153 		ret = 0;
8154 		break;
8155 #ifdef CONFIG_X86_64
8156 	case KVM_HC_CLOCK_PAIRING:
8157 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8158 		break;
8159 #endif
8160 	case KVM_HC_SEND_IPI:
8161 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8162 			break;
8163 
8164 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8165 		break;
8166 	case KVM_HC_SCHED_YIELD:
8167 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8168 			break;
8169 
8170 		kvm_sched_yield(vcpu->kvm, a0);
8171 		ret = 0;
8172 		break;
8173 	default:
8174 		ret = -KVM_ENOSYS;
8175 		break;
8176 	}
8177 out:
8178 	if (!op_64_bit)
8179 		ret = (u32)ret;
8180 	kvm_rax_write(vcpu, ret);
8181 
8182 	++vcpu->stat.hypercalls;
8183 	return kvm_skip_emulated_instruction(vcpu);
8184 }
8185 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8186 
8187 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8188 {
8189 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8190 	char instruction[3];
8191 	unsigned long rip = kvm_rip_read(vcpu);
8192 
8193 	kvm_x86_ops.patch_hypercall(vcpu, instruction);
8194 
8195 	return emulator_write_emulated(ctxt, rip, instruction, 3,
8196 		&ctxt->exception);
8197 }
8198 
8199 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8200 {
8201 	return vcpu->run->request_interrupt_window &&
8202 		likely(!pic_in_kernel(vcpu->kvm));
8203 }
8204 
8205 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8206 {
8207 	struct kvm_run *kvm_run = vcpu->run;
8208 
8209 	/*
8210 	 * if_flag is obsolete and useless, so do not bother
8211 	 * setting it for SEV-ES guests.  Userspace can just
8212 	 * use kvm_run->ready_for_interrupt_injection.
8213 	 */
8214 	kvm_run->if_flag = !vcpu->arch.guest_state_protected
8215 		&& (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8216 
8217 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
8218 	kvm_run->cr8 = kvm_get_cr8(vcpu);
8219 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
8220 	kvm_run->ready_for_interrupt_injection =
8221 		pic_in_kernel(vcpu->kvm) ||
8222 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
8223 }
8224 
8225 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8226 {
8227 	int max_irr, tpr;
8228 
8229 	if (!kvm_x86_ops.update_cr8_intercept)
8230 		return;
8231 
8232 	if (!lapic_in_kernel(vcpu))
8233 		return;
8234 
8235 	if (vcpu->arch.apicv_active)
8236 		return;
8237 
8238 	if (!vcpu->arch.apic->vapic_addr)
8239 		max_irr = kvm_lapic_find_highest_irr(vcpu);
8240 	else
8241 		max_irr = -1;
8242 
8243 	if (max_irr != -1)
8244 		max_irr >>= 4;
8245 
8246 	tpr = kvm_lapic_get_cr8(vcpu);
8247 
8248 	kvm_x86_ops.update_cr8_intercept(vcpu, tpr, max_irr);
8249 }
8250 
8251 static void inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8252 {
8253 	int r;
8254 	bool can_inject = true;
8255 
8256 	/* try to reinject previous events if any */
8257 
8258 	if (vcpu->arch.exception.injected) {
8259 		kvm_x86_ops.queue_exception(vcpu);
8260 		can_inject = false;
8261 	}
8262 	/*
8263 	 * Do not inject an NMI or interrupt if there is a pending
8264 	 * exception.  Exceptions and interrupts are recognized at
8265 	 * instruction boundaries, i.e. the start of an instruction.
8266 	 * Trap-like exceptions, e.g. #DB, have higher priority than
8267 	 * NMIs and interrupts, i.e. traps are recognized before an
8268 	 * NMI/interrupt that's pending on the same instruction.
8269 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8270 	 * priority, but are only generated (pended) during instruction
8271 	 * execution, i.e. a pending fault-like exception means the
8272 	 * fault occurred on the *previous* instruction and must be
8273 	 * serviced prior to recognizing any new events in order to
8274 	 * fully complete the previous instruction.
8275 	 */
8276 	else if (!vcpu->arch.exception.pending) {
8277 		if (vcpu->arch.nmi_injected) {
8278 			kvm_x86_ops.set_nmi(vcpu);
8279 			can_inject = false;
8280 		} else if (vcpu->arch.interrupt.injected) {
8281 			kvm_x86_ops.set_irq(vcpu);
8282 			can_inject = false;
8283 		}
8284 	}
8285 
8286 	WARN_ON_ONCE(vcpu->arch.exception.injected &&
8287 		     vcpu->arch.exception.pending);
8288 
8289 	/*
8290 	 * Call check_nested_events() even if we reinjected a previous event
8291 	 * in order for caller to determine if it should require immediate-exit
8292 	 * from L2 to L1 due to pending L1 events which require exit
8293 	 * from L2 to L1.
8294 	 */
8295 	if (is_guest_mode(vcpu)) {
8296 		r = kvm_x86_ops.nested_ops->check_events(vcpu);
8297 		if (r < 0)
8298 			goto busy;
8299 	}
8300 
8301 	/* try to inject new event if pending */
8302 	if (vcpu->arch.exception.pending) {
8303 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
8304 					vcpu->arch.exception.has_error_code,
8305 					vcpu->arch.exception.error_code);
8306 
8307 		vcpu->arch.exception.pending = false;
8308 		vcpu->arch.exception.injected = true;
8309 
8310 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8311 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8312 					     X86_EFLAGS_RF);
8313 
8314 		if (vcpu->arch.exception.nr == DB_VECTOR) {
8315 			kvm_deliver_exception_payload(vcpu);
8316 			if (vcpu->arch.dr7 & DR7_GD) {
8317 				vcpu->arch.dr7 &= ~DR7_GD;
8318 				kvm_update_dr7(vcpu);
8319 			}
8320 		}
8321 
8322 		kvm_x86_ops.queue_exception(vcpu);
8323 		can_inject = false;
8324 	}
8325 
8326 	/*
8327 	 * Finally, inject interrupt events.  If an event cannot be injected
8328 	 * due to architectural conditions (e.g. IF=0) a window-open exit
8329 	 * will re-request KVM_REQ_EVENT.  Sometimes however an event is pending
8330 	 * and can architecturally be injected, but we cannot do it right now:
8331 	 * an interrupt could have arrived just now and we have to inject it
8332 	 * as a vmexit, or there could already an event in the queue, which is
8333 	 * indicated by can_inject.  In that case we request an immediate exit
8334 	 * in order to make progress and get back here for another iteration.
8335 	 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8336 	 */
8337 	if (vcpu->arch.smi_pending) {
8338 		r = can_inject ? kvm_x86_ops.smi_allowed(vcpu, true) : -EBUSY;
8339 		if (r < 0)
8340 			goto busy;
8341 		if (r) {
8342 			vcpu->arch.smi_pending = false;
8343 			++vcpu->arch.smi_count;
8344 			enter_smm(vcpu);
8345 			can_inject = false;
8346 		} else
8347 			kvm_x86_ops.enable_smi_window(vcpu);
8348 	}
8349 
8350 	if (vcpu->arch.nmi_pending) {
8351 		r = can_inject ? kvm_x86_ops.nmi_allowed(vcpu, true) : -EBUSY;
8352 		if (r < 0)
8353 			goto busy;
8354 		if (r) {
8355 			--vcpu->arch.nmi_pending;
8356 			vcpu->arch.nmi_injected = true;
8357 			kvm_x86_ops.set_nmi(vcpu);
8358 			can_inject = false;
8359 			WARN_ON(kvm_x86_ops.nmi_allowed(vcpu, true) < 0);
8360 		}
8361 		if (vcpu->arch.nmi_pending)
8362 			kvm_x86_ops.enable_nmi_window(vcpu);
8363 	}
8364 
8365 	if (kvm_cpu_has_injectable_intr(vcpu)) {
8366 		r = can_inject ? kvm_x86_ops.interrupt_allowed(vcpu, true) : -EBUSY;
8367 		if (r < 0)
8368 			goto busy;
8369 		if (r) {
8370 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8371 			kvm_x86_ops.set_irq(vcpu);
8372 			WARN_ON(kvm_x86_ops.interrupt_allowed(vcpu, true) < 0);
8373 		}
8374 		if (kvm_cpu_has_injectable_intr(vcpu))
8375 			kvm_x86_ops.enable_irq_window(vcpu);
8376 	}
8377 
8378 	if (is_guest_mode(vcpu) &&
8379 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
8380 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8381 		*req_immediate_exit = true;
8382 
8383 	WARN_ON(vcpu->arch.exception.pending);
8384 	return;
8385 
8386 busy:
8387 	*req_immediate_exit = true;
8388 	return;
8389 }
8390 
8391 static void process_nmi(struct kvm_vcpu *vcpu)
8392 {
8393 	unsigned limit = 2;
8394 
8395 	/*
8396 	 * x86 is limited to one NMI running, and one NMI pending after it.
8397 	 * If an NMI is already in progress, limit further NMIs to just one.
8398 	 * Otherwise, allow two (and we'll inject the first one immediately).
8399 	 */
8400 	if (kvm_x86_ops.get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
8401 		limit = 1;
8402 
8403 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8404 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8405 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8406 }
8407 
8408 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8409 {
8410 	u32 flags = 0;
8411 	flags |= seg->g       << 23;
8412 	flags |= seg->db      << 22;
8413 	flags |= seg->l       << 21;
8414 	flags |= seg->avl     << 20;
8415 	flags |= seg->present << 15;
8416 	flags |= seg->dpl     << 13;
8417 	flags |= seg->s       << 12;
8418 	flags |= seg->type    << 8;
8419 	return flags;
8420 }
8421 
8422 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8423 {
8424 	struct kvm_segment seg;
8425 	int offset;
8426 
8427 	kvm_get_segment(vcpu, &seg, n);
8428 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
8429 
8430 	if (n < 3)
8431 		offset = 0x7f84 + n * 12;
8432 	else
8433 		offset = 0x7f2c + (n - 3) * 12;
8434 
8435 	put_smstate(u32, buf, offset + 8, seg.base);
8436 	put_smstate(u32, buf, offset + 4, seg.limit);
8437 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
8438 }
8439 
8440 #ifdef CONFIG_X86_64
8441 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
8442 {
8443 	struct kvm_segment seg;
8444 	int offset;
8445 	u16 flags;
8446 
8447 	kvm_get_segment(vcpu, &seg, n);
8448 	offset = 0x7e00 + n * 16;
8449 
8450 	flags = enter_smm_get_segment_flags(&seg) >> 8;
8451 	put_smstate(u16, buf, offset, seg.selector);
8452 	put_smstate(u16, buf, offset + 2, flags);
8453 	put_smstate(u32, buf, offset + 4, seg.limit);
8454 	put_smstate(u64, buf, offset + 8, seg.base);
8455 }
8456 #endif
8457 
8458 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
8459 {
8460 	struct desc_ptr dt;
8461 	struct kvm_segment seg;
8462 	unsigned long val;
8463 	int i;
8464 
8465 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
8466 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
8467 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
8468 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
8469 
8470 	for (i = 0; i < 8; i++)
8471 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
8472 
8473 	kvm_get_dr(vcpu, 6, &val);
8474 	put_smstate(u32, buf, 0x7fcc, (u32)val);
8475 	kvm_get_dr(vcpu, 7, &val);
8476 	put_smstate(u32, buf, 0x7fc8, (u32)val);
8477 
8478 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8479 	put_smstate(u32, buf, 0x7fc4, seg.selector);
8480 	put_smstate(u32, buf, 0x7f64, seg.base);
8481 	put_smstate(u32, buf, 0x7f60, seg.limit);
8482 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
8483 
8484 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8485 	put_smstate(u32, buf, 0x7fc0, seg.selector);
8486 	put_smstate(u32, buf, 0x7f80, seg.base);
8487 	put_smstate(u32, buf, 0x7f7c, seg.limit);
8488 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
8489 
8490 	kvm_x86_ops.get_gdt(vcpu, &dt);
8491 	put_smstate(u32, buf, 0x7f74, dt.address);
8492 	put_smstate(u32, buf, 0x7f70, dt.size);
8493 
8494 	kvm_x86_ops.get_idt(vcpu, &dt);
8495 	put_smstate(u32, buf, 0x7f58, dt.address);
8496 	put_smstate(u32, buf, 0x7f54, dt.size);
8497 
8498 	for (i = 0; i < 6; i++)
8499 		enter_smm_save_seg_32(vcpu, buf, i);
8500 
8501 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
8502 
8503 	/* revision id */
8504 	put_smstate(u32, buf, 0x7efc, 0x00020000);
8505 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
8506 }
8507 
8508 #ifdef CONFIG_X86_64
8509 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
8510 {
8511 	struct desc_ptr dt;
8512 	struct kvm_segment seg;
8513 	unsigned long val;
8514 	int i;
8515 
8516 	for (i = 0; i < 16; i++)
8517 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
8518 
8519 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
8520 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
8521 
8522 	kvm_get_dr(vcpu, 6, &val);
8523 	put_smstate(u64, buf, 0x7f68, val);
8524 	kvm_get_dr(vcpu, 7, &val);
8525 	put_smstate(u64, buf, 0x7f60, val);
8526 
8527 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
8528 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
8529 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
8530 
8531 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
8532 
8533 	/* revision id */
8534 	put_smstate(u32, buf, 0x7efc, 0x00020064);
8535 
8536 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
8537 
8538 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8539 	put_smstate(u16, buf, 0x7e90, seg.selector);
8540 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
8541 	put_smstate(u32, buf, 0x7e94, seg.limit);
8542 	put_smstate(u64, buf, 0x7e98, seg.base);
8543 
8544 	kvm_x86_ops.get_idt(vcpu, &dt);
8545 	put_smstate(u32, buf, 0x7e84, dt.size);
8546 	put_smstate(u64, buf, 0x7e88, dt.address);
8547 
8548 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8549 	put_smstate(u16, buf, 0x7e70, seg.selector);
8550 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
8551 	put_smstate(u32, buf, 0x7e74, seg.limit);
8552 	put_smstate(u64, buf, 0x7e78, seg.base);
8553 
8554 	kvm_x86_ops.get_gdt(vcpu, &dt);
8555 	put_smstate(u32, buf, 0x7e64, dt.size);
8556 	put_smstate(u64, buf, 0x7e68, dt.address);
8557 
8558 	for (i = 0; i < 6; i++)
8559 		enter_smm_save_seg_64(vcpu, buf, i);
8560 }
8561 #endif
8562 
8563 static void enter_smm(struct kvm_vcpu *vcpu)
8564 {
8565 	struct kvm_segment cs, ds;
8566 	struct desc_ptr dt;
8567 	char buf[512];
8568 	u32 cr0;
8569 
8570 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
8571 	memset(buf, 0, 512);
8572 #ifdef CONFIG_X86_64
8573 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8574 		enter_smm_save_state_64(vcpu, buf);
8575 	else
8576 #endif
8577 		enter_smm_save_state_32(vcpu, buf);
8578 
8579 	/*
8580 	 * Give pre_enter_smm() a chance to make ISA-specific changes to the
8581 	 * vCPU state (e.g. leave guest mode) after we've saved the state into
8582 	 * the SMM state-save area.
8583 	 */
8584 	kvm_x86_ops.pre_enter_smm(vcpu, buf);
8585 
8586 	vcpu->arch.hflags |= HF_SMM_MASK;
8587 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
8588 
8589 	if (kvm_x86_ops.get_nmi_mask(vcpu))
8590 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
8591 	else
8592 		kvm_x86_ops.set_nmi_mask(vcpu, true);
8593 
8594 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
8595 	kvm_rip_write(vcpu, 0x8000);
8596 
8597 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
8598 	kvm_x86_ops.set_cr0(vcpu, cr0);
8599 	vcpu->arch.cr0 = cr0;
8600 
8601 	kvm_x86_ops.set_cr4(vcpu, 0);
8602 
8603 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
8604 	dt.address = dt.size = 0;
8605 	kvm_x86_ops.set_idt(vcpu, &dt);
8606 
8607 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
8608 
8609 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
8610 	cs.base = vcpu->arch.smbase;
8611 
8612 	ds.selector = 0;
8613 	ds.base = 0;
8614 
8615 	cs.limit    = ds.limit = 0xffffffff;
8616 	cs.type     = ds.type = 0x3;
8617 	cs.dpl      = ds.dpl = 0;
8618 	cs.db       = ds.db = 0;
8619 	cs.s        = ds.s = 1;
8620 	cs.l        = ds.l = 0;
8621 	cs.g        = ds.g = 1;
8622 	cs.avl      = ds.avl = 0;
8623 	cs.present  = ds.present = 1;
8624 	cs.unusable = ds.unusable = 0;
8625 	cs.padding  = ds.padding = 0;
8626 
8627 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8628 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
8629 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
8630 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
8631 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
8632 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
8633 
8634 #ifdef CONFIG_X86_64
8635 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8636 		kvm_x86_ops.set_efer(vcpu, 0);
8637 #endif
8638 
8639 	kvm_update_cpuid_runtime(vcpu);
8640 	kvm_mmu_reset_context(vcpu);
8641 }
8642 
8643 static void process_smi(struct kvm_vcpu *vcpu)
8644 {
8645 	vcpu->arch.smi_pending = true;
8646 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8647 }
8648 
8649 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
8650 				       unsigned long *vcpu_bitmap)
8651 {
8652 	cpumask_var_t cpus;
8653 
8654 	zalloc_cpumask_var(&cpus, GFP_ATOMIC);
8655 
8656 	kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
8657 				    NULL, vcpu_bitmap, cpus);
8658 
8659 	free_cpumask_var(cpus);
8660 }
8661 
8662 void kvm_make_scan_ioapic_request(struct kvm *kvm)
8663 {
8664 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
8665 }
8666 
8667 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
8668 {
8669 	if (!lapic_in_kernel(vcpu))
8670 		return;
8671 
8672 	vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
8673 	kvm_apic_update_apicv(vcpu);
8674 	kvm_x86_ops.refresh_apicv_exec_ctrl(vcpu);
8675 }
8676 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
8677 
8678 /*
8679  * NOTE: Do not hold any lock prior to calling this.
8680  *
8681  * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
8682  * locked, because it calls __x86_set_memory_region() which does
8683  * synchronize_srcu(&kvm->srcu).
8684  */
8685 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
8686 {
8687 	struct kvm_vcpu *except;
8688 	unsigned long old, new, expected;
8689 
8690 	if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
8691 	    !kvm_x86_ops.check_apicv_inhibit_reasons(bit))
8692 		return;
8693 
8694 	old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
8695 	do {
8696 		expected = new = old;
8697 		if (activate)
8698 			__clear_bit(bit, &new);
8699 		else
8700 			__set_bit(bit, &new);
8701 		if (new == old)
8702 			break;
8703 		old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
8704 	} while (old != expected);
8705 
8706 	if (!!old == !!new)
8707 		return;
8708 
8709 	trace_kvm_apicv_update_request(activate, bit);
8710 	if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
8711 		kvm_x86_ops.pre_update_apicv_exec_ctrl(kvm, activate);
8712 
8713 	/*
8714 	 * Sending request to update APICV for all other vcpus,
8715 	 * while update the calling vcpu immediately instead of
8716 	 * waiting for another #VMEXIT to handle the request.
8717 	 */
8718 	except = kvm_get_running_vcpu();
8719 	kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
8720 					 except);
8721 	if (except)
8722 		kvm_vcpu_update_apicv(except);
8723 }
8724 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
8725 
8726 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
8727 {
8728 	if (!kvm_apic_present(vcpu))
8729 		return;
8730 
8731 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
8732 
8733 	if (irqchip_split(vcpu->kvm))
8734 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
8735 	else {
8736 		if (vcpu->arch.apicv_active)
8737 			kvm_x86_ops.sync_pir_to_irr(vcpu);
8738 		if (ioapic_in_kernel(vcpu->kvm))
8739 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
8740 	}
8741 
8742 	if (is_guest_mode(vcpu))
8743 		vcpu->arch.load_eoi_exitmap_pending = true;
8744 	else
8745 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
8746 }
8747 
8748 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8749 {
8750 	u64 eoi_exit_bitmap[4];
8751 
8752 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
8753 		return;
8754 
8755 	bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
8756 		  vcpu_to_synic(vcpu)->vec_bitmap, 256);
8757 	kvm_x86_ops.load_eoi_exitmap(vcpu, eoi_exit_bitmap);
8758 }
8759 
8760 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
8761 					    unsigned long start, unsigned long end)
8762 {
8763 	unsigned long apic_address;
8764 
8765 	/*
8766 	 * The physical address of apic access page is stored in the VMCS.
8767 	 * Update it when it becomes invalid.
8768 	 */
8769 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8770 	if (start <= apic_address && apic_address < end)
8771 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
8772 }
8773 
8774 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
8775 {
8776 	if (!lapic_in_kernel(vcpu))
8777 		return;
8778 
8779 	if (!kvm_x86_ops.set_apic_access_page_addr)
8780 		return;
8781 
8782 	kvm_x86_ops.set_apic_access_page_addr(vcpu);
8783 }
8784 
8785 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
8786 {
8787 	smp_send_reschedule(vcpu->cpu);
8788 }
8789 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
8790 
8791 /*
8792  * Returns 1 to let vcpu_run() continue the guest execution loop without
8793  * exiting to the userspace.  Otherwise, the value will be returned to the
8794  * userspace.
8795  */
8796 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
8797 {
8798 	int r;
8799 	bool req_int_win =
8800 		dm_request_for_irq_injection(vcpu) &&
8801 		kvm_cpu_accept_dm_intr(vcpu);
8802 	fastpath_t exit_fastpath;
8803 
8804 	bool req_immediate_exit = false;
8805 
8806 	/* Forbid vmenter if vcpu dirty ring is soft-full */
8807 	if (unlikely(vcpu->kvm->dirty_ring_size &&
8808 		     kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
8809 		vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
8810 		trace_kvm_dirty_ring_exit(vcpu);
8811 		r = 0;
8812 		goto out;
8813 	}
8814 
8815 	if (kvm_request_pending(vcpu)) {
8816 		if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
8817 			if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
8818 				r = 0;
8819 				goto out;
8820 			}
8821 		}
8822 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
8823 			kvm_mmu_unload(vcpu);
8824 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
8825 			__kvm_migrate_timers(vcpu);
8826 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
8827 			kvm_gen_update_masterclock(vcpu->kvm);
8828 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
8829 			kvm_gen_kvmclock_update(vcpu);
8830 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
8831 			r = kvm_guest_time_update(vcpu);
8832 			if (unlikely(r))
8833 				goto out;
8834 		}
8835 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
8836 			kvm_mmu_sync_roots(vcpu);
8837 		if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
8838 			kvm_mmu_load_pgd(vcpu);
8839 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
8840 			kvm_vcpu_flush_tlb_all(vcpu);
8841 
8842 			/* Flushing all ASIDs flushes the current ASID... */
8843 			kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
8844 		}
8845 		if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
8846 			kvm_vcpu_flush_tlb_current(vcpu);
8847 		if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu))
8848 			kvm_vcpu_flush_tlb_guest(vcpu);
8849 
8850 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
8851 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
8852 			r = 0;
8853 			goto out;
8854 		}
8855 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8856 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
8857 			vcpu->mmio_needed = 0;
8858 			r = 0;
8859 			goto out;
8860 		}
8861 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
8862 			/* Page is swapped out. Do synthetic halt */
8863 			vcpu->arch.apf.halted = true;
8864 			r = 1;
8865 			goto out;
8866 		}
8867 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8868 			record_steal_time(vcpu);
8869 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
8870 			process_smi(vcpu);
8871 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
8872 			process_nmi(vcpu);
8873 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
8874 			kvm_pmu_handle_event(vcpu);
8875 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
8876 			kvm_pmu_deliver_pmi(vcpu);
8877 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8878 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8879 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
8880 				     vcpu->arch.ioapic_handled_vectors)) {
8881 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8882 				vcpu->run->eoi.vector =
8883 						vcpu->arch.pending_ioapic_eoi;
8884 				r = 0;
8885 				goto out;
8886 			}
8887 		}
8888 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8889 			vcpu_scan_ioapic(vcpu);
8890 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8891 			vcpu_load_eoi_exitmap(vcpu);
8892 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8893 			kvm_vcpu_reload_apic_access_page(vcpu);
8894 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8895 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8896 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8897 			r = 0;
8898 			goto out;
8899 		}
8900 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8901 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8902 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8903 			r = 0;
8904 			goto out;
8905 		}
8906 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8907 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8908 			vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8909 			r = 0;
8910 			goto out;
8911 		}
8912 
8913 		/*
8914 		 * KVM_REQ_HV_STIMER has to be processed after
8915 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8916 		 * depend on the guest clock being up-to-date
8917 		 */
8918 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8919 			kvm_hv_process_stimers(vcpu);
8920 		if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
8921 			kvm_vcpu_update_apicv(vcpu);
8922 		if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
8923 			kvm_check_async_pf_completion(vcpu);
8924 		if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
8925 			kvm_x86_ops.msr_filter_changed(vcpu);
8926 	}
8927 
8928 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8929 		++vcpu->stat.req_event;
8930 		kvm_apic_accept_events(vcpu);
8931 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8932 			r = 1;
8933 			goto out;
8934 		}
8935 
8936 		inject_pending_event(vcpu, &req_immediate_exit);
8937 		if (req_int_win)
8938 			kvm_x86_ops.enable_irq_window(vcpu);
8939 
8940 		if (kvm_lapic_enabled(vcpu)) {
8941 			update_cr8_intercept(vcpu);
8942 			kvm_lapic_sync_to_vapic(vcpu);
8943 		}
8944 	}
8945 
8946 	r = kvm_mmu_reload(vcpu);
8947 	if (unlikely(r)) {
8948 		goto cancel_injection;
8949 	}
8950 
8951 	preempt_disable();
8952 
8953 	kvm_x86_ops.prepare_guest_switch(vcpu);
8954 
8955 	/*
8956 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
8957 	 * IPI are then delayed after guest entry, which ensures that they
8958 	 * result in virtual interrupt delivery.
8959 	 */
8960 	local_irq_disable();
8961 	vcpu->mode = IN_GUEST_MODE;
8962 
8963 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8964 
8965 	/*
8966 	 * 1) We should set ->mode before checking ->requests.  Please see
8967 	 * the comment in kvm_vcpu_exiting_guest_mode().
8968 	 *
8969 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
8970 	 * pairs with the memory barrier implicit in pi_test_and_set_on
8971 	 * (see vmx_deliver_posted_interrupt).
8972 	 *
8973 	 * 3) This also orders the write to mode from any reads to the page
8974 	 * tables done while the VCPU is running.  Please see the comment
8975 	 * in kvm_flush_remote_tlbs.
8976 	 */
8977 	smp_mb__after_srcu_read_unlock();
8978 
8979 	/*
8980 	 * This handles the case where a posted interrupt was
8981 	 * notified with kvm_vcpu_kick.
8982 	 */
8983 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8984 		kvm_x86_ops.sync_pir_to_irr(vcpu);
8985 
8986 	if (kvm_vcpu_exit_request(vcpu)) {
8987 		vcpu->mode = OUTSIDE_GUEST_MODE;
8988 		smp_wmb();
8989 		local_irq_enable();
8990 		preempt_enable();
8991 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8992 		r = 1;
8993 		goto cancel_injection;
8994 	}
8995 
8996 	if (req_immediate_exit) {
8997 		kvm_make_request(KVM_REQ_EVENT, vcpu);
8998 		kvm_x86_ops.request_immediate_exit(vcpu);
8999 	}
9000 
9001 	fpregs_assert_state_consistent();
9002 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
9003 		switch_fpu_return();
9004 
9005 	if (unlikely(vcpu->arch.switch_db_regs)) {
9006 		set_debugreg(0, 7);
9007 		set_debugreg(vcpu->arch.eff_db[0], 0);
9008 		set_debugreg(vcpu->arch.eff_db[1], 1);
9009 		set_debugreg(vcpu->arch.eff_db[2], 2);
9010 		set_debugreg(vcpu->arch.eff_db[3], 3);
9011 		set_debugreg(vcpu->arch.dr6, 6);
9012 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9013 	}
9014 
9015 	exit_fastpath = kvm_x86_ops.run(vcpu);
9016 
9017 	/*
9018 	 * Do this here before restoring debug registers on the host.  And
9019 	 * since we do this before handling the vmexit, a DR access vmexit
9020 	 * can (a) read the correct value of the debug registers, (b) set
9021 	 * KVM_DEBUGREG_WONT_EXIT again.
9022 	 */
9023 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9024 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9025 		kvm_x86_ops.sync_dirty_debug_regs(vcpu);
9026 		kvm_update_dr0123(vcpu);
9027 		kvm_update_dr7(vcpu);
9028 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9029 	}
9030 
9031 	/*
9032 	 * If the guest has used debug registers, at least dr7
9033 	 * will be disabled while returning to the host.
9034 	 * If we don't have active breakpoints in the host, we don't
9035 	 * care about the messed up debug address registers. But if
9036 	 * we have some of them active, restore the old state.
9037 	 */
9038 	if (hw_breakpoint_active())
9039 		hw_breakpoint_restore();
9040 
9041 	vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9042 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9043 
9044 	vcpu->mode = OUTSIDE_GUEST_MODE;
9045 	smp_wmb();
9046 
9047 	kvm_x86_ops.handle_exit_irqoff(vcpu);
9048 
9049 	/*
9050 	 * Consume any pending interrupts, including the possible source of
9051 	 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9052 	 * An instruction is required after local_irq_enable() to fully unblock
9053 	 * interrupts on processors that implement an interrupt shadow, the
9054 	 * stat.exits increment will do nicely.
9055 	 */
9056 	kvm_before_interrupt(vcpu);
9057 	local_irq_enable();
9058 	++vcpu->stat.exits;
9059 	local_irq_disable();
9060 	kvm_after_interrupt(vcpu);
9061 
9062 	if (lapic_in_kernel(vcpu)) {
9063 		s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9064 		if (delta != S64_MIN) {
9065 			trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9066 			vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9067 		}
9068 	}
9069 
9070 	local_irq_enable();
9071 	preempt_enable();
9072 
9073 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9074 
9075 	/*
9076 	 * Profile KVM exit RIPs:
9077 	 */
9078 	if (unlikely(prof_on == KVM_PROFILING)) {
9079 		unsigned long rip = kvm_rip_read(vcpu);
9080 		profile_hit(KVM_PROFILING, (void *)rip);
9081 	}
9082 
9083 	if (unlikely(vcpu->arch.tsc_always_catchup))
9084 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9085 
9086 	if (vcpu->arch.apic_attention)
9087 		kvm_lapic_sync_from_vapic(vcpu);
9088 
9089 	r = kvm_x86_ops.handle_exit(vcpu, exit_fastpath);
9090 	return r;
9091 
9092 cancel_injection:
9093 	if (req_immediate_exit)
9094 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9095 	kvm_x86_ops.cancel_injection(vcpu);
9096 	if (unlikely(vcpu->arch.apic_attention))
9097 		kvm_lapic_sync_from_vapic(vcpu);
9098 out:
9099 	return r;
9100 }
9101 
9102 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9103 {
9104 	if (!kvm_arch_vcpu_runnable(vcpu) &&
9105 	    (!kvm_x86_ops.pre_block || kvm_x86_ops.pre_block(vcpu) == 0)) {
9106 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9107 		kvm_vcpu_block(vcpu);
9108 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9109 
9110 		if (kvm_x86_ops.post_block)
9111 			kvm_x86_ops.post_block(vcpu);
9112 
9113 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9114 			return 1;
9115 	}
9116 
9117 	kvm_apic_accept_events(vcpu);
9118 	switch(vcpu->arch.mp_state) {
9119 	case KVM_MP_STATE_HALTED:
9120 	case KVM_MP_STATE_AP_RESET_HOLD:
9121 		vcpu->arch.pv.pv_unhalted = false;
9122 		vcpu->arch.mp_state =
9123 			KVM_MP_STATE_RUNNABLE;
9124 		fallthrough;
9125 	case KVM_MP_STATE_RUNNABLE:
9126 		vcpu->arch.apf.halted = false;
9127 		break;
9128 	case KVM_MP_STATE_INIT_RECEIVED:
9129 		break;
9130 	default:
9131 		return -EINTR;
9132 	}
9133 	return 1;
9134 }
9135 
9136 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9137 {
9138 	if (is_guest_mode(vcpu))
9139 		kvm_x86_ops.nested_ops->check_events(vcpu);
9140 
9141 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9142 		!vcpu->arch.apf.halted);
9143 }
9144 
9145 static int vcpu_run(struct kvm_vcpu *vcpu)
9146 {
9147 	int r;
9148 	struct kvm *kvm = vcpu->kvm;
9149 
9150 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9151 	vcpu->arch.l1tf_flush_l1d = true;
9152 
9153 	for (;;) {
9154 		if (kvm_vcpu_running(vcpu)) {
9155 			r = vcpu_enter_guest(vcpu);
9156 		} else {
9157 			r = vcpu_block(kvm, vcpu);
9158 		}
9159 
9160 		if (r <= 0)
9161 			break;
9162 
9163 		kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
9164 		if (kvm_cpu_has_pending_timer(vcpu))
9165 			kvm_inject_pending_timer_irqs(vcpu);
9166 
9167 		if (dm_request_for_irq_injection(vcpu) &&
9168 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9169 			r = 0;
9170 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9171 			++vcpu->stat.request_irq_exits;
9172 			break;
9173 		}
9174 
9175 		if (__xfer_to_guest_mode_work_pending()) {
9176 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9177 			r = xfer_to_guest_mode_handle_work(vcpu);
9178 			if (r)
9179 				return r;
9180 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9181 		}
9182 	}
9183 
9184 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9185 
9186 	return r;
9187 }
9188 
9189 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9190 {
9191 	int r;
9192 
9193 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9194 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9195 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9196 	return r;
9197 }
9198 
9199 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9200 {
9201 	BUG_ON(!vcpu->arch.pio.count);
9202 
9203 	return complete_emulated_io(vcpu);
9204 }
9205 
9206 /*
9207  * Implements the following, as a state machine:
9208  *
9209  * read:
9210  *   for each fragment
9211  *     for each mmio piece in the fragment
9212  *       write gpa, len
9213  *       exit
9214  *       copy data
9215  *   execute insn
9216  *
9217  * write:
9218  *   for each fragment
9219  *     for each mmio piece in the fragment
9220  *       write gpa, len
9221  *       copy data
9222  *       exit
9223  */
9224 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9225 {
9226 	struct kvm_run *run = vcpu->run;
9227 	struct kvm_mmio_fragment *frag;
9228 	unsigned len;
9229 
9230 	BUG_ON(!vcpu->mmio_needed);
9231 
9232 	/* Complete previous fragment */
9233 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9234 	len = min(8u, frag->len);
9235 	if (!vcpu->mmio_is_write)
9236 		memcpy(frag->data, run->mmio.data, len);
9237 
9238 	if (frag->len <= 8) {
9239 		/* Switch to the next fragment. */
9240 		frag++;
9241 		vcpu->mmio_cur_fragment++;
9242 	} else {
9243 		/* Go forward to the next mmio piece. */
9244 		frag->data += len;
9245 		frag->gpa += len;
9246 		frag->len -= len;
9247 	}
9248 
9249 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9250 		vcpu->mmio_needed = 0;
9251 
9252 		/* FIXME: return into emulator if single-stepping.  */
9253 		if (vcpu->mmio_is_write)
9254 			return 1;
9255 		vcpu->mmio_read_completed = 1;
9256 		return complete_emulated_io(vcpu);
9257 	}
9258 
9259 	run->exit_reason = KVM_EXIT_MMIO;
9260 	run->mmio.phys_addr = frag->gpa;
9261 	if (vcpu->mmio_is_write)
9262 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9263 	run->mmio.len = min(8u, frag->len);
9264 	run->mmio.is_write = vcpu->mmio_is_write;
9265 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9266 	return 0;
9267 }
9268 
9269 static void kvm_save_current_fpu(struct fpu *fpu)
9270 {
9271 	/*
9272 	 * If the target FPU state is not resident in the CPU registers, just
9273 	 * memcpy() from current, else save CPU state directly to the target.
9274 	 */
9275 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
9276 		memcpy(&fpu->state, &current->thread.fpu.state,
9277 		       fpu_kernel_xstate_size);
9278 	else
9279 		copy_fpregs_to_fpstate(fpu);
9280 }
9281 
9282 /* Swap (qemu) user FPU context for the guest FPU context. */
9283 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9284 {
9285 	fpregs_lock();
9286 
9287 	kvm_save_current_fpu(vcpu->arch.user_fpu);
9288 
9289 	/*
9290 	 * Guests with protected state can't have it set by the hypervisor,
9291 	 * so skip trying to set it.
9292 	 */
9293 	if (vcpu->arch.guest_fpu)
9294 		/* PKRU is separately restored in kvm_x86_ops.run. */
9295 		__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
9296 					~XFEATURE_MASK_PKRU);
9297 
9298 	fpregs_mark_activate();
9299 	fpregs_unlock();
9300 
9301 	trace_kvm_fpu(1);
9302 }
9303 
9304 /* When vcpu_run ends, restore user space FPU context. */
9305 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9306 {
9307 	fpregs_lock();
9308 
9309 	/*
9310 	 * Guests with protected state can't have it read by the hypervisor,
9311 	 * so skip trying to save it.
9312 	 */
9313 	if (vcpu->arch.guest_fpu)
9314 		kvm_save_current_fpu(vcpu->arch.guest_fpu);
9315 
9316 	copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
9317 
9318 	fpregs_mark_activate();
9319 	fpregs_unlock();
9320 
9321 	++vcpu->stat.fpu_reload;
9322 	trace_kvm_fpu(0);
9323 }
9324 
9325 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9326 {
9327 	struct kvm_run *kvm_run = vcpu->run;
9328 	int r;
9329 
9330 	vcpu_load(vcpu);
9331 	kvm_sigset_activate(vcpu);
9332 	kvm_load_guest_fpu(vcpu);
9333 
9334 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9335 		if (kvm_run->immediate_exit) {
9336 			r = -EINTR;
9337 			goto out;
9338 		}
9339 		kvm_vcpu_block(vcpu);
9340 		kvm_apic_accept_events(vcpu);
9341 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9342 		r = -EAGAIN;
9343 		if (signal_pending(current)) {
9344 			r = -EINTR;
9345 			kvm_run->exit_reason = KVM_EXIT_INTR;
9346 			++vcpu->stat.signal_exits;
9347 		}
9348 		goto out;
9349 	}
9350 
9351 	if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
9352 		r = -EINVAL;
9353 		goto out;
9354 	}
9355 
9356 	if (kvm_run->kvm_dirty_regs) {
9357 		r = sync_regs(vcpu);
9358 		if (r != 0)
9359 			goto out;
9360 	}
9361 
9362 	/* re-sync apic's tpr */
9363 	if (!lapic_in_kernel(vcpu)) {
9364 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
9365 			r = -EINVAL;
9366 			goto out;
9367 		}
9368 	}
9369 
9370 	if (unlikely(vcpu->arch.complete_userspace_io)) {
9371 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
9372 		vcpu->arch.complete_userspace_io = NULL;
9373 		r = cui(vcpu);
9374 		if (r <= 0)
9375 			goto out;
9376 	} else
9377 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
9378 
9379 	if (kvm_run->immediate_exit)
9380 		r = -EINTR;
9381 	else
9382 		r = vcpu_run(vcpu);
9383 
9384 out:
9385 	kvm_put_guest_fpu(vcpu);
9386 	if (kvm_run->kvm_valid_regs)
9387 		store_regs(vcpu);
9388 	post_kvm_run_save(vcpu);
9389 	kvm_sigset_deactivate(vcpu);
9390 
9391 	vcpu_put(vcpu);
9392 	return r;
9393 }
9394 
9395 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9396 {
9397 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
9398 		/*
9399 		 * We are here if userspace calls get_regs() in the middle of
9400 		 * instruction emulation. Registers state needs to be copied
9401 		 * back from emulation context to vcpu. Userspace shouldn't do
9402 		 * that usually, but some bad designed PV devices (vmware
9403 		 * backdoor interface) need this to work
9404 		 */
9405 		emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
9406 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9407 	}
9408 	regs->rax = kvm_rax_read(vcpu);
9409 	regs->rbx = kvm_rbx_read(vcpu);
9410 	regs->rcx = kvm_rcx_read(vcpu);
9411 	regs->rdx = kvm_rdx_read(vcpu);
9412 	regs->rsi = kvm_rsi_read(vcpu);
9413 	regs->rdi = kvm_rdi_read(vcpu);
9414 	regs->rsp = kvm_rsp_read(vcpu);
9415 	regs->rbp = kvm_rbp_read(vcpu);
9416 #ifdef CONFIG_X86_64
9417 	regs->r8 = kvm_r8_read(vcpu);
9418 	regs->r9 = kvm_r9_read(vcpu);
9419 	regs->r10 = kvm_r10_read(vcpu);
9420 	regs->r11 = kvm_r11_read(vcpu);
9421 	regs->r12 = kvm_r12_read(vcpu);
9422 	regs->r13 = kvm_r13_read(vcpu);
9423 	regs->r14 = kvm_r14_read(vcpu);
9424 	regs->r15 = kvm_r15_read(vcpu);
9425 #endif
9426 
9427 	regs->rip = kvm_rip_read(vcpu);
9428 	regs->rflags = kvm_get_rflags(vcpu);
9429 }
9430 
9431 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9432 {
9433 	vcpu_load(vcpu);
9434 	__get_regs(vcpu, regs);
9435 	vcpu_put(vcpu);
9436 	return 0;
9437 }
9438 
9439 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9440 {
9441 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
9442 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9443 
9444 	kvm_rax_write(vcpu, regs->rax);
9445 	kvm_rbx_write(vcpu, regs->rbx);
9446 	kvm_rcx_write(vcpu, regs->rcx);
9447 	kvm_rdx_write(vcpu, regs->rdx);
9448 	kvm_rsi_write(vcpu, regs->rsi);
9449 	kvm_rdi_write(vcpu, regs->rdi);
9450 	kvm_rsp_write(vcpu, regs->rsp);
9451 	kvm_rbp_write(vcpu, regs->rbp);
9452 #ifdef CONFIG_X86_64
9453 	kvm_r8_write(vcpu, regs->r8);
9454 	kvm_r9_write(vcpu, regs->r9);
9455 	kvm_r10_write(vcpu, regs->r10);
9456 	kvm_r11_write(vcpu, regs->r11);
9457 	kvm_r12_write(vcpu, regs->r12);
9458 	kvm_r13_write(vcpu, regs->r13);
9459 	kvm_r14_write(vcpu, regs->r14);
9460 	kvm_r15_write(vcpu, regs->r15);
9461 #endif
9462 
9463 	kvm_rip_write(vcpu, regs->rip);
9464 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
9465 
9466 	vcpu->arch.exception.pending = false;
9467 
9468 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9469 }
9470 
9471 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9472 {
9473 	vcpu_load(vcpu);
9474 	__set_regs(vcpu, regs);
9475 	vcpu_put(vcpu);
9476 	return 0;
9477 }
9478 
9479 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
9480 {
9481 	struct kvm_segment cs;
9482 
9483 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9484 	*db = cs.db;
9485 	*l = cs.l;
9486 }
9487 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
9488 
9489 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9490 {
9491 	struct desc_ptr dt;
9492 
9493 	if (vcpu->arch.guest_state_protected)
9494 		goto skip_protected_regs;
9495 
9496 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9497 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9498 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9499 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9500 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9501 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9502 
9503 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9504 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9505 
9506 	kvm_x86_ops.get_idt(vcpu, &dt);
9507 	sregs->idt.limit = dt.size;
9508 	sregs->idt.base = dt.address;
9509 	kvm_x86_ops.get_gdt(vcpu, &dt);
9510 	sregs->gdt.limit = dt.size;
9511 	sregs->gdt.base = dt.address;
9512 
9513 	sregs->cr2 = vcpu->arch.cr2;
9514 	sregs->cr3 = kvm_read_cr3(vcpu);
9515 
9516 skip_protected_regs:
9517 	sregs->cr0 = kvm_read_cr0(vcpu);
9518 	sregs->cr4 = kvm_read_cr4(vcpu);
9519 	sregs->cr8 = kvm_get_cr8(vcpu);
9520 	sregs->efer = vcpu->arch.efer;
9521 	sregs->apic_base = kvm_get_apic_base(vcpu);
9522 
9523 	memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
9524 
9525 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
9526 		set_bit(vcpu->arch.interrupt.nr,
9527 			(unsigned long *)sregs->interrupt_bitmap);
9528 }
9529 
9530 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
9531 				  struct kvm_sregs *sregs)
9532 {
9533 	vcpu_load(vcpu);
9534 	__get_sregs(vcpu, sregs);
9535 	vcpu_put(vcpu);
9536 	return 0;
9537 }
9538 
9539 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
9540 				    struct kvm_mp_state *mp_state)
9541 {
9542 	vcpu_load(vcpu);
9543 	if (kvm_mpx_supported())
9544 		kvm_load_guest_fpu(vcpu);
9545 
9546 	kvm_apic_accept_events(vcpu);
9547 	if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
9548 	     vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
9549 	    vcpu->arch.pv.pv_unhalted)
9550 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
9551 	else
9552 		mp_state->mp_state = vcpu->arch.mp_state;
9553 
9554 	if (kvm_mpx_supported())
9555 		kvm_put_guest_fpu(vcpu);
9556 	vcpu_put(vcpu);
9557 	return 0;
9558 }
9559 
9560 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
9561 				    struct kvm_mp_state *mp_state)
9562 {
9563 	int ret = -EINVAL;
9564 
9565 	vcpu_load(vcpu);
9566 
9567 	if (!lapic_in_kernel(vcpu) &&
9568 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
9569 		goto out;
9570 
9571 	/*
9572 	 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
9573 	 * INIT state; latched init should be reported using
9574 	 * KVM_SET_VCPU_EVENTS, so reject it here.
9575 	 */
9576 	if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
9577 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
9578 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
9579 		goto out;
9580 
9581 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
9582 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
9583 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
9584 	} else
9585 		vcpu->arch.mp_state = mp_state->mp_state;
9586 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9587 
9588 	ret = 0;
9589 out:
9590 	vcpu_put(vcpu);
9591 	return ret;
9592 }
9593 
9594 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
9595 		    int reason, bool has_error_code, u32 error_code)
9596 {
9597 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
9598 	int ret;
9599 
9600 	init_emulate_ctxt(vcpu);
9601 
9602 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9603 				   has_error_code, error_code);
9604 	if (ret) {
9605 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
9606 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
9607 		vcpu->run->internal.ndata = 0;
9608 		return 0;
9609 	}
9610 
9611 	kvm_rip_write(vcpu, ctxt->eip);
9612 	kvm_set_rflags(vcpu, ctxt->eflags);
9613 	return 1;
9614 }
9615 EXPORT_SYMBOL_GPL(kvm_task_switch);
9616 
9617 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9618 {
9619 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
9620 		/*
9621 		 * When EFER.LME and CR0.PG are set, the processor is in
9622 		 * 64-bit mode (though maybe in a 32-bit code segment).
9623 		 * CR4.PAE and EFER.LMA must be set.
9624 		 */
9625 		if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
9626 			return false;
9627 		if (sregs->cr3 & vcpu->arch.cr3_lm_rsvd_bits)
9628 			return false;
9629 	} else {
9630 		/*
9631 		 * Not in 64-bit mode: EFER.LMA is clear and the code
9632 		 * segment cannot be 64-bit.
9633 		 */
9634 		if (sregs->efer & EFER_LMA || sregs->cs.l)
9635 			return false;
9636 	}
9637 
9638 	return kvm_is_valid_cr4(vcpu, sregs->cr4);
9639 }
9640 
9641 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9642 {
9643 	struct msr_data apic_base_msr;
9644 	int mmu_reset_needed = 0;
9645 	int pending_vec, max_bits, idx;
9646 	struct desc_ptr dt;
9647 	int ret = -EINVAL;
9648 
9649 	if (!kvm_is_valid_sregs(vcpu, sregs))
9650 		goto out;
9651 
9652 	apic_base_msr.data = sregs->apic_base;
9653 	apic_base_msr.host_initiated = true;
9654 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
9655 		goto out;
9656 
9657 	if (vcpu->arch.guest_state_protected)
9658 		goto skip_protected_regs;
9659 
9660 	dt.size = sregs->idt.limit;
9661 	dt.address = sregs->idt.base;
9662 	kvm_x86_ops.set_idt(vcpu, &dt);
9663 	dt.size = sregs->gdt.limit;
9664 	dt.address = sregs->gdt.base;
9665 	kvm_x86_ops.set_gdt(vcpu, &dt);
9666 
9667 	vcpu->arch.cr2 = sregs->cr2;
9668 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
9669 	vcpu->arch.cr3 = sregs->cr3;
9670 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
9671 
9672 	kvm_set_cr8(vcpu, sregs->cr8);
9673 
9674 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
9675 	kvm_x86_ops.set_efer(vcpu, sregs->efer);
9676 
9677 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
9678 	kvm_x86_ops.set_cr0(vcpu, sregs->cr0);
9679 	vcpu->arch.cr0 = sregs->cr0;
9680 
9681 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
9682 	kvm_x86_ops.set_cr4(vcpu, sregs->cr4);
9683 
9684 	idx = srcu_read_lock(&vcpu->kvm->srcu);
9685 	if (is_pae_paging(vcpu)) {
9686 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
9687 		mmu_reset_needed = 1;
9688 	}
9689 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
9690 
9691 	if (mmu_reset_needed)
9692 		kvm_mmu_reset_context(vcpu);
9693 
9694 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9695 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9696 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9697 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9698 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9699 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9700 
9701 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9702 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9703 
9704 	update_cr8_intercept(vcpu);
9705 
9706 	/* Older userspace won't unhalt the vcpu on reset. */
9707 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9708 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
9709 	    !is_protmode(vcpu))
9710 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9711 
9712 skip_protected_regs:
9713 	max_bits = KVM_NR_INTERRUPTS;
9714 	pending_vec = find_first_bit(
9715 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
9716 	if (pending_vec < max_bits) {
9717 		kvm_queue_interrupt(vcpu, pending_vec, false);
9718 		pr_debug("Set back pending irq %d\n", pending_vec);
9719 	}
9720 
9721 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9722 
9723 	ret = 0;
9724 out:
9725 	return ret;
9726 }
9727 
9728 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
9729 				  struct kvm_sregs *sregs)
9730 {
9731 	int ret;
9732 
9733 	vcpu_load(vcpu);
9734 	ret = __set_sregs(vcpu, sregs);
9735 	vcpu_put(vcpu);
9736 	return ret;
9737 }
9738 
9739 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
9740 					struct kvm_guest_debug *dbg)
9741 {
9742 	unsigned long rflags;
9743 	int i, r;
9744 
9745 	if (vcpu->arch.guest_state_protected)
9746 		return -EINVAL;
9747 
9748 	vcpu_load(vcpu);
9749 
9750 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
9751 		r = -EBUSY;
9752 		if (vcpu->arch.exception.pending)
9753 			goto out;
9754 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
9755 			kvm_queue_exception(vcpu, DB_VECTOR);
9756 		else
9757 			kvm_queue_exception(vcpu, BP_VECTOR);
9758 	}
9759 
9760 	/*
9761 	 * Read rflags as long as potentially injected trace flags are still
9762 	 * filtered out.
9763 	 */
9764 	rflags = kvm_get_rflags(vcpu);
9765 
9766 	vcpu->guest_debug = dbg->control;
9767 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
9768 		vcpu->guest_debug = 0;
9769 
9770 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
9771 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
9772 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
9773 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
9774 	} else {
9775 		for (i = 0; i < KVM_NR_DB_REGS; i++)
9776 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
9777 	}
9778 	kvm_update_dr7(vcpu);
9779 
9780 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9781 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
9782 			get_segment_base(vcpu, VCPU_SREG_CS);
9783 
9784 	/*
9785 	 * Trigger an rflags update that will inject or remove the trace
9786 	 * flags.
9787 	 */
9788 	kvm_set_rflags(vcpu, rflags);
9789 
9790 	kvm_x86_ops.update_exception_bitmap(vcpu);
9791 
9792 	r = 0;
9793 
9794 out:
9795 	vcpu_put(vcpu);
9796 	return r;
9797 }
9798 
9799 /*
9800  * Translate a guest virtual address to a guest physical address.
9801  */
9802 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
9803 				    struct kvm_translation *tr)
9804 {
9805 	unsigned long vaddr = tr->linear_address;
9806 	gpa_t gpa;
9807 	int idx;
9808 
9809 	vcpu_load(vcpu);
9810 
9811 	idx = srcu_read_lock(&vcpu->kvm->srcu);
9812 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
9813 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
9814 	tr->physical_address = gpa;
9815 	tr->valid = gpa != UNMAPPED_GVA;
9816 	tr->writeable = 1;
9817 	tr->usermode = 0;
9818 
9819 	vcpu_put(vcpu);
9820 	return 0;
9821 }
9822 
9823 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9824 {
9825 	struct fxregs_state *fxsave;
9826 
9827 	if (!vcpu->arch.guest_fpu)
9828 		return 0;
9829 
9830 	vcpu_load(vcpu);
9831 
9832 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9833 	memcpy(fpu->fpr, fxsave->st_space, 128);
9834 	fpu->fcw = fxsave->cwd;
9835 	fpu->fsw = fxsave->swd;
9836 	fpu->ftwx = fxsave->twd;
9837 	fpu->last_opcode = fxsave->fop;
9838 	fpu->last_ip = fxsave->rip;
9839 	fpu->last_dp = fxsave->rdp;
9840 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
9841 
9842 	vcpu_put(vcpu);
9843 	return 0;
9844 }
9845 
9846 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9847 {
9848 	struct fxregs_state *fxsave;
9849 
9850 	if (!vcpu->arch.guest_fpu)
9851 		return 0;
9852 
9853 	vcpu_load(vcpu);
9854 
9855 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9856 
9857 	memcpy(fxsave->st_space, fpu->fpr, 128);
9858 	fxsave->cwd = fpu->fcw;
9859 	fxsave->swd = fpu->fsw;
9860 	fxsave->twd = fpu->ftwx;
9861 	fxsave->fop = fpu->last_opcode;
9862 	fxsave->rip = fpu->last_ip;
9863 	fxsave->rdp = fpu->last_dp;
9864 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
9865 
9866 	vcpu_put(vcpu);
9867 	return 0;
9868 }
9869 
9870 static void store_regs(struct kvm_vcpu *vcpu)
9871 {
9872 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
9873 
9874 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
9875 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
9876 
9877 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
9878 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
9879 
9880 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
9881 		kvm_vcpu_ioctl_x86_get_vcpu_events(
9882 				vcpu, &vcpu->run->s.regs.events);
9883 }
9884 
9885 static int sync_regs(struct kvm_vcpu *vcpu)
9886 {
9887 	if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
9888 		return -EINVAL;
9889 
9890 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
9891 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
9892 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
9893 	}
9894 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
9895 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
9896 			return -EINVAL;
9897 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
9898 	}
9899 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
9900 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9901 				vcpu, &vcpu->run->s.regs.events))
9902 			return -EINVAL;
9903 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9904 	}
9905 
9906 	return 0;
9907 }
9908 
9909 static void fx_init(struct kvm_vcpu *vcpu)
9910 {
9911 	if (!vcpu->arch.guest_fpu)
9912 		return;
9913 
9914 	fpstate_init(&vcpu->arch.guest_fpu->state);
9915 	if (boot_cpu_has(X86_FEATURE_XSAVES))
9916 		vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9917 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
9918 
9919 	/*
9920 	 * Ensure guest xcr0 is valid for loading
9921 	 */
9922 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9923 
9924 	vcpu->arch.cr0 |= X86_CR0_ET;
9925 }
9926 
9927 void kvm_free_guest_fpu(struct kvm_vcpu *vcpu)
9928 {
9929 	if (vcpu->arch.guest_fpu) {
9930 		kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9931 		vcpu->arch.guest_fpu = NULL;
9932 	}
9933 }
9934 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu);
9935 
9936 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
9937 {
9938 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9939 		pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
9940 			     "guest TSC will not be reliable\n");
9941 
9942 	return 0;
9943 }
9944 
9945 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
9946 {
9947 	struct page *page;
9948 	int r;
9949 
9950 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9951 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9952 	else
9953 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9954 
9955 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
9956 
9957 	r = kvm_mmu_create(vcpu);
9958 	if (r < 0)
9959 		return r;
9960 
9961 	if (irqchip_in_kernel(vcpu->kvm)) {
9962 		r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9963 		if (r < 0)
9964 			goto fail_mmu_destroy;
9965 		if (kvm_apicv_activated(vcpu->kvm))
9966 			vcpu->arch.apicv_active = true;
9967 	} else
9968 		static_key_slow_inc(&kvm_no_apic_vcpu);
9969 
9970 	r = -ENOMEM;
9971 
9972 	page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
9973 	if (!page)
9974 		goto fail_free_lapic;
9975 	vcpu->arch.pio_data = page_address(page);
9976 
9977 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9978 				       GFP_KERNEL_ACCOUNT);
9979 	if (!vcpu->arch.mce_banks)
9980 		goto fail_free_pio_data;
9981 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9982 
9983 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9984 				GFP_KERNEL_ACCOUNT))
9985 		goto fail_free_mce_banks;
9986 
9987 	if (!alloc_emulate_ctxt(vcpu))
9988 		goto free_wbinvd_dirty_mask;
9989 
9990 	vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
9991 						GFP_KERNEL_ACCOUNT);
9992 	if (!vcpu->arch.user_fpu) {
9993 		pr_err("kvm: failed to allocate userspace's fpu\n");
9994 		goto free_emulate_ctxt;
9995 	}
9996 
9997 	vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
9998 						 GFP_KERNEL_ACCOUNT);
9999 	if (!vcpu->arch.guest_fpu) {
10000 		pr_err("kvm: failed to allocate vcpu's fpu\n");
10001 		goto free_user_fpu;
10002 	}
10003 	fx_init(vcpu);
10004 
10005 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10006 	vcpu->arch.cr3_lm_rsvd_bits = rsvd_bits(cpuid_maxphyaddr(vcpu), 63);
10007 
10008 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10009 
10010 	kvm_async_pf_hash_reset(vcpu);
10011 	kvm_pmu_init(vcpu);
10012 
10013 	vcpu->arch.pending_external_vector = -1;
10014 	vcpu->arch.preempted_in_kernel = false;
10015 
10016 	kvm_hv_vcpu_init(vcpu);
10017 
10018 	r = kvm_x86_ops.vcpu_create(vcpu);
10019 	if (r)
10020 		goto free_guest_fpu;
10021 
10022 	vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10023 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10024 	kvm_vcpu_mtrr_init(vcpu);
10025 	vcpu_load(vcpu);
10026 	kvm_vcpu_reset(vcpu, false);
10027 	kvm_init_mmu(vcpu, false);
10028 	vcpu_put(vcpu);
10029 	return 0;
10030 
10031 free_guest_fpu:
10032 	kvm_free_guest_fpu(vcpu);
10033 free_user_fpu:
10034 	kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10035 free_emulate_ctxt:
10036 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10037 free_wbinvd_dirty_mask:
10038 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10039 fail_free_mce_banks:
10040 	kfree(vcpu->arch.mce_banks);
10041 fail_free_pio_data:
10042 	free_page((unsigned long)vcpu->arch.pio_data);
10043 fail_free_lapic:
10044 	kvm_free_lapic(vcpu);
10045 fail_mmu_destroy:
10046 	kvm_mmu_destroy(vcpu);
10047 	return r;
10048 }
10049 
10050 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10051 {
10052 	struct kvm *kvm = vcpu->kvm;
10053 
10054 	kvm_hv_vcpu_postcreate(vcpu);
10055 
10056 	if (mutex_lock_killable(&vcpu->mutex))
10057 		return;
10058 	vcpu_load(vcpu);
10059 	kvm_synchronize_tsc(vcpu, 0);
10060 	vcpu_put(vcpu);
10061 
10062 	/* poll control enabled by default */
10063 	vcpu->arch.msr_kvm_poll_control = 1;
10064 
10065 	mutex_unlock(&vcpu->mutex);
10066 
10067 	if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10068 		schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10069 						KVMCLOCK_SYNC_PERIOD);
10070 }
10071 
10072 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10073 {
10074 	struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
10075 	int idx;
10076 
10077 	kvm_release_pfn(cache->pfn, cache->dirty, cache);
10078 
10079 	kvmclock_reset(vcpu);
10080 
10081 	kvm_x86_ops.vcpu_free(vcpu);
10082 
10083 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10084 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10085 	kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10086 	kvm_free_guest_fpu(vcpu);
10087 
10088 	kvm_hv_vcpu_uninit(vcpu);
10089 	kvm_pmu_destroy(vcpu);
10090 	kfree(vcpu->arch.mce_banks);
10091 	kvm_free_lapic(vcpu);
10092 	idx = srcu_read_lock(&vcpu->kvm->srcu);
10093 	kvm_mmu_destroy(vcpu);
10094 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
10095 	free_page((unsigned long)vcpu->arch.pio_data);
10096 	kvfree(vcpu->arch.cpuid_entries);
10097 	if (!lapic_in_kernel(vcpu))
10098 		static_key_slow_dec(&kvm_no_apic_vcpu);
10099 }
10100 
10101 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10102 {
10103 	kvm_lapic_reset(vcpu, init_event);
10104 
10105 	vcpu->arch.hflags = 0;
10106 
10107 	vcpu->arch.smi_pending = 0;
10108 	vcpu->arch.smi_count = 0;
10109 	atomic_set(&vcpu->arch.nmi_queued, 0);
10110 	vcpu->arch.nmi_pending = 0;
10111 	vcpu->arch.nmi_injected = false;
10112 	kvm_clear_interrupt_queue(vcpu);
10113 	kvm_clear_exception_queue(vcpu);
10114 
10115 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10116 	kvm_update_dr0123(vcpu);
10117 	vcpu->arch.dr6 = DR6_INIT;
10118 	vcpu->arch.dr7 = DR7_FIXED_1;
10119 	kvm_update_dr7(vcpu);
10120 
10121 	vcpu->arch.cr2 = 0;
10122 
10123 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10124 	vcpu->arch.apf.msr_en_val = 0;
10125 	vcpu->arch.apf.msr_int_val = 0;
10126 	vcpu->arch.st.msr_val = 0;
10127 
10128 	kvmclock_reset(vcpu);
10129 
10130 	kvm_clear_async_pf_completion_queue(vcpu);
10131 	kvm_async_pf_hash_reset(vcpu);
10132 	vcpu->arch.apf.halted = false;
10133 
10134 	if (vcpu->arch.guest_fpu && kvm_mpx_supported()) {
10135 		void *mpx_state_buffer;
10136 
10137 		/*
10138 		 * To avoid have the INIT path from kvm_apic_has_events() that be
10139 		 * called with loaded FPU and does not let userspace fix the state.
10140 		 */
10141 		if (init_event)
10142 			kvm_put_guest_fpu(vcpu);
10143 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10144 					XFEATURE_BNDREGS);
10145 		if (mpx_state_buffer)
10146 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
10147 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10148 					XFEATURE_BNDCSR);
10149 		if (mpx_state_buffer)
10150 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
10151 		if (init_event)
10152 			kvm_load_guest_fpu(vcpu);
10153 	}
10154 
10155 	if (!init_event) {
10156 		kvm_pmu_reset(vcpu);
10157 		vcpu->arch.smbase = 0x30000;
10158 
10159 		vcpu->arch.msr_misc_features_enables = 0;
10160 
10161 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10162 	}
10163 
10164 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10165 	vcpu->arch.regs_avail = ~0;
10166 	vcpu->arch.regs_dirty = ~0;
10167 
10168 	vcpu->arch.ia32_xss = 0;
10169 
10170 	kvm_x86_ops.vcpu_reset(vcpu, init_event);
10171 }
10172 
10173 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10174 {
10175 	struct kvm_segment cs;
10176 
10177 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10178 	cs.selector = vector << 8;
10179 	cs.base = vector << 12;
10180 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10181 	kvm_rip_write(vcpu, 0);
10182 }
10183 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
10184 
10185 int kvm_arch_hardware_enable(void)
10186 {
10187 	struct kvm *kvm;
10188 	struct kvm_vcpu *vcpu;
10189 	int i;
10190 	int ret;
10191 	u64 local_tsc;
10192 	u64 max_tsc = 0;
10193 	bool stable, backwards_tsc = false;
10194 
10195 	kvm_user_return_msr_cpu_online();
10196 	ret = kvm_x86_ops.hardware_enable();
10197 	if (ret != 0)
10198 		return ret;
10199 
10200 	local_tsc = rdtsc();
10201 	stable = !kvm_check_tsc_unstable();
10202 	list_for_each_entry(kvm, &vm_list, vm_list) {
10203 		kvm_for_each_vcpu(i, vcpu, kvm) {
10204 			if (!stable && vcpu->cpu == smp_processor_id())
10205 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10206 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10207 				backwards_tsc = true;
10208 				if (vcpu->arch.last_host_tsc > max_tsc)
10209 					max_tsc = vcpu->arch.last_host_tsc;
10210 			}
10211 		}
10212 	}
10213 
10214 	/*
10215 	 * Sometimes, even reliable TSCs go backwards.  This happens on
10216 	 * platforms that reset TSC during suspend or hibernate actions, but
10217 	 * maintain synchronization.  We must compensate.  Fortunately, we can
10218 	 * detect that condition here, which happens early in CPU bringup,
10219 	 * before any KVM threads can be running.  Unfortunately, we can't
10220 	 * bring the TSCs fully up to date with real time, as we aren't yet far
10221 	 * enough into CPU bringup that we know how much real time has actually
10222 	 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10223 	 * variables that haven't been updated yet.
10224 	 *
10225 	 * So we simply find the maximum observed TSC above, then record the
10226 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
10227 	 * the adjustment will be applied.  Note that we accumulate
10228 	 * adjustments, in case multiple suspend cycles happen before some VCPU
10229 	 * gets a chance to run again.  In the event that no KVM threads get a
10230 	 * chance to run, we will miss the entire elapsed period, as we'll have
10231 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10232 	 * loose cycle time.  This isn't too big a deal, since the loss will be
10233 	 * uniform across all VCPUs (not to mention the scenario is extremely
10234 	 * unlikely). It is possible that a second hibernate recovery happens
10235 	 * much faster than a first, causing the observed TSC here to be
10236 	 * smaller; this would require additional padding adjustment, which is
10237 	 * why we set last_host_tsc to the local tsc observed here.
10238 	 *
10239 	 * N.B. - this code below runs only on platforms with reliable TSC,
10240 	 * as that is the only way backwards_tsc is set above.  Also note
10241 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10242 	 * have the same delta_cyc adjustment applied if backwards_tsc
10243 	 * is detected.  Note further, this adjustment is only done once,
10244 	 * as we reset last_host_tsc on all VCPUs to stop this from being
10245 	 * called multiple times (one for each physical CPU bringup).
10246 	 *
10247 	 * Platforms with unreliable TSCs don't have to deal with this, they
10248 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
10249 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
10250 	 * guarantee that they stay in perfect synchronization.
10251 	 */
10252 	if (backwards_tsc) {
10253 		u64 delta_cyc = max_tsc - local_tsc;
10254 		list_for_each_entry(kvm, &vm_list, vm_list) {
10255 			kvm->arch.backwards_tsc_observed = true;
10256 			kvm_for_each_vcpu(i, vcpu, kvm) {
10257 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
10258 				vcpu->arch.last_host_tsc = local_tsc;
10259 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
10260 			}
10261 
10262 			/*
10263 			 * We have to disable TSC offset matching.. if you were
10264 			 * booting a VM while issuing an S4 host suspend....
10265 			 * you may have some problem.  Solving this issue is
10266 			 * left as an exercise to the reader.
10267 			 */
10268 			kvm->arch.last_tsc_nsec = 0;
10269 			kvm->arch.last_tsc_write = 0;
10270 		}
10271 
10272 	}
10273 	return 0;
10274 }
10275 
10276 void kvm_arch_hardware_disable(void)
10277 {
10278 	kvm_x86_ops.hardware_disable();
10279 	drop_user_return_notifiers();
10280 }
10281 
10282 int kvm_arch_hardware_setup(void *opaque)
10283 {
10284 	struct kvm_x86_init_ops *ops = opaque;
10285 	int r;
10286 
10287 	rdmsrl_safe(MSR_EFER, &host_efer);
10288 
10289 	if (boot_cpu_has(X86_FEATURE_XSAVES))
10290 		rdmsrl(MSR_IA32_XSS, host_xss);
10291 
10292 	r = ops->hardware_setup();
10293 	if (r != 0)
10294 		return r;
10295 
10296 	memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
10297 
10298 	if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
10299 		supported_xss = 0;
10300 
10301 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
10302 	cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
10303 #undef __kvm_cpu_cap_has
10304 
10305 	if (kvm_has_tsc_control) {
10306 		/*
10307 		 * Make sure the user can only configure tsc_khz values that
10308 		 * fit into a signed integer.
10309 		 * A min value is not calculated because it will always
10310 		 * be 1 on all machines.
10311 		 */
10312 		u64 max = min(0x7fffffffULL,
10313 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
10314 		kvm_max_guest_tsc_khz = max;
10315 
10316 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
10317 	}
10318 
10319 	kvm_init_msr_list();
10320 	return 0;
10321 }
10322 
10323 void kvm_arch_hardware_unsetup(void)
10324 {
10325 	kvm_x86_ops.hardware_unsetup();
10326 }
10327 
10328 int kvm_arch_check_processor_compat(void *opaque)
10329 {
10330 	struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
10331 	struct kvm_x86_init_ops *ops = opaque;
10332 
10333 	WARN_ON(!irqs_disabled());
10334 
10335 	if (__cr4_reserved_bits(cpu_has, c) !=
10336 	    __cr4_reserved_bits(cpu_has, &boot_cpu_data))
10337 		return -EIO;
10338 
10339 	return ops->check_processor_compatibility();
10340 }
10341 
10342 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
10343 {
10344 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
10345 }
10346 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
10347 
10348 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
10349 {
10350 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
10351 }
10352 
10353 struct static_key kvm_no_apic_vcpu __read_mostly;
10354 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
10355 
10356 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
10357 {
10358 	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
10359 
10360 	vcpu->arch.l1tf_flush_l1d = true;
10361 	if (pmu->version && unlikely(pmu->event_count)) {
10362 		pmu->need_cleanup = true;
10363 		kvm_make_request(KVM_REQ_PMU, vcpu);
10364 	}
10365 	kvm_x86_ops.sched_in(vcpu, cpu);
10366 }
10367 
10368 void kvm_arch_free_vm(struct kvm *kvm)
10369 {
10370 	kfree(kvm->arch.hyperv.hv_pa_pg);
10371 	vfree(kvm);
10372 }
10373 
10374 
10375 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
10376 {
10377 	if (type)
10378 		return -EINVAL;
10379 
10380 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
10381 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
10382 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
10383 	INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
10384 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
10385 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
10386 
10387 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
10388 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
10389 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
10390 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
10391 		&kvm->arch.irq_sources_bitmap);
10392 
10393 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
10394 	mutex_init(&kvm->arch.apic_map_lock);
10395 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
10396 
10397 	kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
10398 	pvclock_update_vm_gtod_copy(kvm);
10399 
10400 	kvm->arch.guest_can_read_msr_platform_info = true;
10401 
10402 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
10403 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
10404 
10405 	kvm_hv_init_vm(kvm);
10406 	kvm_page_track_init(kvm);
10407 	kvm_mmu_init_vm(kvm);
10408 
10409 	return kvm_x86_ops.vm_init(kvm);
10410 }
10411 
10412 int kvm_arch_post_init_vm(struct kvm *kvm)
10413 {
10414 	return kvm_mmu_post_init_vm(kvm);
10415 }
10416 
10417 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
10418 {
10419 	vcpu_load(vcpu);
10420 	kvm_mmu_unload(vcpu);
10421 	vcpu_put(vcpu);
10422 }
10423 
10424 static void kvm_free_vcpus(struct kvm *kvm)
10425 {
10426 	unsigned int i;
10427 	struct kvm_vcpu *vcpu;
10428 
10429 	/*
10430 	 * Unpin any mmu pages first.
10431 	 */
10432 	kvm_for_each_vcpu(i, vcpu, kvm) {
10433 		kvm_clear_async_pf_completion_queue(vcpu);
10434 		kvm_unload_vcpu_mmu(vcpu);
10435 	}
10436 	kvm_for_each_vcpu(i, vcpu, kvm)
10437 		kvm_vcpu_destroy(vcpu);
10438 
10439 	mutex_lock(&kvm->lock);
10440 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
10441 		kvm->vcpus[i] = NULL;
10442 
10443 	atomic_set(&kvm->online_vcpus, 0);
10444 	mutex_unlock(&kvm->lock);
10445 }
10446 
10447 void kvm_arch_sync_events(struct kvm *kvm)
10448 {
10449 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
10450 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
10451 	kvm_free_pit(kvm);
10452 }
10453 
10454 #define  ERR_PTR_USR(e)  ((void __user *)ERR_PTR(e))
10455 
10456 /**
10457  * __x86_set_memory_region: Setup KVM internal memory slot
10458  *
10459  * @kvm: the kvm pointer to the VM.
10460  * @id: the slot ID to setup.
10461  * @gpa: the GPA to install the slot (unused when @size == 0).
10462  * @size: the size of the slot. Set to zero to uninstall a slot.
10463  *
10464  * This function helps to setup a KVM internal memory slot.  Specify
10465  * @size > 0 to install a new slot, while @size == 0 to uninstall a
10466  * slot.  The return code can be one of the following:
10467  *
10468  *   HVA:           on success (uninstall will return a bogus HVA)
10469  *   -errno:        on error
10470  *
10471  * The caller should always use IS_ERR() to check the return value
10472  * before use.  Note, the KVM internal memory slots are guaranteed to
10473  * remain valid and unchanged until the VM is destroyed, i.e., the
10474  * GPA->HVA translation will not change.  However, the HVA is a user
10475  * address, i.e. its accessibility is not guaranteed, and must be
10476  * accessed via __copy_{to,from}_user().
10477  */
10478 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
10479 				      u32 size)
10480 {
10481 	int i, r;
10482 	unsigned long hva, old_npages;
10483 	struct kvm_memslots *slots = kvm_memslots(kvm);
10484 	struct kvm_memory_slot *slot;
10485 
10486 	/* Called with kvm->slots_lock held.  */
10487 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
10488 		return ERR_PTR_USR(-EINVAL);
10489 
10490 	slot = id_to_memslot(slots, id);
10491 	if (size) {
10492 		if (slot && slot->npages)
10493 			return ERR_PTR_USR(-EEXIST);
10494 
10495 		/*
10496 		 * MAP_SHARED to prevent internal slot pages from being moved
10497 		 * by fork()/COW.
10498 		 */
10499 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
10500 			      MAP_SHARED | MAP_ANONYMOUS, 0);
10501 		if (IS_ERR((void *)hva))
10502 			return (void __user *)hva;
10503 	} else {
10504 		if (!slot || !slot->npages)
10505 			return 0;
10506 
10507 		old_npages = slot->npages;
10508 		hva = slot->userspace_addr;
10509 	}
10510 
10511 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
10512 		struct kvm_userspace_memory_region m;
10513 
10514 		m.slot = id | (i << 16);
10515 		m.flags = 0;
10516 		m.guest_phys_addr = gpa;
10517 		m.userspace_addr = hva;
10518 		m.memory_size = size;
10519 		r = __kvm_set_memory_region(kvm, &m);
10520 		if (r < 0)
10521 			return ERR_PTR_USR(r);
10522 	}
10523 
10524 	if (!size)
10525 		vm_munmap(hva, old_npages * PAGE_SIZE);
10526 
10527 	return (void __user *)hva;
10528 }
10529 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
10530 
10531 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
10532 {
10533 	kvm_mmu_pre_destroy_vm(kvm);
10534 }
10535 
10536 void kvm_arch_destroy_vm(struct kvm *kvm)
10537 {
10538 	u32 i;
10539 
10540 	if (current->mm == kvm->mm) {
10541 		/*
10542 		 * Free memory regions allocated on behalf of userspace,
10543 		 * unless the the memory map has changed due to process exit
10544 		 * or fd copying.
10545 		 */
10546 		mutex_lock(&kvm->slots_lock);
10547 		__x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
10548 					0, 0);
10549 		__x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
10550 					0, 0);
10551 		__x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
10552 		mutex_unlock(&kvm->slots_lock);
10553 	}
10554 	if (kvm_x86_ops.vm_destroy)
10555 		kvm_x86_ops.vm_destroy(kvm);
10556 	for (i = 0; i < kvm->arch.msr_filter.count; i++)
10557 		kfree(kvm->arch.msr_filter.ranges[i].bitmap);
10558 	kvm_pic_destroy(kvm);
10559 	kvm_ioapic_destroy(kvm);
10560 	kvm_free_vcpus(kvm);
10561 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
10562 	kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
10563 	kvm_mmu_uninit_vm(kvm);
10564 	kvm_page_track_cleanup(kvm);
10565 	kvm_hv_destroy_vm(kvm);
10566 }
10567 
10568 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
10569 {
10570 	int i;
10571 
10572 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10573 		kvfree(slot->arch.rmap[i]);
10574 		slot->arch.rmap[i] = NULL;
10575 
10576 		if (i == 0)
10577 			continue;
10578 
10579 		kvfree(slot->arch.lpage_info[i - 1]);
10580 		slot->arch.lpage_info[i - 1] = NULL;
10581 	}
10582 
10583 	kvm_page_track_free_memslot(slot);
10584 }
10585 
10586 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
10587 				      unsigned long npages)
10588 {
10589 	int i;
10590 
10591 	/*
10592 	 * Clear out the previous array pointers for the KVM_MR_MOVE case.  The
10593 	 * old arrays will be freed by __kvm_set_memory_region() if installing
10594 	 * the new memslot is successful.
10595 	 */
10596 	memset(&slot->arch, 0, sizeof(slot->arch));
10597 
10598 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10599 		struct kvm_lpage_info *linfo;
10600 		unsigned long ugfn;
10601 		int lpages;
10602 		int level = i + 1;
10603 
10604 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
10605 				      slot->base_gfn, level) + 1;
10606 
10607 		slot->arch.rmap[i] =
10608 			kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
10609 				 GFP_KERNEL_ACCOUNT);
10610 		if (!slot->arch.rmap[i])
10611 			goto out_free;
10612 		if (i == 0)
10613 			continue;
10614 
10615 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
10616 		if (!linfo)
10617 			goto out_free;
10618 
10619 		slot->arch.lpage_info[i - 1] = linfo;
10620 
10621 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
10622 			linfo[0].disallow_lpage = 1;
10623 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
10624 			linfo[lpages - 1].disallow_lpage = 1;
10625 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
10626 		/*
10627 		 * If the gfn and userspace address are not aligned wrt each
10628 		 * other, disable large page support for this slot.
10629 		 */
10630 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
10631 			unsigned long j;
10632 
10633 			for (j = 0; j < lpages; ++j)
10634 				linfo[j].disallow_lpage = 1;
10635 		}
10636 	}
10637 
10638 	if (kvm_page_track_create_memslot(slot, npages))
10639 		goto out_free;
10640 
10641 	return 0;
10642 
10643 out_free:
10644 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10645 		kvfree(slot->arch.rmap[i]);
10646 		slot->arch.rmap[i] = NULL;
10647 		if (i == 0)
10648 			continue;
10649 
10650 		kvfree(slot->arch.lpage_info[i - 1]);
10651 		slot->arch.lpage_info[i - 1] = NULL;
10652 	}
10653 	return -ENOMEM;
10654 }
10655 
10656 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
10657 {
10658 	struct kvm_vcpu *vcpu;
10659 	int i;
10660 
10661 	/*
10662 	 * memslots->generation has been incremented.
10663 	 * mmio generation may have reached its maximum value.
10664 	 */
10665 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
10666 
10667 	/* Force re-initialization of steal_time cache */
10668 	kvm_for_each_vcpu(i, vcpu, kvm)
10669 		kvm_vcpu_kick(vcpu);
10670 }
10671 
10672 int kvm_arch_prepare_memory_region(struct kvm *kvm,
10673 				struct kvm_memory_slot *memslot,
10674 				const struct kvm_userspace_memory_region *mem,
10675 				enum kvm_mr_change change)
10676 {
10677 	if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
10678 		return kvm_alloc_memslot_metadata(memslot,
10679 						  mem->memory_size >> PAGE_SHIFT);
10680 	return 0;
10681 }
10682 
10683 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
10684 				     struct kvm_memory_slot *old,
10685 				     struct kvm_memory_slot *new,
10686 				     enum kvm_mr_change change)
10687 {
10688 	/*
10689 	 * Nothing to do for RO slots or CREATE/MOVE/DELETE of a slot.
10690 	 * See comments below.
10691 	 */
10692 	if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
10693 		return;
10694 
10695 	/*
10696 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
10697 	 * sptes have to be split.  If live migration is successful, the guest
10698 	 * in the source machine will be destroyed and large sptes will be
10699 	 * created in the destination. However, if the guest continues to run
10700 	 * in the source machine (for example if live migration fails), small
10701 	 * sptes will remain around and cause bad performance.
10702 	 *
10703 	 * Scan sptes if dirty logging has been stopped, dropping those
10704 	 * which can be collapsed into a single large-page spte.  Later
10705 	 * page faults will create the large-page sptes.
10706 	 *
10707 	 * There is no need to do this in any of the following cases:
10708 	 * CREATE:      No dirty mappings will already exist.
10709 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
10710 	 *		kvm_arch_flush_shadow_memslot()
10711 	 */
10712 	if ((old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
10713 	    !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
10714 		kvm_mmu_zap_collapsible_sptes(kvm, new);
10715 
10716 	/*
10717 	 * Enable or disable dirty logging for the slot.
10718 	 *
10719 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of the old
10720 	 * slot have been zapped so no dirty logging updates are needed for
10721 	 * the old slot.
10722 	 * For KVM_MR_CREATE and KVM_MR_MOVE, once the new slot is visible
10723 	 * any mappings that might be created in it will consume the
10724 	 * properties of the new slot and do not need to be updated here.
10725 	 *
10726 	 * When PML is enabled, the kvm_x86_ops dirty logging hooks are
10727 	 * called to enable/disable dirty logging.
10728 	 *
10729 	 * When disabling dirty logging with PML enabled, the D-bit is set
10730 	 * for sptes in the slot in order to prevent unnecessary GPA
10731 	 * logging in the PML buffer (and potential PML buffer full VMEXIT).
10732 	 * This guarantees leaving PML enabled for the guest's lifetime
10733 	 * won't have any additional overhead from PML when the guest is
10734 	 * running with dirty logging disabled.
10735 	 *
10736 	 * When enabling dirty logging, large sptes are write-protected
10737 	 * so they can be split on first write.  New large sptes cannot
10738 	 * be created for this slot until the end of the logging.
10739 	 * See the comments in fast_page_fault().
10740 	 * For small sptes, nothing is done if the dirty log is in the
10741 	 * initial-all-set state.  Otherwise, depending on whether pml
10742 	 * is enabled the D-bit or the W-bit will be cleared.
10743 	 */
10744 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
10745 		if (kvm_x86_ops.slot_enable_log_dirty) {
10746 			kvm_x86_ops.slot_enable_log_dirty(kvm, new);
10747 		} else {
10748 			int level =
10749 				kvm_dirty_log_manual_protect_and_init_set(kvm) ?
10750 				PG_LEVEL_2M : PG_LEVEL_4K;
10751 
10752 			/*
10753 			 * If we're with initial-all-set, we don't need
10754 			 * to write protect any small page because
10755 			 * they're reported as dirty already.  However
10756 			 * we still need to write-protect huge pages
10757 			 * so that the page split can happen lazily on
10758 			 * the first write to the huge page.
10759 			 */
10760 			kvm_mmu_slot_remove_write_access(kvm, new, level);
10761 		}
10762 	} else {
10763 		if (kvm_x86_ops.slot_disable_log_dirty)
10764 			kvm_x86_ops.slot_disable_log_dirty(kvm, new);
10765 	}
10766 }
10767 
10768 void kvm_arch_commit_memory_region(struct kvm *kvm,
10769 				const struct kvm_userspace_memory_region *mem,
10770 				struct kvm_memory_slot *old,
10771 				const struct kvm_memory_slot *new,
10772 				enum kvm_mr_change change)
10773 {
10774 	if (!kvm->arch.n_requested_mmu_pages)
10775 		kvm_mmu_change_mmu_pages(kvm,
10776 				kvm_mmu_calculate_default_mmu_pages(kvm));
10777 
10778 	/*
10779 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
10780 	 */
10781 	kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change);
10782 
10783 	/* Free the arrays associated with the old memslot. */
10784 	if (change == KVM_MR_MOVE)
10785 		kvm_arch_free_memslot(kvm, old);
10786 }
10787 
10788 void kvm_arch_flush_shadow_all(struct kvm *kvm)
10789 {
10790 	kvm_mmu_zap_all(kvm);
10791 }
10792 
10793 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
10794 				   struct kvm_memory_slot *slot)
10795 {
10796 	kvm_page_track_flush_slot(kvm, slot);
10797 }
10798 
10799 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
10800 {
10801 	return (is_guest_mode(vcpu) &&
10802 			kvm_x86_ops.guest_apic_has_interrupt &&
10803 			kvm_x86_ops.guest_apic_has_interrupt(vcpu));
10804 }
10805 
10806 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
10807 {
10808 	if (!list_empty_careful(&vcpu->async_pf.done))
10809 		return true;
10810 
10811 	if (kvm_apic_has_events(vcpu))
10812 		return true;
10813 
10814 	if (vcpu->arch.pv.pv_unhalted)
10815 		return true;
10816 
10817 	if (vcpu->arch.exception.pending)
10818 		return true;
10819 
10820 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10821 	    (vcpu->arch.nmi_pending &&
10822 	     kvm_x86_ops.nmi_allowed(vcpu, false)))
10823 		return true;
10824 
10825 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
10826 	    (vcpu->arch.smi_pending &&
10827 	     kvm_x86_ops.smi_allowed(vcpu, false)))
10828 		return true;
10829 
10830 	if (kvm_arch_interrupt_allowed(vcpu) &&
10831 	    (kvm_cpu_has_interrupt(vcpu) ||
10832 	    kvm_guest_apic_has_interrupt(vcpu)))
10833 		return true;
10834 
10835 	if (kvm_hv_has_stimer_pending(vcpu))
10836 		return true;
10837 
10838 	if (is_guest_mode(vcpu) &&
10839 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
10840 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
10841 		return true;
10842 
10843 	return false;
10844 }
10845 
10846 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
10847 {
10848 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
10849 }
10850 
10851 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
10852 {
10853 	if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
10854 		return true;
10855 
10856 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10857 		kvm_test_request(KVM_REQ_SMI, vcpu) ||
10858 		 kvm_test_request(KVM_REQ_EVENT, vcpu))
10859 		return true;
10860 
10861 	if (vcpu->arch.apicv_active && kvm_x86_ops.dy_apicv_has_pending_interrupt(vcpu))
10862 		return true;
10863 
10864 	return false;
10865 }
10866 
10867 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
10868 {
10869 	return vcpu->arch.preempted_in_kernel;
10870 }
10871 
10872 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
10873 {
10874 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
10875 }
10876 
10877 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
10878 {
10879 	return kvm_x86_ops.interrupt_allowed(vcpu, false);
10880 }
10881 
10882 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
10883 {
10884 	/* Can't read the RIP when guest state is protected, just return 0 */
10885 	if (vcpu->arch.guest_state_protected)
10886 		return 0;
10887 
10888 	if (is_64_bit_mode(vcpu))
10889 		return kvm_rip_read(vcpu);
10890 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
10891 		     kvm_rip_read(vcpu));
10892 }
10893 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
10894 
10895 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
10896 {
10897 	return kvm_get_linear_rip(vcpu) == linear_rip;
10898 }
10899 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
10900 
10901 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
10902 {
10903 	unsigned long rflags;
10904 
10905 	rflags = kvm_x86_ops.get_rflags(vcpu);
10906 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10907 		rflags &= ~X86_EFLAGS_TF;
10908 	return rflags;
10909 }
10910 EXPORT_SYMBOL_GPL(kvm_get_rflags);
10911 
10912 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10913 {
10914 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
10915 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
10916 		rflags |= X86_EFLAGS_TF;
10917 	kvm_x86_ops.set_rflags(vcpu, rflags);
10918 }
10919 
10920 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10921 {
10922 	__kvm_set_rflags(vcpu, rflags);
10923 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10924 }
10925 EXPORT_SYMBOL_GPL(kvm_set_rflags);
10926 
10927 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
10928 {
10929 	int r;
10930 
10931 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
10932 	      work->wakeup_all)
10933 		return;
10934 
10935 	r = kvm_mmu_reload(vcpu);
10936 	if (unlikely(r))
10937 		return;
10938 
10939 	if (!vcpu->arch.mmu->direct_map &&
10940 	      work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
10941 		return;
10942 
10943 	kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
10944 }
10945 
10946 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
10947 {
10948 	BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
10949 
10950 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
10951 }
10952 
10953 static inline u32 kvm_async_pf_next_probe(u32 key)
10954 {
10955 	return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
10956 }
10957 
10958 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10959 {
10960 	u32 key = kvm_async_pf_hash_fn(gfn);
10961 
10962 	while (vcpu->arch.apf.gfns[key] != ~0)
10963 		key = kvm_async_pf_next_probe(key);
10964 
10965 	vcpu->arch.apf.gfns[key] = gfn;
10966 }
10967 
10968 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
10969 {
10970 	int i;
10971 	u32 key = kvm_async_pf_hash_fn(gfn);
10972 
10973 	for (i = 0; i < ASYNC_PF_PER_VCPU &&
10974 		     (vcpu->arch.apf.gfns[key] != gfn &&
10975 		      vcpu->arch.apf.gfns[key] != ~0); i++)
10976 		key = kvm_async_pf_next_probe(key);
10977 
10978 	return key;
10979 }
10980 
10981 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10982 {
10983 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
10984 }
10985 
10986 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10987 {
10988 	u32 i, j, k;
10989 
10990 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
10991 
10992 	if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
10993 		return;
10994 
10995 	while (true) {
10996 		vcpu->arch.apf.gfns[i] = ~0;
10997 		do {
10998 			j = kvm_async_pf_next_probe(j);
10999 			if (vcpu->arch.apf.gfns[j] == ~0)
11000 				return;
11001 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
11002 			/*
11003 			 * k lies cyclically in ]i,j]
11004 			 * |    i.k.j |
11005 			 * |....j i.k.| or  |.k..j i...|
11006 			 */
11007 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
11008 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11009 		i = j;
11010 	}
11011 }
11012 
11013 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11014 {
11015 	u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11016 
11017 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11018 				      sizeof(reason));
11019 }
11020 
11021 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11022 {
11023 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11024 
11025 	return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11026 					     &token, offset, sizeof(token));
11027 }
11028 
11029 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
11030 {
11031 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11032 	u32 val;
11033 
11034 	if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11035 					 &val, offset, sizeof(val)))
11036 		return false;
11037 
11038 	return !val;
11039 }
11040 
11041 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
11042 {
11043 	if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
11044 		return false;
11045 
11046 	if (!kvm_pv_async_pf_enabled(vcpu) ||
11047 	    (vcpu->arch.apf.send_user_only && kvm_x86_ops.get_cpl(vcpu) == 0))
11048 		return false;
11049 
11050 	return true;
11051 }
11052 
11053 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
11054 {
11055 	if (unlikely(!lapic_in_kernel(vcpu) ||
11056 		     kvm_event_needs_reinjection(vcpu) ||
11057 		     vcpu->arch.exception.pending))
11058 		return false;
11059 
11060 	if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
11061 		return false;
11062 
11063 	/*
11064 	 * If interrupts are off we cannot even use an artificial
11065 	 * halt state.
11066 	 */
11067 	return kvm_arch_interrupt_allowed(vcpu);
11068 }
11069 
11070 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
11071 				     struct kvm_async_pf *work)
11072 {
11073 	struct x86_exception fault;
11074 
11075 	trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
11076 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
11077 
11078 	if (kvm_can_deliver_async_pf(vcpu) &&
11079 	    !apf_put_user_notpresent(vcpu)) {
11080 		fault.vector = PF_VECTOR;
11081 		fault.error_code_valid = true;
11082 		fault.error_code = 0;
11083 		fault.nested_page_fault = false;
11084 		fault.address = work->arch.token;
11085 		fault.async_page_fault = true;
11086 		kvm_inject_page_fault(vcpu, &fault);
11087 		return true;
11088 	} else {
11089 		/*
11090 		 * It is not possible to deliver a paravirtualized asynchronous
11091 		 * page fault, but putting the guest in an artificial halt state
11092 		 * can be beneficial nevertheless: if an interrupt arrives, we
11093 		 * can deliver it timely and perhaps the guest will schedule
11094 		 * another process.  When the instruction that triggered a page
11095 		 * fault is retried, hopefully the page will be ready in the host.
11096 		 */
11097 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
11098 		return false;
11099 	}
11100 }
11101 
11102 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
11103 				 struct kvm_async_pf *work)
11104 {
11105 	struct kvm_lapic_irq irq = {
11106 		.delivery_mode = APIC_DM_FIXED,
11107 		.vector = vcpu->arch.apf.vec
11108 	};
11109 
11110 	if (work->wakeup_all)
11111 		work->arch.token = ~0; /* broadcast wakeup */
11112 	else
11113 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
11114 	trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
11115 
11116 	if ((work->wakeup_all || work->notpresent_injected) &&
11117 	    kvm_pv_async_pf_enabled(vcpu) &&
11118 	    !apf_put_user_ready(vcpu, work->arch.token)) {
11119 		vcpu->arch.apf.pageready_pending = true;
11120 		kvm_apic_set_irq(vcpu, &irq, NULL);
11121 	}
11122 
11123 	vcpu->arch.apf.halted = false;
11124 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11125 }
11126 
11127 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
11128 {
11129 	kvm_make_request(KVM_REQ_APF_READY, vcpu);
11130 	if (!vcpu->arch.apf.pageready_pending)
11131 		kvm_vcpu_kick(vcpu);
11132 }
11133 
11134 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
11135 {
11136 	if (!kvm_pv_async_pf_enabled(vcpu))
11137 		return true;
11138 	else
11139 		return apf_pageready_slot_free(vcpu);
11140 }
11141 
11142 void kvm_arch_start_assignment(struct kvm *kvm)
11143 {
11144 	atomic_inc(&kvm->arch.assigned_device_count);
11145 }
11146 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
11147 
11148 void kvm_arch_end_assignment(struct kvm *kvm)
11149 {
11150 	atomic_dec(&kvm->arch.assigned_device_count);
11151 }
11152 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
11153 
11154 bool kvm_arch_has_assigned_device(struct kvm *kvm)
11155 {
11156 	return atomic_read(&kvm->arch.assigned_device_count);
11157 }
11158 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
11159 
11160 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
11161 {
11162 	atomic_inc(&kvm->arch.noncoherent_dma_count);
11163 }
11164 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
11165 
11166 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
11167 {
11168 	atomic_dec(&kvm->arch.noncoherent_dma_count);
11169 }
11170 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
11171 
11172 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
11173 {
11174 	return atomic_read(&kvm->arch.noncoherent_dma_count);
11175 }
11176 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
11177 
11178 bool kvm_arch_has_irq_bypass(void)
11179 {
11180 	return true;
11181 }
11182 
11183 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
11184 				      struct irq_bypass_producer *prod)
11185 {
11186 	struct kvm_kernel_irqfd *irqfd =
11187 		container_of(cons, struct kvm_kernel_irqfd, consumer);
11188 	int ret;
11189 
11190 	irqfd->producer = prod;
11191 	kvm_arch_start_assignment(irqfd->kvm);
11192 	ret = kvm_x86_ops.update_pi_irte(irqfd->kvm,
11193 					 prod->irq, irqfd->gsi, 1);
11194 
11195 	if (ret)
11196 		kvm_arch_end_assignment(irqfd->kvm);
11197 
11198 	return ret;
11199 }
11200 
11201 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
11202 				      struct irq_bypass_producer *prod)
11203 {
11204 	int ret;
11205 	struct kvm_kernel_irqfd *irqfd =
11206 		container_of(cons, struct kvm_kernel_irqfd, consumer);
11207 
11208 	WARN_ON(irqfd->producer != prod);
11209 	irqfd->producer = NULL;
11210 
11211 	/*
11212 	 * When producer of consumer is unregistered, we change back to
11213 	 * remapped mode, so we can re-use the current implementation
11214 	 * when the irq is masked/disabled or the consumer side (KVM
11215 	 * int this case doesn't want to receive the interrupts.
11216 	*/
11217 	ret = kvm_x86_ops.update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
11218 	if (ret)
11219 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
11220 		       " fails: %d\n", irqfd->consumer.token, ret);
11221 
11222 	kvm_arch_end_assignment(irqfd->kvm);
11223 }
11224 
11225 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
11226 				   uint32_t guest_irq, bool set)
11227 {
11228 	return kvm_x86_ops.update_pi_irte(kvm, host_irq, guest_irq, set);
11229 }
11230 
11231 bool kvm_vector_hashing_enabled(void)
11232 {
11233 	return vector_hashing;
11234 }
11235 
11236 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
11237 {
11238 	return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
11239 }
11240 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
11241 
11242 
11243 int kvm_spec_ctrl_test_value(u64 value)
11244 {
11245 	/*
11246 	 * test that setting IA32_SPEC_CTRL to given value
11247 	 * is allowed by the host processor
11248 	 */
11249 
11250 	u64 saved_value;
11251 	unsigned long flags;
11252 	int ret = 0;
11253 
11254 	local_irq_save(flags);
11255 
11256 	if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
11257 		ret = 1;
11258 	else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
11259 		ret = 1;
11260 	else
11261 		wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
11262 
11263 	local_irq_restore(flags);
11264 
11265 	return ret;
11266 }
11267 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
11268 
11269 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
11270 {
11271 	struct x86_exception fault;
11272 	u32 access = error_code &
11273 		(PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
11274 
11275 	if (!(error_code & PFERR_PRESENT_MASK) ||
11276 	    vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
11277 		/*
11278 		 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
11279 		 * tables probably do not match the TLB.  Just proceed
11280 		 * with the error code that the processor gave.
11281 		 */
11282 		fault.vector = PF_VECTOR;
11283 		fault.error_code_valid = true;
11284 		fault.error_code = error_code;
11285 		fault.nested_page_fault = false;
11286 		fault.address = gva;
11287 	}
11288 	vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
11289 }
11290 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
11291 
11292 /*
11293  * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
11294  * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
11295  * indicates whether exit to userspace is needed.
11296  */
11297 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
11298 			      struct x86_exception *e)
11299 {
11300 	if (r == X86EMUL_PROPAGATE_FAULT) {
11301 		kvm_inject_emulated_page_fault(vcpu, e);
11302 		return 1;
11303 	}
11304 
11305 	/*
11306 	 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
11307 	 * while handling a VMX instruction KVM could've handled the request
11308 	 * correctly by exiting to userspace and performing I/O but there
11309 	 * doesn't seem to be a real use-case behind such requests, just return
11310 	 * KVM_EXIT_INTERNAL_ERROR for now.
11311 	 */
11312 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
11313 	vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
11314 	vcpu->run->internal.ndata = 0;
11315 
11316 	return 0;
11317 }
11318 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
11319 
11320 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
11321 {
11322 	bool pcid_enabled;
11323 	struct x86_exception e;
11324 	unsigned i;
11325 	unsigned long roots_to_free = 0;
11326 	struct {
11327 		u64 pcid;
11328 		u64 gla;
11329 	} operand;
11330 	int r;
11331 
11332 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
11333 	if (r != X86EMUL_CONTINUE)
11334 		return kvm_handle_memory_failure(vcpu, r, &e);
11335 
11336 	if (operand.pcid >> 12 != 0) {
11337 		kvm_inject_gp(vcpu, 0);
11338 		return 1;
11339 	}
11340 
11341 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
11342 
11343 	switch (type) {
11344 	case INVPCID_TYPE_INDIV_ADDR:
11345 		if ((!pcid_enabled && (operand.pcid != 0)) ||
11346 		    is_noncanonical_address(operand.gla, vcpu)) {
11347 			kvm_inject_gp(vcpu, 0);
11348 			return 1;
11349 		}
11350 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
11351 		return kvm_skip_emulated_instruction(vcpu);
11352 
11353 	case INVPCID_TYPE_SINGLE_CTXT:
11354 		if (!pcid_enabled && (operand.pcid != 0)) {
11355 			kvm_inject_gp(vcpu, 0);
11356 			return 1;
11357 		}
11358 
11359 		if (kvm_get_active_pcid(vcpu) == operand.pcid) {
11360 			kvm_mmu_sync_roots(vcpu);
11361 			kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
11362 		}
11363 
11364 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
11365 			if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd)
11366 			    == operand.pcid)
11367 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
11368 
11369 		kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
11370 		/*
11371 		 * If neither the current cr3 nor any of the prev_roots use the
11372 		 * given PCID, then nothing needs to be done here because a
11373 		 * resync will happen anyway before switching to any other CR3.
11374 		 */
11375 
11376 		return kvm_skip_emulated_instruction(vcpu);
11377 
11378 	case INVPCID_TYPE_ALL_NON_GLOBAL:
11379 		/*
11380 		 * Currently, KVM doesn't mark global entries in the shadow
11381 		 * page tables, so a non-global flush just degenerates to a
11382 		 * global flush. If needed, we could optimize this later by
11383 		 * keeping track of global entries in shadow page tables.
11384 		 */
11385 
11386 		fallthrough;
11387 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
11388 		kvm_mmu_unload(vcpu);
11389 		return kvm_skip_emulated_instruction(vcpu);
11390 
11391 	default:
11392 		BUG(); /* We have already checked above that type <= 3 */
11393 	}
11394 }
11395 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
11396 
11397 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
11398 {
11399 	struct kvm_run *run = vcpu->run;
11400 	struct kvm_mmio_fragment *frag;
11401 	unsigned int len;
11402 
11403 	BUG_ON(!vcpu->mmio_needed);
11404 
11405 	/* Complete previous fragment */
11406 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
11407 	len = min(8u, frag->len);
11408 	if (!vcpu->mmio_is_write)
11409 		memcpy(frag->data, run->mmio.data, len);
11410 
11411 	if (frag->len <= 8) {
11412 		/* Switch to the next fragment. */
11413 		frag++;
11414 		vcpu->mmio_cur_fragment++;
11415 	} else {
11416 		/* Go forward to the next mmio piece. */
11417 		frag->data += len;
11418 		frag->gpa += len;
11419 		frag->len -= len;
11420 	}
11421 
11422 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
11423 		vcpu->mmio_needed = 0;
11424 
11425 		// VMG change, at this point, we're always done
11426 		// RIP has already been advanced
11427 		return 1;
11428 	}
11429 
11430 	// More MMIO is needed
11431 	run->mmio.phys_addr = frag->gpa;
11432 	run->mmio.len = min(8u, frag->len);
11433 	run->mmio.is_write = vcpu->mmio_is_write;
11434 	if (run->mmio.is_write)
11435 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
11436 	run->exit_reason = KVM_EXIT_MMIO;
11437 
11438 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11439 
11440 	return 0;
11441 }
11442 
11443 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11444 			  void *data)
11445 {
11446 	int handled;
11447 	struct kvm_mmio_fragment *frag;
11448 
11449 	if (!data)
11450 		return -EINVAL;
11451 
11452 	handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11453 	if (handled == bytes)
11454 		return 1;
11455 
11456 	bytes -= handled;
11457 	gpa += handled;
11458 	data += handled;
11459 
11460 	/*TODO: Check if need to increment number of frags */
11461 	frag = vcpu->mmio_fragments;
11462 	vcpu->mmio_nr_fragments = 1;
11463 	frag->len = bytes;
11464 	frag->gpa = gpa;
11465 	frag->data = data;
11466 
11467 	vcpu->mmio_needed = 1;
11468 	vcpu->mmio_cur_fragment = 0;
11469 
11470 	vcpu->run->mmio.phys_addr = gpa;
11471 	vcpu->run->mmio.len = min(8u, frag->len);
11472 	vcpu->run->mmio.is_write = 1;
11473 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
11474 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
11475 
11476 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11477 
11478 	return 0;
11479 }
11480 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
11481 
11482 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11483 			 void *data)
11484 {
11485 	int handled;
11486 	struct kvm_mmio_fragment *frag;
11487 
11488 	if (!data)
11489 		return -EINVAL;
11490 
11491 	handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11492 	if (handled == bytes)
11493 		return 1;
11494 
11495 	bytes -= handled;
11496 	gpa += handled;
11497 	data += handled;
11498 
11499 	/*TODO: Check if need to increment number of frags */
11500 	frag = vcpu->mmio_fragments;
11501 	vcpu->mmio_nr_fragments = 1;
11502 	frag->len = bytes;
11503 	frag->gpa = gpa;
11504 	frag->data = data;
11505 
11506 	vcpu->mmio_needed = 1;
11507 	vcpu->mmio_cur_fragment = 0;
11508 
11509 	vcpu->run->mmio.phys_addr = gpa;
11510 	vcpu->run->mmio.len = min(8u, frag->len);
11511 	vcpu->run->mmio.is_write = 0;
11512 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
11513 
11514 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11515 
11516 	return 0;
11517 }
11518 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
11519 
11520 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
11521 {
11522 	memcpy(vcpu->arch.guest_ins_data, vcpu->arch.pio_data,
11523 	       vcpu->arch.pio.count * vcpu->arch.pio.size);
11524 	vcpu->arch.pio.count = 0;
11525 
11526 	return 1;
11527 }
11528 
11529 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
11530 			   unsigned int port, void *data,  unsigned int count)
11531 {
11532 	int ret;
11533 
11534 	ret = emulator_pio_out_emulated(vcpu->arch.emulate_ctxt, size, port,
11535 					data, count);
11536 	if (ret)
11537 		return ret;
11538 
11539 	vcpu->arch.pio.count = 0;
11540 
11541 	return 0;
11542 }
11543 
11544 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
11545 			  unsigned int port, void *data, unsigned int count)
11546 {
11547 	int ret;
11548 
11549 	ret = emulator_pio_in_emulated(vcpu->arch.emulate_ctxt, size, port,
11550 				       data, count);
11551 	if (ret) {
11552 		vcpu->arch.pio.count = 0;
11553 	} else {
11554 		vcpu->arch.guest_ins_data = data;
11555 		vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
11556 	}
11557 
11558 	return 0;
11559 }
11560 
11561 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
11562 			 unsigned int port, void *data,  unsigned int count,
11563 			 int in)
11564 {
11565 	return in ? kvm_sev_es_ins(vcpu, size, port, data, count)
11566 		  : kvm_sev_es_outs(vcpu, size, port, data, count);
11567 }
11568 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
11569 
11570 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
11571 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
11572 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
11573 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
11574 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
11575 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
11576 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
11577 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
11578 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
11579 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
11580 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
11581 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
11582 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
11583 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
11584 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
11585 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
11586 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
11587 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
11588 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
11589 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
11590 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
11591 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
11592 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
11593 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
11594 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
11595 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
11596 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);
11597