xref: /linux/arch/x86/kvm/x86.c (revision f4d74ef6220c1eda0875da30457bef5c7111ab06)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * derived from drivers/kvm/kvm_main.c
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright (C) 2008 Qumranet, Inc.
9  * Copyright IBM Corporation, 2008
10  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11  *
12  * Authors:
13  *   Avi Kivity   <avi@qumranet.com>
14  *   Yaniv Kamay  <yaniv@qumranet.com>
15  *   Amit Shah    <amit.shah@qumranet.com>
16  *   Ben-Ami Yassour <benami@il.ibm.com>
17  */
18 
19 #include <linux/kvm_host.h>
20 #include "irq.h"
21 #include "mmu.h"
22 #include "i8254.h"
23 #include "tss.h"
24 #include "kvm_cache_regs.h"
25 #include "kvm_emulate.h"
26 #include "x86.h"
27 #include "cpuid.h"
28 #include "pmu.h"
29 #include "hyperv.h"
30 #include "lapic.h"
31 
32 #include <linux/clocksource.h>
33 #include <linux/interrupt.h>
34 #include <linux/kvm.h>
35 #include <linux/fs.h>
36 #include <linux/vmalloc.h>
37 #include <linux/export.h>
38 #include <linux/moduleparam.h>
39 #include <linux/mman.h>
40 #include <linux/highmem.h>
41 #include <linux/iommu.h>
42 #include <linux/intel-iommu.h>
43 #include <linux/cpufreq.h>
44 #include <linux/user-return-notifier.h>
45 #include <linux/srcu.h>
46 #include <linux/slab.h>
47 #include <linux/perf_event.h>
48 #include <linux/uaccess.h>
49 #include <linux/hash.h>
50 #include <linux/pci.h>
51 #include <linux/timekeeper_internal.h>
52 #include <linux/pvclock_gtod.h>
53 #include <linux/kvm_irqfd.h>
54 #include <linux/irqbypass.h>
55 #include <linux/sched/stat.h>
56 #include <linux/sched/isolation.h>
57 #include <linux/mem_encrypt.h>
58 
59 #include <trace/events/kvm.h>
60 
61 #include <asm/debugreg.h>
62 #include <asm/msr.h>
63 #include <asm/desc.h>
64 #include <asm/mce.h>
65 #include <linux/kernel_stat.h>
66 #include <asm/fpu/internal.h> /* Ugh! */
67 #include <asm/pvclock.h>
68 #include <asm/div64.h>
69 #include <asm/irq_remapping.h>
70 #include <asm/mshyperv.h>
71 #include <asm/hypervisor.h>
72 #include <asm/intel_pt.h>
73 #include <asm/emulate_prefix.h>
74 #include <clocksource/hyperv_timer.h>
75 
76 #define CREATE_TRACE_POINTS
77 #include "trace.h"
78 
79 #define MAX_IO_MSRS 256
80 #define KVM_MAX_MCE_BANKS 32
81 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
82 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
83 
84 #define emul_to_vcpu(ctxt) \
85 	((struct kvm_vcpu *)(ctxt)->vcpu)
86 
87 /* EFER defaults:
88  * - enable syscall per default because its emulated by KVM
89  * - enable LME and LMA per default on 64 bit KVM
90  */
91 #ifdef CONFIG_X86_64
92 static
93 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
94 #else
95 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
96 #endif
97 
98 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
99 
100 #define VM_STAT(x, ...) offsetof(struct kvm, stat.x), KVM_STAT_VM, ## __VA_ARGS__
101 #define VCPU_STAT(x, ...) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU, ## __VA_ARGS__
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 enter_smm(struct kvm_vcpu *vcpu);
109 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
110 static void store_regs(struct kvm_vcpu *vcpu);
111 static int sync_regs(struct kvm_vcpu *vcpu);
112 
113 struct kvm_x86_ops kvm_x86_ops __read_mostly;
114 EXPORT_SYMBOL_GPL(kvm_x86_ops);
115 
116 static bool __read_mostly ignore_msrs = 0;
117 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
118 
119 static bool __read_mostly report_ignored_msrs = true;
120 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
121 
122 unsigned int min_timer_period_us = 200;
123 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
124 
125 static bool __read_mostly kvmclock_periodic_sync = true;
126 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
127 
128 bool __read_mostly kvm_has_tsc_control;
129 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
130 u32  __read_mostly kvm_max_guest_tsc_khz;
131 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
132 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
133 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
134 u64  __read_mostly kvm_max_tsc_scaling_ratio;
135 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
136 u64 __read_mostly kvm_default_tsc_scaling_ratio;
137 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
138 
139 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
140 static u32 __read_mostly tsc_tolerance_ppm = 250;
141 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
142 
143 /*
144  * lapic timer advance (tscdeadline mode only) in nanoseconds.  '-1' enables
145  * adaptive tuning starting from default advancment of 1000ns.  '0' disables
146  * advancement entirely.  Any other value is used as-is and disables adaptive
147  * tuning, i.e. allows priveleged userspace to set an exact advancement time.
148  */
149 static int __read_mostly lapic_timer_advance_ns = -1;
150 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
151 
152 static bool __read_mostly vector_hashing = true;
153 module_param(vector_hashing, bool, S_IRUGO);
154 
155 bool __read_mostly enable_vmware_backdoor = false;
156 module_param(enable_vmware_backdoor, bool, S_IRUGO);
157 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
158 
159 static bool __read_mostly force_emulation_prefix = false;
160 module_param(force_emulation_prefix, bool, S_IRUGO);
161 
162 int __read_mostly pi_inject_timer = -1;
163 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
164 
165 #define KVM_NR_SHARED_MSRS 16
166 
167 struct kvm_shared_msrs_global {
168 	int nr;
169 	u32 msrs[KVM_NR_SHARED_MSRS];
170 };
171 
172 struct kvm_shared_msrs {
173 	struct user_return_notifier urn;
174 	bool registered;
175 	struct kvm_shared_msr_values {
176 		u64 host;
177 		u64 curr;
178 	} values[KVM_NR_SHARED_MSRS];
179 };
180 
181 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
182 static struct kvm_shared_msrs __percpu *shared_msrs;
183 
184 #define KVM_SUPPORTED_XCR0     (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
185 				| XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
186 				| XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
187 				| XFEATURE_MASK_PKRU)
188 
189 u64 __read_mostly host_efer;
190 EXPORT_SYMBOL_GPL(host_efer);
191 
192 static u64 __read_mostly host_xss;
193 u64 __read_mostly supported_xss;
194 EXPORT_SYMBOL_GPL(supported_xss);
195 
196 struct kvm_stats_debugfs_item debugfs_entries[] = {
197 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
198 	{ "pf_guest", VCPU_STAT(pf_guest) },
199 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
200 	{ "invlpg", VCPU_STAT(invlpg) },
201 	{ "exits", VCPU_STAT(exits) },
202 	{ "io_exits", VCPU_STAT(io_exits) },
203 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
204 	{ "signal_exits", VCPU_STAT(signal_exits) },
205 	{ "irq_window", VCPU_STAT(irq_window_exits) },
206 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
207 	{ "halt_exits", VCPU_STAT(halt_exits) },
208 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
209 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
210 	{ "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
211 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
212 	{ "hypercalls", VCPU_STAT(hypercalls) },
213 	{ "request_irq", VCPU_STAT(request_irq_exits) },
214 	{ "irq_exits", VCPU_STAT(irq_exits) },
215 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
216 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
217 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
218 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
219 	{ "irq_injections", VCPU_STAT(irq_injections) },
220 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
221 	{ "req_event", VCPU_STAT(req_event) },
222 	{ "l1d_flush", VCPU_STAT(l1d_flush) },
223 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
224 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
225 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
226 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
227 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
228 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
229 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
230 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
231 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
232 	{ "largepages", VM_STAT(lpages, .mode = 0444) },
233 	{ "nx_largepages_splitted", VM_STAT(nx_lpage_splits, .mode = 0444) },
234 	{ "max_mmu_page_hash_collisions",
235 		VM_STAT(max_mmu_page_hash_collisions) },
236 	{ NULL }
237 };
238 
239 u64 __read_mostly host_xcr0;
240 u64 __read_mostly supported_xcr0;
241 EXPORT_SYMBOL_GPL(supported_xcr0);
242 
243 struct kmem_cache *x86_fpu_cache;
244 EXPORT_SYMBOL_GPL(x86_fpu_cache);
245 
246 static struct kmem_cache *x86_emulator_cache;
247 
248 static struct kmem_cache *kvm_alloc_emulator_cache(void)
249 {
250 	unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
251 	unsigned int size = sizeof(struct x86_emulate_ctxt);
252 
253 	return kmem_cache_create_usercopy("x86_emulator", size,
254 					  __alignof__(struct x86_emulate_ctxt),
255 					  SLAB_ACCOUNT, useroffset,
256 					  size - useroffset, NULL);
257 }
258 
259 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
260 
261 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
262 {
263 	int i;
264 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
265 		vcpu->arch.apf.gfns[i] = ~0;
266 }
267 
268 static void kvm_on_user_return(struct user_return_notifier *urn)
269 {
270 	unsigned slot;
271 	struct kvm_shared_msrs *locals
272 		= container_of(urn, struct kvm_shared_msrs, urn);
273 	struct kvm_shared_msr_values *values;
274 	unsigned long flags;
275 
276 	/*
277 	 * Disabling irqs at this point since the following code could be
278 	 * interrupted and executed through kvm_arch_hardware_disable()
279 	 */
280 	local_irq_save(flags);
281 	if (locals->registered) {
282 		locals->registered = false;
283 		user_return_notifier_unregister(urn);
284 	}
285 	local_irq_restore(flags);
286 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
287 		values = &locals->values[slot];
288 		if (values->host != values->curr) {
289 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
290 			values->curr = values->host;
291 		}
292 	}
293 }
294 
295 void kvm_define_shared_msr(unsigned slot, u32 msr)
296 {
297 	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
298 	shared_msrs_global.msrs[slot] = msr;
299 	if (slot >= shared_msrs_global.nr)
300 		shared_msrs_global.nr = slot + 1;
301 }
302 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
303 
304 static void kvm_shared_msr_cpu_online(void)
305 {
306 	unsigned int cpu = smp_processor_id();
307 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
308 	u64 value;
309 	int i;
310 
311 	for (i = 0; i < shared_msrs_global.nr; ++i) {
312 		rdmsrl_safe(shared_msrs_global.msrs[i], &value);
313 		smsr->values[i].host = value;
314 		smsr->values[i].curr = value;
315 	}
316 }
317 
318 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
319 {
320 	unsigned int cpu = smp_processor_id();
321 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
322 	int err;
323 
324 	value = (value & mask) | (smsr->values[slot].host & ~mask);
325 	if (value == smsr->values[slot].curr)
326 		return 0;
327 	err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
328 	if (err)
329 		return 1;
330 
331 	smsr->values[slot].curr = value;
332 	if (!smsr->registered) {
333 		smsr->urn.on_user_return = kvm_on_user_return;
334 		user_return_notifier_register(&smsr->urn);
335 		smsr->registered = true;
336 	}
337 	return 0;
338 }
339 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
340 
341 static void drop_user_return_notifiers(void)
342 {
343 	unsigned int cpu = smp_processor_id();
344 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
345 
346 	if (smsr->registered)
347 		kvm_on_user_return(&smsr->urn);
348 }
349 
350 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
351 {
352 	return vcpu->arch.apic_base;
353 }
354 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
355 
356 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
357 {
358 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
359 }
360 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
361 
362 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
363 {
364 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
365 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
366 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
367 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
368 
369 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
370 		return 1;
371 	if (!msr_info->host_initiated) {
372 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
373 			return 1;
374 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
375 			return 1;
376 	}
377 
378 	kvm_lapic_set_base(vcpu, msr_info->data);
379 	kvm_recalculate_apic_map(vcpu->kvm);
380 	return 0;
381 }
382 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
383 
384 asmlinkage __visible void kvm_spurious_fault(void)
385 {
386 	/* Fault while not rebooting.  We want the trace. */
387 	BUG_ON(!kvm_rebooting);
388 }
389 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
390 
391 #define EXCPT_BENIGN		0
392 #define EXCPT_CONTRIBUTORY	1
393 #define EXCPT_PF		2
394 
395 static int exception_class(int vector)
396 {
397 	switch (vector) {
398 	case PF_VECTOR:
399 		return EXCPT_PF;
400 	case DE_VECTOR:
401 	case TS_VECTOR:
402 	case NP_VECTOR:
403 	case SS_VECTOR:
404 	case GP_VECTOR:
405 		return EXCPT_CONTRIBUTORY;
406 	default:
407 		break;
408 	}
409 	return EXCPT_BENIGN;
410 }
411 
412 #define EXCPT_FAULT		0
413 #define EXCPT_TRAP		1
414 #define EXCPT_ABORT		2
415 #define EXCPT_INTERRUPT		3
416 
417 static int exception_type(int vector)
418 {
419 	unsigned int mask;
420 
421 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
422 		return EXCPT_INTERRUPT;
423 
424 	mask = 1 << vector;
425 
426 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
427 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
428 		return EXCPT_TRAP;
429 
430 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
431 		return EXCPT_ABORT;
432 
433 	/* Reserved exceptions will result in fault */
434 	return EXCPT_FAULT;
435 }
436 
437 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
438 {
439 	unsigned nr = vcpu->arch.exception.nr;
440 	bool has_payload = vcpu->arch.exception.has_payload;
441 	unsigned long payload = vcpu->arch.exception.payload;
442 
443 	if (!has_payload)
444 		return;
445 
446 	switch (nr) {
447 	case DB_VECTOR:
448 		/*
449 		 * "Certain debug exceptions may clear bit 0-3.  The
450 		 * remaining contents of the DR6 register are never
451 		 * cleared by the processor".
452 		 */
453 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
454 		/*
455 		 * DR6.RTM is set by all #DB exceptions that don't clear it.
456 		 */
457 		vcpu->arch.dr6 |= DR6_RTM;
458 		vcpu->arch.dr6 |= payload;
459 		/*
460 		 * Bit 16 should be set in the payload whenever the #DB
461 		 * exception should clear DR6.RTM. This makes the payload
462 		 * compatible with the pending debug exceptions under VMX.
463 		 * Though not currently documented in the SDM, this also
464 		 * makes the payload compatible with the exit qualification
465 		 * for #DB exceptions under VMX.
466 		 */
467 		vcpu->arch.dr6 ^= payload & DR6_RTM;
468 
469 		/*
470 		 * The #DB payload is defined as compatible with the 'pending
471 		 * debug exceptions' field under VMX, not DR6. While bit 12 is
472 		 * defined in the 'pending debug exceptions' field (enabled
473 		 * breakpoint), it is reserved and must be zero in DR6.
474 		 */
475 		vcpu->arch.dr6 &= ~BIT(12);
476 		break;
477 	case PF_VECTOR:
478 		vcpu->arch.cr2 = payload;
479 		break;
480 	}
481 
482 	vcpu->arch.exception.has_payload = false;
483 	vcpu->arch.exception.payload = 0;
484 }
485 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
486 
487 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
488 		unsigned nr, bool has_error, u32 error_code,
489 	        bool has_payload, unsigned long payload, bool reinject)
490 {
491 	u32 prev_nr;
492 	int class1, class2;
493 
494 	kvm_make_request(KVM_REQ_EVENT, vcpu);
495 
496 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
497 	queue:
498 		if (has_error && !is_protmode(vcpu))
499 			has_error = false;
500 		if (reinject) {
501 			/*
502 			 * On vmentry, vcpu->arch.exception.pending is only
503 			 * true if an event injection was blocked by
504 			 * nested_run_pending.  In that case, however,
505 			 * vcpu_enter_guest requests an immediate exit,
506 			 * and the guest shouldn't proceed far enough to
507 			 * need reinjection.
508 			 */
509 			WARN_ON_ONCE(vcpu->arch.exception.pending);
510 			vcpu->arch.exception.injected = true;
511 			if (WARN_ON_ONCE(has_payload)) {
512 				/*
513 				 * A reinjected event has already
514 				 * delivered its payload.
515 				 */
516 				has_payload = false;
517 				payload = 0;
518 			}
519 		} else {
520 			vcpu->arch.exception.pending = true;
521 			vcpu->arch.exception.injected = false;
522 		}
523 		vcpu->arch.exception.has_error_code = has_error;
524 		vcpu->arch.exception.nr = nr;
525 		vcpu->arch.exception.error_code = error_code;
526 		vcpu->arch.exception.has_payload = has_payload;
527 		vcpu->arch.exception.payload = payload;
528 		if (!is_guest_mode(vcpu))
529 			kvm_deliver_exception_payload(vcpu);
530 		return;
531 	}
532 
533 	/* to check exception */
534 	prev_nr = vcpu->arch.exception.nr;
535 	if (prev_nr == DF_VECTOR) {
536 		/* triple fault -> shutdown */
537 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
538 		return;
539 	}
540 	class1 = exception_class(prev_nr);
541 	class2 = exception_class(nr);
542 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
543 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
544 		/*
545 		 * Generate double fault per SDM Table 5-5.  Set
546 		 * exception.pending = true so that the double fault
547 		 * can trigger a nested vmexit.
548 		 */
549 		vcpu->arch.exception.pending = true;
550 		vcpu->arch.exception.injected = false;
551 		vcpu->arch.exception.has_error_code = true;
552 		vcpu->arch.exception.nr = DF_VECTOR;
553 		vcpu->arch.exception.error_code = 0;
554 		vcpu->arch.exception.has_payload = false;
555 		vcpu->arch.exception.payload = 0;
556 	} else
557 		/* replace previous exception with a new one in a hope
558 		   that instruction re-execution will regenerate lost
559 		   exception */
560 		goto queue;
561 }
562 
563 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
564 {
565 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
566 }
567 EXPORT_SYMBOL_GPL(kvm_queue_exception);
568 
569 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
570 {
571 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
572 }
573 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
574 
575 static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
576 				  unsigned long payload)
577 {
578 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
579 }
580 
581 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
582 				    u32 error_code, unsigned long payload)
583 {
584 	kvm_multiple_exception(vcpu, nr, true, error_code,
585 			       true, payload, false);
586 }
587 
588 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
589 {
590 	if (err)
591 		kvm_inject_gp(vcpu, 0);
592 	else
593 		return kvm_skip_emulated_instruction(vcpu);
594 
595 	return 1;
596 }
597 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
598 
599 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
600 {
601 	++vcpu->stat.pf_guest;
602 	vcpu->arch.exception.nested_apf =
603 		is_guest_mode(vcpu) && fault->async_page_fault;
604 	if (vcpu->arch.exception.nested_apf) {
605 		vcpu->arch.apf.nested_apf_token = fault->address;
606 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
607 	} else {
608 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
609 					fault->address);
610 	}
611 }
612 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
613 
614 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
615 {
616 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
617 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
618 	else
619 		vcpu->arch.mmu->inject_page_fault(vcpu, fault);
620 
621 	return fault->nested_page_fault;
622 }
623 
624 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
625 {
626 	atomic_inc(&vcpu->arch.nmi_queued);
627 	kvm_make_request(KVM_REQ_NMI, vcpu);
628 }
629 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
630 
631 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
632 {
633 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
634 }
635 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
636 
637 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
638 {
639 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
640 }
641 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
642 
643 /*
644  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
645  * a #GP and return false.
646  */
647 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
648 {
649 	if (kvm_x86_ops.get_cpl(vcpu) <= required_cpl)
650 		return true;
651 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
652 	return false;
653 }
654 EXPORT_SYMBOL_GPL(kvm_require_cpl);
655 
656 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
657 {
658 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
659 		return true;
660 
661 	kvm_queue_exception(vcpu, UD_VECTOR);
662 	return false;
663 }
664 EXPORT_SYMBOL_GPL(kvm_require_dr);
665 
666 /*
667  * This function will be used to read from the physical memory of the currently
668  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
669  * can read from guest physical or from the guest's guest physical memory.
670  */
671 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
672 			    gfn_t ngfn, void *data, int offset, int len,
673 			    u32 access)
674 {
675 	struct x86_exception exception;
676 	gfn_t real_gfn;
677 	gpa_t ngpa;
678 
679 	ngpa     = gfn_to_gpa(ngfn);
680 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
681 	if (real_gfn == UNMAPPED_GVA)
682 		return -EFAULT;
683 
684 	real_gfn = gpa_to_gfn(real_gfn);
685 
686 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
687 }
688 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
689 
690 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
691 			       void *data, int offset, int len, u32 access)
692 {
693 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
694 				       data, offset, len, access);
695 }
696 
697 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
698 {
699 	return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
700 	       rsvd_bits(1, 2);
701 }
702 
703 /*
704  * Load the pae pdptrs.  Return 1 if they are all valid, 0 otherwise.
705  */
706 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
707 {
708 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
709 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
710 	int i;
711 	int ret;
712 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
713 
714 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
715 				      offset * sizeof(u64), sizeof(pdpte),
716 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
717 	if (ret < 0) {
718 		ret = 0;
719 		goto out;
720 	}
721 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
722 		if ((pdpte[i] & PT_PRESENT_MASK) &&
723 		    (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
724 			ret = 0;
725 			goto out;
726 		}
727 	}
728 	ret = 1;
729 
730 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
731 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
732 
733 out:
734 
735 	return ret;
736 }
737 EXPORT_SYMBOL_GPL(load_pdptrs);
738 
739 bool pdptrs_changed(struct kvm_vcpu *vcpu)
740 {
741 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
742 	int offset;
743 	gfn_t gfn;
744 	int r;
745 
746 	if (!is_pae_paging(vcpu))
747 		return false;
748 
749 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
750 		return true;
751 
752 	gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
753 	offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
754 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
755 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
756 	if (r < 0)
757 		return true;
758 
759 	return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
760 }
761 EXPORT_SYMBOL_GPL(pdptrs_changed);
762 
763 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
764 {
765 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
766 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
767 
768 	cr0 |= X86_CR0_ET;
769 
770 #ifdef CONFIG_X86_64
771 	if (cr0 & 0xffffffff00000000UL)
772 		return 1;
773 #endif
774 
775 	cr0 &= ~CR0_RESERVED_BITS;
776 
777 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
778 		return 1;
779 
780 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
781 		return 1;
782 
783 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
784 #ifdef CONFIG_X86_64
785 		if ((vcpu->arch.efer & EFER_LME)) {
786 			int cs_db, cs_l;
787 
788 			if (!is_pae(vcpu))
789 				return 1;
790 			kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
791 			if (cs_l)
792 				return 1;
793 		} else
794 #endif
795 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
796 						 kvm_read_cr3(vcpu)))
797 			return 1;
798 	}
799 
800 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
801 		return 1;
802 
803 	kvm_x86_ops.set_cr0(vcpu, cr0);
804 
805 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
806 		kvm_clear_async_pf_completion_queue(vcpu);
807 		kvm_async_pf_hash_reset(vcpu);
808 	}
809 
810 	if ((cr0 ^ old_cr0) & update_bits)
811 		kvm_mmu_reset_context(vcpu);
812 
813 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
814 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
815 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
816 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
817 
818 	return 0;
819 }
820 EXPORT_SYMBOL_GPL(kvm_set_cr0);
821 
822 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
823 {
824 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
825 }
826 EXPORT_SYMBOL_GPL(kvm_lmsw);
827 
828 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
829 {
830 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
831 
832 		if (vcpu->arch.xcr0 != host_xcr0)
833 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
834 
835 		if (vcpu->arch.xsaves_enabled &&
836 		    vcpu->arch.ia32_xss != host_xss)
837 			wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
838 	}
839 }
840 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
841 
842 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
843 {
844 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
845 
846 		if (vcpu->arch.xcr0 != host_xcr0)
847 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
848 
849 		if (vcpu->arch.xsaves_enabled &&
850 		    vcpu->arch.ia32_xss != host_xss)
851 			wrmsrl(MSR_IA32_XSS, host_xss);
852 	}
853 
854 }
855 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
856 
857 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
858 {
859 	u64 xcr0 = xcr;
860 	u64 old_xcr0 = vcpu->arch.xcr0;
861 	u64 valid_bits;
862 
863 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
864 	if (index != XCR_XFEATURE_ENABLED_MASK)
865 		return 1;
866 	if (!(xcr0 & XFEATURE_MASK_FP))
867 		return 1;
868 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
869 		return 1;
870 
871 	/*
872 	 * Do not allow the guest to set bits that we do not support
873 	 * saving.  However, xcr0 bit 0 is always set, even if the
874 	 * emulated CPU does not support XSAVE (see fx_init).
875 	 */
876 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
877 	if (xcr0 & ~valid_bits)
878 		return 1;
879 
880 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
881 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
882 		return 1;
883 
884 	if (xcr0 & XFEATURE_MASK_AVX512) {
885 		if (!(xcr0 & XFEATURE_MASK_YMM))
886 			return 1;
887 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
888 			return 1;
889 	}
890 	vcpu->arch.xcr0 = xcr0;
891 
892 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
893 		kvm_update_cpuid(vcpu);
894 	return 0;
895 }
896 
897 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
898 {
899 	if (kvm_x86_ops.get_cpl(vcpu) != 0 ||
900 	    __kvm_set_xcr(vcpu, index, xcr)) {
901 		kvm_inject_gp(vcpu, 0);
902 		return 1;
903 	}
904 	return 0;
905 }
906 EXPORT_SYMBOL_GPL(kvm_set_xcr);
907 
908 #define __cr4_reserved_bits(__cpu_has, __c)		\
909 ({							\
910 	u64 __reserved_bits = CR4_RESERVED_BITS;	\
911 							\
912 	if (!__cpu_has(__c, X86_FEATURE_XSAVE))		\
913 		__reserved_bits |= X86_CR4_OSXSAVE;	\
914 	if (!__cpu_has(__c, X86_FEATURE_SMEP))		\
915 		__reserved_bits |= X86_CR4_SMEP;	\
916 	if (!__cpu_has(__c, X86_FEATURE_SMAP))		\
917 		__reserved_bits |= X86_CR4_SMAP;	\
918 	if (!__cpu_has(__c, X86_FEATURE_FSGSBASE))	\
919 		__reserved_bits |= X86_CR4_FSGSBASE;	\
920 	if (!__cpu_has(__c, X86_FEATURE_PKU))		\
921 		__reserved_bits |= X86_CR4_PKE;		\
922 	if (!__cpu_has(__c, X86_FEATURE_LA57))		\
923 		__reserved_bits |= X86_CR4_LA57;	\
924 	if (!__cpu_has(__c, X86_FEATURE_UMIP))		\
925 		__reserved_bits |= X86_CR4_UMIP;	\
926 	__reserved_bits;				\
927 })
928 
929 static u64 kvm_host_cr4_reserved_bits(struct cpuinfo_x86 *c)
930 {
931 	u64 reserved_bits = __cr4_reserved_bits(cpu_has, c);
932 
933 	if (kvm_cpu_cap_has(X86_FEATURE_LA57))
934 		reserved_bits &= ~X86_CR4_LA57;
935 
936 	if (kvm_cpu_cap_has(X86_FEATURE_UMIP))
937 		reserved_bits &= ~X86_CR4_UMIP;
938 
939 	return reserved_bits;
940 }
941 
942 static int kvm_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
943 {
944 	if (cr4 & cr4_reserved_bits)
945 		return -EINVAL;
946 
947 	if (cr4 & __cr4_reserved_bits(guest_cpuid_has, vcpu))
948 		return -EINVAL;
949 
950 	return 0;
951 }
952 
953 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
954 {
955 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
956 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
957 				   X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
958 
959 	if (kvm_valid_cr4(vcpu, cr4))
960 		return 1;
961 
962 	if (is_long_mode(vcpu)) {
963 		if (!(cr4 & X86_CR4_PAE))
964 			return 1;
965 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
966 		   && ((cr4 ^ old_cr4) & pdptr_bits)
967 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
968 				   kvm_read_cr3(vcpu)))
969 		return 1;
970 
971 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
972 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
973 			return 1;
974 
975 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
976 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
977 			return 1;
978 	}
979 
980 	if (kvm_x86_ops.set_cr4(vcpu, cr4))
981 		return 1;
982 
983 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
984 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
985 		kvm_mmu_reset_context(vcpu);
986 
987 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
988 		kvm_update_cpuid(vcpu);
989 
990 	return 0;
991 }
992 EXPORT_SYMBOL_GPL(kvm_set_cr4);
993 
994 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
995 {
996 	bool skip_tlb_flush = false;
997 #ifdef CONFIG_X86_64
998 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
999 
1000 	if (pcid_enabled) {
1001 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1002 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
1003 	}
1004 #endif
1005 
1006 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
1007 		if (!skip_tlb_flush) {
1008 			kvm_mmu_sync_roots(vcpu);
1009 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1010 		}
1011 		return 0;
1012 	}
1013 
1014 	if (is_long_mode(vcpu) &&
1015 	    (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
1016 		return 1;
1017 	else if (is_pae_paging(vcpu) &&
1018 		 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1019 		return 1;
1020 
1021 	kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
1022 	vcpu->arch.cr3 = cr3;
1023 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1024 
1025 	return 0;
1026 }
1027 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1028 
1029 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1030 {
1031 	if (cr8 & CR8_RESERVED_BITS)
1032 		return 1;
1033 	if (lapic_in_kernel(vcpu))
1034 		kvm_lapic_set_tpr(vcpu, cr8);
1035 	else
1036 		vcpu->arch.cr8 = cr8;
1037 	return 0;
1038 }
1039 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1040 
1041 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1042 {
1043 	if (lapic_in_kernel(vcpu))
1044 		return kvm_lapic_get_cr8(vcpu);
1045 	else
1046 		return vcpu->arch.cr8;
1047 }
1048 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1049 
1050 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1051 {
1052 	int i;
1053 
1054 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1055 		for (i = 0; i < KVM_NR_DB_REGS; i++)
1056 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1057 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1058 	}
1059 }
1060 
1061 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1062 {
1063 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1064 		kvm_x86_ops.set_dr6(vcpu, vcpu->arch.dr6);
1065 }
1066 
1067 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1068 {
1069 	unsigned long dr7;
1070 
1071 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1072 		dr7 = vcpu->arch.guest_debug_dr7;
1073 	else
1074 		dr7 = vcpu->arch.dr7;
1075 	kvm_x86_ops.set_dr7(vcpu, dr7);
1076 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1077 	if (dr7 & DR7_BP_EN_MASK)
1078 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1079 }
1080 
1081 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1082 {
1083 	u64 fixed = DR6_FIXED_1;
1084 
1085 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1086 		fixed |= DR6_RTM;
1087 	return fixed;
1088 }
1089 
1090 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1091 {
1092 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1093 
1094 	switch (dr) {
1095 	case 0 ... 3:
1096 		vcpu->arch.db[array_index_nospec(dr, size)] = val;
1097 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1098 			vcpu->arch.eff_db[dr] = val;
1099 		break;
1100 	case 4:
1101 		/* fall through */
1102 	case 6:
1103 		if (val & 0xffffffff00000000ULL)
1104 			return -1; /* #GP */
1105 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1106 		kvm_update_dr6(vcpu);
1107 		break;
1108 	case 5:
1109 		/* fall through */
1110 	default: /* 7 */
1111 		if (!kvm_dr7_valid(val))
1112 			return -1; /* #GP */
1113 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1114 		kvm_update_dr7(vcpu);
1115 		break;
1116 	}
1117 
1118 	return 0;
1119 }
1120 
1121 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1122 {
1123 	if (__kvm_set_dr(vcpu, dr, val)) {
1124 		kvm_inject_gp(vcpu, 0);
1125 		return 1;
1126 	}
1127 	return 0;
1128 }
1129 EXPORT_SYMBOL_GPL(kvm_set_dr);
1130 
1131 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1132 {
1133 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1134 
1135 	switch (dr) {
1136 	case 0 ... 3:
1137 		*val = vcpu->arch.db[array_index_nospec(dr, size)];
1138 		break;
1139 	case 4:
1140 		/* fall through */
1141 	case 6:
1142 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1143 			*val = vcpu->arch.dr6;
1144 		else
1145 			*val = kvm_x86_ops.get_dr6(vcpu);
1146 		break;
1147 	case 5:
1148 		/* fall through */
1149 	default: /* 7 */
1150 		*val = vcpu->arch.dr7;
1151 		break;
1152 	}
1153 	return 0;
1154 }
1155 EXPORT_SYMBOL_GPL(kvm_get_dr);
1156 
1157 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1158 {
1159 	u32 ecx = kvm_rcx_read(vcpu);
1160 	u64 data;
1161 	int err;
1162 
1163 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1164 	if (err)
1165 		return err;
1166 	kvm_rax_write(vcpu, (u32)data);
1167 	kvm_rdx_write(vcpu, data >> 32);
1168 	return err;
1169 }
1170 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1171 
1172 /*
1173  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1174  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1175  *
1176  * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1177  * extract the supported MSRs from the related const lists.
1178  * msrs_to_save is selected from the msrs_to_save_all to reflect the
1179  * capabilities of the host cpu. This capabilities test skips MSRs that are
1180  * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1181  * may depend on host virtualization features rather than host cpu features.
1182  */
1183 
1184 static const u32 msrs_to_save_all[] = {
1185 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1186 	MSR_STAR,
1187 #ifdef CONFIG_X86_64
1188 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1189 #endif
1190 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1191 	MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1192 	MSR_IA32_SPEC_CTRL,
1193 	MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1194 	MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1195 	MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1196 	MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1197 	MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1198 	MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1199 	MSR_IA32_UMWAIT_CONTROL,
1200 
1201 	MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1202 	MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1203 	MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1204 	MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1205 	MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1206 	MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1207 	MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1208 	MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1209 	MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1210 	MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1211 	MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1212 	MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1213 	MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1214 	MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1215 	MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1216 	MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1217 	MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1218 	MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1219 	MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1220 	MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1221 	MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1222 	MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1223 };
1224 
1225 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1226 static unsigned num_msrs_to_save;
1227 
1228 static const u32 emulated_msrs_all[] = {
1229 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1230 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1231 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1232 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1233 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1234 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1235 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1236 	HV_X64_MSR_RESET,
1237 	HV_X64_MSR_VP_INDEX,
1238 	HV_X64_MSR_VP_RUNTIME,
1239 	HV_X64_MSR_SCONTROL,
1240 	HV_X64_MSR_STIMER0_CONFIG,
1241 	HV_X64_MSR_VP_ASSIST_PAGE,
1242 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1243 	HV_X64_MSR_TSC_EMULATION_STATUS,
1244 
1245 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1246 	MSR_KVM_PV_EOI_EN,
1247 
1248 	MSR_IA32_TSC_ADJUST,
1249 	MSR_IA32_TSCDEADLINE,
1250 	MSR_IA32_ARCH_CAPABILITIES,
1251 	MSR_IA32_MISC_ENABLE,
1252 	MSR_IA32_MCG_STATUS,
1253 	MSR_IA32_MCG_CTL,
1254 	MSR_IA32_MCG_EXT_CTL,
1255 	MSR_IA32_SMBASE,
1256 	MSR_SMI_COUNT,
1257 	MSR_PLATFORM_INFO,
1258 	MSR_MISC_FEATURES_ENABLES,
1259 	MSR_AMD64_VIRT_SPEC_CTRL,
1260 	MSR_IA32_POWER_CTL,
1261 	MSR_IA32_UCODE_REV,
1262 
1263 	/*
1264 	 * The following list leaves out MSRs whose values are determined
1265 	 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1266 	 * We always support the "true" VMX control MSRs, even if the host
1267 	 * processor does not, so I am putting these registers here rather
1268 	 * than in msrs_to_save_all.
1269 	 */
1270 	MSR_IA32_VMX_BASIC,
1271 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1272 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1273 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1274 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1275 	MSR_IA32_VMX_MISC,
1276 	MSR_IA32_VMX_CR0_FIXED0,
1277 	MSR_IA32_VMX_CR4_FIXED0,
1278 	MSR_IA32_VMX_VMCS_ENUM,
1279 	MSR_IA32_VMX_PROCBASED_CTLS2,
1280 	MSR_IA32_VMX_EPT_VPID_CAP,
1281 	MSR_IA32_VMX_VMFUNC,
1282 
1283 	MSR_K7_HWCR,
1284 	MSR_KVM_POLL_CONTROL,
1285 };
1286 
1287 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1288 static unsigned num_emulated_msrs;
1289 
1290 /*
1291  * List of msr numbers which are used to expose MSR-based features that
1292  * can be used by a hypervisor to validate requested CPU features.
1293  */
1294 static const u32 msr_based_features_all[] = {
1295 	MSR_IA32_VMX_BASIC,
1296 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1297 	MSR_IA32_VMX_PINBASED_CTLS,
1298 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1299 	MSR_IA32_VMX_PROCBASED_CTLS,
1300 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1301 	MSR_IA32_VMX_EXIT_CTLS,
1302 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1303 	MSR_IA32_VMX_ENTRY_CTLS,
1304 	MSR_IA32_VMX_MISC,
1305 	MSR_IA32_VMX_CR0_FIXED0,
1306 	MSR_IA32_VMX_CR0_FIXED1,
1307 	MSR_IA32_VMX_CR4_FIXED0,
1308 	MSR_IA32_VMX_CR4_FIXED1,
1309 	MSR_IA32_VMX_VMCS_ENUM,
1310 	MSR_IA32_VMX_PROCBASED_CTLS2,
1311 	MSR_IA32_VMX_EPT_VPID_CAP,
1312 	MSR_IA32_VMX_VMFUNC,
1313 
1314 	MSR_F10H_DECFG,
1315 	MSR_IA32_UCODE_REV,
1316 	MSR_IA32_ARCH_CAPABILITIES,
1317 };
1318 
1319 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1320 static unsigned int num_msr_based_features;
1321 
1322 static u64 kvm_get_arch_capabilities(void)
1323 {
1324 	u64 data = 0;
1325 
1326 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1327 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1328 
1329 	/*
1330 	 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1331 	 * the nested hypervisor runs with NX huge pages.  If it is not,
1332 	 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1333 	 * L1 guests, so it need not worry about its own (L2) guests.
1334 	 */
1335 	data |= ARCH_CAP_PSCHANGE_MC_NO;
1336 
1337 	/*
1338 	 * If we're doing cache flushes (either "always" or "cond")
1339 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1340 	 * If an outer hypervisor is doing the cache flush for us
1341 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1342 	 * capability to the guest too, and if EPT is disabled we're not
1343 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1344 	 * require a nested hypervisor to do a flush of its own.
1345 	 */
1346 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1347 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1348 
1349 	if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1350 		data |= ARCH_CAP_RDCL_NO;
1351 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1352 		data |= ARCH_CAP_SSB_NO;
1353 	if (!boot_cpu_has_bug(X86_BUG_MDS))
1354 		data |= ARCH_CAP_MDS_NO;
1355 
1356 	/*
1357 	 * On TAA affected systems:
1358 	 *      - nothing to do if TSX is disabled on the host.
1359 	 *      - we emulate TSX_CTRL if present on the host.
1360 	 *	  This lets the guest use VERW to clear CPU buffers.
1361 	 */
1362 	if (!boot_cpu_has(X86_FEATURE_RTM))
1363 		data &= ~(ARCH_CAP_TAA_NO | ARCH_CAP_TSX_CTRL_MSR);
1364 	else if (!boot_cpu_has_bug(X86_BUG_TAA))
1365 		data |= ARCH_CAP_TAA_NO;
1366 
1367 	return data;
1368 }
1369 
1370 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1371 {
1372 	switch (msr->index) {
1373 	case MSR_IA32_ARCH_CAPABILITIES:
1374 		msr->data = kvm_get_arch_capabilities();
1375 		break;
1376 	case MSR_IA32_UCODE_REV:
1377 		rdmsrl_safe(msr->index, &msr->data);
1378 		break;
1379 	default:
1380 		if (kvm_x86_ops.get_msr_feature(msr))
1381 			return 1;
1382 	}
1383 	return 0;
1384 }
1385 
1386 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1387 {
1388 	struct kvm_msr_entry msr;
1389 	int r;
1390 
1391 	msr.index = index;
1392 	r = kvm_get_msr_feature(&msr);
1393 	if (r)
1394 		return r;
1395 
1396 	*data = msr.data;
1397 
1398 	return 0;
1399 }
1400 
1401 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1402 {
1403 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1404 		return false;
1405 
1406 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1407 		return false;
1408 
1409 	if (efer & (EFER_LME | EFER_LMA) &&
1410 	    !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1411 		return false;
1412 
1413 	if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1414 		return false;
1415 
1416 	return true;
1417 
1418 }
1419 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1420 {
1421 	if (efer & efer_reserved_bits)
1422 		return false;
1423 
1424 	return __kvm_valid_efer(vcpu, efer);
1425 }
1426 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1427 
1428 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1429 {
1430 	u64 old_efer = vcpu->arch.efer;
1431 	u64 efer = msr_info->data;
1432 
1433 	if (efer & efer_reserved_bits)
1434 		return 1;
1435 
1436 	if (!msr_info->host_initiated) {
1437 		if (!__kvm_valid_efer(vcpu, efer))
1438 			return 1;
1439 
1440 		if (is_paging(vcpu) &&
1441 		    (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1442 			return 1;
1443 	}
1444 
1445 	efer &= ~EFER_LMA;
1446 	efer |= vcpu->arch.efer & EFER_LMA;
1447 
1448 	kvm_x86_ops.set_efer(vcpu, efer);
1449 
1450 	/* Update reserved bits */
1451 	if ((efer ^ old_efer) & EFER_NX)
1452 		kvm_mmu_reset_context(vcpu);
1453 
1454 	return 0;
1455 }
1456 
1457 void kvm_enable_efer_bits(u64 mask)
1458 {
1459        efer_reserved_bits &= ~mask;
1460 }
1461 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1462 
1463 /*
1464  * Write @data into the MSR specified by @index.  Select MSR specific fault
1465  * checks are bypassed if @host_initiated is %true.
1466  * Returns 0 on success, non-0 otherwise.
1467  * Assumes vcpu_load() was already called.
1468  */
1469 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1470 			 bool host_initiated)
1471 {
1472 	struct msr_data msr;
1473 
1474 	switch (index) {
1475 	case MSR_FS_BASE:
1476 	case MSR_GS_BASE:
1477 	case MSR_KERNEL_GS_BASE:
1478 	case MSR_CSTAR:
1479 	case MSR_LSTAR:
1480 		if (is_noncanonical_address(data, vcpu))
1481 			return 1;
1482 		break;
1483 	case MSR_IA32_SYSENTER_EIP:
1484 	case MSR_IA32_SYSENTER_ESP:
1485 		/*
1486 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1487 		 * non-canonical address is written on Intel but not on
1488 		 * AMD (which ignores the top 32-bits, because it does
1489 		 * not implement 64-bit SYSENTER).
1490 		 *
1491 		 * 64-bit code should hence be able to write a non-canonical
1492 		 * value on AMD.  Making the address canonical ensures that
1493 		 * vmentry does not fail on Intel after writing a non-canonical
1494 		 * value, and that something deterministic happens if the guest
1495 		 * invokes 64-bit SYSENTER.
1496 		 */
1497 		data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1498 	}
1499 
1500 	msr.data = data;
1501 	msr.index = index;
1502 	msr.host_initiated = host_initiated;
1503 
1504 	return kvm_x86_ops.set_msr(vcpu, &msr);
1505 }
1506 
1507 /*
1508  * Read the MSR specified by @index into @data.  Select MSR specific fault
1509  * checks are bypassed if @host_initiated is %true.
1510  * Returns 0 on success, non-0 otherwise.
1511  * Assumes vcpu_load() was already called.
1512  */
1513 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1514 		  bool host_initiated)
1515 {
1516 	struct msr_data msr;
1517 	int ret;
1518 
1519 	msr.index = index;
1520 	msr.host_initiated = host_initiated;
1521 
1522 	ret = kvm_x86_ops.get_msr(vcpu, &msr);
1523 	if (!ret)
1524 		*data = msr.data;
1525 	return ret;
1526 }
1527 
1528 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1529 {
1530 	return __kvm_get_msr(vcpu, index, data, false);
1531 }
1532 EXPORT_SYMBOL_GPL(kvm_get_msr);
1533 
1534 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1535 {
1536 	return __kvm_set_msr(vcpu, index, data, false);
1537 }
1538 EXPORT_SYMBOL_GPL(kvm_set_msr);
1539 
1540 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1541 {
1542 	u32 ecx = kvm_rcx_read(vcpu);
1543 	u64 data;
1544 
1545 	if (kvm_get_msr(vcpu, ecx, &data)) {
1546 		trace_kvm_msr_read_ex(ecx);
1547 		kvm_inject_gp(vcpu, 0);
1548 		return 1;
1549 	}
1550 
1551 	trace_kvm_msr_read(ecx, data);
1552 
1553 	kvm_rax_write(vcpu, data & -1u);
1554 	kvm_rdx_write(vcpu, (data >> 32) & -1u);
1555 	return kvm_skip_emulated_instruction(vcpu);
1556 }
1557 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1558 
1559 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1560 {
1561 	u32 ecx = kvm_rcx_read(vcpu);
1562 	u64 data = kvm_read_edx_eax(vcpu);
1563 
1564 	if (kvm_set_msr(vcpu, ecx, data)) {
1565 		trace_kvm_msr_write_ex(ecx, data);
1566 		kvm_inject_gp(vcpu, 0);
1567 		return 1;
1568 	}
1569 
1570 	trace_kvm_msr_write(ecx, data);
1571 	return kvm_skip_emulated_instruction(vcpu);
1572 }
1573 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1574 
1575 /*
1576  * The fast path for frequent and performance sensitive wrmsr emulation,
1577  * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1578  * the latency of virtual IPI by avoiding the expensive bits of transitioning
1579  * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1580  * other cases which must be called after interrupts are enabled on the host.
1581  */
1582 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1583 {
1584 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1585 		return 1;
1586 
1587 	if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1588 		((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1589 		((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1590 		((u32)(data >> 32) != X2APIC_BROADCAST)) {
1591 
1592 		data &= ~(1 << 12);
1593 		kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1594 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1595 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1596 		trace_kvm_apic_write(APIC_ICR, (u32)data);
1597 		return 0;
1598 	}
1599 
1600 	return 1;
1601 }
1602 
1603 enum exit_fastpath_completion handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1604 {
1605 	u32 msr = kvm_rcx_read(vcpu);
1606 	u64 data;
1607 	int ret = 0;
1608 
1609 	switch (msr) {
1610 	case APIC_BASE_MSR + (APIC_ICR >> 4):
1611 		data = kvm_read_edx_eax(vcpu);
1612 		ret = handle_fastpath_set_x2apic_icr_irqoff(vcpu, data);
1613 		break;
1614 	default:
1615 		return EXIT_FASTPATH_NONE;
1616 	}
1617 
1618 	if (!ret) {
1619 		trace_kvm_msr_write(msr, data);
1620 		return EXIT_FASTPATH_SKIP_EMUL_INS;
1621 	}
1622 
1623 	return EXIT_FASTPATH_NONE;
1624 }
1625 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1626 
1627 /*
1628  * Adapt set_msr() to msr_io()'s calling convention
1629  */
1630 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1631 {
1632 	return __kvm_get_msr(vcpu, index, data, true);
1633 }
1634 
1635 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1636 {
1637 	return __kvm_set_msr(vcpu, index, *data, true);
1638 }
1639 
1640 #ifdef CONFIG_X86_64
1641 struct pvclock_clock {
1642 	int vclock_mode;
1643 	u64 cycle_last;
1644 	u64 mask;
1645 	u32 mult;
1646 	u32 shift;
1647 	u64 base_cycles;
1648 	u64 offset;
1649 };
1650 
1651 struct pvclock_gtod_data {
1652 	seqcount_t	seq;
1653 
1654 	struct pvclock_clock clock; /* extract of a clocksource struct */
1655 	struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1656 
1657 	ktime_t		offs_boot;
1658 	u64		wall_time_sec;
1659 };
1660 
1661 static struct pvclock_gtod_data pvclock_gtod_data;
1662 
1663 static void update_pvclock_gtod(struct timekeeper *tk)
1664 {
1665 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1666 
1667 	write_seqcount_begin(&vdata->seq);
1668 
1669 	/* copy pvclock gtod data */
1670 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->vdso_clock_mode;
1671 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1672 	vdata->clock.mask		= tk->tkr_mono.mask;
1673 	vdata->clock.mult		= tk->tkr_mono.mult;
1674 	vdata->clock.shift		= tk->tkr_mono.shift;
1675 	vdata->clock.base_cycles	= tk->tkr_mono.xtime_nsec;
1676 	vdata->clock.offset		= tk->tkr_mono.base;
1677 
1678 	vdata->raw_clock.vclock_mode	= tk->tkr_raw.clock->vdso_clock_mode;
1679 	vdata->raw_clock.cycle_last	= tk->tkr_raw.cycle_last;
1680 	vdata->raw_clock.mask		= tk->tkr_raw.mask;
1681 	vdata->raw_clock.mult		= tk->tkr_raw.mult;
1682 	vdata->raw_clock.shift		= tk->tkr_raw.shift;
1683 	vdata->raw_clock.base_cycles	= tk->tkr_raw.xtime_nsec;
1684 	vdata->raw_clock.offset		= tk->tkr_raw.base;
1685 
1686 	vdata->wall_time_sec            = tk->xtime_sec;
1687 
1688 	vdata->offs_boot		= tk->offs_boot;
1689 
1690 	write_seqcount_end(&vdata->seq);
1691 }
1692 
1693 static s64 get_kvmclock_base_ns(void)
1694 {
1695 	/* Count up from boot time, but with the frequency of the raw clock.  */
1696 	return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
1697 }
1698 #else
1699 static s64 get_kvmclock_base_ns(void)
1700 {
1701 	/* Master clock not used, so we can just use CLOCK_BOOTTIME.  */
1702 	return ktime_get_boottime_ns();
1703 }
1704 #endif
1705 
1706 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1707 {
1708 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1709 	kvm_vcpu_kick(vcpu);
1710 }
1711 
1712 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1713 {
1714 	int version;
1715 	int r;
1716 	struct pvclock_wall_clock wc;
1717 	u64 wall_nsec;
1718 
1719 	if (!wall_clock)
1720 		return;
1721 
1722 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1723 	if (r)
1724 		return;
1725 
1726 	if (version & 1)
1727 		++version;  /* first time write, random junk */
1728 
1729 	++version;
1730 
1731 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1732 		return;
1733 
1734 	/*
1735 	 * The guest calculates current wall clock time by adding
1736 	 * system time (updated by kvm_guest_time_update below) to the
1737 	 * wall clock specified here.  We do the reverse here.
1738 	 */
1739 	wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
1740 
1741 	wc.nsec = do_div(wall_nsec, 1000000000);
1742 	wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
1743 	wc.version = version;
1744 
1745 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1746 
1747 	version++;
1748 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1749 }
1750 
1751 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1752 {
1753 	do_shl32_div32(dividend, divisor);
1754 	return dividend;
1755 }
1756 
1757 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1758 			       s8 *pshift, u32 *pmultiplier)
1759 {
1760 	uint64_t scaled64;
1761 	int32_t  shift = 0;
1762 	uint64_t tps64;
1763 	uint32_t tps32;
1764 
1765 	tps64 = base_hz;
1766 	scaled64 = scaled_hz;
1767 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1768 		tps64 >>= 1;
1769 		shift--;
1770 	}
1771 
1772 	tps32 = (uint32_t)tps64;
1773 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1774 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1775 			scaled64 >>= 1;
1776 		else
1777 			tps32 <<= 1;
1778 		shift++;
1779 	}
1780 
1781 	*pshift = shift;
1782 	*pmultiplier = div_frac(scaled64, tps32);
1783 }
1784 
1785 #ifdef CONFIG_X86_64
1786 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1787 #endif
1788 
1789 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1790 static unsigned long max_tsc_khz;
1791 
1792 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1793 {
1794 	u64 v = (u64)khz * (1000000 + ppm);
1795 	do_div(v, 1000000);
1796 	return v;
1797 }
1798 
1799 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1800 {
1801 	u64 ratio;
1802 
1803 	/* Guest TSC same frequency as host TSC? */
1804 	if (!scale) {
1805 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1806 		return 0;
1807 	}
1808 
1809 	/* TSC scaling supported? */
1810 	if (!kvm_has_tsc_control) {
1811 		if (user_tsc_khz > tsc_khz) {
1812 			vcpu->arch.tsc_catchup = 1;
1813 			vcpu->arch.tsc_always_catchup = 1;
1814 			return 0;
1815 		} else {
1816 			pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
1817 			return -1;
1818 		}
1819 	}
1820 
1821 	/* TSC scaling required  - calculate ratio */
1822 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1823 				user_tsc_khz, tsc_khz);
1824 
1825 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1826 		pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1827 			            user_tsc_khz);
1828 		return -1;
1829 	}
1830 
1831 	vcpu->arch.tsc_scaling_ratio = ratio;
1832 	return 0;
1833 }
1834 
1835 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1836 {
1837 	u32 thresh_lo, thresh_hi;
1838 	int use_scaling = 0;
1839 
1840 	/* tsc_khz can be zero if TSC calibration fails */
1841 	if (user_tsc_khz == 0) {
1842 		/* set tsc_scaling_ratio to a safe value */
1843 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1844 		return -1;
1845 	}
1846 
1847 	/* Compute a scale to convert nanoseconds in TSC cycles */
1848 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1849 			   &vcpu->arch.virtual_tsc_shift,
1850 			   &vcpu->arch.virtual_tsc_mult);
1851 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1852 
1853 	/*
1854 	 * Compute the variation in TSC rate which is acceptable
1855 	 * within the range of tolerance and decide if the
1856 	 * rate being applied is within that bounds of the hardware
1857 	 * rate.  If so, no scaling or compensation need be done.
1858 	 */
1859 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1860 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1861 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1862 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1863 		use_scaling = 1;
1864 	}
1865 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1866 }
1867 
1868 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1869 {
1870 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1871 				      vcpu->arch.virtual_tsc_mult,
1872 				      vcpu->arch.virtual_tsc_shift);
1873 	tsc += vcpu->arch.this_tsc_write;
1874 	return tsc;
1875 }
1876 
1877 static inline int gtod_is_based_on_tsc(int mode)
1878 {
1879 	return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
1880 }
1881 
1882 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1883 {
1884 #ifdef CONFIG_X86_64
1885 	bool vcpus_matched;
1886 	struct kvm_arch *ka = &vcpu->kvm->arch;
1887 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1888 
1889 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1890 			 atomic_read(&vcpu->kvm->online_vcpus));
1891 
1892 	/*
1893 	 * Once the masterclock is enabled, always perform request in
1894 	 * order to update it.
1895 	 *
1896 	 * In order to enable masterclock, the host clocksource must be TSC
1897 	 * and the vcpus need to have matched TSCs.  When that happens,
1898 	 * perform request to enable masterclock.
1899 	 */
1900 	if (ka->use_master_clock ||
1901 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1902 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1903 
1904 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1905 			    atomic_read(&vcpu->kvm->online_vcpus),
1906 		            ka->use_master_clock, gtod->clock.vclock_mode);
1907 #endif
1908 }
1909 
1910 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1911 {
1912 	u64 curr_offset = kvm_x86_ops.read_l1_tsc_offset(vcpu);
1913 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1914 }
1915 
1916 /*
1917  * Multiply tsc by a fixed point number represented by ratio.
1918  *
1919  * The most significant 64-N bits (mult) of ratio represent the
1920  * integral part of the fixed point number; the remaining N bits
1921  * (frac) represent the fractional part, ie. ratio represents a fixed
1922  * point number (mult + frac * 2^(-N)).
1923  *
1924  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1925  */
1926 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1927 {
1928 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1929 }
1930 
1931 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1932 {
1933 	u64 _tsc = tsc;
1934 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
1935 
1936 	if (ratio != kvm_default_tsc_scaling_ratio)
1937 		_tsc = __scale_tsc(ratio, tsc);
1938 
1939 	return _tsc;
1940 }
1941 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1942 
1943 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1944 {
1945 	u64 tsc;
1946 
1947 	tsc = kvm_scale_tsc(vcpu, rdtsc());
1948 
1949 	return target_tsc - tsc;
1950 }
1951 
1952 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1953 {
1954 	u64 tsc_offset = kvm_x86_ops.read_l1_tsc_offset(vcpu);
1955 
1956 	return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1957 }
1958 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1959 
1960 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1961 {
1962 	vcpu->arch.tsc_offset = kvm_x86_ops.write_l1_tsc_offset(vcpu, offset);
1963 }
1964 
1965 static inline bool kvm_check_tsc_unstable(void)
1966 {
1967 #ifdef CONFIG_X86_64
1968 	/*
1969 	 * TSC is marked unstable when we're running on Hyper-V,
1970 	 * 'TSC page' clocksource is good.
1971 	 */
1972 	if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
1973 		return false;
1974 #endif
1975 	return check_tsc_unstable();
1976 }
1977 
1978 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1979 {
1980 	struct kvm *kvm = vcpu->kvm;
1981 	u64 offset, ns, elapsed;
1982 	unsigned long flags;
1983 	bool matched;
1984 	bool already_matched;
1985 	u64 data = msr->data;
1986 	bool synchronizing = false;
1987 
1988 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1989 	offset = kvm_compute_tsc_offset(vcpu, data);
1990 	ns = get_kvmclock_base_ns();
1991 	elapsed = ns - kvm->arch.last_tsc_nsec;
1992 
1993 	if (vcpu->arch.virtual_tsc_khz) {
1994 		if (data == 0 && msr->host_initiated) {
1995 			/*
1996 			 * detection of vcpu initialization -- need to sync
1997 			 * with other vCPUs. This particularly helps to keep
1998 			 * kvm_clock stable after CPU hotplug
1999 			 */
2000 			synchronizing = true;
2001 		} else {
2002 			u64 tsc_exp = kvm->arch.last_tsc_write +
2003 						nsec_to_cycles(vcpu, elapsed);
2004 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2005 			/*
2006 			 * Special case: TSC write with a small delta (1 second)
2007 			 * of virtual cycle time against real time is
2008 			 * interpreted as an attempt to synchronize the CPU.
2009 			 */
2010 			synchronizing = data < tsc_exp + tsc_hz &&
2011 					data + tsc_hz > tsc_exp;
2012 		}
2013 	}
2014 
2015 	/*
2016 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
2017 	 * TSC, we add elapsed time in this computation.  We could let the
2018 	 * compensation code attempt to catch up if we fall behind, but
2019 	 * it's better to try to match offsets from the beginning.
2020          */
2021 	if (synchronizing &&
2022 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2023 		if (!kvm_check_tsc_unstable()) {
2024 			offset = kvm->arch.cur_tsc_offset;
2025 		} else {
2026 			u64 delta = nsec_to_cycles(vcpu, elapsed);
2027 			data += delta;
2028 			offset = kvm_compute_tsc_offset(vcpu, data);
2029 		}
2030 		matched = true;
2031 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2032 	} else {
2033 		/*
2034 		 * We split periods of matched TSC writes into generations.
2035 		 * For each generation, we track the original measured
2036 		 * nanosecond time, offset, and write, so if TSCs are in
2037 		 * sync, we can match exact offset, and if not, we can match
2038 		 * exact software computation in compute_guest_tsc()
2039 		 *
2040 		 * These values are tracked in kvm->arch.cur_xxx variables.
2041 		 */
2042 		kvm->arch.cur_tsc_generation++;
2043 		kvm->arch.cur_tsc_nsec = ns;
2044 		kvm->arch.cur_tsc_write = data;
2045 		kvm->arch.cur_tsc_offset = offset;
2046 		matched = false;
2047 	}
2048 
2049 	/*
2050 	 * We also track th most recent recorded KHZ, write and time to
2051 	 * allow the matching interval to be extended at each write.
2052 	 */
2053 	kvm->arch.last_tsc_nsec = ns;
2054 	kvm->arch.last_tsc_write = data;
2055 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2056 
2057 	vcpu->arch.last_guest_tsc = data;
2058 
2059 	/* Keep track of which generation this VCPU has synchronized to */
2060 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2061 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2062 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2063 
2064 	if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
2065 		update_ia32_tsc_adjust_msr(vcpu, offset);
2066 
2067 	kvm_vcpu_write_tsc_offset(vcpu, offset);
2068 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2069 
2070 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
2071 	if (!matched) {
2072 		kvm->arch.nr_vcpus_matched_tsc = 0;
2073 	} else if (!already_matched) {
2074 		kvm->arch.nr_vcpus_matched_tsc++;
2075 	}
2076 
2077 	kvm_track_tsc_matching(vcpu);
2078 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
2079 }
2080 
2081 EXPORT_SYMBOL_GPL(kvm_write_tsc);
2082 
2083 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2084 					   s64 adjustment)
2085 {
2086 	u64 tsc_offset = kvm_x86_ops.read_l1_tsc_offset(vcpu);
2087 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2088 }
2089 
2090 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2091 {
2092 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2093 		WARN_ON(adjustment < 0);
2094 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2095 	adjust_tsc_offset_guest(vcpu, adjustment);
2096 }
2097 
2098 #ifdef CONFIG_X86_64
2099 
2100 static u64 read_tsc(void)
2101 {
2102 	u64 ret = (u64)rdtsc_ordered();
2103 	u64 last = pvclock_gtod_data.clock.cycle_last;
2104 
2105 	if (likely(ret >= last))
2106 		return ret;
2107 
2108 	/*
2109 	 * GCC likes to generate cmov here, but this branch is extremely
2110 	 * predictable (it's just a function of time and the likely is
2111 	 * very likely) and there's a data dependence, so force GCC
2112 	 * to generate a branch instead.  I don't barrier() because
2113 	 * we don't actually need a barrier, and if this function
2114 	 * ever gets inlined it will generate worse code.
2115 	 */
2116 	asm volatile ("");
2117 	return last;
2118 }
2119 
2120 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2121 			  int *mode)
2122 {
2123 	long v;
2124 	u64 tsc_pg_val;
2125 
2126 	switch (clock->vclock_mode) {
2127 	case VDSO_CLOCKMODE_HVCLOCK:
2128 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2129 						  tsc_timestamp);
2130 		if (tsc_pg_val != U64_MAX) {
2131 			/* TSC page valid */
2132 			*mode = VDSO_CLOCKMODE_HVCLOCK;
2133 			v = (tsc_pg_val - clock->cycle_last) &
2134 				clock->mask;
2135 		} else {
2136 			/* TSC page invalid */
2137 			*mode = VDSO_CLOCKMODE_NONE;
2138 		}
2139 		break;
2140 	case VDSO_CLOCKMODE_TSC:
2141 		*mode = VDSO_CLOCKMODE_TSC;
2142 		*tsc_timestamp = read_tsc();
2143 		v = (*tsc_timestamp - clock->cycle_last) &
2144 			clock->mask;
2145 		break;
2146 	default:
2147 		*mode = VDSO_CLOCKMODE_NONE;
2148 	}
2149 
2150 	if (*mode == VDSO_CLOCKMODE_NONE)
2151 		*tsc_timestamp = v = 0;
2152 
2153 	return v * clock->mult;
2154 }
2155 
2156 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2157 {
2158 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2159 	unsigned long seq;
2160 	int mode;
2161 	u64 ns;
2162 
2163 	do {
2164 		seq = read_seqcount_begin(&gtod->seq);
2165 		ns = gtod->raw_clock.base_cycles;
2166 		ns += vgettsc(&gtod->raw_clock, tsc_timestamp, &mode);
2167 		ns >>= gtod->raw_clock.shift;
2168 		ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2169 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2170 	*t = ns;
2171 
2172 	return mode;
2173 }
2174 
2175 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2176 {
2177 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2178 	unsigned long seq;
2179 	int mode;
2180 	u64 ns;
2181 
2182 	do {
2183 		seq = read_seqcount_begin(&gtod->seq);
2184 		ts->tv_sec = gtod->wall_time_sec;
2185 		ns = gtod->clock.base_cycles;
2186 		ns += vgettsc(&gtod->clock, tsc_timestamp, &mode);
2187 		ns >>= gtod->clock.shift;
2188 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2189 
2190 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2191 	ts->tv_nsec = ns;
2192 
2193 	return mode;
2194 }
2195 
2196 /* returns true if host is using TSC based clocksource */
2197 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2198 {
2199 	/* checked again under seqlock below */
2200 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2201 		return false;
2202 
2203 	return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2204 						      tsc_timestamp));
2205 }
2206 
2207 /* returns true if host is using TSC based clocksource */
2208 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2209 					   u64 *tsc_timestamp)
2210 {
2211 	/* checked again under seqlock below */
2212 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2213 		return false;
2214 
2215 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2216 }
2217 #endif
2218 
2219 /*
2220  *
2221  * Assuming a stable TSC across physical CPUS, and a stable TSC
2222  * across virtual CPUs, the following condition is possible.
2223  * Each numbered line represents an event visible to both
2224  * CPUs at the next numbered event.
2225  *
2226  * "timespecX" represents host monotonic time. "tscX" represents
2227  * RDTSC value.
2228  *
2229  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
2230  *
2231  * 1.  read timespec0,tsc0
2232  * 2.					| timespec1 = timespec0 + N
2233  * 					| tsc1 = tsc0 + M
2234  * 3. transition to guest		| transition to guest
2235  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2236  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
2237  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2238  *
2239  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2240  *
2241  * 	- ret0 < ret1
2242  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2243  *		...
2244  *	- 0 < N - M => M < N
2245  *
2246  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2247  * always the case (the difference between two distinct xtime instances
2248  * might be smaller then the difference between corresponding TSC reads,
2249  * when updating guest vcpus pvclock areas).
2250  *
2251  * To avoid that problem, do not allow visibility of distinct
2252  * system_timestamp/tsc_timestamp values simultaneously: use a master
2253  * copy of host monotonic time values. Update that master copy
2254  * in lockstep.
2255  *
2256  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2257  *
2258  */
2259 
2260 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2261 {
2262 #ifdef CONFIG_X86_64
2263 	struct kvm_arch *ka = &kvm->arch;
2264 	int vclock_mode;
2265 	bool host_tsc_clocksource, vcpus_matched;
2266 
2267 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2268 			atomic_read(&kvm->online_vcpus));
2269 
2270 	/*
2271 	 * If the host uses TSC clock, then passthrough TSC as stable
2272 	 * to the guest.
2273 	 */
2274 	host_tsc_clocksource = kvm_get_time_and_clockread(
2275 					&ka->master_kernel_ns,
2276 					&ka->master_cycle_now);
2277 
2278 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2279 				&& !ka->backwards_tsc_observed
2280 				&& !ka->boot_vcpu_runs_old_kvmclock;
2281 
2282 	if (ka->use_master_clock)
2283 		atomic_set(&kvm_guest_has_master_clock, 1);
2284 
2285 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2286 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2287 					vcpus_matched);
2288 #endif
2289 }
2290 
2291 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2292 {
2293 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2294 }
2295 
2296 static void kvm_gen_update_masterclock(struct kvm *kvm)
2297 {
2298 #ifdef CONFIG_X86_64
2299 	int i;
2300 	struct kvm_vcpu *vcpu;
2301 	struct kvm_arch *ka = &kvm->arch;
2302 
2303 	spin_lock(&ka->pvclock_gtod_sync_lock);
2304 	kvm_make_mclock_inprogress_request(kvm);
2305 	/* no guest entries from this point */
2306 	pvclock_update_vm_gtod_copy(kvm);
2307 
2308 	kvm_for_each_vcpu(i, vcpu, kvm)
2309 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2310 
2311 	/* guest entries allowed */
2312 	kvm_for_each_vcpu(i, vcpu, kvm)
2313 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2314 
2315 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2316 #endif
2317 }
2318 
2319 u64 get_kvmclock_ns(struct kvm *kvm)
2320 {
2321 	struct kvm_arch *ka = &kvm->arch;
2322 	struct pvclock_vcpu_time_info hv_clock;
2323 	u64 ret;
2324 
2325 	spin_lock(&ka->pvclock_gtod_sync_lock);
2326 	if (!ka->use_master_clock) {
2327 		spin_unlock(&ka->pvclock_gtod_sync_lock);
2328 		return get_kvmclock_base_ns() + ka->kvmclock_offset;
2329 	}
2330 
2331 	hv_clock.tsc_timestamp = ka->master_cycle_now;
2332 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2333 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2334 
2335 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2336 	get_cpu();
2337 
2338 	if (__this_cpu_read(cpu_tsc_khz)) {
2339 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2340 				   &hv_clock.tsc_shift,
2341 				   &hv_clock.tsc_to_system_mul);
2342 		ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2343 	} else
2344 		ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2345 
2346 	put_cpu();
2347 
2348 	return ret;
2349 }
2350 
2351 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2352 {
2353 	struct kvm_vcpu_arch *vcpu = &v->arch;
2354 	struct pvclock_vcpu_time_info guest_hv_clock;
2355 
2356 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2357 		&guest_hv_clock, sizeof(guest_hv_clock))))
2358 		return;
2359 
2360 	/* This VCPU is paused, but it's legal for a guest to read another
2361 	 * VCPU's kvmclock, so we really have to follow the specification where
2362 	 * it says that version is odd if data is being modified, and even after
2363 	 * it is consistent.
2364 	 *
2365 	 * Version field updates must be kept separate.  This is because
2366 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2367 	 * writes within a string instruction are weakly ordered.  So there
2368 	 * are three writes overall.
2369 	 *
2370 	 * As a small optimization, only write the version field in the first
2371 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2372 	 * version field is the first in the struct.
2373 	 */
2374 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2375 
2376 	if (guest_hv_clock.version & 1)
2377 		++guest_hv_clock.version;  /* first time write, random junk */
2378 
2379 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2380 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2381 				&vcpu->hv_clock,
2382 				sizeof(vcpu->hv_clock.version));
2383 
2384 	smp_wmb();
2385 
2386 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2387 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2388 
2389 	if (vcpu->pvclock_set_guest_stopped_request) {
2390 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2391 		vcpu->pvclock_set_guest_stopped_request = false;
2392 	}
2393 
2394 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2395 
2396 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2397 				&vcpu->hv_clock,
2398 				sizeof(vcpu->hv_clock));
2399 
2400 	smp_wmb();
2401 
2402 	vcpu->hv_clock.version++;
2403 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2404 				&vcpu->hv_clock,
2405 				sizeof(vcpu->hv_clock.version));
2406 }
2407 
2408 static int kvm_guest_time_update(struct kvm_vcpu *v)
2409 {
2410 	unsigned long flags, tgt_tsc_khz;
2411 	struct kvm_vcpu_arch *vcpu = &v->arch;
2412 	struct kvm_arch *ka = &v->kvm->arch;
2413 	s64 kernel_ns;
2414 	u64 tsc_timestamp, host_tsc;
2415 	u8 pvclock_flags;
2416 	bool use_master_clock;
2417 
2418 	kernel_ns = 0;
2419 	host_tsc = 0;
2420 
2421 	/*
2422 	 * If the host uses TSC clock, then passthrough TSC as stable
2423 	 * to the guest.
2424 	 */
2425 	spin_lock(&ka->pvclock_gtod_sync_lock);
2426 	use_master_clock = ka->use_master_clock;
2427 	if (use_master_clock) {
2428 		host_tsc = ka->master_cycle_now;
2429 		kernel_ns = ka->master_kernel_ns;
2430 	}
2431 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2432 
2433 	/* Keep irq disabled to prevent changes to the clock */
2434 	local_irq_save(flags);
2435 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2436 	if (unlikely(tgt_tsc_khz == 0)) {
2437 		local_irq_restore(flags);
2438 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2439 		return 1;
2440 	}
2441 	if (!use_master_clock) {
2442 		host_tsc = rdtsc();
2443 		kernel_ns = get_kvmclock_base_ns();
2444 	}
2445 
2446 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2447 
2448 	/*
2449 	 * We may have to catch up the TSC to match elapsed wall clock
2450 	 * time for two reasons, even if kvmclock is used.
2451 	 *   1) CPU could have been running below the maximum TSC rate
2452 	 *   2) Broken TSC compensation resets the base at each VCPU
2453 	 *      entry to avoid unknown leaps of TSC even when running
2454 	 *      again on the same CPU.  This may cause apparent elapsed
2455 	 *      time to disappear, and the guest to stand still or run
2456 	 *	very slowly.
2457 	 */
2458 	if (vcpu->tsc_catchup) {
2459 		u64 tsc = compute_guest_tsc(v, kernel_ns);
2460 		if (tsc > tsc_timestamp) {
2461 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2462 			tsc_timestamp = tsc;
2463 		}
2464 	}
2465 
2466 	local_irq_restore(flags);
2467 
2468 	/* With all the info we got, fill in the values */
2469 
2470 	if (kvm_has_tsc_control)
2471 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2472 
2473 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2474 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2475 				   &vcpu->hv_clock.tsc_shift,
2476 				   &vcpu->hv_clock.tsc_to_system_mul);
2477 		vcpu->hw_tsc_khz = tgt_tsc_khz;
2478 	}
2479 
2480 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2481 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2482 	vcpu->last_guest_tsc = tsc_timestamp;
2483 
2484 	/* If the host uses TSC clocksource, then it is stable */
2485 	pvclock_flags = 0;
2486 	if (use_master_clock)
2487 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2488 
2489 	vcpu->hv_clock.flags = pvclock_flags;
2490 
2491 	if (vcpu->pv_time_enabled)
2492 		kvm_setup_pvclock_page(v);
2493 	if (v == kvm_get_vcpu(v->kvm, 0))
2494 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2495 	return 0;
2496 }
2497 
2498 /*
2499  * kvmclock updates which are isolated to a given vcpu, such as
2500  * vcpu->cpu migration, should not allow system_timestamp from
2501  * the rest of the vcpus to remain static. Otherwise ntp frequency
2502  * correction applies to one vcpu's system_timestamp but not
2503  * the others.
2504  *
2505  * So in those cases, request a kvmclock update for all vcpus.
2506  * We need to rate-limit these requests though, as they can
2507  * considerably slow guests that have a large number of vcpus.
2508  * The time for a remote vcpu to update its kvmclock is bound
2509  * by the delay we use to rate-limit the updates.
2510  */
2511 
2512 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2513 
2514 static void kvmclock_update_fn(struct work_struct *work)
2515 {
2516 	int i;
2517 	struct delayed_work *dwork = to_delayed_work(work);
2518 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2519 					   kvmclock_update_work);
2520 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2521 	struct kvm_vcpu *vcpu;
2522 
2523 	kvm_for_each_vcpu(i, vcpu, kvm) {
2524 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2525 		kvm_vcpu_kick(vcpu);
2526 	}
2527 }
2528 
2529 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2530 {
2531 	struct kvm *kvm = v->kvm;
2532 
2533 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2534 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2535 					KVMCLOCK_UPDATE_DELAY);
2536 }
2537 
2538 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2539 
2540 static void kvmclock_sync_fn(struct work_struct *work)
2541 {
2542 	struct delayed_work *dwork = to_delayed_work(work);
2543 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2544 					   kvmclock_sync_work);
2545 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2546 
2547 	if (!kvmclock_periodic_sync)
2548 		return;
2549 
2550 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2551 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2552 					KVMCLOCK_SYNC_PERIOD);
2553 }
2554 
2555 /*
2556  * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2557  */
2558 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2559 {
2560 	/* McStatusWrEn enabled? */
2561 	if (guest_cpuid_is_amd_or_hygon(vcpu))
2562 		return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2563 
2564 	return false;
2565 }
2566 
2567 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2568 {
2569 	u64 mcg_cap = vcpu->arch.mcg_cap;
2570 	unsigned bank_num = mcg_cap & 0xff;
2571 	u32 msr = msr_info->index;
2572 	u64 data = msr_info->data;
2573 
2574 	switch (msr) {
2575 	case MSR_IA32_MCG_STATUS:
2576 		vcpu->arch.mcg_status = data;
2577 		break;
2578 	case MSR_IA32_MCG_CTL:
2579 		if (!(mcg_cap & MCG_CTL_P) &&
2580 		    (data || !msr_info->host_initiated))
2581 			return 1;
2582 		if (data != 0 && data != ~(u64)0)
2583 			return 1;
2584 		vcpu->arch.mcg_ctl = data;
2585 		break;
2586 	default:
2587 		if (msr >= MSR_IA32_MC0_CTL &&
2588 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2589 			u32 offset = array_index_nospec(
2590 				msr - MSR_IA32_MC0_CTL,
2591 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2592 
2593 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2594 			 * some Linux kernels though clear bit 10 in bank 4 to
2595 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2596 			 * this to avoid an uncatched #GP in the guest
2597 			 */
2598 			if ((offset & 0x3) == 0 &&
2599 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2600 				return -1;
2601 
2602 			/* MCi_STATUS */
2603 			if (!msr_info->host_initiated &&
2604 			    (offset & 0x3) == 1 && data != 0) {
2605 				if (!can_set_mci_status(vcpu))
2606 					return -1;
2607 			}
2608 
2609 			vcpu->arch.mce_banks[offset] = data;
2610 			break;
2611 		}
2612 		return 1;
2613 	}
2614 	return 0;
2615 }
2616 
2617 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2618 {
2619 	struct kvm *kvm = vcpu->kvm;
2620 	int lm = is_long_mode(vcpu);
2621 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2622 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2623 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2624 		: kvm->arch.xen_hvm_config.blob_size_32;
2625 	u32 page_num = data & ~PAGE_MASK;
2626 	u64 page_addr = data & PAGE_MASK;
2627 	u8 *page;
2628 	int r;
2629 
2630 	r = -E2BIG;
2631 	if (page_num >= blob_size)
2632 		goto out;
2633 	r = -ENOMEM;
2634 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2635 	if (IS_ERR(page)) {
2636 		r = PTR_ERR(page);
2637 		goto out;
2638 	}
2639 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2640 		goto out_free;
2641 	r = 0;
2642 out_free:
2643 	kfree(page);
2644 out:
2645 	return r;
2646 }
2647 
2648 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2649 {
2650 	gpa_t gpa = data & ~0x3f;
2651 
2652 	/* Bits 3:5 are reserved, Should be zero */
2653 	if (data & 0x38)
2654 		return 1;
2655 
2656 	vcpu->arch.apf.msr_val = data;
2657 
2658 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
2659 		kvm_clear_async_pf_completion_queue(vcpu);
2660 		kvm_async_pf_hash_reset(vcpu);
2661 		return 0;
2662 	}
2663 
2664 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2665 					sizeof(u32)))
2666 		return 1;
2667 
2668 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2669 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2670 	kvm_async_pf_wakeup_all(vcpu);
2671 	return 0;
2672 }
2673 
2674 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2675 {
2676 	vcpu->arch.pv_time_enabled = false;
2677 	vcpu->arch.time = 0;
2678 }
2679 
2680 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2681 {
2682 	++vcpu->stat.tlb_flush;
2683 	kvm_x86_ops.tlb_flush(vcpu, invalidate_gpa);
2684 }
2685 
2686 static void record_steal_time(struct kvm_vcpu *vcpu)
2687 {
2688 	struct kvm_host_map map;
2689 	struct kvm_steal_time *st;
2690 
2691 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2692 		return;
2693 
2694 	/* -EAGAIN is returned in atomic context so we can just return. */
2695 	if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
2696 			&map, &vcpu->arch.st.cache, false))
2697 		return;
2698 
2699 	st = map.hva +
2700 		offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
2701 
2702 	/*
2703 	 * Doing a TLB flush here, on the guest's behalf, can avoid
2704 	 * expensive IPIs.
2705 	 */
2706 	trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2707 		st->preempted & KVM_VCPU_FLUSH_TLB);
2708 	if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
2709 		kvm_vcpu_flush_tlb(vcpu, false);
2710 
2711 	vcpu->arch.st.preempted = 0;
2712 
2713 	if (st->version & 1)
2714 		st->version += 1;  /* first time write, random junk */
2715 
2716 	st->version += 1;
2717 
2718 	smp_wmb();
2719 
2720 	st->steal += current->sched_info.run_delay -
2721 		vcpu->arch.st.last_steal;
2722 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2723 
2724 	smp_wmb();
2725 
2726 	st->version += 1;
2727 
2728 	kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
2729 }
2730 
2731 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2732 {
2733 	bool pr = false;
2734 	u32 msr = msr_info->index;
2735 	u64 data = msr_info->data;
2736 
2737 	switch (msr) {
2738 	case MSR_AMD64_NB_CFG:
2739 	case MSR_IA32_UCODE_WRITE:
2740 	case MSR_VM_HSAVE_PA:
2741 	case MSR_AMD64_PATCH_LOADER:
2742 	case MSR_AMD64_BU_CFG2:
2743 	case MSR_AMD64_DC_CFG:
2744 	case MSR_F15H_EX_CFG:
2745 		break;
2746 
2747 	case MSR_IA32_UCODE_REV:
2748 		if (msr_info->host_initiated)
2749 			vcpu->arch.microcode_version = data;
2750 		break;
2751 	case MSR_IA32_ARCH_CAPABILITIES:
2752 		if (!msr_info->host_initiated)
2753 			return 1;
2754 		vcpu->arch.arch_capabilities = data;
2755 		break;
2756 	case MSR_EFER:
2757 		return set_efer(vcpu, msr_info);
2758 	case MSR_K7_HWCR:
2759 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2760 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2761 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2762 
2763 		/* Handle McStatusWrEn */
2764 		if (data == BIT_ULL(18)) {
2765 			vcpu->arch.msr_hwcr = data;
2766 		} else if (data != 0) {
2767 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2768 				    data);
2769 			return 1;
2770 		}
2771 		break;
2772 	case MSR_FAM10H_MMIO_CONF_BASE:
2773 		if (data != 0) {
2774 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2775 				    "0x%llx\n", data);
2776 			return 1;
2777 		}
2778 		break;
2779 	case MSR_IA32_DEBUGCTLMSR:
2780 		if (!data) {
2781 			/* We support the non-activated case already */
2782 			break;
2783 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2784 			/* Values other than LBR and BTF are vendor-specific,
2785 			   thus reserved and should throw a #GP */
2786 			return 1;
2787 		}
2788 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2789 			    __func__, data);
2790 		break;
2791 	case 0x200 ... 0x2ff:
2792 		return kvm_mtrr_set_msr(vcpu, msr, data);
2793 	case MSR_IA32_APICBASE:
2794 		return kvm_set_apic_base(vcpu, msr_info);
2795 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2796 		return kvm_x2apic_msr_write(vcpu, msr, data);
2797 	case MSR_IA32_TSCDEADLINE:
2798 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2799 		break;
2800 	case MSR_IA32_TSC_ADJUST:
2801 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2802 			if (!msr_info->host_initiated) {
2803 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2804 				adjust_tsc_offset_guest(vcpu, adj);
2805 			}
2806 			vcpu->arch.ia32_tsc_adjust_msr = data;
2807 		}
2808 		break;
2809 	case MSR_IA32_MISC_ENABLE:
2810 		if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
2811 		    ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
2812 			if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
2813 				return 1;
2814 			vcpu->arch.ia32_misc_enable_msr = data;
2815 			kvm_update_cpuid(vcpu);
2816 		} else {
2817 			vcpu->arch.ia32_misc_enable_msr = data;
2818 		}
2819 		break;
2820 	case MSR_IA32_SMBASE:
2821 		if (!msr_info->host_initiated)
2822 			return 1;
2823 		vcpu->arch.smbase = data;
2824 		break;
2825 	case MSR_IA32_POWER_CTL:
2826 		vcpu->arch.msr_ia32_power_ctl = data;
2827 		break;
2828 	case MSR_IA32_TSC:
2829 		kvm_write_tsc(vcpu, msr_info);
2830 		break;
2831 	case MSR_IA32_XSS:
2832 		if (!msr_info->host_initiated &&
2833 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
2834 			return 1;
2835 		/*
2836 		 * KVM supports exposing PT to the guest, but does not support
2837 		 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
2838 		 * XSAVES/XRSTORS to save/restore PT MSRs.
2839 		 */
2840 		if (data & ~supported_xss)
2841 			return 1;
2842 		vcpu->arch.ia32_xss = data;
2843 		break;
2844 	case MSR_SMI_COUNT:
2845 		if (!msr_info->host_initiated)
2846 			return 1;
2847 		vcpu->arch.smi_count = data;
2848 		break;
2849 	case MSR_KVM_WALL_CLOCK_NEW:
2850 	case MSR_KVM_WALL_CLOCK:
2851 		vcpu->kvm->arch.wall_clock = data;
2852 		kvm_write_wall_clock(vcpu->kvm, data);
2853 		break;
2854 	case MSR_KVM_SYSTEM_TIME_NEW:
2855 	case MSR_KVM_SYSTEM_TIME: {
2856 		struct kvm_arch *ka = &vcpu->kvm->arch;
2857 
2858 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2859 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2860 
2861 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2862 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2863 
2864 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2865 		}
2866 
2867 		vcpu->arch.time = data;
2868 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2869 
2870 		/* we verify if the enable bit is set... */
2871 		vcpu->arch.pv_time_enabled = false;
2872 		if (!(data & 1))
2873 			break;
2874 
2875 		if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2876 		     &vcpu->arch.pv_time, data & ~1ULL,
2877 		     sizeof(struct pvclock_vcpu_time_info)))
2878 			vcpu->arch.pv_time_enabled = true;
2879 
2880 		break;
2881 	}
2882 	case MSR_KVM_ASYNC_PF_EN:
2883 		if (kvm_pv_enable_async_pf(vcpu, data))
2884 			return 1;
2885 		break;
2886 	case MSR_KVM_STEAL_TIME:
2887 
2888 		if (unlikely(!sched_info_on()))
2889 			return 1;
2890 
2891 		if (data & KVM_STEAL_RESERVED_MASK)
2892 			return 1;
2893 
2894 		vcpu->arch.st.msr_val = data;
2895 
2896 		if (!(data & KVM_MSR_ENABLED))
2897 			break;
2898 
2899 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2900 
2901 		break;
2902 	case MSR_KVM_PV_EOI_EN:
2903 		if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2904 			return 1;
2905 		break;
2906 
2907 	case MSR_KVM_POLL_CONTROL:
2908 		/* only enable bit supported */
2909 		if (data & (-1ULL << 1))
2910 			return 1;
2911 
2912 		vcpu->arch.msr_kvm_poll_control = data;
2913 		break;
2914 
2915 	case MSR_IA32_MCG_CTL:
2916 	case MSR_IA32_MCG_STATUS:
2917 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2918 		return set_msr_mce(vcpu, msr_info);
2919 
2920 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2921 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2922 		pr = true; /* fall through */
2923 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2924 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2925 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2926 			return kvm_pmu_set_msr(vcpu, msr_info);
2927 
2928 		if (pr || data != 0)
2929 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2930 				    "0x%x data 0x%llx\n", msr, data);
2931 		break;
2932 	case MSR_K7_CLK_CTL:
2933 		/*
2934 		 * Ignore all writes to this no longer documented MSR.
2935 		 * Writes are only relevant for old K7 processors,
2936 		 * all pre-dating SVM, but a recommended workaround from
2937 		 * AMD for these chips. It is possible to specify the
2938 		 * affected processor models on the command line, hence
2939 		 * the need to ignore the workaround.
2940 		 */
2941 		break;
2942 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2943 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2944 	case HV_X64_MSR_CRASH_CTL:
2945 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2946 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2947 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
2948 	case HV_X64_MSR_TSC_EMULATION_STATUS:
2949 		return kvm_hv_set_msr_common(vcpu, msr, data,
2950 					     msr_info->host_initiated);
2951 	case MSR_IA32_BBL_CR_CTL3:
2952 		/* Drop writes to this legacy MSR -- see rdmsr
2953 		 * counterpart for further detail.
2954 		 */
2955 		if (report_ignored_msrs)
2956 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2957 				msr, data);
2958 		break;
2959 	case MSR_AMD64_OSVW_ID_LENGTH:
2960 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2961 			return 1;
2962 		vcpu->arch.osvw.length = data;
2963 		break;
2964 	case MSR_AMD64_OSVW_STATUS:
2965 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2966 			return 1;
2967 		vcpu->arch.osvw.status = data;
2968 		break;
2969 	case MSR_PLATFORM_INFO:
2970 		if (!msr_info->host_initiated ||
2971 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2972 		     cpuid_fault_enabled(vcpu)))
2973 			return 1;
2974 		vcpu->arch.msr_platform_info = data;
2975 		break;
2976 	case MSR_MISC_FEATURES_ENABLES:
2977 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2978 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2979 		     !supports_cpuid_fault(vcpu)))
2980 			return 1;
2981 		vcpu->arch.msr_misc_features_enables = data;
2982 		break;
2983 	default:
2984 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2985 			return xen_hvm_config(vcpu, data);
2986 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2987 			return kvm_pmu_set_msr(vcpu, msr_info);
2988 		if (!ignore_msrs) {
2989 			vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2990 				    msr, data);
2991 			return 1;
2992 		} else {
2993 			if (report_ignored_msrs)
2994 				vcpu_unimpl(vcpu,
2995 					"ignored wrmsr: 0x%x data 0x%llx\n",
2996 					msr, data);
2997 			break;
2998 		}
2999 	}
3000 	return 0;
3001 }
3002 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3003 
3004 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3005 {
3006 	u64 data;
3007 	u64 mcg_cap = vcpu->arch.mcg_cap;
3008 	unsigned bank_num = mcg_cap & 0xff;
3009 
3010 	switch (msr) {
3011 	case MSR_IA32_P5_MC_ADDR:
3012 	case MSR_IA32_P5_MC_TYPE:
3013 		data = 0;
3014 		break;
3015 	case MSR_IA32_MCG_CAP:
3016 		data = vcpu->arch.mcg_cap;
3017 		break;
3018 	case MSR_IA32_MCG_CTL:
3019 		if (!(mcg_cap & MCG_CTL_P) && !host)
3020 			return 1;
3021 		data = vcpu->arch.mcg_ctl;
3022 		break;
3023 	case MSR_IA32_MCG_STATUS:
3024 		data = vcpu->arch.mcg_status;
3025 		break;
3026 	default:
3027 		if (msr >= MSR_IA32_MC0_CTL &&
3028 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
3029 			u32 offset = array_index_nospec(
3030 				msr - MSR_IA32_MC0_CTL,
3031 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3032 
3033 			data = vcpu->arch.mce_banks[offset];
3034 			break;
3035 		}
3036 		return 1;
3037 	}
3038 	*pdata = data;
3039 	return 0;
3040 }
3041 
3042 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3043 {
3044 	switch (msr_info->index) {
3045 	case MSR_IA32_PLATFORM_ID:
3046 	case MSR_IA32_EBL_CR_POWERON:
3047 	case MSR_IA32_DEBUGCTLMSR:
3048 	case MSR_IA32_LASTBRANCHFROMIP:
3049 	case MSR_IA32_LASTBRANCHTOIP:
3050 	case MSR_IA32_LASTINTFROMIP:
3051 	case MSR_IA32_LASTINTTOIP:
3052 	case MSR_K8_SYSCFG:
3053 	case MSR_K8_TSEG_ADDR:
3054 	case MSR_K8_TSEG_MASK:
3055 	case MSR_VM_HSAVE_PA:
3056 	case MSR_K8_INT_PENDING_MSG:
3057 	case MSR_AMD64_NB_CFG:
3058 	case MSR_FAM10H_MMIO_CONF_BASE:
3059 	case MSR_AMD64_BU_CFG2:
3060 	case MSR_IA32_PERF_CTL:
3061 	case MSR_AMD64_DC_CFG:
3062 	case MSR_F15H_EX_CFG:
3063 		msr_info->data = 0;
3064 		break;
3065 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3066 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3067 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3068 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3069 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3070 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3071 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3072 		msr_info->data = 0;
3073 		break;
3074 	case MSR_IA32_UCODE_REV:
3075 		msr_info->data = vcpu->arch.microcode_version;
3076 		break;
3077 	case MSR_IA32_ARCH_CAPABILITIES:
3078 		if (!msr_info->host_initiated &&
3079 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3080 			return 1;
3081 		msr_info->data = vcpu->arch.arch_capabilities;
3082 		break;
3083 	case MSR_IA32_POWER_CTL:
3084 		msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3085 		break;
3086 	case MSR_IA32_TSC:
3087 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
3088 		break;
3089 	case MSR_MTRRcap:
3090 	case 0x200 ... 0x2ff:
3091 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3092 	case 0xcd: /* fsb frequency */
3093 		msr_info->data = 3;
3094 		break;
3095 		/*
3096 		 * MSR_EBC_FREQUENCY_ID
3097 		 * Conservative value valid for even the basic CPU models.
3098 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3099 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3100 		 * and 266MHz for model 3, or 4. Set Core Clock
3101 		 * Frequency to System Bus Frequency Ratio to 1 (bits
3102 		 * 31:24) even though these are only valid for CPU
3103 		 * models > 2, however guests may end up dividing or
3104 		 * multiplying by zero otherwise.
3105 		 */
3106 	case MSR_EBC_FREQUENCY_ID:
3107 		msr_info->data = 1 << 24;
3108 		break;
3109 	case MSR_IA32_APICBASE:
3110 		msr_info->data = kvm_get_apic_base(vcpu);
3111 		break;
3112 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
3113 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3114 	case MSR_IA32_TSCDEADLINE:
3115 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3116 		break;
3117 	case MSR_IA32_TSC_ADJUST:
3118 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3119 		break;
3120 	case MSR_IA32_MISC_ENABLE:
3121 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3122 		break;
3123 	case MSR_IA32_SMBASE:
3124 		if (!msr_info->host_initiated)
3125 			return 1;
3126 		msr_info->data = vcpu->arch.smbase;
3127 		break;
3128 	case MSR_SMI_COUNT:
3129 		msr_info->data = vcpu->arch.smi_count;
3130 		break;
3131 	case MSR_IA32_PERF_STATUS:
3132 		/* TSC increment by tick */
3133 		msr_info->data = 1000ULL;
3134 		/* CPU multiplier */
3135 		msr_info->data |= (((uint64_t)4ULL) << 40);
3136 		break;
3137 	case MSR_EFER:
3138 		msr_info->data = vcpu->arch.efer;
3139 		break;
3140 	case MSR_KVM_WALL_CLOCK:
3141 	case MSR_KVM_WALL_CLOCK_NEW:
3142 		msr_info->data = vcpu->kvm->arch.wall_clock;
3143 		break;
3144 	case MSR_KVM_SYSTEM_TIME:
3145 	case MSR_KVM_SYSTEM_TIME_NEW:
3146 		msr_info->data = vcpu->arch.time;
3147 		break;
3148 	case MSR_KVM_ASYNC_PF_EN:
3149 		msr_info->data = vcpu->arch.apf.msr_val;
3150 		break;
3151 	case MSR_KVM_STEAL_TIME:
3152 		msr_info->data = vcpu->arch.st.msr_val;
3153 		break;
3154 	case MSR_KVM_PV_EOI_EN:
3155 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
3156 		break;
3157 	case MSR_KVM_POLL_CONTROL:
3158 		msr_info->data = vcpu->arch.msr_kvm_poll_control;
3159 		break;
3160 	case MSR_IA32_P5_MC_ADDR:
3161 	case MSR_IA32_P5_MC_TYPE:
3162 	case MSR_IA32_MCG_CAP:
3163 	case MSR_IA32_MCG_CTL:
3164 	case MSR_IA32_MCG_STATUS:
3165 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3166 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3167 				   msr_info->host_initiated);
3168 	case MSR_IA32_XSS:
3169 		if (!msr_info->host_initiated &&
3170 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3171 			return 1;
3172 		msr_info->data = vcpu->arch.ia32_xss;
3173 		break;
3174 	case MSR_K7_CLK_CTL:
3175 		/*
3176 		 * Provide expected ramp-up count for K7. All other
3177 		 * are set to zero, indicating minimum divisors for
3178 		 * every field.
3179 		 *
3180 		 * This prevents guest kernels on AMD host with CPU
3181 		 * type 6, model 8 and higher from exploding due to
3182 		 * the rdmsr failing.
3183 		 */
3184 		msr_info->data = 0x20000000;
3185 		break;
3186 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3187 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3188 	case HV_X64_MSR_CRASH_CTL:
3189 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3190 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3191 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3192 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3193 		return kvm_hv_get_msr_common(vcpu,
3194 					     msr_info->index, &msr_info->data,
3195 					     msr_info->host_initiated);
3196 	case MSR_IA32_BBL_CR_CTL3:
3197 		/* This legacy MSR exists but isn't fully documented in current
3198 		 * silicon.  It is however accessed by winxp in very narrow
3199 		 * scenarios where it sets bit #19, itself documented as
3200 		 * a "reserved" bit.  Best effort attempt to source coherent
3201 		 * read data here should the balance of the register be
3202 		 * interpreted by the guest:
3203 		 *
3204 		 * L2 cache control register 3: 64GB range, 256KB size,
3205 		 * enabled, latency 0x1, configured
3206 		 */
3207 		msr_info->data = 0xbe702111;
3208 		break;
3209 	case MSR_AMD64_OSVW_ID_LENGTH:
3210 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3211 			return 1;
3212 		msr_info->data = vcpu->arch.osvw.length;
3213 		break;
3214 	case MSR_AMD64_OSVW_STATUS:
3215 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3216 			return 1;
3217 		msr_info->data = vcpu->arch.osvw.status;
3218 		break;
3219 	case MSR_PLATFORM_INFO:
3220 		if (!msr_info->host_initiated &&
3221 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3222 			return 1;
3223 		msr_info->data = vcpu->arch.msr_platform_info;
3224 		break;
3225 	case MSR_MISC_FEATURES_ENABLES:
3226 		msr_info->data = vcpu->arch.msr_misc_features_enables;
3227 		break;
3228 	case MSR_K7_HWCR:
3229 		msr_info->data = vcpu->arch.msr_hwcr;
3230 		break;
3231 	default:
3232 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3233 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3234 		if (!ignore_msrs) {
3235 			vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
3236 					       msr_info->index);
3237 			return 1;
3238 		} else {
3239 			if (report_ignored_msrs)
3240 				vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
3241 					msr_info->index);
3242 			msr_info->data = 0;
3243 		}
3244 		break;
3245 	}
3246 	return 0;
3247 }
3248 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3249 
3250 /*
3251  * Read or write a bunch of msrs. All parameters are kernel addresses.
3252  *
3253  * @return number of msrs set successfully.
3254  */
3255 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3256 		    struct kvm_msr_entry *entries,
3257 		    int (*do_msr)(struct kvm_vcpu *vcpu,
3258 				  unsigned index, u64 *data))
3259 {
3260 	int i;
3261 
3262 	for (i = 0; i < msrs->nmsrs; ++i)
3263 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
3264 			break;
3265 
3266 	return i;
3267 }
3268 
3269 /*
3270  * Read or write a bunch of msrs. Parameters are user addresses.
3271  *
3272  * @return number of msrs set successfully.
3273  */
3274 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3275 		  int (*do_msr)(struct kvm_vcpu *vcpu,
3276 				unsigned index, u64 *data),
3277 		  int writeback)
3278 {
3279 	struct kvm_msrs msrs;
3280 	struct kvm_msr_entry *entries;
3281 	int r, n;
3282 	unsigned size;
3283 
3284 	r = -EFAULT;
3285 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3286 		goto out;
3287 
3288 	r = -E2BIG;
3289 	if (msrs.nmsrs >= MAX_IO_MSRS)
3290 		goto out;
3291 
3292 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3293 	entries = memdup_user(user_msrs->entries, size);
3294 	if (IS_ERR(entries)) {
3295 		r = PTR_ERR(entries);
3296 		goto out;
3297 	}
3298 
3299 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3300 	if (r < 0)
3301 		goto out_free;
3302 
3303 	r = -EFAULT;
3304 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
3305 		goto out_free;
3306 
3307 	r = n;
3308 
3309 out_free:
3310 	kfree(entries);
3311 out:
3312 	return r;
3313 }
3314 
3315 static inline bool kvm_can_mwait_in_guest(void)
3316 {
3317 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
3318 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
3319 		boot_cpu_has(X86_FEATURE_ARAT);
3320 }
3321 
3322 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3323 {
3324 	int r = 0;
3325 
3326 	switch (ext) {
3327 	case KVM_CAP_IRQCHIP:
3328 	case KVM_CAP_HLT:
3329 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3330 	case KVM_CAP_SET_TSS_ADDR:
3331 	case KVM_CAP_EXT_CPUID:
3332 	case KVM_CAP_EXT_EMUL_CPUID:
3333 	case KVM_CAP_CLOCKSOURCE:
3334 	case KVM_CAP_PIT:
3335 	case KVM_CAP_NOP_IO_DELAY:
3336 	case KVM_CAP_MP_STATE:
3337 	case KVM_CAP_SYNC_MMU:
3338 	case KVM_CAP_USER_NMI:
3339 	case KVM_CAP_REINJECT_CONTROL:
3340 	case KVM_CAP_IRQ_INJECT_STATUS:
3341 	case KVM_CAP_IOEVENTFD:
3342 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
3343 	case KVM_CAP_PIT2:
3344 	case KVM_CAP_PIT_STATE2:
3345 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3346 	case KVM_CAP_XEN_HVM:
3347 	case KVM_CAP_VCPU_EVENTS:
3348 	case KVM_CAP_HYPERV:
3349 	case KVM_CAP_HYPERV_VAPIC:
3350 	case KVM_CAP_HYPERV_SPIN:
3351 	case KVM_CAP_HYPERV_SYNIC:
3352 	case KVM_CAP_HYPERV_SYNIC2:
3353 	case KVM_CAP_HYPERV_VP_INDEX:
3354 	case KVM_CAP_HYPERV_EVENTFD:
3355 	case KVM_CAP_HYPERV_TLBFLUSH:
3356 	case KVM_CAP_HYPERV_SEND_IPI:
3357 	case KVM_CAP_HYPERV_CPUID:
3358 	case KVM_CAP_PCI_SEGMENT:
3359 	case KVM_CAP_DEBUGREGS:
3360 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
3361 	case KVM_CAP_XSAVE:
3362 	case KVM_CAP_ASYNC_PF:
3363 	case KVM_CAP_GET_TSC_KHZ:
3364 	case KVM_CAP_KVMCLOCK_CTRL:
3365 	case KVM_CAP_READONLY_MEM:
3366 	case KVM_CAP_HYPERV_TIME:
3367 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3368 	case KVM_CAP_TSC_DEADLINE_TIMER:
3369 	case KVM_CAP_DISABLE_QUIRKS:
3370 	case KVM_CAP_SET_BOOT_CPU_ID:
3371  	case KVM_CAP_SPLIT_IRQCHIP:
3372 	case KVM_CAP_IMMEDIATE_EXIT:
3373 	case KVM_CAP_PMU_EVENT_FILTER:
3374 	case KVM_CAP_GET_MSR_FEATURES:
3375 	case KVM_CAP_MSR_PLATFORM_INFO:
3376 	case KVM_CAP_EXCEPTION_PAYLOAD:
3377 		r = 1;
3378 		break;
3379 	case KVM_CAP_SYNC_REGS:
3380 		r = KVM_SYNC_X86_VALID_FIELDS;
3381 		break;
3382 	case KVM_CAP_ADJUST_CLOCK:
3383 		r = KVM_CLOCK_TSC_STABLE;
3384 		break;
3385 	case KVM_CAP_X86_DISABLE_EXITS:
3386 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3387 		      KVM_X86_DISABLE_EXITS_CSTATE;
3388 		if(kvm_can_mwait_in_guest())
3389 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
3390 		break;
3391 	case KVM_CAP_X86_SMM:
3392 		/* SMBASE is usually relocated above 1M on modern chipsets,
3393 		 * and SMM handlers might indeed rely on 4G segment limits,
3394 		 * so do not report SMM to be available if real mode is
3395 		 * emulated via vm86 mode.  Still, do not go to great lengths
3396 		 * to avoid userspace's usage of the feature, because it is a
3397 		 * fringe case that is not enabled except via specific settings
3398 		 * of the module parameters.
3399 		 */
3400 		r = kvm_x86_ops.has_emulated_msr(MSR_IA32_SMBASE);
3401 		break;
3402 	case KVM_CAP_VAPIC:
3403 		r = !kvm_x86_ops.cpu_has_accelerated_tpr();
3404 		break;
3405 	case KVM_CAP_NR_VCPUS:
3406 		r = KVM_SOFT_MAX_VCPUS;
3407 		break;
3408 	case KVM_CAP_MAX_VCPUS:
3409 		r = KVM_MAX_VCPUS;
3410 		break;
3411 	case KVM_CAP_MAX_VCPU_ID:
3412 		r = KVM_MAX_VCPU_ID;
3413 		break;
3414 	case KVM_CAP_PV_MMU:	/* obsolete */
3415 		r = 0;
3416 		break;
3417 	case KVM_CAP_MCE:
3418 		r = KVM_MAX_MCE_BANKS;
3419 		break;
3420 	case KVM_CAP_XCRS:
3421 		r = boot_cpu_has(X86_FEATURE_XSAVE);
3422 		break;
3423 	case KVM_CAP_TSC_CONTROL:
3424 		r = kvm_has_tsc_control;
3425 		break;
3426 	case KVM_CAP_X2APIC_API:
3427 		r = KVM_X2APIC_API_VALID_FLAGS;
3428 		break;
3429 	case KVM_CAP_NESTED_STATE:
3430 		r = kvm_x86_ops.get_nested_state ?
3431 			kvm_x86_ops.get_nested_state(NULL, NULL, 0) : 0;
3432 		break;
3433 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3434 		r = kvm_x86_ops.enable_direct_tlbflush != NULL;
3435 		break;
3436 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3437 		r = kvm_x86_ops.nested_enable_evmcs != NULL;
3438 		break;
3439 	default:
3440 		break;
3441 	}
3442 	return r;
3443 
3444 }
3445 
3446 long kvm_arch_dev_ioctl(struct file *filp,
3447 			unsigned int ioctl, unsigned long arg)
3448 {
3449 	void __user *argp = (void __user *)arg;
3450 	long r;
3451 
3452 	switch (ioctl) {
3453 	case KVM_GET_MSR_INDEX_LIST: {
3454 		struct kvm_msr_list __user *user_msr_list = argp;
3455 		struct kvm_msr_list msr_list;
3456 		unsigned n;
3457 
3458 		r = -EFAULT;
3459 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3460 			goto out;
3461 		n = msr_list.nmsrs;
3462 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3463 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3464 			goto out;
3465 		r = -E2BIG;
3466 		if (n < msr_list.nmsrs)
3467 			goto out;
3468 		r = -EFAULT;
3469 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3470 				 num_msrs_to_save * sizeof(u32)))
3471 			goto out;
3472 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3473 				 &emulated_msrs,
3474 				 num_emulated_msrs * sizeof(u32)))
3475 			goto out;
3476 		r = 0;
3477 		break;
3478 	}
3479 	case KVM_GET_SUPPORTED_CPUID:
3480 	case KVM_GET_EMULATED_CPUID: {
3481 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3482 		struct kvm_cpuid2 cpuid;
3483 
3484 		r = -EFAULT;
3485 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3486 			goto out;
3487 
3488 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3489 					    ioctl);
3490 		if (r)
3491 			goto out;
3492 
3493 		r = -EFAULT;
3494 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3495 			goto out;
3496 		r = 0;
3497 		break;
3498 	}
3499 	case KVM_X86_GET_MCE_CAP_SUPPORTED:
3500 		r = -EFAULT;
3501 		if (copy_to_user(argp, &kvm_mce_cap_supported,
3502 				 sizeof(kvm_mce_cap_supported)))
3503 			goto out;
3504 		r = 0;
3505 		break;
3506 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3507 		struct kvm_msr_list __user *user_msr_list = argp;
3508 		struct kvm_msr_list msr_list;
3509 		unsigned int n;
3510 
3511 		r = -EFAULT;
3512 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3513 			goto out;
3514 		n = msr_list.nmsrs;
3515 		msr_list.nmsrs = num_msr_based_features;
3516 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3517 			goto out;
3518 		r = -E2BIG;
3519 		if (n < msr_list.nmsrs)
3520 			goto out;
3521 		r = -EFAULT;
3522 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
3523 				 num_msr_based_features * sizeof(u32)))
3524 			goto out;
3525 		r = 0;
3526 		break;
3527 	}
3528 	case KVM_GET_MSRS:
3529 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
3530 		break;
3531 	default:
3532 		r = -EINVAL;
3533 		break;
3534 	}
3535 out:
3536 	return r;
3537 }
3538 
3539 static void wbinvd_ipi(void *garbage)
3540 {
3541 	wbinvd();
3542 }
3543 
3544 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3545 {
3546 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3547 }
3548 
3549 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3550 {
3551 	/* Address WBINVD may be executed by guest */
3552 	if (need_emulate_wbinvd(vcpu)) {
3553 		if (kvm_x86_ops.has_wbinvd_exit())
3554 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3555 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3556 			smp_call_function_single(vcpu->cpu,
3557 					wbinvd_ipi, NULL, 1);
3558 	}
3559 
3560 	kvm_x86_ops.vcpu_load(vcpu, cpu);
3561 
3562 	/* Apply any externally detected TSC adjustments (due to suspend) */
3563 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3564 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3565 		vcpu->arch.tsc_offset_adjustment = 0;
3566 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3567 	}
3568 
3569 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3570 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3571 				rdtsc() - vcpu->arch.last_host_tsc;
3572 		if (tsc_delta < 0)
3573 			mark_tsc_unstable("KVM discovered backwards TSC");
3574 
3575 		if (kvm_check_tsc_unstable()) {
3576 			u64 offset = kvm_compute_tsc_offset(vcpu,
3577 						vcpu->arch.last_guest_tsc);
3578 			kvm_vcpu_write_tsc_offset(vcpu, offset);
3579 			vcpu->arch.tsc_catchup = 1;
3580 		}
3581 
3582 		if (kvm_lapic_hv_timer_in_use(vcpu))
3583 			kvm_lapic_restart_hv_timer(vcpu);
3584 
3585 		/*
3586 		 * On a host with synchronized TSC, there is no need to update
3587 		 * kvmclock on vcpu->cpu migration
3588 		 */
3589 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3590 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3591 		if (vcpu->cpu != cpu)
3592 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3593 		vcpu->cpu = cpu;
3594 	}
3595 
3596 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3597 }
3598 
3599 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3600 {
3601 	struct kvm_host_map map;
3602 	struct kvm_steal_time *st;
3603 
3604 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3605 		return;
3606 
3607 	if (vcpu->arch.st.preempted)
3608 		return;
3609 
3610 	if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
3611 			&vcpu->arch.st.cache, true))
3612 		return;
3613 
3614 	st = map.hva +
3615 		offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3616 
3617 	st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
3618 
3619 	kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
3620 }
3621 
3622 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3623 {
3624 	int idx;
3625 
3626 	if (vcpu->preempted)
3627 		vcpu->arch.preempted_in_kernel = !kvm_x86_ops.get_cpl(vcpu);
3628 
3629 	/*
3630 	 * Disable page faults because we're in atomic context here.
3631 	 * kvm_write_guest_offset_cached() would call might_fault()
3632 	 * that relies on pagefault_disable() to tell if there's a
3633 	 * bug. NOTE: the write to guest memory may not go through if
3634 	 * during postcopy live migration or if there's heavy guest
3635 	 * paging.
3636 	 */
3637 	pagefault_disable();
3638 	/*
3639 	 * kvm_memslots() will be called by
3640 	 * kvm_write_guest_offset_cached() so take the srcu lock.
3641 	 */
3642 	idx = srcu_read_lock(&vcpu->kvm->srcu);
3643 	kvm_steal_time_set_preempted(vcpu);
3644 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
3645 	pagefault_enable();
3646 	kvm_x86_ops.vcpu_put(vcpu);
3647 	vcpu->arch.last_host_tsc = rdtsc();
3648 	/*
3649 	 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3650 	 * on every vmexit, but if not, we might have a stale dr6 from the
3651 	 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3652 	 */
3653 	set_debugreg(0, 6);
3654 }
3655 
3656 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3657 				    struct kvm_lapic_state *s)
3658 {
3659 	if (vcpu->arch.apicv_active)
3660 		kvm_x86_ops.sync_pir_to_irr(vcpu);
3661 
3662 	return kvm_apic_get_state(vcpu, s);
3663 }
3664 
3665 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3666 				    struct kvm_lapic_state *s)
3667 {
3668 	int r;
3669 
3670 	r = kvm_apic_set_state(vcpu, s);
3671 	if (r)
3672 		return r;
3673 	update_cr8_intercept(vcpu);
3674 
3675 	return 0;
3676 }
3677 
3678 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3679 {
3680 	return (!lapic_in_kernel(vcpu) ||
3681 		kvm_apic_accept_pic_intr(vcpu));
3682 }
3683 
3684 /*
3685  * if userspace requested an interrupt window, check that the
3686  * interrupt window is open.
3687  *
3688  * No need to exit to userspace if we already have an interrupt queued.
3689  */
3690 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3691 {
3692 	return kvm_arch_interrupt_allowed(vcpu) &&
3693 		!kvm_cpu_has_interrupt(vcpu) &&
3694 		!kvm_event_needs_reinjection(vcpu) &&
3695 		kvm_cpu_accept_dm_intr(vcpu);
3696 }
3697 
3698 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3699 				    struct kvm_interrupt *irq)
3700 {
3701 	if (irq->irq >= KVM_NR_INTERRUPTS)
3702 		return -EINVAL;
3703 
3704 	if (!irqchip_in_kernel(vcpu->kvm)) {
3705 		kvm_queue_interrupt(vcpu, irq->irq, false);
3706 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3707 		return 0;
3708 	}
3709 
3710 	/*
3711 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3712 	 * fail for in-kernel 8259.
3713 	 */
3714 	if (pic_in_kernel(vcpu->kvm))
3715 		return -ENXIO;
3716 
3717 	if (vcpu->arch.pending_external_vector != -1)
3718 		return -EEXIST;
3719 
3720 	vcpu->arch.pending_external_vector = irq->irq;
3721 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3722 	return 0;
3723 }
3724 
3725 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3726 {
3727 	kvm_inject_nmi(vcpu);
3728 
3729 	return 0;
3730 }
3731 
3732 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3733 {
3734 	kvm_make_request(KVM_REQ_SMI, vcpu);
3735 
3736 	return 0;
3737 }
3738 
3739 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3740 					   struct kvm_tpr_access_ctl *tac)
3741 {
3742 	if (tac->flags)
3743 		return -EINVAL;
3744 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
3745 	return 0;
3746 }
3747 
3748 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3749 					u64 mcg_cap)
3750 {
3751 	int r;
3752 	unsigned bank_num = mcg_cap & 0xff, bank;
3753 
3754 	r = -EINVAL;
3755 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3756 		goto out;
3757 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3758 		goto out;
3759 	r = 0;
3760 	vcpu->arch.mcg_cap = mcg_cap;
3761 	/* Init IA32_MCG_CTL to all 1s */
3762 	if (mcg_cap & MCG_CTL_P)
3763 		vcpu->arch.mcg_ctl = ~(u64)0;
3764 	/* Init IA32_MCi_CTL to all 1s */
3765 	for (bank = 0; bank < bank_num; bank++)
3766 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3767 
3768 	kvm_x86_ops.setup_mce(vcpu);
3769 out:
3770 	return r;
3771 }
3772 
3773 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3774 				      struct kvm_x86_mce *mce)
3775 {
3776 	u64 mcg_cap = vcpu->arch.mcg_cap;
3777 	unsigned bank_num = mcg_cap & 0xff;
3778 	u64 *banks = vcpu->arch.mce_banks;
3779 
3780 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3781 		return -EINVAL;
3782 	/*
3783 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3784 	 * reporting is disabled
3785 	 */
3786 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3787 	    vcpu->arch.mcg_ctl != ~(u64)0)
3788 		return 0;
3789 	banks += 4 * mce->bank;
3790 	/*
3791 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3792 	 * reporting is disabled for the bank
3793 	 */
3794 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3795 		return 0;
3796 	if (mce->status & MCI_STATUS_UC) {
3797 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3798 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3799 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3800 			return 0;
3801 		}
3802 		if (banks[1] & MCI_STATUS_VAL)
3803 			mce->status |= MCI_STATUS_OVER;
3804 		banks[2] = mce->addr;
3805 		banks[3] = mce->misc;
3806 		vcpu->arch.mcg_status = mce->mcg_status;
3807 		banks[1] = mce->status;
3808 		kvm_queue_exception(vcpu, MC_VECTOR);
3809 	} else if (!(banks[1] & MCI_STATUS_VAL)
3810 		   || !(banks[1] & MCI_STATUS_UC)) {
3811 		if (banks[1] & MCI_STATUS_VAL)
3812 			mce->status |= MCI_STATUS_OVER;
3813 		banks[2] = mce->addr;
3814 		banks[3] = mce->misc;
3815 		banks[1] = mce->status;
3816 	} else
3817 		banks[1] |= MCI_STATUS_OVER;
3818 	return 0;
3819 }
3820 
3821 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3822 					       struct kvm_vcpu_events *events)
3823 {
3824 	process_nmi(vcpu);
3825 
3826 	/*
3827 	 * In guest mode, payload delivery should be deferred,
3828 	 * so that the L1 hypervisor can intercept #PF before
3829 	 * CR2 is modified (or intercept #DB before DR6 is
3830 	 * modified under nVMX). Unless the per-VM capability,
3831 	 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
3832 	 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
3833 	 * opportunistically defer the exception payload, deliver it if the
3834 	 * capability hasn't been requested before processing a
3835 	 * KVM_GET_VCPU_EVENTS.
3836 	 */
3837 	if (!vcpu->kvm->arch.exception_payload_enabled &&
3838 	    vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
3839 		kvm_deliver_exception_payload(vcpu);
3840 
3841 	/*
3842 	 * The API doesn't provide the instruction length for software
3843 	 * exceptions, so don't report them. As long as the guest RIP
3844 	 * isn't advanced, we should expect to encounter the exception
3845 	 * again.
3846 	 */
3847 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3848 		events->exception.injected = 0;
3849 		events->exception.pending = 0;
3850 	} else {
3851 		events->exception.injected = vcpu->arch.exception.injected;
3852 		events->exception.pending = vcpu->arch.exception.pending;
3853 		/*
3854 		 * For ABI compatibility, deliberately conflate
3855 		 * pending and injected exceptions when
3856 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3857 		 */
3858 		if (!vcpu->kvm->arch.exception_payload_enabled)
3859 			events->exception.injected |=
3860 				vcpu->arch.exception.pending;
3861 	}
3862 	events->exception.nr = vcpu->arch.exception.nr;
3863 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3864 	events->exception.error_code = vcpu->arch.exception.error_code;
3865 	events->exception_has_payload = vcpu->arch.exception.has_payload;
3866 	events->exception_payload = vcpu->arch.exception.payload;
3867 
3868 	events->interrupt.injected =
3869 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3870 	events->interrupt.nr = vcpu->arch.interrupt.nr;
3871 	events->interrupt.soft = 0;
3872 	events->interrupt.shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
3873 
3874 	events->nmi.injected = vcpu->arch.nmi_injected;
3875 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
3876 	events->nmi.masked = kvm_x86_ops.get_nmi_mask(vcpu);
3877 	events->nmi.pad = 0;
3878 
3879 	events->sipi_vector = 0; /* never valid when reporting to user space */
3880 
3881 	events->smi.smm = is_smm(vcpu);
3882 	events->smi.pending = vcpu->arch.smi_pending;
3883 	events->smi.smm_inside_nmi =
3884 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3885 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3886 
3887 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3888 			 | KVM_VCPUEVENT_VALID_SHADOW
3889 			 | KVM_VCPUEVENT_VALID_SMM);
3890 	if (vcpu->kvm->arch.exception_payload_enabled)
3891 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3892 
3893 	memset(&events->reserved, 0, sizeof(events->reserved));
3894 }
3895 
3896 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
3897 
3898 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3899 					      struct kvm_vcpu_events *events)
3900 {
3901 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3902 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3903 			      | KVM_VCPUEVENT_VALID_SHADOW
3904 			      | KVM_VCPUEVENT_VALID_SMM
3905 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
3906 		return -EINVAL;
3907 
3908 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3909 		if (!vcpu->kvm->arch.exception_payload_enabled)
3910 			return -EINVAL;
3911 		if (events->exception.pending)
3912 			events->exception.injected = 0;
3913 		else
3914 			events->exception_has_payload = 0;
3915 	} else {
3916 		events->exception.pending = 0;
3917 		events->exception_has_payload = 0;
3918 	}
3919 
3920 	if ((events->exception.injected || events->exception.pending) &&
3921 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3922 		return -EINVAL;
3923 
3924 	/* INITs are latched while in SMM */
3925 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3926 	    (events->smi.smm || events->smi.pending) &&
3927 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3928 		return -EINVAL;
3929 
3930 	process_nmi(vcpu);
3931 	vcpu->arch.exception.injected = events->exception.injected;
3932 	vcpu->arch.exception.pending = events->exception.pending;
3933 	vcpu->arch.exception.nr = events->exception.nr;
3934 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3935 	vcpu->arch.exception.error_code = events->exception.error_code;
3936 	vcpu->arch.exception.has_payload = events->exception_has_payload;
3937 	vcpu->arch.exception.payload = events->exception_payload;
3938 
3939 	vcpu->arch.interrupt.injected = events->interrupt.injected;
3940 	vcpu->arch.interrupt.nr = events->interrupt.nr;
3941 	vcpu->arch.interrupt.soft = events->interrupt.soft;
3942 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3943 		kvm_x86_ops.set_interrupt_shadow(vcpu,
3944 						  events->interrupt.shadow);
3945 
3946 	vcpu->arch.nmi_injected = events->nmi.injected;
3947 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3948 		vcpu->arch.nmi_pending = events->nmi.pending;
3949 	kvm_x86_ops.set_nmi_mask(vcpu, events->nmi.masked);
3950 
3951 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3952 	    lapic_in_kernel(vcpu))
3953 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
3954 
3955 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3956 		if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
3957 			if (events->smi.smm)
3958 				vcpu->arch.hflags |= HF_SMM_MASK;
3959 			else
3960 				vcpu->arch.hflags &= ~HF_SMM_MASK;
3961 			kvm_smm_changed(vcpu);
3962 		}
3963 
3964 		vcpu->arch.smi_pending = events->smi.pending;
3965 
3966 		if (events->smi.smm) {
3967 			if (events->smi.smm_inside_nmi)
3968 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3969 			else
3970 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3971 		}
3972 
3973 		if (lapic_in_kernel(vcpu)) {
3974 			if (events->smi.latched_init)
3975 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3976 			else
3977 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3978 		}
3979 	}
3980 
3981 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3982 
3983 	return 0;
3984 }
3985 
3986 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3987 					     struct kvm_debugregs *dbgregs)
3988 {
3989 	unsigned long val;
3990 
3991 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3992 	kvm_get_dr(vcpu, 6, &val);
3993 	dbgregs->dr6 = val;
3994 	dbgregs->dr7 = vcpu->arch.dr7;
3995 	dbgregs->flags = 0;
3996 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3997 }
3998 
3999 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4000 					    struct kvm_debugregs *dbgregs)
4001 {
4002 	if (dbgregs->flags)
4003 		return -EINVAL;
4004 
4005 	if (dbgregs->dr6 & ~0xffffffffull)
4006 		return -EINVAL;
4007 	if (dbgregs->dr7 & ~0xffffffffull)
4008 		return -EINVAL;
4009 
4010 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4011 	kvm_update_dr0123(vcpu);
4012 	vcpu->arch.dr6 = dbgregs->dr6;
4013 	kvm_update_dr6(vcpu);
4014 	vcpu->arch.dr7 = dbgregs->dr7;
4015 	kvm_update_dr7(vcpu);
4016 
4017 	return 0;
4018 }
4019 
4020 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4021 
4022 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4023 {
4024 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4025 	u64 xstate_bv = xsave->header.xfeatures;
4026 	u64 valid;
4027 
4028 	/*
4029 	 * Copy legacy XSAVE area, to avoid complications with CPUID
4030 	 * leaves 0 and 1 in the loop below.
4031 	 */
4032 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4033 
4034 	/* Set XSTATE_BV */
4035 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4036 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4037 
4038 	/*
4039 	 * Copy each region from the possibly compacted offset to the
4040 	 * non-compacted offset.
4041 	 */
4042 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4043 	while (valid) {
4044 		u64 xfeature_mask = valid & -valid;
4045 		int xfeature_nr = fls64(xfeature_mask) - 1;
4046 		void *src = get_xsave_addr(xsave, xfeature_nr);
4047 
4048 		if (src) {
4049 			u32 size, offset, ecx, edx;
4050 			cpuid_count(XSTATE_CPUID, xfeature_nr,
4051 				    &size, &offset, &ecx, &edx);
4052 			if (xfeature_nr == XFEATURE_PKRU)
4053 				memcpy(dest + offset, &vcpu->arch.pkru,
4054 				       sizeof(vcpu->arch.pkru));
4055 			else
4056 				memcpy(dest + offset, src, size);
4057 
4058 		}
4059 
4060 		valid -= xfeature_mask;
4061 	}
4062 }
4063 
4064 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4065 {
4066 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4067 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4068 	u64 valid;
4069 
4070 	/*
4071 	 * Copy legacy XSAVE area, to avoid complications with CPUID
4072 	 * leaves 0 and 1 in the loop below.
4073 	 */
4074 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
4075 
4076 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
4077 	xsave->header.xfeatures = xstate_bv;
4078 	if (boot_cpu_has(X86_FEATURE_XSAVES))
4079 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4080 
4081 	/*
4082 	 * Copy each region from the non-compacted offset to the
4083 	 * possibly compacted offset.
4084 	 */
4085 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4086 	while (valid) {
4087 		u64 xfeature_mask = valid & -valid;
4088 		int xfeature_nr = fls64(xfeature_mask) - 1;
4089 		void *dest = get_xsave_addr(xsave, xfeature_nr);
4090 
4091 		if (dest) {
4092 			u32 size, offset, ecx, edx;
4093 			cpuid_count(XSTATE_CPUID, xfeature_nr,
4094 				    &size, &offset, &ecx, &edx);
4095 			if (xfeature_nr == XFEATURE_PKRU)
4096 				memcpy(&vcpu->arch.pkru, src + offset,
4097 				       sizeof(vcpu->arch.pkru));
4098 			else
4099 				memcpy(dest, src + offset, size);
4100 		}
4101 
4102 		valid -= xfeature_mask;
4103 	}
4104 }
4105 
4106 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4107 					 struct kvm_xsave *guest_xsave)
4108 {
4109 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4110 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4111 		fill_xsave((u8 *) guest_xsave->region, vcpu);
4112 	} else {
4113 		memcpy(guest_xsave->region,
4114 			&vcpu->arch.guest_fpu->state.fxsave,
4115 			sizeof(struct fxregs_state));
4116 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4117 			XFEATURE_MASK_FPSSE;
4118 	}
4119 }
4120 
4121 #define XSAVE_MXCSR_OFFSET 24
4122 
4123 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4124 					struct kvm_xsave *guest_xsave)
4125 {
4126 	u64 xstate_bv =
4127 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4128 	u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4129 
4130 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4131 		/*
4132 		 * Here we allow setting states that are not present in
4133 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
4134 		 * with old userspace.
4135 		 */
4136 		if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4137 			return -EINVAL;
4138 		load_xsave(vcpu, (u8 *)guest_xsave->region);
4139 	} else {
4140 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4141 			mxcsr & ~mxcsr_feature_mask)
4142 			return -EINVAL;
4143 		memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4144 			guest_xsave->region, sizeof(struct fxregs_state));
4145 	}
4146 	return 0;
4147 }
4148 
4149 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4150 					struct kvm_xcrs *guest_xcrs)
4151 {
4152 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4153 		guest_xcrs->nr_xcrs = 0;
4154 		return;
4155 	}
4156 
4157 	guest_xcrs->nr_xcrs = 1;
4158 	guest_xcrs->flags = 0;
4159 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4160 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4161 }
4162 
4163 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4164 				       struct kvm_xcrs *guest_xcrs)
4165 {
4166 	int i, r = 0;
4167 
4168 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
4169 		return -EINVAL;
4170 
4171 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4172 		return -EINVAL;
4173 
4174 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4175 		/* Only support XCR0 currently */
4176 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4177 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4178 				guest_xcrs->xcrs[i].value);
4179 			break;
4180 		}
4181 	if (r)
4182 		r = -EINVAL;
4183 	return r;
4184 }
4185 
4186 /*
4187  * kvm_set_guest_paused() indicates to the guest kernel that it has been
4188  * stopped by the hypervisor.  This function will be called from the host only.
4189  * EINVAL is returned when the host attempts to set the flag for a guest that
4190  * does not support pv clocks.
4191  */
4192 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4193 {
4194 	if (!vcpu->arch.pv_time_enabled)
4195 		return -EINVAL;
4196 	vcpu->arch.pvclock_set_guest_stopped_request = true;
4197 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4198 	return 0;
4199 }
4200 
4201 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4202 				     struct kvm_enable_cap *cap)
4203 {
4204 	int r;
4205 	uint16_t vmcs_version;
4206 	void __user *user_ptr;
4207 
4208 	if (cap->flags)
4209 		return -EINVAL;
4210 
4211 	switch (cap->cap) {
4212 	case KVM_CAP_HYPERV_SYNIC2:
4213 		if (cap->args[0])
4214 			return -EINVAL;
4215 		/* fall through */
4216 
4217 	case KVM_CAP_HYPERV_SYNIC:
4218 		if (!irqchip_in_kernel(vcpu->kvm))
4219 			return -EINVAL;
4220 		return kvm_hv_activate_synic(vcpu, cap->cap ==
4221 					     KVM_CAP_HYPERV_SYNIC2);
4222 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4223 		if (!kvm_x86_ops.nested_enable_evmcs)
4224 			return -ENOTTY;
4225 		r = kvm_x86_ops.nested_enable_evmcs(vcpu, &vmcs_version);
4226 		if (!r) {
4227 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
4228 			if (copy_to_user(user_ptr, &vmcs_version,
4229 					 sizeof(vmcs_version)))
4230 				r = -EFAULT;
4231 		}
4232 		return r;
4233 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4234 		if (!kvm_x86_ops.enable_direct_tlbflush)
4235 			return -ENOTTY;
4236 
4237 		return kvm_x86_ops.enable_direct_tlbflush(vcpu);
4238 
4239 	default:
4240 		return -EINVAL;
4241 	}
4242 }
4243 
4244 long kvm_arch_vcpu_ioctl(struct file *filp,
4245 			 unsigned int ioctl, unsigned long arg)
4246 {
4247 	struct kvm_vcpu *vcpu = filp->private_data;
4248 	void __user *argp = (void __user *)arg;
4249 	int r;
4250 	union {
4251 		struct kvm_lapic_state *lapic;
4252 		struct kvm_xsave *xsave;
4253 		struct kvm_xcrs *xcrs;
4254 		void *buffer;
4255 	} u;
4256 
4257 	vcpu_load(vcpu);
4258 
4259 	u.buffer = NULL;
4260 	switch (ioctl) {
4261 	case KVM_GET_LAPIC: {
4262 		r = -EINVAL;
4263 		if (!lapic_in_kernel(vcpu))
4264 			goto out;
4265 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4266 				GFP_KERNEL_ACCOUNT);
4267 
4268 		r = -ENOMEM;
4269 		if (!u.lapic)
4270 			goto out;
4271 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4272 		if (r)
4273 			goto out;
4274 		r = -EFAULT;
4275 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4276 			goto out;
4277 		r = 0;
4278 		break;
4279 	}
4280 	case KVM_SET_LAPIC: {
4281 		r = -EINVAL;
4282 		if (!lapic_in_kernel(vcpu))
4283 			goto out;
4284 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
4285 		if (IS_ERR(u.lapic)) {
4286 			r = PTR_ERR(u.lapic);
4287 			goto out_nofree;
4288 		}
4289 
4290 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4291 		break;
4292 	}
4293 	case KVM_INTERRUPT: {
4294 		struct kvm_interrupt irq;
4295 
4296 		r = -EFAULT;
4297 		if (copy_from_user(&irq, argp, sizeof(irq)))
4298 			goto out;
4299 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4300 		break;
4301 	}
4302 	case KVM_NMI: {
4303 		r = kvm_vcpu_ioctl_nmi(vcpu);
4304 		break;
4305 	}
4306 	case KVM_SMI: {
4307 		r = kvm_vcpu_ioctl_smi(vcpu);
4308 		break;
4309 	}
4310 	case KVM_SET_CPUID: {
4311 		struct kvm_cpuid __user *cpuid_arg = argp;
4312 		struct kvm_cpuid cpuid;
4313 
4314 		r = -EFAULT;
4315 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4316 			goto out;
4317 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4318 		break;
4319 	}
4320 	case KVM_SET_CPUID2: {
4321 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4322 		struct kvm_cpuid2 cpuid;
4323 
4324 		r = -EFAULT;
4325 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4326 			goto out;
4327 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4328 					      cpuid_arg->entries);
4329 		break;
4330 	}
4331 	case KVM_GET_CPUID2: {
4332 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4333 		struct kvm_cpuid2 cpuid;
4334 
4335 		r = -EFAULT;
4336 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4337 			goto out;
4338 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4339 					      cpuid_arg->entries);
4340 		if (r)
4341 			goto out;
4342 		r = -EFAULT;
4343 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4344 			goto out;
4345 		r = 0;
4346 		break;
4347 	}
4348 	case KVM_GET_MSRS: {
4349 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4350 		r = msr_io(vcpu, argp, do_get_msr, 1);
4351 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4352 		break;
4353 	}
4354 	case KVM_SET_MSRS: {
4355 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4356 		r = msr_io(vcpu, argp, do_set_msr, 0);
4357 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4358 		break;
4359 	}
4360 	case KVM_TPR_ACCESS_REPORTING: {
4361 		struct kvm_tpr_access_ctl tac;
4362 
4363 		r = -EFAULT;
4364 		if (copy_from_user(&tac, argp, sizeof(tac)))
4365 			goto out;
4366 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4367 		if (r)
4368 			goto out;
4369 		r = -EFAULT;
4370 		if (copy_to_user(argp, &tac, sizeof(tac)))
4371 			goto out;
4372 		r = 0;
4373 		break;
4374 	};
4375 	case KVM_SET_VAPIC_ADDR: {
4376 		struct kvm_vapic_addr va;
4377 		int idx;
4378 
4379 		r = -EINVAL;
4380 		if (!lapic_in_kernel(vcpu))
4381 			goto out;
4382 		r = -EFAULT;
4383 		if (copy_from_user(&va, argp, sizeof(va)))
4384 			goto out;
4385 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4386 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4387 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4388 		break;
4389 	}
4390 	case KVM_X86_SETUP_MCE: {
4391 		u64 mcg_cap;
4392 
4393 		r = -EFAULT;
4394 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4395 			goto out;
4396 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4397 		break;
4398 	}
4399 	case KVM_X86_SET_MCE: {
4400 		struct kvm_x86_mce mce;
4401 
4402 		r = -EFAULT;
4403 		if (copy_from_user(&mce, argp, sizeof(mce)))
4404 			goto out;
4405 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4406 		break;
4407 	}
4408 	case KVM_GET_VCPU_EVENTS: {
4409 		struct kvm_vcpu_events events;
4410 
4411 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4412 
4413 		r = -EFAULT;
4414 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4415 			break;
4416 		r = 0;
4417 		break;
4418 	}
4419 	case KVM_SET_VCPU_EVENTS: {
4420 		struct kvm_vcpu_events events;
4421 
4422 		r = -EFAULT;
4423 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4424 			break;
4425 
4426 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4427 		break;
4428 	}
4429 	case KVM_GET_DEBUGREGS: {
4430 		struct kvm_debugregs dbgregs;
4431 
4432 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4433 
4434 		r = -EFAULT;
4435 		if (copy_to_user(argp, &dbgregs,
4436 				 sizeof(struct kvm_debugregs)))
4437 			break;
4438 		r = 0;
4439 		break;
4440 	}
4441 	case KVM_SET_DEBUGREGS: {
4442 		struct kvm_debugregs dbgregs;
4443 
4444 		r = -EFAULT;
4445 		if (copy_from_user(&dbgregs, argp,
4446 				   sizeof(struct kvm_debugregs)))
4447 			break;
4448 
4449 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4450 		break;
4451 	}
4452 	case KVM_GET_XSAVE: {
4453 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4454 		r = -ENOMEM;
4455 		if (!u.xsave)
4456 			break;
4457 
4458 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4459 
4460 		r = -EFAULT;
4461 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4462 			break;
4463 		r = 0;
4464 		break;
4465 	}
4466 	case KVM_SET_XSAVE: {
4467 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
4468 		if (IS_ERR(u.xsave)) {
4469 			r = PTR_ERR(u.xsave);
4470 			goto out_nofree;
4471 		}
4472 
4473 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4474 		break;
4475 	}
4476 	case KVM_GET_XCRS: {
4477 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4478 		r = -ENOMEM;
4479 		if (!u.xcrs)
4480 			break;
4481 
4482 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4483 
4484 		r = -EFAULT;
4485 		if (copy_to_user(argp, u.xcrs,
4486 				 sizeof(struct kvm_xcrs)))
4487 			break;
4488 		r = 0;
4489 		break;
4490 	}
4491 	case KVM_SET_XCRS: {
4492 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4493 		if (IS_ERR(u.xcrs)) {
4494 			r = PTR_ERR(u.xcrs);
4495 			goto out_nofree;
4496 		}
4497 
4498 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4499 		break;
4500 	}
4501 	case KVM_SET_TSC_KHZ: {
4502 		u32 user_tsc_khz;
4503 
4504 		r = -EINVAL;
4505 		user_tsc_khz = (u32)arg;
4506 
4507 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4508 			goto out;
4509 
4510 		if (user_tsc_khz == 0)
4511 			user_tsc_khz = tsc_khz;
4512 
4513 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4514 			r = 0;
4515 
4516 		goto out;
4517 	}
4518 	case KVM_GET_TSC_KHZ: {
4519 		r = vcpu->arch.virtual_tsc_khz;
4520 		goto out;
4521 	}
4522 	case KVM_KVMCLOCK_CTRL: {
4523 		r = kvm_set_guest_paused(vcpu);
4524 		goto out;
4525 	}
4526 	case KVM_ENABLE_CAP: {
4527 		struct kvm_enable_cap cap;
4528 
4529 		r = -EFAULT;
4530 		if (copy_from_user(&cap, argp, sizeof(cap)))
4531 			goto out;
4532 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4533 		break;
4534 	}
4535 	case KVM_GET_NESTED_STATE: {
4536 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4537 		u32 user_data_size;
4538 
4539 		r = -EINVAL;
4540 		if (!kvm_x86_ops.get_nested_state)
4541 			break;
4542 
4543 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4544 		r = -EFAULT;
4545 		if (get_user(user_data_size, &user_kvm_nested_state->size))
4546 			break;
4547 
4548 		r = kvm_x86_ops.get_nested_state(vcpu, user_kvm_nested_state,
4549 						  user_data_size);
4550 		if (r < 0)
4551 			break;
4552 
4553 		if (r > user_data_size) {
4554 			if (put_user(r, &user_kvm_nested_state->size))
4555 				r = -EFAULT;
4556 			else
4557 				r = -E2BIG;
4558 			break;
4559 		}
4560 
4561 		r = 0;
4562 		break;
4563 	}
4564 	case KVM_SET_NESTED_STATE: {
4565 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4566 		struct kvm_nested_state kvm_state;
4567 		int idx;
4568 
4569 		r = -EINVAL;
4570 		if (!kvm_x86_ops.set_nested_state)
4571 			break;
4572 
4573 		r = -EFAULT;
4574 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4575 			break;
4576 
4577 		r = -EINVAL;
4578 		if (kvm_state.size < sizeof(kvm_state))
4579 			break;
4580 
4581 		if (kvm_state.flags &
4582 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4583 		      | KVM_STATE_NESTED_EVMCS))
4584 			break;
4585 
4586 		/* nested_run_pending implies guest_mode.  */
4587 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4588 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4589 			break;
4590 
4591 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4592 		r = kvm_x86_ops.set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4593 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4594 		break;
4595 	}
4596 	case KVM_GET_SUPPORTED_HV_CPUID: {
4597 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4598 		struct kvm_cpuid2 cpuid;
4599 
4600 		r = -EFAULT;
4601 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4602 			goto out;
4603 
4604 		r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4605 						cpuid_arg->entries);
4606 		if (r)
4607 			goto out;
4608 
4609 		r = -EFAULT;
4610 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4611 			goto out;
4612 		r = 0;
4613 		break;
4614 	}
4615 	default:
4616 		r = -EINVAL;
4617 	}
4618 out:
4619 	kfree(u.buffer);
4620 out_nofree:
4621 	vcpu_put(vcpu);
4622 	return r;
4623 }
4624 
4625 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4626 {
4627 	return VM_FAULT_SIGBUS;
4628 }
4629 
4630 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4631 {
4632 	int ret;
4633 
4634 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
4635 		return -EINVAL;
4636 	ret = kvm_x86_ops.set_tss_addr(kvm, addr);
4637 	return ret;
4638 }
4639 
4640 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4641 					      u64 ident_addr)
4642 {
4643 	return kvm_x86_ops.set_identity_map_addr(kvm, ident_addr);
4644 }
4645 
4646 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4647 					 unsigned long kvm_nr_mmu_pages)
4648 {
4649 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4650 		return -EINVAL;
4651 
4652 	mutex_lock(&kvm->slots_lock);
4653 
4654 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4655 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4656 
4657 	mutex_unlock(&kvm->slots_lock);
4658 	return 0;
4659 }
4660 
4661 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4662 {
4663 	return kvm->arch.n_max_mmu_pages;
4664 }
4665 
4666 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4667 {
4668 	struct kvm_pic *pic = kvm->arch.vpic;
4669 	int r;
4670 
4671 	r = 0;
4672 	switch (chip->chip_id) {
4673 	case KVM_IRQCHIP_PIC_MASTER:
4674 		memcpy(&chip->chip.pic, &pic->pics[0],
4675 			sizeof(struct kvm_pic_state));
4676 		break;
4677 	case KVM_IRQCHIP_PIC_SLAVE:
4678 		memcpy(&chip->chip.pic, &pic->pics[1],
4679 			sizeof(struct kvm_pic_state));
4680 		break;
4681 	case KVM_IRQCHIP_IOAPIC:
4682 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
4683 		break;
4684 	default:
4685 		r = -EINVAL;
4686 		break;
4687 	}
4688 	return r;
4689 }
4690 
4691 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4692 {
4693 	struct kvm_pic *pic = kvm->arch.vpic;
4694 	int r;
4695 
4696 	r = 0;
4697 	switch (chip->chip_id) {
4698 	case KVM_IRQCHIP_PIC_MASTER:
4699 		spin_lock(&pic->lock);
4700 		memcpy(&pic->pics[0], &chip->chip.pic,
4701 			sizeof(struct kvm_pic_state));
4702 		spin_unlock(&pic->lock);
4703 		break;
4704 	case KVM_IRQCHIP_PIC_SLAVE:
4705 		spin_lock(&pic->lock);
4706 		memcpy(&pic->pics[1], &chip->chip.pic,
4707 			sizeof(struct kvm_pic_state));
4708 		spin_unlock(&pic->lock);
4709 		break;
4710 	case KVM_IRQCHIP_IOAPIC:
4711 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
4712 		break;
4713 	default:
4714 		r = -EINVAL;
4715 		break;
4716 	}
4717 	kvm_pic_update_irq(pic);
4718 	return r;
4719 }
4720 
4721 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4722 {
4723 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4724 
4725 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4726 
4727 	mutex_lock(&kps->lock);
4728 	memcpy(ps, &kps->channels, sizeof(*ps));
4729 	mutex_unlock(&kps->lock);
4730 	return 0;
4731 }
4732 
4733 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4734 {
4735 	int i;
4736 	struct kvm_pit *pit = kvm->arch.vpit;
4737 
4738 	mutex_lock(&pit->pit_state.lock);
4739 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4740 	for (i = 0; i < 3; i++)
4741 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4742 	mutex_unlock(&pit->pit_state.lock);
4743 	return 0;
4744 }
4745 
4746 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4747 {
4748 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
4749 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4750 		sizeof(ps->channels));
4751 	ps->flags = kvm->arch.vpit->pit_state.flags;
4752 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4753 	memset(&ps->reserved, 0, sizeof(ps->reserved));
4754 	return 0;
4755 }
4756 
4757 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4758 {
4759 	int start = 0;
4760 	int i;
4761 	u32 prev_legacy, cur_legacy;
4762 	struct kvm_pit *pit = kvm->arch.vpit;
4763 
4764 	mutex_lock(&pit->pit_state.lock);
4765 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4766 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4767 	if (!prev_legacy && cur_legacy)
4768 		start = 1;
4769 	memcpy(&pit->pit_state.channels, &ps->channels,
4770 	       sizeof(pit->pit_state.channels));
4771 	pit->pit_state.flags = ps->flags;
4772 	for (i = 0; i < 3; i++)
4773 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4774 				   start && i == 0);
4775 	mutex_unlock(&pit->pit_state.lock);
4776 	return 0;
4777 }
4778 
4779 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4780 				 struct kvm_reinject_control *control)
4781 {
4782 	struct kvm_pit *pit = kvm->arch.vpit;
4783 
4784 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
4785 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4786 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
4787 	 */
4788 	mutex_lock(&pit->pit_state.lock);
4789 	kvm_pit_set_reinject(pit, control->pit_reinject);
4790 	mutex_unlock(&pit->pit_state.lock);
4791 
4792 	return 0;
4793 }
4794 
4795 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
4796 {
4797 	/*
4798 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4799 	 */
4800 	if (kvm_x86_ops.flush_log_dirty)
4801 		kvm_x86_ops.flush_log_dirty(kvm);
4802 }
4803 
4804 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4805 			bool line_status)
4806 {
4807 	if (!irqchip_in_kernel(kvm))
4808 		return -ENXIO;
4809 
4810 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4811 					irq_event->irq, irq_event->level,
4812 					line_status);
4813 	return 0;
4814 }
4815 
4816 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4817 			    struct kvm_enable_cap *cap)
4818 {
4819 	int r;
4820 
4821 	if (cap->flags)
4822 		return -EINVAL;
4823 
4824 	switch (cap->cap) {
4825 	case KVM_CAP_DISABLE_QUIRKS:
4826 		kvm->arch.disabled_quirks = cap->args[0];
4827 		r = 0;
4828 		break;
4829 	case KVM_CAP_SPLIT_IRQCHIP: {
4830 		mutex_lock(&kvm->lock);
4831 		r = -EINVAL;
4832 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4833 			goto split_irqchip_unlock;
4834 		r = -EEXIST;
4835 		if (irqchip_in_kernel(kvm))
4836 			goto split_irqchip_unlock;
4837 		if (kvm->created_vcpus)
4838 			goto split_irqchip_unlock;
4839 		r = kvm_setup_empty_irq_routing(kvm);
4840 		if (r)
4841 			goto split_irqchip_unlock;
4842 		/* Pairs with irqchip_in_kernel. */
4843 		smp_wmb();
4844 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4845 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4846 		r = 0;
4847 split_irqchip_unlock:
4848 		mutex_unlock(&kvm->lock);
4849 		break;
4850 	}
4851 	case KVM_CAP_X2APIC_API:
4852 		r = -EINVAL;
4853 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4854 			break;
4855 
4856 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4857 			kvm->arch.x2apic_format = true;
4858 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4859 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
4860 
4861 		r = 0;
4862 		break;
4863 	case KVM_CAP_X86_DISABLE_EXITS:
4864 		r = -EINVAL;
4865 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4866 			break;
4867 
4868 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4869 			kvm_can_mwait_in_guest())
4870 			kvm->arch.mwait_in_guest = true;
4871 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4872 			kvm->arch.hlt_in_guest = true;
4873 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4874 			kvm->arch.pause_in_guest = true;
4875 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
4876 			kvm->arch.cstate_in_guest = true;
4877 		r = 0;
4878 		break;
4879 	case KVM_CAP_MSR_PLATFORM_INFO:
4880 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4881 		r = 0;
4882 		break;
4883 	case KVM_CAP_EXCEPTION_PAYLOAD:
4884 		kvm->arch.exception_payload_enabled = cap->args[0];
4885 		r = 0;
4886 		break;
4887 	default:
4888 		r = -EINVAL;
4889 		break;
4890 	}
4891 	return r;
4892 }
4893 
4894 long kvm_arch_vm_ioctl(struct file *filp,
4895 		       unsigned int ioctl, unsigned long arg)
4896 {
4897 	struct kvm *kvm = filp->private_data;
4898 	void __user *argp = (void __user *)arg;
4899 	int r = -ENOTTY;
4900 	/*
4901 	 * This union makes it completely explicit to gcc-3.x
4902 	 * that these two variables' stack usage should be
4903 	 * combined, not added together.
4904 	 */
4905 	union {
4906 		struct kvm_pit_state ps;
4907 		struct kvm_pit_state2 ps2;
4908 		struct kvm_pit_config pit_config;
4909 	} u;
4910 
4911 	switch (ioctl) {
4912 	case KVM_SET_TSS_ADDR:
4913 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4914 		break;
4915 	case KVM_SET_IDENTITY_MAP_ADDR: {
4916 		u64 ident_addr;
4917 
4918 		mutex_lock(&kvm->lock);
4919 		r = -EINVAL;
4920 		if (kvm->created_vcpus)
4921 			goto set_identity_unlock;
4922 		r = -EFAULT;
4923 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4924 			goto set_identity_unlock;
4925 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4926 set_identity_unlock:
4927 		mutex_unlock(&kvm->lock);
4928 		break;
4929 	}
4930 	case KVM_SET_NR_MMU_PAGES:
4931 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4932 		break;
4933 	case KVM_GET_NR_MMU_PAGES:
4934 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4935 		break;
4936 	case KVM_CREATE_IRQCHIP: {
4937 		mutex_lock(&kvm->lock);
4938 
4939 		r = -EEXIST;
4940 		if (irqchip_in_kernel(kvm))
4941 			goto create_irqchip_unlock;
4942 
4943 		r = -EINVAL;
4944 		if (kvm->created_vcpus)
4945 			goto create_irqchip_unlock;
4946 
4947 		r = kvm_pic_init(kvm);
4948 		if (r)
4949 			goto create_irqchip_unlock;
4950 
4951 		r = kvm_ioapic_init(kvm);
4952 		if (r) {
4953 			kvm_pic_destroy(kvm);
4954 			goto create_irqchip_unlock;
4955 		}
4956 
4957 		r = kvm_setup_default_irq_routing(kvm);
4958 		if (r) {
4959 			kvm_ioapic_destroy(kvm);
4960 			kvm_pic_destroy(kvm);
4961 			goto create_irqchip_unlock;
4962 		}
4963 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4964 		smp_wmb();
4965 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4966 	create_irqchip_unlock:
4967 		mutex_unlock(&kvm->lock);
4968 		break;
4969 	}
4970 	case KVM_CREATE_PIT:
4971 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4972 		goto create_pit;
4973 	case KVM_CREATE_PIT2:
4974 		r = -EFAULT;
4975 		if (copy_from_user(&u.pit_config, argp,
4976 				   sizeof(struct kvm_pit_config)))
4977 			goto out;
4978 	create_pit:
4979 		mutex_lock(&kvm->lock);
4980 		r = -EEXIST;
4981 		if (kvm->arch.vpit)
4982 			goto create_pit_unlock;
4983 		r = -ENOMEM;
4984 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4985 		if (kvm->arch.vpit)
4986 			r = 0;
4987 	create_pit_unlock:
4988 		mutex_unlock(&kvm->lock);
4989 		break;
4990 	case KVM_GET_IRQCHIP: {
4991 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4992 		struct kvm_irqchip *chip;
4993 
4994 		chip = memdup_user(argp, sizeof(*chip));
4995 		if (IS_ERR(chip)) {
4996 			r = PTR_ERR(chip);
4997 			goto out;
4998 		}
4999 
5000 		r = -ENXIO;
5001 		if (!irqchip_kernel(kvm))
5002 			goto get_irqchip_out;
5003 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5004 		if (r)
5005 			goto get_irqchip_out;
5006 		r = -EFAULT;
5007 		if (copy_to_user(argp, chip, sizeof(*chip)))
5008 			goto get_irqchip_out;
5009 		r = 0;
5010 	get_irqchip_out:
5011 		kfree(chip);
5012 		break;
5013 	}
5014 	case KVM_SET_IRQCHIP: {
5015 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5016 		struct kvm_irqchip *chip;
5017 
5018 		chip = memdup_user(argp, sizeof(*chip));
5019 		if (IS_ERR(chip)) {
5020 			r = PTR_ERR(chip);
5021 			goto out;
5022 		}
5023 
5024 		r = -ENXIO;
5025 		if (!irqchip_kernel(kvm))
5026 			goto set_irqchip_out;
5027 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5028 	set_irqchip_out:
5029 		kfree(chip);
5030 		break;
5031 	}
5032 	case KVM_GET_PIT: {
5033 		r = -EFAULT;
5034 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5035 			goto out;
5036 		r = -ENXIO;
5037 		if (!kvm->arch.vpit)
5038 			goto out;
5039 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5040 		if (r)
5041 			goto out;
5042 		r = -EFAULT;
5043 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5044 			goto out;
5045 		r = 0;
5046 		break;
5047 	}
5048 	case KVM_SET_PIT: {
5049 		r = -EFAULT;
5050 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5051 			goto out;
5052 		r = -ENXIO;
5053 		if (!kvm->arch.vpit)
5054 			goto out;
5055 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5056 		break;
5057 	}
5058 	case KVM_GET_PIT2: {
5059 		r = -ENXIO;
5060 		if (!kvm->arch.vpit)
5061 			goto out;
5062 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5063 		if (r)
5064 			goto out;
5065 		r = -EFAULT;
5066 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5067 			goto out;
5068 		r = 0;
5069 		break;
5070 	}
5071 	case KVM_SET_PIT2: {
5072 		r = -EFAULT;
5073 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5074 			goto out;
5075 		r = -ENXIO;
5076 		if (!kvm->arch.vpit)
5077 			goto out;
5078 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5079 		break;
5080 	}
5081 	case KVM_REINJECT_CONTROL: {
5082 		struct kvm_reinject_control control;
5083 		r =  -EFAULT;
5084 		if (copy_from_user(&control, argp, sizeof(control)))
5085 			goto out;
5086 		r = -ENXIO;
5087 		if (!kvm->arch.vpit)
5088 			goto out;
5089 		r = kvm_vm_ioctl_reinject(kvm, &control);
5090 		break;
5091 	}
5092 	case KVM_SET_BOOT_CPU_ID:
5093 		r = 0;
5094 		mutex_lock(&kvm->lock);
5095 		if (kvm->created_vcpus)
5096 			r = -EBUSY;
5097 		else
5098 			kvm->arch.bsp_vcpu_id = arg;
5099 		mutex_unlock(&kvm->lock);
5100 		break;
5101 	case KVM_XEN_HVM_CONFIG: {
5102 		struct kvm_xen_hvm_config xhc;
5103 		r = -EFAULT;
5104 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
5105 			goto out;
5106 		r = -EINVAL;
5107 		if (xhc.flags)
5108 			goto out;
5109 		memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5110 		r = 0;
5111 		break;
5112 	}
5113 	case KVM_SET_CLOCK: {
5114 		struct kvm_clock_data user_ns;
5115 		u64 now_ns;
5116 
5117 		r = -EFAULT;
5118 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5119 			goto out;
5120 
5121 		r = -EINVAL;
5122 		if (user_ns.flags)
5123 			goto out;
5124 
5125 		r = 0;
5126 		/*
5127 		 * TODO: userspace has to take care of races with VCPU_RUN, so
5128 		 * kvm_gen_update_masterclock() can be cut down to locked
5129 		 * pvclock_update_vm_gtod_copy().
5130 		 */
5131 		kvm_gen_update_masterclock(kvm);
5132 		now_ns = get_kvmclock_ns(kvm);
5133 		kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5134 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5135 		break;
5136 	}
5137 	case KVM_GET_CLOCK: {
5138 		struct kvm_clock_data user_ns;
5139 		u64 now_ns;
5140 
5141 		now_ns = get_kvmclock_ns(kvm);
5142 		user_ns.clock = now_ns;
5143 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5144 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5145 
5146 		r = -EFAULT;
5147 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5148 			goto out;
5149 		r = 0;
5150 		break;
5151 	}
5152 	case KVM_MEMORY_ENCRYPT_OP: {
5153 		r = -ENOTTY;
5154 		if (kvm_x86_ops.mem_enc_op)
5155 			r = kvm_x86_ops.mem_enc_op(kvm, argp);
5156 		break;
5157 	}
5158 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
5159 		struct kvm_enc_region region;
5160 
5161 		r = -EFAULT;
5162 		if (copy_from_user(&region, argp, sizeof(region)))
5163 			goto out;
5164 
5165 		r = -ENOTTY;
5166 		if (kvm_x86_ops.mem_enc_reg_region)
5167 			r = kvm_x86_ops.mem_enc_reg_region(kvm, &region);
5168 		break;
5169 	}
5170 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5171 		struct kvm_enc_region region;
5172 
5173 		r = -EFAULT;
5174 		if (copy_from_user(&region, argp, sizeof(region)))
5175 			goto out;
5176 
5177 		r = -ENOTTY;
5178 		if (kvm_x86_ops.mem_enc_unreg_region)
5179 			r = kvm_x86_ops.mem_enc_unreg_region(kvm, &region);
5180 		break;
5181 	}
5182 	case KVM_HYPERV_EVENTFD: {
5183 		struct kvm_hyperv_eventfd hvevfd;
5184 
5185 		r = -EFAULT;
5186 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5187 			goto out;
5188 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5189 		break;
5190 	}
5191 	case KVM_SET_PMU_EVENT_FILTER:
5192 		r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5193 		break;
5194 	default:
5195 		r = -ENOTTY;
5196 	}
5197 out:
5198 	return r;
5199 }
5200 
5201 static void kvm_init_msr_list(void)
5202 {
5203 	struct x86_pmu_capability x86_pmu;
5204 	u32 dummy[2];
5205 	unsigned i;
5206 
5207 	BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5208 			 "Please update the fixed PMCs in msrs_to_saved_all[]");
5209 
5210 	perf_get_x86_pmu_capability(&x86_pmu);
5211 
5212 	num_msrs_to_save = 0;
5213 	num_emulated_msrs = 0;
5214 	num_msr_based_features = 0;
5215 
5216 	for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5217 		if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5218 			continue;
5219 
5220 		/*
5221 		 * Even MSRs that are valid in the host may not be exposed
5222 		 * to the guests in some cases.
5223 		 */
5224 		switch (msrs_to_save_all[i]) {
5225 		case MSR_IA32_BNDCFGS:
5226 			if (!kvm_mpx_supported())
5227 				continue;
5228 			break;
5229 		case MSR_TSC_AUX:
5230 			if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
5231 				continue;
5232 			break;
5233 		case MSR_IA32_RTIT_CTL:
5234 		case MSR_IA32_RTIT_STATUS:
5235 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
5236 				continue;
5237 			break;
5238 		case MSR_IA32_RTIT_CR3_MATCH:
5239 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5240 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5241 				continue;
5242 			break;
5243 		case MSR_IA32_RTIT_OUTPUT_BASE:
5244 		case MSR_IA32_RTIT_OUTPUT_MASK:
5245 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5246 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5247 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5248 				continue;
5249 			break;
5250 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
5251 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5252 				msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5253 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5254 				continue;
5255 			break;
5256 		case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5257 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5258 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5259 				continue;
5260 			break;
5261 		case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5262 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5263 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5264 				continue;
5265 		}
5266 		default:
5267 			break;
5268 		}
5269 
5270 		msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5271 	}
5272 
5273 	for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5274 		if (!kvm_x86_ops.has_emulated_msr(emulated_msrs_all[i]))
5275 			continue;
5276 
5277 		emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5278 	}
5279 
5280 	for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5281 		struct kvm_msr_entry msr;
5282 
5283 		msr.index = msr_based_features_all[i];
5284 		if (kvm_get_msr_feature(&msr))
5285 			continue;
5286 
5287 		msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
5288 	}
5289 }
5290 
5291 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5292 			   const void *v)
5293 {
5294 	int handled = 0;
5295 	int n;
5296 
5297 	do {
5298 		n = min(len, 8);
5299 		if (!(lapic_in_kernel(vcpu) &&
5300 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5301 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5302 			break;
5303 		handled += n;
5304 		addr += n;
5305 		len -= n;
5306 		v += n;
5307 	} while (len);
5308 
5309 	return handled;
5310 }
5311 
5312 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5313 {
5314 	int handled = 0;
5315 	int n;
5316 
5317 	do {
5318 		n = min(len, 8);
5319 		if (!(lapic_in_kernel(vcpu) &&
5320 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5321 					 addr, n, v))
5322 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5323 			break;
5324 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5325 		handled += n;
5326 		addr += n;
5327 		len -= n;
5328 		v += n;
5329 	} while (len);
5330 
5331 	return handled;
5332 }
5333 
5334 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5335 			struct kvm_segment *var, int seg)
5336 {
5337 	kvm_x86_ops.set_segment(vcpu, var, seg);
5338 }
5339 
5340 void kvm_get_segment(struct kvm_vcpu *vcpu,
5341 		     struct kvm_segment *var, int seg)
5342 {
5343 	kvm_x86_ops.get_segment(vcpu, var, seg);
5344 }
5345 
5346 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5347 			   struct x86_exception *exception)
5348 {
5349 	gpa_t t_gpa;
5350 
5351 	BUG_ON(!mmu_is_nested(vcpu));
5352 
5353 	/* NPT walks are always user-walks */
5354 	access |= PFERR_USER_MASK;
5355 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5356 
5357 	return t_gpa;
5358 }
5359 
5360 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5361 			      struct x86_exception *exception)
5362 {
5363 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5364 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5365 }
5366 
5367  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5368 				struct x86_exception *exception)
5369 {
5370 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5371 	access |= PFERR_FETCH_MASK;
5372 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5373 }
5374 
5375 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5376 			       struct x86_exception *exception)
5377 {
5378 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5379 	access |= PFERR_WRITE_MASK;
5380 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5381 }
5382 
5383 /* uses this to access any guest's mapped memory without checking CPL */
5384 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5385 				struct x86_exception *exception)
5386 {
5387 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5388 }
5389 
5390 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5391 				      struct kvm_vcpu *vcpu, u32 access,
5392 				      struct x86_exception *exception)
5393 {
5394 	void *data = val;
5395 	int r = X86EMUL_CONTINUE;
5396 
5397 	while (bytes) {
5398 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5399 							    exception);
5400 		unsigned offset = addr & (PAGE_SIZE-1);
5401 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5402 		int ret;
5403 
5404 		if (gpa == UNMAPPED_GVA)
5405 			return X86EMUL_PROPAGATE_FAULT;
5406 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5407 					       offset, toread);
5408 		if (ret < 0) {
5409 			r = X86EMUL_IO_NEEDED;
5410 			goto out;
5411 		}
5412 
5413 		bytes -= toread;
5414 		data += toread;
5415 		addr += toread;
5416 	}
5417 out:
5418 	return r;
5419 }
5420 
5421 /* used for instruction fetching */
5422 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5423 				gva_t addr, void *val, unsigned int bytes,
5424 				struct x86_exception *exception)
5425 {
5426 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5427 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5428 	unsigned offset;
5429 	int ret;
5430 
5431 	/* Inline kvm_read_guest_virt_helper for speed.  */
5432 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5433 						    exception);
5434 	if (unlikely(gpa == UNMAPPED_GVA))
5435 		return X86EMUL_PROPAGATE_FAULT;
5436 
5437 	offset = addr & (PAGE_SIZE-1);
5438 	if (WARN_ON(offset + bytes > PAGE_SIZE))
5439 		bytes = (unsigned)PAGE_SIZE - offset;
5440 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5441 				       offset, bytes);
5442 	if (unlikely(ret < 0))
5443 		return X86EMUL_IO_NEEDED;
5444 
5445 	return X86EMUL_CONTINUE;
5446 }
5447 
5448 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5449 			       gva_t addr, void *val, unsigned int bytes,
5450 			       struct x86_exception *exception)
5451 {
5452 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5453 
5454 	/*
5455 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5456 	 * is returned, but our callers are not ready for that and they blindly
5457 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
5458 	 * uninitialized kernel stack memory into cr2 and error code.
5459 	 */
5460 	memset(exception, 0, sizeof(*exception));
5461 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5462 					  exception);
5463 }
5464 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5465 
5466 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5467 			     gva_t addr, void *val, unsigned int bytes,
5468 			     struct x86_exception *exception, bool system)
5469 {
5470 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5471 	u32 access = 0;
5472 
5473 	if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
5474 		access |= PFERR_USER_MASK;
5475 
5476 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5477 }
5478 
5479 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5480 		unsigned long addr, void *val, unsigned int bytes)
5481 {
5482 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5483 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5484 
5485 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5486 }
5487 
5488 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5489 				      struct kvm_vcpu *vcpu, u32 access,
5490 				      struct x86_exception *exception)
5491 {
5492 	void *data = val;
5493 	int r = X86EMUL_CONTINUE;
5494 
5495 	while (bytes) {
5496 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5497 							     access,
5498 							     exception);
5499 		unsigned offset = addr & (PAGE_SIZE-1);
5500 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5501 		int ret;
5502 
5503 		if (gpa == UNMAPPED_GVA)
5504 			return X86EMUL_PROPAGATE_FAULT;
5505 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5506 		if (ret < 0) {
5507 			r = X86EMUL_IO_NEEDED;
5508 			goto out;
5509 		}
5510 
5511 		bytes -= towrite;
5512 		data += towrite;
5513 		addr += towrite;
5514 	}
5515 out:
5516 	return r;
5517 }
5518 
5519 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5520 			      unsigned int bytes, struct x86_exception *exception,
5521 			      bool system)
5522 {
5523 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5524 	u32 access = PFERR_WRITE_MASK;
5525 
5526 	if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
5527 		access |= PFERR_USER_MASK;
5528 
5529 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5530 					   access, exception);
5531 }
5532 
5533 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5534 				unsigned int bytes, struct x86_exception *exception)
5535 {
5536 	/* kvm_write_guest_virt_system can pull in tons of pages. */
5537 	vcpu->arch.l1tf_flush_l1d = true;
5538 
5539 	/*
5540 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5541 	 * is returned, but our callers are not ready for that and they blindly
5542 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
5543 	 * uninitialized kernel stack memory into cr2 and error code.
5544 	 */
5545 	memset(exception, 0, sizeof(*exception));
5546 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5547 					   PFERR_WRITE_MASK, exception);
5548 }
5549 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5550 
5551 int handle_ud(struct kvm_vcpu *vcpu)
5552 {
5553 	static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
5554 	int emul_type = EMULTYPE_TRAP_UD;
5555 	char sig[5]; /* ud2; .ascii "kvm" */
5556 	struct x86_exception e;
5557 
5558 	if (force_emulation_prefix &&
5559 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5560 				sig, sizeof(sig), &e) == 0 &&
5561 	    memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
5562 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5563 		emul_type = EMULTYPE_TRAP_UD_FORCED;
5564 	}
5565 
5566 	return kvm_emulate_instruction(vcpu, emul_type);
5567 }
5568 EXPORT_SYMBOL_GPL(handle_ud);
5569 
5570 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5571 			    gpa_t gpa, bool write)
5572 {
5573 	/* For APIC access vmexit */
5574 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5575 		return 1;
5576 
5577 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5578 		trace_vcpu_match_mmio(gva, gpa, write, true);
5579 		return 1;
5580 	}
5581 
5582 	return 0;
5583 }
5584 
5585 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5586 				gpa_t *gpa, struct x86_exception *exception,
5587 				bool write)
5588 {
5589 	u32 access = ((kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5590 		| (write ? PFERR_WRITE_MASK : 0);
5591 
5592 	/*
5593 	 * currently PKRU is only applied to ept enabled guest so
5594 	 * there is no pkey in EPT page table for L1 guest or EPT
5595 	 * shadow page table for L2 guest.
5596 	 */
5597 	if (vcpu_match_mmio_gva(vcpu, gva)
5598 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5599 				 vcpu->arch.mmio_access, 0, access)) {
5600 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5601 					(gva & (PAGE_SIZE - 1));
5602 		trace_vcpu_match_mmio(gva, *gpa, write, false);
5603 		return 1;
5604 	}
5605 
5606 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5607 
5608 	if (*gpa == UNMAPPED_GVA)
5609 		return -1;
5610 
5611 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5612 }
5613 
5614 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5615 			const void *val, int bytes)
5616 {
5617 	int ret;
5618 
5619 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5620 	if (ret < 0)
5621 		return 0;
5622 	kvm_page_track_write(vcpu, gpa, val, bytes);
5623 	return 1;
5624 }
5625 
5626 struct read_write_emulator_ops {
5627 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5628 				  int bytes);
5629 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5630 				  void *val, int bytes);
5631 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5632 			       int bytes, void *val);
5633 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5634 				    void *val, int bytes);
5635 	bool write;
5636 };
5637 
5638 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5639 {
5640 	if (vcpu->mmio_read_completed) {
5641 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5642 			       vcpu->mmio_fragments[0].gpa, val);
5643 		vcpu->mmio_read_completed = 0;
5644 		return 1;
5645 	}
5646 
5647 	return 0;
5648 }
5649 
5650 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5651 			void *val, int bytes)
5652 {
5653 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5654 }
5655 
5656 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5657 			 void *val, int bytes)
5658 {
5659 	return emulator_write_phys(vcpu, gpa, val, bytes);
5660 }
5661 
5662 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5663 {
5664 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5665 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
5666 }
5667 
5668 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5669 			  void *val, int bytes)
5670 {
5671 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5672 	return X86EMUL_IO_NEEDED;
5673 }
5674 
5675 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5676 			   void *val, int bytes)
5677 {
5678 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5679 
5680 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5681 	return X86EMUL_CONTINUE;
5682 }
5683 
5684 static const struct read_write_emulator_ops read_emultor = {
5685 	.read_write_prepare = read_prepare,
5686 	.read_write_emulate = read_emulate,
5687 	.read_write_mmio = vcpu_mmio_read,
5688 	.read_write_exit_mmio = read_exit_mmio,
5689 };
5690 
5691 static const struct read_write_emulator_ops write_emultor = {
5692 	.read_write_emulate = write_emulate,
5693 	.read_write_mmio = write_mmio,
5694 	.read_write_exit_mmio = write_exit_mmio,
5695 	.write = true,
5696 };
5697 
5698 static int emulator_read_write_onepage(unsigned long addr, void *val,
5699 				       unsigned int bytes,
5700 				       struct x86_exception *exception,
5701 				       struct kvm_vcpu *vcpu,
5702 				       const struct read_write_emulator_ops *ops)
5703 {
5704 	gpa_t gpa;
5705 	int handled, ret;
5706 	bool write = ops->write;
5707 	struct kvm_mmio_fragment *frag;
5708 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
5709 
5710 	/*
5711 	 * If the exit was due to a NPF we may already have a GPA.
5712 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5713 	 * Note, this cannot be used on string operations since string
5714 	 * operation using rep will only have the initial GPA from the NPF
5715 	 * occurred.
5716 	 */
5717 	if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
5718 	    (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
5719 		gpa = ctxt->gpa_val;
5720 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5721 	} else {
5722 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5723 		if (ret < 0)
5724 			return X86EMUL_PROPAGATE_FAULT;
5725 	}
5726 
5727 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5728 		return X86EMUL_CONTINUE;
5729 
5730 	/*
5731 	 * Is this MMIO handled locally?
5732 	 */
5733 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5734 	if (handled == bytes)
5735 		return X86EMUL_CONTINUE;
5736 
5737 	gpa += handled;
5738 	bytes -= handled;
5739 	val += handled;
5740 
5741 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5742 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5743 	frag->gpa = gpa;
5744 	frag->data = val;
5745 	frag->len = bytes;
5746 	return X86EMUL_CONTINUE;
5747 }
5748 
5749 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5750 			unsigned long addr,
5751 			void *val, unsigned int bytes,
5752 			struct x86_exception *exception,
5753 			const struct read_write_emulator_ops *ops)
5754 {
5755 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5756 	gpa_t gpa;
5757 	int rc;
5758 
5759 	if (ops->read_write_prepare &&
5760 		  ops->read_write_prepare(vcpu, val, bytes))
5761 		return X86EMUL_CONTINUE;
5762 
5763 	vcpu->mmio_nr_fragments = 0;
5764 
5765 	/* Crossing a page boundary? */
5766 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5767 		int now;
5768 
5769 		now = -addr & ~PAGE_MASK;
5770 		rc = emulator_read_write_onepage(addr, val, now, exception,
5771 						 vcpu, ops);
5772 
5773 		if (rc != X86EMUL_CONTINUE)
5774 			return rc;
5775 		addr += now;
5776 		if (ctxt->mode != X86EMUL_MODE_PROT64)
5777 			addr = (u32)addr;
5778 		val += now;
5779 		bytes -= now;
5780 	}
5781 
5782 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
5783 					 vcpu, ops);
5784 	if (rc != X86EMUL_CONTINUE)
5785 		return rc;
5786 
5787 	if (!vcpu->mmio_nr_fragments)
5788 		return rc;
5789 
5790 	gpa = vcpu->mmio_fragments[0].gpa;
5791 
5792 	vcpu->mmio_needed = 1;
5793 	vcpu->mmio_cur_fragment = 0;
5794 
5795 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5796 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5797 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
5798 	vcpu->run->mmio.phys_addr = gpa;
5799 
5800 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5801 }
5802 
5803 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5804 				  unsigned long addr,
5805 				  void *val,
5806 				  unsigned int bytes,
5807 				  struct x86_exception *exception)
5808 {
5809 	return emulator_read_write(ctxt, addr, val, bytes,
5810 				   exception, &read_emultor);
5811 }
5812 
5813 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5814 			    unsigned long addr,
5815 			    const void *val,
5816 			    unsigned int bytes,
5817 			    struct x86_exception *exception)
5818 {
5819 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
5820 				   exception, &write_emultor);
5821 }
5822 
5823 #define CMPXCHG_TYPE(t, ptr, old, new) \
5824 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5825 
5826 #ifdef CONFIG_X86_64
5827 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5828 #else
5829 #  define CMPXCHG64(ptr, old, new) \
5830 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5831 #endif
5832 
5833 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5834 				     unsigned long addr,
5835 				     const void *old,
5836 				     const void *new,
5837 				     unsigned int bytes,
5838 				     struct x86_exception *exception)
5839 {
5840 	struct kvm_host_map map;
5841 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5842 	gpa_t gpa;
5843 	char *kaddr;
5844 	bool exchanged;
5845 
5846 	/* guests cmpxchg8b have to be emulated atomically */
5847 	if (bytes > 8 || (bytes & (bytes - 1)))
5848 		goto emul_write;
5849 
5850 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5851 
5852 	if (gpa == UNMAPPED_GVA ||
5853 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5854 		goto emul_write;
5855 
5856 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5857 		goto emul_write;
5858 
5859 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
5860 		goto emul_write;
5861 
5862 	kaddr = map.hva + offset_in_page(gpa);
5863 
5864 	switch (bytes) {
5865 	case 1:
5866 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5867 		break;
5868 	case 2:
5869 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5870 		break;
5871 	case 4:
5872 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5873 		break;
5874 	case 8:
5875 		exchanged = CMPXCHG64(kaddr, old, new);
5876 		break;
5877 	default:
5878 		BUG();
5879 	}
5880 
5881 	kvm_vcpu_unmap(vcpu, &map, true);
5882 
5883 	if (!exchanged)
5884 		return X86EMUL_CMPXCHG_FAILED;
5885 
5886 	kvm_page_track_write(vcpu, gpa, new, bytes);
5887 
5888 	return X86EMUL_CONTINUE;
5889 
5890 emul_write:
5891 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5892 
5893 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5894 }
5895 
5896 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5897 {
5898 	int r = 0, i;
5899 
5900 	for (i = 0; i < vcpu->arch.pio.count; i++) {
5901 		if (vcpu->arch.pio.in)
5902 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5903 					    vcpu->arch.pio.size, pd);
5904 		else
5905 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5906 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
5907 					     pd);
5908 		if (r)
5909 			break;
5910 		pd += vcpu->arch.pio.size;
5911 	}
5912 	return r;
5913 }
5914 
5915 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5916 			       unsigned short port, void *val,
5917 			       unsigned int count, bool in)
5918 {
5919 	vcpu->arch.pio.port = port;
5920 	vcpu->arch.pio.in = in;
5921 	vcpu->arch.pio.count  = count;
5922 	vcpu->arch.pio.size = size;
5923 
5924 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5925 		vcpu->arch.pio.count = 0;
5926 		return 1;
5927 	}
5928 
5929 	vcpu->run->exit_reason = KVM_EXIT_IO;
5930 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5931 	vcpu->run->io.size = size;
5932 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5933 	vcpu->run->io.count = count;
5934 	vcpu->run->io.port = port;
5935 
5936 	return 0;
5937 }
5938 
5939 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
5940 			   unsigned short port, void *val, unsigned int count)
5941 {
5942 	int ret;
5943 
5944 	if (vcpu->arch.pio.count)
5945 		goto data_avail;
5946 
5947 	memset(vcpu->arch.pio_data, 0, size * count);
5948 
5949 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5950 	if (ret) {
5951 data_avail:
5952 		memcpy(val, vcpu->arch.pio_data, size * count);
5953 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5954 		vcpu->arch.pio.count = 0;
5955 		return 1;
5956 	}
5957 
5958 	return 0;
5959 }
5960 
5961 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5962 				    int size, unsigned short port, void *val,
5963 				    unsigned int count)
5964 {
5965 	return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
5966 
5967 }
5968 
5969 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
5970 			    unsigned short port, const void *val,
5971 			    unsigned int count)
5972 {
5973 	memcpy(vcpu->arch.pio_data, val, size * count);
5974 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5975 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5976 }
5977 
5978 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5979 				     int size, unsigned short port,
5980 				     const void *val, unsigned int count)
5981 {
5982 	return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
5983 }
5984 
5985 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5986 {
5987 	return kvm_x86_ops.get_segment_base(vcpu, seg);
5988 }
5989 
5990 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5991 {
5992 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5993 }
5994 
5995 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5996 {
5997 	if (!need_emulate_wbinvd(vcpu))
5998 		return X86EMUL_CONTINUE;
5999 
6000 	if (kvm_x86_ops.has_wbinvd_exit()) {
6001 		int cpu = get_cpu();
6002 
6003 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6004 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
6005 				wbinvd_ipi, NULL, 1);
6006 		put_cpu();
6007 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6008 	} else
6009 		wbinvd();
6010 	return X86EMUL_CONTINUE;
6011 }
6012 
6013 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6014 {
6015 	kvm_emulate_wbinvd_noskip(vcpu);
6016 	return kvm_skip_emulated_instruction(vcpu);
6017 }
6018 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6019 
6020 
6021 
6022 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6023 {
6024 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6025 }
6026 
6027 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6028 			   unsigned long *dest)
6029 {
6030 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6031 }
6032 
6033 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6034 			   unsigned long value)
6035 {
6036 
6037 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6038 }
6039 
6040 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6041 {
6042 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6043 }
6044 
6045 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6046 {
6047 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6048 	unsigned long value;
6049 
6050 	switch (cr) {
6051 	case 0:
6052 		value = kvm_read_cr0(vcpu);
6053 		break;
6054 	case 2:
6055 		value = vcpu->arch.cr2;
6056 		break;
6057 	case 3:
6058 		value = kvm_read_cr3(vcpu);
6059 		break;
6060 	case 4:
6061 		value = kvm_read_cr4(vcpu);
6062 		break;
6063 	case 8:
6064 		value = kvm_get_cr8(vcpu);
6065 		break;
6066 	default:
6067 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
6068 		return 0;
6069 	}
6070 
6071 	return value;
6072 }
6073 
6074 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6075 {
6076 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6077 	int res = 0;
6078 
6079 	switch (cr) {
6080 	case 0:
6081 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6082 		break;
6083 	case 2:
6084 		vcpu->arch.cr2 = val;
6085 		break;
6086 	case 3:
6087 		res = kvm_set_cr3(vcpu, val);
6088 		break;
6089 	case 4:
6090 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6091 		break;
6092 	case 8:
6093 		res = kvm_set_cr8(vcpu, val);
6094 		break;
6095 	default:
6096 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
6097 		res = -1;
6098 	}
6099 
6100 	return res;
6101 }
6102 
6103 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6104 {
6105 	return kvm_x86_ops.get_cpl(emul_to_vcpu(ctxt));
6106 }
6107 
6108 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6109 {
6110 	kvm_x86_ops.get_gdt(emul_to_vcpu(ctxt), dt);
6111 }
6112 
6113 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6114 {
6115 	kvm_x86_ops.get_idt(emul_to_vcpu(ctxt), dt);
6116 }
6117 
6118 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6119 {
6120 	kvm_x86_ops.set_gdt(emul_to_vcpu(ctxt), dt);
6121 }
6122 
6123 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6124 {
6125 	kvm_x86_ops.set_idt(emul_to_vcpu(ctxt), dt);
6126 }
6127 
6128 static unsigned long emulator_get_cached_segment_base(
6129 	struct x86_emulate_ctxt *ctxt, int seg)
6130 {
6131 	return get_segment_base(emul_to_vcpu(ctxt), seg);
6132 }
6133 
6134 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6135 				 struct desc_struct *desc, u32 *base3,
6136 				 int seg)
6137 {
6138 	struct kvm_segment var;
6139 
6140 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6141 	*selector = var.selector;
6142 
6143 	if (var.unusable) {
6144 		memset(desc, 0, sizeof(*desc));
6145 		if (base3)
6146 			*base3 = 0;
6147 		return false;
6148 	}
6149 
6150 	if (var.g)
6151 		var.limit >>= 12;
6152 	set_desc_limit(desc, var.limit);
6153 	set_desc_base(desc, (unsigned long)var.base);
6154 #ifdef CONFIG_X86_64
6155 	if (base3)
6156 		*base3 = var.base >> 32;
6157 #endif
6158 	desc->type = var.type;
6159 	desc->s = var.s;
6160 	desc->dpl = var.dpl;
6161 	desc->p = var.present;
6162 	desc->avl = var.avl;
6163 	desc->l = var.l;
6164 	desc->d = var.db;
6165 	desc->g = var.g;
6166 
6167 	return true;
6168 }
6169 
6170 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6171 				 struct desc_struct *desc, u32 base3,
6172 				 int seg)
6173 {
6174 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6175 	struct kvm_segment var;
6176 
6177 	var.selector = selector;
6178 	var.base = get_desc_base(desc);
6179 #ifdef CONFIG_X86_64
6180 	var.base |= ((u64)base3) << 32;
6181 #endif
6182 	var.limit = get_desc_limit(desc);
6183 	if (desc->g)
6184 		var.limit = (var.limit << 12) | 0xfff;
6185 	var.type = desc->type;
6186 	var.dpl = desc->dpl;
6187 	var.db = desc->d;
6188 	var.s = desc->s;
6189 	var.l = desc->l;
6190 	var.g = desc->g;
6191 	var.avl = desc->avl;
6192 	var.present = desc->p;
6193 	var.unusable = !var.present;
6194 	var.padding = 0;
6195 
6196 	kvm_set_segment(vcpu, &var, seg);
6197 	return;
6198 }
6199 
6200 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6201 			    u32 msr_index, u64 *pdata)
6202 {
6203 	return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
6204 }
6205 
6206 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6207 			    u32 msr_index, u64 data)
6208 {
6209 	return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
6210 }
6211 
6212 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6213 {
6214 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6215 
6216 	return vcpu->arch.smbase;
6217 }
6218 
6219 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6220 {
6221 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6222 
6223 	vcpu->arch.smbase = smbase;
6224 }
6225 
6226 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6227 			      u32 pmc)
6228 {
6229 	return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6230 }
6231 
6232 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6233 			     u32 pmc, u64 *pdata)
6234 {
6235 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6236 }
6237 
6238 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6239 {
6240 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
6241 }
6242 
6243 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6244 			      struct x86_instruction_info *info,
6245 			      enum x86_intercept_stage stage)
6246 {
6247 	return kvm_x86_ops.check_intercept(emul_to_vcpu(ctxt), info, stage,
6248 					    &ctxt->exception);
6249 }
6250 
6251 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6252 			      u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
6253 			      bool exact_only)
6254 {
6255 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
6256 }
6257 
6258 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
6259 {
6260 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
6261 }
6262 
6263 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
6264 {
6265 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
6266 }
6267 
6268 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
6269 {
6270 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
6271 }
6272 
6273 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6274 {
6275 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
6276 }
6277 
6278 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6279 {
6280 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6281 }
6282 
6283 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6284 {
6285 	kvm_x86_ops.set_nmi_mask(emul_to_vcpu(ctxt), masked);
6286 }
6287 
6288 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6289 {
6290 	return emul_to_vcpu(ctxt)->arch.hflags;
6291 }
6292 
6293 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6294 {
6295 	emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6296 }
6297 
6298 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6299 				  const char *smstate)
6300 {
6301 	return kvm_x86_ops.pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6302 }
6303 
6304 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6305 {
6306 	kvm_smm_changed(emul_to_vcpu(ctxt));
6307 }
6308 
6309 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6310 {
6311 	return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6312 }
6313 
6314 static const struct x86_emulate_ops emulate_ops = {
6315 	.read_gpr            = emulator_read_gpr,
6316 	.write_gpr           = emulator_write_gpr,
6317 	.read_std            = emulator_read_std,
6318 	.write_std           = emulator_write_std,
6319 	.read_phys           = kvm_read_guest_phys_system,
6320 	.fetch               = kvm_fetch_guest_virt,
6321 	.read_emulated       = emulator_read_emulated,
6322 	.write_emulated      = emulator_write_emulated,
6323 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
6324 	.invlpg              = emulator_invlpg,
6325 	.pio_in_emulated     = emulator_pio_in_emulated,
6326 	.pio_out_emulated    = emulator_pio_out_emulated,
6327 	.get_segment         = emulator_get_segment,
6328 	.set_segment         = emulator_set_segment,
6329 	.get_cached_segment_base = emulator_get_cached_segment_base,
6330 	.get_gdt             = emulator_get_gdt,
6331 	.get_idt	     = emulator_get_idt,
6332 	.set_gdt             = emulator_set_gdt,
6333 	.set_idt	     = emulator_set_idt,
6334 	.get_cr              = emulator_get_cr,
6335 	.set_cr              = emulator_set_cr,
6336 	.cpl                 = emulator_get_cpl,
6337 	.get_dr              = emulator_get_dr,
6338 	.set_dr              = emulator_set_dr,
6339 	.get_smbase          = emulator_get_smbase,
6340 	.set_smbase          = emulator_set_smbase,
6341 	.set_msr             = emulator_set_msr,
6342 	.get_msr             = emulator_get_msr,
6343 	.check_pmc	     = emulator_check_pmc,
6344 	.read_pmc            = emulator_read_pmc,
6345 	.halt                = emulator_halt,
6346 	.wbinvd              = emulator_wbinvd,
6347 	.fix_hypercall       = emulator_fix_hypercall,
6348 	.intercept           = emulator_intercept,
6349 	.get_cpuid           = emulator_get_cpuid,
6350 	.guest_has_long_mode = emulator_guest_has_long_mode,
6351 	.guest_has_movbe     = emulator_guest_has_movbe,
6352 	.guest_has_fxsr      = emulator_guest_has_fxsr,
6353 	.set_nmi_mask        = emulator_set_nmi_mask,
6354 	.get_hflags          = emulator_get_hflags,
6355 	.set_hflags          = emulator_set_hflags,
6356 	.pre_leave_smm       = emulator_pre_leave_smm,
6357 	.post_leave_smm      = emulator_post_leave_smm,
6358 	.set_xcr             = emulator_set_xcr,
6359 };
6360 
6361 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6362 {
6363 	u32 int_shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
6364 	/*
6365 	 * an sti; sti; sequence only disable interrupts for the first
6366 	 * instruction. So, if the last instruction, be it emulated or
6367 	 * not, left the system with the INT_STI flag enabled, it
6368 	 * means that the last instruction is an sti. We should not
6369 	 * leave the flag on in this case. The same goes for mov ss
6370 	 */
6371 	if (int_shadow & mask)
6372 		mask = 0;
6373 	if (unlikely(int_shadow || mask)) {
6374 		kvm_x86_ops.set_interrupt_shadow(vcpu, mask);
6375 		if (!mask)
6376 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6377 	}
6378 }
6379 
6380 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6381 {
6382 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6383 	if (ctxt->exception.vector == PF_VECTOR)
6384 		return kvm_propagate_fault(vcpu, &ctxt->exception);
6385 
6386 	if (ctxt->exception.error_code_valid)
6387 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6388 				      ctxt->exception.error_code);
6389 	else
6390 		kvm_queue_exception(vcpu, ctxt->exception.vector);
6391 	return false;
6392 }
6393 
6394 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
6395 {
6396 	struct x86_emulate_ctxt *ctxt;
6397 
6398 	ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
6399 	if (!ctxt) {
6400 		pr_err("kvm: failed to allocate vcpu's emulator\n");
6401 		return NULL;
6402 	}
6403 
6404 	ctxt->vcpu = vcpu;
6405 	ctxt->ops = &emulate_ops;
6406 	vcpu->arch.emulate_ctxt = ctxt;
6407 
6408 	return ctxt;
6409 }
6410 
6411 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6412 {
6413 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6414 	int cs_db, cs_l;
6415 
6416 	kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6417 
6418 	ctxt->gpa_available = false;
6419 	ctxt->eflags = kvm_get_rflags(vcpu);
6420 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6421 
6422 	ctxt->eip = kvm_rip_read(vcpu);
6423 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
6424 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
6425 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
6426 		     cs_db				? X86EMUL_MODE_PROT32 :
6427 							  X86EMUL_MODE_PROT16;
6428 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6429 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6430 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6431 
6432 	init_decode_cache(ctxt);
6433 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6434 }
6435 
6436 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6437 {
6438 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6439 	int ret;
6440 
6441 	init_emulate_ctxt(vcpu);
6442 
6443 	ctxt->op_bytes = 2;
6444 	ctxt->ad_bytes = 2;
6445 	ctxt->_eip = ctxt->eip + inc_eip;
6446 	ret = emulate_int_real(ctxt, irq);
6447 
6448 	if (ret != X86EMUL_CONTINUE) {
6449 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6450 	} else {
6451 		ctxt->eip = ctxt->_eip;
6452 		kvm_rip_write(vcpu, ctxt->eip);
6453 		kvm_set_rflags(vcpu, ctxt->eflags);
6454 	}
6455 }
6456 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6457 
6458 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6459 {
6460 	++vcpu->stat.insn_emulation_fail;
6461 	trace_kvm_emulate_insn_failed(vcpu);
6462 
6463 	if (emulation_type & EMULTYPE_VMWARE_GP) {
6464 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6465 		return 1;
6466 	}
6467 
6468 	if (emulation_type & EMULTYPE_SKIP) {
6469 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6470 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6471 		vcpu->run->internal.ndata = 0;
6472 		return 0;
6473 	}
6474 
6475 	kvm_queue_exception(vcpu, UD_VECTOR);
6476 
6477 	if (!is_guest_mode(vcpu) && kvm_x86_ops.get_cpl(vcpu) == 0) {
6478 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6479 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6480 		vcpu->run->internal.ndata = 0;
6481 		return 0;
6482 	}
6483 
6484 	return 1;
6485 }
6486 
6487 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
6488 				  bool write_fault_to_shadow_pgtable,
6489 				  int emulation_type)
6490 {
6491 	gpa_t gpa = cr2_or_gpa;
6492 	kvm_pfn_t pfn;
6493 
6494 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
6495 		return false;
6496 
6497 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
6498 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
6499 		return false;
6500 
6501 	if (!vcpu->arch.mmu->direct_map) {
6502 		/*
6503 		 * Write permission should be allowed since only
6504 		 * write access need to be emulated.
6505 		 */
6506 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
6507 
6508 		/*
6509 		 * If the mapping is invalid in guest, let cpu retry
6510 		 * it to generate fault.
6511 		 */
6512 		if (gpa == UNMAPPED_GVA)
6513 			return true;
6514 	}
6515 
6516 	/*
6517 	 * Do not retry the unhandleable instruction if it faults on the
6518 	 * readonly host memory, otherwise it will goto a infinite loop:
6519 	 * retry instruction -> write #PF -> emulation fail -> retry
6520 	 * instruction -> ...
6521 	 */
6522 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6523 
6524 	/*
6525 	 * If the instruction failed on the error pfn, it can not be fixed,
6526 	 * report the error to userspace.
6527 	 */
6528 	if (is_error_noslot_pfn(pfn))
6529 		return false;
6530 
6531 	kvm_release_pfn_clean(pfn);
6532 
6533 	/* The instructions are well-emulated on direct mmu. */
6534 	if (vcpu->arch.mmu->direct_map) {
6535 		unsigned int indirect_shadow_pages;
6536 
6537 		spin_lock(&vcpu->kvm->mmu_lock);
6538 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6539 		spin_unlock(&vcpu->kvm->mmu_lock);
6540 
6541 		if (indirect_shadow_pages)
6542 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6543 
6544 		return true;
6545 	}
6546 
6547 	/*
6548 	 * if emulation was due to access to shadowed page table
6549 	 * and it failed try to unshadow page and re-enter the
6550 	 * guest to let CPU execute the instruction.
6551 	 */
6552 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6553 
6554 	/*
6555 	 * If the access faults on its page table, it can not
6556 	 * be fixed by unprotecting shadow page and it should
6557 	 * be reported to userspace.
6558 	 */
6559 	return !write_fault_to_shadow_pgtable;
6560 }
6561 
6562 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6563 			      gpa_t cr2_or_gpa,  int emulation_type)
6564 {
6565 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6566 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
6567 
6568 	last_retry_eip = vcpu->arch.last_retry_eip;
6569 	last_retry_addr = vcpu->arch.last_retry_addr;
6570 
6571 	/*
6572 	 * If the emulation is caused by #PF and it is non-page_table
6573 	 * writing instruction, it means the VM-EXIT is caused by shadow
6574 	 * page protected, we can zap the shadow page and retry this
6575 	 * instruction directly.
6576 	 *
6577 	 * Note: if the guest uses a non-page-table modifying instruction
6578 	 * on the PDE that points to the instruction, then we will unmap
6579 	 * the instruction and go to an infinite loop. So, we cache the
6580 	 * last retried eip and the last fault address, if we meet the eip
6581 	 * and the address again, we can break out of the potential infinite
6582 	 * loop.
6583 	 */
6584 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6585 
6586 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
6587 		return false;
6588 
6589 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
6590 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
6591 		return false;
6592 
6593 	if (x86_page_table_writing_insn(ctxt))
6594 		return false;
6595 
6596 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
6597 		return false;
6598 
6599 	vcpu->arch.last_retry_eip = ctxt->eip;
6600 	vcpu->arch.last_retry_addr = cr2_or_gpa;
6601 
6602 	if (!vcpu->arch.mmu->direct_map)
6603 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
6604 
6605 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6606 
6607 	return true;
6608 }
6609 
6610 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6611 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6612 
6613 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6614 {
6615 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6616 		/* This is a good place to trace that we are exiting SMM.  */
6617 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6618 
6619 		/* Process a latched INIT or SMI, if any.  */
6620 		kvm_make_request(KVM_REQ_EVENT, vcpu);
6621 	}
6622 
6623 	kvm_mmu_reset_context(vcpu);
6624 }
6625 
6626 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6627 				unsigned long *db)
6628 {
6629 	u32 dr6 = 0;
6630 	int i;
6631 	u32 enable, rwlen;
6632 
6633 	enable = dr7;
6634 	rwlen = dr7 >> 16;
6635 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6636 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6637 			dr6 |= (1 << i);
6638 	return dr6;
6639 }
6640 
6641 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
6642 {
6643 	struct kvm_run *kvm_run = vcpu->run;
6644 
6645 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6646 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6647 		kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6648 		kvm_run->debug.arch.exception = DB_VECTOR;
6649 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
6650 		return 0;
6651 	}
6652 	kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6653 	return 1;
6654 }
6655 
6656 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6657 {
6658 	unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
6659 	int r;
6660 
6661 	r = kvm_x86_ops.skip_emulated_instruction(vcpu);
6662 	if (unlikely(!r))
6663 		return 0;
6664 
6665 	/*
6666 	 * rflags is the old, "raw" value of the flags.  The new value has
6667 	 * not been saved yet.
6668 	 *
6669 	 * This is correct even for TF set by the guest, because "the
6670 	 * processor will not generate this exception after the instruction
6671 	 * that sets the TF flag".
6672 	 */
6673 	if (unlikely(rflags & X86_EFLAGS_TF))
6674 		r = kvm_vcpu_do_singlestep(vcpu);
6675 	return r;
6676 }
6677 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6678 
6679 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6680 {
6681 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6682 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6683 		struct kvm_run *kvm_run = vcpu->run;
6684 		unsigned long eip = kvm_get_linear_rip(vcpu);
6685 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6686 					   vcpu->arch.guest_debug_dr7,
6687 					   vcpu->arch.eff_db);
6688 
6689 		if (dr6 != 0) {
6690 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6691 			kvm_run->debug.arch.pc = eip;
6692 			kvm_run->debug.arch.exception = DB_VECTOR;
6693 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
6694 			*r = 0;
6695 			return true;
6696 		}
6697 	}
6698 
6699 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6700 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6701 		unsigned long eip = kvm_get_linear_rip(vcpu);
6702 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6703 					   vcpu->arch.dr7,
6704 					   vcpu->arch.db);
6705 
6706 		if (dr6 != 0) {
6707 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
6708 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
6709 			kvm_queue_exception(vcpu, DB_VECTOR);
6710 			*r = 1;
6711 			return true;
6712 		}
6713 	}
6714 
6715 	return false;
6716 }
6717 
6718 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6719 {
6720 	switch (ctxt->opcode_len) {
6721 	case 1:
6722 		switch (ctxt->b) {
6723 		case 0xe4:	/* IN */
6724 		case 0xe5:
6725 		case 0xec:
6726 		case 0xed:
6727 		case 0xe6:	/* OUT */
6728 		case 0xe7:
6729 		case 0xee:
6730 		case 0xef:
6731 		case 0x6c:	/* INS */
6732 		case 0x6d:
6733 		case 0x6e:	/* OUTS */
6734 		case 0x6f:
6735 			return true;
6736 		}
6737 		break;
6738 	case 2:
6739 		switch (ctxt->b) {
6740 		case 0x33:	/* RDPMC */
6741 			return true;
6742 		}
6743 		break;
6744 	}
6745 
6746 	return false;
6747 }
6748 
6749 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
6750 			    int emulation_type, void *insn, int insn_len)
6751 {
6752 	int r;
6753 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6754 	bool writeback = true;
6755 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6756 
6757 	vcpu->arch.l1tf_flush_l1d = true;
6758 
6759 	/*
6760 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
6761 	 * never reused.
6762 	 */
6763 	vcpu->arch.write_fault_to_shadow_pgtable = false;
6764 	kvm_clear_exception_queue(vcpu);
6765 
6766 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6767 		init_emulate_ctxt(vcpu);
6768 
6769 		/*
6770 		 * We will reenter on the same instruction since
6771 		 * we do not set complete_userspace_io.  This does not
6772 		 * handle watchpoints yet, those would be handled in
6773 		 * the emulate_ops.
6774 		 */
6775 		if (!(emulation_type & EMULTYPE_SKIP) &&
6776 		    kvm_vcpu_check_breakpoint(vcpu, &r))
6777 			return r;
6778 
6779 		ctxt->interruptibility = 0;
6780 		ctxt->have_exception = false;
6781 		ctxt->exception.vector = -1;
6782 		ctxt->perm_ok = false;
6783 
6784 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6785 
6786 		r = x86_decode_insn(ctxt, insn, insn_len);
6787 
6788 		trace_kvm_emulate_insn_start(vcpu);
6789 		++vcpu->stat.insn_emulation;
6790 		if (r != EMULATION_OK)  {
6791 			if ((emulation_type & EMULTYPE_TRAP_UD) ||
6792 			    (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
6793 				kvm_queue_exception(vcpu, UD_VECTOR);
6794 				return 1;
6795 			}
6796 			if (reexecute_instruction(vcpu, cr2_or_gpa,
6797 						  write_fault_to_spt,
6798 						  emulation_type))
6799 				return 1;
6800 			if (ctxt->have_exception) {
6801 				/*
6802 				 * #UD should result in just EMULATION_FAILED, and trap-like
6803 				 * exception should not be encountered during decode.
6804 				 */
6805 				WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
6806 					     exception_type(ctxt->exception.vector) == EXCPT_TRAP);
6807 				inject_emulated_exception(vcpu);
6808 				return 1;
6809 			}
6810 			return handle_emulation_failure(vcpu, emulation_type);
6811 		}
6812 	}
6813 
6814 	if ((emulation_type & EMULTYPE_VMWARE_GP) &&
6815 	    !is_vmware_backdoor_opcode(ctxt)) {
6816 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6817 		return 1;
6818 	}
6819 
6820 	/*
6821 	 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
6822 	 * for kvm_skip_emulated_instruction().  The caller is responsible for
6823 	 * updating interruptibility state and injecting single-step #DBs.
6824 	 */
6825 	if (emulation_type & EMULTYPE_SKIP) {
6826 		kvm_rip_write(vcpu, ctxt->_eip);
6827 		if (ctxt->eflags & X86_EFLAGS_RF)
6828 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6829 		return 1;
6830 	}
6831 
6832 	if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
6833 		return 1;
6834 
6835 	/* this is needed for vmware backdoor interface to work since it
6836 	   changes registers values  during IO operation */
6837 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6838 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6839 		emulator_invalidate_register_cache(ctxt);
6840 	}
6841 
6842 restart:
6843 	if (emulation_type & EMULTYPE_PF) {
6844 		/* Save the faulting GPA (cr2) in the address field */
6845 		ctxt->exception.address = cr2_or_gpa;
6846 
6847 		/* With shadow page tables, cr2 contains a GVA or nGPA. */
6848 		if (vcpu->arch.mmu->direct_map) {
6849 			ctxt->gpa_available = true;
6850 			ctxt->gpa_val = cr2_or_gpa;
6851 		}
6852 	} else {
6853 		/* Sanitize the address out of an abundance of paranoia. */
6854 		ctxt->exception.address = 0;
6855 	}
6856 
6857 	r = x86_emulate_insn(ctxt);
6858 
6859 	if (r == EMULATION_INTERCEPTED)
6860 		return 1;
6861 
6862 	if (r == EMULATION_FAILED) {
6863 		if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
6864 					emulation_type))
6865 			return 1;
6866 
6867 		return handle_emulation_failure(vcpu, emulation_type);
6868 	}
6869 
6870 	if (ctxt->have_exception) {
6871 		r = 1;
6872 		if (inject_emulated_exception(vcpu))
6873 			return r;
6874 	} else if (vcpu->arch.pio.count) {
6875 		if (!vcpu->arch.pio.in) {
6876 			/* FIXME: return into emulator if single-stepping.  */
6877 			vcpu->arch.pio.count = 0;
6878 		} else {
6879 			writeback = false;
6880 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
6881 		}
6882 		r = 0;
6883 	} else if (vcpu->mmio_needed) {
6884 		++vcpu->stat.mmio_exits;
6885 
6886 		if (!vcpu->mmio_is_write)
6887 			writeback = false;
6888 		r = 0;
6889 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6890 	} else if (r == EMULATION_RESTART)
6891 		goto restart;
6892 	else
6893 		r = 1;
6894 
6895 	if (writeback) {
6896 		unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
6897 		toggle_interruptibility(vcpu, ctxt->interruptibility);
6898 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6899 		if (!ctxt->have_exception ||
6900 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
6901 			kvm_rip_write(vcpu, ctxt->eip);
6902 			if (r && ctxt->tf)
6903 				r = kvm_vcpu_do_singlestep(vcpu);
6904 			if (kvm_x86_ops.update_emulated_instruction)
6905 				kvm_x86_ops.update_emulated_instruction(vcpu);
6906 			__kvm_set_rflags(vcpu, ctxt->eflags);
6907 		}
6908 
6909 		/*
6910 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6911 		 * do nothing, and it will be requested again as soon as
6912 		 * the shadow expires.  But we still need to check here,
6913 		 * because POPF has no interrupt shadow.
6914 		 */
6915 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6916 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6917 	} else
6918 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6919 
6920 	return r;
6921 }
6922 
6923 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6924 {
6925 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6926 }
6927 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6928 
6929 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6930 					void *insn, int insn_len)
6931 {
6932 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6933 }
6934 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6935 
6936 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
6937 {
6938 	vcpu->arch.pio.count = 0;
6939 	return 1;
6940 }
6941 
6942 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
6943 {
6944 	vcpu->arch.pio.count = 0;
6945 
6946 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
6947 		return 1;
6948 
6949 	return kvm_skip_emulated_instruction(vcpu);
6950 }
6951 
6952 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6953 			    unsigned short port)
6954 {
6955 	unsigned long val = kvm_rax_read(vcpu);
6956 	int ret = emulator_pio_out(vcpu, size, port, &val, 1);
6957 
6958 	if (ret)
6959 		return ret;
6960 
6961 	/*
6962 	 * Workaround userspace that relies on old KVM behavior of %rip being
6963 	 * incremented prior to exiting to userspace to handle "OUT 0x7e".
6964 	 */
6965 	if (port == 0x7e &&
6966 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
6967 		vcpu->arch.complete_userspace_io =
6968 			complete_fast_pio_out_port_0x7e;
6969 		kvm_skip_emulated_instruction(vcpu);
6970 	} else {
6971 		vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6972 		vcpu->arch.complete_userspace_io = complete_fast_pio_out;
6973 	}
6974 	return 0;
6975 }
6976 
6977 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6978 {
6979 	unsigned long val;
6980 
6981 	/* We should only ever be called with arch.pio.count equal to 1 */
6982 	BUG_ON(vcpu->arch.pio.count != 1);
6983 
6984 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
6985 		vcpu->arch.pio.count = 0;
6986 		return 1;
6987 	}
6988 
6989 	/* For size less than 4 we merge, else we zero extend */
6990 	val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
6991 
6992 	/*
6993 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
6994 	 * the copy and tracing
6995 	 */
6996 	emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
6997 	kvm_rax_write(vcpu, val);
6998 
6999 	return kvm_skip_emulated_instruction(vcpu);
7000 }
7001 
7002 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7003 			   unsigned short port)
7004 {
7005 	unsigned long val;
7006 	int ret;
7007 
7008 	/* For size less than 4 we merge, else we zero extend */
7009 	val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7010 
7011 	ret = emulator_pio_in(vcpu, size, port, &val, 1);
7012 	if (ret) {
7013 		kvm_rax_write(vcpu, val);
7014 		return ret;
7015 	}
7016 
7017 	vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7018 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7019 
7020 	return 0;
7021 }
7022 
7023 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7024 {
7025 	int ret;
7026 
7027 	if (in)
7028 		ret = kvm_fast_pio_in(vcpu, size, port);
7029 	else
7030 		ret = kvm_fast_pio_out(vcpu, size, port);
7031 	return ret && kvm_skip_emulated_instruction(vcpu);
7032 }
7033 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7034 
7035 static int kvmclock_cpu_down_prep(unsigned int cpu)
7036 {
7037 	__this_cpu_write(cpu_tsc_khz, 0);
7038 	return 0;
7039 }
7040 
7041 static void tsc_khz_changed(void *data)
7042 {
7043 	struct cpufreq_freqs *freq = data;
7044 	unsigned long khz = 0;
7045 
7046 	if (data)
7047 		khz = freq->new;
7048 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7049 		khz = cpufreq_quick_get(raw_smp_processor_id());
7050 	if (!khz)
7051 		khz = tsc_khz;
7052 	__this_cpu_write(cpu_tsc_khz, khz);
7053 }
7054 
7055 #ifdef CONFIG_X86_64
7056 static void kvm_hyperv_tsc_notifier(void)
7057 {
7058 	struct kvm *kvm;
7059 	struct kvm_vcpu *vcpu;
7060 	int cpu;
7061 
7062 	mutex_lock(&kvm_lock);
7063 	list_for_each_entry(kvm, &vm_list, vm_list)
7064 		kvm_make_mclock_inprogress_request(kvm);
7065 
7066 	hyperv_stop_tsc_emulation();
7067 
7068 	/* TSC frequency always matches when on Hyper-V */
7069 	for_each_present_cpu(cpu)
7070 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7071 	kvm_max_guest_tsc_khz = tsc_khz;
7072 
7073 	list_for_each_entry(kvm, &vm_list, vm_list) {
7074 		struct kvm_arch *ka = &kvm->arch;
7075 
7076 		spin_lock(&ka->pvclock_gtod_sync_lock);
7077 
7078 		pvclock_update_vm_gtod_copy(kvm);
7079 
7080 		kvm_for_each_vcpu(cpu, vcpu, kvm)
7081 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7082 
7083 		kvm_for_each_vcpu(cpu, vcpu, kvm)
7084 			kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7085 
7086 		spin_unlock(&ka->pvclock_gtod_sync_lock);
7087 	}
7088 	mutex_unlock(&kvm_lock);
7089 }
7090 #endif
7091 
7092 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7093 {
7094 	struct kvm *kvm;
7095 	struct kvm_vcpu *vcpu;
7096 	int i, send_ipi = 0;
7097 
7098 	/*
7099 	 * We allow guests to temporarily run on slowing clocks,
7100 	 * provided we notify them after, or to run on accelerating
7101 	 * clocks, provided we notify them before.  Thus time never
7102 	 * goes backwards.
7103 	 *
7104 	 * However, we have a problem.  We can't atomically update
7105 	 * the frequency of a given CPU from this function; it is
7106 	 * merely a notifier, which can be called from any CPU.
7107 	 * Changing the TSC frequency at arbitrary points in time
7108 	 * requires a recomputation of local variables related to
7109 	 * the TSC for each VCPU.  We must flag these local variables
7110 	 * to be updated and be sure the update takes place with the
7111 	 * new frequency before any guests proceed.
7112 	 *
7113 	 * Unfortunately, the combination of hotplug CPU and frequency
7114 	 * change creates an intractable locking scenario; the order
7115 	 * of when these callouts happen is undefined with respect to
7116 	 * CPU hotplug, and they can race with each other.  As such,
7117 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7118 	 * undefined; you can actually have a CPU frequency change take
7119 	 * place in between the computation of X and the setting of the
7120 	 * variable.  To protect against this problem, all updates of
7121 	 * the per_cpu tsc_khz variable are done in an interrupt
7122 	 * protected IPI, and all callers wishing to update the value
7123 	 * must wait for a synchronous IPI to complete (which is trivial
7124 	 * if the caller is on the CPU already).  This establishes the
7125 	 * necessary total order on variable updates.
7126 	 *
7127 	 * Note that because a guest time update may take place
7128 	 * anytime after the setting of the VCPU's request bit, the
7129 	 * correct TSC value must be set before the request.  However,
7130 	 * to ensure the update actually makes it to any guest which
7131 	 * starts running in hardware virtualization between the set
7132 	 * and the acquisition of the spinlock, we must also ping the
7133 	 * CPU after setting the request bit.
7134 	 *
7135 	 */
7136 
7137 	smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7138 
7139 	mutex_lock(&kvm_lock);
7140 	list_for_each_entry(kvm, &vm_list, vm_list) {
7141 		kvm_for_each_vcpu(i, vcpu, kvm) {
7142 			if (vcpu->cpu != cpu)
7143 				continue;
7144 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7145 			if (vcpu->cpu != raw_smp_processor_id())
7146 				send_ipi = 1;
7147 		}
7148 	}
7149 	mutex_unlock(&kvm_lock);
7150 
7151 	if (freq->old < freq->new && send_ipi) {
7152 		/*
7153 		 * We upscale the frequency.  Must make the guest
7154 		 * doesn't see old kvmclock values while running with
7155 		 * the new frequency, otherwise we risk the guest sees
7156 		 * time go backwards.
7157 		 *
7158 		 * In case we update the frequency for another cpu
7159 		 * (which might be in guest context) send an interrupt
7160 		 * to kick the cpu out of guest context.  Next time
7161 		 * guest context is entered kvmclock will be updated,
7162 		 * so the guest will not see stale values.
7163 		 */
7164 		smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7165 	}
7166 }
7167 
7168 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7169 				     void *data)
7170 {
7171 	struct cpufreq_freqs *freq = data;
7172 	int cpu;
7173 
7174 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7175 		return 0;
7176 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7177 		return 0;
7178 
7179 	for_each_cpu(cpu, freq->policy->cpus)
7180 		__kvmclock_cpufreq_notifier(freq, cpu);
7181 
7182 	return 0;
7183 }
7184 
7185 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7186 	.notifier_call  = kvmclock_cpufreq_notifier
7187 };
7188 
7189 static int kvmclock_cpu_online(unsigned int cpu)
7190 {
7191 	tsc_khz_changed(NULL);
7192 	return 0;
7193 }
7194 
7195 static void kvm_timer_init(void)
7196 {
7197 	max_tsc_khz = tsc_khz;
7198 
7199 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7200 #ifdef CONFIG_CPU_FREQ
7201 		struct cpufreq_policy *policy;
7202 		int cpu;
7203 
7204 		cpu = get_cpu();
7205 		policy = cpufreq_cpu_get(cpu);
7206 		if (policy) {
7207 			if (policy->cpuinfo.max_freq)
7208 				max_tsc_khz = policy->cpuinfo.max_freq;
7209 			cpufreq_cpu_put(policy);
7210 		}
7211 		put_cpu();
7212 #endif
7213 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7214 					  CPUFREQ_TRANSITION_NOTIFIER);
7215 	}
7216 
7217 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7218 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
7219 }
7220 
7221 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7222 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7223 
7224 int kvm_is_in_guest(void)
7225 {
7226 	return __this_cpu_read(current_vcpu) != NULL;
7227 }
7228 
7229 static int kvm_is_user_mode(void)
7230 {
7231 	int user_mode = 3;
7232 
7233 	if (__this_cpu_read(current_vcpu))
7234 		user_mode = kvm_x86_ops.get_cpl(__this_cpu_read(current_vcpu));
7235 
7236 	return user_mode != 0;
7237 }
7238 
7239 static unsigned long kvm_get_guest_ip(void)
7240 {
7241 	unsigned long ip = 0;
7242 
7243 	if (__this_cpu_read(current_vcpu))
7244 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7245 
7246 	return ip;
7247 }
7248 
7249 static void kvm_handle_intel_pt_intr(void)
7250 {
7251 	struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7252 
7253 	kvm_make_request(KVM_REQ_PMI, vcpu);
7254 	__set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7255 			(unsigned long *)&vcpu->arch.pmu.global_status);
7256 }
7257 
7258 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7259 	.is_in_guest		= kvm_is_in_guest,
7260 	.is_user_mode		= kvm_is_user_mode,
7261 	.get_guest_ip		= kvm_get_guest_ip,
7262 	.handle_intel_pt_intr	= kvm_handle_intel_pt_intr,
7263 };
7264 
7265 #ifdef CONFIG_X86_64
7266 static void pvclock_gtod_update_fn(struct work_struct *work)
7267 {
7268 	struct kvm *kvm;
7269 
7270 	struct kvm_vcpu *vcpu;
7271 	int i;
7272 
7273 	mutex_lock(&kvm_lock);
7274 	list_for_each_entry(kvm, &vm_list, vm_list)
7275 		kvm_for_each_vcpu(i, vcpu, kvm)
7276 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7277 	atomic_set(&kvm_guest_has_master_clock, 0);
7278 	mutex_unlock(&kvm_lock);
7279 }
7280 
7281 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7282 
7283 /*
7284  * Notification about pvclock gtod data update.
7285  */
7286 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7287 			       void *priv)
7288 {
7289 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7290 	struct timekeeper *tk = priv;
7291 
7292 	update_pvclock_gtod(tk);
7293 
7294 	/* disable master clock if host does not trust, or does not
7295 	 * use, TSC based clocksource.
7296 	 */
7297 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7298 	    atomic_read(&kvm_guest_has_master_clock) != 0)
7299 		queue_work(system_long_wq, &pvclock_gtod_work);
7300 
7301 	return 0;
7302 }
7303 
7304 static struct notifier_block pvclock_gtod_notifier = {
7305 	.notifier_call = pvclock_gtod_notify,
7306 };
7307 #endif
7308 
7309 int kvm_arch_init(void *opaque)
7310 {
7311 	struct kvm_x86_init_ops *ops = opaque;
7312 	int r;
7313 
7314 	if (kvm_x86_ops.hardware_enable) {
7315 		printk(KERN_ERR "kvm: already loaded the other module\n");
7316 		r = -EEXIST;
7317 		goto out;
7318 	}
7319 
7320 	if (!ops->cpu_has_kvm_support()) {
7321 		pr_err_ratelimited("kvm: no hardware support\n");
7322 		r = -EOPNOTSUPP;
7323 		goto out;
7324 	}
7325 	if (ops->disabled_by_bios()) {
7326 		pr_err_ratelimited("kvm: disabled by bios\n");
7327 		r = -EOPNOTSUPP;
7328 		goto out;
7329 	}
7330 
7331 	/*
7332 	 * KVM explicitly assumes that the guest has an FPU and
7333 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7334 	 * vCPU's FPU state as a fxregs_state struct.
7335 	 */
7336 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7337 		printk(KERN_ERR "kvm: inadequate fpu\n");
7338 		r = -EOPNOTSUPP;
7339 		goto out;
7340 	}
7341 
7342 	r = -ENOMEM;
7343 	x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7344 					  __alignof__(struct fpu), SLAB_ACCOUNT,
7345 					  NULL);
7346 	if (!x86_fpu_cache) {
7347 		printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7348 		goto out;
7349 	}
7350 
7351 	x86_emulator_cache = kvm_alloc_emulator_cache();
7352 	if (!x86_emulator_cache) {
7353 		pr_err("kvm: failed to allocate cache for x86 emulator\n");
7354 		goto out_free_x86_fpu_cache;
7355 	}
7356 
7357 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
7358 	if (!shared_msrs) {
7359 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
7360 		goto out_free_x86_emulator_cache;
7361 	}
7362 
7363 	r = kvm_mmu_module_init();
7364 	if (r)
7365 		goto out_free_percpu;
7366 
7367 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7368 			PT_DIRTY_MASK, PT64_NX_MASK, 0,
7369 			PT_PRESENT_MASK, 0, sme_me_mask);
7370 	kvm_timer_init();
7371 
7372 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
7373 
7374 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
7375 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7376 		supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
7377 	}
7378 
7379 	kvm_lapic_init();
7380 	if (pi_inject_timer == -1)
7381 		pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7382 #ifdef CONFIG_X86_64
7383 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7384 
7385 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7386 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7387 #endif
7388 
7389 	return 0;
7390 
7391 out_free_percpu:
7392 	free_percpu(shared_msrs);
7393 out_free_x86_emulator_cache:
7394 	kmem_cache_destroy(x86_emulator_cache);
7395 out_free_x86_fpu_cache:
7396 	kmem_cache_destroy(x86_fpu_cache);
7397 out:
7398 	return r;
7399 }
7400 
7401 void kvm_arch_exit(void)
7402 {
7403 #ifdef CONFIG_X86_64
7404 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7405 		clear_hv_tscchange_cb();
7406 #endif
7407 	kvm_lapic_exit();
7408 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7409 
7410 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7411 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7412 					    CPUFREQ_TRANSITION_NOTIFIER);
7413 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7414 #ifdef CONFIG_X86_64
7415 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7416 #endif
7417 	kvm_x86_ops.hardware_enable = NULL;
7418 	kvm_mmu_module_exit();
7419 	free_percpu(shared_msrs);
7420 	kmem_cache_destroy(x86_fpu_cache);
7421 }
7422 
7423 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7424 {
7425 	++vcpu->stat.halt_exits;
7426 	if (lapic_in_kernel(vcpu)) {
7427 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7428 		return 1;
7429 	} else {
7430 		vcpu->run->exit_reason = KVM_EXIT_HLT;
7431 		return 0;
7432 	}
7433 }
7434 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7435 
7436 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7437 {
7438 	int ret = kvm_skip_emulated_instruction(vcpu);
7439 	/*
7440 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7441 	 * KVM_EXIT_DEBUG here.
7442 	 */
7443 	return kvm_vcpu_halt(vcpu) && ret;
7444 }
7445 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7446 
7447 #ifdef CONFIG_X86_64
7448 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7449 			        unsigned long clock_type)
7450 {
7451 	struct kvm_clock_pairing clock_pairing;
7452 	struct timespec64 ts;
7453 	u64 cycle;
7454 	int ret;
7455 
7456 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7457 		return -KVM_EOPNOTSUPP;
7458 
7459 	if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7460 		return -KVM_EOPNOTSUPP;
7461 
7462 	clock_pairing.sec = ts.tv_sec;
7463 	clock_pairing.nsec = ts.tv_nsec;
7464 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7465 	clock_pairing.flags = 0;
7466 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7467 
7468 	ret = 0;
7469 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7470 			    sizeof(struct kvm_clock_pairing)))
7471 		ret = -KVM_EFAULT;
7472 
7473 	return ret;
7474 }
7475 #endif
7476 
7477 /*
7478  * kvm_pv_kick_cpu_op:  Kick a vcpu.
7479  *
7480  * @apicid - apicid of vcpu to be kicked.
7481  */
7482 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7483 {
7484 	struct kvm_lapic_irq lapic_irq;
7485 
7486 	lapic_irq.shorthand = APIC_DEST_NOSHORT;
7487 	lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
7488 	lapic_irq.level = 0;
7489 	lapic_irq.dest_id = apicid;
7490 	lapic_irq.msi_redir_hint = false;
7491 
7492 	lapic_irq.delivery_mode = APIC_DM_REMRD;
7493 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7494 }
7495 
7496 bool kvm_apicv_activated(struct kvm *kvm)
7497 {
7498 	return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
7499 }
7500 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
7501 
7502 void kvm_apicv_init(struct kvm *kvm, bool enable)
7503 {
7504 	if (enable)
7505 		clear_bit(APICV_INHIBIT_REASON_DISABLE,
7506 			  &kvm->arch.apicv_inhibit_reasons);
7507 	else
7508 		set_bit(APICV_INHIBIT_REASON_DISABLE,
7509 			&kvm->arch.apicv_inhibit_reasons);
7510 }
7511 EXPORT_SYMBOL_GPL(kvm_apicv_init);
7512 
7513 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
7514 {
7515 	struct kvm_vcpu *target = NULL;
7516 	struct kvm_apic_map *map;
7517 
7518 	rcu_read_lock();
7519 	map = rcu_dereference(kvm->arch.apic_map);
7520 
7521 	if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
7522 		target = map->phys_map[dest_id]->vcpu;
7523 
7524 	rcu_read_unlock();
7525 
7526 	if (target && READ_ONCE(target->ready))
7527 		kvm_vcpu_yield_to(target);
7528 }
7529 
7530 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7531 {
7532 	unsigned long nr, a0, a1, a2, a3, ret;
7533 	int op_64_bit;
7534 
7535 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
7536 		return kvm_hv_hypercall(vcpu);
7537 
7538 	nr = kvm_rax_read(vcpu);
7539 	a0 = kvm_rbx_read(vcpu);
7540 	a1 = kvm_rcx_read(vcpu);
7541 	a2 = kvm_rdx_read(vcpu);
7542 	a3 = kvm_rsi_read(vcpu);
7543 
7544 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
7545 
7546 	op_64_bit = is_64_bit_mode(vcpu);
7547 	if (!op_64_bit) {
7548 		nr &= 0xFFFFFFFF;
7549 		a0 &= 0xFFFFFFFF;
7550 		a1 &= 0xFFFFFFFF;
7551 		a2 &= 0xFFFFFFFF;
7552 		a3 &= 0xFFFFFFFF;
7553 	}
7554 
7555 	if (kvm_x86_ops.get_cpl(vcpu) != 0) {
7556 		ret = -KVM_EPERM;
7557 		goto out;
7558 	}
7559 
7560 	switch (nr) {
7561 	case KVM_HC_VAPIC_POLL_IRQ:
7562 		ret = 0;
7563 		break;
7564 	case KVM_HC_KICK_CPU:
7565 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7566 		kvm_sched_yield(vcpu->kvm, a1);
7567 		ret = 0;
7568 		break;
7569 #ifdef CONFIG_X86_64
7570 	case KVM_HC_CLOCK_PAIRING:
7571 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7572 		break;
7573 #endif
7574 	case KVM_HC_SEND_IPI:
7575 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7576 		break;
7577 	case KVM_HC_SCHED_YIELD:
7578 		kvm_sched_yield(vcpu->kvm, a0);
7579 		ret = 0;
7580 		break;
7581 	default:
7582 		ret = -KVM_ENOSYS;
7583 		break;
7584 	}
7585 out:
7586 	if (!op_64_bit)
7587 		ret = (u32)ret;
7588 	kvm_rax_write(vcpu, ret);
7589 
7590 	++vcpu->stat.hypercalls;
7591 	return kvm_skip_emulated_instruction(vcpu);
7592 }
7593 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7594 
7595 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7596 {
7597 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7598 	char instruction[3];
7599 	unsigned long rip = kvm_rip_read(vcpu);
7600 
7601 	kvm_x86_ops.patch_hypercall(vcpu, instruction);
7602 
7603 	return emulator_write_emulated(ctxt, rip, instruction, 3,
7604 		&ctxt->exception);
7605 }
7606 
7607 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7608 {
7609 	return vcpu->run->request_interrupt_window &&
7610 		likely(!pic_in_kernel(vcpu->kvm));
7611 }
7612 
7613 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7614 {
7615 	struct kvm_run *kvm_run = vcpu->run;
7616 
7617 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7618 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7619 	kvm_run->cr8 = kvm_get_cr8(vcpu);
7620 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
7621 	kvm_run->ready_for_interrupt_injection =
7622 		pic_in_kernel(vcpu->kvm) ||
7623 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
7624 }
7625 
7626 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7627 {
7628 	int max_irr, tpr;
7629 
7630 	if (!kvm_x86_ops.update_cr8_intercept)
7631 		return;
7632 
7633 	if (!lapic_in_kernel(vcpu))
7634 		return;
7635 
7636 	if (vcpu->arch.apicv_active)
7637 		return;
7638 
7639 	if (!vcpu->arch.apic->vapic_addr)
7640 		max_irr = kvm_lapic_find_highest_irr(vcpu);
7641 	else
7642 		max_irr = -1;
7643 
7644 	if (max_irr != -1)
7645 		max_irr >>= 4;
7646 
7647 	tpr = kvm_lapic_get_cr8(vcpu);
7648 
7649 	kvm_x86_ops.update_cr8_intercept(vcpu, tpr, max_irr);
7650 }
7651 
7652 static int inject_pending_event(struct kvm_vcpu *vcpu)
7653 {
7654 	int r;
7655 
7656 	/* try to reinject previous events if any */
7657 
7658 	if (vcpu->arch.exception.injected)
7659 		kvm_x86_ops.queue_exception(vcpu);
7660 	/*
7661 	 * Do not inject an NMI or interrupt if there is a pending
7662 	 * exception.  Exceptions and interrupts are recognized at
7663 	 * instruction boundaries, i.e. the start of an instruction.
7664 	 * Trap-like exceptions, e.g. #DB, have higher priority than
7665 	 * NMIs and interrupts, i.e. traps are recognized before an
7666 	 * NMI/interrupt that's pending on the same instruction.
7667 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7668 	 * priority, but are only generated (pended) during instruction
7669 	 * execution, i.e. a pending fault-like exception means the
7670 	 * fault occurred on the *previous* instruction and must be
7671 	 * serviced prior to recognizing any new events in order to
7672 	 * fully complete the previous instruction.
7673 	 */
7674 	else if (!vcpu->arch.exception.pending) {
7675 		if (vcpu->arch.nmi_injected)
7676 			kvm_x86_ops.set_nmi(vcpu);
7677 		else if (vcpu->arch.interrupt.injected)
7678 			kvm_x86_ops.set_irq(vcpu);
7679 	}
7680 
7681 	/*
7682 	 * Call check_nested_events() even if we reinjected a previous event
7683 	 * in order for caller to determine if it should require immediate-exit
7684 	 * from L2 to L1 due to pending L1 events which require exit
7685 	 * from L2 to L1.
7686 	 */
7687 	if (is_guest_mode(vcpu) && kvm_x86_ops.check_nested_events) {
7688 		r = kvm_x86_ops.check_nested_events(vcpu);
7689 		if (r != 0)
7690 			return r;
7691 	}
7692 
7693 	/* try to inject new event if pending */
7694 	if (vcpu->arch.exception.pending) {
7695 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
7696 					vcpu->arch.exception.has_error_code,
7697 					vcpu->arch.exception.error_code);
7698 
7699 		WARN_ON_ONCE(vcpu->arch.exception.injected);
7700 		vcpu->arch.exception.pending = false;
7701 		vcpu->arch.exception.injected = true;
7702 
7703 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7704 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7705 					     X86_EFLAGS_RF);
7706 
7707 		if (vcpu->arch.exception.nr == DB_VECTOR) {
7708 			/*
7709 			 * This code assumes that nSVM doesn't use
7710 			 * check_nested_events(). If it does, the
7711 			 * DR6/DR7 changes should happen before L1
7712 			 * gets a #VMEXIT for an intercepted #DB in
7713 			 * L2.  (Under VMX, on the other hand, the
7714 			 * DR6/DR7 changes should not happen in the
7715 			 * event of a VM-exit to L1 for an intercepted
7716 			 * #DB in L2.)
7717 			 */
7718 			kvm_deliver_exception_payload(vcpu);
7719 			if (vcpu->arch.dr7 & DR7_GD) {
7720 				vcpu->arch.dr7 &= ~DR7_GD;
7721 				kvm_update_dr7(vcpu);
7722 			}
7723 		}
7724 
7725 		kvm_x86_ops.queue_exception(vcpu);
7726 	}
7727 
7728 	/* Don't consider new event if we re-injected an event */
7729 	if (kvm_event_needs_reinjection(vcpu))
7730 		return 0;
7731 
7732 	if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7733 	    kvm_x86_ops.smi_allowed(vcpu)) {
7734 		vcpu->arch.smi_pending = false;
7735 		++vcpu->arch.smi_count;
7736 		enter_smm(vcpu);
7737 	} else if (vcpu->arch.nmi_pending && kvm_x86_ops.nmi_allowed(vcpu)) {
7738 		--vcpu->arch.nmi_pending;
7739 		vcpu->arch.nmi_injected = true;
7740 		kvm_x86_ops.set_nmi(vcpu);
7741 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
7742 		/*
7743 		 * Because interrupts can be injected asynchronously, we are
7744 		 * calling check_nested_events again here to avoid a race condition.
7745 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7746 		 * proposal and current concerns.  Perhaps we should be setting
7747 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
7748 		 */
7749 		if (is_guest_mode(vcpu) && kvm_x86_ops.check_nested_events) {
7750 			r = kvm_x86_ops.check_nested_events(vcpu);
7751 			if (r != 0)
7752 				return r;
7753 		}
7754 		if (kvm_x86_ops.interrupt_allowed(vcpu)) {
7755 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7756 					    false);
7757 			kvm_x86_ops.set_irq(vcpu);
7758 		}
7759 	}
7760 
7761 	return 0;
7762 }
7763 
7764 static void process_nmi(struct kvm_vcpu *vcpu)
7765 {
7766 	unsigned limit = 2;
7767 
7768 	/*
7769 	 * x86 is limited to one NMI running, and one NMI pending after it.
7770 	 * If an NMI is already in progress, limit further NMIs to just one.
7771 	 * Otherwise, allow two (and we'll inject the first one immediately).
7772 	 */
7773 	if (kvm_x86_ops.get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7774 		limit = 1;
7775 
7776 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7777 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7778 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7779 }
7780 
7781 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7782 {
7783 	u32 flags = 0;
7784 	flags |= seg->g       << 23;
7785 	flags |= seg->db      << 22;
7786 	flags |= seg->l       << 21;
7787 	flags |= seg->avl     << 20;
7788 	flags |= seg->present << 15;
7789 	flags |= seg->dpl     << 13;
7790 	flags |= seg->s       << 12;
7791 	flags |= seg->type    << 8;
7792 	return flags;
7793 }
7794 
7795 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7796 {
7797 	struct kvm_segment seg;
7798 	int offset;
7799 
7800 	kvm_get_segment(vcpu, &seg, n);
7801 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7802 
7803 	if (n < 3)
7804 		offset = 0x7f84 + n * 12;
7805 	else
7806 		offset = 0x7f2c + (n - 3) * 12;
7807 
7808 	put_smstate(u32, buf, offset + 8, seg.base);
7809 	put_smstate(u32, buf, offset + 4, seg.limit);
7810 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7811 }
7812 
7813 #ifdef CONFIG_X86_64
7814 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7815 {
7816 	struct kvm_segment seg;
7817 	int offset;
7818 	u16 flags;
7819 
7820 	kvm_get_segment(vcpu, &seg, n);
7821 	offset = 0x7e00 + n * 16;
7822 
7823 	flags = enter_smm_get_segment_flags(&seg) >> 8;
7824 	put_smstate(u16, buf, offset, seg.selector);
7825 	put_smstate(u16, buf, offset + 2, flags);
7826 	put_smstate(u32, buf, offset + 4, seg.limit);
7827 	put_smstate(u64, buf, offset + 8, seg.base);
7828 }
7829 #endif
7830 
7831 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7832 {
7833 	struct desc_ptr dt;
7834 	struct kvm_segment seg;
7835 	unsigned long val;
7836 	int i;
7837 
7838 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7839 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7840 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7841 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7842 
7843 	for (i = 0; i < 8; i++)
7844 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7845 
7846 	kvm_get_dr(vcpu, 6, &val);
7847 	put_smstate(u32, buf, 0x7fcc, (u32)val);
7848 	kvm_get_dr(vcpu, 7, &val);
7849 	put_smstate(u32, buf, 0x7fc8, (u32)val);
7850 
7851 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7852 	put_smstate(u32, buf, 0x7fc4, seg.selector);
7853 	put_smstate(u32, buf, 0x7f64, seg.base);
7854 	put_smstate(u32, buf, 0x7f60, seg.limit);
7855 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7856 
7857 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7858 	put_smstate(u32, buf, 0x7fc0, seg.selector);
7859 	put_smstate(u32, buf, 0x7f80, seg.base);
7860 	put_smstate(u32, buf, 0x7f7c, seg.limit);
7861 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7862 
7863 	kvm_x86_ops.get_gdt(vcpu, &dt);
7864 	put_smstate(u32, buf, 0x7f74, dt.address);
7865 	put_smstate(u32, buf, 0x7f70, dt.size);
7866 
7867 	kvm_x86_ops.get_idt(vcpu, &dt);
7868 	put_smstate(u32, buf, 0x7f58, dt.address);
7869 	put_smstate(u32, buf, 0x7f54, dt.size);
7870 
7871 	for (i = 0; i < 6; i++)
7872 		enter_smm_save_seg_32(vcpu, buf, i);
7873 
7874 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7875 
7876 	/* revision id */
7877 	put_smstate(u32, buf, 0x7efc, 0x00020000);
7878 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7879 }
7880 
7881 #ifdef CONFIG_X86_64
7882 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7883 {
7884 	struct desc_ptr dt;
7885 	struct kvm_segment seg;
7886 	unsigned long val;
7887 	int i;
7888 
7889 	for (i = 0; i < 16; i++)
7890 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7891 
7892 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7893 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7894 
7895 	kvm_get_dr(vcpu, 6, &val);
7896 	put_smstate(u64, buf, 0x7f68, val);
7897 	kvm_get_dr(vcpu, 7, &val);
7898 	put_smstate(u64, buf, 0x7f60, val);
7899 
7900 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7901 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7902 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7903 
7904 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7905 
7906 	/* revision id */
7907 	put_smstate(u32, buf, 0x7efc, 0x00020064);
7908 
7909 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7910 
7911 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7912 	put_smstate(u16, buf, 0x7e90, seg.selector);
7913 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7914 	put_smstate(u32, buf, 0x7e94, seg.limit);
7915 	put_smstate(u64, buf, 0x7e98, seg.base);
7916 
7917 	kvm_x86_ops.get_idt(vcpu, &dt);
7918 	put_smstate(u32, buf, 0x7e84, dt.size);
7919 	put_smstate(u64, buf, 0x7e88, dt.address);
7920 
7921 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7922 	put_smstate(u16, buf, 0x7e70, seg.selector);
7923 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7924 	put_smstate(u32, buf, 0x7e74, seg.limit);
7925 	put_smstate(u64, buf, 0x7e78, seg.base);
7926 
7927 	kvm_x86_ops.get_gdt(vcpu, &dt);
7928 	put_smstate(u32, buf, 0x7e64, dt.size);
7929 	put_smstate(u64, buf, 0x7e68, dt.address);
7930 
7931 	for (i = 0; i < 6; i++)
7932 		enter_smm_save_seg_64(vcpu, buf, i);
7933 }
7934 #endif
7935 
7936 static void enter_smm(struct kvm_vcpu *vcpu)
7937 {
7938 	struct kvm_segment cs, ds;
7939 	struct desc_ptr dt;
7940 	char buf[512];
7941 	u32 cr0;
7942 
7943 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7944 	memset(buf, 0, 512);
7945 #ifdef CONFIG_X86_64
7946 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7947 		enter_smm_save_state_64(vcpu, buf);
7948 	else
7949 #endif
7950 		enter_smm_save_state_32(vcpu, buf);
7951 
7952 	/*
7953 	 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7954 	 * vCPU state (e.g. leave guest mode) after we've saved the state into
7955 	 * the SMM state-save area.
7956 	 */
7957 	kvm_x86_ops.pre_enter_smm(vcpu, buf);
7958 
7959 	vcpu->arch.hflags |= HF_SMM_MASK;
7960 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7961 
7962 	if (kvm_x86_ops.get_nmi_mask(vcpu))
7963 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7964 	else
7965 		kvm_x86_ops.set_nmi_mask(vcpu, true);
7966 
7967 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7968 	kvm_rip_write(vcpu, 0x8000);
7969 
7970 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7971 	kvm_x86_ops.set_cr0(vcpu, cr0);
7972 	vcpu->arch.cr0 = cr0;
7973 
7974 	kvm_x86_ops.set_cr4(vcpu, 0);
7975 
7976 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
7977 	dt.address = dt.size = 0;
7978 	kvm_x86_ops.set_idt(vcpu, &dt);
7979 
7980 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7981 
7982 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7983 	cs.base = vcpu->arch.smbase;
7984 
7985 	ds.selector = 0;
7986 	ds.base = 0;
7987 
7988 	cs.limit    = ds.limit = 0xffffffff;
7989 	cs.type     = ds.type = 0x3;
7990 	cs.dpl      = ds.dpl = 0;
7991 	cs.db       = ds.db = 0;
7992 	cs.s        = ds.s = 1;
7993 	cs.l        = ds.l = 0;
7994 	cs.g        = ds.g = 1;
7995 	cs.avl      = ds.avl = 0;
7996 	cs.present  = ds.present = 1;
7997 	cs.unusable = ds.unusable = 0;
7998 	cs.padding  = ds.padding = 0;
7999 
8000 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8001 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
8002 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
8003 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
8004 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
8005 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
8006 
8007 #ifdef CONFIG_X86_64
8008 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8009 		kvm_x86_ops.set_efer(vcpu, 0);
8010 #endif
8011 
8012 	kvm_update_cpuid(vcpu);
8013 	kvm_mmu_reset_context(vcpu);
8014 }
8015 
8016 static void process_smi(struct kvm_vcpu *vcpu)
8017 {
8018 	vcpu->arch.smi_pending = true;
8019 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8020 }
8021 
8022 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
8023 				       unsigned long *vcpu_bitmap)
8024 {
8025 	cpumask_var_t cpus;
8026 
8027 	zalloc_cpumask_var(&cpus, GFP_ATOMIC);
8028 
8029 	kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
8030 				    vcpu_bitmap, cpus);
8031 
8032 	free_cpumask_var(cpus);
8033 }
8034 
8035 void kvm_make_scan_ioapic_request(struct kvm *kvm)
8036 {
8037 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
8038 }
8039 
8040 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
8041 {
8042 	if (!lapic_in_kernel(vcpu))
8043 		return;
8044 
8045 	vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
8046 	kvm_apic_update_apicv(vcpu);
8047 	kvm_x86_ops.refresh_apicv_exec_ctrl(vcpu);
8048 }
8049 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
8050 
8051 /*
8052  * NOTE: Do not hold any lock prior to calling this.
8053  *
8054  * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
8055  * locked, because it calls __x86_set_memory_region() which does
8056  * synchronize_srcu(&kvm->srcu).
8057  */
8058 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
8059 {
8060 	unsigned long old, new, expected;
8061 
8062 	if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
8063 	    !kvm_x86_ops.check_apicv_inhibit_reasons(bit))
8064 		return;
8065 
8066 	old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
8067 	do {
8068 		expected = new = old;
8069 		if (activate)
8070 			__clear_bit(bit, &new);
8071 		else
8072 			__set_bit(bit, &new);
8073 		if (new == old)
8074 			break;
8075 		old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
8076 	} while (old != expected);
8077 
8078 	if (!!old == !!new)
8079 		return;
8080 
8081 	trace_kvm_apicv_update_request(activate, bit);
8082 	if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
8083 		kvm_x86_ops.pre_update_apicv_exec_ctrl(kvm, activate);
8084 	kvm_make_all_cpus_request(kvm, KVM_REQ_APICV_UPDATE);
8085 }
8086 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
8087 
8088 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
8089 {
8090 	if (!kvm_apic_present(vcpu))
8091 		return;
8092 
8093 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
8094 
8095 	if (irqchip_split(vcpu->kvm))
8096 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
8097 	else {
8098 		if (vcpu->arch.apicv_active)
8099 			kvm_x86_ops.sync_pir_to_irr(vcpu);
8100 		if (ioapic_in_kernel(vcpu->kvm))
8101 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
8102 	}
8103 
8104 	if (is_guest_mode(vcpu))
8105 		vcpu->arch.load_eoi_exitmap_pending = true;
8106 	else
8107 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
8108 }
8109 
8110 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8111 {
8112 	u64 eoi_exit_bitmap[4];
8113 
8114 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
8115 		return;
8116 
8117 	bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
8118 		  vcpu_to_synic(vcpu)->vec_bitmap, 256);
8119 	kvm_x86_ops.load_eoi_exitmap(vcpu, eoi_exit_bitmap);
8120 }
8121 
8122 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
8123 		unsigned long start, unsigned long end,
8124 		bool blockable)
8125 {
8126 	unsigned long apic_address;
8127 
8128 	/*
8129 	 * The physical address of apic access page is stored in the VMCS.
8130 	 * Update it when it becomes invalid.
8131 	 */
8132 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8133 	if (start <= apic_address && apic_address < end)
8134 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
8135 
8136 	return 0;
8137 }
8138 
8139 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
8140 {
8141 	struct page *page = NULL;
8142 
8143 	if (!lapic_in_kernel(vcpu))
8144 		return;
8145 
8146 	if (!kvm_x86_ops.set_apic_access_page_addr)
8147 		return;
8148 
8149 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8150 	if (is_error_page(page))
8151 		return;
8152 	kvm_x86_ops.set_apic_access_page_addr(vcpu, page_to_phys(page));
8153 
8154 	/*
8155 	 * Do not pin apic access page in memory, the MMU notifier
8156 	 * will call us again if it is migrated or swapped out.
8157 	 */
8158 	put_page(page);
8159 }
8160 
8161 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
8162 {
8163 	smp_send_reschedule(vcpu->cpu);
8164 }
8165 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
8166 
8167 /*
8168  * Returns 1 to let vcpu_run() continue the guest execution loop without
8169  * exiting to the userspace.  Otherwise, the value will be returned to the
8170  * userspace.
8171  */
8172 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
8173 {
8174 	int r;
8175 	bool req_int_win =
8176 		dm_request_for_irq_injection(vcpu) &&
8177 		kvm_cpu_accept_dm_intr(vcpu);
8178 	enum exit_fastpath_completion exit_fastpath = EXIT_FASTPATH_NONE;
8179 
8180 	bool req_immediate_exit = false;
8181 
8182 	if (kvm_request_pending(vcpu)) {
8183 		if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu)) {
8184 			if (unlikely(!kvm_x86_ops.get_vmcs12_pages(vcpu))) {
8185 				r = 0;
8186 				goto out;
8187 			}
8188 		}
8189 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
8190 			kvm_mmu_unload(vcpu);
8191 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
8192 			__kvm_migrate_timers(vcpu);
8193 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
8194 			kvm_gen_update_masterclock(vcpu->kvm);
8195 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
8196 			kvm_gen_kvmclock_update(vcpu);
8197 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
8198 			r = kvm_guest_time_update(vcpu);
8199 			if (unlikely(r))
8200 				goto out;
8201 		}
8202 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
8203 			kvm_mmu_sync_roots(vcpu);
8204 		if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
8205 			kvm_mmu_load_pgd(vcpu);
8206 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
8207 			kvm_vcpu_flush_tlb(vcpu, true);
8208 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
8209 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
8210 			r = 0;
8211 			goto out;
8212 		}
8213 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8214 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
8215 			vcpu->mmio_needed = 0;
8216 			r = 0;
8217 			goto out;
8218 		}
8219 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
8220 			/* Page is swapped out. Do synthetic halt */
8221 			vcpu->arch.apf.halted = true;
8222 			r = 1;
8223 			goto out;
8224 		}
8225 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8226 			record_steal_time(vcpu);
8227 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
8228 			process_smi(vcpu);
8229 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
8230 			process_nmi(vcpu);
8231 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
8232 			kvm_pmu_handle_event(vcpu);
8233 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
8234 			kvm_pmu_deliver_pmi(vcpu);
8235 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8236 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8237 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
8238 				     vcpu->arch.ioapic_handled_vectors)) {
8239 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8240 				vcpu->run->eoi.vector =
8241 						vcpu->arch.pending_ioapic_eoi;
8242 				r = 0;
8243 				goto out;
8244 			}
8245 		}
8246 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8247 			vcpu_scan_ioapic(vcpu);
8248 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8249 			vcpu_load_eoi_exitmap(vcpu);
8250 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8251 			kvm_vcpu_reload_apic_access_page(vcpu);
8252 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8253 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8254 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8255 			r = 0;
8256 			goto out;
8257 		}
8258 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8259 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8260 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8261 			r = 0;
8262 			goto out;
8263 		}
8264 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8265 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8266 			vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8267 			r = 0;
8268 			goto out;
8269 		}
8270 
8271 		/*
8272 		 * KVM_REQ_HV_STIMER has to be processed after
8273 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8274 		 * depend on the guest clock being up-to-date
8275 		 */
8276 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8277 			kvm_hv_process_stimers(vcpu);
8278 		if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
8279 			kvm_vcpu_update_apicv(vcpu);
8280 	}
8281 
8282 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8283 		++vcpu->stat.req_event;
8284 		kvm_apic_accept_events(vcpu);
8285 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8286 			r = 1;
8287 			goto out;
8288 		}
8289 
8290 		if (inject_pending_event(vcpu) != 0)
8291 			req_immediate_exit = true;
8292 		else {
8293 			/* Enable SMI/NMI/IRQ window open exits if needed.
8294 			 *
8295 			 * SMIs have three cases:
8296 			 * 1) They can be nested, and then there is nothing to
8297 			 *    do here because RSM will cause a vmexit anyway.
8298 			 * 2) There is an ISA-specific reason why SMI cannot be
8299 			 *    injected, and the moment when this changes can be
8300 			 *    intercepted.
8301 			 * 3) Or the SMI can be pending because
8302 			 *    inject_pending_event has completed the injection
8303 			 *    of an IRQ or NMI from the previous vmexit, and
8304 			 *    then we request an immediate exit to inject the
8305 			 *    SMI.
8306 			 */
8307 			if (vcpu->arch.smi_pending && !is_smm(vcpu))
8308 				if (!kvm_x86_ops.enable_smi_window(vcpu))
8309 					req_immediate_exit = true;
8310 			if (vcpu->arch.nmi_pending)
8311 				kvm_x86_ops.enable_nmi_window(vcpu);
8312 			if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
8313 				kvm_x86_ops.enable_irq_window(vcpu);
8314 			WARN_ON(vcpu->arch.exception.pending);
8315 		}
8316 
8317 		if (kvm_lapic_enabled(vcpu)) {
8318 			update_cr8_intercept(vcpu);
8319 			kvm_lapic_sync_to_vapic(vcpu);
8320 		}
8321 	}
8322 
8323 	r = kvm_mmu_reload(vcpu);
8324 	if (unlikely(r)) {
8325 		goto cancel_injection;
8326 	}
8327 
8328 	preempt_disable();
8329 
8330 	kvm_x86_ops.prepare_guest_switch(vcpu);
8331 
8332 	/*
8333 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
8334 	 * IPI are then delayed after guest entry, which ensures that they
8335 	 * result in virtual interrupt delivery.
8336 	 */
8337 	local_irq_disable();
8338 	vcpu->mode = IN_GUEST_MODE;
8339 
8340 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8341 
8342 	/*
8343 	 * 1) We should set ->mode before checking ->requests.  Please see
8344 	 * the comment in kvm_vcpu_exiting_guest_mode().
8345 	 *
8346 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
8347 	 * pairs with the memory barrier implicit in pi_test_and_set_on
8348 	 * (see vmx_deliver_posted_interrupt).
8349 	 *
8350 	 * 3) This also orders the write to mode from any reads to the page
8351 	 * tables done while the VCPU is running.  Please see the comment
8352 	 * in kvm_flush_remote_tlbs.
8353 	 */
8354 	smp_mb__after_srcu_read_unlock();
8355 
8356 	/*
8357 	 * This handles the case where a posted interrupt was
8358 	 * notified with kvm_vcpu_kick.
8359 	 */
8360 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8361 		kvm_x86_ops.sync_pir_to_irr(vcpu);
8362 
8363 	if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
8364 	    || need_resched() || signal_pending(current)) {
8365 		vcpu->mode = OUTSIDE_GUEST_MODE;
8366 		smp_wmb();
8367 		local_irq_enable();
8368 		preempt_enable();
8369 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8370 		r = 1;
8371 		goto cancel_injection;
8372 	}
8373 
8374 	if (req_immediate_exit) {
8375 		kvm_make_request(KVM_REQ_EVENT, vcpu);
8376 		kvm_x86_ops.request_immediate_exit(vcpu);
8377 	}
8378 
8379 	trace_kvm_entry(vcpu->vcpu_id);
8380 	guest_enter_irqoff();
8381 
8382 	fpregs_assert_state_consistent();
8383 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
8384 		switch_fpu_return();
8385 
8386 	if (unlikely(vcpu->arch.switch_db_regs)) {
8387 		set_debugreg(0, 7);
8388 		set_debugreg(vcpu->arch.eff_db[0], 0);
8389 		set_debugreg(vcpu->arch.eff_db[1], 1);
8390 		set_debugreg(vcpu->arch.eff_db[2], 2);
8391 		set_debugreg(vcpu->arch.eff_db[3], 3);
8392 		set_debugreg(vcpu->arch.dr6, 6);
8393 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8394 	}
8395 
8396 	kvm_x86_ops.run(vcpu);
8397 
8398 	/*
8399 	 * Do this here before restoring debug registers on the host.  And
8400 	 * since we do this before handling the vmexit, a DR access vmexit
8401 	 * can (a) read the correct value of the debug registers, (b) set
8402 	 * KVM_DEBUGREG_WONT_EXIT again.
8403 	 */
8404 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8405 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8406 		kvm_x86_ops.sync_dirty_debug_regs(vcpu);
8407 		kvm_update_dr0123(vcpu);
8408 		kvm_update_dr6(vcpu);
8409 		kvm_update_dr7(vcpu);
8410 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8411 	}
8412 
8413 	/*
8414 	 * If the guest has used debug registers, at least dr7
8415 	 * will be disabled while returning to the host.
8416 	 * If we don't have active breakpoints in the host, we don't
8417 	 * care about the messed up debug address registers. But if
8418 	 * we have some of them active, restore the old state.
8419 	 */
8420 	if (hw_breakpoint_active())
8421 		hw_breakpoint_restore();
8422 
8423 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8424 
8425 	vcpu->mode = OUTSIDE_GUEST_MODE;
8426 	smp_wmb();
8427 
8428 	kvm_x86_ops.handle_exit_irqoff(vcpu, &exit_fastpath);
8429 
8430 	/*
8431 	 * Consume any pending interrupts, including the possible source of
8432 	 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8433 	 * An instruction is required after local_irq_enable() to fully unblock
8434 	 * interrupts on processors that implement an interrupt shadow, the
8435 	 * stat.exits increment will do nicely.
8436 	 */
8437 	kvm_before_interrupt(vcpu);
8438 	local_irq_enable();
8439 	++vcpu->stat.exits;
8440 	local_irq_disable();
8441 	kvm_after_interrupt(vcpu);
8442 
8443 	guest_exit_irqoff();
8444 	if (lapic_in_kernel(vcpu)) {
8445 		s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8446 		if (delta != S64_MIN) {
8447 			trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8448 			vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8449 		}
8450 	}
8451 
8452 	local_irq_enable();
8453 	preempt_enable();
8454 
8455 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8456 
8457 	/*
8458 	 * Profile KVM exit RIPs:
8459 	 */
8460 	if (unlikely(prof_on == KVM_PROFILING)) {
8461 		unsigned long rip = kvm_rip_read(vcpu);
8462 		profile_hit(KVM_PROFILING, (void *)rip);
8463 	}
8464 
8465 	if (unlikely(vcpu->arch.tsc_always_catchup))
8466 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8467 
8468 	if (vcpu->arch.apic_attention)
8469 		kvm_lapic_sync_from_vapic(vcpu);
8470 
8471 	r = kvm_x86_ops.handle_exit(vcpu, exit_fastpath);
8472 	return r;
8473 
8474 cancel_injection:
8475 	kvm_x86_ops.cancel_injection(vcpu);
8476 	if (unlikely(vcpu->arch.apic_attention))
8477 		kvm_lapic_sync_from_vapic(vcpu);
8478 out:
8479 	return r;
8480 }
8481 
8482 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8483 {
8484 	if (!kvm_arch_vcpu_runnable(vcpu) &&
8485 	    (!kvm_x86_ops.pre_block || kvm_x86_ops.pre_block(vcpu) == 0)) {
8486 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8487 		kvm_vcpu_block(vcpu);
8488 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8489 
8490 		if (kvm_x86_ops.post_block)
8491 			kvm_x86_ops.post_block(vcpu);
8492 
8493 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8494 			return 1;
8495 	}
8496 
8497 	kvm_apic_accept_events(vcpu);
8498 	switch(vcpu->arch.mp_state) {
8499 	case KVM_MP_STATE_HALTED:
8500 		vcpu->arch.pv.pv_unhalted = false;
8501 		vcpu->arch.mp_state =
8502 			KVM_MP_STATE_RUNNABLE;
8503 		/* fall through */
8504 	case KVM_MP_STATE_RUNNABLE:
8505 		vcpu->arch.apf.halted = false;
8506 		break;
8507 	case KVM_MP_STATE_INIT_RECEIVED:
8508 		break;
8509 	default:
8510 		return -EINTR;
8511 	}
8512 	return 1;
8513 }
8514 
8515 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8516 {
8517 	if (is_guest_mode(vcpu) && kvm_x86_ops.check_nested_events)
8518 		kvm_x86_ops.check_nested_events(vcpu);
8519 
8520 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8521 		!vcpu->arch.apf.halted);
8522 }
8523 
8524 static int vcpu_run(struct kvm_vcpu *vcpu)
8525 {
8526 	int r;
8527 	struct kvm *kvm = vcpu->kvm;
8528 
8529 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8530 	vcpu->arch.l1tf_flush_l1d = true;
8531 
8532 	for (;;) {
8533 		if (kvm_vcpu_running(vcpu)) {
8534 			r = vcpu_enter_guest(vcpu);
8535 		} else {
8536 			r = vcpu_block(kvm, vcpu);
8537 		}
8538 
8539 		if (r <= 0)
8540 			break;
8541 
8542 		kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8543 		if (kvm_cpu_has_pending_timer(vcpu))
8544 			kvm_inject_pending_timer_irqs(vcpu);
8545 
8546 		if (dm_request_for_irq_injection(vcpu) &&
8547 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8548 			r = 0;
8549 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8550 			++vcpu->stat.request_irq_exits;
8551 			break;
8552 		}
8553 
8554 		kvm_check_async_pf_completion(vcpu);
8555 
8556 		if (signal_pending(current)) {
8557 			r = -EINTR;
8558 			vcpu->run->exit_reason = KVM_EXIT_INTR;
8559 			++vcpu->stat.signal_exits;
8560 			break;
8561 		}
8562 		if (need_resched()) {
8563 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8564 			cond_resched();
8565 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8566 		}
8567 	}
8568 
8569 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8570 
8571 	return r;
8572 }
8573 
8574 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8575 {
8576 	int r;
8577 
8578 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8579 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8580 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8581 	return r;
8582 }
8583 
8584 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8585 {
8586 	BUG_ON(!vcpu->arch.pio.count);
8587 
8588 	return complete_emulated_io(vcpu);
8589 }
8590 
8591 /*
8592  * Implements the following, as a state machine:
8593  *
8594  * read:
8595  *   for each fragment
8596  *     for each mmio piece in the fragment
8597  *       write gpa, len
8598  *       exit
8599  *       copy data
8600  *   execute insn
8601  *
8602  * write:
8603  *   for each fragment
8604  *     for each mmio piece in the fragment
8605  *       write gpa, len
8606  *       copy data
8607  *       exit
8608  */
8609 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8610 {
8611 	struct kvm_run *run = vcpu->run;
8612 	struct kvm_mmio_fragment *frag;
8613 	unsigned len;
8614 
8615 	BUG_ON(!vcpu->mmio_needed);
8616 
8617 	/* Complete previous fragment */
8618 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8619 	len = min(8u, frag->len);
8620 	if (!vcpu->mmio_is_write)
8621 		memcpy(frag->data, run->mmio.data, len);
8622 
8623 	if (frag->len <= 8) {
8624 		/* Switch to the next fragment. */
8625 		frag++;
8626 		vcpu->mmio_cur_fragment++;
8627 	} else {
8628 		/* Go forward to the next mmio piece. */
8629 		frag->data += len;
8630 		frag->gpa += len;
8631 		frag->len -= len;
8632 	}
8633 
8634 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8635 		vcpu->mmio_needed = 0;
8636 
8637 		/* FIXME: return into emulator if single-stepping.  */
8638 		if (vcpu->mmio_is_write)
8639 			return 1;
8640 		vcpu->mmio_read_completed = 1;
8641 		return complete_emulated_io(vcpu);
8642 	}
8643 
8644 	run->exit_reason = KVM_EXIT_MMIO;
8645 	run->mmio.phys_addr = frag->gpa;
8646 	if (vcpu->mmio_is_write)
8647 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8648 	run->mmio.len = min(8u, frag->len);
8649 	run->mmio.is_write = vcpu->mmio_is_write;
8650 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8651 	return 0;
8652 }
8653 
8654 static void kvm_save_current_fpu(struct fpu *fpu)
8655 {
8656 	/*
8657 	 * If the target FPU state is not resident in the CPU registers, just
8658 	 * memcpy() from current, else save CPU state directly to the target.
8659 	 */
8660 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
8661 		memcpy(&fpu->state, &current->thread.fpu.state,
8662 		       fpu_kernel_xstate_size);
8663 	else
8664 		copy_fpregs_to_fpstate(fpu);
8665 }
8666 
8667 /* Swap (qemu) user FPU context for the guest FPU context. */
8668 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8669 {
8670 	fpregs_lock();
8671 
8672 	kvm_save_current_fpu(vcpu->arch.user_fpu);
8673 
8674 	/* PKRU is separately restored in kvm_x86_ops.run.  */
8675 	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8676 				~XFEATURE_MASK_PKRU);
8677 
8678 	fpregs_mark_activate();
8679 	fpregs_unlock();
8680 
8681 	trace_kvm_fpu(1);
8682 }
8683 
8684 /* When vcpu_run ends, restore user space FPU context. */
8685 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8686 {
8687 	fpregs_lock();
8688 
8689 	kvm_save_current_fpu(vcpu->arch.guest_fpu);
8690 
8691 	copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
8692 
8693 	fpregs_mark_activate();
8694 	fpregs_unlock();
8695 
8696 	++vcpu->stat.fpu_reload;
8697 	trace_kvm_fpu(0);
8698 }
8699 
8700 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8701 {
8702 	int r;
8703 
8704 	vcpu_load(vcpu);
8705 	kvm_sigset_activate(vcpu);
8706 	kvm_load_guest_fpu(vcpu);
8707 
8708 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8709 		if (kvm_run->immediate_exit) {
8710 			r = -EINTR;
8711 			goto out;
8712 		}
8713 		kvm_vcpu_block(vcpu);
8714 		kvm_apic_accept_events(vcpu);
8715 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8716 		r = -EAGAIN;
8717 		if (signal_pending(current)) {
8718 			r = -EINTR;
8719 			vcpu->run->exit_reason = KVM_EXIT_INTR;
8720 			++vcpu->stat.signal_exits;
8721 		}
8722 		goto out;
8723 	}
8724 
8725 	if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8726 		r = -EINVAL;
8727 		goto out;
8728 	}
8729 
8730 	if (vcpu->run->kvm_dirty_regs) {
8731 		r = sync_regs(vcpu);
8732 		if (r != 0)
8733 			goto out;
8734 	}
8735 
8736 	/* re-sync apic's tpr */
8737 	if (!lapic_in_kernel(vcpu)) {
8738 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8739 			r = -EINVAL;
8740 			goto out;
8741 		}
8742 	}
8743 
8744 	if (unlikely(vcpu->arch.complete_userspace_io)) {
8745 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8746 		vcpu->arch.complete_userspace_io = NULL;
8747 		r = cui(vcpu);
8748 		if (r <= 0)
8749 			goto out;
8750 	} else
8751 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8752 
8753 	if (kvm_run->immediate_exit)
8754 		r = -EINTR;
8755 	else
8756 		r = vcpu_run(vcpu);
8757 
8758 out:
8759 	kvm_put_guest_fpu(vcpu);
8760 	if (vcpu->run->kvm_valid_regs)
8761 		store_regs(vcpu);
8762 	post_kvm_run_save(vcpu);
8763 	kvm_sigset_deactivate(vcpu);
8764 
8765 	vcpu_put(vcpu);
8766 	return r;
8767 }
8768 
8769 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8770 {
8771 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8772 		/*
8773 		 * We are here if userspace calls get_regs() in the middle of
8774 		 * instruction emulation. Registers state needs to be copied
8775 		 * back from emulation context to vcpu. Userspace shouldn't do
8776 		 * that usually, but some bad designed PV devices (vmware
8777 		 * backdoor interface) need this to work
8778 		 */
8779 		emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
8780 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8781 	}
8782 	regs->rax = kvm_rax_read(vcpu);
8783 	regs->rbx = kvm_rbx_read(vcpu);
8784 	regs->rcx = kvm_rcx_read(vcpu);
8785 	regs->rdx = kvm_rdx_read(vcpu);
8786 	regs->rsi = kvm_rsi_read(vcpu);
8787 	regs->rdi = kvm_rdi_read(vcpu);
8788 	regs->rsp = kvm_rsp_read(vcpu);
8789 	regs->rbp = kvm_rbp_read(vcpu);
8790 #ifdef CONFIG_X86_64
8791 	regs->r8 = kvm_r8_read(vcpu);
8792 	regs->r9 = kvm_r9_read(vcpu);
8793 	regs->r10 = kvm_r10_read(vcpu);
8794 	regs->r11 = kvm_r11_read(vcpu);
8795 	regs->r12 = kvm_r12_read(vcpu);
8796 	regs->r13 = kvm_r13_read(vcpu);
8797 	regs->r14 = kvm_r14_read(vcpu);
8798 	regs->r15 = kvm_r15_read(vcpu);
8799 #endif
8800 
8801 	regs->rip = kvm_rip_read(vcpu);
8802 	regs->rflags = kvm_get_rflags(vcpu);
8803 }
8804 
8805 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8806 {
8807 	vcpu_load(vcpu);
8808 	__get_regs(vcpu, regs);
8809 	vcpu_put(vcpu);
8810 	return 0;
8811 }
8812 
8813 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8814 {
8815 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8816 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8817 
8818 	kvm_rax_write(vcpu, regs->rax);
8819 	kvm_rbx_write(vcpu, regs->rbx);
8820 	kvm_rcx_write(vcpu, regs->rcx);
8821 	kvm_rdx_write(vcpu, regs->rdx);
8822 	kvm_rsi_write(vcpu, regs->rsi);
8823 	kvm_rdi_write(vcpu, regs->rdi);
8824 	kvm_rsp_write(vcpu, regs->rsp);
8825 	kvm_rbp_write(vcpu, regs->rbp);
8826 #ifdef CONFIG_X86_64
8827 	kvm_r8_write(vcpu, regs->r8);
8828 	kvm_r9_write(vcpu, regs->r9);
8829 	kvm_r10_write(vcpu, regs->r10);
8830 	kvm_r11_write(vcpu, regs->r11);
8831 	kvm_r12_write(vcpu, regs->r12);
8832 	kvm_r13_write(vcpu, regs->r13);
8833 	kvm_r14_write(vcpu, regs->r14);
8834 	kvm_r15_write(vcpu, regs->r15);
8835 #endif
8836 
8837 	kvm_rip_write(vcpu, regs->rip);
8838 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8839 
8840 	vcpu->arch.exception.pending = false;
8841 
8842 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8843 }
8844 
8845 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8846 {
8847 	vcpu_load(vcpu);
8848 	__set_regs(vcpu, regs);
8849 	vcpu_put(vcpu);
8850 	return 0;
8851 }
8852 
8853 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8854 {
8855 	struct kvm_segment cs;
8856 
8857 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8858 	*db = cs.db;
8859 	*l = cs.l;
8860 }
8861 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8862 
8863 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8864 {
8865 	struct desc_ptr dt;
8866 
8867 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8868 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8869 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8870 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8871 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8872 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8873 
8874 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8875 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8876 
8877 	kvm_x86_ops.get_idt(vcpu, &dt);
8878 	sregs->idt.limit = dt.size;
8879 	sregs->idt.base = dt.address;
8880 	kvm_x86_ops.get_gdt(vcpu, &dt);
8881 	sregs->gdt.limit = dt.size;
8882 	sregs->gdt.base = dt.address;
8883 
8884 	sregs->cr0 = kvm_read_cr0(vcpu);
8885 	sregs->cr2 = vcpu->arch.cr2;
8886 	sregs->cr3 = kvm_read_cr3(vcpu);
8887 	sregs->cr4 = kvm_read_cr4(vcpu);
8888 	sregs->cr8 = kvm_get_cr8(vcpu);
8889 	sregs->efer = vcpu->arch.efer;
8890 	sregs->apic_base = kvm_get_apic_base(vcpu);
8891 
8892 	memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8893 
8894 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8895 		set_bit(vcpu->arch.interrupt.nr,
8896 			(unsigned long *)sregs->interrupt_bitmap);
8897 }
8898 
8899 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8900 				  struct kvm_sregs *sregs)
8901 {
8902 	vcpu_load(vcpu);
8903 	__get_sregs(vcpu, sregs);
8904 	vcpu_put(vcpu);
8905 	return 0;
8906 }
8907 
8908 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8909 				    struct kvm_mp_state *mp_state)
8910 {
8911 	vcpu_load(vcpu);
8912 	if (kvm_mpx_supported())
8913 		kvm_load_guest_fpu(vcpu);
8914 
8915 	kvm_apic_accept_events(vcpu);
8916 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8917 					vcpu->arch.pv.pv_unhalted)
8918 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8919 	else
8920 		mp_state->mp_state = vcpu->arch.mp_state;
8921 
8922 	if (kvm_mpx_supported())
8923 		kvm_put_guest_fpu(vcpu);
8924 	vcpu_put(vcpu);
8925 	return 0;
8926 }
8927 
8928 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8929 				    struct kvm_mp_state *mp_state)
8930 {
8931 	int ret = -EINVAL;
8932 
8933 	vcpu_load(vcpu);
8934 
8935 	if (!lapic_in_kernel(vcpu) &&
8936 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8937 		goto out;
8938 
8939 	/*
8940 	 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
8941 	 * INIT state; latched init should be reported using
8942 	 * KVM_SET_VCPU_EVENTS, so reject it here.
8943 	 */
8944 	if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
8945 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8946 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8947 		goto out;
8948 
8949 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8950 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8951 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8952 	} else
8953 		vcpu->arch.mp_state = mp_state->mp_state;
8954 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8955 
8956 	ret = 0;
8957 out:
8958 	vcpu_put(vcpu);
8959 	return ret;
8960 }
8961 
8962 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8963 		    int reason, bool has_error_code, u32 error_code)
8964 {
8965 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8966 	int ret;
8967 
8968 	init_emulate_ctxt(vcpu);
8969 
8970 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8971 				   has_error_code, error_code);
8972 	if (ret) {
8973 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8974 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
8975 		vcpu->run->internal.ndata = 0;
8976 		return 0;
8977 	}
8978 
8979 	kvm_rip_write(vcpu, ctxt->eip);
8980 	kvm_set_rflags(vcpu, ctxt->eflags);
8981 	return 1;
8982 }
8983 EXPORT_SYMBOL_GPL(kvm_task_switch);
8984 
8985 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8986 {
8987 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8988 		/*
8989 		 * When EFER.LME and CR0.PG are set, the processor is in
8990 		 * 64-bit mode (though maybe in a 32-bit code segment).
8991 		 * CR4.PAE and EFER.LMA must be set.
8992 		 */
8993 		if (!(sregs->cr4 & X86_CR4_PAE)
8994 		    || !(sregs->efer & EFER_LMA))
8995 			return -EINVAL;
8996 	} else {
8997 		/*
8998 		 * Not in 64-bit mode: EFER.LMA is clear and the code
8999 		 * segment cannot be 64-bit.
9000 		 */
9001 		if (sregs->efer & EFER_LMA || sregs->cs.l)
9002 			return -EINVAL;
9003 	}
9004 
9005 	return kvm_valid_cr4(vcpu, sregs->cr4);
9006 }
9007 
9008 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9009 {
9010 	struct msr_data apic_base_msr;
9011 	int mmu_reset_needed = 0;
9012 	int cpuid_update_needed = 0;
9013 	int pending_vec, max_bits, idx;
9014 	struct desc_ptr dt;
9015 	int ret = -EINVAL;
9016 
9017 	if (kvm_valid_sregs(vcpu, sregs))
9018 		goto out;
9019 
9020 	apic_base_msr.data = sregs->apic_base;
9021 	apic_base_msr.host_initiated = true;
9022 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
9023 		goto out;
9024 
9025 	dt.size = sregs->idt.limit;
9026 	dt.address = sregs->idt.base;
9027 	kvm_x86_ops.set_idt(vcpu, &dt);
9028 	dt.size = sregs->gdt.limit;
9029 	dt.address = sregs->gdt.base;
9030 	kvm_x86_ops.set_gdt(vcpu, &dt);
9031 
9032 	vcpu->arch.cr2 = sregs->cr2;
9033 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
9034 	vcpu->arch.cr3 = sregs->cr3;
9035 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
9036 
9037 	kvm_set_cr8(vcpu, sregs->cr8);
9038 
9039 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
9040 	kvm_x86_ops.set_efer(vcpu, sregs->efer);
9041 
9042 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
9043 	kvm_x86_ops.set_cr0(vcpu, sregs->cr0);
9044 	vcpu->arch.cr0 = sregs->cr0;
9045 
9046 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
9047 	cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
9048 				(X86_CR4_OSXSAVE | X86_CR4_PKE));
9049 	kvm_x86_ops.set_cr4(vcpu, sregs->cr4);
9050 	if (cpuid_update_needed)
9051 		kvm_update_cpuid(vcpu);
9052 
9053 	idx = srcu_read_lock(&vcpu->kvm->srcu);
9054 	if (is_pae_paging(vcpu)) {
9055 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
9056 		mmu_reset_needed = 1;
9057 	}
9058 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
9059 
9060 	if (mmu_reset_needed)
9061 		kvm_mmu_reset_context(vcpu);
9062 
9063 	max_bits = KVM_NR_INTERRUPTS;
9064 	pending_vec = find_first_bit(
9065 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
9066 	if (pending_vec < max_bits) {
9067 		kvm_queue_interrupt(vcpu, pending_vec, false);
9068 		pr_debug("Set back pending irq %d\n", pending_vec);
9069 	}
9070 
9071 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9072 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9073 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9074 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9075 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9076 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9077 
9078 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9079 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9080 
9081 	update_cr8_intercept(vcpu);
9082 
9083 	/* Older userspace won't unhalt the vcpu on reset. */
9084 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9085 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
9086 	    !is_protmode(vcpu))
9087 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9088 
9089 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9090 
9091 	ret = 0;
9092 out:
9093 	return ret;
9094 }
9095 
9096 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
9097 				  struct kvm_sregs *sregs)
9098 {
9099 	int ret;
9100 
9101 	vcpu_load(vcpu);
9102 	ret = __set_sregs(vcpu, sregs);
9103 	vcpu_put(vcpu);
9104 	return ret;
9105 }
9106 
9107 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
9108 					struct kvm_guest_debug *dbg)
9109 {
9110 	unsigned long rflags;
9111 	int i, r;
9112 
9113 	vcpu_load(vcpu);
9114 
9115 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
9116 		r = -EBUSY;
9117 		if (vcpu->arch.exception.pending)
9118 			goto out;
9119 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
9120 			kvm_queue_exception(vcpu, DB_VECTOR);
9121 		else
9122 			kvm_queue_exception(vcpu, BP_VECTOR);
9123 	}
9124 
9125 	/*
9126 	 * Read rflags as long as potentially injected trace flags are still
9127 	 * filtered out.
9128 	 */
9129 	rflags = kvm_get_rflags(vcpu);
9130 
9131 	vcpu->guest_debug = dbg->control;
9132 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
9133 		vcpu->guest_debug = 0;
9134 
9135 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
9136 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
9137 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
9138 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
9139 	} else {
9140 		for (i = 0; i < KVM_NR_DB_REGS; i++)
9141 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
9142 	}
9143 	kvm_update_dr7(vcpu);
9144 
9145 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9146 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
9147 			get_segment_base(vcpu, VCPU_SREG_CS);
9148 
9149 	/*
9150 	 * Trigger an rflags update that will inject or remove the trace
9151 	 * flags.
9152 	 */
9153 	kvm_set_rflags(vcpu, rflags);
9154 
9155 	kvm_x86_ops.update_bp_intercept(vcpu);
9156 
9157 	r = 0;
9158 
9159 out:
9160 	vcpu_put(vcpu);
9161 	return r;
9162 }
9163 
9164 /*
9165  * Translate a guest virtual address to a guest physical address.
9166  */
9167 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
9168 				    struct kvm_translation *tr)
9169 {
9170 	unsigned long vaddr = tr->linear_address;
9171 	gpa_t gpa;
9172 	int idx;
9173 
9174 	vcpu_load(vcpu);
9175 
9176 	idx = srcu_read_lock(&vcpu->kvm->srcu);
9177 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
9178 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
9179 	tr->physical_address = gpa;
9180 	tr->valid = gpa != UNMAPPED_GVA;
9181 	tr->writeable = 1;
9182 	tr->usermode = 0;
9183 
9184 	vcpu_put(vcpu);
9185 	return 0;
9186 }
9187 
9188 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9189 {
9190 	struct fxregs_state *fxsave;
9191 
9192 	vcpu_load(vcpu);
9193 
9194 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9195 	memcpy(fpu->fpr, fxsave->st_space, 128);
9196 	fpu->fcw = fxsave->cwd;
9197 	fpu->fsw = fxsave->swd;
9198 	fpu->ftwx = fxsave->twd;
9199 	fpu->last_opcode = fxsave->fop;
9200 	fpu->last_ip = fxsave->rip;
9201 	fpu->last_dp = fxsave->rdp;
9202 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
9203 
9204 	vcpu_put(vcpu);
9205 	return 0;
9206 }
9207 
9208 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9209 {
9210 	struct fxregs_state *fxsave;
9211 
9212 	vcpu_load(vcpu);
9213 
9214 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9215 
9216 	memcpy(fxsave->st_space, fpu->fpr, 128);
9217 	fxsave->cwd = fpu->fcw;
9218 	fxsave->swd = fpu->fsw;
9219 	fxsave->twd = fpu->ftwx;
9220 	fxsave->fop = fpu->last_opcode;
9221 	fxsave->rip = fpu->last_ip;
9222 	fxsave->rdp = fpu->last_dp;
9223 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
9224 
9225 	vcpu_put(vcpu);
9226 	return 0;
9227 }
9228 
9229 static void store_regs(struct kvm_vcpu *vcpu)
9230 {
9231 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
9232 
9233 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
9234 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
9235 
9236 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
9237 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
9238 
9239 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
9240 		kvm_vcpu_ioctl_x86_get_vcpu_events(
9241 				vcpu, &vcpu->run->s.regs.events);
9242 }
9243 
9244 static int sync_regs(struct kvm_vcpu *vcpu)
9245 {
9246 	if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
9247 		return -EINVAL;
9248 
9249 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
9250 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
9251 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
9252 	}
9253 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
9254 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
9255 			return -EINVAL;
9256 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
9257 	}
9258 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
9259 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9260 				vcpu, &vcpu->run->s.regs.events))
9261 			return -EINVAL;
9262 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9263 	}
9264 
9265 	return 0;
9266 }
9267 
9268 static void fx_init(struct kvm_vcpu *vcpu)
9269 {
9270 	fpstate_init(&vcpu->arch.guest_fpu->state);
9271 	if (boot_cpu_has(X86_FEATURE_XSAVES))
9272 		vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9273 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
9274 
9275 	/*
9276 	 * Ensure guest xcr0 is valid for loading
9277 	 */
9278 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9279 
9280 	vcpu->arch.cr0 |= X86_CR0_ET;
9281 }
9282 
9283 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
9284 {
9285 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9286 		pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
9287 			     "guest TSC will not be reliable\n");
9288 
9289 	return 0;
9290 }
9291 
9292 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
9293 {
9294 	struct page *page;
9295 	int r;
9296 
9297 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9298 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9299 	else
9300 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9301 
9302 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
9303 
9304 	r = kvm_mmu_create(vcpu);
9305 	if (r < 0)
9306 		return r;
9307 
9308 	if (irqchip_in_kernel(vcpu->kvm)) {
9309 		r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9310 		if (r < 0)
9311 			goto fail_mmu_destroy;
9312 		if (kvm_apicv_activated(vcpu->kvm))
9313 			vcpu->arch.apicv_active = true;
9314 	} else
9315 		static_key_slow_inc(&kvm_no_apic_vcpu);
9316 
9317 	r = -ENOMEM;
9318 
9319 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9320 	if (!page)
9321 		goto fail_free_lapic;
9322 	vcpu->arch.pio_data = page_address(page);
9323 
9324 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9325 				       GFP_KERNEL_ACCOUNT);
9326 	if (!vcpu->arch.mce_banks)
9327 		goto fail_free_pio_data;
9328 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9329 
9330 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9331 				GFP_KERNEL_ACCOUNT))
9332 		goto fail_free_mce_banks;
9333 
9334 	if (!alloc_emulate_ctxt(vcpu))
9335 		goto free_wbinvd_dirty_mask;
9336 
9337 	vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
9338 						GFP_KERNEL_ACCOUNT);
9339 	if (!vcpu->arch.user_fpu) {
9340 		pr_err("kvm: failed to allocate userspace's fpu\n");
9341 		goto free_emulate_ctxt;
9342 	}
9343 
9344 	vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
9345 						 GFP_KERNEL_ACCOUNT);
9346 	if (!vcpu->arch.guest_fpu) {
9347 		pr_err("kvm: failed to allocate vcpu's fpu\n");
9348 		goto free_user_fpu;
9349 	}
9350 	fx_init(vcpu);
9351 
9352 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9353 
9354 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9355 
9356 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9357 
9358 	kvm_async_pf_hash_reset(vcpu);
9359 	kvm_pmu_init(vcpu);
9360 
9361 	vcpu->arch.pending_external_vector = -1;
9362 	vcpu->arch.preempted_in_kernel = false;
9363 
9364 	kvm_hv_vcpu_init(vcpu);
9365 
9366 	r = kvm_x86_ops.vcpu_create(vcpu);
9367 	if (r)
9368 		goto free_guest_fpu;
9369 
9370 	vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
9371 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
9372 	kvm_vcpu_mtrr_init(vcpu);
9373 	vcpu_load(vcpu);
9374 	kvm_vcpu_reset(vcpu, false);
9375 	kvm_init_mmu(vcpu, false);
9376 	vcpu_put(vcpu);
9377 	return 0;
9378 
9379 free_guest_fpu:
9380 	kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9381 free_user_fpu:
9382 	kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9383 free_emulate_ctxt:
9384 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
9385 free_wbinvd_dirty_mask:
9386 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9387 fail_free_mce_banks:
9388 	kfree(vcpu->arch.mce_banks);
9389 fail_free_pio_data:
9390 	free_page((unsigned long)vcpu->arch.pio_data);
9391 fail_free_lapic:
9392 	kvm_free_lapic(vcpu);
9393 fail_mmu_destroy:
9394 	kvm_mmu_destroy(vcpu);
9395 	return r;
9396 }
9397 
9398 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
9399 {
9400 	struct msr_data msr;
9401 	struct kvm *kvm = vcpu->kvm;
9402 
9403 	kvm_hv_vcpu_postcreate(vcpu);
9404 
9405 	if (mutex_lock_killable(&vcpu->mutex))
9406 		return;
9407 	vcpu_load(vcpu);
9408 	msr.data = 0x0;
9409 	msr.index = MSR_IA32_TSC;
9410 	msr.host_initiated = true;
9411 	kvm_write_tsc(vcpu, &msr);
9412 	vcpu_put(vcpu);
9413 
9414 	/* poll control enabled by default */
9415 	vcpu->arch.msr_kvm_poll_control = 1;
9416 
9417 	mutex_unlock(&vcpu->mutex);
9418 
9419 	if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
9420 		schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9421 						KVMCLOCK_SYNC_PERIOD);
9422 }
9423 
9424 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9425 {
9426 	struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
9427 	int idx;
9428 
9429 	kvm_release_pfn(cache->pfn, cache->dirty, cache);
9430 
9431 	kvmclock_reset(vcpu);
9432 
9433 	kvm_x86_ops.vcpu_free(vcpu);
9434 
9435 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
9436 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9437 	kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9438 	kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9439 
9440 	kvm_hv_vcpu_uninit(vcpu);
9441 	kvm_pmu_destroy(vcpu);
9442 	kfree(vcpu->arch.mce_banks);
9443 	kvm_free_lapic(vcpu);
9444 	idx = srcu_read_lock(&vcpu->kvm->srcu);
9445 	kvm_mmu_destroy(vcpu);
9446 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
9447 	free_page((unsigned long)vcpu->arch.pio_data);
9448 	if (!lapic_in_kernel(vcpu))
9449 		static_key_slow_dec(&kvm_no_apic_vcpu);
9450 }
9451 
9452 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
9453 {
9454 	kvm_lapic_reset(vcpu, init_event);
9455 
9456 	vcpu->arch.hflags = 0;
9457 
9458 	vcpu->arch.smi_pending = 0;
9459 	vcpu->arch.smi_count = 0;
9460 	atomic_set(&vcpu->arch.nmi_queued, 0);
9461 	vcpu->arch.nmi_pending = 0;
9462 	vcpu->arch.nmi_injected = false;
9463 	kvm_clear_interrupt_queue(vcpu);
9464 	kvm_clear_exception_queue(vcpu);
9465 
9466 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
9467 	kvm_update_dr0123(vcpu);
9468 	vcpu->arch.dr6 = DR6_INIT;
9469 	kvm_update_dr6(vcpu);
9470 	vcpu->arch.dr7 = DR7_FIXED_1;
9471 	kvm_update_dr7(vcpu);
9472 
9473 	vcpu->arch.cr2 = 0;
9474 
9475 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9476 	vcpu->arch.apf.msr_val = 0;
9477 	vcpu->arch.st.msr_val = 0;
9478 
9479 	kvmclock_reset(vcpu);
9480 
9481 	kvm_clear_async_pf_completion_queue(vcpu);
9482 	kvm_async_pf_hash_reset(vcpu);
9483 	vcpu->arch.apf.halted = false;
9484 
9485 	if (kvm_mpx_supported()) {
9486 		void *mpx_state_buffer;
9487 
9488 		/*
9489 		 * To avoid have the INIT path from kvm_apic_has_events() that be
9490 		 * called with loaded FPU and does not let userspace fix the state.
9491 		 */
9492 		if (init_event)
9493 			kvm_put_guest_fpu(vcpu);
9494 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9495 					XFEATURE_BNDREGS);
9496 		if (mpx_state_buffer)
9497 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
9498 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9499 					XFEATURE_BNDCSR);
9500 		if (mpx_state_buffer)
9501 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
9502 		if (init_event)
9503 			kvm_load_guest_fpu(vcpu);
9504 	}
9505 
9506 	if (!init_event) {
9507 		kvm_pmu_reset(vcpu);
9508 		vcpu->arch.smbase = 0x30000;
9509 
9510 		vcpu->arch.msr_misc_features_enables = 0;
9511 
9512 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9513 	}
9514 
9515 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
9516 	vcpu->arch.regs_avail = ~0;
9517 	vcpu->arch.regs_dirty = ~0;
9518 
9519 	vcpu->arch.ia32_xss = 0;
9520 
9521 	kvm_x86_ops.vcpu_reset(vcpu, init_event);
9522 }
9523 
9524 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
9525 {
9526 	struct kvm_segment cs;
9527 
9528 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9529 	cs.selector = vector << 8;
9530 	cs.base = vector << 12;
9531 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9532 	kvm_rip_write(vcpu, 0);
9533 }
9534 
9535 int kvm_arch_hardware_enable(void)
9536 {
9537 	struct kvm *kvm;
9538 	struct kvm_vcpu *vcpu;
9539 	int i;
9540 	int ret;
9541 	u64 local_tsc;
9542 	u64 max_tsc = 0;
9543 	bool stable, backwards_tsc = false;
9544 
9545 	kvm_shared_msr_cpu_online();
9546 	ret = kvm_x86_ops.hardware_enable();
9547 	if (ret != 0)
9548 		return ret;
9549 
9550 	local_tsc = rdtsc();
9551 	stable = !kvm_check_tsc_unstable();
9552 	list_for_each_entry(kvm, &vm_list, vm_list) {
9553 		kvm_for_each_vcpu(i, vcpu, kvm) {
9554 			if (!stable && vcpu->cpu == smp_processor_id())
9555 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9556 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
9557 				backwards_tsc = true;
9558 				if (vcpu->arch.last_host_tsc > max_tsc)
9559 					max_tsc = vcpu->arch.last_host_tsc;
9560 			}
9561 		}
9562 	}
9563 
9564 	/*
9565 	 * Sometimes, even reliable TSCs go backwards.  This happens on
9566 	 * platforms that reset TSC during suspend or hibernate actions, but
9567 	 * maintain synchronization.  We must compensate.  Fortunately, we can
9568 	 * detect that condition here, which happens early in CPU bringup,
9569 	 * before any KVM threads can be running.  Unfortunately, we can't
9570 	 * bring the TSCs fully up to date with real time, as we aren't yet far
9571 	 * enough into CPU bringup that we know how much real time has actually
9572 	 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
9573 	 * variables that haven't been updated yet.
9574 	 *
9575 	 * So we simply find the maximum observed TSC above, then record the
9576 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
9577 	 * the adjustment will be applied.  Note that we accumulate
9578 	 * adjustments, in case multiple suspend cycles happen before some VCPU
9579 	 * gets a chance to run again.  In the event that no KVM threads get a
9580 	 * chance to run, we will miss the entire elapsed period, as we'll have
9581 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
9582 	 * loose cycle time.  This isn't too big a deal, since the loss will be
9583 	 * uniform across all VCPUs (not to mention the scenario is extremely
9584 	 * unlikely). It is possible that a second hibernate recovery happens
9585 	 * much faster than a first, causing the observed TSC here to be
9586 	 * smaller; this would require additional padding adjustment, which is
9587 	 * why we set last_host_tsc to the local tsc observed here.
9588 	 *
9589 	 * N.B. - this code below runs only on platforms with reliable TSC,
9590 	 * as that is the only way backwards_tsc is set above.  Also note
9591 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
9592 	 * have the same delta_cyc adjustment applied if backwards_tsc
9593 	 * is detected.  Note further, this adjustment is only done once,
9594 	 * as we reset last_host_tsc on all VCPUs to stop this from being
9595 	 * called multiple times (one for each physical CPU bringup).
9596 	 *
9597 	 * Platforms with unreliable TSCs don't have to deal with this, they
9598 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
9599 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
9600 	 * guarantee that they stay in perfect synchronization.
9601 	 */
9602 	if (backwards_tsc) {
9603 		u64 delta_cyc = max_tsc - local_tsc;
9604 		list_for_each_entry(kvm, &vm_list, vm_list) {
9605 			kvm->arch.backwards_tsc_observed = true;
9606 			kvm_for_each_vcpu(i, vcpu, kvm) {
9607 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
9608 				vcpu->arch.last_host_tsc = local_tsc;
9609 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9610 			}
9611 
9612 			/*
9613 			 * We have to disable TSC offset matching.. if you were
9614 			 * booting a VM while issuing an S4 host suspend....
9615 			 * you may have some problem.  Solving this issue is
9616 			 * left as an exercise to the reader.
9617 			 */
9618 			kvm->arch.last_tsc_nsec = 0;
9619 			kvm->arch.last_tsc_write = 0;
9620 		}
9621 
9622 	}
9623 	return 0;
9624 }
9625 
9626 void kvm_arch_hardware_disable(void)
9627 {
9628 	kvm_x86_ops.hardware_disable();
9629 	drop_user_return_notifiers();
9630 }
9631 
9632 int kvm_arch_hardware_setup(void *opaque)
9633 {
9634 	struct kvm_x86_init_ops *ops = opaque;
9635 	int r;
9636 
9637 	rdmsrl_safe(MSR_EFER, &host_efer);
9638 
9639 	if (boot_cpu_has(X86_FEATURE_XSAVES))
9640 		rdmsrl(MSR_IA32_XSS, host_xss);
9641 
9642 	r = ops->hardware_setup();
9643 	if (r != 0)
9644 		return r;
9645 
9646 	memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
9647 
9648 	if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
9649 		supported_xss = 0;
9650 
9651 	cr4_reserved_bits = kvm_host_cr4_reserved_bits(&boot_cpu_data);
9652 
9653 	if (kvm_has_tsc_control) {
9654 		/*
9655 		 * Make sure the user can only configure tsc_khz values that
9656 		 * fit into a signed integer.
9657 		 * A min value is not calculated because it will always
9658 		 * be 1 on all machines.
9659 		 */
9660 		u64 max = min(0x7fffffffULL,
9661 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9662 		kvm_max_guest_tsc_khz = max;
9663 
9664 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9665 	}
9666 
9667 	kvm_init_msr_list();
9668 	return 0;
9669 }
9670 
9671 void kvm_arch_hardware_unsetup(void)
9672 {
9673 	kvm_x86_ops.hardware_unsetup();
9674 }
9675 
9676 int kvm_arch_check_processor_compat(void *opaque)
9677 {
9678 	struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
9679 	struct kvm_x86_init_ops *ops = opaque;
9680 
9681 	WARN_ON(!irqs_disabled());
9682 
9683 	if (kvm_host_cr4_reserved_bits(c) != cr4_reserved_bits)
9684 		return -EIO;
9685 
9686 	return ops->check_processor_compatibility();
9687 }
9688 
9689 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9690 {
9691 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9692 }
9693 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9694 
9695 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9696 {
9697 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9698 }
9699 
9700 struct static_key kvm_no_apic_vcpu __read_mostly;
9701 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9702 
9703 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9704 {
9705 	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
9706 
9707 	vcpu->arch.l1tf_flush_l1d = true;
9708 	if (pmu->version && unlikely(pmu->event_count)) {
9709 		pmu->need_cleanup = true;
9710 		kvm_make_request(KVM_REQ_PMU, vcpu);
9711 	}
9712 	kvm_x86_ops.sched_in(vcpu, cpu);
9713 }
9714 
9715 void kvm_arch_free_vm(struct kvm *kvm)
9716 {
9717 	kfree(kvm->arch.hyperv.hv_pa_pg);
9718 	vfree(kvm);
9719 }
9720 
9721 
9722 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9723 {
9724 	if (type)
9725 		return -EINVAL;
9726 
9727 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9728 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9729 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9730 	INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
9731 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9732 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9733 
9734 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9735 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9736 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9737 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9738 		&kvm->arch.irq_sources_bitmap);
9739 
9740 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9741 	mutex_init(&kvm->arch.apic_map_lock);
9742 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9743 
9744 	kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
9745 	pvclock_update_vm_gtod_copy(kvm);
9746 
9747 	kvm->arch.guest_can_read_msr_platform_info = true;
9748 
9749 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9750 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9751 
9752 	kvm_hv_init_vm(kvm);
9753 	kvm_page_track_init(kvm);
9754 	kvm_mmu_init_vm(kvm);
9755 
9756 	return kvm_x86_ops.vm_init(kvm);
9757 }
9758 
9759 int kvm_arch_post_init_vm(struct kvm *kvm)
9760 {
9761 	return kvm_mmu_post_init_vm(kvm);
9762 }
9763 
9764 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9765 {
9766 	vcpu_load(vcpu);
9767 	kvm_mmu_unload(vcpu);
9768 	vcpu_put(vcpu);
9769 }
9770 
9771 static void kvm_free_vcpus(struct kvm *kvm)
9772 {
9773 	unsigned int i;
9774 	struct kvm_vcpu *vcpu;
9775 
9776 	/*
9777 	 * Unpin any mmu pages first.
9778 	 */
9779 	kvm_for_each_vcpu(i, vcpu, kvm) {
9780 		kvm_clear_async_pf_completion_queue(vcpu);
9781 		kvm_unload_vcpu_mmu(vcpu);
9782 	}
9783 	kvm_for_each_vcpu(i, vcpu, kvm)
9784 		kvm_vcpu_destroy(vcpu);
9785 
9786 	mutex_lock(&kvm->lock);
9787 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9788 		kvm->vcpus[i] = NULL;
9789 
9790 	atomic_set(&kvm->online_vcpus, 0);
9791 	mutex_unlock(&kvm->lock);
9792 }
9793 
9794 void kvm_arch_sync_events(struct kvm *kvm)
9795 {
9796 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9797 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9798 	kvm_free_pit(kvm);
9799 }
9800 
9801 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9802 {
9803 	int i, r;
9804 	unsigned long hva, uninitialized_var(old_npages);
9805 	struct kvm_memslots *slots = kvm_memslots(kvm);
9806 	struct kvm_memory_slot *slot;
9807 
9808 	/* Called with kvm->slots_lock held.  */
9809 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9810 		return -EINVAL;
9811 
9812 	slot = id_to_memslot(slots, id);
9813 	if (size) {
9814 		if (slot && slot->npages)
9815 			return -EEXIST;
9816 
9817 		/*
9818 		 * MAP_SHARED to prevent internal slot pages from being moved
9819 		 * by fork()/COW.
9820 		 */
9821 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9822 			      MAP_SHARED | MAP_ANONYMOUS, 0);
9823 		if (IS_ERR((void *)hva))
9824 			return PTR_ERR((void *)hva);
9825 	} else {
9826 		if (!slot || !slot->npages)
9827 			return 0;
9828 
9829 		/*
9830 		 * Stuff a non-canonical value to catch use-after-delete.  This
9831 		 * ends up being 0 on 32-bit KVM, but there's no better
9832 		 * alternative.
9833 		 */
9834 		hva = (unsigned long)(0xdeadull << 48);
9835 		old_npages = slot->npages;
9836 	}
9837 
9838 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9839 		struct kvm_userspace_memory_region m;
9840 
9841 		m.slot = id | (i << 16);
9842 		m.flags = 0;
9843 		m.guest_phys_addr = gpa;
9844 		m.userspace_addr = hva;
9845 		m.memory_size = size;
9846 		r = __kvm_set_memory_region(kvm, &m);
9847 		if (r < 0)
9848 			return r;
9849 	}
9850 
9851 	if (!size)
9852 		vm_munmap(hva, old_npages * PAGE_SIZE);
9853 
9854 	return 0;
9855 }
9856 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9857 
9858 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
9859 {
9860 	kvm_mmu_pre_destroy_vm(kvm);
9861 }
9862 
9863 void kvm_arch_destroy_vm(struct kvm *kvm)
9864 {
9865 	if (current->mm == kvm->mm) {
9866 		/*
9867 		 * Free memory regions allocated on behalf of userspace,
9868 		 * unless the the memory map has changed due to process exit
9869 		 * or fd copying.
9870 		 */
9871 		mutex_lock(&kvm->slots_lock);
9872 		__x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
9873 					0, 0);
9874 		__x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
9875 					0, 0);
9876 		__x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9877 		mutex_unlock(&kvm->slots_lock);
9878 	}
9879 	if (kvm_x86_ops.vm_destroy)
9880 		kvm_x86_ops.vm_destroy(kvm);
9881 	kvm_pic_destroy(kvm);
9882 	kvm_ioapic_destroy(kvm);
9883 	kvm_free_vcpus(kvm);
9884 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9885 	kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
9886 	kvm_mmu_uninit_vm(kvm);
9887 	kvm_page_track_cleanup(kvm);
9888 	kvm_hv_destroy_vm(kvm);
9889 }
9890 
9891 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
9892 {
9893 	int i;
9894 
9895 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9896 		kvfree(slot->arch.rmap[i]);
9897 		slot->arch.rmap[i] = NULL;
9898 
9899 		if (i == 0)
9900 			continue;
9901 
9902 		kvfree(slot->arch.lpage_info[i - 1]);
9903 		slot->arch.lpage_info[i - 1] = NULL;
9904 	}
9905 
9906 	kvm_page_track_free_memslot(slot);
9907 }
9908 
9909 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
9910 				      unsigned long npages)
9911 {
9912 	int i;
9913 
9914 	/*
9915 	 * Clear out the previous array pointers for the KVM_MR_MOVE case.  The
9916 	 * old arrays will be freed by __kvm_set_memory_region() if installing
9917 	 * the new memslot is successful.
9918 	 */
9919 	memset(&slot->arch, 0, sizeof(slot->arch));
9920 
9921 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9922 		struct kvm_lpage_info *linfo;
9923 		unsigned long ugfn;
9924 		int lpages;
9925 		int level = i + 1;
9926 
9927 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
9928 				      slot->base_gfn, level) + 1;
9929 
9930 		slot->arch.rmap[i] =
9931 			kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9932 				 GFP_KERNEL_ACCOUNT);
9933 		if (!slot->arch.rmap[i])
9934 			goto out_free;
9935 		if (i == 0)
9936 			continue;
9937 
9938 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
9939 		if (!linfo)
9940 			goto out_free;
9941 
9942 		slot->arch.lpage_info[i - 1] = linfo;
9943 
9944 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9945 			linfo[0].disallow_lpage = 1;
9946 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9947 			linfo[lpages - 1].disallow_lpage = 1;
9948 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
9949 		/*
9950 		 * If the gfn and userspace address are not aligned wrt each
9951 		 * other, disable large page support for this slot.
9952 		 */
9953 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
9954 			unsigned long j;
9955 
9956 			for (j = 0; j < lpages; ++j)
9957 				linfo[j].disallow_lpage = 1;
9958 		}
9959 	}
9960 
9961 	if (kvm_page_track_create_memslot(slot, npages))
9962 		goto out_free;
9963 
9964 	return 0;
9965 
9966 out_free:
9967 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9968 		kvfree(slot->arch.rmap[i]);
9969 		slot->arch.rmap[i] = NULL;
9970 		if (i == 0)
9971 			continue;
9972 
9973 		kvfree(slot->arch.lpage_info[i - 1]);
9974 		slot->arch.lpage_info[i - 1] = NULL;
9975 	}
9976 	return -ENOMEM;
9977 }
9978 
9979 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
9980 {
9981 	struct kvm_vcpu *vcpu;
9982 	int i;
9983 
9984 	/*
9985 	 * memslots->generation has been incremented.
9986 	 * mmio generation may have reached its maximum value.
9987 	 */
9988 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
9989 
9990 	/* Force re-initialization of steal_time cache */
9991 	kvm_for_each_vcpu(i, vcpu, kvm)
9992 		kvm_vcpu_kick(vcpu);
9993 }
9994 
9995 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9996 				struct kvm_memory_slot *memslot,
9997 				const struct kvm_userspace_memory_region *mem,
9998 				enum kvm_mr_change change)
9999 {
10000 	if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
10001 		return kvm_alloc_memslot_metadata(memslot,
10002 						  mem->memory_size >> PAGE_SHIFT);
10003 	return 0;
10004 }
10005 
10006 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
10007 				     struct kvm_memory_slot *new)
10008 {
10009 	/* Still write protect RO slot */
10010 	if (new->flags & KVM_MEM_READONLY) {
10011 		kvm_mmu_slot_remove_write_access(kvm, new, PT_PAGE_TABLE_LEVEL);
10012 		return;
10013 	}
10014 
10015 	/*
10016 	 * Call kvm_x86_ops dirty logging hooks when they are valid.
10017 	 *
10018 	 * kvm_x86_ops.slot_disable_log_dirty is called when:
10019 	 *
10020 	 *  - KVM_MR_CREATE with dirty logging is disabled
10021 	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
10022 	 *
10023 	 * The reason is, in case of PML, we need to set D-bit for any slots
10024 	 * with dirty logging disabled in order to eliminate unnecessary GPA
10025 	 * logging in PML buffer (and potential PML buffer full VMEXIT). This
10026 	 * guarantees leaving PML enabled during guest's lifetime won't have
10027 	 * any additional overhead from PML when guest is running with dirty
10028 	 * logging disabled for memory slots.
10029 	 *
10030 	 * kvm_x86_ops.slot_enable_log_dirty is called when switching new slot
10031 	 * to dirty logging mode.
10032 	 *
10033 	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
10034 	 *
10035 	 * In case of write protect:
10036 	 *
10037 	 * Write protect all pages for dirty logging.
10038 	 *
10039 	 * All the sptes including the large sptes which point to this
10040 	 * slot are set to readonly. We can not create any new large
10041 	 * spte on this slot until the end of the logging.
10042 	 *
10043 	 * See the comments in fast_page_fault().
10044 	 */
10045 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
10046 		if (kvm_x86_ops.slot_enable_log_dirty) {
10047 			kvm_x86_ops.slot_enable_log_dirty(kvm, new);
10048 		} else {
10049 			int level =
10050 				kvm_dirty_log_manual_protect_and_init_set(kvm) ?
10051 				PT_DIRECTORY_LEVEL : PT_PAGE_TABLE_LEVEL;
10052 
10053 			/*
10054 			 * If we're with initial-all-set, we don't need
10055 			 * to write protect any small page because
10056 			 * they're reported as dirty already.  However
10057 			 * we still need to write-protect huge pages
10058 			 * so that the page split can happen lazily on
10059 			 * the first write to the huge page.
10060 			 */
10061 			kvm_mmu_slot_remove_write_access(kvm, new, level);
10062 		}
10063 	} else {
10064 		if (kvm_x86_ops.slot_disable_log_dirty)
10065 			kvm_x86_ops.slot_disable_log_dirty(kvm, new);
10066 	}
10067 }
10068 
10069 void kvm_arch_commit_memory_region(struct kvm *kvm,
10070 				const struct kvm_userspace_memory_region *mem,
10071 				struct kvm_memory_slot *old,
10072 				const struct kvm_memory_slot *new,
10073 				enum kvm_mr_change change)
10074 {
10075 	if (!kvm->arch.n_requested_mmu_pages)
10076 		kvm_mmu_change_mmu_pages(kvm,
10077 				kvm_mmu_calculate_default_mmu_pages(kvm));
10078 
10079 	/*
10080 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
10081 	 * sptes have to be split.  If live migration is successful, the guest
10082 	 * in the source machine will be destroyed and large sptes will be
10083 	 * created in the destination. However, if the guest continues to run
10084 	 * in the source machine (for example if live migration fails), small
10085 	 * sptes will remain around and cause bad performance.
10086 	 *
10087 	 * Scan sptes if dirty logging has been stopped, dropping those
10088 	 * which can be collapsed into a single large-page spte.  Later
10089 	 * page faults will create the large-page sptes.
10090 	 *
10091 	 * There is no need to do this in any of the following cases:
10092 	 * CREATE:	No dirty mappings will already exist.
10093 	 * MOVE/DELETE:	The old mappings will already have been cleaned up by
10094 	 *		kvm_arch_flush_shadow_memslot()
10095 	 */
10096 	if (change == KVM_MR_FLAGS_ONLY &&
10097 		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
10098 		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
10099 		kvm_mmu_zap_collapsible_sptes(kvm, new);
10100 
10101 	/*
10102 	 * Set up write protection and/or dirty logging for the new slot.
10103 	 *
10104 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
10105 	 * been zapped so no dirty logging staff is needed for old slot. For
10106 	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
10107 	 * new and it's also covered when dealing with the new slot.
10108 	 *
10109 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
10110 	 */
10111 	if (change != KVM_MR_DELETE)
10112 		kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
10113 
10114 	/* Free the arrays associated with the old memslot. */
10115 	if (change == KVM_MR_MOVE)
10116 		kvm_arch_free_memslot(kvm, old);
10117 }
10118 
10119 void kvm_arch_flush_shadow_all(struct kvm *kvm)
10120 {
10121 	kvm_mmu_zap_all(kvm);
10122 }
10123 
10124 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
10125 				   struct kvm_memory_slot *slot)
10126 {
10127 	kvm_page_track_flush_slot(kvm, slot);
10128 }
10129 
10130 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
10131 {
10132 	return (is_guest_mode(vcpu) &&
10133 			kvm_x86_ops.guest_apic_has_interrupt &&
10134 			kvm_x86_ops.guest_apic_has_interrupt(vcpu));
10135 }
10136 
10137 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
10138 {
10139 	if (!list_empty_careful(&vcpu->async_pf.done))
10140 		return true;
10141 
10142 	if (kvm_apic_has_events(vcpu))
10143 		return true;
10144 
10145 	if (vcpu->arch.pv.pv_unhalted)
10146 		return true;
10147 
10148 	if (vcpu->arch.exception.pending)
10149 		return true;
10150 
10151 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10152 	    (vcpu->arch.nmi_pending &&
10153 	     kvm_x86_ops.nmi_allowed(vcpu)))
10154 		return true;
10155 
10156 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
10157 	    (vcpu->arch.smi_pending && !is_smm(vcpu)))
10158 		return true;
10159 
10160 	if (kvm_arch_interrupt_allowed(vcpu) &&
10161 	    (kvm_cpu_has_interrupt(vcpu) ||
10162 	    kvm_guest_apic_has_interrupt(vcpu)))
10163 		return true;
10164 
10165 	if (kvm_hv_has_stimer_pending(vcpu))
10166 		return true;
10167 
10168 	return false;
10169 }
10170 
10171 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
10172 {
10173 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
10174 }
10175 
10176 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
10177 {
10178 	if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
10179 		return true;
10180 
10181 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10182 		kvm_test_request(KVM_REQ_SMI, vcpu) ||
10183 		 kvm_test_request(KVM_REQ_EVENT, vcpu))
10184 		return true;
10185 
10186 	if (vcpu->arch.apicv_active && kvm_x86_ops.dy_apicv_has_pending_interrupt(vcpu))
10187 		return true;
10188 
10189 	return false;
10190 }
10191 
10192 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
10193 {
10194 	return vcpu->arch.preempted_in_kernel;
10195 }
10196 
10197 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
10198 {
10199 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
10200 }
10201 
10202 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
10203 {
10204 	return kvm_x86_ops.interrupt_allowed(vcpu);
10205 }
10206 
10207 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
10208 {
10209 	if (is_64_bit_mode(vcpu))
10210 		return kvm_rip_read(vcpu);
10211 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
10212 		     kvm_rip_read(vcpu));
10213 }
10214 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
10215 
10216 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
10217 {
10218 	return kvm_get_linear_rip(vcpu) == linear_rip;
10219 }
10220 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
10221 
10222 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
10223 {
10224 	unsigned long rflags;
10225 
10226 	rflags = kvm_x86_ops.get_rflags(vcpu);
10227 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10228 		rflags &= ~X86_EFLAGS_TF;
10229 	return rflags;
10230 }
10231 EXPORT_SYMBOL_GPL(kvm_get_rflags);
10232 
10233 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10234 {
10235 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
10236 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
10237 		rflags |= X86_EFLAGS_TF;
10238 	kvm_x86_ops.set_rflags(vcpu, rflags);
10239 }
10240 
10241 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10242 {
10243 	__kvm_set_rflags(vcpu, rflags);
10244 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10245 }
10246 EXPORT_SYMBOL_GPL(kvm_set_rflags);
10247 
10248 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
10249 {
10250 	int r;
10251 
10252 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
10253 	      work->wakeup_all)
10254 		return;
10255 
10256 	r = kvm_mmu_reload(vcpu);
10257 	if (unlikely(r))
10258 		return;
10259 
10260 	if (!vcpu->arch.mmu->direct_map &&
10261 	      work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
10262 		return;
10263 
10264 	kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
10265 }
10266 
10267 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
10268 {
10269 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
10270 }
10271 
10272 static inline u32 kvm_async_pf_next_probe(u32 key)
10273 {
10274 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
10275 }
10276 
10277 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10278 {
10279 	u32 key = kvm_async_pf_hash_fn(gfn);
10280 
10281 	while (vcpu->arch.apf.gfns[key] != ~0)
10282 		key = kvm_async_pf_next_probe(key);
10283 
10284 	vcpu->arch.apf.gfns[key] = gfn;
10285 }
10286 
10287 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
10288 {
10289 	int i;
10290 	u32 key = kvm_async_pf_hash_fn(gfn);
10291 
10292 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
10293 		     (vcpu->arch.apf.gfns[key] != gfn &&
10294 		      vcpu->arch.apf.gfns[key] != ~0); i++)
10295 		key = kvm_async_pf_next_probe(key);
10296 
10297 	return key;
10298 }
10299 
10300 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10301 {
10302 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
10303 }
10304 
10305 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10306 {
10307 	u32 i, j, k;
10308 
10309 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
10310 	while (true) {
10311 		vcpu->arch.apf.gfns[i] = ~0;
10312 		do {
10313 			j = kvm_async_pf_next_probe(j);
10314 			if (vcpu->arch.apf.gfns[j] == ~0)
10315 				return;
10316 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
10317 			/*
10318 			 * k lies cyclically in ]i,j]
10319 			 * |    i.k.j |
10320 			 * |....j i.k.| or  |.k..j i...|
10321 			 */
10322 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
10323 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
10324 		i = j;
10325 	}
10326 }
10327 
10328 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
10329 {
10330 
10331 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
10332 				      sizeof(val));
10333 }
10334 
10335 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
10336 {
10337 
10338 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
10339 				      sizeof(u32));
10340 }
10341 
10342 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
10343 {
10344 	if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
10345 		return false;
10346 
10347 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
10348 	    (vcpu->arch.apf.send_user_only &&
10349 	     kvm_x86_ops.get_cpl(vcpu) == 0))
10350 		return false;
10351 
10352 	return true;
10353 }
10354 
10355 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
10356 {
10357 	if (unlikely(!lapic_in_kernel(vcpu) ||
10358 		     kvm_event_needs_reinjection(vcpu) ||
10359 		     vcpu->arch.exception.pending))
10360 		return false;
10361 
10362 	if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
10363 		return false;
10364 
10365 	/*
10366 	 * If interrupts are off we cannot even use an artificial
10367 	 * halt state.
10368 	 */
10369 	return kvm_x86_ops.interrupt_allowed(vcpu);
10370 }
10371 
10372 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
10373 				     struct kvm_async_pf *work)
10374 {
10375 	struct x86_exception fault;
10376 
10377 	trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
10378 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
10379 
10380 	if (kvm_can_deliver_async_pf(vcpu) &&
10381 	    !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
10382 		fault.vector = PF_VECTOR;
10383 		fault.error_code_valid = true;
10384 		fault.error_code = 0;
10385 		fault.nested_page_fault = false;
10386 		fault.address = work->arch.token;
10387 		fault.async_page_fault = true;
10388 		kvm_inject_page_fault(vcpu, &fault);
10389 	} else {
10390 		/*
10391 		 * It is not possible to deliver a paravirtualized asynchronous
10392 		 * page fault, but putting the guest in an artificial halt state
10393 		 * can be beneficial nevertheless: if an interrupt arrives, we
10394 		 * can deliver it timely and perhaps the guest will schedule
10395 		 * another process.  When the instruction that triggered a page
10396 		 * fault is retried, hopefully the page will be ready in the host.
10397 		 */
10398 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
10399 	}
10400 }
10401 
10402 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
10403 				 struct kvm_async_pf *work)
10404 {
10405 	struct x86_exception fault;
10406 	u32 val;
10407 
10408 	if (work->wakeup_all)
10409 		work->arch.token = ~0; /* broadcast wakeup */
10410 	else
10411 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10412 	trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
10413 
10414 	if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
10415 	    !apf_get_user(vcpu, &val)) {
10416 		if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
10417 		    vcpu->arch.exception.pending &&
10418 		    vcpu->arch.exception.nr == PF_VECTOR &&
10419 		    !apf_put_user(vcpu, 0)) {
10420 			vcpu->arch.exception.injected = false;
10421 			vcpu->arch.exception.pending = false;
10422 			vcpu->arch.exception.nr = 0;
10423 			vcpu->arch.exception.has_error_code = false;
10424 			vcpu->arch.exception.error_code = 0;
10425 			vcpu->arch.exception.has_payload = false;
10426 			vcpu->arch.exception.payload = 0;
10427 		} else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
10428 			fault.vector = PF_VECTOR;
10429 			fault.error_code_valid = true;
10430 			fault.error_code = 0;
10431 			fault.nested_page_fault = false;
10432 			fault.address = work->arch.token;
10433 			fault.async_page_fault = true;
10434 			kvm_inject_page_fault(vcpu, &fault);
10435 		}
10436 	}
10437 	vcpu->arch.apf.halted = false;
10438 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10439 }
10440 
10441 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
10442 {
10443 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
10444 		return true;
10445 	else
10446 		return kvm_can_do_async_pf(vcpu);
10447 }
10448 
10449 void kvm_arch_start_assignment(struct kvm *kvm)
10450 {
10451 	atomic_inc(&kvm->arch.assigned_device_count);
10452 }
10453 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
10454 
10455 void kvm_arch_end_assignment(struct kvm *kvm)
10456 {
10457 	atomic_dec(&kvm->arch.assigned_device_count);
10458 }
10459 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
10460 
10461 bool kvm_arch_has_assigned_device(struct kvm *kvm)
10462 {
10463 	return atomic_read(&kvm->arch.assigned_device_count);
10464 }
10465 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
10466 
10467 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
10468 {
10469 	atomic_inc(&kvm->arch.noncoherent_dma_count);
10470 }
10471 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
10472 
10473 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
10474 {
10475 	atomic_dec(&kvm->arch.noncoherent_dma_count);
10476 }
10477 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
10478 
10479 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
10480 {
10481 	return atomic_read(&kvm->arch.noncoherent_dma_count);
10482 }
10483 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
10484 
10485 bool kvm_arch_has_irq_bypass(void)
10486 {
10487 	return true;
10488 }
10489 
10490 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
10491 				      struct irq_bypass_producer *prod)
10492 {
10493 	struct kvm_kernel_irqfd *irqfd =
10494 		container_of(cons, struct kvm_kernel_irqfd, consumer);
10495 
10496 	irqfd->producer = prod;
10497 
10498 	return kvm_x86_ops.update_pi_irte(irqfd->kvm,
10499 					   prod->irq, irqfd->gsi, 1);
10500 }
10501 
10502 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
10503 				      struct irq_bypass_producer *prod)
10504 {
10505 	int ret;
10506 	struct kvm_kernel_irqfd *irqfd =
10507 		container_of(cons, struct kvm_kernel_irqfd, consumer);
10508 
10509 	WARN_ON(irqfd->producer != prod);
10510 	irqfd->producer = NULL;
10511 
10512 	/*
10513 	 * When producer of consumer is unregistered, we change back to
10514 	 * remapped mode, so we can re-use the current implementation
10515 	 * when the irq is masked/disabled or the consumer side (KVM
10516 	 * int this case doesn't want to receive the interrupts.
10517 	*/
10518 	ret = kvm_x86_ops.update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
10519 	if (ret)
10520 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
10521 		       " fails: %d\n", irqfd->consumer.token, ret);
10522 }
10523 
10524 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
10525 				   uint32_t guest_irq, bool set)
10526 {
10527 	return kvm_x86_ops.update_pi_irte(kvm, host_irq, guest_irq, set);
10528 }
10529 
10530 bool kvm_vector_hashing_enabled(void)
10531 {
10532 	return vector_hashing;
10533 }
10534 
10535 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
10536 {
10537 	return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
10538 }
10539 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
10540 
10541 u64 kvm_spec_ctrl_valid_bits(struct kvm_vcpu *vcpu)
10542 {
10543 	uint64_t bits = SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD;
10544 
10545 	/* The STIBP bit doesn't fault even if it's not advertised */
10546 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL) &&
10547 	    !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS))
10548 		bits &= ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP);
10549 	if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL) &&
10550 	    !boot_cpu_has(X86_FEATURE_AMD_IBRS))
10551 		bits &= ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP);
10552 
10553 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL_SSBD) &&
10554 	    !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
10555 		bits &= ~SPEC_CTRL_SSBD;
10556 	if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
10557 	    !boot_cpu_has(X86_FEATURE_AMD_SSBD))
10558 		bits &= ~SPEC_CTRL_SSBD;
10559 
10560 	return bits;
10561 }
10562 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_valid_bits);
10563 
10564 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
10565 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
10566 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
10567 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
10568 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
10569 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
10570 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
10571 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
10572 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
10573 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
10574 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
10575 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
10576 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
10577 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
10578 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
10579 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
10580 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
10581 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
10582 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
10583 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
10584 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
10585 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
10586