xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 6197e5b7)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * derived from drivers/kvm/kvm_main.c
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright (C) 2008 Qumranet, Inc.
9  * Copyright IBM Corporation, 2008
10  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11  *
12  * Authors:
13  *   Avi Kivity   <avi@qumranet.com>
14  *   Yaniv Kamay  <yaniv@qumranet.com>
15  *   Amit Shah    <amit.shah@qumranet.com>
16  *   Ben-Ami Yassour <benami@il.ibm.com>
17  */
18 
19 #include <linux/kvm_host.h>
20 #include "irq.h"
21 #include "ioapic.h"
22 #include "mmu.h"
23 #include "i8254.h"
24 #include "tss.h"
25 #include "kvm_cache_regs.h"
26 #include "kvm_emulate.h"
27 #include "x86.h"
28 #include "cpuid.h"
29 #include "pmu.h"
30 #include "hyperv.h"
31 #include "lapic.h"
32 
33 #include <linux/clocksource.h>
34 #include <linux/interrupt.h>
35 #include <linux/kvm.h>
36 #include <linux/fs.h>
37 #include <linux/vmalloc.h>
38 #include <linux/export.h>
39 #include <linux/moduleparam.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <linux/sched/stat.h>
57 #include <linux/sched/isolation.h>
58 #include <linux/mem_encrypt.h>
59 #include <linux/entry-kvm.h>
60 
61 #include <trace/events/kvm.h>
62 
63 #include <asm/debugreg.h>
64 #include <asm/msr.h>
65 #include <asm/desc.h>
66 #include <asm/mce.h>
67 #include <linux/kernel_stat.h>
68 #include <asm/fpu/internal.h> /* Ugh! */
69 #include <asm/pvclock.h>
70 #include <asm/div64.h>
71 #include <asm/irq_remapping.h>
72 #include <asm/mshyperv.h>
73 #include <asm/hypervisor.h>
74 #include <asm/tlbflush.h>
75 #include <asm/intel_pt.h>
76 #include <asm/emulate_prefix.h>
77 #include <clocksource/hyperv_timer.h>
78 
79 #define CREATE_TRACE_POINTS
80 #include "trace.h"
81 
82 #define MAX_IO_MSRS 256
83 #define KVM_MAX_MCE_BANKS 32
84 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
85 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
86 
87 #define emul_to_vcpu(ctxt) \
88 	((struct kvm_vcpu *)(ctxt)->vcpu)
89 
90 /* EFER defaults:
91  * - enable syscall per default because its emulated by KVM
92  * - enable LME and LMA per default on 64 bit KVM
93  */
94 #ifdef CONFIG_X86_64
95 static
96 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
97 #else
98 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
99 #endif
100 
101 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
102 
103 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
104                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
105 
106 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
107 static void process_nmi(struct kvm_vcpu *vcpu);
108 static void process_smi(struct kvm_vcpu *vcpu);
109 static void enter_smm(struct kvm_vcpu *vcpu);
110 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
111 static void store_regs(struct kvm_vcpu *vcpu);
112 static int sync_regs(struct kvm_vcpu *vcpu);
113 
114 struct kvm_x86_ops kvm_x86_ops __read_mostly;
115 EXPORT_SYMBOL_GPL(kvm_x86_ops);
116 
117 static bool __read_mostly ignore_msrs = 0;
118 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
119 
120 static bool __read_mostly report_ignored_msrs = true;
121 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
122 
123 unsigned int min_timer_period_us = 200;
124 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
125 
126 static bool __read_mostly kvmclock_periodic_sync = true;
127 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
128 
129 bool __read_mostly kvm_has_tsc_control;
130 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
131 u32  __read_mostly kvm_max_guest_tsc_khz;
132 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
133 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
134 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
135 u64  __read_mostly kvm_max_tsc_scaling_ratio;
136 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
137 u64 __read_mostly kvm_default_tsc_scaling_ratio;
138 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
139 
140 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
141 static u32 __read_mostly tsc_tolerance_ppm = 250;
142 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
143 
144 /*
145  * lapic timer advance (tscdeadline mode only) in nanoseconds.  '-1' enables
146  * adaptive tuning starting from default advancment of 1000ns.  '0' disables
147  * advancement entirely.  Any other value is used as-is and disables adaptive
148  * tuning, i.e. allows priveleged userspace to set an exact advancement time.
149  */
150 static int __read_mostly lapic_timer_advance_ns = -1;
151 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
152 
153 static bool __read_mostly vector_hashing = true;
154 module_param(vector_hashing, bool, S_IRUGO);
155 
156 bool __read_mostly enable_vmware_backdoor = false;
157 module_param(enable_vmware_backdoor, bool, S_IRUGO);
158 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
159 
160 static bool __read_mostly force_emulation_prefix = false;
161 module_param(force_emulation_prefix, bool, S_IRUGO);
162 
163 int __read_mostly pi_inject_timer = -1;
164 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
165 
166 /*
167  * Restoring the host value for MSRs that are only consumed when running in
168  * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
169  * returns to userspace, i.e. the kernel can run with the guest's value.
170  */
171 #define KVM_MAX_NR_USER_RETURN_MSRS 16
172 
173 struct kvm_user_return_msrs_global {
174 	int nr;
175 	u32 msrs[KVM_MAX_NR_USER_RETURN_MSRS];
176 };
177 
178 struct kvm_user_return_msrs {
179 	struct user_return_notifier urn;
180 	bool registered;
181 	struct kvm_user_return_msr_values {
182 		u64 host;
183 		u64 curr;
184 	} values[KVM_MAX_NR_USER_RETURN_MSRS];
185 };
186 
187 static struct kvm_user_return_msrs_global __read_mostly user_return_msrs_global;
188 static struct kvm_user_return_msrs __percpu *user_return_msrs;
189 
190 #define KVM_SUPPORTED_XCR0     (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
191 				| XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
192 				| XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
193 				| XFEATURE_MASK_PKRU)
194 
195 u64 __read_mostly host_efer;
196 EXPORT_SYMBOL_GPL(host_efer);
197 
198 bool __read_mostly allow_smaller_maxphyaddr = 0;
199 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
200 
201 u64 __read_mostly host_xss;
202 EXPORT_SYMBOL_GPL(host_xss);
203 u64 __read_mostly supported_xss;
204 EXPORT_SYMBOL_GPL(supported_xss);
205 
206 struct kvm_stats_debugfs_item debugfs_entries[] = {
207 	VCPU_STAT("pf_fixed", pf_fixed),
208 	VCPU_STAT("pf_guest", pf_guest),
209 	VCPU_STAT("tlb_flush", tlb_flush),
210 	VCPU_STAT("invlpg", invlpg),
211 	VCPU_STAT("exits", exits),
212 	VCPU_STAT("io_exits", io_exits),
213 	VCPU_STAT("mmio_exits", mmio_exits),
214 	VCPU_STAT("signal_exits", signal_exits),
215 	VCPU_STAT("irq_window", irq_window_exits),
216 	VCPU_STAT("nmi_window", nmi_window_exits),
217 	VCPU_STAT("halt_exits", halt_exits),
218 	VCPU_STAT("halt_successful_poll", halt_successful_poll),
219 	VCPU_STAT("halt_attempted_poll", halt_attempted_poll),
220 	VCPU_STAT("halt_poll_invalid", halt_poll_invalid),
221 	VCPU_STAT("halt_wakeup", halt_wakeup),
222 	VCPU_STAT("hypercalls", hypercalls),
223 	VCPU_STAT("request_irq", request_irq_exits),
224 	VCPU_STAT("irq_exits", irq_exits),
225 	VCPU_STAT("host_state_reload", host_state_reload),
226 	VCPU_STAT("fpu_reload", fpu_reload),
227 	VCPU_STAT("insn_emulation", insn_emulation),
228 	VCPU_STAT("insn_emulation_fail", insn_emulation_fail),
229 	VCPU_STAT("irq_injections", irq_injections),
230 	VCPU_STAT("nmi_injections", nmi_injections),
231 	VCPU_STAT("req_event", req_event),
232 	VCPU_STAT("l1d_flush", l1d_flush),
233 	VCPU_STAT("halt_poll_success_ns", halt_poll_success_ns),
234 	VCPU_STAT("halt_poll_fail_ns", halt_poll_fail_ns),
235 	VM_STAT("mmu_shadow_zapped", mmu_shadow_zapped),
236 	VM_STAT("mmu_pte_write", mmu_pte_write),
237 	VM_STAT("mmu_pte_updated", mmu_pte_updated),
238 	VM_STAT("mmu_pde_zapped", mmu_pde_zapped),
239 	VM_STAT("mmu_flooded", mmu_flooded),
240 	VM_STAT("mmu_recycled", mmu_recycled),
241 	VM_STAT("mmu_cache_miss", mmu_cache_miss),
242 	VM_STAT("mmu_unsync", mmu_unsync),
243 	VM_STAT("remote_tlb_flush", remote_tlb_flush),
244 	VM_STAT("largepages", lpages, .mode = 0444),
245 	VM_STAT("nx_largepages_splitted", nx_lpage_splits, .mode = 0444),
246 	VM_STAT("max_mmu_page_hash_collisions", max_mmu_page_hash_collisions),
247 	{ NULL }
248 };
249 
250 u64 __read_mostly host_xcr0;
251 u64 __read_mostly supported_xcr0;
252 EXPORT_SYMBOL_GPL(supported_xcr0);
253 
254 static struct kmem_cache *x86_fpu_cache;
255 
256 static struct kmem_cache *x86_emulator_cache;
257 
258 /*
259  * When called, it means the previous get/set msr reached an invalid msr.
260  * Return true if we want to ignore/silent this failed msr access.
261  */
262 static bool kvm_msr_ignored_check(struct kvm_vcpu *vcpu, u32 msr,
263 				  u64 data, bool write)
264 {
265 	const char *op = write ? "wrmsr" : "rdmsr";
266 
267 	if (ignore_msrs) {
268 		if (report_ignored_msrs)
269 			kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
270 				      op, msr, data);
271 		/* Mask the error */
272 		return true;
273 	} else {
274 		kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
275 				      op, msr, data);
276 		return false;
277 	}
278 }
279 
280 static struct kmem_cache *kvm_alloc_emulator_cache(void)
281 {
282 	unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
283 	unsigned int size = sizeof(struct x86_emulate_ctxt);
284 
285 	return kmem_cache_create_usercopy("x86_emulator", size,
286 					  __alignof__(struct x86_emulate_ctxt),
287 					  SLAB_ACCOUNT, useroffset,
288 					  size - useroffset, NULL);
289 }
290 
291 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
292 
293 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
294 {
295 	int i;
296 	for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
297 		vcpu->arch.apf.gfns[i] = ~0;
298 }
299 
300 static void kvm_on_user_return(struct user_return_notifier *urn)
301 {
302 	unsigned slot;
303 	struct kvm_user_return_msrs *msrs
304 		= container_of(urn, struct kvm_user_return_msrs, urn);
305 	struct kvm_user_return_msr_values *values;
306 	unsigned long flags;
307 
308 	/*
309 	 * Disabling irqs at this point since the following code could be
310 	 * interrupted and executed through kvm_arch_hardware_disable()
311 	 */
312 	local_irq_save(flags);
313 	if (msrs->registered) {
314 		msrs->registered = false;
315 		user_return_notifier_unregister(urn);
316 	}
317 	local_irq_restore(flags);
318 	for (slot = 0; slot < user_return_msrs_global.nr; ++slot) {
319 		values = &msrs->values[slot];
320 		if (values->host != values->curr) {
321 			wrmsrl(user_return_msrs_global.msrs[slot], values->host);
322 			values->curr = values->host;
323 		}
324 	}
325 }
326 
327 void kvm_define_user_return_msr(unsigned slot, u32 msr)
328 {
329 	BUG_ON(slot >= KVM_MAX_NR_USER_RETURN_MSRS);
330 	user_return_msrs_global.msrs[slot] = msr;
331 	if (slot >= user_return_msrs_global.nr)
332 		user_return_msrs_global.nr = slot + 1;
333 }
334 EXPORT_SYMBOL_GPL(kvm_define_user_return_msr);
335 
336 static void kvm_user_return_msr_cpu_online(void)
337 {
338 	unsigned int cpu = smp_processor_id();
339 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
340 	u64 value;
341 	int i;
342 
343 	for (i = 0; i < user_return_msrs_global.nr; ++i) {
344 		rdmsrl_safe(user_return_msrs_global.msrs[i], &value);
345 		msrs->values[i].host = value;
346 		msrs->values[i].curr = value;
347 	}
348 }
349 
350 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
351 {
352 	unsigned int cpu = smp_processor_id();
353 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
354 	int err;
355 
356 	value = (value & mask) | (msrs->values[slot].host & ~mask);
357 	if (value == msrs->values[slot].curr)
358 		return 0;
359 	err = wrmsrl_safe(user_return_msrs_global.msrs[slot], value);
360 	if (err)
361 		return 1;
362 
363 	msrs->values[slot].curr = value;
364 	if (!msrs->registered) {
365 		msrs->urn.on_user_return = kvm_on_user_return;
366 		user_return_notifier_register(&msrs->urn);
367 		msrs->registered = true;
368 	}
369 	return 0;
370 }
371 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
372 
373 static void drop_user_return_notifiers(void)
374 {
375 	unsigned int cpu = smp_processor_id();
376 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
377 
378 	if (msrs->registered)
379 		kvm_on_user_return(&msrs->urn);
380 }
381 
382 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
383 {
384 	return vcpu->arch.apic_base;
385 }
386 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
387 
388 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
389 {
390 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
391 }
392 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
393 
394 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
395 {
396 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
397 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
398 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
399 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
400 
401 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
402 		return 1;
403 	if (!msr_info->host_initiated) {
404 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
405 			return 1;
406 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
407 			return 1;
408 	}
409 
410 	kvm_lapic_set_base(vcpu, msr_info->data);
411 	kvm_recalculate_apic_map(vcpu->kvm);
412 	return 0;
413 }
414 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
415 
416 asmlinkage __visible noinstr void kvm_spurious_fault(void)
417 {
418 	/* Fault while not rebooting.  We want the trace. */
419 	BUG_ON(!kvm_rebooting);
420 }
421 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
422 
423 #define EXCPT_BENIGN		0
424 #define EXCPT_CONTRIBUTORY	1
425 #define EXCPT_PF		2
426 
427 static int exception_class(int vector)
428 {
429 	switch (vector) {
430 	case PF_VECTOR:
431 		return EXCPT_PF;
432 	case DE_VECTOR:
433 	case TS_VECTOR:
434 	case NP_VECTOR:
435 	case SS_VECTOR:
436 	case GP_VECTOR:
437 		return EXCPT_CONTRIBUTORY;
438 	default:
439 		break;
440 	}
441 	return EXCPT_BENIGN;
442 }
443 
444 #define EXCPT_FAULT		0
445 #define EXCPT_TRAP		1
446 #define EXCPT_ABORT		2
447 #define EXCPT_INTERRUPT		3
448 
449 static int exception_type(int vector)
450 {
451 	unsigned int mask;
452 
453 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
454 		return EXCPT_INTERRUPT;
455 
456 	mask = 1 << vector;
457 
458 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
459 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
460 		return EXCPT_TRAP;
461 
462 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
463 		return EXCPT_ABORT;
464 
465 	/* Reserved exceptions will result in fault */
466 	return EXCPT_FAULT;
467 }
468 
469 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
470 {
471 	unsigned nr = vcpu->arch.exception.nr;
472 	bool has_payload = vcpu->arch.exception.has_payload;
473 	unsigned long payload = vcpu->arch.exception.payload;
474 
475 	if (!has_payload)
476 		return;
477 
478 	switch (nr) {
479 	case DB_VECTOR:
480 		/*
481 		 * "Certain debug exceptions may clear bit 0-3.  The
482 		 * remaining contents of the DR6 register are never
483 		 * cleared by the processor".
484 		 */
485 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
486 		/*
487 		 * DR6.RTM is set by all #DB exceptions that don't clear it.
488 		 */
489 		vcpu->arch.dr6 |= DR6_RTM;
490 		vcpu->arch.dr6 |= payload;
491 		/*
492 		 * Bit 16 should be set in the payload whenever the #DB
493 		 * exception should clear DR6.RTM. This makes the payload
494 		 * compatible with the pending debug exceptions under VMX.
495 		 * Though not currently documented in the SDM, this also
496 		 * makes the payload compatible with the exit qualification
497 		 * for #DB exceptions under VMX.
498 		 */
499 		vcpu->arch.dr6 ^= payload & DR6_RTM;
500 
501 		/*
502 		 * The #DB payload is defined as compatible with the 'pending
503 		 * debug exceptions' field under VMX, not DR6. While bit 12 is
504 		 * defined in the 'pending debug exceptions' field (enabled
505 		 * breakpoint), it is reserved and must be zero in DR6.
506 		 */
507 		vcpu->arch.dr6 &= ~BIT(12);
508 		break;
509 	case PF_VECTOR:
510 		vcpu->arch.cr2 = payload;
511 		break;
512 	}
513 
514 	vcpu->arch.exception.has_payload = false;
515 	vcpu->arch.exception.payload = 0;
516 }
517 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
518 
519 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
520 		unsigned nr, bool has_error, u32 error_code,
521 	        bool has_payload, unsigned long payload, bool reinject)
522 {
523 	u32 prev_nr;
524 	int class1, class2;
525 
526 	kvm_make_request(KVM_REQ_EVENT, vcpu);
527 
528 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
529 	queue:
530 		if (has_error && !is_protmode(vcpu))
531 			has_error = false;
532 		if (reinject) {
533 			/*
534 			 * On vmentry, vcpu->arch.exception.pending is only
535 			 * true if an event injection was blocked by
536 			 * nested_run_pending.  In that case, however,
537 			 * vcpu_enter_guest requests an immediate exit,
538 			 * and the guest shouldn't proceed far enough to
539 			 * need reinjection.
540 			 */
541 			WARN_ON_ONCE(vcpu->arch.exception.pending);
542 			vcpu->arch.exception.injected = true;
543 			if (WARN_ON_ONCE(has_payload)) {
544 				/*
545 				 * A reinjected event has already
546 				 * delivered its payload.
547 				 */
548 				has_payload = false;
549 				payload = 0;
550 			}
551 		} else {
552 			vcpu->arch.exception.pending = true;
553 			vcpu->arch.exception.injected = false;
554 		}
555 		vcpu->arch.exception.has_error_code = has_error;
556 		vcpu->arch.exception.nr = nr;
557 		vcpu->arch.exception.error_code = error_code;
558 		vcpu->arch.exception.has_payload = has_payload;
559 		vcpu->arch.exception.payload = payload;
560 		if (!is_guest_mode(vcpu))
561 			kvm_deliver_exception_payload(vcpu);
562 		return;
563 	}
564 
565 	/* to check exception */
566 	prev_nr = vcpu->arch.exception.nr;
567 	if (prev_nr == DF_VECTOR) {
568 		/* triple fault -> shutdown */
569 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
570 		return;
571 	}
572 	class1 = exception_class(prev_nr);
573 	class2 = exception_class(nr);
574 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
575 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
576 		/*
577 		 * Generate double fault per SDM Table 5-5.  Set
578 		 * exception.pending = true so that the double fault
579 		 * can trigger a nested vmexit.
580 		 */
581 		vcpu->arch.exception.pending = true;
582 		vcpu->arch.exception.injected = false;
583 		vcpu->arch.exception.has_error_code = true;
584 		vcpu->arch.exception.nr = DF_VECTOR;
585 		vcpu->arch.exception.error_code = 0;
586 		vcpu->arch.exception.has_payload = false;
587 		vcpu->arch.exception.payload = 0;
588 	} else
589 		/* replace previous exception with a new one in a hope
590 		   that instruction re-execution will regenerate lost
591 		   exception */
592 		goto queue;
593 }
594 
595 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
596 {
597 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
598 }
599 EXPORT_SYMBOL_GPL(kvm_queue_exception);
600 
601 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
602 {
603 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
604 }
605 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
606 
607 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
608 			   unsigned long payload)
609 {
610 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
611 }
612 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
613 
614 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
615 				    u32 error_code, unsigned long payload)
616 {
617 	kvm_multiple_exception(vcpu, nr, true, error_code,
618 			       true, payload, false);
619 }
620 
621 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
622 {
623 	if (err)
624 		kvm_inject_gp(vcpu, 0);
625 	else
626 		return kvm_skip_emulated_instruction(vcpu);
627 
628 	return 1;
629 }
630 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
631 
632 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
633 {
634 	++vcpu->stat.pf_guest;
635 	vcpu->arch.exception.nested_apf =
636 		is_guest_mode(vcpu) && fault->async_page_fault;
637 	if (vcpu->arch.exception.nested_apf) {
638 		vcpu->arch.apf.nested_apf_token = fault->address;
639 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
640 	} else {
641 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
642 					fault->address);
643 	}
644 }
645 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
646 
647 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
648 				    struct x86_exception *fault)
649 {
650 	struct kvm_mmu *fault_mmu;
651 	WARN_ON_ONCE(fault->vector != PF_VECTOR);
652 
653 	fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
654 					       vcpu->arch.walk_mmu;
655 
656 	/*
657 	 * Invalidate the TLB entry for the faulting address, if it exists,
658 	 * else the access will fault indefinitely (and to emulate hardware).
659 	 */
660 	if ((fault->error_code & PFERR_PRESENT_MASK) &&
661 	    !(fault->error_code & PFERR_RSVD_MASK))
662 		kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
663 				       fault_mmu->root_hpa);
664 
665 	fault_mmu->inject_page_fault(vcpu, fault);
666 	return fault->nested_page_fault;
667 }
668 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
669 
670 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
671 {
672 	atomic_inc(&vcpu->arch.nmi_queued);
673 	kvm_make_request(KVM_REQ_NMI, vcpu);
674 }
675 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
676 
677 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
678 {
679 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
680 }
681 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
682 
683 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
684 {
685 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
686 }
687 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
688 
689 /*
690  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
691  * a #GP and return false.
692  */
693 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
694 {
695 	if (kvm_x86_ops.get_cpl(vcpu) <= required_cpl)
696 		return true;
697 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
698 	return false;
699 }
700 EXPORT_SYMBOL_GPL(kvm_require_cpl);
701 
702 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
703 {
704 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
705 		return true;
706 
707 	kvm_queue_exception(vcpu, UD_VECTOR);
708 	return false;
709 }
710 EXPORT_SYMBOL_GPL(kvm_require_dr);
711 
712 /*
713  * This function will be used to read from the physical memory of the currently
714  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
715  * can read from guest physical or from the guest's guest physical memory.
716  */
717 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
718 			    gfn_t ngfn, void *data, int offset, int len,
719 			    u32 access)
720 {
721 	struct x86_exception exception;
722 	gfn_t real_gfn;
723 	gpa_t ngpa;
724 
725 	ngpa     = gfn_to_gpa(ngfn);
726 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
727 	if (real_gfn == UNMAPPED_GVA)
728 		return -EFAULT;
729 
730 	real_gfn = gpa_to_gfn(real_gfn);
731 
732 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
733 }
734 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
735 
736 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
737 			       void *data, int offset, int len, u32 access)
738 {
739 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
740 				       data, offset, len, access);
741 }
742 
743 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
744 {
745 	return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
746 	       rsvd_bits(1, 2);
747 }
748 
749 /*
750  * Load the pae pdptrs.  Return 1 if they are all valid, 0 otherwise.
751  */
752 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
753 {
754 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
755 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
756 	int i;
757 	int ret;
758 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
759 
760 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
761 				      offset * sizeof(u64), sizeof(pdpte),
762 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
763 	if (ret < 0) {
764 		ret = 0;
765 		goto out;
766 	}
767 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
768 		if ((pdpte[i] & PT_PRESENT_MASK) &&
769 		    (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
770 			ret = 0;
771 			goto out;
772 		}
773 	}
774 	ret = 1;
775 
776 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
777 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
778 
779 out:
780 
781 	return ret;
782 }
783 EXPORT_SYMBOL_GPL(load_pdptrs);
784 
785 bool pdptrs_changed(struct kvm_vcpu *vcpu)
786 {
787 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
788 	int offset;
789 	gfn_t gfn;
790 	int r;
791 
792 	if (!is_pae_paging(vcpu))
793 		return false;
794 
795 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
796 		return true;
797 
798 	gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
799 	offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
800 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
801 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
802 	if (r < 0)
803 		return true;
804 
805 	return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
806 }
807 EXPORT_SYMBOL_GPL(pdptrs_changed);
808 
809 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
810 {
811 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
812 
813 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
814 		kvm_clear_async_pf_completion_queue(vcpu);
815 		kvm_async_pf_hash_reset(vcpu);
816 	}
817 
818 	if ((cr0 ^ old_cr0) & update_bits)
819 		kvm_mmu_reset_context(vcpu);
820 
821 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
822 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
823 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
824 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
825 }
826 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
827 
828 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
829 {
830 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
831 	unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
832 
833 	cr0 |= X86_CR0_ET;
834 
835 #ifdef CONFIG_X86_64
836 	if (cr0 & 0xffffffff00000000UL)
837 		return 1;
838 #endif
839 
840 	cr0 &= ~CR0_RESERVED_BITS;
841 
842 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
843 		return 1;
844 
845 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
846 		return 1;
847 
848 #ifdef CONFIG_X86_64
849 	if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
850 	    (cr0 & X86_CR0_PG)) {
851 		int cs_db, cs_l;
852 
853 		if (!is_pae(vcpu))
854 			return 1;
855 		kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
856 		if (cs_l)
857 			return 1;
858 	}
859 #endif
860 	if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
861 	    is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
862 	    !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
863 		return 1;
864 
865 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
866 		return 1;
867 
868 	kvm_x86_ops.set_cr0(vcpu, cr0);
869 
870 	kvm_post_set_cr0(vcpu, old_cr0, cr0);
871 
872 	return 0;
873 }
874 EXPORT_SYMBOL_GPL(kvm_set_cr0);
875 
876 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
877 {
878 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
879 }
880 EXPORT_SYMBOL_GPL(kvm_lmsw);
881 
882 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
883 {
884 	if (vcpu->arch.guest_state_protected)
885 		return;
886 
887 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
888 
889 		if (vcpu->arch.xcr0 != host_xcr0)
890 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
891 
892 		if (vcpu->arch.xsaves_enabled &&
893 		    vcpu->arch.ia32_xss != host_xss)
894 			wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
895 	}
896 
897 	if (static_cpu_has(X86_FEATURE_PKU) &&
898 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
899 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
900 	    vcpu->arch.pkru != vcpu->arch.host_pkru)
901 		__write_pkru(vcpu->arch.pkru);
902 }
903 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
904 
905 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
906 {
907 	if (vcpu->arch.guest_state_protected)
908 		return;
909 
910 	if (static_cpu_has(X86_FEATURE_PKU) &&
911 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
912 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
913 		vcpu->arch.pkru = rdpkru();
914 		if (vcpu->arch.pkru != vcpu->arch.host_pkru)
915 			__write_pkru(vcpu->arch.host_pkru);
916 	}
917 
918 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
919 
920 		if (vcpu->arch.xcr0 != host_xcr0)
921 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
922 
923 		if (vcpu->arch.xsaves_enabled &&
924 		    vcpu->arch.ia32_xss != host_xss)
925 			wrmsrl(MSR_IA32_XSS, host_xss);
926 	}
927 
928 }
929 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
930 
931 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
932 {
933 	u64 xcr0 = xcr;
934 	u64 old_xcr0 = vcpu->arch.xcr0;
935 	u64 valid_bits;
936 
937 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
938 	if (index != XCR_XFEATURE_ENABLED_MASK)
939 		return 1;
940 	if (!(xcr0 & XFEATURE_MASK_FP))
941 		return 1;
942 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
943 		return 1;
944 
945 	/*
946 	 * Do not allow the guest to set bits that we do not support
947 	 * saving.  However, xcr0 bit 0 is always set, even if the
948 	 * emulated CPU does not support XSAVE (see fx_init).
949 	 */
950 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
951 	if (xcr0 & ~valid_bits)
952 		return 1;
953 
954 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
955 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
956 		return 1;
957 
958 	if (xcr0 & XFEATURE_MASK_AVX512) {
959 		if (!(xcr0 & XFEATURE_MASK_YMM))
960 			return 1;
961 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
962 			return 1;
963 	}
964 	vcpu->arch.xcr0 = xcr0;
965 
966 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
967 		kvm_update_cpuid_runtime(vcpu);
968 	return 0;
969 }
970 
971 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
972 {
973 	if (kvm_x86_ops.get_cpl(vcpu) != 0 ||
974 	    __kvm_set_xcr(vcpu, index, xcr)) {
975 		kvm_inject_gp(vcpu, 0);
976 		return 1;
977 	}
978 	return 0;
979 }
980 EXPORT_SYMBOL_GPL(kvm_set_xcr);
981 
982 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
983 {
984 	if (cr4 & cr4_reserved_bits)
985 		return false;
986 
987 	if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
988 		return false;
989 
990 	return kvm_x86_ops.is_valid_cr4(vcpu, cr4);
991 }
992 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
993 
994 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
995 {
996 	unsigned long mmu_role_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
997 				      X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
998 
999 	if (((cr4 ^ old_cr4) & mmu_role_bits) ||
1000 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1001 		kvm_mmu_reset_context(vcpu);
1002 }
1003 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1004 
1005 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1006 {
1007 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
1008 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1009 				   X86_CR4_SMEP;
1010 
1011 	if (!kvm_is_valid_cr4(vcpu, cr4))
1012 		return 1;
1013 
1014 	if (is_long_mode(vcpu)) {
1015 		if (!(cr4 & X86_CR4_PAE))
1016 			return 1;
1017 		if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1018 			return 1;
1019 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1020 		   && ((cr4 ^ old_cr4) & pdptr_bits)
1021 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1022 				   kvm_read_cr3(vcpu)))
1023 		return 1;
1024 
1025 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1026 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1027 			return 1;
1028 
1029 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1030 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1031 			return 1;
1032 	}
1033 
1034 	kvm_x86_ops.set_cr4(vcpu, cr4);
1035 
1036 	kvm_post_set_cr4(vcpu, old_cr4, cr4);
1037 
1038 	return 0;
1039 }
1040 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1041 
1042 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1043 {
1044 	bool skip_tlb_flush = false;
1045 #ifdef CONFIG_X86_64
1046 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1047 
1048 	if (pcid_enabled) {
1049 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1050 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
1051 	}
1052 #endif
1053 
1054 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
1055 		if (!skip_tlb_flush) {
1056 			kvm_mmu_sync_roots(vcpu);
1057 			kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1058 		}
1059 		return 0;
1060 	}
1061 
1062 	if (is_long_mode(vcpu) &&
1063 	    (cr3 & vcpu->arch.cr3_lm_rsvd_bits))
1064 		return 1;
1065 	else if (is_pae_paging(vcpu) &&
1066 		 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1067 		return 1;
1068 
1069 	kvm_mmu_new_pgd(vcpu, cr3, skip_tlb_flush, skip_tlb_flush);
1070 	vcpu->arch.cr3 = cr3;
1071 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1072 
1073 	return 0;
1074 }
1075 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1076 
1077 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1078 {
1079 	if (cr8 & CR8_RESERVED_BITS)
1080 		return 1;
1081 	if (lapic_in_kernel(vcpu))
1082 		kvm_lapic_set_tpr(vcpu, cr8);
1083 	else
1084 		vcpu->arch.cr8 = cr8;
1085 	return 0;
1086 }
1087 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1088 
1089 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1090 {
1091 	if (lapic_in_kernel(vcpu))
1092 		return kvm_lapic_get_cr8(vcpu);
1093 	else
1094 		return vcpu->arch.cr8;
1095 }
1096 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1097 
1098 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1099 {
1100 	int i;
1101 
1102 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1103 		for (i = 0; i < KVM_NR_DB_REGS; i++)
1104 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1105 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1106 	}
1107 }
1108 
1109 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1110 {
1111 	unsigned long dr7;
1112 
1113 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1114 		dr7 = vcpu->arch.guest_debug_dr7;
1115 	else
1116 		dr7 = vcpu->arch.dr7;
1117 	kvm_x86_ops.set_dr7(vcpu, dr7);
1118 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1119 	if (dr7 & DR7_BP_EN_MASK)
1120 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1121 }
1122 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1123 
1124 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1125 {
1126 	u64 fixed = DR6_FIXED_1;
1127 
1128 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1129 		fixed |= DR6_RTM;
1130 	return fixed;
1131 }
1132 
1133 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1134 {
1135 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1136 
1137 	switch (dr) {
1138 	case 0 ... 3:
1139 		vcpu->arch.db[array_index_nospec(dr, size)] = val;
1140 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1141 			vcpu->arch.eff_db[dr] = val;
1142 		break;
1143 	case 4:
1144 	case 6:
1145 		if (!kvm_dr6_valid(val))
1146 			return -1; /* #GP */
1147 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1148 		break;
1149 	case 5:
1150 	default: /* 7 */
1151 		if (!kvm_dr7_valid(val))
1152 			return -1; /* #GP */
1153 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1154 		kvm_update_dr7(vcpu);
1155 		break;
1156 	}
1157 
1158 	return 0;
1159 }
1160 
1161 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1162 {
1163 	if (__kvm_set_dr(vcpu, dr, val)) {
1164 		kvm_inject_gp(vcpu, 0);
1165 		return 1;
1166 	}
1167 	return 0;
1168 }
1169 EXPORT_SYMBOL_GPL(kvm_set_dr);
1170 
1171 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1172 {
1173 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1174 
1175 	switch (dr) {
1176 	case 0 ... 3:
1177 		*val = vcpu->arch.db[array_index_nospec(dr, size)];
1178 		break;
1179 	case 4:
1180 	case 6:
1181 		*val = vcpu->arch.dr6;
1182 		break;
1183 	case 5:
1184 	default: /* 7 */
1185 		*val = vcpu->arch.dr7;
1186 		break;
1187 	}
1188 	return 0;
1189 }
1190 EXPORT_SYMBOL_GPL(kvm_get_dr);
1191 
1192 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1193 {
1194 	u32 ecx = kvm_rcx_read(vcpu);
1195 	u64 data;
1196 	int err;
1197 
1198 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1199 	if (err)
1200 		return err;
1201 	kvm_rax_write(vcpu, (u32)data);
1202 	kvm_rdx_write(vcpu, data >> 32);
1203 	return err;
1204 }
1205 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1206 
1207 /*
1208  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1209  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1210  *
1211  * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1212  * extract the supported MSRs from the related const lists.
1213  * msrs_to_save is selected from the msrs_to_save_all to reflect the
1214  * capabilities of the host cpu. This capabilities test skips MSRs that are
1215  * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1216  * may depend on host virtualization features rather than host cpu features.
1217  */
1218 
1219 static const u32 msrs_to_save_all[] = {
1220 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1221 	MSR_STAR,
1222 #ifdef CONFIG_X86_64
1223 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1224 #endif
1225 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1226 	MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1227 	MSR_IA32_SPEC_CTRL,
1228 	MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1229 	MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1230 	MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1231 	MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1232 	MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1233 	MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1234 	MSR_IA32_UMWAIT_CONTROL,
1235 
1236 	MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1237 	MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1238 	MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1239 	MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1240 	MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1241 	MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1242 	MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1243 	MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1244 	MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1245 	MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1246 	MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1247 	MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1248 	MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1249 	MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1250 	MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1251 	MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1252 	MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1253 	MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1254 	MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1255 	MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1256 	MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1257 	MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1258 };
1259 
1260 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1261 static unsigned num_msrs_to_save;
1262 
1263 static const u32 emulated_msrs_all[] = {
1264 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1265 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1266 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1267 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1268 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1269 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1270 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1271 	HV_X64_MSR_RESET,
1272 	HV_X64_MSR_VP_INDEX,
1273 	HV_X64_MSR_VP_RUNTIME,
1274 	HV_X64_MSR_SCONTROL,
1275 	HV_X64_MSR_STIMER0_CONFIG,
1276 	HV_X64_MSR_VP_ASSIST_PAGE,
1277 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1278 	HV_X64_MSR_TSC_EMULATION_STATUS,
1279 	HV_X64_MSR_SYNDBG_OPTIONS,
1280 	HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1281 	HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1282 	HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1283 
1284 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1285 	MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1286 
1287 	MSR_IA32_TSC_ADJUST,
1288 	MSR_IA32_TSCDEADLINE,
1289 	MSR_IA32_ARCH_CAPABILITIES,
1290 	MSR_IA32_PERF_CAPABILITIES,
1291 	MSR_IA32_MISC_ENABLE,
1292 	MSR_IA32_MCG_STATUS,
1293 	MSR_IA32_MCG_CTL,
1294 	MSR_IA32_MCG_EXT_CTL,
1295 	MSR_IA32_SMBASE,
1296 	MSR_SMI_COUNT,
1297 	MSR_PLATFORM_INFO,
1298 	MSR_MISC_FEATURES_ENABLES,
1299 	MSR_AMD64_VIRT_SPEC_CTRL,
1300 	MSR_IA32_POWER_CTL,
1301 	MSR_IA32_UCODE_REV,
1302 
1303 	/*
1304 	 * The following list leaves out MSRs whose values are determined
1305 	 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1306 	 * We always support the "true" VMX control MSRs, even if the host
1307 	 * processor does not, so I am putting these registers here rather
1308 	 * than in msrs_to_save_all.
1309 	 */
1310 	MSR_IA32_VMX_BASIC,
1311 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1312 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1313 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1314 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1315 	MSR_IA32_VMX_MISC,
1316 	MSR_IA32_VMX_CR0_FIXED0,
1317 	MSR_IA32_VMX_CR4_FIXED0,
1318 	MSR_IA32_VMX_VMCS_ENUM,
1319 	MSR_IA32_VMX_PROCBASED_CTLS2,
1320 	MSR_IA32_VMX_EPT_VPID_CAP,
1321 	MSR_IA32_VMX_VMFUNC,
1322 
1323 	MSR_K7_HWCR,
1324 	MSR_KVM_POLL_CONTROL,
1325 };
1326 
1327 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1328 static unsigned num_emulated_msrs;
1329 
1330 /*
1331  * List of msr numbers which are used to expose MSR-based features that
1332  * can be used by a hypervisor to validate requested CPU features.
1333  */
1334 static const u32 msr_based_features_all[] = {
1335 	MSR_IA32_VMX_BASIC,
1336 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1337 	MSR_IA32_VMX_PINBASED_CTLS,
1338 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1339 	MSR_IA32_VMX_PROCBASED_CTLS,
1340 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1341 	MSR_IA32_VMX_EXIT_CTLS,
1342 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1343 	MSR_IA32_VMX_ENTRY_CTLS,
1344 	MSR_IA32_VMX_MISC,
1345 	MSR_IA32_VMX_CR0_FIXED0,
1346 	MSR_IA32_VMX_CR0_FIXED1,
1347 	MSR_IA32_VMX_CR4_FIXED0,
1348 	MSR_IA32_VMX_CR4_FIXED1,
1349 	MSR_IA32_VMX_VMCS_ENUM,
1350 	MSR_IA32_VMX_PROCBASED_CTLS2,
1351 	MSR_IA32_VMX_EPT_VPID_CAP,
1352 	MSR_IA32_VMX_VMFUNC,
1353 
1354 	MSR_F10H_DECFG,
1355 	MSR_IA32_UCODE_REV,
1356 	MSR_IA32_ARCH_CAPABILITIES,
1357 	MSR_IA32_PERF_CAPABILITIES,
1358 };
1359 
1360 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1361 static unsigned int num_msr_based_features;
1362 
1363 static u64 kvm_get_arch_capabilities(void)
1364 {
1365 	u64 data = 0;
1366 
1367 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1368 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1369 
1370 	/*
1371 	 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1372 	 * the nested hypervisor runs with NX huge pages.  If it is not,
1373 	 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1374 	 * L1 guests, so it need not worry about its own (L2) guests.
1375 	 */
1376 	data |= ARCH_CAP_PSCHANGE_MC_NO;
1377 
1378 	/*
1379 	 * If we're doing cache flushes (either "always" or "cond")
1380 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1381 	 * If an outer hypervisor is doing the cache flush for us
1382 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1383 	 * capability to the guest too, and if EPT is disabled we're not
1384 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1385 	 * require a nested hypervisor to do a flush of its own.
1386 	 */
1387 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1388 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1389 
1390 	if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1391 		data |= ARCH_CAP_RDCL_NO;
1392 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1393 		data |= ARCH_CAP_SSB_NO;
1394 	if (!boot_cpu_has_bug(X86_BUG_MDS))
1395 		data |= ARCH_CAP_MDS_NO;
1396 
1397 	if (!boot_cpu_has(X86_FEATURE_RTM)) {
1398 		/*
1399 		 * If RTM=0 because the kernel has disabled TSX, the host might
1400 		 * have TAA_NO or TSX_CTRL.  Clear TAA_NO (the guest sees RTM=0
1401 		 * and therefore knows that there cannot be TAA) but keep
1402 		 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1403 		 * and we want to allow migrating those guests to tsx=off hosts.
1404 		 */
1405 		data &= ~ARCH_CAP_TAA_NO;
1406 	} else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1407 		data |= ARCH_CAP_TAA_NO;
1408 	} else {
1409 		/*
1410 		 * Nothing to do here; we emulate TSX_CTRL if present on the
1411 		 * host so the guest can choose between disabling TSX or
1412 		 * using VERW to clear CPU buffers.
1413 		 */
1414 	}
1415 
1416 	return data;
1417 }
1418 
1419 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1420 {
1421 	switch (msr->index) {
1422 	case MSR_IA32_ARCH_CAPABILITIES:
1423 		msr->data = kvm_get_arch_capabilities();
1424 		break;
1425 	case MSR_IA32_UCODE_REV:
1426 		rdmsrl_safe(msr->index, &msr->data);
1427 		break;
1428 	default:
1429 		return kvm_x86_ops.get_msr_feature(msr);
1430 	}
1431 	return 0;
1432 }
1433 
1434 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1435 {
1436 	struct kvm_msr_entry msr;
1437 	int r;
1438 
1439 	msr.index = index;
1440 	r = kvm_get_msr_feature(&msr);
1441 
1442 	if (r == KVM_MSR_RET_INVALID) {
1443 		/* Unconditionally clear the output for simplicity */
1444 		*data = 0;
1445 		if (kvm_msr_ignored_check(vcpu, index, 0, false))
1446 			r = 0;
1447 	}
1448 
1449 	if (r)
1450 		return r;
1451 
1452 	*data = msr.data;
1453 
1454 	return 0;
1455 }
1456 
1457 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1458 {
1459 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1460 		return false;
1461 
1462 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1463 		return false;
1464 
1465 	if (efer & (EFER_LME | EFER_LMA) &&
1466 	    !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1467 		return false;
1468 
1469 	if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1470 		return false;
1471 
1472 	return true;
1473 
1474 }
1475 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1476 {
1477 	if (efer & efer_reserved_bits)
1478 		return false;
1479 
1480 	return __kvm_valid_efer(vcpu, efer);
1481 }
1482 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1483 
1484 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1485 {
1486 	u64 old_efer = vcpu->arch.efer;
1487 	u64 efer = msr_info->data;
1488 	int r;
1489 
1490 	if (efer & efer_reserved_bits)
1491 		return 1;
1492 
1493 	if (!msr_info->host_initiated) {
1494 		if (!__kvm_valid_efer(vcpu, efer))
1495 			return 1;
1496 
1497 		if (is_paging(vcpu) &&
1498 		    (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1499 			return 1;
1500 	}
1501 
1502 	efer &= ~EFER_LMA;
1503 	efer |= vcpu->arch.efer & EFER_LMA;
1504 
1505 	r = kvm_x86_ops.set_efer(vcpu, efer);
1506 	if (r) {
1507 		WARN_ON(r > 0);
1508 		return r;
1509 	}
1510 
1511 	/* Update reserved bits */
1512 	if ((efer ^ old_efer) & EFER_NX)
1513 		kvm_mmu_reset_context(vcpu);
1514 
1515 	return 0;
1516 }
1517 
1518 void kvm_enable_efer_bits(u64 mask)
1519 {
1520        efer_reserved_bits &= ~mask;
1521 }
1522 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1523 
1524 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1525 {
1526 	struct kvm *kvm = vcpu->kvm;
1527 	struct msr_bitmap_range *ranges = kvm->arch.msr_filter.ranges;
1528 	u32 count = kvm->arch.msr_filter.count;
1529 	u32 i;
1530 	bool r = kvm->arch.msr_filter.default_allow;
1531 	int idx;
1532 
1533 	/* MSR filtering not set up or x2APIC enabled, allow everything */
1534 	if (!count || (index >= 0x800 && index <= 0x8ff))
1535 		return true;
1536 
1537 	/* Prevent collision with set_msr_filter */
1538 	idx = srcu_read_lock(&kvm->srcu);
1539 
1540 	for (i = 0; i < count; i++) {
1541 		u32 start = ranges[i].base;
1542 		u32 end = start + ranges[i].nmsrs;
1543 		u32 flags = ranges[i].flags;
1544 		unsigned long *bitmap = ranges[i].bitmap;
1545 
1546 		if ((index >= start) && (index < end) && (flags & type)) {
1547 			r = !!test_bit(index - start, bitmap);
1548 			break;
1549 		}
1550 	}
1551 
1552 	srcu_read_unlock(&kvm->srcu, idx);
1553 
1554 	return r;
1555 }
1556 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1557 
1558 /*
1559  * Write @data into the MSR specified by @index.  Select MSR specific fault
1560  * checks are bypassed if @host_initiated is %true.
1561  * Returns 0 on success, non-0 otherwise.
1562  * Assumes vcpu_load() was already called.
1563  */
1564 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1565 			 bool host_initiated)
1566 {
1567 	struct msr_data msr;
1568 
1569 	if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1570 		return KVM_MSR_RET_FILTERED;
1571 
1572 	switch (index) {
1573 	case MSR_FS_BASE:
1574 	case MSR_GS_BASE:
1575 	case MSR_KERNEL_GS_BASE:
1576 	case MSR_CSTAR:
1577 	case MSR_LSTAR:
1578 		if (is_noncanonical_address(data, vcpu))
1579 			return 1;
1580 		break;
1581 	case MSR_IA32_SYSENTER_EIP:
1582 	case MSR_IA32_SYSENTER_ESP:
1583 		/*
1584 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1585 		 * non-canonical address is written on Intel but not on
1586 		 * AMD (which ignores the top 32-bits, because it does
1587 		 * not implement 64-bit SYSENTER).
1588 		 *
1589 		 * 64-bit code should hence be able to write a non-canonical
1590 		 * value on AMD.  Making the address canonical ensures that
1591 		 * vmentry does not fail on Intel after writing a non-canonical
1592 		 * value, and that something deterministic happens if the guest
1593 		 * invokes 64-bit SYSENTER.
1594 		 */
1595 		data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1596 	}
1597 
1598 	msr.data = data;
1599 	msr.index = index;
1600 	msr.host_initiated = host_initiated;
1601 
1602 	return kvm_x86_ops.set_msr(vcpu, &msr);
1603 }
1604 
1605 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1606 				     u32 index, u64 data, bool host_initiated)
1607 {
1608 	int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1609 
1610 	if (ret == KVM_MSR_RET_INVALID)
1611 		if (kvm_msr_ignored_check(vcpu, index, data, true))
1612 			ret = 0;
1613 
1614 	return ret;
1615 }
1616 
1617 /*
1618  * Read the MSR specified by @index into @data.  Select MSR specific fault
1619  * checks are bypassed if @host_initiated is %true.
1620  * Returns 0 on success, non-0 otherwise.
1621  * Assumes vcpu_load() was already called.
1622  */
1623 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1624 		  bool host_initiated)
1625 {
1626 	struct msr_data msr;
1627 	int ret;
1628 
1629 	if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1630 		return KVM_MSR_RET_FILTERED;
1631 
1632 	msr.index = index;
1633 	msr.host_initiated = host_initiated;
1634 
1635 	ret = kvm_x86_ops.get_msr(vcpu, &msr);
1636 	if (!ret)
1637 		*data = msr.data;
1638 	return ret;
1639 }
1640 
1641 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1642 				     u32 index, u64 *data, bool host_initiated)
1643 {
1644 	int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1645 
1646 	if (ret == KVM_MSR_RET_INVALID) {
1647 		/* Unconditionally clear *data for simplicity */
1648 		*data = 0;
1649 		if (kvm_msr_ignored_check(vcpu, index, 0, false))
1650 			ret = 0;
1651 	}
1652 
1653 	return ret;
1654 }
1655 
1656 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1657 {
1658 	return kvm_get_msr_ignored_check(vcpu, index, data, false);
1659 }
1660 EXPORT_SYMBOL_GPL(kvm_get_msr);
1661 
1662 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1663 {
1664 	return kvm_set_msr_ignored_check(vcpu, index, data, false);
1665 }
1666 EXPORT_SYMBOL_GPL(kvm_set_msr);
1667 
1668 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1669 {
1670 	int err = vcpu->run->msr.error;
1671 	if (!err) {
1672 		kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1673 		kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1674 	}
1675 
1676 	return kvm_x86_ops.complete_emulated_msr(vcpu, err);
1677 }
1678 
1679 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1680 {
1681 	return kvm_x86_ops.complete_emulated_msr(vcpu, vcpu->run->msr.error);
1682 }
1683 
1684 static u64 kvm_msr_reason(int r)
1685 {
1686 	switch (r) {
1687 	case KVM_MSR_RET_INVALID:
1688 		return KVM_MSR_EXIT_REASON_UNKNOWN;
1689 	case KVM_MSR_RET_FILTERED:
1690 		return KVM_MSR_EXIT_REASON_FILTER;
1691 	default:
1692 		return KVM_MSR_EXIT_REASON_INVAL;
1693 	}
1694 }
1695 
1696 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1697 			      u32 exit_reason, u64 data,
1698 			      int (*completion)(struct kvm_vcpu *vcpu),
1699 			      int r)
1700 {
1701 	u64 msr_reason = kvm_msr_reason(r);
1702 
1703 	/* Check if the user wanted to know about this MSR fault */
1704 	if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1705 		return 0;
1706 
1707 	vcpu->run->exit_reason = exit_reason;
1708 	vcpu->run->msr.error = 0;
1709 	memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1710 	vcpu->run->msr.reason = msr_reason;
1711 	vcpu->run->msr.index = index;
1712 	vcpu->run->msr.data = data;
1713 	vcpu->arch.complete_userspace_io = completion;
1714 
1715 	return 1;
1716 }
1717 
1718 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1719 {
1720 	return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1721 				   complete_emulated_rdmsr, r);
1722 }
1723 
1724 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1725 {
1726 	return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1727 				   complete_emulated_wrmsr, r);
1728 }
1729 
1730 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1731 {
1732 	u32 ecx = kvm_rcx_read(vcpu);
1733 	u64 data;
1734 	int r;
1735 
1736 	r = kvm_get_msr(vcpu, ecx, &data);
1737 
1738 	/* MSR read failed? See if we should ask user space */
1739 	if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1740 		/* Bounce to user space */
1741 		return 0;
1742 	}
1743 
1744 	if (!r) {
1745 		trace_kvm_msr_read(ecx, data);
1746 
1747 		kvm_rax_write(vcpu, data & -1u);
1748 		kvm_rdx_write(vcpu, (data >> 32) & -1u);
1749 	} else {
1750 		trace_kvm_msr_read_ex(ecx);
1751 	}
1752 
1753 	return kvm_x86_ops.complete_emulated_msr(vcpu, r);
1754 }
1755 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1756 
1757 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1758 {
1759 	u32 ecx = kvm_rcx_read(vcpu);
1760 	u64 data = kvm_read_edx_eax(vcpu);
1761 	int r;
1762 
1763 	r = kvm_set_msr(vcpu, ecx, data);
1764 
1765 	/* MSR write failed? See if we should ask user space */
1766 	if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1767 		/* Bounce to user space */
1768 		return 0;
1769 
1770 	/* Signal all other negative errors to userspace */
1771 	if (r < 0)
1772 		return r;
1773 
1774 	if (!r)
1775 		trace_kvm_msr_write(ecx, data);
1776 	else
1777 		trace_kvm_msr_write_ex(ecx, data);
1778 
1779 	return kvm_x86_ops.complete_emulated_msr(vcpu, r);
1780 }
1781 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1782 
1783 bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1784 {
1785 	xfer_to_guest_mode_prepare();
1786 	return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1787 		xfer_to_guest_mode_work_pending();
1788 }
1789 EXPORT_SYMBOL_GPL(kvm_vcpu_exit_request);
1790 
1791 /*
1792  * The fast path for frequent and performance sensitive wrmsr emulation,
1793  * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1794  * the latency of virtual IPI by avoiding the expensive bits of transitioning
1795  * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1796  * other cases which must be called after interrupts are enabled on the host.
1797  */
1798 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1799 {
1800 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1801 		return 1;
1802 
1803 	if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1804 		((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1805 		((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1806 		((u32)(data >> 32) != X2APIC_BROADCAST)) {
1807 
1808 		data &= ~(1 << 12);
1809 		kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1810 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1811 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1812 		trace_kvm_apic_write(APIC_ICR, (u32)data);
1813 		return 0;
1814 	}
1815 
1816 	return 1;
1817 }
1818 
1819 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1820 {
1821 	if (!kvm_can_use_hv_timer(vcpu))
1822 		return 1;
1823 
1824 	kvm_set_lapic_tscdeadline_msr(vcpu, data);
1825 	return 0;
1826 }
1827 
1828 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1829 {
1830 	u32 msr = kvm_rcx_read(vcpu);
1831 	u64 data;
1832 	fastpath_t ret = EXIT_FASTPATH_NONE;
1833 
1834 	switch (msr) {
1835 	case APIC_BASE_MSR + (APIC_ICR >> 4):
1836 		data = kvm_read_edx_eax(vcpu);
1837 		if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1838 			kvm_skip_emulated_instruction(vcpu);
1839 			ret = EXIT_FASTPATH_EXIT_HANDLED;
1840 		}
1841 		break;
1842 	case MSR_IA32_TSCDEADLINE:
1843 		data = kvm_read_edx_eax(vcpu);
1844 		if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1845 			kvm_skip_emulated_instruction(vcpu);
1846 			ret = EXIT_FASTPATH_REENTER_GUEST;
1847 		}
1848 		break;
1849 	default:
1850 		break;
1851 	}
1852 
1853 	if (ret != EXIT_FASTPATH_NONE)
1854 		trace_kvm_msr_write(msr, data);
1855 
1856 	return ret;
1857 }
1858 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1859 
1860 /*
1861  * Adapt set_msr() to msr_io()'s calling convention
1862  */
1863 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1864 {
1865 	return kvm_get_msr_ignored_check(vcpu, index, data, true);
1866 }
1867 
1868 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1869 {
1870 	return kvm_set_msr_ignored_check(vcpu, index, *data, true);
1871 }
1872 
1873 #ifdef CONFIG_X86_64
1874 struct pvclock_clock {
1875 	int vclock_mode;
1876 	u64 cycle_last;
1877 	u64 mask;
1878 	u32 mult;
1879 	u32 shift;
1880 	u64 base_cycles;
1881 	u64 offset;
1882 };
1883 
1884 struct pvclock_gtod_data {
1885 	seqcount_t	seq;
1886 
1887 	struct pvclock_clock clock; /* extract of a clocksource struct */
1888 	struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1889 
1890 	ktime_t		offs_boot;
1891 	u64		wall_time_sec;
1892 };
1893 
1894 static struct pvclock_gtod_data pvclock_gtod_data;
1895 
1896 static void update_pvclock_gtod(struct timekeeper *tk)
1897 {
1898 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1899 
1900 	write_seqcount_begin(&vdata->seq);
1901 
1902 	/* copy pvclock gtod data */
1903 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->vdso_clock_mode;
1904 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1905 	vdata->clock.mask		= tk->tkr_mono.mask;
1906 	vdata->clock.mult		= tk->tkr_mono.mult;
1907 	vdata->clock.shift		= tk->tkr_mono.shift;
1908 	vdata->clock.base_cycles	= tk->tkr_mono.xtime_nsec;
1909 	vdata->clock.offset		= tk->tkr_mono.base;
1910 
1911 	vdata->raw_clock.vclock_mode	= tk->tkr_raw.clock->vdso_clock_mode;
1912 	vdata->raw_clock.cycle_last	= tk->tkr_raw.cycle_last;
1913 	vdata->raw_clock.mask		= tk->tkr_raw.mask;
1914 	vdata->raw_clock.mult		= tk->tkr_raw.mult;
1915 	vdata->raw_clock.shift		= tk->tkr_raw.shift;
1916 	vdata->raw_clock.base_cycles	= tk->tkr_raw.xtime_nsec;
1917 	vdata->raw_clock.offset		= tk->tkr_raw.base;
1918 
1919 	vdata->wall_time_sec            = tk->xtime_sec;
1920 
1921 	vdata->offs_boot		= tk->offs_boot;
1922 
1923 	write_seqcount_end(&vdata->seq);
1924 }
1925 
1926 static s64 get_kvmclock_base_ns(void)
1927 {
1928 	/* Count up from boot time, but with the frequency of the raw clock.  */
1929 	return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
1930 }
1931 #else
1932 static s64 get_kvmclock_base_ns(void)
1933 {
1934 	/* Master clock not used, so we can just use CLOCK_BOOTTIME.  */
1935 	return ktime_get_boottime_ns();
1936 }
1937 #endif
1938 
1939 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1940 {
1941 	int version;
1942 	int r;
1943 	struct pvclock_wall_clock wc;
1944 	u64 wall_nsec;
1945 
1946 	kvm->arch.wall_clock = wall_clock;
1947 
1948 	if (!wall_clock)
1949 		return;
1950 
1951 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1952 	if (r)
1953 		return;
1954 
1955 	if (version & 1)
1956 		++version;  /* first time write, random junk */
1957 
1958 	++version;
1959 
1960 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1961 		return;
1962 
1963 	/*
1964 	 * The guest calculates current wall clock time by adding
1965 	 * system time (updated by kvm_guest_time_update below) to the
1966 	 * wall clock specified here.  We do the reverse here.
1967 	 */
1968 	wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
1969 
1970 	wc.nsec = do_div(wall_nsec, 1000000000);
1971 	wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
1972 	wc.version = version;
1973 
1974 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1975 
1976 	version++;
1977 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1978 }
1979 
1980 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
1981 				  bool old_msr, bool host_initiated)
1982 {
1983 	struct kvm_arch *ka = &vcpu->kvm->arch;
1984 
1985 	if (vcpu->vcpu_id == 0 && !host_initiated) {
1986 		if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
1987 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1988 
1989 		ka->boot_vcpu_runs_old_kvmclock = old_msr;
1990 	}
1991 
1992 	vcpu->arch.time = system_time;
1993 	kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
1994 
1995 	/* we verify if the enable bit is set... */
1996 	vcpu->arch.pv_time_enabled = false;
1997 	if (!(system_time & 1))
1998 		return;
1999 
2000 	if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2001 				       &vcpu->arch.pv_time, system_time & ~1ULL,
2002 				       sizeof(struct pvclock_vcpu_time_info)))
2003 		vcpu->arch.pv_time_enabled = true;
2004 
2005 	return;
2006 }
2007 
2008 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2009 {
2010 	do_shl32_div32(dividend, divisor);
2011 	return dividend;
2012 }
2013 
2014 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2015 			       s8 *pshift, u32 *pmultiplier)
2016 {
2017 	uint64_t scaled64;
2018 	int32_t  shift = 0;
2019 	uint64_t tps64;
2020 	uint32_t tps32;
2021 
2022 	tps64 = base_hz;
2023 	scaled64 = scaled_hz;
2024 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2025 		tps64 >>= 1;
2026 		shift--;
2027 	}
2028 
2029 	tps32 = (uint32_t)tps64;
2030 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2031 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2032 			scaled64 >>= 1;
2033 		else
2034 			tps32 <<= 1;
2035 		shift++;
2036 	}
2037 
2038 	*pshift = shift;
2039 	*pmultiplier = div_frac(scaled64, tps32);
2040 }
2041 
2042 #ifdef CONFIG_X86_64
2043 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2044 #endif
2045 
2046 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2047 static unsigned long max_tsc_khz;
2048 
2049 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2050 {
2051 	u64 v = (u64)khz * (1000000 + ppm);
2052 	do_div(v, 1000000);
2053 	return v;
2054 }
2055 
2056 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2057 {
2058 	u64 ratio;
2059 
2060 	/* Guest TSC same frequency as host TSC? */
2061 	if (!scale) {
2062 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2063 		return 0;
2064 	}
2065 
2066 	/* TSC scaling supported? */
2067 	if (!kvm_has_tsc_control) {
2068 		if (user_tsc_khz > tsc_khz) {
2069 			vcpu->arch.tsc_catchup = 1;
2070 			vcpu->arch.tsc_always_catchup = 1;
2071 			return 0;
2072 		} else {
2073 			pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2074 			return -1;
2075 		}
2076 	}
2077 
2078 	/* TSC scaling required  - calculate ratio */
2079 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2080 				user_tsc_khz, tsc_khz);
2081 
2082 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2083 		pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2084 			            user_tsc_khz);
2085 		return -1;
2086 	}
2087 
2088 	vcpu->arch.tsc_scaling_ratio = ratio;
2089 	return 0;
2090 }
2091 
2092 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2093 {
2094 	u32 thresh_lo, thresh_hi;
2095 	int use_scaling = 0;
2096 
2097 	/* tsc_khz can be zero if TSC calibration fails */
2098 	if (user_tsc_khz == 0) {
2099 		/* set tsc_scaling_ratio to a safe value */
2100 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2101 		return -1;
2102 	}
2103 
2104 	/* Compute a scale to convert nanoseconds in TSC cycles */
2105 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2106 			   &vcpu->arch.virtual_tsc_shift,
2107 			   &vcpu->arch.virtual_tsc_mult);
2108 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2109 
2110 	/*
2111 	 * Compute the variation in TSC rate which is acceptable
2112 	 * within the range of tolerance and decide if the
2113 	 * rate being applied is within that bounds of the hardware
2114 	 * rate.  If so, no scaling or compensation need be done.
2115 	 */
2116 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2117 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2118 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2119 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2120 		use_scaling = 1;
2121 	}
2122 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2123 }
2124 
2125 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2126 {
2127 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2128 				      vcpu->arch.virtual_tsc_mult,
2129 				      vcpu->arch.virtual_tsc_shift);
2130 	tsc += vcpu->arch.this_tsc_write;
2131 	return tsc;
2132 }
2133 
2134 static inline int gtod_is_based_on_tsc(int mode)
2135 {
2136 	return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2137 }
2138 
2139 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2140 {
2141 #ifdef CONFIG_X86_64
2142 	bool vcpus_matched;
2143 	struct kvm_arch *ka = &vcpu->kvm->arch;
2144 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2145 
2146 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2147 			 atomic_read(&vcpu->kvm->online_vcpus));
2148 
2149 	/*
2150 	 * Once the masterclock is enabled, always perform request in
2151 	 * order to update it.
2152 	 *
2153 	 * In order to enable masterclock, the host clocksource must be TSC
2154 	 * and the vcpus need to have matched TSCs.  When that happens,
2155 	 * perform request to enable masterclock.
2156 	 */
2157 	if (ka->use_master_clock ||
2158 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2159 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2160 
2161 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2162 			    atomic_read(&vcpu->kvm->online_vcpus),
2163 		            ka->use_master_clock, gtod->clock.vclock_mode);
2164 #endif
2165 }
2166 
2167 /*
2168  * Multiply tsc by a fixed point number represented by ratio.
2169  *
2170  * The most significant 64-N bits (mult) of ratio represent the
2171  * integral part of the fixed point number; the remaining N bits
2172  * (frac) represent the fractional part, ie. ratio represents a fixed
2173  * point number (mult + frac * 2^(-N)).
2174  *
2175  * N equals to kvm_tsc_scaling_ratio_frac_bits.
2176  */
2177 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2178 {
2179 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2180 }
2181 
2182 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
2183 {
2184 	u64 _tsc = tsc;
2185 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
2186 
2187 	if (ratio != kvm_default_tsc_scaling_ratio)
2188 		_tsc = __scale_tsc(ratio, tsc);
2189 
2190 	return _tsc;
2191 }
2192 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2193 
2194 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2195 {
2196 	u64 tsc;
2197 
2198 	tsc = kvm_scale_tsc(vcpu, rdtsc());
2199 
2200 	return target_tsc - tsc;
2201 }
2202 
2203 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2204 {
2205 	return vcpu->arch.l1_tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
2206 }
2207 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2208 
2209 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2210 {
2211 	vcpu->arch.l1_tsc_offset = offset;
2212 	vcpu->arch.tsc_offset = kvm_x86_ops.write_l1_tsc_offset(vcpu, offset);
2213 }
2214 
2215 static inline bool kvm_check_tsc_unstable(void)
2216 {
2217 #ifdef CONFIG_X86_64
2218 	/*
2219 	 * TSC is marked unstable when we're running on Hyper-V,
2220 	 * 'TSC page' clocksource is good.
2221 	 */
2222 	if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2223 		return false;
2224 #endif
2225 	return check_tsc_unstable();
2226 }
2227 
2228 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2229 {
2230 	struct kvm *kvm = vcpu->kvm;
2231 	u64 offset, ns, elapsed;
2232 	unsigned long flags;
2233 	bool matched;
2234 	bool already_matched;
2235 	bool synchronizing = false;
2236 
2237 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2238 	offset = kvm_compute_tsc_offset(vcpu, data);
2239 	ns = get_kvmclock_base_ns();
2240 	elapsed = ns - kvm->arch.last_tsc_nsec;
2241 
2242 	if (vcpu->arch.virtual_tsc_khz) {
2243 		if (data == 0) {
2244 			/*
2245 			 * detection of vcpu initialization -- need to sync
2246 			 * with other vCPUs. This particularly helps to keep
2247 			 * kvm_clock stable after CPU hotplug
2248 			 */
2249 			synchronizing = true;
2250 		} else {
2251 			u64 tsc_exp = kvm->arch.last_tsc_write +
2252 						nsec_to_cycles(vcpu, elapsed);
2253 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2254 			/*
2255 			 * Special case: TSC write with a small delta (1 second)
2256 			 * of virtual cycle time against real time is
2257 			 * interpreted as an attempt to synchronize the CPU.
2258 			 */
2259 			synchronizing = data < tsc_exp + tsc_hz &&
2260 					data + tsc_hz > tsc_exp;
2261 		}
2262 	}
2263 
2264 	/*
2265 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
2266 	 * TSC, we add elapsed time in this computation.  We could let the
2267 	 * compensation code attempt to catch up if we fall behind, but
2268 	 * it's better to try to match offsets from the beginning.
2269          */
2270 	if (synchronizing &&
2271 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2272 		if (!kvm_check_tsc_unstable()) {
2273 			offset = kvm->arch.cur_tsc_offset;
2274 		} else {
2275 			u64 delta = nsec_to_cycles(vcpu, elapsed);
2276 			data += delta;
2277 			offset = kvm_compute_tsc_offset(vcpu, data);
2278 		}
2279 		matched = true;
2280 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2281 	} else {
2282 		/*
2283 		 * We split periods of matched TSC writes into generations.
2284 		 * For each generation, we track the original measured
2285 		 * nanosecond time, offset, and write, so if TSCs are in
2286 		 * sync, we can match exact offset, and if not, we can match
2287 		 * exact software computation in compute_guest_tsc()
2288 		 *
2289 		 * These values are tracked in kvm->arch.cur_xxx variables.
2290 		 */
2291 		kvm->arch.cur_tsc_generation++;
2292 		kvm->arch.cur_tsc_nsec = ns;
2293 		kvm->arch.cur_tsc_write = data;
2294 		kvm->arch.cur_tsc_offset = offset;
2295 		matched = false;
2296 	}
2297 
2298 	/*
2299 	 * We also track th most recent recorded KHZ, write and time to
2300 	 * allow the matching interval to be extended at each write.
2301 	 */
2302 	kvm->arch.last_tsc_nsec = ns;
2303 	kvm->arch.last_tsc_write = data;
2304 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2305 
2306 	vcpu->arch.last_guest_tsc = data;
2307 
2308 	/* Keep track of which generation this VCPU has synchronized to */
2309 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2310 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2311 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2312 
2313 	kvm_vcpu_write_tsc_offset(vcpu, offset);
2314 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2315 
2316 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
2317 	if (!matched) {
2318 		kvm->arch.nr_vcpus_matched_tsc = 0;
2319 	} else if (!already_matched) {
2320 		kvm->arch.nr_vcpus_matched_tsc++;
2321 	}
2322 
2323 	kvm_track_tsc_matching(vcpu);
2324 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
2325 }
2326 
2327 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2328 					   s64 adjustment)
2329 {
2330 	u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2331 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2332 }
2333 
2334 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2335 {
2336 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2337 		WARN_ON(adjustment < 0);
2338 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2339 	adjust_tsc_offset_guest(vcpu, adjustment);
2340 }
2341 
2342 #ifdef CONFIG_X86_64
2343 
2344 static u64 read_tsc(void)
2345 {
2346 	u64 ret = (u64)rdtsc_ordered();
2347 	u64 last = pvclock_gtod_data.clock.cycle_last;
2348 
2349 	if (likely(ret >= last))
2350 		return ret;
2351 
2352 	/*
2353 	 * GCC likes to generate cmov here, but this branch is extremely
2354 	 * predictable (it's just a function of time and the likely is
2355 	 * very likely) and there's a data dependence, so force GCC
2356 	 * to generate a branch instead.  I don't barrier() because
2357 	 * we don't actually need a barrier, and if this function
2358 	 * ever gets inlined it will generate worse code.
2359 	 */
2360 	asm volatile ("");
2361 	return last;
2362 }
2363 
2364 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2365 			  int *mode)
2366 {
2367 	long v;
2368 	u64 tsc_pg_val;
2369 
2370 	switch (clock->vclock_mode) {
2371 	case VDSO_CLOCKMODE_HVCLOCK:
2372 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2373 						  tsc_timestamp);
2374 		if (tsc_pg_val != U64_MAX) {
2375 			/* TSC page valid */
2376 			*mode = VDSO_CLOCKMODE_HVCLOCK;
2377 			v = (tsc_pg_val - clock->cycle_last) &
2378 				clock->mask;
2379 		} else {
2380 			/* TSC page invalid */
2381 			*mode = VDSO_CLOCKMODE_NONE;
2382 		}
2383 		break;
2384 	case VDSO_CLOCKMODE_TSC:
2385 		*mode = VDSO_CLOCKMODE_TSC;
2386 		*tsc_timestamp = read_tsc();
2387 		v = (*tsc_timestamp - clock->cycle_last) &
2388 			clock->mask;
2389 		break;
2390 	default:
2391 		*mode = VDSO_CLOCKMODE_NONE;
2392 	}
2393 
2394 	if (*mode == VDSO_CLOCKMODE_NONE)
2395 		*tsc_timestamp = v = 0;
2396 
2397 	return v * clock->mult;
2398 }
2399 
2400 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2401 {
2402 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2403 	unsigned long seq;
2404 	int mode;
2405 	u64 ns;
2406 
2407 	do {
2408 		seq = read_seqcount_begin(&gtod->seq);
2409 		ns = gtod->raw_clock.base_cycles;
2410 		ns += vgettsc(&gtod->raw_clock, tsc_timestamp, &mode);
2411 		ns >>= gtod->raw_clock.shift;
2412 		ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2413 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2414 	*t = ns;
2415 
2416 	return mode;
2417 }
2418 
2419 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2420 {
2421 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2422 	unsigned long seq;
2423 	int mode;
2424 	u64 ns;
2425 
2426 	do {
2427 		seq = read_seqcount_begin(&gtod->seq);
2428 		ts->tv_sec = gtod->wall_time_sec;
2429 		ns = gtod->clock.base_cycles;
2430 		ns += vgettsc(&gtod->clock, tsc_timestamp, &mode);
2431 		ns >>= gtod->clock.shift;
2432 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2433 
2434 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2435 	ts->tv_nsec = ns;
2436 
2437 	return mode;
2438 }
2439 
2440 /* returns true if host is using TSC based clocksource */
2441 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2442 {
2443 	/* checked again under seqlock below */
2444 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2445 		return false;
2446 
2447 	return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2448 						      tsc_timestamp));
2449 }
2450 
2451 /* returns true if host is using TSC based clocksource */
2452 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2453 					   u64 *tsc_timestamp)
2454 {
2455 	/* checked again under seqlock below */
2456 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2457 		return false;
2458 
2459 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2460 }
2461 #endif
2462 
2463 /*
2464  *
2465  * Assuming a stable TSC across physical CPUS, and a stable TSC
2466  * across virtual CPUs, the following condition is possible.
2467  * Each numbered line represents an event visible to both
2468  * CPUs at the next numbered event.
2469  *
2470  * "timespecX" represents host monotonic time. "tscX" represents
2471  * RDTSC value.
2472  *
2473  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
2474  *
2475  * 1.  read timespec0,tsc0
2476  * 2.					| timespec1 = timespec0 + N
2477  * 					| tsc1 = tsc0 + M
2478  * 3. transition to guest		| transition to guest
2479  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2480  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
2481  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2482  *
2483  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2484  *
2485  * 	- ret0 < ret1
2486  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2487  *		...
2488  *	- 0 < N - M => M < N
2489  *
2490  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2491  * always the case (the difference between two distinct xtime instances
2492  * might be smaller then the difference between corresponding TSC reads,
2493  * when updating guest vcpus pvclock areas).
2494  *
2495  * To avoid that problem, do not allow visibility of distinct
2496  * system_timestamp/tsc_timestamp values simultaneously: use a master
2497  * copy of host monotonic time values. Update that master copy
2498  * in lockstep.
2499  *
2500  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2501  *
2502  */
2503 
2504 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2505 {
2506 #ifdef CONFIG_X86_64
2507 	struct kvm_arch *ka = &kvm->arch;
2508 	int vclock_mode;
2509 	bool host_tsc_clocksource, vcpus_matched;
2510 
2511 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2512 			atomic_read(&kvm->online_vcpus));
2513 
2514 	/*
2515 	 * If the host uses TSC clock, then passthrough TSC as stable
2516 	 * to the guest.
2517 	 */
2518 	host_tsc_clocksource = kvm_get_time_and_clockread(
2519 					&ka->master_kernel_ns,
2520 					&ka->master_cycle_now);
2521 
2522 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2523 				&& !ka->backwards_tsc_observed
2524 				&& !ka->boot_vcpu_runs_old_kvmclock;
2525 
2526 	if (ka->use_master_clock)
2527 		atomic_set(&kvm_guest_has_master_clock, 1);
2528 
2529 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2530 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2531 					vcpus_matched);
2532 #endif
2533 }
2534 
2535 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2536 {
2537 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2538 }
2539 
2540 static void kvm_gen_update_masterclock(struct kvm *kvm)
2541 {
2542 #ifdef CONFIG_X86_64
2543 	int i;
2544 	struct kvm_vcpu *vcpu;
2545 	struct kvm_arch *ka = &kvm->arch;
2546 
2547 	spin_lock(&ka->pvclock_gtod_sync_lock);
2548 	kvm_make_mclock_inprogress_request(kvm);
2549 	/* no guest entries from this point */
2550 	pvclock_update_vm_gtod_copy(kvm);
2551 
2552 	kvm_for_each_vcpu(i, vcpu, kvm)
2553 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2554 
2555 	/* guest entries allowed */
2556 	kvm_for_each_vcpu(i, vcpu, kvm)
2557 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2558 
2559 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2560 #endif
2561 }
2562 
2563 u64 get_kvmclock_ns(struct kvm *kvm)
2564 {
2565 	struct kvm_arch *ka = &kvm->arch;
2566 	struct pvclock_vcpu_time_info hv_clock;
2567 	u64 ret;
2568 
2569 	spin_lock(&ka->pvclock_gtod_sync_lock);
2570 	if (!ka->use_master_clock) {
2571 		spin_unlock(&ka->pvclock_gtod_sync_lock);
2572 		return get_kvmclock_base_ns() + ka->kvmclock_offset;
2573 	}
2574 
2575 	hv_clock.tsc_timestamp = ka->master_cycle_now;
2576 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2577 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2578 
2579 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2580 	get_cpu();
2581 
2582 	if (__this_cpu_read(cpu_tsc_khz)) {
2583 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2584 				   &hv_clock.tsc_shift,
2585 				   &hv_clock.tsc_to_system_mul);
2586 		ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2587 	} else
2588 		ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2589 
2590 	put_cpu();
2591 
2592 	return ret;
2593 }
2594 
2595 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2596 {
2597 	struct kvm_vcpu_arch *vcpu = &v->arch;
2598 	struct pvclock_vcpu_time_info guest_hv_clock;
2599 
2600 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2601 		&guest_hv_clock, sizeof(guest_hv_clock))))
2602 		return;
2603 
2604 	/* This VCPU is paused, but it's legal for a guest to read another
2605 	 * VCPU's kvmclock, so we really have to follow the specification where
2606 	 * it says that version is odd if data is being modified, and even after
2607 	 * it is consistent.
2608 	 *
2609 	 * Version field updates must be kept separate.  This is because
2610 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2611 	 * writes within a string instruction are weakly ordered.  So there
2612 	 * are three writes overall.
2613 	 *
2614 	 * As a small optimization, only write the version field in the first
2615 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2616 	 * version field is the first in the struct.
2617 	 */
2618 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2619 
2620 	if (guest_hv_clock.version & 1)
2621 		++guest_hv_clock.version;  /* first time write, random junk */
2622 
2623 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2624 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2625 				&vcpu->hv_clock,
2626 				sizeof(vcpu->hv_clock.version));
2627 
2628 	smp_wmb();
2629 
2630 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2631 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2632 
2633 	if (vcpu->pvclock_set_guest_stopped_request) {
2634 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2635 		vcpu->pvclock_set_guest_stopped_request = false;
2636 	}
2637 
2638 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2639 
2640 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2641 				&vcpu->hv_clock,
2642 				sizeof(vcpu->hv_clock));
2643 
2644 	smp_wmb();
2645 
2646 	vcpu->hv_clock.version++;
2647 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2648 				&vcpu->hv_clock,
2649 				sizeof(vcpu->hv_clock.version));
2650 }
2651 
2652 static int kvm_guest_time_update(struct kvm_vcpu *v)
2653 {
2654 	unsigned long flags, tgt_tsc_khz;
2655 	struct kvm_vcpu_arch *vcpu = &v->arch;
2656 	struct kvm_arch *ka = &v->kvm->arch;
2657 	s64 kernel_ns;
2658 	u64 tsc_timestamp, host_tsc;
2659 	u8 pvclock_flags;
2660 	bool use_master_clock;
2661 
2662 	kernel_ns = 0;
2663 	host_tsc = 0;
2664 
2665 	/*
2666 	 * If the host uses TSC clock, then passthrough TSC as stable
2667 	 * to the guest.
2668 	 */
2669 	spin_lock(&ka->pvclock_gtod_sync_lock);
2670 	use_master_clock = ka->use_master_clock;
2671 	if (use_master_clock) {
2672 		host_tsc = ka->master_cycle_now;
2673 		kernel_ns = ka->master_kernel_ns;
2674 	}
2675 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2676 
2677 	/* Keep irq disabled to prevent changes to the clock */
2678 	local_irq_save(flags);
2679 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2680 	if (unlikely(tgt_tsc_khz == 0)) {
2681 		local_irq_restore(flags);
2682 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2683 		return 1;
2684 	}
2685 	if (!use_master_clock) {
2686 		host_tsc = rdtsc();
2687 		kernel_ns = get_kvmclock_base_ns();
2688 	}
2689 
2690 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2691 
2692 	/*
2693 	 * We may have to catch up the TSC to match elapsed wall clock
2694 	 * time for two reasons, even if kvmclock is used.
2695 	 *   1) CPU could have been running below the maximum TSC rate
2696 	 *   2) Broken TSC compensation resets the base at each VCPU
2697 	 *      entry to avoid unknown leaps of TSC even when running
2698 	 *      again on the same CPU.  This may cause apparent elapsed
2699 	 *      time to disappear, and the guest to stand still or run
2700 	 *	very slowly.
2701 	 */
2702 	if (vcpu->tsc_catchup) {
2703 		u64 tsc = compute_guest_tsc(v, kernel_ns);
2704 		if (tsc > tsc_timestamp) {
2705 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2706 			tsc_timestamp = tsc;
2707 		}
2708 	}
2709 
2710 	local_irq_restore(flags);
2711 
2712 	/* With all the info we got, fill in the values */
2713 
2714 	if (kvm_has_tsc_control)
2715 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2716 
2717 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2718 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2719 				   &vcpu->hv_clock.tsc_shift,
2720 				   &vcpu->hv_clock.tsc_to_system_mul);
2721 		vcpu->hw_tsc_khz = tgt_tsc_khz;
2722 	}
2723 
2724 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2725 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2726 	vcpu->last_guest_tsc = tsc_timestamp;
2727 
2728 	/* If the host uses TSC clocksource, then it is stable */
2729 	pvclock_flags = 0;
2730 	if (use_master_clock)
2731 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2732 
2733 	vcpu->hv_clock.flags = pvclock_flags;
2734 
2735 	if (vcpu->pv_time_enabled)
2736 		kvm_setup_pvclock_page(v);
2737 	if (v == kvm_get_vcpu(v->kvm, 0))
2738 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2739 	return 0;
2740 }
2741 
2742 /*
2743  * kvmclock updates which are isolated to a given vcpu, such as
2744  * vcpu->cpu migration, should not allow system_timestamp from
2745  * the rest of the vcpus to remain static. Otherwise ntp frequency
2746  * correction applies to one vcpu's system_timestamp but not
2747  * the others.
2748  *
2749  * So in those cases, request a kvmclock update for all vcpus.
2750  * We need to rate-limit these requests though, as they can
2751  * considerably slow guests that have a large number of vcpus.
2752  * The time for a remote vcpu to update its kvmclock is bound
2753  * by the delay we use to rate-limit the updates.
2754  */
2755 
2756 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2757 
2758 static void kvmclock_update_fn(struct work_struct *work)
2759 {
2760 	int i;
2761 	struct delayed_work *dwork = to_delayed_work(work);
2762 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2763 					   kvmclock_update_work);
2764 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2765 	struct kvm_vcpu *vcpu;
2766 
2767 	kvm_for_each_vcpu(i, vcpu, kvm) {
2768 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2769 		kvm_vcpu_kick(vcpu);
2770 	}
2771 }
2772 
2773 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2774 {
2775 	struct kvm *kvm = v->kvm;
2776 
2777 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2778 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2779 					KVMCLOCK_UPDATE_DELAY);
2780 }
2781 
2782 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2783 
2784 static void kvmclock_sync_fn(struct work_struct *work)
2785 {
2786 	struct delayed_work *dwork = to_delayed_work(work);
2787 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2788 					   kvmclock_sync_work);
2789 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2790 
2791 	if (!kvmclock_periodic_sync)
2792 		return;
2793 
2794 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2795 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2796 					KVMCLOCK_SYNC_PERIOD);
2797 }
2798 
2799 /*
2800  * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2801  */
2802 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2803 {
2804 	/* McStatusWrEn enabled? */
2805 	if (guest_cpuid_is_amd_or_hygon(vcpu))
2806 		return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2807 
2808 	return false;
2809 }
2810 
2811 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2812 {
2813 	u64 mcg_cap = vcpu->arch.mcg_cap;
2814 	unsigned bank_num = mcg_cap & 0xff;
2815 	u32 msr = msr_info->index;
2816 	u64 data = msr_info->data;
2817 
2818 	switch (msr) {
2819 	case MSR_IA32_MCG_STATUS:
2820 		vcpu->arch.mcg_status = data;
2821 		break;
2822 	case MSR_IA32_MCG_CTL:
2823 		if (!(mcg_cap & MCG_CTL_P) &&
2824 		    (data || !msr_info->host_initiated))
2825 			return 1;
2826 		if (data != 0 && data != ~(u64)0)
2827 			return 1;
2828 		vcpu->arch.mcg_ctl = data;
2829 		break;
2830 	default:
2831 		if (msr >= MSR_IA32_MC0_CTL &&
2832 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2833 			u32 offset = array_index_nospec(
2834 				msr - MSR_IA32_MC0_CTL,
2835 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2836 
2837 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2838 			 * some Linux kernels though clear bit 10 in bank 4 to
2839 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2840 			 * this to avoid an uncatched #GP in the guest
2841 			 */
2842 			if ((offset & 0x3) == 0 &&
2843 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2844 				return -1;
2845 
2846 			/* MCi_STATUS */
2847 			if (!msr_info->host_initiated &&
2848 			    (offset & 0x3) == 1 && data != 0) {
2849 				if (!can_set_mci_status(vcpu))
2850 					return -1;
2851 			}
2852 
2853 			vcpu->arch.mce_banks[offset] = data;
2854 			break;
2855 		}
2856 		return 1;
2857 	}
2858 	return 0;
2859 }
2860 
2861 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2862 {
2863 	struct kvm *kvm = vcpu->kvm;
2864 	int lm = is_long_mode(vcpu);
2865 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2866 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2867 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2868 		: kvm->arch.xen_hvm_config.blob_size_32;
2869 	u32 page_num = data & ~PAGE_MASK;
2870 	u64 page_addr = data & PAGE_MASK;
2871 	u8 *page;
2872 
2873 	if (page_num >= blob_size)
2874 		return 1;
2875 
2876 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2877 	if (IS_ERR(page))
2878 		return PTR_ERR(page);
2879 
2880 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE)) {
2881 		kfree(page);
2882 		return 1;
2883 	}
2884 	return 0;
2885 }
2886 
2887 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
2888 {
2889 	u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
2890 
2891 	return (vcpu->arch.apf.msr_en_val & mask) == mask;
2892 }
2893 
2894 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2895 {
2896 	gpa_t gpa = data & ~0x3f;
2897 
2898 	/* Bits 4:5 are reserved, Should be zero */
2899 	if (data & 0x30)
2900 		return 1;
2901 
2902 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
2903 	    (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
2904 		return 1;
2905 
2906 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
2907 	    (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
2908 		return 1;
2909 
2910 	if (!lapic_in_kernel(vcpu))
2911 		return data ? 1 : 0;
2912 
2913 	vcpu->arch.apf.msr_en_val = data;
2914 
2915 	if (!kvm_pv_async_pf_enabled(vcpu)) {
2916 		kvm_clear_async_pf_completion_queue(vcpu);
2917 		kvm_async_pf_hash_reset(vcpu);
2918 		return 0;
2919 	}
2920 
2921 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2922 					sizeof(u64)))
2923 		return 1;
2924 
2925 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2926 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2927 
2928 	kvm_async_pf_wakeup_all(vcpu);
2929 
2930 	return 0;
2931 }
2932 
2933 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
2934 {
2935 	/* Bits 8-63 are reserved */
2936 	if (data >> 8)
2937 		return 1;
2938 
2939 	if (!lapic_in_kernel(vcpu))
2940 		return 1;
2941 
2942 	vcpu->arch.apf.msr_int_val = data;
2943 
2944 	vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
2945 
2946 	return 0;
2947 }
2948 
2949 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2950 {
2951 	vcpu->arch.pv_time_enabled = false;
2952 	vcpu->arch.time = 0;
2953 }
2954 
2955 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
2956 {
2957 	++vcpu->stat.tlb_flush;
2958 	kvm_x86_ops.tlb_flush_all(vcpu);
2959 }
2960 
2961 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
2962 {
2963 	++vcpu->stat.tlb_flush;
2964 	kvm_x86_ops.tlb_flush_guest(vcpu);
2965 }
2966 
2967 static void record_steal_time(struct kvm_vcpu *vcpu)
2968 {
2969 	struct kvm_host_map map;
2970 	struct kvm_steal_time *st;
2971 
2972 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2973 		return;
2974 
2975 	/* -EAGAIN is returned in atomic context so we can just return. */
2976 	if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
2977 			&map, &vcpu->arch.st.cache, false))
2978 		return;
2979 
2980 	st = map.hva +
2981 		offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
2982 
2983 	/*
2984 	 * Doing a TLB flush here, on the guest's behalf, can avoid
2985 	 * expensive IPIs.
2986 	 */
2987 	if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
2988 		trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2989 				       st->preempted & KVM_VCPU_FLUSH_TLB);
2990 		if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
2991 			kvm_vcpu_flush_tlb_guest(vcpu);
2992 	}
2993 
2994 	vcpu->arch.st.preempted = 0;
2995 
2996 	if (st->version & 1)
2997 		st->version += 1;  /* first time write, random junk */
2998 
2999 	st->version += 1;
3000 
3001 	smp_wmb();
3002 
3003 	st->steal += current->sched_info.run_delay -
3004 		vcpu->arch.st.last_steal;
3005 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
3006 
3007 	smp_wmb();
3008 
3009 	st->version += 1;
3010 
3011 	kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
3012 }
3013 
3014 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3015 {
3016 	bool pr = false;
3017 	u32 msr = msr_info->index;
3018 	u64 data = msr_info->data;
3019 
3020 	switch (msr) {
3021 	case MSR_AMD64_NB_CFG:
3022 	case MSR_IA32_UCODE_WRITE:
3023 	case MSR_VM_HSAVE_PA:
3024 	case MSR_AMD64_PATCH_LOADER:
3025 	case MSR_AMD64_BU_CFG2:
3026 	case MSR_AMD64_DC_CFG:
3027 	case MSR_F15H_EX_CFG:
3028 		break;
3029 
3030 	case MSR_IA32_UCODE_REV:
3031 		if (msr_info->host_initiated)
3032 			vcpu->arch.microcode_version = data;
3033 		break;
3034 	case MSR_IA32_ARCH_CAPABILITIES:
3035 		if (!msr_info->host_initiated)
3036 			return 1;
3037 		vcpu->arch.arch_capabilities = data;
3038 		break;
3039 	case MSR_IA32_PERF_CAPABILITIES: {
3040 		struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3041 
3042 		if (!msr_info->host_initiated)
3043 			return 1;
3044 		if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3045 			return 1;
3046 		if (data & ~msr_ent.data)
3047 			return 1;
3048 
3049 		vcpu->arch.perf_capabilities = data;
3050 
3051 		return 0;
3052 		}
3053 	case MSR_EFER:
3054 		return set_efer(vcpu, msr_info);
3055 	case MSR_K7_HWCR:
3056 		data &= ~(u64)0x40;	/* ignore flush filter disable */
3057 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
3058 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
3059 
3060 		/* Handle McStatusWrEn */
3061 		if (data == BIT_ULL(18)) {
3062 			vcpu->arch.msr_hwcr = data;
3063 		} else if (data != 0) {
3064 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3065 				    data);
3066 			return 1;
3067 		}
3068 		break;
3069 	case MSR_FAM10H_MMIO_CONF_BASE:
3070 		if (data != 0) {
3071 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3072 				    "0x%llx\n", data);
3073 			return 1;
3074 		}
3075 		break;
3076 	case MSR_IA32_DEBUGCTLMSR:
3077 		if (!data) {
3078 			/* We support the non-activated case already */
3079 			break;
3080 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
3081 			/* Values other than LBR and BTF are vendor-specific,
3082 			   thus reserved and should throw a #GP */
3083 			return 1;
3084 		} else if (report_ignored_msrs)
3085 			vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
3086 				    __func__, data);
3087 		break;
3088 	case 0x200 ... 0x2ff:
3089 		return kvm_mtrr_set_msr(vcpu, msr, data);
3090 	case MSR_IA32_APICBASE:
3091 		return kvm_set_apic_base(vcpu, msr_info);
3092 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3093 		return kvm_x2apic_msr_write(vcpu, msr, data);
3094 	case MSR_IA32_TSCDEADLINE:
3095 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
3096 		break;
3097 	case MSR_IA32_TSC_ADJUST:
3098 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3099 			if (!msr_info->host_initiated) {
3100 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3101 				adjust_tsc_offset_guest(vcpu, adj);
3102 			}
3103 			vcpu->arch.ia32_tsc_adjust_msr = data;
3104 		}
3105 		break;
3106 	case MSR_IA32_MISC_ENABLE:
3107 		if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3108 		    ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3109 			if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3110 				return 1;
3111 			vcpu->arch.ia32_misc_enable_msr = data;
3112 			kvm_update_cpuid_runtime(vcpu);
3113 		} else {
3114 			vcpu->arch.ia32_misc_enable_msr = data;
3115 		}
3116 		break;
3117 	case MSR_IA32_SMBASE:
3118 		if (!msr_info->host_initiated)
3119 			return 1;
3120 		vcpu->arch.smbase = data;
3121 		break;
3122 	case MSR_IA32_POWER_CTL:
3123 		vcpu->arch.msr_ia32_power_ctl = data;
3124 		break;
3125 	case MSR_IA32_TSC:
3126 		if (msr_info->host_initiated) {
3127 			kvm_synchronize_tsc(vcpu, data);
3128 		} else {
3129 			u64 adj = kvm_compute_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3130 			adjust_tsc_offset_guest(vcpu, adj);
3131 			vcpu->arch.ia32_tsc_adjust_msr += adj;
3132 		}
3133 		break;
3134 	case MSR_IA32_XSS:
3135 		if (!msr_info->host_initiated &&
3136 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3137 			return 1;
3138 		/*
3139 		 * KVM supports exposing PT to the guest, but does not support
3140 		 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3141 		 * XSAVES/XRSTORS to save/restore PT MSRs.
3142 		 */
3143 		if (data & ~supported_xss)
3144 			return 1;
3145 		vcpu->arch.ia32_xss = data;
3146 		break;
3147 	case MSR_SMI_COUNT:
3148 		if (!msr_info->host_initiated)
3149 			return 1;
3150 		vcpu->arch.smi_count = data;
3151 		break;
3152 	case MSR_KVM_WALL_CLOCK_NEW:
3153 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3154 			return 1;
3155 
3156 		kvm_write_wall_clock(vcpu->kvm, data);
3157 		break;
3158 	case MSR_KVM_WALL_CLOCK:
3159 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3160 			return 1;
3161 
3162 		kvm_write_wall_clock(vcpu->kvm, data);
3163 		break;
3164 	case MSR_KVM_SYSTEM_TIME_NEW:
3165 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3166 			return 1;
3167 
3168 		kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3169 		break;
3170 	case MSR_KVM_SYSTEM_TIME:
3171 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3172 			return 1;
3173 
3174 		kvm_write_system_time(vcpu, data, true,  msr_info->host_initiated);
3175 		break;
3176 	case MSR_KVM_ASYNC_PF_EN:
3177 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3178 			return 1;
3179 
3180 		if (kvm_pv_enable_async_pf(vcpu, data))
3181 			return 1;
3182 		break;
3183 	case MSR_KVM_ASYNC_PF_INT:
3184 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3185 			return 1;
3186 
3187 		if (kvm_pv_enable_async_pf_int(vcpu, data))
3188 			return 1;
3189 		break;
3190 	case MSR_KVM_ASYNC_PF_ACK:
3191 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3192 			return 1;
3193 		if (data & 0x1) {
3194 			vcpu->arch.apf.pageready_pending = false;
3195 			kvm_check_async_pf_completion(vcpu);
3196 		}
3197 		break;
3198 	case MSR_KVM_STEAL_TIME:
3199 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3200 			return 1;
3201 
3202 		if (unlikely(!sched_info_on()))
3203 			return 1;
3204 
3205 		if (data & KVM_STEAL_RESERVED_MASK)
3206 			return 1;
3207 
3208 		vcpu->arch.st.msr_val = data;
3209 
3210 		if (!(data & KVM_MSR_ENABLED))
3211 			break;
3212 
3213 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3214 
3215 		break;
3216 	case MSR_KVM_PV_EOI_EN:
3217 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3218 			return 1;
3219 
3220 		if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3221 			return 1;
3222 		break;
3223 
3224 	case MSR_KVM_POLL_CONTROL:
3225 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3226 			return 1;
3227 
3228 		/* only enable bit supported */
3229 		if (data & (-1ULL << 1))
3230 			return 1;
3231 
3232 		vcpu->arch.msr_kvm_poll_control = data;
3233 		break;
3234 
3235 	case MSR_IA32_MCG_CTL:
3236 	case MSR_IA32_MCG_STATUS:
3237 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3238 		return set_msr_mce(vcpu, msr_info);
3239 
3240 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3241 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3242 		pr = true;
3243 		fallthrough;
3244 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3245 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3246 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3247 			return kvm_pmu_set_msr(vcpu, msr_info);
3248 
3249 		if (pr || data != 0)
3250 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3251 				    "0x%x data 0x%llx\n", msr, data);
3252 		break;
3253 	case MSR_K7_CLK_CTL:
3254 		/*
3255 		 * Ignore all writes to this no longer documented MSR.
3256 		 * Writes are only relevant for old K7 processors,
3257 		 * all pre-dating SVM, but a recommended workaround from
3258 		 * AMD for these chips. It is possible to specify the
3259 		 * affected processor models on the command line, hence
3260 		 * the need to ignore the workaround.
3261 		 */
3262 		break;
3263 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3264 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3265 	case HV_X64_MSR_SYNDBG_OPTIONS:
3266 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3267 	case HV_X64_MSR_CRASH_CTL:
3268 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3269 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3270 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3271 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3272 		return kvm_hv_set_msr_common(vcpu, msr, data,
3273 					     msr_info->host_initiated);
3274 	case MSR_IA32_BBL_CR_CTL3:
3275 		/* Drop writes to this legacy MSR -- see rdmsr
3276 		 * counterpart for further detail.
3277 		 */
3278 		if (report_ignored_msrs)
3279 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3280 				msr, data);
3281 		break;
3282 	case MSR_AMD64_OSVW_ID_LENGTH:
3283 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3284 			return 1;
3285 		vcpu->arch.osvw.length = data;
3286 		break;
3287 	case MSR_AMD64_OSVW_STATUS:
3288 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3289 			return 1;
3290 		vcpu->arch.osvw.status = data;
3291 		break;
3292 	case MSR_PLATFORM_INFO:
3293 		if (!msr_info->host_initiated ||
3294 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3295 		     cpuid_fault_enabled(vcpu)))
3296 			return 1;
3297 		vcpu->arch.msr_platform_info = data;
3298 		break;
3299 	case MSR_MISC_FEATURES_ENABLES:
3300 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3301 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3302 		     !supports_cpuid_fault(vcpu)))
3303 			return 1;
3304 		vcpu->arch.msr_misc_features_enables = data;
3305 		break;
3306 	default:
3307 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
3308 			return xen_hvm_config(vcpu, data);
3309 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3310 			return kvm_pmu_set_msr(vcpu, msr_info);
3311 		return KVM_MSR_RET_INVALID;
3312 	}
3313 	return 0;
3314 }
3315 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3316 
3317 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3318 {
3319 	u64 data;
3320 	u64 mcg_cap = vcpu->arch.mcg_cap;
3321 	unsigned bank_num = mcg_cap & 0xff;
3322 
3323 	switch (msr) {
3324 	case MSR_IA32_P5_MC_ADDR:
3325 	case MSR_IA32_P5_MC_TYPE:
3326 		data = 0;
3327 		break;
3328 	case MSR_IA32_MCG_CAP:
3329 		data = vcpu->arch.mcg_cap;
3330 		break;
3331 	case MSR_IA32_MCG_CTL:
3332 		if (!(mcg_cap & MCG_CTL_P) && !host)
3333 			return 1;
3334 		data = vcpu->arch.mcg_ctl;
3335 		break;
3336 	case MSR_IA32_MCG_STATUS:
3337 		data = vcpu->arch.mcg_status;
3338 		break;
3339 	default:
3340 		if (msr >= MSR_IA32_MC0_CTL &&
3341 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
3342 			u32 offset = array_index_nospec(
3343 				msr - MSR_IA32_MC0_CTL,
3344 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3345 
3346 			data = vcpu->arch.mce_banks[offset];
3347 			break;
3348 		}
3349 		return 1;
3350 	}
3351 	*pdata = data;
3352 	return 0;
3353 }
3354 
3355 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3356 {
3357 	switch (msr_info->index) {
3358 	case MSR_IA32_PLATFORM_ID:
3359 	case MSR_IA32_EBL_CR_POWERON:
3360 	case MSR_IA32_DEBUGCTLMSR:
3361 	case MSR_IA32_LASTBRANCHFROMIP:
3362 	case MSR_IA32_LASTBRANCHTOIP:
3363 	case MSR_IA32_LASTINTFROMIP:
3364 	case MSR_IA32_LASTINTTOIP:
3365 	case MSR_K8_SYSCFG:
3366 	case MSR_K8_TSEG_ADDR:
3367 	case MSR_K8_TSEG_MASK:
3368 	case MSR_VM_HSAVE_PA:
3369 	case MSR_K8_INT_PENDING_MSG:
3370 	case MSR_AMD64_NB_CFG:
3371 	case MSR_FAM10H_MMIO_CONF_BASE:
3372 	case MSR_AMD64_BU_CFG2:
3373 	case MSR_IA32_PERF_CTL:
3374 	case MSR_AMD64_DC_CFG:
3375 	case MSR_F15H_EX_CFG:
3376 	/*
3377 	 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3378 	 * limit) MSRs. Just return 0, as we do not want to expose the host
3379 	 * data here. Do not conditionalize this on CPUID, as KVM does not do
3380 	 * so for existing CPU-specific MSRs.
3381 	 */
3382 	case MSR_RAPL_POWER_UNIT:
3383 	case MSR_PP0_ENERGY_STATUS:	/* Power plane 0 (core) */
3384 	case MSR_PP1_ENERGY_STATUS:	/* Power plane 1 (graphics uncore) */
3385 	case MSR_PKG_ENERGY_STATUS:	/* Total package */
3386 	case MSR_DRAM_ENERGY_STATUS:	/* DRAM controller */
3387 		msr_info->data = 0;
3388 		break;
3389 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3390 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3391 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3392 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3393 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3394 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3395 			return kvm_pmu_get_msr(vcpu, msr_info);
3396 		msr_info->data = 0;
3397 		break;
3398 	case MSR_IA32_UCODE_REV:
3399 		msr_info->data = vcpu->arch.microcode_version;
3400 		break;
3401 	case MSR_IA32_ARCH_CAPABILITIES:
3402 		if (!msr_info->host_initiated &&
3403 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3404 			return 1;
3405 		msr_info->data = vcpu->arch.arch_capabilities;
3406 		break;
3407 	case MSR_IA32_PERF_CAPABILITIES:
3408 		if (!msr_info->host_initiated &&
3409 		    !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3410 			return 1;
3411 		msr_info->data = vcpu->arch.perf_capabilities;
3412 		break;
3413 	case MSR_IA32_POWER_CTL:
3414 		msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3415 		break;
3416 	case MSR_IA32_TSC: {
3417 		/*
3418 		 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3419 		 * even when not intercepted. AMD manual doesn't explicitly
3420 		 * state this but appears to behave the same.
3421 		 *
3422 		 * On userspace reads and writes, however, we unconditionally
3423 		 * return L1's TSC value to ensure backwards-compatible
3424 		 * behavior for migration.
3425 		 */
3426 		u64 tsc_offset = msr_info->host_initiated ? vcpu->arch.l1_tsc_offset :
3427 							    vcpu->arch.tsc_offset;
3428 
3429 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + tsc_offset;
3430 		break;
3431 	}
3432 	case MSR_MTRRcap:
3433 	case 0x200 ... 0x2ff:
3434 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3435 	case 0xcd: /* fsb frequency */
3436 		msr_info->data = 3;
3437 		break;
3438 		/*
3439 		 * MSR_EBC_FREQUENCY_ID
3440 		 * Conservative value valid for even the basic CPU models.
3441 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3442 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3443 		 * and 266MHz for model 3, or 4. Set Core Clock
3444 		 * Frequency to System Bus Frequency Ratio to 1 (bits
3445 		 * 31:24) even though these are only valid for CPU
3446 		 * models > 2, however guests may end up dividing or
3447 		 * multiplying by zero otherwise.
3448 		 */
3449 	case MSR_EBC_FREQUENCY_ID:
3450 		msr_info->data = 1 << 24;
3451 		break;
3452 	case MSR_IA32_APICBASE:
3453 		msr_info->data = kvm_get_apic_base(vcpu);
3454 		break;
3455 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3456 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3457 	case MSR_IA32_TSCDEADLINE:
3458 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3459 		break;
3460 	case MSR_IA32_TSC_ADJUST:
3461 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3462 		break;
3463 	case MSR_IA32_MISC_ENABLE:
3464 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3465 		break;
3466 	case MSR_IA32_SMBASE:
3467 		if (!msr_info->host_initiated)
3468 			return 1;
3469 		msr_info->data = vcpu->arch.smbase;
3470 		break;
3471 	case MSR_SMI_COUNT:
3472 		msr_info->data = vcpu->arch.smi_count;
3473 		break;
3474 	case MSR_IA32_PERF_STATUS:
3475 		/* TSC increment by tick */
3476 		msr_info->data = 1000ULL;
3477 		/* CPU multiplier */
3478 		msr_info->data |= (((uint64_t)4ULL) << 40);
3479 		break;
3480 	case MSR_EFER:
3481 		msr_info->data = vcpu->arch.efer;
3482 		break;
3483 	case MSR_KVM_WALL_CLOCK:
3484 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3485 			return 1;
3486 
3487 		msr_info->data = vcpu->kvm->arch.wall_clock;
3488 		break;
3489 	case MSR_KVM_WALL_CLOCK_NEW:
3490 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3491 			return 1;
3492 
3493 		msr_info->data = vcpu->kvm->arch.wall_clock;
3494 		break;
3495 	case MSR_KVM_SYSTEM_TIME:
3496 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3497 			return 1;
3498 
3499 		msr_info->data = vcpu->arch.time;
3500 		break;
3501 	case MSR_KVM_SYSTEM_TIME_NEW:
3502 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3503 			return 1;
3504 
3505 		msr_info->data = vcpu->arch.time;
3506 		break;
3507 	case MSR_KVM_ASYNC_PF_EN:
3508 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3509 			return 1;
3510 
3511 		msr_info->data = vcpu->arch.apf.msr_en_val;
3512 		break;
3513 	case MSR_KVM_ASYNC_PF_INT:
3514 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3515 			return 1;
3516 
3517 		msr_info->data = vcpu->arch.apf.msr_int_val;
3518 		break;
3519 	case MSR_KVM_ASYNC_PF_ACK:
3520 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3521 			return 1;
3522 
3523 		msr_info->data = 0;
3524 		break;
3525 	case MSR_KVM_STEAL_TIME:
3526 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3527 			return 1;
3528 
3529 		msr_info->data = vcpu->arch.st.msr_val;
3530 		break;
3531 	case MSR_KVM_PV_EOI_EN:
3532 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3533 			return 1;
3534 
3535 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
3536 		break;
3537 	case MSR_KVM_POLL_CONTROL:
3538 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3539 			return 1;
3540 
3541 		msr_info->data = vcpu->arch.msr_kvm_poll_control;
3542 		break;
3543 	case MSR_IA32_P5_MC_ADDR:
3544 	case MSR_IA32_P5_MC_TYPE:
3545 	case MSR_IA32_MCG_CAP:
3546 	case MSR_IA32_MCG_CTL:
3547 	case MSR_IA32_MCG_STATUS:
3548 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3549 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3550 				   msr_info->host_initiated);
3551 	case MSR_IA32_XSS:
3552 		if (!msr_info->host_initiated &&
3553 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3554 			return 1;
3555 		msr_info->data = vcpu->arch.ia32_xss;
3556 		break;
3557 	case MSR_K7_CLK_CTL:
3558 		/*
3559 		 * Provide expected ramp-up count for K7. All other
3560 		 * are set to zero, indicating minimum divisors for
3561 		 * every field.
3562 		 *
3563 		 * This prevents guest kernels on AMD host with CPU
3564 		 * type 6, model 8 and higher from exploding due to
3565 		 * the rdmsr failing.
3566 		 */
3567 		msr_info->data = 0x20000000;
3568 		break;
3569 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3570 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3571 	case HV_X64_MSR_SYNDBG_OPTIONS:
3572 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3573 	case HV_X64_MSR_CRASH_CTL:
3574 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3575 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3576 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3577 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3578 		return kvm_hv_get_msr_common(vcpu,
3579 					     msr_info->index, &msr_info->data,
3580 					     msr_info->host_initiated);
3581 	case MSR_IA32_BBL_CR_CTL3:
3582 		/* This legacy MSR exists but isn't fully documented in current
3583 		 * silicon.  It is however accessed by winxp in very narrow
3584 		 * scenarios where it sets bit #19, itself documented as
3585 		 * a "reserved" bit.  Best effort attempt to source coherent
3586 		 * read data here should the balance of the register be
3587 		 * interpreted by the guest:
3588 		 *
3589 		 * L2 cache control register 3: 64GB range, 256KB size,
3590 		 * enabled, latency 0x1, configured
3591 		 */
3592 		msr_info->data = 0xbe702111;
3593 		break;
3594 	case MSR_AMD64_OSVW_ID_LENGTH:
3595 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3596 			return 1;
3597 		msr_info->data = vcpu->arch.osvw.length;
3598 		break;
3599 	case MSR_AMD64_OSVW_STATUS:
3600 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3601 			return 1;
3602 		msr_info->data = vcpu->arch.osvw.status;
3603 		break;
3604 	case MSR_PLATFORM_INFO:
3605 		if (!msr_info->host_initiated &&
3606 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3607 			return 1;
3608 		msr_info->data = vcpu->arch.msr_platform_info;
3609 		break;
3610 	case MSR_MISC_FEATURES_ENABLES:
3611 		msr_info->data = vcpu->arch.msr_misc_features_enables;
3612 		break;
3613 	case MSR_K7_HWCR:
3614 		msr_info->data = vcpu->arch.msr_hwcr;
3615 		break;
3616 	default:
3617 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3618 			return kvm_pmu_get_msr(vcpu, msr_info);
3619 		return KVM_MSR_RET_INVALID;
3620 	}
3621 	return 0;
3622 }
3623 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3624 
3625 /*
3626  * Read or write a bunch of msrs. All parameters are kernel addresses.
3627  *
3628  * @return number of msrs set successfully.
3629  */
3630 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3631 		    struct kvm_msr_entry *entries,
3632 		    int (*do_msr)(struct kvm_vcpu *vcpu,
3633 				  unsigned index, u64 *data))
3634 {
3635 	int i;
3636 
3637 	for (i = 0; i < msrs->nmsrs; ++i)
3638 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
3639 			break;
3640 
3641 	return i;
3642 }
3643 
3644 /*
3645  * Read or write a bunch of msrs. Parameters are user addresses.
3646  *
3647  * @return number of msrs set successfully.
3648  */
3649 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3650 		  int (*do_msr)(struct kvm_vcpu *vcpu,
3651 				unsigned index, u64 *data),
3652 		  int writeback)
3653 {
3654 	struct kvm_msrs msrs;
3655 	struct kvm_msr_entry *entries;
3656 	int r, n;
3657 	unsigned size;
3658 
3659 	r = -EFAULT;
3660 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3661 		goto out;
3662 
3663 	r = -E2BIG;
3664 	if (msrs.nmsrs >= MAX_IO_MSRS)
3665 		goto out;
3666 
3667 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3668 	entries = memdup_user(user_msrs->entries, size);
3669 	if (IS_ERR(entries)) {
3670 		r = PTR_ERR(entries);
3671 		goto out;
3672 	}
3673 
3674 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3675 	if (r < 0)
3676 		goto out_free;
3677 
3678 	r = -EFAULT;
3679 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
3680 		goto out_free;
3681 
3682 	r = n;
3683 
3684 out_free:
3685 	kfree(entries);
3686 out:
3687 	return r;
3688 }
3689 
3690 static inline bool kvm_can_mwait_in_guest(void)
3691 {
3692 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
3693 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
3694 		boot_cpu_has(X86_FEATURE_ARAT);
3695 }
3696 
3697 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
3698 					    struct kvm_cpuid2 __user *cpuid_arg)
3699 {
3700 	struct kvm_cpuid2 cpuid;
3701 	int r;
3702 
3703 	r = -EFAULT;
3704 	if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3705 		return r;
3706 
3707 	r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3708 	if (r)
3709 		return r;
3710 
3711 	r = -EFAULT;
3712 	if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3713 		return r;
3714 
3715 	return 0;
3716 }
3717 
3718 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3719 {
3720 	int r = 0;
3721 
3722 	switch (ext) {
3723 	case KVM_CAP_IRQCHIP:
3724 	case KVM_CAP_HLT:
3725 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3726 	case KVM_CAP_SET_TSS_ADDR:
3727 	case KVM_CAP_EXT_CPUID:
3728 	case KVM_CAP_EXT_EMUL_CPUID:
3729 	case KVM_CAP_CLOCKSOURCE:
3730 	case KVM_CAP_PIT:
3731 	case KVM_CAP_NOP_IO_DELAY:
3732 	case KVM_CAP_MP_STATE:
3733 	case KVM_CAP_SYNC_MMU:
3734 	case KVM_CAP_USER_NMI:
3735 	case KVM_CAP_REINJECT_CONTROL:
3736 	case KVM_CAP_IRQ_INJECT_STATUS:
3737 	case KVM_CAP_IOEVENTFD:
3738 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
3739 	case KVM_CAP_PIT2:
3740 	case KVM_CAP_PIT_STATE2:
3741 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3742 	case KVM_CAP_XEN_HVM:
3743 	case KVM_CAP_VCPU_EVENTS:
3744 	case KVM_CAP_HYPERV:
3745 	case KVM_CAP_HYPERV_VAPIC:
3746 	case KVM_CAP_HYPERV_SPIN:
3747 	case KVM_CAP_HYPERV_SYNIC:
3748 	case KVM_CAP_HYPERV_SYNIC2:
3749 	case KVM_CAP_HYPERV_VP_INDEX:
3750 	case KVM_CAP_HYPERV_EVENTFD:
3751 	case KVM_CAP_HYPERV_TLBFLUSH:
3752 	case KVM_CAP_HYPERV_SEND_IPI:
3753 	case KVM_CAP_HYPERV_CPUID:
3754 	case KVM_CAP_SYS_HYPERV_CPUID:
3755 	case KVM_CAP_PCI_SEGMENT:
3756 	case KVM_CAP_DEBUGREGS:
3757 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
3758 	case KVM_CAP_XSAVE:
3759 	case KVM_CAP_ASYNC_PF:
3760 	case KVM_CAP_ASYNC_PF_INT:
3761 	case KVM_CAP_GET_TSC_KHZ:
3762 	case KVM_CAP_KVMCLOCK_CTRL:
3763 	case KVM_CAP_READONLY_MEM:
3764 	case KVM_CAP_HYPERV_TIME:
3765 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3766 	case KVM_CAP_TSC_DEADLINE_TIMER:
3767 	case KVM_CAP_DISABLE_QUIRKS:
3768 	case KVM_CAP_SET_BOOT_CPU_ID:
3769  	case KVM_CAP_SPLIT_IRQCHIP:
3770 	case KVM_CAP_IMMEDIATE_EXIT:
3771 	case KVM_CAP_PMU_EVENT_FILTER:
3772 	case KVM_CAP_GET_MSR_FEATURES:
3773 	case KVM_CAP_MSR_PLATFORM_INFO:
3774 	case KVM_CAP_EXCEPTION_PAYLOAD:
3775 	case KVM_CAP_SET_GUEST_DEBUG:
3776 	case KVM_CAP_LAST_CPU:
3777 	case KVM_CAP_X86_USER_SPACE_MSR:
3778 	case KVM_CAP_X86_MSR_FILTER:
3779 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
3780 		r = 1;
3781 		break;
3782 	case KVM_CAP_SYNC_REGS:
3783 		r = KVM_SYNC_X86_VALID_FIELDS;
3784 		break;
3785 	case KVM_CAP_ADJUST_CLOCK:
3786 		r = KVM_CLOCK_TSC_STABLE;
3787 		break;
3788 	case KVM_CAP_X86_DISABLE_EXITS:
3789 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3790 		      KVM_X86_DISABLE_EXITS_CSTATE;
3791 		if(kvm_can_mwait_in_guest())
3792 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
3793 		break;
3794 	case KVM_CAP_X86_SMM:
3795 		/* SMBASE is usually relocated above 1M on modern chipsets,
3796 		 * and SMM handlers might indeed rely on 4G segment limits,
3797 		 * so do not report SMM to be available if real mode is
3798 		 * emulated via vm86 mode.  Still, do not go to great lengths
3799 		 * to avoid userspace's usage of the feature, because it is a
3800 		 * fringe case that is not enabled except via specific settings
3801 		 * of the module parameters.
3802 		 */
3803 		r = kvm_x86_ops.has_emulated_msr(kvm, MSR_IA32_SMBASE);
3804 		break;
3805 	case KVM_CAP_VAPIC:
3806 		r = !kvm_x86_ops.cpu_has_accelerated_tpr();
3807 		break;
3808 	case KVM_CAP_NR_VCPUS:
3809 		r = KVM_SOFT_MAX_VCPUS;
3810 		break;
3811 	case KVM_CAP_MAX_VCPUS:
3812 		r = KVM_MAX_VCPUS;
3813 		break;
3814 	case KVM_CAP_MAX_VCPU_ID:
3815 		r = KVM_MAX_VCPU_ID;
3816 		break;
3817 	case KVM_CAP_PV_MMU:	/* obsolete */
3818 		r = 0;
3819 		break;
3820 	case KVM_CAP_MCE:
3821 		r = KVM_MAX_MCE_BANKS;
3822 		break;
3823 	case KVM_CAP_XCRS:
3824 		r = boot_cpu_has(X86_FEATURE_XSAVE);
3825 		break;
3826 	case KVM_CAP_TSC_CONTROL:
3827 		r = kvm_has_tsc_control;
3828 		break;
3829 	case KVM_CAP_X2APIC_API:
3830 		r = KVM_X2APIC_API_VALID_FLAGS;
3831 		break;
3832 	case KVM_CAP_NESTED_STATE:
3833 		r = kvm_x86_ops.nested_ops->get_state ?
3834 			kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
3835 		break;
3836 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3837 		r = kvm_x86_ops.enable_direct_tlbflush != NULL;
3838 		break;
3839 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3840 		r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
3841 		break;
3842 	case KVM_CAP_SMALLER_MAXPHYADDR:
3843 		r = (int) allow_smaller_maxphyaddr;
3844 		break;
3845 	case KVM_CAP_STEAL_TIME:
3846 		r = sched_info_on();
3847 		break;
3848 	default:
3849 		break;
3850 	}
3851 	return r;
3852 
3853 }
3854 
3855 long kvm_arch_dev_ioctl(struct file *filp,
3856 			unsigned int ioctl, unsigned long arg)
3857 {
3858 	void __user *argp = (void __user *)arg;
3859 	long r;
3860 
3861 	switch (ioctl) {
3862 	case KVM_GET_MSR_INDEX_LIST: {
3863 		struct kvm_msr_list __user *user_msr_list = argp;
3864 		struct kvm_msr_list msr_list;
3865 		unsigned n;
3866 
3867 		r = -EFAULT;
3868 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3869 			goto out;
3870 		n = msr_list.nmsrs;
3871 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3872 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3873 			goto out;
3874 		r = -E2BIG;
3875 		if (n < msr_list.nmsrs)
3876 			goto out;
3877 		r = -EFAULT;
3878 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3879 				 num_msrs_to_save * sizeof(u32)))
3880 			goto out;
3881 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3882 				 &emulated_msrs,
3883 				 num_emulated_msrs * sizeof(u32)))
3884 			goto out;
3885 		r = 0;
3886 		break;
3887 	}
3888 	case KVM_GET_SUPPORTED_CPUID:
3889 	case KVM_GET_EMULATED_CPUID: {
3890 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3891 		struct kvm_cpuid2 cpuid;
3892 
3893 		r = -EFAULT;
3894 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3895 			goto out;
3896 
3897 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3898 					    ioctl);
3899 		if (r)
3900 			goto out;
3901 
3902 		r = -EFAULT;
3903 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3904 			goto out;
3905 		r = 0;
3906 		break;
3907 	}
3908 	case KVM_X86_GET_MCE_CAP_SUPPORTED:
3909 		r = -EFAULT;
3910 		if (copy_to_user(argp, &kvm_mce_cap_supported,
3911 				 sizeof(kvm_mce_cap_supported)))
3912 			goto out;
3913 		r = 0;
3914 		break;
3915 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3916 		struct kvm_msr_list __user *user_msr_list = argp;
3917 		struct kvm_msr_list msr_list;
3918 		unsigned int n;
3919 
3920 		r = -EFAULT;
3921 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3922 			goto out;
3923 		n = msr_list.nmsrs;
3924 		msr_list.nmsrs = num_msr_based_features;
3925 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3926 			goto out;
3927 		r = -E2BIG;
3928 		if (n < msr_list.nmsrs)
3929 			goto out;
3930 		r = -EFAULT;
3931 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
3932 				 num_msr_based_features * sizeof(u32)))
3933 			goto out;
3934 		r = 0;
3935 		break;
3936 	}
3937 	case KVM_GET_MSRS:
3938 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
3939 		break;
3940 	case KVM_GET_SUPPORTED_HV_CPUID:
3941 		r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
3942 		break;
3943 	default:
3944 		r = -EINVAL;
3945 		break;
3946 	}
3947 out:
3948 	return r;
3949 }
3950 
3951 static void wbinvd_ipi(void *garbage)
3952 {
3953 	wbinvd();
3954 }
3955 
3956 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3957 {
3958 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3959 }
3960 
3961 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3962 {
3963 	/* Address WBINVD may be executed by guest */
3964 	if (need_emulate_wbinvd(vcpu)) {
3965 		if (kvm_x86_ops.has_wbinvd_exit())
3966 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3967 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3968 			smp_call_function_single(vcpu->cpu,
3969 					wbinvd_ipi, NULL, 1);
3970 	}
3971 
3972 	kvm_x86_ops.vcpu_load(vcpu, cpu);
3973 
3974 	/* Save host pkru register if supported */
3975 	vcpu->arch.host_pkru = read_pkru();
3976 
3977 	/* Apply any externally detected TSC adjustments (due to suspend) */
3978 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3979 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3980 		vcpu->arch.tsc_offset_adjustment = 0;
3981 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3982 	}
3983 
3984 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3985 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3986 				rdtsc() - vcpu->arch.last_host_tsc;
3987 		if (tsc_delta < 0)
3988 			mark_tsc_unstable("KVM discovered backwards TSC");
3989 
3990 		if (kvm_check_tsc_unstable()) {
3991 			u64 offset = kvm_compute_tsc_offset(vcpu,
3992 						vcpu->arch.last_guest_tsc);
3993 			kvm_vcpu_write_tsc_offset(vcpu, offset);
3994 			vcpu->arch.tsc_catchup = 1;
3995 		}
3996 
3997 		if (kvm_lapic_hv_timer_in_use(vcpu))
3998 			kvm_lapic_restart_hv_timer(vcpu);
3999 
4000 		/*
4001 		 * On a host with synchronized TSC, there is no need to update
4002 		 * kvmclock on vcpu->cpu migration
4003 		 */
4004 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4005 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4006 		if (vcpu->cpu != cpu)
4007 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4008 		vcpu->cpu = cpu;
4009 	}
4010 
4011 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4012 }
4013 
4014 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4015 {
4016 	struct kvm_host_map map;
4017 	struct kvm_steal_time *st;
4018 
4019 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4020 		return;
4021 
4022 	if (vcpu->arch.st.preempted)
4023 		return;
4024 
4025 	if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
4026 			&vcpu->arch.st.cache, true))
4027 		return;
4028 
4029 	st = map.hva +
4030 		offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
4031 
4032 	st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4033 
4034 	kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
4035 }
4036 
4037 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4038 {
4039 	int idx;
4040 
4041 	if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4042 		vcpu->arch.preempted_in_kernel = !kvm_x86_ops.get_cpl(vcpu);
4043 
4044 	/*
4045 	 * Disable page faults because we're in atomic context here.
4046 	 * kvm_write_guest_offset_cached() would call might_fault()
4047 	 * that relies on pagefault_disable() to tell if there's a
4048 	 * bug. NOTE: the write to guest memory may not go through if
4049 	 * during postcopy live migration or if there's heavy guest
4050 	 * paging.
4051 	 */
4052 	pagefault_disable();
4053 	/*
4054 	 * kvm_memslots() will be called by
4055 	 * kvm_write_guest_offset_cached() so take the srcu lock.
4056 	 */
4057 	idx = srcu_read_lock(&vcpu->kvm->srcu);
4058 	kvm_steal_time_set_preempted(vcpu);
4059 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
4060 	pagefault_enable();
4061 	kvm_x86_ops.vcpu_put(vcpu);
4062 	vcpu->arch.last_host_tsc = rdtsc();
4063 	/*
4064 	 * If userspace has set any breakpoints or watchpoints, dr6 is restored
4065 	 * on every vmexit, but if not, we might have a stale dr6 from the
4066 	 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
4067 	 */
4068 	set_debugreg(0, 6);
4069 }
4070 
4071 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4072 				    struct kvm_lapic_state *s)
4073 {
4074 	if (vcpu->arch.apicv_active)
4075 		kvm_x86_ops.sync_pir_to_irr(vcpu);
4076 
4077 	return kvm_apic_get_state(vcpu, s);
4078 }
4079 
4080 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4081 				    struct kvm_lapic_state *s)
4082 {
4083 	int r;
4084 
4085 	r = kvm_apic_set_state(vcpu, s);
4086 	if (r)
4087 		return r;
4088 	update_cr8_intercept(vcpu);
4089 
4090 	return 0;
4091 }
4092 
4093 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4094 {
4095 	/*
4096 	 * We can accept userspace's request for interrupt injection
4097 	 * as long as we have a place to store the interrupt number.
4098 	 * The actual injection will happen when the CPU is able to
4099 	 * deliver the interrupt.
4100 	 */
4101 	if (kvm_cpu_has_extint(vcpu))
4102 		return false;
4103 
4104 	/* Acknowledging ExtINT does not happen if LINT0 is masked.  */
4105 	return (!lapic_in_kernel(vcpu) ||
4106 		kvm_apic_accept_pic_intr(vcpu));
4107 }
4108 
4109 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4110 {
4111 	return kvm_arch_interrupt_allowed(vcpu) &&
4112 		kvm_cpu_accept_dm_intr(vcpu);
4113 }
4114 
4115 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4116 				    struct kvm_interrupt *irq)
4117 {
4118 	if (irq->irq >= KVM_NR_INTERRUPTS)
4119 		return -EINVAL;
4120 
4121 	if (!irqchip_in_kernel(vcpu->kvm)) {
4122 		kvm_queue_interrupt(vcpu, irq->irq, false);
4123 		kvm_make_request(KVM_REQ_EVENT, vcpu);
4124 		return 0;
4125 	}
4126 
4127 	/*
4128 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4129 	 * fail for in-kernel 8259.
4130 	 */
4131 	if (pic_in_kernel(vcpu->kvm))
4132 		return -ENXIO;
4133 
4134 	if (vcpu->arch.pending_external_vector != -1)
4135 		return -EEXIST;
4136 
4137 	vcpu->arch.pending_external_vector = irq->irq;
4138 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4139 	return 0;
4140 }
4141 
4142 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4143 {
4144 	kvm_inject_nmi(vcpu);
4145 
4146 	return 0;
4147 }
4148 
4149 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4150 {
4151 	kvm_make_request(KVM_REQ_SMI, vcpu);
4152 
4153 	return 0;
4154 }
4155 
4156 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4157 					   struct kvm_tpr_access_ctl *tac)
4158 {
4159 	if (tac->flags)
4160 		return -EINVAL;
4161 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
4162 	return 0;
4163 }
4164 
4165 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4166 					u64 mcg_cap)
4167 {
4168 	int r;
4169 	unsigned bank_num = mcg_cap & 0xff, bank;
4170 
4171 	r = -EINVAL;
4172 	if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4173 		goto out;
4174 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4175 		goto out;
4176 	r = 0;
4177 	vcpu->arch.mcg_cap = mcg_cap;
4178 	/* Init IA32_MCG_CTL to all 1s */
4179 	if (mcg_cap & MCG_CTL_P)
4180 		vcpu->arch.mcg_ctl = ~(u64)0;
4181 	/* Init IA32_MCi_CTL to all 1s */
4182 	for (bank = 0; bank < bank_num; bank++)
4183 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4184 
4185 	kvm_x86_ops.setup_mce(vcpu);
4186 out:
4187 	return r;
4188 }
4189 
4190 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4191 				      struct kvm_x86_mce *mce)
4192 {
4193 	u64 mcg_cap = vcpu->arch.mcg_cap;
4194 	unsigned bank_num = mcg_cap & 0xff;
4195 	u64 *banks = vcpu->arch.mce_banks;
4196 
4197 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4198 		return -EINVAL;
4199 	/*
4200 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4201 	 * reporting is disabled
4202 	 */
4203 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4204 	    vcpu->arch.mcg_ctl != ~(u64)0)
4205 		return 0;
4206 	banks += 4 * mce->bank;
4207 	/*
4208 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4209 	 * reporting is disabled for the bank
4210 	 */
4211 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4212 		return 0;
4213 	if (mce->status & MCI_STATUS_UC) {
4214 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4215 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4216 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4217 			return 0;
4218 		}
4219 		if (banks[1] & MCI_STATUS_VAL)
4220 			mce->status |= MCI_STATUS_OVER;
4221 		banks[2] = mce->addr;
4222 		banks[3] = mce->misc;
4223 		vcpu->arch.mcg_status = mce->mcg_status;
4224 		banks[1] = mce->status;
4225 		kvm_queue_exception(vcpu, MC_VECTOR);
4226 	} else if (!(banks[1] & MCI_STATUS_VAL)
4227 		   || !(banks[1] & MCI_STATUS_UC)) {
4228 		if (banks[1] & MCI_STATUS_VAL)
4229 			mce->status |= MCI_STATUS_OVER;
4230 		banks[2] = mce->addr;
4231 		banks[3] = mce->misc;
4232 		banks[1] = mce->status;
4233 	} else
4234 		banks[1] |= MCI_STATUS_OVER;
4235 	return 0;
4236 }
4237 
4238 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4239 					       struct kvm_vcpu_events *events)
4240 {
4241 	process_nmi(vcpu);
4242 
4243 	if (kvm_check_request(KVM_REQ_SMI, vcpu))
4244 		process_smi(vcpu);
4245 
4246 	/*
4247 	 * In guest mode, payload delivery should be deferred,
4248 	 * so that the L1 hypervisor can intercept #PF before
4249 	 * CR2 is modified (or intercept #DB before DR6 is
4250 	 * modified under nVMX). Unless the per-VM capability,
4251 	 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4252 	 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4253 	 * opportunistically defer the exception payload, deliver it if the
4254 	 * capability hasn't been requested before processing a
4255 	 * KVM_GET_VCPU_EVENTS.
4256 	 */
4257 	if (!vcpu->kvm->arch.exception_payload_enabled &&
4258 	    vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4259 		kvm_deliver_exception_payload(vcpu);
4260 
4261 	/*
4262 	 * The API doesn't provide the instruction length for software
4263 	 * exceptions, so don't report them. As long as the guest RIP
4264 	 * isn't advanced, we should expect to encounter the exception
4265 	 * again.
4266 	 */
4267 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4268 		events->exception.injected = 0;
4269 		events->exception.pending = 0;
4270 	} else {
4271 		events->exception.injected = vcpu->arch.exception.injected;
4272 		events->exception.pending = vcpu->arch.exception.pending;
4273 		/*
4274 		 * For ABI compatibility, deliberately conflate
4275 		 * pending and injected exceptions when
4276 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4277 		 */
4278 		if (!vcpu->kvm->arch.exception_payload_enabled)
4279 			events->exception.injected |=
4280 				vcpu->arch.exception.pending;
4281 	}
4282 	events->exception.nr = vcpu->arch.exception.nr;
4283 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4284 	events->exception.error_code = vcpu->arch.exception.error_code;
4285 	events->exception_has_payload = vcpu->arch.exception.has_payload;
4286 	events->exception_payload = vcpu->arch.exception.payload;
4287 
4288 	events->interrupt.injected =
4289 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4290 	events->interrupt.nr = vcpu->arch.interrupt.nr;
4291 	events->interrupt.soft = 0;
4292 	events->interrupt.shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
4293 
4294 	events->nmi.injected = vcpu->arch.nmi_injected;
4295 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
4296 	events->nmi.masked = kvm_x86_ops.get_nmi_mask(vcpu);
4297 	events->nmi.pad = 0;
4298 
4299 	events->sipi_vector = 0; /* never valid when reporting to user space */
4300 
4301 	events->smi.smm = is_smm(vcpu);
4302 	events->smi.pending = vcpu->arch.smi_pending;
4303 	events->smi.smm_inside_nmi =
4304 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4305 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4306 
4307 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4308 			 | KVM_VCPUEVENT_VALID_SHADOW
4309 			 | KVM_VCPUEVENT_VALID_SMM);
4310 	if (vcpu->kvm->arch.exception_payload_enabled)
4311 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4312 
4313 	memset(&events->reserved, 0, sizeof(events->reserved));
4314 }
4315 
4316 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
4317 
4318 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4319 					      struct kvm_vcpu_events *events)
4320 {
4321 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4322 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4323 			      | KVM_VCPUEVENT_VALID_SHADOW
4324 			      | KVM_VCPUEVENT_VALID_SMM
4325 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
4326 		return -EINVAL;
4327 
4328 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4329 		if (!vcpu->kvm->arch.exception_payload_enabled)
4330 			return -EINVAL;
4331 		if (events->exception.pending)
4332 			events->exception.injected = 0;
4333 		else
4334 			events->exception_has_payload = 0;
4335 	} else {
4336 		events->exception.pending = 0;
4337 		events->exception_has_payload = 0;
4338 	}
4339 
4340 	if ((events->exception.injected || events->exception.pending) &&
4341 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4342 		return -EINVAL;
4343 
4344 	/* INITs are latched while in SMM */
4345 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4346 	    (events->smi.smm || events->smi.pending) &&
4347 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4348 		return -EINVAL;
4349 
4350 	process_nmi(vcpu);
4351 	vcpu->arch.exception.injected = events->exception.injected;
4352 	vcpu->arch.exception.pending = events->exception.pending;
4353 	vcpu->arch.exception.nr = events->exception.nr;
4354 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4355 	vcpu->arch.exception.error_code = events->exception.error_code;
4356 	vcpu->arch.exception.has_payload = events->exception_has_payload;
4357 	vcpu->arch.exception.payload = events->exception_payload;
4358 
4359 	vcpu->arch.interrupt.injected = events->interrupt.injected;
4360 	vcpu->arch.interrupt.nr = events->interrupt.nr;
4361 	vcpu->arch.interrupt.soft = events->interrupt.soft;
4362 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4363 		kvm_x86_ops.set_interrupt_shadow(vcpu,
4364 						  events->interrupt.shadow);
4365 
4366 	vcpu->arch.nmi_injected = events->nmi.injected;
4367 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4368 		vcpu->arch.nmi_pending = events->nmi.pending;
4369 	kvm_x86_ops.set_nmi_mask(vcpu, events->nmi.masked);
4370 
4371 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4372 	    lapic_in_kernel(vcpu))
4373 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
4374 
4375 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4376 		if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
4377 			if (events->smi.smm)
4378 				vcpu->arch.hflags |= HF_SMM_MASK;
4379 			else
4380 				vcpu->arch.hflags &= ~HF_SMM_MASK;
4381 			kvm_smm_changed(vcpu);
4382 		}
4383 
4384 		vcpu->arch.smi_pending = events->smi.pending;
4385 
4386 		if (events->smi.smm) {
4387 			if (events->smi.smm_inside_nmi)
4388 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4389 			else
4390 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4391 		}
4392 
4393 		if (lapic_in_kernel(vcpu)) {
4394 			if (events->smi.latched_init)
4395 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4396 			else
4397 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4398 		}
4399 	}
4400 
4401 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4402 
4403 	return 0;
4404 }
4405 
4406 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4407 					     struct kvm_debugregs *dbgregs)
4408 {
4409 	unsigned long val;
4410 
4411 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4412 	kvm_get_dr(vcpu, 6, &val);
4413 	dbgregs->dr6 = val;
4414 	dbgregs->dr7 = vcpu->arch.dr7;
4415 	dbgregs->flags = 0;
4416 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4417 }
4418 
4419 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4420 					    struct kvm_debugregs *dbgregs)
4421 {
4422 	if (dbgregs->flags)
4423 		return -EINVAL;
4424 
4425 	if (dbgregs->dr6 & ~0xffffffffull)
4426 		return -EINVAL;
4427 	if (dbgregs->dr7 & ~0xffffffffull)
4428 		return -EINVAL;
4429 
4430 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4431 	kvm_update_dr0123(vcpu);
4432 	vcpu->arch.dr6 = dbgregs->dr6;
4433 	vcpu->arch.dr7 = dbgregs->dr7;
4434 	kvm_update_dr7(vcpu);
4435 
4436 	return 0;
4437 }
4438 
4439 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4440 
4441 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4442 {
4443 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4444 	u64 xstate_bv = xsave->header.xfeatures;
4445 	u64 valid;
4446 
4447 	/*
4448 	 * Copy legacy XSAVE area, to avoid complications with CPUID
4449 	 * leaves 0 and 1 in the loop below.
4450 	 */
4451 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4452 
4453 	/* Set XSTATE_BV */
4454 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4455 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4456 
4457 	/*
4458 	 * Copy each region from the possibly compacted offset to the
4459 	 * non-compacted offset.
4460 	 */
4461 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4462 	while (valid) {
4463 		u64 xfeature_mask = valid & -valid;
4464 		int xfeature_nr = fls64(xfeature_mask) - 1;
4465 		void *src = get_xsave_addr(xsave, xfeature_nr);
4466 
4467 		if (src) {
4468 			u32 size, offset, ecx, edx;
4469 			cpuid_count(XSTATE_CPUID, xfeature_nr,
4470 				    &size, &offset, &ecx, &edx);
4471 			if (xfeature_nr == XFEATURE_PKRU)
4472 				memcpy(dest + offset, &vcpu->arch.pkru,
4473 				       sizeof(vcpu->arch.pkru));
4474 			else
4475 				memcpy(dest + offset, src, size);
4476 
4477 		}
4478 
4479 		valid -= xfeature_mask;
4480 	}
4481 }
4482 
4483 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4484 {
4485 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4486 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4487 	u64 valid;
4488 
4489 	/*
4490 	 * Copy legacy XSAVE area, to avoid complications with CPUID
4491 	 * leaves 0 and 1 in the loop below.
4492 	 */
4493 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
4494 
4495 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
4496 	xsave->header.xfeatures = xstate_bv;
4497 	if (boot_cpu_has(X86_FEATURE_XSAVES))
4498 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4499 
4500 	/*
4501 	 * Copy each region from the non-compacted offset to the
4502 	 * possibly compacted offset.
4503 	 */
4504 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4505 	while (valid) {
4506 		u64 xfeature_mask = valid & -valid;
4507 		int xfeature_nr = fls64(xfeature_mask) - 1;
4508 		void *dest = get_xsave_addr(xsave, xfeature_nr);
4509 
4510 		if (dest) {
4511 			u32 size, offset, ecx, edx;
4512 			cpuid_count(XSTATE_CPUID, xfeature_nr,
4513 				    &size, &offset, &ecx, &edx);
4514 			if (xfeature_nr == XFEATURE_PKRU)
4515 				memcpy(&vcpu->arch.pkru, src + offset,
4516 				       sizeof(vcpu->arch.pkru));
4517 			else
4518 				memcpy(dest, src + offset, size);
4519 		}
4520 
4521 		valid -= xfeature_mask;
4522 	}
4523 }
4524 
4525 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4526 					 struct kvm_xsave *guest_xsave)
4527 {
4528 	if (!vcpu->arch.guest_fpu)
4529 		return;
4530 
4531 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4532 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4533 		fill_xsave((u8 *) guest_xsave->region, vcpu);
4534 	} else {
4535 		memcpy(guest_xsave->region,
4536 			&vcpu->arch.guest_fpu->state.fxsave,
4537 			sizeof(struct fxregs_state));
4538 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4539 			XFEATURE_MASK_FPSSE;
4540 	}
4541 }
4542 
4543 #define XSAVE_MXCSR_OFFSET 24
4544 
4545 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4546 					struct kvm_xsave *guest_xsave)
4547 {
4548 	u64 xstate_bv;
4549 	u32 mxcsr;
4550 
4551 	if (!vcpu->arch.guest_fpu)
4552 		return 0;
4553 
4554 	xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4555 	mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4556 
4557 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4558 		/*
4559 		 * Here we allow setting states that are not present in
4560 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
4561 		 * with old userspace.
4562 		 */
4563 		if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4564 			return -EINVAL;
4565 		load_xsave(vcpu, (u8 *)guest_xsave->region);
4566 	} else {
4567 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4568 			mxcsr & ~mxcsr_feature_mask)
4569 			return -EINVAL;
4570 		memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4571 			guest_xsave->region, sizeof(struct fxregs_state));
4572 	}
4573 	return 0;
4574 }
4575 
4576 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4577 					struct kvm_xcrs *guest_xcrs)
4578 {
4579 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4580 		guest_xcrs->nr_xcrs = 0;
4581 		return;
4582 	}
4583 
4584 	guest_xcrs->nr_xcrs = 1;
4585 	guest_xcrs->flags = 0;
4586 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4587 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4588 }
4589 
4590 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4591 				       struct kvm_xcrs *guest_xcrs)
4592 {
4593 	int i, r = 0;
4594 
4595 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
4596 		return -EINVAL;
4597 
4598 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4599 		return -EINVAL;
4600 
4601 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4602 		/* Only support XCR0 currently */
4603 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4604 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4605 				guest_xcrs->xcrs[i].value);
4606 			break;
4607 		}
4608 	if (r)
4609 		r = -EINVAL;
4610 	return r;
4611 }
4612 
4613 /*
4614  * kvm_set_guest_paused() indicates to the guest kernel that it has been
4615  * stopped by the hypervisor.  This function will be called from the host only.
4616  * EINVAL is returned when the host attempts to set the flag for a guest that
4617  * does not support pv clocks.
4618  */
4619 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4620 {
4621 	if (!vcpu->arch.pv_time_enabled)
4622 		return -EINVAL;
4623 	vcpu->arch.pvclock_set_guest_stopped_request = true;
4624 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4625 	return 0;
4626 }
4627 
4628 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4629 				     struct kvm_enable_cap *cap)
4630 {
4631 	int r;
4632 	uint16_t vmcs_version;
4633 	void __user *user_ptr;
4634 
4635 	if (cap->flags)
4636 		return -EINVAL;
4637 
4638 	switch (cap->cap) {
4639 	case KVM_CAP_HYPERV_SYNIC2:
4640 		if (cap->args[0])
4641 			return -EINVAL;
4642 		fallthrough;
4643 
4644 	case KVM_CAP_HYPERV_SYNIC:
4645 		if (!irqchip_in_kernel(vcpu->kvm))
4646 			return -EINVAL;
4647 		return kvm_hv_activate_synic(vcpu, cap->cap ==
4648 					     KVM_CAP_HYPERV_SYNIC2);
4649 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4650 		if (!kvm_x86_ops.nested_ops->enable_evmcs)
4651 			return -ENOTTY;
4652 		r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4653 		if (!r) {
4654 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
4655 			if (copy_to_user(user_ptr, &vmcs_version,
4656 					 sizeof(vmcs_version)))
4657 				r = -EFAULT;
4658 		}
4659 		return r;
4660 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4661 		if (!kvm_x86_ops.enable_direct_tlbflush)
4662 			return -ENOTTY;
4663 
4664 		return kvm_x86_ops.enable_direct_tlbflush(vcpu);
4665 
4666 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4667 		vcpu->arch.pv_cpuid.enforce = cap->args[0];
4668 		if (vcpu->arch.pv_cpuid.enforce)
4669 			kvm_update_pv_runtime(vcpu);
4670 
4671 		return 0;
4672 
4673 	default:
4674 		return -EINVAL;
4675 	}
4676 }
4677 
4678 long kvm_arch_vcpu_ioctl(struct file *filp,
4679 			 unsigned int ioctl, unsigned long arg)
4680 {
4681 	struct kvm_vcpu *vcpu = filp->private_data;
4682 	void __user *argp = (void __user *)arg;
4683 	int r;
4684 	union {
4685 		struct kvm_lapic_state *lapic;
4686 		struct kvm_xsave *xsave;
4687 		struct kvm_xcrs *xcrs;
4688 		void *buffer;
4689 	} u;
4690 
4691 	vcpu_load(vcpu);
4692 
4693 	u.buffer = NULL;
4694 	switch (ioctl) {
4695 	case KVM_GET_LAPIC: {
4696 		r = -EINVAL;
4697 		if (!lapic_in_kernel(vcpu))
4698 			goto out;
4699 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4700 				GFP_KERNEL_ACCOUNT);
4701 
4702 		r = -ENOMEM;
4703 		if (!u.lapic)
4704 			goto out;
4705 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4706 		if (r)
4707 			goto out;
4708 		r = -EFAULT;
4709 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4710 			goto out;
4711 		r = 0;
4712 		break;
4713 	}
4714 	case KVM_SET_LAPIC: {
4715 		r = -EINVAL;
4716 		if (!lapic_in_kernel(vcpu))
4717 			goto out;
4718 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
4719 		if (IS_ERR(u.lapic)) {
4720 			r = PTR_ERR(u.lapic);
4721 			goto out_nofree;
4722 		}
4723 
4724 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4725 		break;
4726 	}
4727 	case KVM_INTERRUPT: {
4728 		struct kvm_interrupt irq;
4729 
4730 		r = -EFAULT;
4731 		if (copy_from_user(&irq, argp, sizeof(irq)))
4732 			goto out;
4733 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4734 		break;
4735 	}
4736 	case KVM_NMI: {
4737 		r = kvm_vcpu_ioctl_nmi(vcpu);
4738 		break;
4739 	}
4740 	case KVM_SMI: {
4741 		r = kvm_vcpu_ioctl_smi(vcpu);
4742 		break;
4743 	}
4744 	case KVM_SET_CPUID: {
4745 		struct kvm_cpuid __user *cpuid_arg = argp;
4746 		struct kvm_cpuid cpuid;
4747 
4748 		r = -EFAULT;
4749 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4750 			goto out;
4751 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4752 		break;
4753 	}
4754 	case KVM_SET_CPUID2: {
4755 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4756 		struct kvm_cpuid2 cpuid;
4757 
4758 		r = -EFAULT;
4759 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4760 			goto out;
4761 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4762 					      cpuid_arg->entries);
4763 		break;
4764 	}
4765 	case KVM_GET_CPUID2: {
4766 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4767 		struct kvm_cpuid2 cpuid;
4768 
4769 		r = -EFAULT;
4770 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4771 			goto out;
4772 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4773 					      cpuid_arg->entries);
4774 		if (r)
4775 			goto out;
4776 		r = -EFAULT;
4777 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4778 			goto out;
4779 		r = 0;
4780 		break;
4781 	}
4782 	case KVM_GET_MSRS: {
4783 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4784 		r = msr_io(vcpu, argp, do_get_msr, 1);
4785 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4786 		break;
4787 	}
4788 	case KVM_SET_MSRS: {
4789 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4790 		r = msr_io(vcpu, argp, do_set_msr, 0);
4791 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4792 		break;
4793 	}
4794 	case KVM_TPR_ACCESS_REPORTING: {
4795 		struct kvm_tpr_access_ctl tac;
4796 
4797 		r = -EFAULT;
4798 		if (copy_from_user(&tac, argp, sizeof(tac)))
4799 			goto out;
4800 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4801 		if (r)
4802 			goto out;
4803 		r = -EFAULT;
4804 		if (copy_to_user(argp, &tac, sizeof(tac)))
4805 			goto out;
4806 		r = 0;
4807 		break;
4808 	};
4809 	case KVM_SET_VAPIC_ADDR: {
4810 		struct kvm_vapic_addr va;
4811 		int idx;
4812 
4813 		r = -EINVAL;
4814 		if (!lapic_in_kernel(vcpu))
4815 			goto out;
4816 		r = -EFAULT;
4817 		if (copy_from_user(&va, argp, sizeof(va)))
4818 			goto out;
4819 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4820 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4821 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4822 		break;
4823 	}
4824 	case KVM_X86_SETUP_MCE: {
4825 		u64 mcg_cap;
4826 
4827 		r = -EFAULT;
4828 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4829 			goto out;
4830 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4831 		break;
4832 	}
4833 	case KVM_X86_SET_MCE: {
4834 		struct kvm_x86_mce mce;
4835 
4836 		r = -EFAULT;
4837 		if (copy_from_user(&mce, argp, sizeof(mce)))
4838 			goto out;
4839 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4840 		break;
4841 	}
4842 	case KVM_GET_VCPU_EVENTS: {
4843 		struct kvm_vcpu_events events;
4844 
4845 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4846 
4847 		r = -EFAULT;
4848 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4849 			break;
4850 		r = 0;
4851 		break;
4852 	}
4853 	case KVM_SET_VCPU_EVENTS: {
4854 		struct kvm_vcpu_events events;
4855 
4856 		r = -EFAULT;
4857 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4858 			break;
4859 
4860 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4861 		break;
4862 	}
4863 	case KVM_GET_DEBUGREGS: {
4864 		struct kvm_debugregs dbgregs;
4865 
4866 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4867 
4868 		r = -EFAULT;
4869 		if (copy_to_user(argp, &dbgregs,
4870 				 sizeof(struct kvm_debugregs)))
4871 			break;
4872 		r = 0;
4873 		break;
4874 	}
4875 	case KVM_SET_DEBUGREGS: {
4876 		struct kvm_debugregs dbgregs;
4877 
4878 		r = -EFAULT;
4879 		if (copy_from_user(&dbgregs, argp,
4880 				   sizeof(struct kvm_debugregs)))
4881 			break;
4882 
4883 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4884 		break;
4885 	}
4886 	case KVM_GET_XSAVE: {
4887 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4888 		r = -ENOMEM;
4889 		if (!u.xsave)
4890 			break;
4891 
4892 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4893 
4894 		r = -EFAULT;
4895 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4896 			break;
4897 		r = 0;
4898 		break;
4899 	}
4900 	case KVM_SET_XSAVE: {
4901 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
4902 		if (IS_ERR(u.xsave)) {
4903 			r = PTR_ERR(u.xsave);
4904 			goto out_nofree;
4905 		}
4906 
4907 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4908 		break;
4909 	}
4910 	case KVM_GET_XCRS: {
4911 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4912 		r = -ENOMEM;
4913 		if (!u.xcrs)
4914 			break;
4915 
4916 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4917 
4918 		r = -EFAULT;
4919 		if (copy_to_user(argp, u.xcrs,
4920 				 sizeof(struct kvm_xcrs)))
4921 			break;
4922 		r = 0;
4923 		break;
4924 	}
4925 	case KVM_SET_XCRS: {
4926 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4927 		if (IS_ERR(u.xcrs)) {
4928 			r = PTR_ERR(u.xcrs);
4929 			goto out_nofree;
4930 		}
4931 
4932 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4933 		break;
4934 	}
4935 	case KVM_SET_TSC_KHZ: {
4936 		u32 user_tsc_khz;
4937 
4938 		r = -EINVAL;
4939 		user_tsc_khz = (u32)arg;
4940 
4941 		if (kvm_has_tsc_control &&
4942 		    user_tsc_khz >= kvm_max_guest_tsc_khz)
4943 			goto out;
4944 
4945 		if (user_tsc_khz == 0)
4946 			user_tsc_khz = tsc_khz;
4947 
4948 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4949 			r = 0;
4950 
4951 		goto out;
4952 	}
4953 	case KVM_GET_TSC_KHZ: {
4954 		r = vcpu->arch.virtual_tsc_khz;
4955 		goto out;
4956 	}
4957 	case KVM_KVMCLOCK_CTRL: {
4958 		r = kvm_set_guest_paused(vcpu);
4959 		goto out;
4960 	}
4961 	case KVM_ENABLE_CAP: {
4962 		struct kvm_enable_cap cap;
4963 
4964 		r = -EFAULT;
4965 		if (copy_from_user(&cap, argp, sizeof(cap)))
4966 			goto out;
4967 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4968 		break;
4969 	}
4970 	case KVM_GET_NESTED_STATE: {
4971 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4972 		u32 user_data_size;
4973 
4974 		r = -EINVAL;
4975 		if (!kvm_x86_ops.nested_ops->get_state)
4976 			break;
4977 
4978 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4979 		r = -EFAULT;
4980 		if (get_user(user_data_size, &user_kvm_nested_state->size))
4981 			break;
4982 
4983 		r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
4984 						     user_data_size);
4985 		if (r < 0)
4986 			break;
4987 
4988 		if (r > user_data_size) {
4989 			if (put_user(r, &user_kvm_nested_state->size))
4990 				r = -EFAULT;
4991 			else
4992 				r = -E2BIG;
4993 			break;
4994 		}
4995 
4996 		r = 0;
4997 		break;
4998 	}
4999 	case KVM_SET_NESTED_STATE: {
5000 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5001 		struct kvm_nested_state kvm_state;
5002 		int idx;
5003 
5004 		r = -EINVAL;
5005 		if (!kvm_x86_ops.nested_ops->set_state)
5006 			break;
5007 
5008 		r = -EFAULT;
5009 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5010 			break;
5011 
5012 		r = -EINVAL;
5013 		if (kvm_state.size < sizeof(kvm_state))
5014 			break;
5015 
5016 		if (kvm_state.flags &
5017 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5018 		      | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5019 		      | KVM_STATE_NESTED_GIF_SET))
5020 			break;
5021 
5022 		/* nested_run_pending implies guest_mode.  */
5023 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5024 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5025 			break;
5026 
5027 		idx = srcu_read_lock(&vcpu->kvm->srcu);
5028 		r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5029 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5030 		break;
5031 	}
5032 	case KVM_GET_SUPPORTED_HV_CPUID:
5033 		r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5034 		break;
5035 	default:
5036 		r = -EINVAL;
5037 	}
5038 out:
5039 	kfree(u.buffer);
5040 out_nofree:
5041 	vcpu_put(vcpu);
5042 	return r;
5043 }
5044 
5045 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5046 {
5047 	return VM_FAULT_SIGBUS;
5048 }
5049 
5050 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5051 {
5052 	int ret;
5053 
5054 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
5055 		return -EINVAL;
5056 	ret = kvm_x86_ops.set_tss_addr(kvm, addr);
5057 	return ret;
5058 }
5059 
5060 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5061 					      u64 ident_addr)
5062 {
5063 	return kvm_x86_ops.set_identity_map_addr(kvm, ident_addr);
5064 }
5065 
5066 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5067 					 unsigned long kvm_nr_mmu_pages)
5068 {
5069 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5070 		return -EINVAL;
5071 
5072 	mutex_lock(&kvm->slots_lock);
5073 
5074 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5075 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5076 
5077 	mutex_unlock(&kvm->slots_lock);
5078 	return 0;
5079 }
5080 
5081 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5082 {
5083 	return kvm->arch.n_max_mmu_pages;
5084 }
5085 
5086 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5087 {
5088 	struct kvm_pic *pic = kvm->arch.vpic;
5089 	int r;
5090 
5091 	r = 0;
5092 	switch (chip->chip_id) {
5093 	case KVM_IRQCHIP_PIC_MASTER:
5094 		memcpy(&chip->chip.pic, &pic->pics[0],
5095 			sizeof(struct kvm_pic_state));
5096 		break;
5097 	case KVM_IRQCHIP_PIC_SLAVE:
5098 		memcpy(&chip->chip.pic, &pic->pics[1],
5099 			sizeof(struct kvm_pic_state));
5100 		break;
5101 	case KVM_IRQCHIP_IOAPIC:
5102 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
5103 		break;
5104 	default:
5105 		r = -EINVAL;
5106 		break;
5107 	}
5108 	return r;
5109 }
5110 
5111 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5112 {
5113 	struct kvm_pic *pic = kvm->arch.vpic;
5114 	int r;
5115 
5116 	r = 0;
5117 	switch (chip->chip_id) {
5118 	case KVM_IRQCHIP_PIC_MASTER:
5119 		spin_lock(&pic->lock);
5120 		memcpy(&pic->pics[0], &chip->chip.pic,
5121 			sizeof(struct kvm_pic_state));
5122 		spin_unlock(&pic->lock);
5123 		break;
5124 	case KVM_IRQCHIP_PIC_SLAVE:
5125 		spin_lock(&pic->lock);
5126 		memcpy(&pic->pics[1], &chip->chip.pic,
5127 			sizeof(struct kvm_pic_state));
5128 		spin_unlock(&pic->lock);
5129 		break;
5130 	case KVM_IRQCHIP_IOAPIC:
5131 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
5132 		break;
5133 	default:
5134 		r = -EINVAL;
5135 		break;
5136 	}
5137 	kvm_pic_update_irq(pic);
5138 	return r;
5139 }
5140 
5141 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5142 {
5143 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5144 
5145 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5146 
5147 	mutex_lock(&kps->lock);
5148 	memcpy(ps, &kps->channels, sizeof(*ps));
5149 	mutex_unlock(&kps->lock);
5150 	return 0;
5151 }
5152 
5153 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5154 {
5155 	int i;
5156 	struct kvm_pit *pit = kvm->arch.vpit;
5157 
5158 	mutex_lock(&pit->pit_state.lock);
5159 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5160 	for (i = 0; i < 3; i++)
5161 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5162 	mutex_unlock(&pit->pit_state.lock);
5163 	return 0;
5164 }
5165 
5166 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5167 {
5168 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
5169 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5170 		sizeof(ps->channels));
5171 	ps->flags = kvm->arch.vpit->pit_state.flags;
5172 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5173 	memset(&ps->reserved, 0, sizeof(ps->reserved));
5174 	return 0;
5175 }
5176 
5177 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5178 {
5179 	int start = 0;
5180 	int i;
5181 	u32 prev_legacy, cur_legacy;
5182 	struct kvm_pit *pit = kvm->arch.vpit;
5183 
5184 	mutex_lock(&pit->pit_state.lock);
5185 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5186 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5187 	if (!prev_legacy && cur_legacy)
5188 		start = 1;
5189 	memcpy(&pit->pit_state.channels, &ps->channels,
5190 	       sizeof(pit->pit_state.channels));
5191 	pit->pit_state.flags = ps->flags;
5192 	for (i = 0; i < 3; i++)
5193 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5194 				   start && i == 0);
5195 	mutex_unlock(&pit->pit_state.lock);
5196 	return 0;
5197 }
5198 
5199 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5200 				 struct kvm_reinject_control *control)
5201 {
5202 	struct kvm_pit *pit = kvm->arch.vpit;
5203 
5204 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
5205 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5206 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
5207 	 */
5208 	mutex_lock(&pit->pit_state.lock);
5209 	kvm_pit_set_reinject(pit, control->pit_reinject);
5210 	mutex_unlock(&pit->pit_state.lock);
5211 
5212 	return 0;
5213 }
5214 
5215 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5216 {
5217 	/*
5218 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
5219 	 */
5220 	if (kvm_x86_ops.flush_log_dirty)
5221 		kvm_x86_ops.flush_log_dirty(kvm);
5222 }
5223 
5224 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5225 			bool line_status)
5226 {
5227 	if (!irqchip_in_kernel(kvm))
5228 		return -ENXIO;
5229 
5230 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5231 					irq_event->irq, irq_event->level,
5232 					line_status);
5233 	return 0;
5234 }
5235 
5236 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5237 			    struct kvm_enable_cap *cap)
5238 {
5239 	int r;
5240 
5241 	if (cap->flags)
5242 		return -EINVAL;
5243 
5244 	switch (cap->cap) {
5245 	case KVM_CAP_DISABLE_QUIRKS:
5246 		kvm->arch.disabled_quirks = cap->args[0];
5247 		r = 0;
5248 		break;
5249 	case KVM_CAP_SPLIT_IRQCHIP: {
5250 		mutex_lock(&kvm->lock);
5251 		r = -EINVAL;
5252 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5253 			goto split_irqchip_unlock;
5254 		r = -EEXIST;
5255 		if (irqchip_in_kernel(kvm))
5256 			goto split_irqchip_unlock;
5257 		if (kvm->created_vcpus)
5258 			goto split_irqchip_unlock;
5259 		r = kvm_setup_empty_irq_routing(kvm);
5260 		if (r)
5261 			goto split_irqchip_unlock;
5262 		/* Pairs with irqchip_in_kernel. */
5263 		smp_wmb();
5264 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5265 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5266 		r = 0;
5267 split_irqchip_unlock:
5268 		mutex_unlock(&kvm->lock);
5269 		break;
5270 	}
5271 	case KVM_CAP_X2APIC_API:
5272 		r = -EINVAL;
5273 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5274 			break;
5275 
5276 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5277 			kvm->arch.x2apic_format = true;
5278 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5279 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
5280 
5281 		r = 0;
5282 		break;
5283 	case KVM_CAP_X86_DISABLE_EXITS:
5284 		r = -EINVAL;
5285 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5286 			break;
5287 
5288 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5289 			kvm_can_mwait_in_guest())
5290 			kvm->arch.mwait_in_guest = true;
5291 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5292 			kvm->arch.hlt_in_guest = true;
5293 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5294 			kvm->arch.pause_in_guest = true;
5295 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5296 			kvm->arch.cstate_in_guest = true;
5297 		r = 0;
5298 		break;
5299 	case KVM_CAP_MSR_PLATFORM_INFO:
5300 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5301 		r = 0;
5302 		break;
5303 	case KVM_CAP_EXCEPTION_PAYLOAD:
5304 		kvm->arch.exception_payload_enabled = cap->args[0];
5305 		r = 0;
5306 		break;
5307 	case KVM_CAP_X86_USER_SPACE_MSR:
5308 		kvm->arch.user_space_msr_mask = cap->args[0];
5309 		r = 0;
5310 		break;
5311 	default:
5312 		r = -EINVAL;
5313 		break;
5314 	}
5315 	return r;
5316 }
5317 
5318 static void kvm_clear_msr_filter(struct kvm *kvm)
5319 {
5320 	u32 i;
5321 	u32 count = kvm->arch.msr_filter.count;
5322 	struct msr_bitmap_range ranges[16];
5323 
5324 	mutex_lock(&kvm->lock);
5325 	kvm->arch.msr_filter.count = 0;
5326 	memcpy(ranges, kvm->arch.msr_filter.ranges, count * sizeof(ranges[0]));
5327 	mutex_unlock(&kvm->lock);
5328 	synchronize_srcu(&kvm->srcu);
5329 
5330 	for (i = 0; i < count; i++)
5331 		kfree(ranges[i].bitmap);
5332 }
5333 
5334 static int kvm_add_msr_filter(struct kvm *kvm, struct kvm_msr_filter_range *user_range)
5335 {
5336 	struct msr_bitmap_range *ranges = kvm->arch.msr_filter.ranges;
5337 	struct msr_bitmap_range range;
5338 	unsigned long *bitmap = NULL;
5339 	size_t bitmap_size;
5340 	int r;
5341 
5342 	if (!user_range->nmsrs)
5343 		return 0;
5344 
5345 	bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5346 	if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5347 		return -EINVAL;
5348 
5349 	bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5350 	if (IS_ERR(bitmap))
5351 		return PTR_ERR(bitmap);
5352 
5353 	range = (struct msr_bitmap_range) {
5354 		.flags = user_range->flags,
5355 		.base = user_range->base,
5356 		.nmsrs = user_range->nmsrs,
5357 		.bitmap = bitmap,
5358 	};
5359 
5360 	if (range.flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE)) {
5361 		r = -EINVAL;
5362 		goto err;
5363 	}
5364 
5365 	if (!range.flags) {
5366 		r = -EINVAL;
5367 		goto err;
5368 	}
5369 
5370 	/* Everything ok, add this range identifier to our global pool */
5371 	ranges[kvm->arch.msr_filter.count] = range;
5372 	/* Make sure we filled the array before we tell anyone to walk it */
5373 	smp_wmb();
5374 	kvm->arch.msr_filter.count++;
5375 
5376 	return 0;
5377 err:
5378 	kfree(bitmap);
5379 	return r;
5380 }
5381 
5382 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5383 {
5384 	struct kvm_msr_filter __user *user_msr_filter = argp;
5385 	struct kvm_msr_filter filter;
5386 	bool default_allow;
5387 	int r = 0;
5388 	bool empty = true;
5389 	u32 i;
5390 
5391 	if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5392 		return -EFAULT;
5393 
5394 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5395 		empty &= !filter.ranges[i].nmsrs;
5396 
5397 	default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5398 	if (empty && !default_allow)
5399 		return -EINVAL;
5400 
5401 	kvm_clear_msr_filter(kvm);
5402 
5403 	kvm->arch.msr_filter.default_allow = default_allow;
5404 
5405 	/*
5406 	 * Protect from concurrent calls to this function that could trigger
5407 	 * a TOCTOU violation on kvm->arch.msr_filter.count.
5408 	 */
5409 	mutex_lock(&kvm->lock);
5410 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5411 		r = kvm_add_msr_filter(kvm, &filter.ranges[i]);
5412 		if (r)
5413 			break;
5414 	}
5415 
5416 	kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5417 	mutex_unlock(&kvm->lock);
5418 
5419 	return r;
5420 }
5421 
5422 long kvm_arch_vm_ioctl(struct file *filp,
5423 		       unsigned int ioctl, unsigned long arg)
5424 {
5425 	struct kvm *kvm = filp->private_data;
5426 	void __user *argp = (void __user *)arg;
5427 	int r = -ENOTTY;
5428 	/*
5429 	 * This union makes it completely explicit to gcc-3.x
5430 	 * that these two variables' stack usage should be
5431 	 * combined, not added together.
5432 	 */
5433 	union {
5434 		struct kvm_pit_state ps;
5435 		struct kvm_pit_state2 ps2;
5436 		struct kvm_pit_config pit_config;
5437 	} u;
5438 
5439 	switch (ioctl) {
5440 	case KVM_SET_TSS_ADDR:
5441 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5442 		break;
5443 	case KVM_SET_IDENTITY_MAP_ADDR: {
5444 		u64 ident_addr;
5445 
5446 		mutex_lock(&kvm->lock);
5447 		r = -EINVAL;
5448 		if (kvm->created_vcpus)
5449 			goto set_identity_unlock;
5450 		r = -EFAULT;
5451 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5452 			goto set_identity_unlock;
5453 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5454 set_identity_unlock:
5455 		mutex_unlock(&kvm->lock);
5456 		break;
5457 	}
5458 	case KVM_SET_NR_MMU_PAGES:
5459 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5460 		break;
5461 	case KVM_GET_NR_MMU_PAGES:
5462 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5463 		break;
5464 	case KVM_CREATE_IRQCHIP: {
5465 		mutex_lock(&kvm->lock);
5466 
5467 		r = -EEXIST;
5468 		if (irqchip_in_kernel(kvm))
5469 			goto create_irqchip_unlock;
5470 
5471 		r = -EINVAL;
5472 		if (kvm->created_vcpus)
5473 			goto create_irqchip_unlock;
5474 
5475 		r = kvm_pic_init(kvm);
5476 		if (r)
5477 			goto create_irqchip_unlock;
5478 
5479 		r = kvm_ioapic_init(kvm);
5480 		if (r) {
5481 			kvm_pic_destroy(kvm);
5482 			goto create_irqchip_unlock;
5483 		}
5484 
5485 		r = kvm_setup_default_irq_routing(kvm);
5486 		if (r) {
5487 			kvm_ioapic_destroy(kvm);
5488 			kvm_pic_destroy(kvm);
5489 			goto create_irqchip_unlock;
5490 		}
5491 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5492 		smp_wmb();
5493 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5494 	create_irqchip_unlock:
5495 		mutex_unlock(&kvm->lock);
5496 		break;
5497 	}
5498 	case KVM_CREATE_PIT:
5499 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5500 		goto create_pit;
5501 	case KVM_CREATE_PIT2:
5502 		r = -EFAULT;
5503 		if (copy_from_user(&u.pit_config, argp,
5504 				   sizeof(struct kvm_pit_config)))
5505 			goto out;
5506 	create_pit:
5507 		mutex_lock(&kvm->lock);
5508 		r = -EEXIST;
5509 		if (kvm->arch.vpit)
5510 			goto create_pit_unlock;
5511 		r = -ENOMEM;
5512 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5513 		if (kvm->arch.vpit)
5514 			r = 0;
5515 	create_pit_unlock:
5516 		mutex_unlock(&kvm->lock);
5517 		break;
5518 	case KVM_GET_IRQCHIP: {
5519 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5520 		struct kvm_irqchip *chip;
5521 
5522 		chip = memdup_user(argp, sizeof(*chip));
5523 		if (IS_ERR(chip)) {
5524 			r = PTR_ERR(chip);
5525 			goto out;
5526 		}
5527 
5528 		r = -ENXIO;
5529 		if (!irqchip_kernel(kvm))
5530 			goto get_irqchip_out;
5531 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5532 		if (r)
5533 			goto get_irqchip_out;
5534 		r = -EFAULT;
5535 		if (copy_to_user(argp, chip, sizeof(*chip)))
5536 			goto get_irqchip_out;
5537 		r = 0;
5538 	get_irqchip_out:
5539 		kfree(chip);
5540 		break;
5541 	}
5542 	case KVM_SET_IRQCHIP: {
5543 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5544 		struct kvm_irqchip *chip;
5545 
5546 		chip = memdup_user(argp, sizeof(*chip));
5547 		if (IS_ERR(chip)) {
5548 			r = PTR_ERR(chip);
5549 			goto out;
5550 		}
5551 
5552 		r = -ENXIO;
5553 		if (!irqchip_kernel(kvm))
5554 			goto set_irqchip_out;
5555 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5556 	set_irqchip_out:
5557 		kfree(chip);
5558 		break;
5559 	}
5560 	case KVM_GET_PIT: {
5561 		r = -EFAULT;
5562 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5563 			goto out;
5564 		r = -ENXIO;
5565 		if (!kvm->arch.vpit)
5566 			goto out;
5567 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5568 		if (r)
5569 			goto out;
5570 		r = -EFAULT;
5571 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5572 			goto out;
5573 		r = 0;
5574 		break;
5575 	}
5576 	case KVM_SET_PIT: {
5577 		r = -EFAULT;
5578 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5579 			goto out;
5580 		mutex_lock(&kvm->lock);
5581 		r = -ENXIO;
5582 		if (!kvm->arch.vpit)
5583 			goto set_pit_out;
5584 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5585 set_pit_out:
5586 		mutex_unlock(&kvm->lock);
5587 		break;
5588 	}
5589 	case KVM_GET_PIT2: {
5590 		r = -ENXIO;
5591 		if (!kvm->arch.vpit)
5592 			goto out;
5593 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5594 		if (r)
5595 			goto out;
5596 		r = -EFAULT;
5597 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5598 			goto out;
5599 		r = 0;
5600 		break;
5601 	}
5602 	case KVM_SET_PIT2: {
5603 		r = -EFAULT;
5604 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5605 			goto out;
5606 		mutex_lock(&kvm->lock);
5607 		r = -ENXIO;
5608 		if (!kvm->arch.vpit)
5609 			goto set_pit2_out;
5610 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5611 set_pit2_out:
5612 		mutex_unlock(&kvm->lock);
5613 		break;
5614 	}
5615 	case KVM_REINJECT_CONTROL: {
5616 		struct kvm_reinject_control control;
5617 		r =  -EFAULT;
5618 		if (copy_from_user(&control, argp, sizeof(control)))
5619 			goto out;
5620 		r = -ENXIO;
5621 		if (!kvm->arch.vpit)
5622 			goto out;
5623 		r = kvm_vm_ioctl_reinject(kvm, &control);
5624 		break;
5625 	}
5626 	case KVM_SET_BOOT_CPU_ID:
5627 		r = 0;
5628 		mutex_lock(&kvm->lock);
5629 		if (kvm->created_vcpus)
5630 			r = -EBUSY;
5631 		else
5632 			kvm->arch.bsp_vcpu_id = arg;
5633 		mutex_unlock(&kvm->lock);
5634 		break;
5635 	case KVM_XEN_HVM_CONFIG: {
5636 		struct kvm_xen_hvm_config xhc;
5637 		r = -EFAULT;
5638 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
5639 			goto out;
5640 		r = -EINVAL;
5641 		if (xhc.flags)
5642 			goto out;
5643 		memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5644 		r = 0;
5645 		break;
5646 	}
5647 	case KVM_SET_CLOCK: {
5648 		struct kvm_clock_data user_ns;
5649 		u64 now_ns;
5650 
5651 		r = -EFAULT;
5652 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5653 			goto out;
5654 
5655 		r = -EINVAL;
5656 		if (user_ns.flags)
5657 			goto out;
5658 
5659 		r = 0;
5660 		/*
5661 		 * TODO: userspace has to take care of races with VCPU_RUN, so
5662 		 * kvm_gen_update_masterclock() can be cut down to locked
5663 		 * pvclock_update_vm_gtod_copy().
5664 		 */
5665 		kvm_gen_update_masterclock(kvm);
5666 		now_ns = get_kvmclock_ns(kvm);
5667 		kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5668 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5669 		break;
5670 	}
5671 	case KVM_GET_CLOCK: {
5672 		struct kvm_clock_data user_ns;
5673 		u64 now_ns;
5674 
5675 		now_ns = get_kvmclock_ns(kvm);
5676 		user_ns.clock = now_ns;
5677 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5678 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5679 
5680 		r = -EFAULT;
5681 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5682 			goto out;
5683 		r = 0;
5684 		break;
5685 	}
5686 	case KVM_MEMORY_ENCRYPT_OP: {
5687 		r = -ENOTTY;
5688 		if (kvm_x86_ops.mem_enc_op)
5689 			r = kvm_x86_ops.mem_enc_op(kvm, argp);
5690 		break;
5691 	}
5692 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
5693 		struct kvm_enc_region region;
5694 
5695 		r = -EFAULT;
5696 		if (copy_from_user(&region, argp, sizeof(region)))
5697 			goto out;
5698 
5699 		r = -ENOTTY;
5700 		if (kvm_x86_ops.mem_enc_reg_region)
5701 			r = kvm_x86_ops.mem_enc_reg_region(kvm, &region);
5702 		break;
5703 	}
5704 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5705 		struct kvm_enc_region region;
5706 
5707 		r = -EFAULT;
5708 		if (copy_from_user(&region, argp, sizeof(region)))
5709 			goto out;
5710 
5711 		r = -ENOTTY;
5712 		if (kvm_x86_ops.mem_enc_unreg_region)
5713 			r = kvm_x86_ops.mem_enc_unreg_region(kvm, &region);
5714 		break;
5715 	}
5716 	case KVM_HYPERV_EVENTFD: {
5717 		struct kvm_hyperv_eventfd hvevfd;
5718 
5719 		r = -EFAULT;
5720 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5721 			goto out;
5722 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5723 		break;
5724 	}
5725 	case KVM_SET_PMU_EVENT_FILTER:
5726 		r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5727 		break;
5728 	case KVM_X86_SET_MSR_FILTER:
5729 		r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
5730 		break;
5731 	default:
5732 		r = -ENOTTY;
5733 	}
5734 out:
5735 	return r;
5736 }
5737 
5738 static void kvm_init_msr_list(void)
5739 {
5740 	struct x86_pmu_capability x86_pmu;
5741 	u32 dummy[2];
5742 	unsigned i;
5743 
5744 	BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5745 			 "Please update the fixed PMCs in msrs_to_saved_all[]");
5746 
5747 	perf_get_x86_pmu_capability(&x86_pmu);
5748 
5749 	num_msrs_to_save = 0;
5750 	num_emulated_msrs = 0;
5751 	num_msr_based_features = 0;
5752 
5753 	for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5754 		if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5755 			continue;
5756 
5757 		/*
5758 		 * Even MSRs that are valid in the host may not be exposed
5759 		 * to the guests in some cases.
5760 		 */
5761 		switch (msrs_to_save_all[i]) {
5762 		case MSR_IA32_BNDCFGS:
5763 			if (!kvm_mpx_supported())
5764 				continue;
5765 			break;
5766 		case MSR_TSC_AUX:
5767 			if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
5768 				continue;
5769 			break;
5770 		case MSR_IA32_UMWAIT_CONTROL:
5771 			if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
5772 				continue;
5773 			break;
5774 		case MSR_IA32_RTIT_CTL:
5775 		case MSR_IA32_RTIT_STATUS:
5776 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
5777 				continue;
5778 			break;
5779 		case MSR_IA32_RTIT_CR3_MATCH:
5780 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5781 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5782 				continue;
5783 			break;
5784 		case MSR_IA32_RTIT_OUTPUT_BASE:
5785 		case MSR_IA32_RTIT_OUTPUT_MASK:
5786 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5787 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5788 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5789 				continue;
5790 			break;
5791 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
5792 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5793 				msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5794 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5795 				continue;
5796 			break;
5797 		case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5798 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5799 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5800 				continue;
5801 			break;
5802 		case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5803 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5804 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5805 				continue;
5806 			break;
5807 		default:
5808 			break;
5809 		}
5810 
5811 		msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5812 	}
5813 
5814 	for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5815 		if (!kvm_x86_ops.has_emulated_msr(NULL, emulated_msrs_all[i]))
5816 			continue;
5817 
5818 		emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5819 	}
5820 
5821 	for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5822 		struct kvm_msr_entry msr;
5823 
5824 		msr.index = msr_based_features_all[i];
5825 		if (kvm_get_msr_feature(&msr))
5826 			continue;
5827 
5828 		msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
5829 	}
5830 }
5831 
5832 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5833 			   const void *v)
5834 {
5835 	int handled = 0;
5836 	int n;
5837 
5838 	do {
5839 		n = min(len, 8);
5840 		if (!(lapic_in_kernel(vcpu) &&
5841 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5842 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5843 			break;
5844 		handled += n;
5845 		addr += n;
5846 		len -= n;
5847 		v += n;
5848 	} while (len);
5849 
5850 	return handled;
5851 }
5852 
5853 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5854 {
5855 	int handled = 0;
5856 	int n;
5857 
5858 	do {
5859 		n = min(len, 8);
5860 		if (!(lapic_in_kernel(vcpu) &&
5861 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5862 					 addr, n, v))
5863 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5864 			break;
5865 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5866 		handled += n;
5867 		addr += n;
5868 		len -= n;
5869 		v += n;
5870 	} while (len);
5871 
5872 	return handled;
5873 }
5874 
5875 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5876 			struct kvm_segment *var, int seg)
5877 {
5878 	kvm_x86_ops.set_segment(vcpu, var, seg);
5879 }
5880 
5881 void kvm_get_segment(struct kvm_vcpu *vcpu,
5882 		     struct kvm_segment *var, int seg)
5883 {
5884 	kvm_x86_ops.get_segment(vcpu, var, seg);
5885 }
5886 
5887 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5888 			   struct x86_exception *exception)
5889 {
5890 	gpa_t t_gpa;
5891 
5892 	BUG_ON(!mmu_is_nested(vcpu));
5893 
5894 	/* NPT walks are always user-walks */
5895 	access |= PFERR_USER_MASK;
5896 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5897 
5898 	return t_gpa;
5899 }
5900 
5901 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5902 			      struct x86_exception *exception)
5903 {
5904 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5905 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5906 }
5907 
5908  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5909 				struct x86_exception *exception)
5910 {
5911 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5912 	access |= PFERR_FETCH_MASK;
5913 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5914 }
5915 
5916 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5917 			       struct x86_exception *exception)
5918 {
5919 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5920 	access |= PFERR_WRITE_MASK;
5921 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5922 }
5923 
5924 /* uses this to access any guest's mapped memory without checking CPL */
5925 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5926 				struct x86_exception *exception)
5927 {
5928 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5929 }
5930 
5931 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5932 				      struct kvm_vcpu *vcpu, u32 access,
5933 				      struct x86_exception *exception)
5934 {
5935 	void *data = val;
5936 	int r = X86EMUL_CONTINUE;
5937 
5938 	while (bytes) {
5939 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5940 							    exception);
5941 		unsigned offset = addr & (PAGE_SIZE-1);
5942 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5943 		int ret;
5944 
5945 		if (gpa == UNMAPPED_GVA)
5946 			return X86EMUL_PROPAGATE_FAULT;
5947 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5948 					       offset, toread);
5949 		if (ret < 0) {
5950 			r = X86EMUL_IO_NEEDED;
5951 			goto out;
5952 		}
5953 
5954 		bytes -= toread;
5955 		data += toread;
5956 		addr += toread;
5957 	}
5958 out:
5959 	return r;
5960 }
5961 
5962 /* used for instruction fetching */
5963 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5964 				gva_t addr, void *val, unsigned int bytes,
5965 				struct x86_exception *exception)
5966 {
5967 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5968 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5969 	unsigned offset;
5970 	int ret;
5971 
5972 	/* Inline kvm_read_guest_virt_helper for speed.  */
5973 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5974 						    exception);
5975 	if (unlikely(gpa == UNMAPPED_GVA))
5976 		return X86EMUL_PROPAGATE_FAULT;
5977 
5978 	offset = addr & (PAGE_SIZE-1);
5979 	if (WARN_ON(offset + bytes > PAGE_SIZE))
5980 		bytes = (unsigned)PAGE_SIZE - offset;
5981 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5982 				       offset, bytes);
5983 	if (unlikely(ret < 0))
5984 		return X86EMUL_IO_NEEDED;
5985 
5986 	return X86EMUL_CONTINUE;
5987 }
5988 
5989 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5990 			       gva_t addr, void *val, unsigned int bytes,
5991 			       struct x86_exception *exception)
5992 {
5993 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5994 
5995 	/*
5996 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5997 	 * is returned, but our callers are not ready for that and they blindly
5998 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
5999 	 * uninitialized kernel stack memory into cr2 and error code.
6000 	 */
6001 	memset(exception, 0, sizeof(*exception));
6002 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6003 					  exception);
6004 }
6005 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6006 
6007 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6008 			     gva_t addr, void *val, unsigned int bytes,
6009 			     struct x86_exception *exception, bool system)
6010 {
6011 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6012 	u32 access = 0;
6013 
6014 	if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
6015 		access |= PFERR_USER_MASK;
6016 
6017 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6018 }
6019 
6020 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6021 		unsigned long addr, void *val, unsigned int bytes)
6022 {
6023 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6024 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6025 
6026 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6027 }
6028 
6029 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6030 				      struct kvm_vcpu *vcpu, u32 access,
6031 				      struct x86_exception *exception)
6032 {
6033 	void *data = val;
6034 	int r = X86EMUL_CONTINUE;
6035 
6036 	while (bytes) {
6037 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6038 							     access,
6039 							     exception);
6040 		unsigned offset = addr & (PAGE_SIZE-1);
6041 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6042 		int ret;
6043 
6044 		if (gpa == UNMAPPED_GVA)
6045 			return X86EMUL_PROPAGATE_FAULT;
6046 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6047 		if (ret < 0) {
6048 			r = X86EMUL_IO_NEEDED;
6049 			goto out;
6050 		}
6051 
6052 		bytes -= towrite;
6053 		data += towrite;
6054 		addr += towrite;
6055 	}
6056 out:
6057 	return r;
6058 }
6059 
6060 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6061 			      unsigned int bytes, struct x86_exception *exception,
6062 			      bool system)
6063 {
6064 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6065 	u32 access = PFERR_WRITE_MASK;
6066 
6067 	if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
6068 		access |= PFERR_USER_MASK;
6069 
6070 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6071 					   access, exception);
6072 }
6073 
6074 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6075 				unsigned int bytes, struct x86_exception *exception)
6076 {
6077 	/* kvm_write_guest_virt_system can pull in tons of pages. */
6078 	vcpu->arch.l1tf_flush_l1d = true;
6079 
6080 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6081 					   PFERR_WRITE_MASK, exception);
6082 }
6083 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6084 
6085 int handle_ud(struct kvm_vcpu *vcpu)
6086 {
6087 	static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6088 	int emul_type = EMULTYPE_TRAP_UD;
6089 	char sig[5]; /* ud2; .ascii "kvm" */
6090 	struct x86_exception e;
6091 
6092 	if (unlikely(!kvm_x86_ops.can_emulate_instruction(vcpu, NULL, 0)))
6093 		return 1;
6094 
6095 	if (force_emulation_prefix &&
6096 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6097 				sig, sizeof(sig), &e) == 0 &&
6098 	    memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6099 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6100 		emul_type = EMULTYPE_TRAP_UD_FORCED;
6101 	}
6102 
6103 	return kvm_emulate_instruction(vcpu, emul_type);
6104 }
6105 EXPORT_SYMBOL_GPL(handle_ud);
6106 
6107 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6108 			    gpa_t gpa, bool write)
6109 {
6110 	/* For APIC access vmexit */
6111 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6112 		return 1;
6113 
6114 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6115 		trace_vcpu_match_mmio(gva, gpa, write, true);
6116 		return 1;
6117 	}
6118 
6119 	return 0;
6120 }
6121 
6122 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6123 				gpa_t *gpa, struct x86_exception *exception,
6124 				bool write)
6125 {
6126 	u32 access = ((kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
6127 		| (write ? PFERR_WRITE_MASK : 0);
6128 
6129 	/*
6130 	 * currently PKRU is only applied to ept enabled guest so
6131 	 * there is no pkey in EPT page table for L1 guest or EPT
6132 	 * shadow page table for L2 guest.
6133 	 */
6134 	if (vcpu_match_mmio_gva(vcpu, gva)
6135 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
6136 				 vcpu->arch.mmio_access, 0, access)) {
6137 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6138 					(gva & (PAGE_SIZE - 1));
6139 		trace_vcpu_match_mmio(gva, *gpa, write, false);
6140 		return 1;
6141 	}
6142 
6143 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6144 
6145 	if (*gpa == UNMAPPED_GVA)
6146 		return -1;
6147 
6148 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6149 }
6150 
6151 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6152 			const void *val, int bytes)
6153 {
6154 	int ret;
6155 
6156 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6157 	if (ret < 0)
6158 		return 0;
6159 	kvm_page_track_write(vcpu, gpa, val, bytes);
6160 	return 1;
6161 }
6162 
6163 struct read_write_emulator_ops {
6164 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6165 				  int bytes);
6166 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6167 				  void *val, int bytes);
6168 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6169 			       int bytes, void *val);
6170 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6171 				    void *val, int bytes);
6172 	bool write;
6173 };
6174 
6175 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6176 {
6177 	if (vcpu->mmio_read_completed) {
6178 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6179 			       vcpu->mmio_fragments[0].gpa, val);
6180 		vcpu->mmio_read_completed = 0;
6181 		return 1;
6182 	}
6183 
6184 	return 0;
6185 }
6186 
6187 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6188 			void *val, int bytes)
6189 {
6190 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6191 }
6192 
6193 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6194 			 void *val, int bytes)
6195 {
6196 	return emulator_write_phys(vcpu, gpa, val, bytes);
6197 }
6198 
6199 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6200 {
6201 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6202 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
6203 }
6204 
6205 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6206 			  void *val, int bytes)
6207 {
6208 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6209 	return X86EMUL_IO_NEEDED;
6210 }
6211 
6212 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6213 			   void *val, int bytes)
6214 {
6215 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6216 
6217 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6218 	return X86EMUL_CONTINUE;
6219 }
6220 
6221 static const struct read_write_emulator_ops read_emultor = {
6222 	.read_write_prepare = read_prepare,
6223 	.read_write_emulate = read_emulate,
6224 	.read_write_mmio = vcpu_mmio_read,
6225 	.read_write_exit_mmio = read_exit_mmio,
6226 };
6227 
6228 static const struct read_write_emulator_ops write_emultor = {
6229 	.read_write_emulate = write_emulate,
6230 	.read_write_mmio = write_mmio,
6231 	.read_write_exit_mmio = write_exit_mmio,
6232 	.write = true,
6233 };
6234 
6235 static int emulator_read_write_onepage(unsigned long addr, void *val,
6236 				       unsigned int bytes,
6237 				       struct x86_exception *exception,
6238 				       struct kvm_vcpu *vcpu,
6239 				       const struct read_write_emulator_ops *ops)
6240 {
6241 	gpa_t gpa;
6242 	int handled, ret;
6243 	bool write = ops->write;
6244 	struct kvm_mmio_fragment *frag;
6245 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6246 
6247 	/*
6248 	 * If the exit was due to a NPF we may already have a GPA.
6249 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6250 	 * Note, this cannot be used on string operations since string
6251 	 * operation using rep will only have the initial GPA from the NPF
6252 	 * occurred.
6253 	 */
6254 	if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6255 	    (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6256 		gpa = ctxt->gpa_val;
6257 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6258 	} else {
6259 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6260 		if (ret < 0)
6261 			return X86EMUL_PROPAGATE_FAULT;
6262 	}
6263 
6264 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6265 		return X86EMUL_CONTINUE;
6266 
6267 	/*
6268 	 * Is this MMIO handled locally?
6269 	 */
6270 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6271 	if (handled == bytes)
6272 		return X86EMUL_CONTINUE;
6273 
6274 	gpa += handled;
6275 	bytes -= handled;
6276 	val += handled;
6277 
6278 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6279 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6280 	frag->gpa = gpa;
6281 	frag->data = val;
6282 	frag->len = bytes;
6283 	return X86EMUL_CONTINUE;
6284 }
6285 
6286 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6287 			unsigned long addr,
6288 			void *val, unsigned int bytes,
6289 			struct x86_exception *exception,
6290 			const struct read_write_emulator_ops *ops)
6291 {
6292 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6293 	gpa_t gpa;
6294 	int rc;
6295 
6296 	if (ops->read_write_prepare &&
6297 		  ops->read_write_prepare(vcpu, val, bytes))
6298 		return X86EMUL_CONTINUE;
6299 
6300 	vcpu->mmio_nr_fragments = 0;
6301 
6302 	/* Crossing a page boundary? */
6303 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6304 		int now;
6305 
6306 		now = -addr & ~PAGE_MASK;
6307 		rc = emulator_read_write_onepage(addr, val, now, exception,
6308 						 vcpu, ops);
6309 
6310 		if (rc != X86EMUL_CONTINUE)
6311 			return rc;
6312 		addr += now;
6313 		if (ctxt->mode != X86EMUL_MODE_PROT64)
6314 			addr = (u32)addr;
6315 		val += now;
6316 		bytes -= now;
6317 	}
6318 
6319 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
6320 					 vcpu, ops);
6321 	if (rc != X86EMUL_CONTINUE)
6322 		return rc;
6323 
6324 	if (!vcpu->mmio_nr_fragments)
6325 		return rc;
6326 
6327 	gpa = vcpu->mmio_fragments[0].gpa;
6328 
6329 	vcpu->mmio_needed = 1;
6330 	vcpu->mmio_cur_fragment = 0;
6331 
6332 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6333 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6334 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
6335 	vcpu->run->mmio.phys_addr = gpa;
6336 
6337 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6338 }
6339 
6340 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6341 				  unsigned long addr,
6342 				  void *val,
6343 				  unsigned int bytes,
6344 				  struct x86_exception *exception)
6345 {
6346 	return emulator_read_write(ctxt, addr, val, bytes,
6347 				   exception, &read_emultor);
6348 }
6349 
6350 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6351 			    unsigned long addr,
6352 			    const void *val,
6353 			    unsigned int bytes,
6354 			    struct x86_exception *exception)
6355 {
6356 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
6357 				   exception, &write_emultor);
6358 }
6359 
6360 #define CMPXCHG_TYPE(t, ptr, old, new) \
6361 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6362 
6363 #ifdef CONFIG_X86_64
6364 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6365 #else
6366 #  define CMPXCHG64(ptr, old, new) \
6367 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6368 #endif
6369 
6370 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6371 				     unsigned long addr,
6372 				     const void *old,
6373 				     const void *new,
6374 				     unsigned int bytes,
6375 				     struct x86_exception *exception)
6376 {
6377 	struct kvm_host_map map;
6378 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6379 	u64 page_line_mask;
6380 	gpa_t gpa;
6381 	char *kaddr;
6382 	bool exchanged;
6383 
6384 	/* guests cmpxchg8b have to be emulated atomically */
6385 	if (bytes > 8 || (bytes & (bytes - 1)))
6386 		goto emul_write;
6387 
6388 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6389 
6390 	if (gpa == UNMAPPED_GVA ||
6391 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6392 		goto emul_write;
6393 
6394 	/*
6395 	 * Emulate the atomic as a straight write to avoid #AC if SLD is
6396 	 * enabled in the host and the access splits a cache line.
6397 	 */
6398 	if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6399 		page_line_mask = ~(cache_line_size() - 1);
6400 	else
6401 		page_line_mask = PAGE_MASK;
6402 
6403 	if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6404 		goto emul_write;
6405 
6406 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6407 		goto emul_write;
6408 
6409 	kaddr = map.hva + offset_in_page(gpa);
6410 
6411 	switch (bytes) {
6412 	case 1:
6413 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6414 		break;
6415 	case 2:
6416 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6417 		break;
6418 	case 4:
6419 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6420 		break;
6421 	case 8:
6422 		exchanged = CMPXCHG64(kaddr, old, new);
6423 		break;
6424 	default:
6425 		BUG();
6426 	}
6427 
6428 	kvm_vcpu_unmap(vcpu, &map, true);
6429 
6430 	if (!exchanged)
6431 		return X86EMUL_CMPXCHG_FAILED;
6432 
6433 	kvm_page_track_write(vcpu, gpa, new, bytes);
6434 
6435 	return X86EMUL_CONTINUE;
6436 
6437 emul_write:
6438 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6439 
6440 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6441 }
6442 
6443 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6444 {
6445 	int r = 0, i;
6446 
6447 	for (i = 0; i < vcpu->arch.pio.count; i++) {
6448 		if (vcpu->arch.pio.in)
6449 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6450 					    vcpu->arch.pio.size, pd);
6451 		else
6452 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6453 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
6454 					     pd);
6455 		if (r)
6456 			break;
6457 		pd += vcpu->arch.pio.size;
6458 	}
6459 	return r;
6460 }
6461 
6462 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6463 			       unsigned short port, void *val,
6464 			       unsigned int count, bool in)
6465 {
6466 	vcpu->arch.pio.port = port;
6467 	vcpu->arch.pio.in = in;
6468 	vcpu->arch.pio.count  = count;
6469 	vcpu->arch.pio.size = size;
6470 
6471 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6472 		vcpu->arch.pio.count = 0;
6473 		return 1;
6474 	}
6475 
6476 	vcpu->run->exit_reason = KVM_EXIT_IO;
6477 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6478 	vcpu->run->io.size = size;
6479 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6480 	vcpu->run->io.count = count;
6481 	vcpu->run->io.port = port;
6482 
6483 	return 0;
6484 }
6485 
6486 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6487 			   unsigned short port, void *val, unsigned int count)
6488 {
6489 	int ret;
6490 
6491 	if (vcpu->arch.pio.count)
6492 		goto data_avail;
6493 
6494 	memset(vcpu->arch.pio_data, 0, size * count);
6495 
6496 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6497 	if (ret) {
6498 data_avail:
6499 		memcpy(val, vcpu->arch.pio_data, size * count);
6500 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6501 		vcpu->arch.pio.count = 0;
6502 		return 1;
6503 	}
6504 
6505 	return 0;
6506 }
6507 
6508 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6509 				    int size, unsigned short port, void *val,
6510 				    unsigned int count)
6511 {
6512 	return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6513 
6514 }
6515 
6516 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6517 			    unsigned short port, const void *val,
6518 			    unsigned int count)
6519 {
6520 	memcpy(vcpu->arch.pio_data, val, size * count);
6521 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6522 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6523 }
6524 
6525 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6526 				     int size, unsigned short port,
6527 				     const void *val, unsigned int count)
6528 {
6529 	return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6530 }
6531 
6532 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6533 {
6534 	return kvm_x86_ops.get_segment_base(vcpu, seg);
6535 }
6536 
6537 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6538 {
6539 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6540 }
6541 
6542 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6543 {
6544 	if (!need_emulate_wbinvd(vcpu))
6545 		return X86EMUL_CONTINUE;
6546 
6547 	if (kvm_x86_ops.has_wbinvd_exit()) {
6548 		int cpu = get_cpu();
6549 
6550 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6551 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
6552 				wbinvd_ipi, NULL, 1);
6553 		put_cpu();
6554 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6555 	} else
6556 		wbinvd();
6557 	return X86EMUL_CONTINUE;
6558 }
6559 
6560 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6561 {
6562 	kvm_emulate_wbinvd_noskip(vcpu);
6563 	return kvm_skip_emulated_instruction(vcpu);
6564 }
6565 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6566 
6567 
6568 
6569 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6570 {
6571 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6572 }
6573 
6574 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6575 			   unsigned long *dest)
6576 {
6577 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6578 }
6579 
6580 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6581 			   unsigned long value)
6582 {
6583 
6584 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6585 }
6586 
6587 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6588 {
6589 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6590 }
6591 
6592 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6593 {
6594 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6595 	unsigned long value;
6596 
6597 	switch (cr) {
6598 	case 0:
6599 		value = kvm_read_cr0(vcpu);
6600 		break;
6601 	case 2:
6602 		value = vcpu->arch.cr2;
6603 		break;
6604 	case 3:
6605 		value = kvm_read_cr3(vcpu);
6606 		break;
6607 	case 4:
6608 		value = kvm_read_cr4(vcpu);
6609 		break;
6610 	case 8:
6611 		value = kvm_get_cr8(vcpu);
6612 		break;
6613 	default:
6614 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
6615 		return 0;
6616 	}
6617 
6618 	return value;
6619 }
6620 
6621 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6622 {
6623 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6624 	int res = 0;
6625 
6626 	switch (cr) {
6627 	case 0:
6628 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6629 		break;
6630 	case 2:
6631 		vcpu->arch.cr2 = val;
6632 		break;
6633 	case 3:
6634 		res = kvm_set_cr3(vcpu, val);
6635 		break;
6636 	case 4:
6637 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6638 		break;
6639 	case 8:
6640 		res = kvm_set_cr8(vcpu, val);
6641 		break;
6642 	default:
6643 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
6644 		res = -1;
6645 	}
6646 
6647 	return res;
6648 }
6649 
6650 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6651 {
6652 	return kvm_x86_ops.get_cpl(emul_to_vcpu(ctxt));
6653 }
6654 
6655 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6656 {
6657 	kvm_x86_ops.get_gdt(emul_to_vcpu(ctxt), dt);
6658 }
6659 
6660 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6661 {
6662 	kvm_x86_ops.get_idt(emul_to_vcpu(ctxt), dt);
6663 }
6664 
6665 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6666 {
6667 	kvm_x86_ops.set_gdt(emul_to_vcpu(ctxt), dt);
6668 }
6669 
6670 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6671 {
6672 	kvm_x86_ops.set_idt(emul_to_vcpu(ctxt), dt);
6673 }
6674 
6675 static unsigned long emulator_get_cached_segment_base(
6676 	struct x86_emulate_ctxt *ctxt, int seg)
6677 {
6678 	return get_segment_base(emul_to_vcpu(ctxt), seg);
6679 }
6680 
6681 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6682 				 struct desc_struct *desc, u32 *base3,
6683 				 int seg)
6684 {
6685 	struct kvm_segment var;
6686 
6687 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6688 	*selector = var.selector;
6689 
6690 	if (var.unusable) {
6691 		memset(desc, 0, sizeof(*desc));
6692 		if (base3)
6693 			*base3 = 0;
6694 		return false;
6695 	}
6696 
6697 	if (var.g)
6698 		var.limit >>= 12;
6699 	set_desc_limit(desc, var.limit);
6700 	set_desc_base(desc, (unsigned long)var.base);
6701 #ifdef CONFIG_X86_64
6702 	if (base3)
6703 		*base3 = var.base >> 32;
6704 #endif
6705 	desc->type = var.type;
6706 	desc->s = var.s;
6707 	desc->dpl = var.dpl;
6708 	desc->p = var.present;
6709 	desc->avl = var.avl;
6710 	desc->l = var.l;
6711 	desc->d = var.db;
6712 	desc->g = var.g;
6713 
6714 	return true;
6715 }
6716 
6717 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6718 				 struct desc_struct *desc, u32 base3,
6719 				 int seg)
6720 {
6721 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6722 	struct kvm_segment var;
6723 
6724 	var.selector = selector;
6725 	var.base = get_desc_base(desc);
6726 #ifdef CONFIG_X86_64
6727 	var.base |= ((u64)base3) << 32;
6728 #endif
6729 	var.limit = get_desc_limit(desc);
6730 	if (desc->g)
6731 		var.limit = (var.limit << 12) | 0xfff;
6732 	var.type = desc->type;
6733 	var.dpl = desc->dpl;
6734 	var.db = desc->d;
6735 	var.s = desc->s;
6736 	var.l = desc->l;
6737 	var.g = desc->g;
6738 	var.avl = desc->avl;
6739 	var.present = desc->p;
6740 	var.unusable = !var.present;
6741 	var.padding = 0;
6742 
6743 	kvm_set_segment(vcpu, &var, seg);
6744 	return;
6745 }
6746 
6747 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6748 			    u32 msr_index, u64 *pdata)
6749 {
6750 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6751 	int r;
6752 
6753 	r = kvm_get_msr(vcpu, msr_index, pdata);
6754 
6755 	if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
6756 		/* Bounce to user space */
6757 		return X86EMUL_IO_NEEDED;
6758 	}
6759 
6760 	return r;
6761 }
6762 
6763 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6764 			    u32 msr_index, u64 data)
6765 {
6766 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6767 	int r;
6768 
6769 	r = kvm_set_msr(vcpu, msr_index, data);
6770 
6771 	if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
6772 		/* Bounce to user space */
6773 		return X86EMUL_IO_NEEDED;
6774 	}
6775 
6776 	return r;
6777 }
6778 
6779 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6780 {
6781 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6782 
6783 	return vcpu->arch.smbase;
6784 }
6785 
6786 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6787 {
6788 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6789 
6790 	vcpu->arch.smbase = smbase;
6791 }
6792 
6793 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6794 			      u32 pmc)
6795 {
6796 	return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6797 }
6798 
6799 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6800 			     u32 pmc, u64 *pdata)
6801 {
6802 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6803 }
6804 
6805 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6806 {
6807 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
6808 }
6809 
6810 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6811 			      struct x86_instruction_info *info,
6812 			      enum x86_intercept_stage stage)
6813 {
6814 	return kvm_x86_ops.check_intercept(emul_to_vcpu(ctxt), info, stage,
6815 					    &ctxt->exception);
6816 }
6817 
6818 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6819 			      u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
6820 			      bool exact_only)
6821 {
6822 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
6823 }
6824 
6825 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
6826 {
6827 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
6828 }
6829 
6830 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
6831 {
6832 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
6833 }
6834 
6835 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
6836 {
6837 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
6838 }
6839 
6840 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6841 {
6842 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
6843 }
6844 
6845 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6846 {
6847 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6848 }
6849 
6850 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6851 {
6852 	kvm_x86_ops.set_nmi_mask(emul_to_vcpu(ctxt), masked);
6853 }
6854 
6855 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6856 {
6857 	return emul_to_vcpu(ctxt)->arch.hflags;
6858 }
6859 
6860 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6861 {
6862 	emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6863 }
6864 
6865 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6866 				  const char *smstate)
6867 {
6868 	return kvm_x86_ops.pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6869 }
6870 
6871 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6872 {
6873 	kvm_smm_changed(emul_to_vcpu(ctxt));
6874 }
6875 
6876 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6877 {
6878 	return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6879 }
6880 
6881 static const struct x86_emulate_ops emulate_ops = {
6882 	.read_gpr            = emulator_read_gpr,
6883 	.write_gpr           = emulator_write_gpr,
6884 	.read_std            = emulator_read_std,
6885 	.write_std           = emulator_write_std,
6886 	.read_phys           = kvm_read_guest_phys_system,
6887 	.fetch               = kvm_fetch_guest_virt,
6888 	.read_emulated       = emulator_read_emulated,
6889 	.write_emulated      = emulator_write_emulated,
6890 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
6891 	.invlpg              = emulator_invlpg,
6892 	.pio_in_emulated     = emulator_pio_in_emulated,
6893 	.pio_out_emulated    = emulator_pio_out_emulated,
6894 	.get_segment         = emulator_get_segment,
6895 	.set_segment         = emulator_set_segment,
6896 	.get_cached_segment_base = emulator_get_cached_segment_base,
6897 	.get_gdt             = emulator_get_gdt,
6898 	.get_idt	     = emulator_get_idt,
6899 	.set_gdt             = emulator_set_gdt,
6900 	.set_idt	     = emulator_set_idt,
6901 	.get_cr              = emulator_get_cr,
6902 	.set_cr              = emulator_set_cr,
6903 	.cpl                 = emulator_get_cpl,
6904 	.get_dr              = emulator_get_dr,
6905 	.set_dr              = emulator_set_dr,
6906 	.get_smbase          = emulator_get_smbase,
6907 	.set_smbase          = emulator_set_smbase,
6908 	.set_msr             = emulator_set_msr,
6909 	.get_msr             = emulator_get_msr,
6910 	.check_pmc	     = emulator_check_pmc,
6911 	.read_pmc            = emulator_read_pmc,
6912 	.halt                = emulator_halt,
6913 	.wbinvd              = emulator_wbinvd,
6914 	.fix_hypercall       = emulator_fix_hypercall,
6915 	.intercept           = emulator_intercept,
6916 	.get_cpuid           = emulator_get_cpuid,
6917 	.guest_has_long_mode = emulator_guest_has_long_mode,
6918 	.guest_has_movbe     = emulator_guest_has_movbe,
6919 	.guest_has_fxsr      = emulator_guest_has_fxsr,
6920 	.set_nmi_mask        = emulator_set_nmi_mask,
6921 	.get_hflags          = emulator_get_hflags,
6922 	.set_hflags          = emulator_set_hflags,
6923 	.pre_leave_smm       = emulator_pre_leave_smm,
6924 	.post_leave_smm      = emulator_post_leave_smm,
6925 	.set_xcr             = emulator_set_xcr,
6926 };
6927 
6928 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6929 {
6930 	u32 int_shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
6931 	/*
6932 	 * an sti; sti; sequence only disable interrupts for the first
6933 	 * instruction. So, if the last instruction, be it emulated or
6934 	 * not, left the system with the INT_STI flag enabled, it
6935 	 * means that the last instruction is an sti. We should not
6936 	 * leave the flag on in this case. The same goes for mov ss
6937 	 */
6938 	if (int_shadow & mask)
6939 		mask = 0;
6940 	if (unlikely(int_shadow || mask)) {
6941 		kvm_x86_ops.set_interrupt_shadow(vcpu, mask);
6942 		if (!mask)
6943 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6944 	}
6945 }
6946 
6947 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6948 {
6949 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6950 	if (ctxt->exception.vector == PF_VECTOR)
6951 		return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
6952 
6953 	if (ctxt->exception.error_code_valid)
6954 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6955 				      ctxt->exception.error_code);
6956 	else
6957 		kvm_queue_exception(vcpu, ctxt->exception.vector);
6958 	return false;
6959 }
6960 
6961 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
6962 {
6963 	struct x86_emulate_ctxt *ctxt;
6964 
6965 	ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
6966 	if (!ctxt) {
6967 		pr_err("kvm: failed to allocate vcpu's emulator\n");
6968 		return NULL;
6969 	}
6970 
6971 	ctxt->vcpu = vcpu;
6972 	ctxt->ops = &emulate_ops;
6973 	vcpu->arch.emulate_ctxt = ctxt;
6974 
6975 	return ctxt;
6976 }
6977 
6978 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6979 {
6980 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6981 	int cs_db, cs_l;
6982 
6983 	kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6984 
6985 	ctxt->gpa_available = false;
6986 	ctxt->eflags = kvm_get_rflags(vcpu);
6987 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6988 
6989 	ctxt->eip = kvm_rip_read(vcpu);
6990 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
6991 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
6992 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
6993 		     cs_db				? X86EMUL_MODE_PROT32 :
6994 							  X86EMUL_MODE_PROT16;
6995 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6996 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6997 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6998 
6999 	init_decode_cache(ctxt);
7000 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7001 }
7002 
7003 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7004 {
7005 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7006 	int ret;
7007 
7008 	init_emulate_ctxt(vcpu);
7009 
7010 	ctxt->op_bytes = 2;
7011 	ctxt->ad_bytes = 2;
7012 	ctxt->_eip = ctxt->eip + inc_eip;
7013 	ret = emulate_int_real(ctxt, irq);
7014 
7015 	if (ret != X86EMUL_CONTINUE) {
7016 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7017 	} else {
7018 		ctxt->eip = ctxt->_eip;
7019 		kvm_rip_write(vcpu, ctxt->eip);
7020 		kvm_set_rflags(vcpu, ctxt->eflags);
7021 	}
7022 }
7023 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7024 
7025 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7026 {
7027 	++vcpu->stat.insn_emulation_fail;
7028 	trace_kvm_emulate_insn_failed(vcpu);
7029 
7030 	if (emulation_type & EMULTYPE_VMWARE_GP) {
7031 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7032 		return 1;
7033 	}
7034 
7035 	if (emulation_type & EMULTYPE_SKIP) {
7036 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7037 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7038 		vcpu->run->internal.ndata = 0;
7039 		return 0;
7040 	}
7041 
7042 	kvm_queue_exception(vcpu, UD_VECTOR);
7043 
7044 	if (!is_guest_mode(vcpu) && kvm_x86_ops.get_cpl(vcpu) == 0) {
7045 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7046 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7047 		vcpu->run->internal.ndata = 0;
7048 		return 0;
7049 	}
7050 
7051 	return 1;
7052 }
7053 
7054 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7055 				  bool write_fault_to_shadow_pgtable,
7056 				  int emulation_type)
7057 {
7058 	gpa_t gpa = cr2_or_gpa;
7059 	kvm_pfn_t pfn;
7060 
7061 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7062 		return false;
7063 
7064 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7065 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7066 		return false;
7067 
7068 	if (!vcpu->arch.mmu->direct_map) {
7069 		/*
7070 		 * Write permission should be allowed since only
7071 		 * write access need to be emulated.
7072 		 */
7073 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7074 
7075 		/*
7076 		 * If the mapping is invalid in guest, let cpu retry
7077 		 * it to generate fault.
7078 		 */
7079 		if (gpa == UNMAPPED_GVA)
7080 			return true;
7081 	}
7082 
7083 	/*
7084 	 * Do not retry the unhandleable instruction if it faults on the
7085 	 * readonly host memory, otherwise it will goto a infinite loop:
7086 	 * retry instruction -> write #PF -> emulation fail -> retry
7087 	 * instruction -> ...
7088 	 */
7089 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7090 
7091 	/*
7092 	 * If the instruction failed on the error pfn, it can not be fixed,
7093 	 * report the error to userspace.
7094 	 */
7095 	if (is_error_noslot_pfn(pfn))
7096 		return false;
7097 
7098 	kvm_release_pfn_clean(pfn);
7099 
7100 	/* The instructions are well-emulated on direct mmu. */
7101 	if (vcpu->arch.mmu->direct_map) {
7102 		unsigned int indirect_shadow_pages;
7103 
7104 		spin_lock(&vcpu->kvm->mmu_lock);
7105 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7106 		spin_unlock(&vcpu->kvm->mmu_lock);
7107 
7108 		if (indirect_shadow_pages)
7109 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7110 
7111 		return true;
7112 	}
7113 
7114 	/*
7115 	 * if emulation was due to access to shadowed page table
7116 	 * and it failed try to unshadow page and re-enter the
7117 	 * guest to let CPU execute the instruction.
7118 	 */
7119 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7120 
7121 	/*
7122 	 * If the access faults on its page table, it can not
7123 	 * be fixed by unprotecting shadow page and it should
7124 	 * be reported to userspace.
7125 	 */
7126 	return !write_fault_to_shadow_pgtable;
7127 }
7128 
7129 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7130 			      gpa_t cr2_or_gpa,  int emulation_type)
7131 {
7132 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7133 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7134 
7135 	last_retry_eip = vcpu->arch.last_retry_eip;
7136 	last_retry_addr = vcpu->arch.last_retry_addr;
7137 
7138 	/*
7139 	 * If the emulation is caused by #PF and it is non-page_table
7140 	 * writing instruction, it means the VM-EXIT is caused by shadow
7141 	 * page protected, we can zap the shadow page and retry this
7142 	 * instruction directly.
7143 	 *
7144 	 * Note: if the guest uses a non-page-table modifying instruction
7145 	 * on the PDE that points to the instruction, then we will unmap
7146 	 * the instruction and go to an infinite loop. So, we cache the
7147 	 * last retried eip and the last fault address, if we meet the eip
7148 	 * and the address again, we can break out of the potential infinite
7149 	 * loop.
7150 	 */
7151 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7152 
7153 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7154 		return false;
7155 
7156 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7157 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7158 		return false;
7159 
7160 	if (x86_page_table_writing_insn(ctxt))
7161 		return false;
7162 
7163 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7164 		return false;
7165 
7166 	vcpu->arch.last_retry_eip = ctxt->eip;
7167 	vcpu->arch.last_retry_addr = cr2_or_gpa;
7168 
7169 	if (!vcpu->arch.mmu->direct_map)
7170 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7171 
7172 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7173 
7174 	return true;
7175 }
7176 
7177 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7178 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7179 
7180 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
7181 {
7182 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
7183 		/* This is a good place to trace that we are exiting SMM.  */
7184 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
7185 
7186 		/* Process a latched INIT or SMI, if any.  */
7187 		kvm_make_request(KVM_REQ_EVENT, vcpu);
7188 	}
7189 
7190 	kvm_mmu_reset_context(vcpu);
7191 }
7192 
7193 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7194 				unsigned long *db)
7195 {
7196 	u32 dr6 = 0;
7197 	int i;
7198 	u32 enable, rwlen;
7199 
7200 	enable = dr7;
7201 	rwlen = dr7 >> 16;
7202 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7203 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7204 			dr6 |= (1 << i);
7205 	return dr6;
7206 }
7207 
7208 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7209 {
7210 	struct kvm_run *kvm_run = vcpu->run;
7211 
7212 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7213 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
7214 		kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7215 		kvm_run->debug.arch.exception = DB_VECTOR;
7216 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
7217 		return 0;
7218 	}
7219 	kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7220 	return 1;
7221 }
7222 
7223 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7224 {
7225 	unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
7226 	int r;
7227 
7228 	r = kvm_x86_ops.skip_emulated_instruction(vcpu);
7229 	if (unlikely(!r))
7230 		return 0;
7231 
7232 	/*
7233 	 * rflags is the old, "raw" value of the flags.  The new value has
7234 	 * not been saved yet.
7235 	 *
7236 	 * This is correct even for TF set by the guest, because "the
7237 	 * processor will not generate this exception after the instruction
7238 	 * that sets the TF flag".
7239 	 */
7240 	if (unlikely(rflags & X86_EFLAGS_TF))
7241 		r = kvm_vcpu_do_singlestep(vcpu);
7242 	return r;
7243 }
7244 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7245 
7246 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7247 {
7248 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7249 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7250 		struct kvm_run *kvm_run = vcpu->run;
7251 		unsigned long eip = kvm_get_linear_rip(vcpu);
7252 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7253 					   vcpu->arch.guest_debug_dr7,
7254 					   vcpu->arch.eff_db);
7255 
7256 		if (dr6 != 0) {
7257 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
7258 			kvm_run->debug.arch.pc = eip;
7259 			kvm_run->debug.arch.exception = DB_VECTOR;
7260 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
7261 			*r = 0;
7262 			return true;
7263 		}
7264 	}
7265 
7266 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7267 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7268 		unsigned long eip = kvm_get_linear_rip(vcpu);
7269 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7270 					   vcpu->arch.dr7,
7271 					   vcpu->arch.db);
7272 
7273 		if (dr6 != 0) {
7274 			kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7275 			*r = 1;
7276 			return true;
7277 		}
7278 	}
7279 
7280 	return false;
7281 }
7282 
7283 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7284 {
7285 	switch (ctxt->opcode_len) {
7286 	case 1:
7287 		switch (ctxt->b) {
7288 		case 0xe4:	/* IN */
7289 		case 0xe5:
7290 		case 0xec:
7291 		case 0xed:
7292 		case 0xe6:	/* OUT */
7293 		case 0xe7:
7294 		case 0xee:
7295 		case 0xef:
7296 		case 0x6c:	/* INS */
7297 		case 0x6d:
7298 		case 0x6e:	/* OUTS */
7299 		case 0x6f:
7300 			return true;
7301 		}
7302 		break;
7303 	case 2:
7304 		switch (ctxt->b) {
7305 		case 0x33:	/* RDPMC */
7306 			return true;
7307 		}
7308 		break;
7309 	}
7310 
7311 	return false;
7312 }
7313 
7314 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7315 			    int emulation_type, void *insn, int insn_len)
7316 {
7317 	int r;
7318 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7319 	bool writeback = true;
7320 	bool write_fault_to_spt;
7321 
7322 	if (unlikely(!kvm_x86_ops.can_emulate_instruction(vcpu, insn, insn_len)))
7323 		return 1;
7324 
7325 	vcpu->arch.l1tf_flush_l1d = true;
7326 
7327 	/*
7328 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
7329 	 * never reused.
7330 	 */
7331 	write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7332 	vcpu->arch.write_fault_to_shadow_pgtable = false;
7333 	kvm_clear_exception_queue(vcpu);
7334 
7335 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7336 		init_emulate_ctxt(vcpu);
7337 
7338 		/*
7339 		 * We will reenter on the same instruction since
7340 		 * we do not set complete_userspace_io.  This does not
7341 		 * handle watchpoints yet, those would be handled in
7342 		 * the emulate_ops.
7343 		 */
7344 		if (!(emulation_type & EMULTYPE_SKIP) &&
7345 		    kvm_vcpu_check_breakpoint(vcpu, &r))
7346 			return r;
7347 
7348 		ctxt->interruptibility = 0;
7349 		ctxt->have_exception = false;
7350 		ctxt->exception.vector = -1;
7351 		ctxt->perm_ok = false;
7352 
7353 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
7354 
7355 		r = x86_decode_insn(ctxt, insn, insn_len);
7356 
7357 		trace_kvm_emulate_insn_start(vcpu);
7358 		++vcpu->stat.insn_emulation;
7359 		if (r != EMULATION_OK)  {
7360 			if ((emulation_type & EMULTYPE_TRAP_UD) ||
7361 			    (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7362 				kvm_queue_exception(vcpu, UD_VECTOR);
7363 				return 1;
7364 			}
7365 			if (reexecute_instruction(vcpu, cr2_or_gpa,
7366 						  write_fault_to_spt,
7367 						  emulation_type))
7368 				return 1;
7369 			if (ctxt->have_exception) {
7370 				/*
7371 				 * #UD should result in just EMULATION_FAILED, and trap-like
7372 				 * exception should not be encountered during decode.
7373 				 */
7374 				WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7375 					     exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7376 				inject_emulated_exception(vcpu);
7377 				return 1;
7378 			}
7379 			return handle_emulation_failure(vcpu, emulation_type);
7380 		}
7381 	}
7382 
7383 	if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7384 	    !is_vmware_backdoor_opcode(ctxt)) {
7385 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7386 		return 1;
7387 	}
7388 
7389 	/*
7390 	 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7391 	 * for kvm_skip_emulated_instruction().  The caller is responsible for
7392 	 * updating interruptibility state and injecting single-step #DBs.
7393 	 */
7394 	if (emulation_type & EMULTYPE_SKIP) {
7395 		kvm_rip_write(vcpu, ctxt->_eip);
7396 		if (ctxt->eflags & X86_EFLAGS_RF)
7397 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7398 		return 1;
7399 	}
7400 
7401 	if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7402 		return 1;
7403 
7404 	/* this is needed for vmware backdoor interface to work since it
7405 	   changes registers values  during IO operation */
7406 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7407 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7408 		emulator_invalidate_register_cache(ctxt);
7409 	}
7410 
7411 restart:
7412 	if (emulation_type & EMULTYPE_PF) {
7413 		/* Save the faulting GPA (cr2) in the address field */
7414 		ctxt->exception.address = cr2_or_gpa;
7415 
7416 		/* With shadow page tables, cr2 contains a GVA or nGPA. */
7417 		if (vcpu->arch.mmu->direct_map) {
7418 			ctxt->gpa_available = true;
7419 			ctxt->gpa_val = cr2_or_gpa;
7420 		}
7421 	} else {
7422 		/* Sanitize the address out of an abundance of paranoia. */
7423 		ctxt->exception.address = 0;
7424 	}
7425 
7426 	r = x86_emulate_insn(ctxt);
7427 
7428 	if (r == EMULATION_INTERCEPTED)
7429 		return 1;
7430 
7431 	if (r == EMULATION_FAILED) {
7432 		if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7433 					emulation_type))
7434 			return 1;
7435 
7436 		return handle_emulation_failure(vcpu, emulation_type);
7437 	}
7438 
7439 	if (ctxt->have_exception) {
7440 		r = 1;
7441 		if (inject_emulated_exception(vcpu))
7442 			return r;
7443 	} else if (vcpu->arch.pio.count) {
7444 		if (!vcpu->arch.pio.in) {
7445 			/* FIXME: return into emulator if single-stepping.  */
7446 			vcpu->arch.pio.count = 0;
7447 		} else {
7448 			writeback = false;
7449 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
7450 		}
7451 		r = 0;
7452 	} else if (vcpu->mmio_needed) {
7453 		++vcpu->stat.mmio_exits;
7454 
7455 		if (!vcpu->mmio_is_write)
7456 			writeback = false;
7457 		r = 0;
7458 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7459 	} else if (r == EMULATION_RESTART)
7460 		goto restart;
7461 	else
7462 		r = 1;
7463 
7464 	if (writeback) {
7465 		unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
7466 		toggle_interruptibility(vcpu, ctxt->interruptibility);
7467 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7468 		if (!ctxt->have_exception ||
7469 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7470 			kvm_rip_write(vcpu, ctxt->eip);
7471 			if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7472 				r = kvm_vcpu_do_singlestep(vcpu);
7473 			if (kvm_x86_ops.update_emulated_instruction)
7474 				kvm_x86_ops.update_emulated_instruction(vcpu);
7475 			__kvm_set_rflags(vcpu, ctxt->eflags);
7476 		}
7477 
7478 		/*
7479 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7480 		 * do nothing, and it will be requested again as soon as
7481 		 * the shadow expires.  But we still need to check here,
7482 		 * because POPF has no interrupt shadow.
7483 		 */
7484 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7485 			kvm_make_request(KVM_REQ_EVENT, vcpu);
7486 	} else
7487 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7488 
7489 	return r;
7490 }
7491 
7492 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7493 {
7494 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7495 }
7496 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7497 
7498 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7499 					void *insn, int insn_len)
7500 {
7501 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7502 }
7503 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7504 
7505 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7506 {
7507 	vcpu->arch.pio.count = 0;
7508 	return 1;
7509 }
7510 
7511 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7512 {
7513 	vcpu->arch.pio.count = 0;
7514 
7515 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7516 		return 1;
7517 
7518 	return kvm_skip_emulated_instruction(vcpu);
7519 }
7520 
7521 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7522 			    unsigned short port)
7523 {
7524 	unsigned long val = kvm_rax_read(vcpu);
7525 	int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7526 
7527 	if (ret)
7528 		return ret;
7529 
7530 	/*
7531 	 * Workaround userspace that relies on old KVM behavior of %rip being
7532 	 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7533 	 */
7534 	if (port == 0x7e &&
7535 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
7536 		vcpu->arch.complete_userspace_io =
7537 			complete_fast_pio_out_port_0x7e;
7538 		kvm_skip_emulated_instruction(vcpu);
7539 	} else {
7540 		vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7541 		vcpu->arch.complete_userspace_io = complete_fast_pio_out;
7542 	}
7543 	return 0;
7544 }
7545 
7546 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7547 {
7548 	unsigned long val;
7549 
7550 	/* We should only ever be called with arch.pio.count equal to 1 */
7551 	BUG_ON(vcpu->arch.pio.count != 1);
7552 
7553 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7554 		vcpu->arch.pio.count = 0;
7555 		return 1;
7556 	}
7557 
7558 	/* For size less than 4 we merge, else we zero extend */
7559 	val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7560 
7561 	/*
7562 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7563 	 * the copy and tracing
7564 	 */
7565 	emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7566 	kvm_rax_write(vcpu, val);
7567 
7568 	return kvm_skip_emulated_instruction(vcpu);
7569 }
7570 
7571 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7572 			   unsigned short port)
7573 {
7574 	unsigned long val;
7575 	int ret;
7576 
7577 	/* For size less than 4 we merge, else we zero extend */
7578 	val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7579 
7580 	ret = emulator_pio_in(vcpu, size, port, &val, 1);
7581 	if (ret) {
7582 		kvm_rax_write(vcpu, val);
7583 		return ret;
7584 	}
7585 
7586 	vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7587 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7588 
7589 	return 0;
7590 }
7591 
7592 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7593 {
7594 	int ret;
7595 
7596 	if (in)
7597 		ret = kvm_fast_pio_in(vcpu, size, port);
7598 	else
7599 		ret = kvm_fast_pio_out(vcpu, size, port);
7600 	return ret && kvm_skip_emulated_instruction(vcpu);
7601 }
7602 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7603 
7604 static int kvmclock_cpu_down_prep(unsigned int cpu)
7605 {
7606 	__this_cpu_write(cpu_tsc_khz, 0);
7607 	return 0;
7608 }
7609 
7610 static void tsc_khz_changed(void *data)
7611 {
7612 	struct cpufreq_freqs *freq = data;
7613 	unsigned long khz = 0;
7614 
7615 	if (data)
7616 		khz = freq->new;
7617 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7618 		khz = cpufreq_quick_get(raw_smp_processor_id());
7619 	if (!khz)
7620 		khz = tsc_khz;
7621 	__this_cpu_write(cpu_tsc_khz, khz);
7622 }
7623 
7624 #ifdef CONFIG_X86_64
7625 static void kvm_hyperv_tsc_notifier(void)
7626 {
7627 	struct kvm *kvm;
7628 	struct kvm_vcpu *vcpu;
7629 	int cpu;
7630 
7631 	mutex_lock(&kvm_lock);
7632 	list_for_each_entry(kvm, &vm_list, vm_list)
7633 		kvm_make_mclock_inprogress_request(kvm);
7634 
7635 	hyperv_stop_tsc_emulation();
7636 
7637 	/* TSC frequency always matches when on Hyper-V */
7638 	for_each_present_cpu(cpu)
7639 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7640 	kvm_max_guest_tsc_khz = tsc_khz;
7641 
7642 	list_for_each_entry(kvm, &vm_list, vm_list) {
7643 		struct kvm_arch *ka = &kvm->arch;
7644 
7645 		spin_lock(&ka->pvclock_gtod_sync_lock);
7646 
7647 		pvclock_update_vm_gtod_copy(kvm);
7648 
7649 		kvm_for_each_vcpu(cpu, vcpu, kvm)
7650 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7651 
7652 		kvm_for_each_vcpu(cpu, vcpu, kvm)
7653 			kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7654 
7655 		spin_unlock(&ka->pvclock_gtod_sync_lock);
7656 	}
7657 	mutex_unlock(&kvm_lock);
7658 }
7659 #endif
7660 
7661 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7662 {
7663 	struct kvm *kvm;
7664 	struct kvm_vcpu *vcpu;
7665 	int i, send_ipi = 0;
7666 
7667 	/*
7668 	 * We allow guests to temporarily run on slowing clocks,
7669 	 * provided we notify them after, or to run on accelerating
7670 	 * clocks, provided we notify them before.  Thus time never
7671 	 * goes backwards.
7672 	 *
7673 	 * However, we have a problem.  We can't atomically update
7674 	 * the frequency of a given CPU from this function; it is
7675 	 * merely a notifier, which can be called from any CPU.
7676 	 * Changing the TSC frequency at arbitrary points in time
7677 	 * requires a recomputation of local variables related to
7678 	 * the TSC for each VCPU.  We must flag these local variables
7679 	 * to be updated and be sure the update takes place with the
7680 	 * new frequency before any guests proceed.
7681 	 *
7682 	 * Unfortunately, the combination of hotplug CPU and frequency
7683 	 * change creates an intractable locking scenario; the order
7684 	 * of when these callouts happen is undefined with respect to
7685 	 * CPU hotplug, and they can race with each other.  As such,
7686 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7687 	 * undefined; you can actually have a CPU frequency change take
7688 	 * place in between the computation of X and the setting of the
7689 	 * variable.  To protect against this problem, all updates of
7690 	 * the per_cpu tsc_khz variable are done in an interrupt
7691 	 * protected IPI, and all callers wishing to update the value
7692 	 * must wait for a synchronous IPI to complete (which is trivial
7693 	 * if the caller is on the CPU already).  This establishes the
7694 	 * necessary total order on variable updates.
7695 	 *
7696 	 * Note that because a guest time update may take place
7697 	 * anytime after the setting of the VCPU's request bit, the
7698 	 * correct TSC value must be set before the request.  However,
7699 	 * to ensure the update actually makes it to any guest which
7700 	 * starts running in hardware virtualization between the set
7701 	 * and the acquisition of the spinlock, we must also ping the
7702 	 * CPU after setting the request bit.
7703 	 *
7704 	 */
7705 
7706 	smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7707 
7708 	mutex_lock(&kvm_lock);
7709 	list_for_each_entry(kvm, &vm_list, vm_list) {
7710 		kvm_for_each_vcpu(i, vcpu, kvm) {
7711 			if (vcpu->cpu != cpu)
7712 				continue;
7713 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7714 			if (vcpu->cpu != raw_smp_processor_id())
7715 				send_ipi = 1;
7716 		}
7717 	}
7718 	mutex_unlock(&kvm_lock);
7719 
7720 	if (freq->old < freq->new && send_ipi) {
7721 		/*
7722 		 * We upscale the frequency.  Must make the guest
7723 		 * doesn't see old kvmclock values while running with
7724 		 * the new frequency, otherwise we risk the guest sees
7725 		 * time go backwards.
7726 		 *
7727 		 * In case we update the frequency for another cpu
7728 		 * (which might be in guest context) send an interrupt
7729 		 * to kick the cpu out of guest context.  Next time
7730 		 * guest context is entered kvmclock will be updated,
7731 		 * so the guest will not see stale values.
7732 		 */
7733 		smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7734 	}
7735 }
7736 
7737 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7738 				     void *data)
7739 {
7740 	struct cpufreq_freqs *freq = data;
7741 	int cpu;
7742 
7743 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7744 		return 0;
7745 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7746 		return 0;
7747 
7748 	for_each_cpu(cpu, freq->policy->cpus)
7749 		__kvmclock_cpufreq_notifier(freq, cpu);
7750 
7751 	return 0;
7752 }
7753 
7754 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7755 	.notifier_call  = kvmclock_cpufreq_notifier
7756 };
7757 
7758 static int kvmclock_cpu_online(unsigned int cpu)
7759 {
7760 	tsc_khz_changed(NULL);
7761 	return 0;
7762 }
7763 
7764 static void kvm_timer_init(void)
7765 {
7766 	max_tsc_khz = tsc_khz;
7767 
7768 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7769 #ifdef CONFIG_CPU_FREQ
7770 		struct cpufreq_policy *policy;
7771 		int cpu;
7772 
7773 		cpu = get_cpu();
7774 		policy = cpufreq_cpu_get(cpu);
7775 		if (policy) {
7776 			if (policy->cpuinfo.max_freq)
7777 				max_tsc_khz = policy->cpuinfo.max_freq;
7778 			cpufreq_cpu_put(policy);
7779 		}
7780 		put_cpu();
7781 #endif
7782 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7783 					  CPUFREQ_TRANSITION_NOTIFIER);
7784 	}
7785 
7786 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7787 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
7788 }
7789 
7790 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7791 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7792 
7793 int kvm_is_in_guest(void)
7794 {
7795 	return __this_cpu_read(current_vcpu) != NULL;
7796 }
7797 
7798 static int kvm_is_user_mode(void)
7799 {
7800 	int user_mode = 3;
7801 
7802 	if (__this_cpu_read(current_vcpu))
7803 		user_mode = kvm_x86_ops.get_cpl(__this_cpu_read(current_vcpu));
7804 
7805 	return user_mode != 0;
7806 }
7807 
7808 static unsigned long kvm_get_guest_ip(void)
7809 {
7810 	unsigned long ip = 0;
7811 
7812 	if (__this_cpu_read(current_vcpu))
7813 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7814 
7815 	return ip;
7816 }
7817 
7818 static void kvm_handle_intel_pt_intr(void)
7819 {
7820 	struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7821 
7822 	kvm_make_request(KVM_REQ_PMI, vcpu);
7823 	__set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7824 			(unsigned long *)&vcpu->arch.pmu.global_status);
7825 }
7826 
7827 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7828 	.is_in_guest		= kvm_is_in_guest,
7829 	.is_user_mode		= kvm_is_user_mode,
7830 	.get_guest_ip		= kvm_get_guest_ip,
7831 	.handle_intel_pt_intr	= kvm_handle_intel_pt_intr,
7832 };
7833 
7834 #ifdef CONFIG_X86_64
7835 static void pvclock_gtod_update_fn(struct work_struct *work)
7836 {
7837 	struct kvm *kvm;
7838 
7839 	struct kvm_vcpu *vcpu;
7840 	int i;
7841 
7842 	mutex_lock(&kvm_lock);
7843 	list_for_each_entry(kvm, &vm_list, vm_list)
7844 		kvm_for_each_vcpu(i, vcpu, kvm)
7845 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7846 	atomic_set(&kvm_guest_has_master_clock, 0);
7847 	mutex_unlock(&kvm_lock);
7848 }
7849 
7850 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7851 
7852 /*
7853  * Notification about pvclock gtod data update.
7854  */
7855 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7856 			       void *priv)
7857 {
7858 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7859 	struct timekeeper *tk = priv;
7860 
7861 	update_pvclock_gtod(tk);
7862 
7863 	/* disable master clock if host does not trust, or does not
7864 	 * use, TSC based clocksource.
7865 	 */
7866 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7867 	    atomic_read(&kvm_guest_has_master_clock) != 0)
7868 		queue_work(system_long_wq, &pvclock_gtod_work);
7869 
7870 	return 0;
7871 }
7872 
7873 static struct notifier_block pvclock_gtod_notifier = {
7874 	.notifier_call = pvclock_gtod_notify,
7875 };
7876 #endif
7877 
7878 int kvm_arch_init(void *opaque)
7879 {
7880 	struct kvm_x86_init_ops *ops = opaque;
7881 	int r;
7882 
7883 	if (kvm_x86_ops.hardware_enable) {
7884 		printk(KERN_ERR "kvm: already loaded the other module\n");
7885 		r = -EEXIST;
7886 		goto out;
7887 	}
7888 
7889 	if (!ops->cpu_has_kvm_support()) {
7890 		pr_err_ratelimited("kvm: no hardware support\n");
7891 		r = -EOPNOTSUPP;
7892 		goto out;
7893 	}
7894 	if (ops->disabled_by_bios()) {
7895 		pr_err_ratelimited("kvm: disabled by bios\n");
7896 		r = -EOPNOTSUPP;
7897 		goto out;
7898 	}
7899 
7900 	/*
7901 	 * KVM explicitly assumes that the guest has an FPU and
7902 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7903 	 * vCPU's FPU state as a fxregs_state struct.
7904 	 */
7905 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7906 		printk(KERN_ERR "kvm: inadequate fpu\n");
7907 		r = -EOPNOTSUPP;
7908 		goto out;
7909 	}
7910 
7911 	r = -ENOMEM;
7912 	x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7913 					  __alignof__(struct fpu), SLAB_ACCOUNT,
7914 					  NULL);
7915 	if (!x86_fpu_cache) {
7916 		printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7917 		goto out;
7918 	}
7919 
7920 	x86_emulator_cache = kvm_alloc_emulator_cache();
7921 	if (!x86_emulator_cache) {
7922 		pr_err("kvm: failed to allocate cache for x86 emulator\n");
7923 		goto out_free_x86_fpu_cache;
7924 	}
7925 
7926 	user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
7927 	if (!user_return_msrs) {
7928 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
7929 		goto out_free_x86_emulator_cache;
7930 	}
7931 
7932 	r = kvm_mmu_module_init();
7933 	if (r)
7934 		goto out_free_percpu;
7935 
7936 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7937 			PT_DIRTY_MASK, PT64_NX_MASK, 0,
7938 			PT_PRESENT_MASK, 0, sme_me_mask);
7939 	kvm_timer_init();
7940 
7941 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
7942 
7943 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
7944 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7945 		supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
7946 	}
7947 
7948 	kvm_lapic_init();
7949 	if (pi_inject_timer == -1)
7950 		pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7951 #ifdef CONFIG_X86_64
7952 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7953 
7954 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7955 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7956 #endif
7957 
7958 	return 0;
7959 
7960 out_free_percpu:
7961 	free_percpu(user_return_msrs);
7962 out_free_x86_emulator_cache:
7963 	kmem_cache_destroy(x86_emulator_cache);
7964 out_free_x86_fpu_cache:
7965 	kmem_cache_destroy(x86_fpu_cache);
7966 out:
7967 	return r;
7968 }
7969 
7970 void kvm_arch_exit(void)
7971 {
7972 #ifdef CONFIG_X86_64
7973 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7974 		clear_hv_tscchange_cb();
7975 #endif
7976 	kvm_lapic_exit();
7977 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7978 
7979 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7980 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7981 					    CPUFREQ_TRANSITION_NOTIFIER);
7982 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7983 #ifdef CONFIG_X86_64
7984 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7985 #endif
7986 	kvm_x86_ops.hardware_enable = NULL;
7987 	kvm_mmu_module_exit();
7988 	free_percpu(user_return_msrs);
7989 	kmem_cache_destroy(x86_fpu_cache);
7990 }
7991 
7992 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
7993 {
7994 	++vcpu->stat.halt_exits;
7995 	if (lapic_in_kernel(vcpu)) {
7996 		vcpu->arch.mp_state = state;
7997 		return 1;
7998 	} else {
7999 		vcpu->run->exit_reason = reason;
8000 		return 0;
8001 	}
8002 }
8003 
8004 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8005 {
8006 	return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8007 }
8008 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8009 
8010 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8011 {
8012 	int ret = kvm_skip_emulated_instruction(vcpu);
8013 	/*
8014 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8015 	 * KVM_EXIT_DEBUG here.
8016 	 */
8017 	return kvm_vcpu_halt(vcpu) && ret;
8018 }
8019 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8020 
8021 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8022 {
8023 	int ret = kvm_skip_emulated_instruction(vcpu);
8024 
8025 	return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8026 }
8027 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8028 
8029 #ifdef CONFIG_X86_64
8030 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8031 			        unsigned long clock_type)
8032 {
8033 	struct kvm_clock_pairing clock_pairing;
8034 	struct timespec64 ts;
8035 	u64 cycle;
8036 	int ret;
8037 
8038 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8039 		return -KVM_EOPNOTSUPP;
8040 
8041 	if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
8042 		return -KVM_EOPNOTSUPP;
8043 
8044 	clock_pairing.sec = ts.tv_sec;
8045 	clock_pairing.nsec = ts.tv_nsec;
8046 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8047 	clock_pairing.flags = 0;
8048 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8049 
8050 	ret = 0;
8051 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8052 			    sizeof(struct kvm_clock_pairing)))
8053 		ret = -KVM_EFAULT;
8054 
8055 	return ret;
8056 }
8057 #endif
8058 
8059 /*
8060  * kvm_pv_kick_cpu_op:  Kick a vcpu.
8061  *
8062  * @apicid - apicid of vcpu to be kicked.
8063  */
8064 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8065 {
8066 	struct kvm_lapic_irq lapic_irq;
8067 
8068 	lapic_irq.shorthand = APIC_DEST_NOSHORT;
8069 	lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8070 	lapic_irq.level = 0;
8071 	lapic_irq.dest_id = apicid;
8072 	lapic_irq.msi_redir_hint = false;
8073 
8074 	lapic_irq.delivery_mode = APIC_DM_REMRD;
8075 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8076 }
8077 
8078 bool kvm_apicv_activated(struct kvm *kvm)
8079 {
8080 	return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8081 }
8082 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8083 
8084 void kvm_apicv_init(struct kvm *kvm, bool enable)
8085 {
8086 	if (enable)
8087 		clear_bit(APICV_INHIBIT_REASON_DISABLE,
8088 			  &kvm->arch.apicv_inhibit_reasons);
8089 	else
8090 		set_bit(APICV_INHIBIT_REASON_DISABLE,
8091 			&kvm->arch.apicv_inhibit_reasons);
8092 }
8093 EXPORT_SYMBOL_GPL(kvm_apicv_init);
8094 
8095 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
8096 {
8097 	struct kvm_vcpu *target = NULL;
8098 	struct kvm_apic_map *map;
8099 
8100 	rcu_read_lock();
8101 	map = rcu_dereference(kvm->arch.apic_map);
8102 
8103 	if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8104 		target = map->phys_map[dest_id]->vcpu;
8105 
8106 	rcu_read_unlock();
8107 
8108 	if (target && READ_ONCE(target->ready))
8109 		kvm_vcpu_yield_to(target);
8110 }
8111 
8112 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8113 {
8114 	unsigned long nr, a0, a1, a2, a3, ret;
8115 	int op_64_bit;
8116 
8117 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
8118 		return kvm_hv_hypercall(vcpu);
8119 
8120 	nr = kvm_rax_read(vcpu);
8121 	a0 = kvm_rbx_read(vcpu);
8122 	a1 = kvm_rcx_read(vcpu);
8123 	a2 = kvm_rdx_read(vcpu);
8124 	a3 = kvm_rsi_read(vcpu);
8125 
8126 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
8127 
8128 	op_64_bit = is_64_bit_mode(vcpu);
8129 	if (!op_64_bit) {
8130 		nr &= 0xFFFFFFFF;
8131 		a0 &= 0xFFFFFFFF;
8132 		a1 &= 0xFFFFFFFF;
8133 		a2 &= 0xFFFFFFFF;
8134 		a3 &= 0xFFFFFFFF;
8135 	}
8136 
8137 	if (kvm_x86_ops.get_cpl(vcpu) != 0) {
8138 		ret = -KVM_EPERM;
8139 		goto out;
8140 	}
8141 
8142 	ret = -KVM_ENOSYS;
8143 
8144 	switch (nr) {
8145 	case KVM_HC_VAPIC_POLL_IRQ:
8146 		ret = 0;
8147 		break;
8148 	case KVM_HC_KICK_CPU:
8149 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8150 			break;
8151 
8152 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8153 		kvm_sched_yield(vcpu->kvm, a1);
8154 		ret = 0;
8155 		break;
8156 #ifdef CONFIG_X86_64
8157 	case KVM_HC_CLOCK_PAIRING:
8158 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8159 		break;
8160 #endif
8161 	case KVM_HC_SEND_IPI:
8162 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8163 			break;
8164 
8165 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8166 		break;
8167 	case KVM_HC_SCHED_YIELD:
8168 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8169 			break;
8170 
8171 		kvm_sched_yield(vcpu->kvm, a0);
8172 		ret = 0;
8173 		break;
8174 	default:
8175 		ret = -KVM_ENOSYS;
8176 		break;
8177 	}
8178 out:
8179 	if (!op_64_bit)
8180 		ret = (u32)ret;
8181 	kvm_rax_write(vcpu, ret);
8182 
8183 	++vcpu->stat.hypercalls;
8184 	return kvm_skip_emulated_instruction(vcpu);
8185 }
8186 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8187 
8188 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8189 {
8190 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8191 	char instruction[3];
8192 	unsigned long rip = kvm_rip_read(vcpu);
8193 
8194 	kvm_x86_ops.patch_hypercall(vcpu, instruction);
8195 
8196 	return emulator_write_emulated(ctxt, rip, instruction, 3,
8197 		&ctxt->exception);
8198 }
8199 
8200 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8201 {
8202 	return vcpu->run->request_interrupt_window &&
8203 		likely(!pic_in_kernel(vcpu->kvm));
8204 }
8205 
8206 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8207 {
8208 	struct kvm_run *kvm_run = vcpu->run;
8209 
8210 	/*
8211 	 * if_flag is obsolete and useless, so do not bother
8212 	 * setting it for SEV-ES guests.  Userspace can just
8213 	 * use kvm_run->ready_for_interrupt_injection.
8214 	 */
8215 	kvm_run->if_flag = !vcpu->arch.guest_state_protected
8216 		&& (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8217 
8218 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
8219 	kvm_run->cr8 = kvm_get_cr8(vcpu);
8220 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
8221 	kvm_run->ready_for_interrupt_injection =
8222 		pic_in_kernel(vcpu->kvm) ||
8223 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
8224 }
8225 
8226 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8227 {
8228 	int max_irr, tpr;
8229 
8230 	if (!kvm_x86_ops.update_cr8_intercept)
8231 		return;
8232 
8233 	if (!lapic_in_kernel(vcpu))
8234 		return;
8235 
8236 	if (vcpu->arch.apicv_active)
8237 		return;
8238 
8239 	if (!vcpu->arch.apic->vapic_addr)
8240 		max_irr = kvm_lapic_find_highest_irr(vcpu);
8241 	else
8242 		max_irr = -1;
8243 
8244 	if (max_irr != -1)
8245 		max_irr >>= 4;
8246 
8247 	tpr = kvm_lapic_get_cr8(vcpu);
8248 
8249 	kvm_x86_ops.update_cr8_intercept(vcpu, tpr, max_irr);
8250 }
8251 
8252 static void inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8253 {
8254 	int r;
8255 	bool can_inject = true;
8256 
8257 	/* try to reinject previous events if any */
8258 
8259 	if (vcpu->arch.exception.injected) {
8260 		kvm_x86_ops.queue_exception(vcpu);
8261 		can_inject = false;
8262 	}
8263 	/*
8264 	 * Do not inject an NMI or interrupt if there is a pending
8265 	 * exception.  Exceptions and interrupts are recognized at
8266 	 * instruction boundaries, i.e. the start of an instruction.
8267 	 * Trap-like exceptions, e.g. #DB, have higher priority than
8268 	 * NMIs and interrupts, i.e. traps are recognized before an
8269 	 * NMI/interrupt that's pending on the same instruction.
8270 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8271 	 * priority, but are only generated (pended) during instruction
8272 	 * execution, i.e. a pending fault-like exception means the
8273 	 * fault occurred on the *previous* instruction and must be
8274 	 * serviced prior to recognizing any new events in order to
8275 	 * fully complete the previous instruction.
8276 	 */
8277 	else if (!vcpu->arch.exception.pending) {
8278 		if (vcpu->arch.nmi_injected) {
8279 			kvm_x86_ops.set_nmi(vcpu);
8280 			can_inject = false;
8281 		} else if (vcpu->arch.interrupt.injected) {
8282 			kvm_x86_ops.set_irq(vcpu);
8283 			can_inject = false;
8284 		}
8285 	}
8286 
8287 	WARN_ON_ONCE(vcpu->arch.exception.injected &&
8288 		     vcpu->arch.exception.pending);
8289 
8290 	/*
8291 	 * Call check_nested_events() even if we reinjected a previous event
8292 	 * in order for caller to determine if it should require immediate-exit
8293 	 * from L2 to L1 due to pending L1 events which require exit
8294 	 * from L2 to L1.
8295 	 */
8296 	if (is_guest_mode(vcpu)) {
8297 		r = kvm_x86_ops.nested_ops->check_events(vcpu);
8298 		if (r < 0)
8299 			goto busy;
8300 	}
8301 
8302 	/* try to inject new event if pending */
8303 	if (vcpu->arch.exception.pending) {
8304 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
8305 					vcpu->arch.exception.has_error_code,
8306 					vcpu->arch.exception.error_code);
8307 
8308 		vcpu->arch.exception.pending = false;
8309 		vcpu->arch.exception.injected = true;
8310 
8311 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8312 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8313 					     X86_EFLAGS_RF);
8314 
8315 		if (vcpu->arch.exception.nr == DB_VECTOR) {
8316 			kvm_deliver_exception_payload(vcpu);
8317 			if (vcpu->arch.dr7 & DR7_GD) {
8318 				vcpu->arch.dr7 &= ~DR7_GD;
8319 				kvm_update_dr7(vcpu);
8320 			}
8321 		}
8322 
8323 		kvm_x86_ops.queue_exception(vcpu);
8324 		can_inject = false;
8325 	}
8326 
8327 	/*
8328 	 * Finally, inject interrupt events.  If an event cannot be injected
8329 	 * due to architectural conditions (e.g. IF=0) a window-open exit
8330 	 * will re-request KVM_REQ_EVENT.  Sometimes however an event is pending
8331 	 * and can architecturally be injected, but we cannot do it right now:
8332 	 * an interrupt could have arrived just now and we have to inject it
8333 	 * as a vmexit, or there could already an event in the queue, which is
8334 	 * indicated by can_inject.  In that case we request an immediate exit
8335 	 * in order to make progress and get back here for another iteration.
8336 	 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8337 	 */
8338 	if (vcpu->arch.smi_pending) {
8339 		r = can_inject ? kvm_x86_ops.smi_allowed(vcpu, true) : -EBUSY;
8340 		if (r < 0)
8341 			goto busy;
8342 		if (r) {
8343 			vcpu->arch.smi_pending = false;
8344 			++vcpu->arch.smi_count;
8345 			enter_smm(vcpu);
8346 			can_inject = false;
8347 		} else
8348 			kvm_x86_ops.enable_smi_window(vcpu);
8349 	}
8350 
8351 	if (vcpu->arch.nmi_pending) {
8352 		r = can_inject ? kvm_x86_ops.nmi_allowed(vcpu, true) : -EBUSY;
8353 		if (r < 0)
8354 			goto busy;
8355 		if (r) {
8356 			--vcpu->arch.nmi_pending;
8357 			vcpu->arch.nmi_injected = true;
8358 			kvm_x86_ops.set_nmi(vcpu);
8359 			can_inject = false;
8360 			WARN_ON(kvm_x86_ops.nmi_allowed(vcpu, true) < 0);
8361 		}
8362 		if (vcpu->arch.nmi_pending)
8363 			kvm_x86_ops.enable_nmi_window(vcpu);
8364 	}
8365 
8366 	if (kvm_cpu_has_injectable_intr(vcpu)) {
8367 		r = can_inject ? kvm_x86_ops.interrupt_allowed(vcpu, true) : -EBUSY;
8368 		if (r < 0)
8369 			goto busy;
8370 		if (r) {
8371 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8372 			kvm_x86_ops.set_irq(vcpu);
8373 			WARN_ON(kvm_x86_ops.interrupt_allowed(vcpu, true) < 0);
8374 		}
8375 		if (kvm_cpu_has_injectable_intr(vcpu))
8376 			kvm_x86_ops.enable_irq_window(vcpu);
8377 	}
8378 
8379 	if (is_guest_mode(vcpu) &&
8380 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
8381 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8382 		*req_immediate_exit = true;
8383 
8384 	WARN_ON(vcpu->arch.exception.pending);
8385 	return;
8386 
8387 busy:
8388 	*req_immediate_exit = true;
8389 	return;
8390 }
8391 
8392 static void process_nmi(struct kvm_vcpu *vcpu)
8393 {
8394 	unsigned limit = 2;
8395 
8396 	/*
8397 	 * x86 is limited to one NMI running, and one NMI pending after it.
8398 	 * If an NMI is already in progress, limit further NMIs to just one.
8399 	 * Otherwise, allow two (and we'll inject the first one immediately).
8400 	 */
8401 	if (kvm_x86_ops.get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
8402 		limit = 1;
8403 
8404 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8405 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8406 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8407 }
8408 
8409 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8410 {
8411 	u32 flags = 0;
8412 	flags |= seg->g       << 23;
8413 	flags |= seg->db      << 22;
8414 	flags |= seg->l       << 21;
8415 	flags |= seg->avl     << 20;
8416 	flags |= seg->present << 15;
8417 	flags |= seg->dpl     << 13;
8418 	flags |= seg->s       << 12;
8419 	flags |= seg->type    << 8;
8420 	return flags;
8421 }
8422 
8423 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8424 {
8425 	struct kvm_segment seg;
8426 	int offset;
8427 
8428 	kvm_get_segment(vcpu, &seg, n);
8429 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
8430 
8431 	if (n < 3)
8432 		offset = 0x7f84 + n * 12;
8433 	else
8434 		offset = 0x7f2c + (n - 3) * 12;
8435 
8436 	put_smstate(u32, buf, offset + 8, seg.base);
8437 	put_smstate(u32, buf, offset + 4, seg.limit);
8438 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
8439 }
8440 
8441 #ifdef CONFIG_X86_64
8442 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
8443 {
8444 	struct kvm_segment seg;
8445 	int offset;
8446 	u16 flags;
8447 
8448 	kvm_get_segment(vcpu, &seg, n);
8449 	offset = 0x7e00 + n * 16;
8450 
8451 	flags = enter_smm_get_segment_flags(&seg) >> 8;
8452 	put_smstate(u16, buf, offset, seg.selector);
8453 	put_smstate(u16, buf, offset + 2, flags);
8454 	put_smstate(u32, buf, offset + 4, seg.limit);
8455 	put_smstate(u64, buf, offset + 8, seg.base);
8456 }
8457 #endif
8458 
8459 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
8460 {
8461 	struct desc_ptr dt;
8462 	struct kvm_segment seg;
8463 	unsigned long val;
8464 	int i;
8465 
8466 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
8467 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
8468 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
8469 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
8470 
8471 	for (i = 0; i < 8; i++)
8472 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
8473 
8474 	kvm_get_dr(vcpu, 6, &val);
8475 	put_smstate(u32, buf, 0x7fcc, (u32)val);
8476 	kvm_get_dr(vcpu, 7, &val);
8477 	put_smstate(u32, buf, 0x7fc8, (u32)val);
8478 
8479 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8480 	put_smstate(u32, buf, 0x7fc4, seg.selector);
8481 	put_smstate(u32, buf, 0x7f64, seg.base);
8482 	put_smstate(u32, buf, 0x7f60, seg.limit);
8483 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
8484 
8485 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8486 	put_smstate(u32, buf, 0x7fc0, seg.selector);
8487 	put_smstate(u32, buf, 0x7f80, seg.base);
8488 	put_smstate(u32, buf, 0x7f7c, seg.limit);
8489 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
8490 
8491 	kvm_x86_ops.get_gdt(vcpu, &dt);
8492 	put_smstate(u32, buf, 0x7f74, dt.address);
8493 	put_smstate(u32, buf, 0x7f70, dt.size);
8494 
8495 	kvm_x86_ops.get_idt(vcpu, &dt);
8496 	put_smstate(u32, buf, 0x7f58, dt.address);
8497 	put_smstate(u32, buf, 0x7f54, dt.size);
8498 
8499 	for (i = 0; i < 6; i++)
8500 		enter_smm_save_seg_32(vcpu, buf, i);
8501 
8502 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
8503 
8504 	/* revision id */
8505 	put_smstate(u32, buf, 0x7efc, 0x00020000);
8506 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
8507 }
8508 
8509 #ifdef CONFIG_X86_64
8510 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
8511 {
8512 	struct desc_ptr dt;
8513 	struct kvm_segment seg;
8514 	unsigned long val;
8515 	int i;
8516 
8517 	for (i = 0; i < 16; i++)
8518 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
8519 
8520 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
8521 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
8522 
8523 	kvm_get_dr(vcpu, 6, &val);
8524 	put_smstate(u64, buf, 0x7f68, val);
8525 	kvm_get_dr(vcpu, 7, &val);
8526 	put_smstate(u64, buf, 0x7f60, val);
8527 
8528 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
8529 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
8530 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
8531 
8532 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
8533 
8534 	/* revision id */
8535 	put_smstate(u32, buf, 0x7efc, 0x00020064);
8536 
8537 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
8538 
8539 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8540 	put_smstate(u16, buf, 0x7e90, seg.selector);
8541 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
8542 	put_smstate(u32, buf, 0x7e94, seg.limit);
8543 	put_smstate(u64, buf, 0x7e98, seg.base);
8544 
8545 	kvm_x86_ops.get_idt(vcpu, &dt);
8546 	put_smstate(u32, buf, 0x7e84, dt.size);
8547 	put_smstate(u64, buf, 0x7e88, dt.address);
8548 
8549 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8550 	put_smstate(u16, buf, 0x7e70, seg.selector);
8551 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
8552 	put_smstate(u32, buf, 0x7e74, seg.limit);
8553 	put_smstate(u64, buf, 0x7e78, seg.base);
8554 
8555 	kvm_x86_ops.get_gdt(vcpu, &dt);
8556 	put_smstate(u32, buf, 0x7e64, dt.size);
8557 	put_smstate(u64, buf, 0x7e68, dt.address);
8558 
8559 	for (i = 0; i < 6; i++)
8560 		enter_smm_save_seg_64(vcpu, buf, i);
8561 }
8562 #endif
8563 
8564 static void enter_smm(struct kvm_vcpu *vcpu)
8565 {
8566 	struct kvm_segment cs, ds;
8567 	struct desc_ptr dt;
8568 	char buf[512];
8569 	u32 cr0;
8570 
8571 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
8572 	memset(buf, 0, 512);
8573 #ifdef CONFIG_X86_64
8574 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8575 		enter_smm_save_state_64(vcpu, buf);
8576 	else
8577 #endif
8578 		enter_smm_save_state_32(vcpu, buf);
8579 
8580 	/*
8581 	 * Give pre_enter_smm() a chance to make ISA-specific changes to the
8582 	 * vCPU state (e.g. leave guest mode) after we've saved the state into
8583 	 * the SMM state-save area.
8584 	 */
8585 	kvm_x86_ops.pre_enter_smm(vcpu, buf);
8586 
8587 	vcpu->arch.hflags |= HF_SMM_MASK;
8588 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
8589 
8590 	if (kvm_x86_ops.get_nmi_mask(vcpu))
8591 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
8592 	else
8593 		kvm_x86_ops.set_nmi_mask(vcpu, true);
8594 
8595 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
8596 	kvm_rip_write(vcpu, 0x8000);
8597 
8598 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
8599 	kvm_x86_ops.set_cr0(vcpu, cr0);
8600 	vcpu->arch.cr0 = cr0;
8601 
8602 	kvm_x86_ops.set_cr4(vcpu, 0);
8603 
8604 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
8605 	dt.address = dt.size = 0;
8606 	kvm_x86_ops.set_idt(vcpu, &dt);
8607 
8608 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
8609 
8610 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
8611 	cs.base = vcpu->arch.smbase;
8612 
8613 	ds.selector = 0;
8614 	ds.base = 0;
8615 
8616 	cs.limit    = ds.limit = 0xffffffff;
8617 	cs.type     = ds.type = 0x3;
8618 	cs.dpl      = ds.dpl = 0;
8619 	cs.db       = ds.db = 0;
8620 	cs.s        = ds.s = 1;
8621 	cs.l        = ds.l = 0;
8622 	cs.g        = ds.g = 1;
8623 	cs.avl      = ds.avl = 0;
8624 	cs.present  = ds.present = 1;
8625 	cs.unusable = ds.unusable = 0;
8626 	cs.padding  = ds.padding = 0;
8627 
8628 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8629 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
8630 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
8631 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
8632 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
8633 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
8634 
8635 #ifdef CONFIG_X86_64
8636 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8637 		kvm_x86_ops.set_efer(vcpu, 0);
8638 #endif
8639 
8640 	kvm_update_cpuid_runtime(vcpu);
8641 	kvm_mmu_reset_context(vcpu);
8642 }
8643 
8644 static void process_smi(struct kvm_vcpu *vcpu)
8645 {
8646 	vcpu->arch.smi_pending = true;
8647 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8648 }
8649 
8650 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
8651 				       unsigned long *vcpu_bitmap)
8652 {
8653 	cpumask_var_t cpus;
8654 
8655 	zalloc_cpumask_var(&cpus, GFP_ATOMIC);
8656 
8657 	kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
8658 				    NULL, vcpu_bitmap, cpus);
8659 
8660 	free_cpumask_var(cpus);
8661 }
8662 
8663 void kvm_make_scan_ioapic_request(struct kvm *kvm)
8664 {
8665 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
8666 }
8667 
8668 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
8669 {
8670 	if (!lapic_in_kernel(vcpu))
8671 		return;
8672 
8673 	vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
8674 	kvm_apic_update_apicv(vcpu);
8675 	kvm_x86_ops.refresh_apicv_exec_ctrl(vcpu);
8676 }
8677 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
8678 
8679 /*
8680  * NOTE: Do not hold any lock prior to calling this.
8681  *
8682  * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
8683  * locked, because it calls __x86_set_memory_region() which does
8684  * synchronize_srcu(&kvm->srcu).
8685  */
8686 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
8687 {
8688 	struct kvm_vcpu *except;
8689 	unsigned long old, new, expected;
8690 
8691 	if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
8692 	    !kvm_x86_ops.check_apicv_inhibit_reasons(bit))
8693 		return;
8694 
8695 	old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
8696 	do {
8697 		expected = new = old;
8698 		if (activate)
8699 			__clear_bit(bit, &new);
8700 		else
8701 			__set_bit(bit, &new);
8702 		if (new == old)
8703 			break;
8704 		old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
8705 	} while (old != expected);
8706 
8707 	if (!!old == !!new)
8708 		return;
8709 
8710 	trace_kvm_apicv_update_request(activate, bit);
8711 	if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
8712 		kvm_x86_ops.pre_update_apicv_exec_ctrl(kvm, activate);
8713 
8714 	/*
8715 	 * Sending request to update APICV for all other vcpus,
8716 	 * while update the calling vcpu immediately instead of
8717 	 * waiting for another #VMEXIT to handle the request.
8718 	 */
8719 	except = kvm_get_running_vcpu();
8720 	kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
8721 					 except);
8722 	if (except)
8723 		kvm_vcpu_update_apicv(except);
8724 }
8725 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
8726 
8727 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
8728 {
8729 	if (!kvm_apic_present(vcpu))
8730 		return;
8731 
8732 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
8733 
8734 	if (irqchip_split(vcpu->kvm))
8735 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
8736 	else {
8737 		if (vcpu->arch.apicv_active)
8738 			kvm_x86_ops.sync_pir_to_irr(vcpu);
8739 		if (ioapic_in_kernel(vcpu->kvm))
8740 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
8741 	}
8742 
8743 	if (is_guest_mode(vcpu))
8744 		vcpu->arch.load_eoi_exitmap_pending = true;
8745 	else
8746 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
8747 }
8748 
8749 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8750 {
8751 	u64 eoi_exit_bitmap[4];
8752 
8753 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
8754 		return;
8755 
8756 	bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
8757 		  vcpu_to_synic(vcpu)->vec_bitmap, 256);
8758 	kvm_x86_ops.load_eoi_exitmap(vcpu, eoi_exit_bitmap);
8759 }
8760 
8761 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
8762 					    unsigned long start, unsigned long end)
8763 {
8764 	unsigned long apic_address;
8765 
8766 	/*
8767 	 * The physical address of apic access page is stored in the VMCS.
8768 	 * Update it when it becomes invalid.
8769 	 */
8770 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8771 	if (start <= apic_address && apic_address < end)
8772 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
8773 }
8774 
8775 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
8776 {
8777 	if (!lapic_in_kernel(vcpu))
8778 		return;
8779 
8780 	if (!kvm_x86_ops.set_apic_access_page_addr)
8781 		return;
8782 
8783 	kvm_x86_ops.set_apic_access_page_addr(vcpu);
8784 }
8785 
8786 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
8787 {
8788 	smp_send_reschedule(vcpu->cpu);
8789 }
8790 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
8791 
8792 /*
8793  * Returns 1 to let vcpu_run() continue the guest execution loop without
8794  * exiting to the userspace.  Otherwise, the value will be returned to the
8795  * userspace.
8796  */
8797 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
8798 {
8799 	int r;
8800 	bool req_int_win =
8801 		dm_request_for_irq_injection(vcpu) &&
8802 		kvm_cpu_accept_dm_intr(vcpu);
8803 	fastpath_t exit_fastpath;
8804 
8805 	bool req_immediate_exit = false;
8806 
8807 	/* Forbid vmenter if vcpu dirty ring is soft-full */
8808 	if (unlikely(vcpu->kvm->dirty_ring_size &&
8809 		     kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
8810 		vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
8811 		trace_kvm_dirty_ring_exit(vcpu);
8812 		r = 0;
8813 		goto out;
8814 	}
8815 
8816 	if (kvm_request_pending(vcpu)) {
8817 		if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
8818 			if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
8819 				r = 0;
8820 				goto out;
8821 			}
8822 		}
8823 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
8824 			kvm_mmu_unload(vcpu);
8825 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
8826 			__kvm_migrate_timers(vcpu);
8827 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
8828 			kvm_gen_update_masterclock(vcpu->kvm);
8829 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
8830 			kvm_gen_kvmclock_update(vcpu);
8831 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
8832 			r = kvm_guest_time_update(vcpu);
8833 			if (unlikely(r))
8834 				goto out;
8835 		}
8836 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
8837 			kvm_mmu_sync_roots(vcpu);
8838 		if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
8839 			kvm_mmu_load_pgd(vcpu);
8840 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
8841 			kvm_vcpu_flush_tlb_all(vcpu);
8842 
8843 			/* Flushing all ASIDs flushes the current ASID... */
8844 			kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
8845 		}
8846 		if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
8847 			kvm_vcpu_flush_tlb_current(vcpu);
8848 		if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu))
8849 			kvm_vcpu_flush_tlb_guest(vcpu);
8850 
8851 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
8852 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
8853 			r = 0;
8854 			goto out;
8855 		}
8856 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8857 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
8858 			vcpu->mmio_needed = 0;
8859 			r = 0;
8860 			goto out;
8861 		}
8862 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
8863 			/* Page is swapped out. Do synthetic halt */
8864 			vcpu->arch.apf.halted = true;
8865 			r = 1;
8866 			goto out;
8867 		}
8868 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8869 			record_steal_time(vcpu);
8870 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
8871 			process_smi(vcpu);
8872 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
8873 			process_nmi(vcpu);
8874 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
8875 			kvm_pmu_handle_event(vcpu);
8876 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
8877 			kvm_pmu_deliver_pmi(vcpu);
8878 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8879 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8880 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
8881 				     vcpu->arch.ioapic_handled_vectors)) {
8882 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8883 				vcpu->run->eoi.vector =
8884 						vcpu->arch.pending_ioapic_eoi;
8885 				r = 0;
8886 				goto out;
8887 			}
8888 		}
8889 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8890 			vcpu_scan_ioapic(vcpu);
8891 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8892 			vcpu_load_eoi_exitmap(vcpu);
8893 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8894 			kvm_vcpu_reload_apic_access_page(vcpu);
8895 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8896 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8897 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8898 			r = 0;
8899 			goto out;
8900 		}
8901 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8902 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8903 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8904 			r = 0;
8905 			goto out;
8906 		}
8907 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8908 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8909 			vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8910 			r = 0;
8911 			goto out;
8912 		}
8913 
8914 		/*
8915 		 * KVM_REQ_HV_STIMER has to be processed after
8916 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8917 		 * depend on the guest clock being up-to-date
8918 		 */
8919 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8920 			kvm_hv_process_stimers(vcpu);
8921 		if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
8922 			kvm_vcpu_update_apicv(vcpu);
8923 		if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
8924 			kvm_check_async_pf_completion(vcpu);
8925 		if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
8926 			kvm_x86_ops.msr_filter_changed(vcpu);
8927 	}
8928 
8929 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8930 		++vcpu->stat.req_event;
8931 		kvm_apic_accept_events(vcpu);
8932 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8933 			r = 1;
8934 			goto out;
8935 		}
8936 
8937 		inject_pending_event(vcpu, &req_immediate_exit);
8938 		if (req_int_win)
8939 			kvm_x86_ops.enable_irq_window(vcpu);
8940 
8941 		if (kvm_lapic_enabled(vcpu)) {
8942 			update_cr8_intercept(vcpu);
8943 			kvm_lapic_sync_to_vapic(vcpu);
8944 		}
8945 	}
8946 
8947 	r = kvm_mmu_reload(vcpu);
8948 	if (unlikely(r)) {
8949 		goto cancel_injection;
8950 	}
8951 
8952 	preempt_disable();
8953 
8954 	kvm_x86_ops.prepare_guest_switch(vcpu);
8955 
8956 	/*
8957 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
8958 	 * IPI are then delayed after guest entry, which ensures that they
8959 	 * result in virtual interrupt delivery.
8960 	 */
8961 	local_irq_disable();
8962 	vcpu->mode = IN_GUEST_MODE;
8963 
8964 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8965 
8966 	/*
8967 	 * 1) We should set ->mode before checking ->requests.  Please see
8968 	 * the comment in kvm_vcpu_exiting_guest_mode().
8969 	 *
8970 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
8971 	 * pairs with the memory barrier implicit in pi_test_and_set_on
8972 	 * (see vmx_deliver_posted_interrupt).
8973 	 *
8974 	 * 3) This also orders the write to mode from any reads to the page
8975 	 * tables done while the VCPU is running.  Please see the comment
8976 	 * in kvm_flush_remote_tlbs.
8977 	 */
8978 	smp_mb__after_srcu_read_unlock();
8979 
8980 	/*
8981 	 * This handles the case where a posted interrupt was
8982 	 * notified with kvm_vcpu_kick.
8983 	 */
8984 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8985 		kvm_x86_ops.sync_pir_to_irr(vcpu);
8986 
8987 	if (kvm_vcpu_exit_request(vcpu)) {
8988 		vcpu->mode = OUTSIDE_GUEST_MODE;
8989 		smp_wmb();
8990 		local_irq_enable();
8991 		preempt_enable();
8992 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8993 		r = 1;
8994 		goto cancel_injection;
8995 	}
8996 
8997 	if (req_immediate_exit) {
8998 		kvm_make_request(KVM_REQ_EVENT, vcpu);
8999 		kvm_x86_ops.request_immediate_exit(vcpu);
9000 	}
9001 
9002 	fpregs_assert_state_consistent();
9003 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
9004 		switch_fpu_return();
9005 
9006 	if (unlikely(vcpu->arch.switch_db_regs)) {
9007 		set_debugreg(0, 7);
9008 		set_debugreg(vcpu->arch.eff_db[0], 0);
9009 		set_debugreg(vcpu->arch.eff_db[1], 1);
9010 		set_debugreg(vcpu->arch.eff_db[2], 2);
9011 		set_debugreg(vcpu->arch.eff_db[3], 3);
9012 		set_debugreg(vcpu->arch.dr6, 6);
9013 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9014 	}
9015 
9016 	exit_fastpath = kvm_x86_ops.run(vcpu);
9017 
9018 	/*
9019 	 * Do this here before restoring debug registers on the host.  And
9020 	 * since we do this before handling the vmexit, a DR access vmexit
9021 	 * can (a) read the correct value of the debug registers, (b) set
9022 	 * KVM_DEBUGREG_WONT_EXIT again.
9023 	 */
9024 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9025 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9026 		kvm_x86_ops.sync_dirty_debug_regs(vcpu);
9027 		kvm_update_dr0123(vcpu);
9028 		kvm_update_dr7(vcpu);
9029 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9030 	}
9031 
9032 	/*
9033 	 * If the guest has used debug registers, at least dr7
9034 	 * will be disabled while returning to the host.
9035 	 * If we don't have active breakpoints in the host, we don't
9036 	 * care about the messed up debug address registers. But if
9037 	 * we have some of them active, restore the old state.
9038 	 */
9039 	if (hw_breakpoint_active())
9040 		hw_breakpoint_restore();
9041 
9042 	vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9043 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9044 
9045 	vcpu->mode = OUTSIDE_GUEST_MODE;
9046 	smp_wmb();
9047 
9048 	kvm_x86_ops.handle_exit_irqoff(vcpu);
9049 
9050 	/*
9051 	 * Consume any pending interrupts, including the possible source of
9052 	 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9053 	 * An instruction is required after local_irq_enable() to fully unblock
9054 	 * interrupts on processors that implement an interrupt shadow, the
9055 	 * stat.exits increment will do nicely.
9056 	 */
9057 	kvm_before_interrupt(vcpu);
9058 	local_irq_enable();
9059 	++vcpu->stat.exits;
9060 	local_irq_disable();
9061 	kvm_after_interrupt(vcpu);
9062 
9063 	if (lapic_in_kernel(vcpu)) {
9064 		s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9065 		if (delta != S64_MIN) {
9066 			trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9067 			vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9068 		}
9069 	}
9070 
9071 	local_irq_enable();
9072 	preempt_enable();
9073 
9074 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9075 
9076 	/*
9077 	 * Profile KVM exit RIPs:
9078 	 */
9079 	if (unlikely(prof_on == KVM_PROFILING)) {
9080 		unsigned long rip = kvm_rip_read(vcpu);
9081 		profile_hit(KVM_PROFILING, (void *)rip);
9082 	}
9083 
9084 	if (unlikely(vcpu->arch.tsc_always_catchup))
9085 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9086 
9087 	if (vcpu->arch.apic_attention)
9088 		kvm_lapic_sync_from_vapic(vcpu);
9089 
9090 	r = kvm_x86_ops.handle_exit(vcpu, exit_fastpath);
9091 	return r;
9092 
9093 cancel_injection:
9094 	if (req_immediate_exit)
9095 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9096 	kvm_x86_ops.cancel_injection(vcpu);
9097 	if (unlikely(vcpu->arch.apic_attention))
9098 		kvm_lapic_sync_from_vapic(vcpu);
9099 out:
9100 	return r;
9101 }
9102 
9103 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9104 {
9105 	if (!kvm_arch_vcpu_runnable(vcpu) &&
9106 	    (!kvm_x86_ops.pre_block || kvm_x86_ops.pre_block(vcpu) == 0)) {
9107 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9108 		kvm_vcpu_block(vcpu);
9109 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9110 
9111 		if (kvm_x86_ops.post_block)
9112 			kvm_x86_ops.post_block(vcpu);
9113 
9114 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9115 			return 1;
9116 	}
9117 
9118 	kvm_apic_accept_events(vcpu);
9119 	switch(vcpu->arch.mp_state) {
9120 	case KVM_MP_STATE_HALTED:
9121 	case KVM_MP_STATE_AP_RESET_HOLD:
9122 		vcpu->arch.pv.pv_unhalted = false;
9123 		vcpu->arch.mp_state =
9124 			KVM_MP_STATE_RUNNABLE;
9125 		fallthrough;
9126 	case KVM_MP_STATE_RUNNABLE:
9127 		vcpu->arch.apf.halted = false;
9128 		break;
9129 	case KVM_MP_STATE_INIT_RECEIVED:
9130 		break;
9131 	default:
9132 		return -EINTR;
9133 	}
9134 	return 1;
9135 }
9136 
9137 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9138 {
9139 	if (is_guest_mode(vcpu))
9140 		kvm_x86_ops.nested_ops->check_events(vcpu);
9141 
9142 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9143 		!vcpu->arch.apf.halted);
9144 }
9145 
9146 static int vcpu_run(struct kvm_vcpu *vcpu)
9147 {
9148 	int r;
9149 	struct kvm *kvm = vcpu->kvm;
9150 
9151 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9152 	vcpu->arch.l1tf_flush_l1d = true;
9153 
9154 	for (;;) {
9155 		if (kvm_vcpu_running(vcpu)) {
9156 			r = vcpu_enter_guest(vcpu);
9157 		} else {
9158 			r = vcpu_block(kvm, vcpu);
9159 		}
9160 
9161 		if (r <= 0)
9162 			break;
9163 
9164 		kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
9165 		if (kvm_cpu_has_pending_timer(vcpu))
9166 			kvm_inject_pending_timer_irqs(vcpu);
9167 
9168 		if (dm_request_for_irq_injection(vcpu) &&
9169 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9170 			r = 0;
9171 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9172 			++vcpu->stat.request_irq_exits;
9173 			break;
9174 		}
9175 
9176 		if (__xfer_to_guest_mode_work_pending()) {
9177 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9178 			r = xfer_to_guest_mode_handle_work(vcpu);
9179 			if (r)
9180 				return r;
9181 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9182 		}
9183 	}
9184 
9185 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9186 
9187 	return r;
9188 }
9189 
9190 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9191 {
9192 	int r;
9193 
9194 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9195 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9196 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9197 	return r;
9198 }
9199 
9200 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9201 {
9202 	BUG_ON(!vcpu->arch.pio.count);
9203 
9204 	return complete_emulated_io(vcpu);
9205 }
9206 
9207 /*
9208  * Implements the following, as a state machine:
9209  *
9210  * read:
9211  *   for each fragment
9212  *     for each mmio piece in the fragment
9213  *       write gpa, len
9214  *       exit
9215  *       copy data
9216  *   execute insn
9217  *
9218  * write:
9219  *   for each fragment
9220  *     for each mmio piece in the fragment
9221  *       write gpa, len
9222  *       copy data
9223  *       exit
9224  */
9225 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9226 {
9227 	struct kvm_run *run = vcpu->run;
9228 	struct kvm_mmio_fragment *frag;
9229 	unsigned len;
9230 
9231 	BUG_ON(!vcpu->mmio_needed);
9232 
9233 	/* Complete previous fragment */
9234 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9235 	len = min(8u, frag->len);
9236 	if (!vcpu->mmio_is_write)
9237 		memcpy(frag->data, run->mmio.data, len);
9238 
9239 	if (frag->len <= 8) {
9240 		/* Switch to the next fragment. */
9241 		frag++;
9242 		vcpu->mmio_cur_fragment++;
9243 	} else {
9244 		/* Go forward to the next mmio piece. */
9245 		frag->data += len;
9246 		frag->gpa += len;
9247 		frag->len -= len;
9248 	}
9249 
9250 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9251 		vcpu->mmio_needed = 0;
9252 
9253 		/* FIXME: return into emulator if single-stepping.  */
9254 		if (vcpu->mmio_is_write)
9255 			return 1;
9256 		vcpu->mmio_read_completed = 1;
9257 		return complete_emulated_io(vcpu);
9258 	}
9259 
9260 	run->exit_reason = KVM_EXIT_MMIO;
9261 	run->mmio.phys_addr = frag->gpa;
9262 	if (vcpu->mmio_is_write)
9263 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9264 	run->mmio.len = min(8u, frag->len);
9265 	run->mmio.is_write = vcpu->mmio_is_write;
9266 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9267 	return 0;
9268 }
9269 
9270 static void kvm_save_current_fpu(struct fpu *fpu)
9271 {
9272 	/*
9273 	 * If the target FPU state is not resident in the CPU registers, just
9274 	 * memcpy() from current, else save CPU state directly to the target.
9275 	 */
9276 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
9277 		memcpy(&fpu->state, &current->thread.fpu.state,
9278 		       fpu_kernel_xstate_size);
9279 	else
9280 		copy_fpregs_to_fpstate(fpu);
9281 }
9282 
9283 /* Swap (qemu) user FPU context for the guest FPU context. */
9284 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9285 {
9286 	fpregs_lock();
9287 
9288 	kvm_save_current_fpu(vcpu->arch.user_fpu);
9289 
9290 	/*
9291 	 * Guests with protected state can't have it set by the hypervisor,
9292 	 * so skip trying to set it.
9293 	 */
9294 	if (vcpu->arch.guest_fpu)
9295 		/* PKRU is separately restored in kvm_x86_ops.run. */
9296 		__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
9297 					~XFEATURE_MASK_PKRU);
9298 
9299 	fpregs_mark_activate();
9300 	fpregs_unlock();
9301 
9302 	trace_kvm_fpu(1);
9303 }
9304 
9305 /* When vcpu_run ends, restore user space FPU context. */
9306 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9307 {
9308 	fpregs_lock();
9309 
9310 	/*
9311 	 * Guests with protected state can't have it read by the hypervisor,
9312 	 * so skip trying to save it.
9313 	 */
9314 	if (vcpu->arch.guest_fpu)
9315 		kvm_save_current_fpu(vcpu->arch.guest_fpu);
9316 
9317 	copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
9318 
9319 	fpregs_mark_activate();
9320 	fpregs_unlock();
9321 
9322 	++vcpu->stat.fpu_reload;
9323 	trace_kvm_fpu(0);
9324 }
9325 
9326 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9327 {
9328 	struct kvm_run *kvm_run = vcpu->run;
9329 	int r;
9330 
9331 	vcpu_load(vcpu);
9332 	kvm_sigset_activate(vcpu);
9333 	kvm_load_guest_fpu(vcpu);
9334 
9335 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9336 		if (kvm_run->immediate_exit) {
9337 			r = -EINTR;
9338 			goto out;
9339 		}
9340 		kvm_vcpu_block(vcpu);
9341 		kvm_apic_accept_events(vcpu);
9342 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9343 		r = -EAGAIN;
9344 		if (signal_pending(current)) {
9345 			r = -EINTR;
9346 			kvm_run->exit_reason = KVM_EXIT_INTR;
9347 			++vcpu->stat.signal_exits;
9348 		}
9349 		goto out;
9350 	}
9351 
9352 	if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
9353 		r = -EINVAL;
9354 		goto out;
9355 	}
9356 
9357 	if (kvm_run->kvm_dirty_regs) {
9358 		r = sync_regs(vcpu);
9359 		if (r != 0)
9360 			goto out;
9361 	}
9362 
9363 	/* re-sync apic's tpr */
9364 	if (!lapic_in_kernel(vcpu)) {
9365 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
9366 			r = -EINVAL;
9367 			goto out;
9368 		}
9369 	}
9370 
9371 	if (unlikely(vcpu->arch.complete_userspace_io)) {
9372 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
9373 		vcpu->arch.complete_userspace_io = NULL;
9374 		r = cui(vcpu);
9375 		if (r <= 0)
9376 			goto out;
9377 	} else
9378 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
9379 
9380 	if (kvm_run->immediate_exit)
9381 		r = -EINTR;
9382 	else
9383 		r = vcpu_run(vcpu);
9384 
9385 out:
9386 	kvm_put_guest_fpu(vcpu);
9387 	if (kvm_run->kvm_valid_regs)
9388 		store_regs(vcpu);
9389 	post_kvm_run_save(vcpu);
9390 	kvm_sigset_deactivate(vcpu);
9391 
9392 	vcpu_put(vcpu);
9393 	return r;
9394 }
9395 
9396 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9397 {
9398 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
9399 		/*
9400 		 * We are here if userspace calls get_regs() in the middle of
9401 		 * instruction emulation. Registers state needs to be copied
9402 		 * back from emulation context to vcpu. Userspace shouldn't do
9403 		 * that usually, but some bad designed PV devices (vmware
9404 		 * backdoor interface) need this to work
9405 		 */
9406 		emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
9407 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9408 	}
9409 	regs->rax = kvm_rax_read(vcpu);
9410 	regs->rbx = kvm_rbx_read(vcpu);
9411 	regs->rcx = kvm_rcx_read(vcpu);
9412 	regs->rdx = kvm_rdx_read(vcpu);
9413 	regs->rsi = kvm_rsi_read(vcpu);
9414 	regs->rdi = kvm_rdi_read(vcpu);
9415 	regs->rsp = kvm_rsp_read(vcpu);
9416 	regs->rbp = kvm_rbp_read(vcpu);
9417 #ifdef CONFIG_X86_64
9418 	regs->r8 = kvm_r8_read(vcpu);
9419 	regs->r9 = kvm_r9_read(vcpu);
9420 	regs->r10 = kvm_r10_read(vcpu);
9421 	regs->r11 = kvm_r11_read(vcpu);
9422 	regs->r12 = kvm_r12_read(vcpu);
9423 	regs->r13 = kvm_r13_read(vcpu);
9424 	regs->r14 = kvm_r14_read(vcpu);
9425 	regs->r15 = kvm_r15_read(vcpu);
9426 #endif
9427 
9428 	regs->rip = kvm_rip_read(vcpu);
9429 	regs->rflags = kvm_get_rflags(vcpu);
9430 }
9431 
9432 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9433 {
9434 	vcpu_load(vcpu);
9435 	__get_regs(vcpu, regs);
9436 	vcpu_put(vcpu);
9437 	return 0;
9438 }
9439 
9440 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9441 {
9442 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
9443 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9444 
9445 	kvm_rax_write(vcpu, regs->rax);
9446 	kvm_rbx_write(vcpu, regs->rbx);
9447 	kvm_rcx_write(vcpu, regs->rcx);
9448 	kvm_rdx_write(vcpu, regs->rdx);
9449 	kvm_rsi_write(vcpu, regs->rsi);
9450 	kvm_rdi_write(vcpu, regs->rdi);
9451 	kvm_rsp_write(vcpu, regs->rsp);
9452 	kvm_rbp_write(vcpu, regs->rbp);
9453 #ifdef CONFIG_X86_64
9454 	kvm_r8_write(vcpu, regs->r8);
9455 	kvm_r9_write(vcpu, regs->r9);
9456 	kvm_r10_write(vcpu, regs->r10);
9457 	kvm_r11_write(vcpu, regs->r11);
9458 	kvm_r12_write(vcpu, regs->r12);
9459 	kvm_r13_write(vcpu, regs->r13);
9460 	kvm_r14_write(vcpu, regs->r14);
9461 	kvm_r15_write(vcpu, regs->r15);
9462 #endif
9463 
9464 	kvm_rip_write(vcpu, regs->rip);
9465 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
9466 
9467 	vcpu->arch.exception.pending = false;
9468 
9469 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9470 }
9471 
9472 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9473 {
9474 	vcpu_load(vcpu);
9475 	__set_regs(vcpu, regs);
9476 	vcpu_put(vcpu);
9477 	return 0;
9478 }
9479 
9480 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
9481 {
9482 	struct kvm_segment cs;
9483 
9484 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9485 	*db = cs.db;
9486 	*l = cs.l;
9487 }
9488 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
9489 
9490 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9491 {
9492 	struct desc_ptr dt;
9493 
9494 	if (vcpu->arch.guest_state_protected)
9495 		goto skip_protected_regs;
9496 
9497 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9498 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9499 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9500 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9501 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9502 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9503 
9504 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9505 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9506 
9507 	kvm_x86_ops.get_idt(vcpu, &dt);
9508 	sregs->idt.limit = dt.size;
9509 	sregs->idt.base = dt.address;
9510 	kvm_x86_ops.get_gdt(vcpu, &dt);
9511 	sregs->gdt.limit = dt.size;
9512 	sregs->gdt.base = dt.address;
9513 
9514 	sregs->cr2 = vcpu->arch.cr2;
9515 	sregs->cr3 = kvm_read_cr3(vcpu);
9516 
9517 skip_protected_regs:
9518 	sregs->cr0 = kvm_read_cr0(vcpu);
9519 	sregs->cr4 = kvm_read_cr4(vcpu);
9520 	sregs->cr8 = kvm_get_cr8(vcpu);
9521 	sregs->efer = vcpu->arch.efer;
9522 	sregs->apic_base = kvm_get_apic_base(vcpu);
9523 
9524 	memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
9525 
9526 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
9527 		set_bit(vcpu->arch.interrupt.nr,
9528 			(unsigned long *)sregs->interrupt_bitmap);
9529 }
9530 
9531 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
9532 				  struct kvm_sregs *sregs)
9533 {
9534 	vcpu_load(vcpu);
9535 	__get_sregs(vcpu, sregs);
9536 	vcpu_put(vcpu);
9537 	return 0;
9538 }
9539 
9540 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
9541 				    struct kvm_mp_state *mp_state)
9542 {
9543 	vcpu_load(vcpu);
9544 	if (kvm_mpx_supported())
9545 		kvm_load_guest_fpu(vcpu);
9546 
9547 	kvm_apic_accept_events(vcpu);
9548 	if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
9549 	     vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
9550 	    vcpu->arch.pv.pv_unhalted)
9551 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
9552 	else
9553 		mp_state->mp_state = vcpu->arch.mp_state;
9554 
9555 	if (kvm_mpx_supported())
9556 		kvm_put_guest_fpu(vcpu);
9557 	vcpu_put(vcpu);
9558 	return 0;
9559 }
9560 
9561 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
9562 				    struct kvm_mp_state *mp_state)
9563 {
9564 	int ret = -EINVAL;
9565 
9566 	vcpu_load(vcpu);
9567 
9568 	if (!lapic_in_kernel(vcpu) &&
9569 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
9570 		goto out;
9571 
9572 	/*
9573 	 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
9574 	 * INIT state; latched init should be reported using
9575 	 * KVM_SET_VCPU_EVENTS, so reject it here.
9576 	 */
9577 	if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
9578 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
9579 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
9580 		goto out;
9581 
9582 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
9583 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
9584 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
9585 	} else
9586 		vcpu->arch.mp_state = mp_state->mp_state;
9587 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9588 
9589 	ret = 0;
9590 out:
9591 	vcpu_put(vcpu);
9592 	return ret;
9593 }
9594 
9595 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
9596 		    int reason, bool has_error_code, u32 error_code)
9597 {
9598 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
9599 	int ret;
9600 
9601 	init_emulate_ctxt(vcpu);
9602 
9603 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9604 				   has_error_code, error_code);
9605 	if (ret) {
9606 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
9607 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
9608 		vcpu->run->internal.ndata = 0;
9609 		return 0;
9610 	}
9611 
9612 	kvm_rip_write(vcpu, ctxt->eip);
9613 	kvm_set_rflags(vcpu, ctxt->eflags);
9614 	return 1;
9615 }
9616 EXPORT_SYMBOL_GPL(kvm_task_switch);
9617 
9618 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9619 {
9620 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
9621 		/*
9622 		 * When EFER.LME and CR0.PG are set, the processor is in
9623 		 * 64-bit mode (though maybe in a 32-bit code segment).
9624 		 * CR4.PAE and EFER.LMA must be set.
9625 		 */
9626 		if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
9627 			return false;
9628 		if (sregs->cr3 & vcpu->arch.cr3_lm_rsvd_bits)
9629 			return false;
9630 	} else {
9631 		/*
9632 		 * Not in 64-bit mode: EFER.LMA is clear and the code
9633 		 * segment cannot be 64-bit.
9634 		 */
9635 		if (sregs->efer & EFER_LMA || sregs->cs.l)
9636 			return false;
9637 	}
9638 
9639 	return kvm_is_valid_cr4(vcpu, sregs->cr4);
9640 }
9641 
9642 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9643 {
9644 	struct msr_data apic_base_msr;
9645 	int mmu_reset_needed = 0;
9646 	int pending_vec, max_bits, idx;
9647 	struct desc_ptr dt;
9648 	int ret = -EINVAL;
9649 
9650 	if (!kvm_is_valid_sregs(vcpu, sregs))
9651 		goto out;
9652 
9653 	apic_base_msr.data = sregs->apic_base;
9654 	apic_base_msr.host_initiated = true;
9655 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
9656 		goto out;
9657 
9658 	if (vcpu->arch.guest_state_protected)
9659 		goto skip_protected_regs;
9660 
9661 	dt.size = sregs->idt.limit;
9662 	dt.address = sregs->idt.base;
9663 	kvm_x86_ops.set_idt(vcpu, &dt);
9664 	dt.size = sregs->gdt.limit;
9665 	dt.address = sregs->gdt.base;
9666 	kvm_x86_ops.set_gdt(vcpu, &dt);
9667 
9668 	vcpu->arch.cr2 = sregs->cr2;
9669 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
9670 	vcpu->arch.cr3 = sregs->cr3;
9671 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
9672 
9673 	kvm_set_cr8(vcpu, sregs->cr8);
9674 
9675 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
9676 	kvm_x86_ops.set_efer(vcpu, sregs->efer);
9677 
9678 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
9679 	kvm_x86_ops.set_cr0(vcpu, sregs->cr0);
9680 	vcpu->arch.cr0 = sregs->cr0;
9681 
9682 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
9683 	kvm_x86_ops.set_cr4(vcpu, sregs->cr4);
9684 
9685 	idx = srcu_read_lock(&vcpu->kvm->srcu);
9686 	if (is_pae_paging(vcpu)) {
9687 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
9688 		mmu_reset_needed = 1;
9689 	}
9690 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
9691 
9692 	if (mmu_reset_needed)
9693 		kvm_mmu_reset_context(vcpu);
9694 
9695 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9696 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9697 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9698 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9699 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9700 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9701 
9702 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9703 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9704 
9705 	update_cr8_intercept(vcpu);
9706 
9707 	/* Older userspace won't unhalt the vcpu on reset. */
9708 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9709 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
9710 	    !is_protmode(vcpu))
9711 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9712 
9713 skip_protected_regs:
9714 	max_bits = KVM_NR_INTERRUPTS;
9715 	pending_vec = find_first_bit(
9716 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
9717 	if (pending_vec < max_bits) {
9718 		kvm_queue_interrupt(vcpu, pending_vec, false);
9719 		pr_debug("Set back pending irq %d\n", pending_vec);
9720 	}
9721 
9722 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9723 
9724 	ret = 0;
9725 out:
9726 	return ret;
9727 }
9728 
9729 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
9730 				  struct kvm_sregs *sregs)
9731 {
9732 	int ret;
9733 
9734 	vcpu_load(vcpu);
9735 	ret = __set_sregs(vcpu, sregs);
9736 	vcpu_put(vcpu);
9737 	return ret;
9738 }
9739 
9740 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
9741 					struct kvm_guest_debug *dbg)
9742 {
9743 	unsigned long rflags;
9744 	int i, r;
9745 
9746 	if (vcpu->arch.guest_state_protected)
9747 		return -EINVAL;
9748 
9749 	vcpu_load(vcpu);
9750 
9751 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
9752 		r = -EBUSY;
9753 		if (vcpu->arch.exception.pending)
9754 			goto out;
9755 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
9756 			kvm_queue_exception(vcpu, DB_VECTOR);
9757 		else
9758 			kvm_queue_exception(vcpu, BP_VECTOR);
9759 	}
9760 
9761 	/*
9762 	 * Read rflags as long as potentially injected trace flags are still
9763 	 * filtered out.
9764 	 */
9765 	rflags = kvm_get_rflags(vcpu);
9766 
9767 	vcpu->guest_debug = dbg->control;
9768 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
9769 		vcpu->guest_debug = 0;
9770 
9771 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
9772 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
9773 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
9774 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
9775 	} else {
9776 		for (i = 0; i < KVM_NR_DB_REGS; i++)
9777 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
9778 	}
9779 	kvm_update_dr7(vcpu);
9780 
9781 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9782 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
9783 			get_segment_base(vcpu, VCPU_SREG_CS);
9784 
9785 	/*
9786 	 * Trigger an rflags update that will inject or remove the trace
9787 	 * flags.
9788 	 */
9789 	kvm_set_rflags(vcpu, rflags);
9790 
9791 	kvm_x86_ops.update_exception_bitmap(vcpu);
9792 
9793 	r = 0;
9794 
9795 out:
9796 	vcpu_put(vcpu);
9797 	return r;
9798 }
9799 
9800 /*
9801  * Translate a guest virtual address to a guest physical address.
9802  */
9803 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
9804 				    struct kvm_translation *tr)
9805 {
9806 	unsigned long vaddr = tr->linear_address;
9807 	gpa_t gpa;
9808 	int idx;
9809 
9810 	vcpu_load(vcpu);
9811 
9812 	idx = srcu_read_lock(&vcpu->kvm->srcu);
9813 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
9814 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
9815 	tr->physical_address = gpa;
9816 	tr->valid = gpa != UNMAPPED_GVA;
9817 	tr->writeable = 1;
9818 	tr->usermode = 0;
9819 
9820 	vcpu_put(vcpu);
9821 	return 0;
9822 }
9823 
9824 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9825 {
9826 	struct fxregs_state *fxsave;
9827 
9828 	if (!vcpu->arch.guest_fpu)
9829 		return 0;
9830 
9831 	vcpu_load(vcpu);
9832 
9833 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9834 	memcpy(fpu->fpr, fxsave->st_space, 128);
9835 	fpu->fcw = fxsave->cwd;
9836 	fpu->fsw = fxsave->swd;
9837 	fpu->ftwx = fxsave->twd;
9838 	fpu->last_opcode = fxsave->fop;
9839 	fpu->last_ip = fxsave->rip;
9840 	fpu->last_dp = fxsave->rdp;
9841 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
9842 
9843 	vcpu_put(vcpu);
9844 	return 0;
9845 }
9846 
9847 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9848 {
9849 	struct fxregs_state *fxsave;
9850 
9851 	if (!vcpu->arch.guest_fpu)
9852 		return 0;
9853 
9854 	vcpu_load(vcpu);
9855 
9856 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9857 
9858 	memcpy(fxsave->st_space, fpu->fpr, 128);
9859 	fxsave->cwd = fpu->fcw;
9860 	fxsave->swd = fpu->fsw;
9861 	fxsave->twd = fpu->ftwx;
9862 	fxsave->fop = fpu->last_opcode;
9863 	fxsave->rip = fpu->last_ip;
9864 	fxsave->rdp = fpu->last_dp;
9865 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
9866 
9867 	vcpu_put(vcpu);
9868 	return 0;
9869 }
9870 
9871 static void store_regs(struct kvm_vcpu *vcpu)
9872 {
9873 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
9874 
9875 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
9876 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
9877 
9878 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
9879 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
9880 
9881 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
9882 		kvm_vcpu_ioctl_x86_get_vcpu_events(
9883 				vcpu, &vcpu->run->s.regs.events);
9884 }
9885 
9886 static int sync_regs(struct kvm_vcpu *vcpu)
9887 {
9888 	if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
9889 		return -EINVAL;
9890 
9891 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
9892 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
9893 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
9894 	}
9895 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
9896 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
9897 			return -EINVAL;
9898 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
9899 	}
9900 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
9901 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9902 				vcpu, &vcpu->run->s.regs.events))
9903 			return -EINVAL;
9904 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9905 	}
9906 
9907 	return 0;
9908 }
9909 
9910 static void fx_init(struct kvm_vcpu *vcpu)
9911 {
9912 	if (!vcpu->arch.guest_fpu)
9913 		return;
9914 
9915 	fpstate_init(&vcpu->arch.guest_fpu->state);
9916 	if (boot_cpu_has(X86_FEATURE_XSAVES))
9917 		vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9918 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
9919 
9920 	/*
9921 	 * Ensure guest xcr0 is valid for loading
9922 	 */
9923 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9924 
9925 	vcpu->arch.cr0 |= X86_CR0_ET;
9926 }
9927 
9928 void kvm_free_guest_fpu(struct kvm_vcpu *vcpu)
9929 {
9930 	if (vcpu->arch.guest_fpu) {
9931 		kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9932 		vcpu->arch.guest_fpu = NULL;
9933 	}
9934 }
9935 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu);
9936 
9937 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
9938 {
9939 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9940 		pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
9941 			     "guest TSC will not be reliable\n");
9942 
9943 	return 0;
9944 }
9945 
9946 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
9947 {
9948 	struct page *page;
9949 	int r;
9950 
9951 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9952 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9953 	else
9954 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9955 
9956 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
9957 
9958 	r = kvm_mmu_create(vcpu);
9959 	if (r < 0)
9960 		return r;
9961 
9962 	if (irqchip_in_kernel(vcpu->kvm)) {
9963 		r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9964 		if (r < 0)
9965 			goto fail_mmu_destroy;
9966 		if (kvm_apicv_activated(vcpu->kvm))
9967 			vcpu->arch.apicv_active = true;
9968 	} else
9969 		static_key_slow_inc(&kvm_no_apic_vcpu);
9970 
9971 	r = -ENOMEM;
9972 
9973 	page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
9974 	if (!page)
9975 		goto fail_free_lapic;
9976 	vcpu->arch.pio_data = page_address(page);
9977 
9978 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9979 				       GFP_KERNEL_ACCOUNT);
9980 	if (!vcpu->arch.mce_banks)
9981 		goto fail_free_pio_data;
9982 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9983 
9984 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9985 				GFP_KERNEL_ACCOUNT))
9986 		goto fail_free_mce_banks;
9987 
9988 	if (!alloc_emulate_ctxt(vcpu))
9989 		goto free_wbinvd_dirty_mask;
9990 
9991 	vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
9992 						GFP_KERNEL_ACCOUNT);
9993 	if (!vcpu->arch.user_fpu) {
9994 		pr_err("kvm: failed to allocate userspace's fpu\n");
9995 		goto free_emulate_ctxt;
9996 	}
9997 
9998 	vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
9999 						 GFP_KERNEL_ACCOUNT);
10000 	if (!vcpu->arch.guest_fpu) {
10001 		pr_err("kvm: failed to allocate vcpu's fpu\n");
10002 		goto free_user_fpu;
10003 	}
10004 	fx_init(vcpu);
10005 
10006 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10007 	vcpu->arch.cr3_lm_rsvd_bits = rsvd_bits(cpuid_maxphyaddr(vcpu), 63);
10008 
10009 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10010 
10011 	kvm_async_pf_hash_reset(vcpu);
10012 	kvm_pmu_init(vcpu);
10013 
10014 	vcpu->arch.pending_external_vector = -1;
10015 	vcpu->arch.preempted_in_kernel = false;
10016 
10017 	kvm_hv_vcpu_init(vcpu);
10018 
10019 	r = kvm_x86_ops.vcpu_create(vcpu);
10020 	if (r)
10021 		goto free_guest_fpu;
10022 
10023 	vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10024 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10025 	kvm_vcpu_mtrr_init(vcpu);
10026 	vcpu_load(vcpu);
10027 	kvm_vcpu_reset(vcpu, false);
10028 	kvm_init_mmu(vcpu, false);
10029 	vcpu_put(vcpu);
10030 	return 0;
10031 
10032 free_guest_fpu:
10033 	kvm_free_guest_fpu(vcpu);
10034 free_user_fpu:
10035 	kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10036 free_emulate_ctxt:
10037 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10038 free_wbinvd_dirty_mask:
10039 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10040 fail_free_mce_banks:
10041 	kfree(vcpu->arch.mce_banks);
10042 fail_free_pio_data:
10043 	free_page((unsigned long)vcpu->arch.pio_data);
10044 fail_free_lapic:
10045 	kvm_free_lapic(vcpu);
10046 fail_mmu_destroy:
10047 	kvm_mmu_destroy(vcpu);
10048 	return r;
10049 }
10050 
10051 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10052 {
10053 	struct kvm *kvm = vcpu->kvm;
10054 
10055 	kvm_hv_vcpu_postcreate(vcpu);
10056 
10057 	if (mutex_lock_killable(&vcpu->mutex))
10058 		return;
10059 	vcpu_load(vcpu);
10060 	kvm_synchronize_tsc(vcpu, 0);
10061 	vcpu_put(vcpu);
10062 
10063 	/* poll control enabled by default */
10064 	vcpu->arch.msr_kvm_poll_control = 1;
10065 
10066 	mutex_unlock(&vcpu->mutex);
10067 
10068 	if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10069 		schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10070 						KVMCLOCK_SYNC_PERIOD);
10071 }
10072 
10073 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10074 {
10075 	struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
10076 	int idx;
10077 
10078 	kvm_release_pfn(cache->pfn, cache->dirty, cache);
10079 
10080 	kvmclock_reset(vcpu);
10081 
10082 	kvm_x86_ops.vcpu_free(vcpu);
10083 
10084 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10085 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10086 	kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10087 	kvm_free_guest_fpu(vcpu);
10088 
10089 	kvm_hv_vcpu_uninit(vcpu);
10090 	kvm_pmu_destroy(vcpu);
10091 	kfree(vcpu->arch.mce_banks);
10092 	kvm_free_lapic(vcpu);
10093 	idx = srcu_read_lock(&vcpu->kvm->srcu);
10094 	kvm_mmu_destroy(vcpu);
10095 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
10096 	free_page((unsigned long)vcpu->arch.pio_data);
10097 	kvfree(vcpu->arch.cpuid_entries);
10098 	if (!lapic_in_kernel(vcpu))
10099 		static_key_slow_dec(&kvm_no_apic_vcpu);
10100 }
10101 
10102 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10103 {
10104 	kvm_lapic_reset(vcpu, init_event);
10105 
10106 	vcpu->arch.hflags = 0;
10107 
10108 	vcpu->arch.smi_pending = 0;
10109 	vcpu->arch.smi_count = 0;
10110 	atomic_set(&vcpu->arch.nmi_queued, 0);
10111 	vcpu->arch.nmi_pending = 0;
10112 	vcpu->arch.nmi_injected = false;
10113 	kvm_clear_interrupt_queue(vcpu);
10114 	kvm_clear_exception_queue(vcpu);
10115 
10116 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10117 	kvm_update_dr0123(vcpu);
10118 	vcpu->arch.dr6 = DR6_INIT;
10119 	vcpu->arch.dr7 = DR7_FIXED_1;
10120 	kvm_update_dr7(vcpu);
10121 
10122 	vcpu->arch.cr2 = 0;
10123 
10124 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10125 	vcpu->arch.apf.msr_en_val = 0;
10126 	vcpu->arch.apf.msr_int_val = 0;
10127 	vcpu->arch.st.msr_val = 0;
10128 
10129 	kvmclock_reset(vcpu);
10130 
10131 	kvm_clear_async_pf_completion_queue(vcpu);
10132 	kvm_async_pf_hash_reset(vcpu);
10133 	vcpu->arch.apf.halted = false;
10134 
10135 	if (vcpu->arch.guest_fpu && kvm_mpx_supported()) {
10136 		void *mpx_state_buffer;
10137 
10138 		/*
10139 		 * To avoid have the INIT path from kvm_apic_has_events() that be
10140 		 * called with loaded FPU and does not let userspace fix the state.
10141 		 */
10142 		if (init_event)
10143 			kvm_put_guest_fpu(vcpu);
10144 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10145 					XFEATURE_BNDREGS);
10146 		if (mpx_state_buffer)
10147 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
10148 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10149 					XFEATURE_BNDCSR);
10150 		if (mpx_state_buffer)
10151 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
10152 		if (init_event)
10153 			kvm_load_guest_fpu(vcpu);
10154 	}
10155 
10156 	if (!init_event) {
10157 		kvm_pmu_reset(vcpu);
10158 		vcpu->arch.smbase = 0x30000;
10159 
10160 		vcpu->arch.msr_misc_features_enables = 0;
10161 
10162 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10163 	}
10164 
10165 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10166 	vcpu->arch.regs_avail = ~0;
10167 	vcpu->arch.regs_dirty = ~0;
10168 
10169 	vcpu->arch.ia32_xss = 0;
10170 
10171 	kvm_x86_ops.vcpu_reset(vcpu, init_event);
10172 }
10173 
10174 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10175 {
10176 	struct kvm_segment cs;
10177 
10178 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10179 	cs.selector = vector << 8;
10180 	cs.base = vector << 12;
10181 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10182 	kvm_rip_write(vcpu, 0);
10183 }
10184 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
10185 
10186 int kvm_arch_hardware_enable(void)
10187 {
10188 	struct kvm *kvm;
10189 	struct kvm_vcpu *vcpu;
10190 	int i;
10191 	int ret;
10192 	u64 local_tsc;
10193 	u64 max_tsc = 0;
10194 	bool stable, backwards_tsc = false;
10195 
10196 	kvm_user_return_msr_cpu_online();
10197 	ret = kvm_x86_ops.hardware_enable();
10198 	if (ret != 0)
10199 		return ret;
10200 
10201 	local_tsc = rdtsc();
10202 	stable = !kvm_check_tsc_unstable();
10203 	list_for_each_entry(kvm, &vm_list, vm_list) {
10204 		kvm_for_each_vcpu(i, vcpu, kvm) {
10205 			if (!stable && vcpu->cpu == smp_processor_id())
10206 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10207 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10208 				backwards_tsc = true;
10209 				if (vcpu->arch.last_host_tsc > max_tsc)
10210 					max_tsc = vcpu->arch.last_host_tsc;
10211 			}
10212 		}
10213 	}
10214 
10215 	/*
10216 	 * Sometimes, even reliable TSCs go backwards.  This happens on
10217 	 * platforms that reset TSC during suspend or hibernate actions, but
10218 	 * maintain synchronization.  We must compensate.  Fortunately, we can
10219 	 * detect that condition here, which happens early in CPU bringup,
10220 	 * before any KVM threads can be running.  Unfortunately, we can't
10221 	 * bring the TSCs fully up to date with real time, as we aren't yet far
10222 	 * enough into CPU bringup that we know how much real time has actually
10223 	 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10224 	 * variables that haven't been updated yet.
10225 	 *
10226 	 * So we simply find the maximum observed TSC above, then record the
10227 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
10228 	 * the adjustment will be applied.  Note that we accumulate
10229 	 * adjustments, in case multiple suspend cycles happen before some VCPU
10230 	 * gets a chance to run again.  In the event that no KVM threads get a
10231 	 * chance to run, we will miss the entire elapsed period, as we'll have
10232 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10233 	 * loose cycle time.  This isn't too big a deal, since the loss will be
10234 	 * uniform across all VCPUs (not to mention the scenario is extremely
10235 	 * unlikely). It is possible that a second hibernate recovery happens
10236 	 * much faster than a first, causing the observed TSC here to be
10237 	 * smaller; this would require additional padding adjustment, which is
10238 	 * why we set last_host_tsc to the local tsc observed here.
10239 	 *
10240 	 * N.B. - this code below runs only on platforms with reliable TSC,
10241 	 * as that is the only way backwards_tsc is set above.  Also note
10242 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10243 	 * have the same delta_cyc adjustment applied if backwards_tsc
10244 	 * is detected.  Note further, this adjustment is only done once,
10245 	 * as we reset last_host_tsc on all VCPUs to stop this from being
10246 	 * called multiple times (one for each physical CPU bringup).
10247 	 *
10248 	 * Platforms with unreliable TSCs don't have to deal with this, they
10249 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
10250 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
10251 	 * guarantee that they stay in perfect synchronization.
10252 	 */
10253 	if (backwards_tsc) {
10254 		u64 delta_cyc = max_tsc - local_tsc;
10255 		list_for_each_entry(kvm, &vm_list, vm_list) {
10256 			kvm->arch.backwards_tsc_observed = true;
10257 			kvm_for_each_vcpu(i, vcpu, kvm) {
10258 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
10259 				vcpu->arch.last_host_tsc = local_tsc;
10260 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
10261 			}
10262 
10263 			/*
10264 			 * We have to disable TSC offset matching.. if you were
10265 			 * booting a VM while issuing an S4 host suspend....
10266 			 * you may have some problem.  Solving this issue is
10267 			 * left as an exercise to the reader.
10268 			 */
10269 			kvm->arch.last_tsc_nsec = 0;
10270 			kvm->arch.last_tsc_write = 0;
10271 		}
10272 
10273 	}
10274 	return 0;
10275 }
10276 
10277 void kvm_arch_hardware_disable(void)
10278 {
10279 	kvm_x86_ops.hardware_disable();
10280 	drop_user_return_notifiers();
10281 }
10282 
10283 int kvm_arch_hardware_setup(void *opaque)
10284 {
10285 	struct kvm_x86_init_ops *ops = opaque;
10286 	int r;
10287 
10288 	rdmsrl_safe(MSR_EFER, &host_efer);
10289 
10290 	if (boot_cpu_has(X86_FEATURE_XSAVES))
10291 		rdmsrl(MSR_IA32_XSS, host_xss);
10292 
10293 	r = ops->hardware_setup();
10294 	if (r != 0)
10295 		return r;
10296 
10297 	memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
10298 
10299 	if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
10300 		supported_xss = 0;
10301 
10302 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
10303 	cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
10304 #undef __kvm_cpu_cap_has
10305 
10306 	if (kvm_has_tsc_control) {
10307 		/*
10308 		 * Make sure the user can only configure tsc_khz values that
10309 		 * fit into a signed integer.
10310 		 * A min value is not calculated because it will always
10311 		 * be 1 on all machines.
10312 		 */
10313 		u64 max = min(0x7fffffffULL,
10314 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
10315 		kvm_max_guest_tsc_khz = max;
10316 
10317 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
10318 	}
10319 
10320 	kvm_init_msr_list();
10321 	return 0;
10322 }
10323 
10324 void kvm_arch_hardware_unsetup(void)
10325 {
10326 	kvm_x86_ops.hardware_unsetup();
10327 }
10328 
10329 int kvm_arch_check_processor_compat(void *opaque)
10330 {
10331 	struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
10332 	struct kvm_x86_init_ops *ops = opaque;
10333 
10334 	WARN_ON(!irqs_disabled());
10335 
10336 	if (__cr4_reserved_bits(cpu_has, c) !=
10337 	    __cr4_reserved_bits(cpu_has, &boot_cpu_data))
10338 		return -EIO;
10339 
10340 	return ops->check_processor_compatibility();
10341 }
10342 
10343 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
10344 {
10345 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
10346 }
10347 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
10348 
10349 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
10350 {
10351 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
10352 }
10353 
10354 struct static_key kvm_no_apic_vcpu __read_mostly;
10355 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
10356 
10357 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
10358 {
10359 	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
10360 
10361 	vcpu->arch.l1tf_flush_l1d = true;
10362 	if (pmu->version && unlikely(pmu->event_count)) {
10363 		pmu->need_cleanup = true;
10364 		kvm_make_request(KVM_REQ_PMU, vcpu);
10365 	}
10366 	kvm_x86_ops.sched_in(vcpu, cpu);
10367 }
10368 
10369 void kvm_arch_free_vm(struct kvm *kvm)
10370 {
10371 	kfree(kvm->arch.hyperv.hv_pa_pg);
10372 	vfree(kvm);
10373 }
10374 
10375 
10376 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
10377 {
10378 	if (type)
10379 		return -EINVAL;
10380 
10381 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
10382 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
10383 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
10384 	INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
10385 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
10386 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
10387 
10388 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
10389 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
10390 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
10391 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
10392 		&kvm->arch.irq_sources_bitmap);
10393 
10394 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
10395 	mutex_init(&kvm->arch.apic_map_lock);
10396 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
10397 
10398 	kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
10399 	pvclock_update_vm_gtod_copy(kvm);
10400 
10401 	kvm->arch.guest_can_read_msr_platform_info = true;
10402 
10403 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
10404 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
10405 
10406 	kvm_hv_init_vm(kvm);
10407 	kvm_page_track_init(kvm);
10408 	kvm_mmu_init_vm(kvm);
10409 
10410 	return kvm_x86_ops.vm_init(kvm);
10411 }
10412 
10413 int kvm_arch_post_init_vm(struct kvm *kvm)
10414 {
10415 	return kvm_mmu_post_init_vm(kvm);
10416 }
10417 
10418 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
10419 {
10420 	vcpu_load(vcpu);
10421 	kvm_mmu_unload(vcpu);
10422 	vcpu_put(vcpu);
10423 }
10424 
10425 static void kvm_free_vcpus(struct kvm *kvm)
10426 {
10427 	unsigned int i;
10428 	struct kvm_vcpu *vcpu;
10429 
10430 	/*
10431 	 * Unpin any mmu pages first.
10432 	 */
10433 	kvm_for_each_vcpu(i, vcpu, kvm) {
10434 		kvm_clear_async_pf_completion_queue(vcpu);
10435 		kvm_unload_vcpu_mmu(vcpu);
10436 	}
10437 	kvm_for_each_vcpu(i, vcpu, kvm)
10438 		kvm_vcpu_destroy(vcpu);
10439 
10440 	mutex_lock(&kvm->lock);
10441 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
10442 		kvm->vcpus[i] = NULL;
10443 
10444 	atomic_set(&kvm->online_vcpus, 0);
10445 	mutex_unlock(&kvm->lock);
10446 }
10447 
10448 void kvm_arch_sync_events(struct kvm *kvm)
10449 {
10450 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
10451 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
10452 	kvm_free_pit(kvm);
10453 }
10454 
10455 #define  ERR_PTR_USR(e)  ((void __user *)ERR_PTR(e))
10456 
10457 /**
10458  * __x86_set_memory_region: Setup KVM internal memory slot
10459  *
10460  * @kvm: the kvm pointer to the VM.
10461  * @id: the slot ID to setup.
10462  * @gpa: the GPA to install the slot (unused when @size == 0).
10463  * @size: the size of the slot. Set to zero to uninstall a slot.
10464  *
10465  * This function helps to setup a KVM internal memory slot.  Specify
10466  * @size > 0 to install a new slot, while @size == 0 to uninstall a
10467  * slot.  The return code can be one of the following:
10468  *
10469  *   HVA:           on success (uninstall will return a bogus HVA)
10470  *   -errno:        on error
10471  *
10472  * The caller should always use IS_ERR() to check the return value
10473  * before use.  Note, the KVM internal memory slots are guaranteed to
10474  * remain valid and unchanged until the VM is destroyed, i.e., the
10475  * GPA->HVA translation will not change.  However, the HVA is a user
10476  * address, i.e. its accessibility is not guaranteed, and must be
10477  * accessed via __copy_{to,from}_user().
10478  */
10479 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
10480 				      u32 size)
10481 {
10482 	int i, r;
10483 	unsigned long hva, old_npages;
10484 	struct kvm_memslots *slots = kvm_memslots(kvm);
10485 	struct kvm_memory_slot *slot;
10486 
10487 	/* Called with kvm->slots_lock held.  */
10488 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
10489 		return ERR_PTR_USR(-EINVAL);
10490 
10491 	slot = id_to_memslot(slots, id);
10492 	if (size) {
10493 		if (slot && slot->npages)
10494 			return ERR_PTR_USR(-EEXIST);
10495 
10496 		/*
10497 		 * MAP_SHARED to prevent internal slot pages from being moved
10498 		 * by fork()/COW.
10499 		 */
10500 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
10501 			      MAP_SHARED | MAP_ANONYMOUS, 0);
10502 		if (IS_ERR((void *)hva))
10503 			return (void __user *)hva;
10504 	} else {
10505 		if (!slot || !slot->npages)
10506 			return 0;
10507 
10508 		old_npages = slot->npages;
10509 		hva = slot->userspace_addr;
10510 	}
10511 
10512 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
10513 		struct kvm_userspace_memory_region m;
10514 
10515 		m.slot = id | (i << 16);
10516 		m.flags = 0;
10517 		m.guest_phys_addr = gpa;
10518 		m.userspace_addr = hva;
10519 		m.memory_size = size;
10520 		r = __kvm_set_memory_region(kvm, &m);
10521 		if (r < 0)
10522 			return ERR_PTR_USR(r);
10523 	}
10524 
10525 	if (!size)
10526 		vm_munmap(hva, old_npages * PAGE_SIZE);
10527 
10528 	return (void __user *)hva;
10529 }
10530 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
10531 
10532 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
10533 {
10534 	kvm_mmu_pre_destroy_vm(kvm);
10535 }
10536 
10537 void kvm_arch_destroy_vm(struct kvm *kvm)
10538 {
10539 	u32 i;
10540 
10541 	if (current->mm == kvm->mm) {
10542 		/*
10543 		 * Free memory regions allocated on behalf of userspace,
10544 		 * unless the the memory map has changed due to process exit
10545 		 * or fd copying.
10546 		 */
10547 		mutex_lock(&kvm->slots_lock);
10548 		__x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
10549 					0, 0);
10550 		__x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
10551 					0, 0);
10552 		__x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
10553 		mutex_unlock(&kvm->slots_lock);
10554 	}
10555 	if (kvm_x86_ops.vm_destroy)
10556 		kvm_x86_ops.vm_destroy(kvm);
10557 	for (i = 0; i < kvm->arch.msr_filter.count; i++)
10558 		kfree(kvm->arch.msr_filter.ranges[i].bitmap);
10559 	kvm_pic_destroy(kvm);
10560 	kvm_ioapic_destroy(kvm);
10561 	kvm_free_vcpus(kvm);
10562 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
10563 	kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
10564 	kvm_mmu_uninit_vm(kvm);
10565 	kvm_page_track_cleanup(kvm);
10566 	kvm_hv_destroy_vm(kvm);
10567 }
10568 
10569 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
10570 {
10571 	int i;
10572 
10573 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10574 		kvfree(slot->arch.rmap[i]);
10575 		slot->arch.rmap[i] = NULL;
10576 
10577 		if (i == 0)
10578 			continue;
10579 
10580 		kvfree(slot->arch.lpage_info[i - 1]);
10581 		slot->arch.lpage_info[i - 1] = NULL;
10582 	}
10583 
10584 	kvm_page_track_free_memslot(slot);
10585 }
10586 
10587 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
10588 				      unsigned long npages)
10589 {
10590 	int i;
10591 
10592 	/*
10593 	 * Clear out the previous array pointers for the KVM_MR_MOVE case.  The
10594 	 * old arrays will be freed by __kvm_set_memory_region() if installing
10595 	 * the new memslot is successful.
10596 	 */
10597 	memset(&slot->arch, 0, sizeof(slot->arch));
10598 
10599 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10600 		struct kvm_lpage_info *linfo;
10601 		unsigned long ugfn;
10602 		int lpages;
10603 		int level = i + 1;
10604 
10605 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
10606 				      slot->base_gfn, level) + 1;
10607 
10608 		slot->arch.rmap[i] =
10609 			kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
10610 				 GFP_KERNEL_ACCOUNT);
10611 		if (!slot->arch.rmap[i])
10612 			goto out_free;
10613 		if (i == 0)
10614 			continue;
10615 
10616 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
10617 		if (!linfo)
10618 			goto out_free;
10619 
10620 		slot->arch.lpage_info[i - 1] = linfo;
10621 
10622 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
10623 			linfo[0].disallow_lpage = 1;
10624 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
10625 			linfo[lpages - 1].disallow_lpage = 1;
10626 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
10627 		/*
10628 		 * If the gfn and userspace address are not aligned wrt each
10629 		 * other, disable large page support for this slot.
10630 		 */
10631 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
10632 			unsigned long j;
10633 
10634 			for (j = 0; j < lpages; ++j)
10635 				linfo[j].disallow_lpage = 1;
10636 		}
10637 	}
10638 
10639 	if (kvm_page_track_create_memslot(slot, npages))
10640 		goto out_free;
10641 
10642 	return 0;
10643 
10644 out_free:
10645 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10646 		kvfree(slot->arch.rmap[i]);
10647 		slot->arch.rmap[i] = NULL;
10648 		if (i == 0)
10649 			continue;
10650 
10651 		kvfree(slot->arch.lpage_info[i - 1]);
10652 		slot->arch.lpage_info[i - 1] = NULL;
10653 	}
10654 	return -ENOMEM;
10655 }
10656 
10657 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
10658 {
10659 	struct kvm_vcpu *vcpu;
10660 	int i;
10661 
10662 	/*
10663 	 * memslots->generation has been incremented.
10664 	 * mmio generation may have reached its maximum value.
10665 	 */
10666 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
10667 
10668 	/* Force re-initialization of steal_time cache */
10669 	kvm_for_each_vcpu(i, vcpu, kvm)
10670 		kvm_vcpu_kick(vcpu);
10671 }
10672 
10673 int kvm_arch_prepare_memory_region(struct kvm *kvm,
10674 				struct kvm_memory_slot *memslot,
10675 				const struct kvm_userspace_memory_region *mem,
10676 				enum kvm_mr_change change)
10677 {
10678 	if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
10679 		return kvm_alloc_memslot_metadata(memslot,
10680 						  mem->memory_size >> PAGE_SHIFT);
10681 	return 0;
10682 }
10683 
10684 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
10685 				     struct kvm_memory_slot *old,
10686 				     struct kvm_memory_slot *new,
10687 				     enum kvm_mr_change change)
10688 {
10689 	/*
10690 	 * Nothing to do for RO slots or CREATE/MOVE/DELETE of a slot.
10691 	 * See comments below.
10692 	 */
10693 	if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
10694 		return;
10695 
10696 	/*
10697 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
10698 	 * sptes have to be split.  If live migration is successful, the guest
10699 	 * in the source machine will be destroyed and large sptes will be
10700 	 * created in the destination. However, if the guest continues to run
10701 	 * in the source machine (for example if live migration fails), small
10702 	 * sptes will remain around and cause bad performance.
10703 	 *
10704 	 * Scan sptes if dirty logging has been stopped, dropping those
10705 	 * which can be collapsed into a single large-page spte.  Later
10706 	 * page faults will create the large-page sptes.
10707 	 *
10708 	 * There is no need to do this in any of the following cases:
10709 	 * CREATE:      No dirty mappings will already exist.
10710 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
10711 	 *		kvm_arch_flush_shadow_memslot()
10712 	 */
10713 	if ((old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
10714 	    !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
10715 		kvm_mmu_zap_collapsible_sptes(kvm, new);
10716 
10717 	/*
10718 	 * Enable or disable dirty logging for the slot.
10719 	 *
10720 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of the old
10721 	 * slot have been zapped so no dirty logging updates are needed for
10722 	 * the old slot.
10723 	 * For KVM_MR_CREATE and KVM_MR_MOVE, once the new slot is visible
10724 	 * any mappings that might be created in it will consume the
10725 	 * properties of the new slot and do not need to be updated here.
10726 	 *
10727 	 * When PML is enabled, the kvm_x86_ops dirty logging hooks are
10728 	 * called to enable/disable dirty logging.
10729 	 *
10730 	 * When disabling dirty logging with PML enabled, the D-bit is set
10731 	 * for sptes in the slot in order to prevent unnecessary GPA
10732 	 * logging in the PML buffer (and potential PML buffer full VMEXIT).
10733 	 * This guarantees leaving PML enabled for the guest's lifetime
10734 	 * won't have any additional overhead from PML when the guest is
10735 	 * running with dirty logging disabled.
10736 	 *
10737 	 * When enabling dirty logging, large sptes are write-protected
10738 	 * so they can be split on first write.  New large sptes cannot
10739 	 * be created for this slot until the end of the logging.
10740 	 * See the comments in fast_page_fault().
10741 	 * For small sptes, nothing is done if the dirty log is in the
10742 	 * initial-all-set state.  Otherwise, depending on whether pml
10743 	 * is enabled the D-bit or the W-bit will be cleared.
10744 	 */
10745 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
10746 		if (kvm_x86_ops.slot_enable_log_dirty) {
10747 			kvm_x86_ops.slot_enable_log_dirty(kvm, new);
10748 		} else {
10749 			int level =
10750 				kvm_dirty_log_manual_protect_and_init_set(kvm) ?
10751 				PG_LEVEL_2M : PG_LEVEL_4K;
10752 
10753 			/*
10754 			 * If we're with initial-all-set, we don't need
10755 			 * to write protect any small page because
10756 			 * they're reported as dirty already.  However
10757 			 * we still need to write-protect huge pages
10758 			 * so that the page split can happen lazily on
10759 			 * the first write to the huge page.
10760 			 */
10761 			kvm_mmu_slot_remove_write_access(kvm, new, level);
10762 		}
10763 	} else {
10764 		if (kvm_x86_ops.slot_disable_log_dirty)
10765 			kvm_x86_ops.slot_disable_log_dirty(kvm, new);
10766 	}
10767 }
10768 
10769 void kvm_arch_commit_memory_region(struct kvm *kvm,
10770 				const struct kvm_userspace_memory_region *mem,
10771 				struct kvm_memory_slot *old,
10772 				const struct kvm_memory_slot *new,
10773 				enum kvm_mr_change change)
10774 {
10775 	if (!kvm->arch.n_requested_mmu_pages)
10776 		kvm_mmu_change_mmu_pages(kvm,
10777 				kvm_mmu_calculate_default_mmu_pages(kvm));
10778 
10779 	/*
10780 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
10781 	 */
10782 	kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change);
10783 
10784 	/* Free the arrays associated with the old memslot. */
10785 	if (change == KVM_MR_MOVE)
10786 		kvm_arch_free_memslot(kvm, old);
10787 }
10788 
10789 void kvm_arch_flush_shadow_all(struct kvm *kvm)
10790 {
10791 	kvm_mmu_zap_all(kvm);
10792 }
10793 
10794 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
10795 				   struct kvm_memory_slot *slot)
10796 {
10797 	kvm_page_track_flush_slot(kvm, slot);
10798 }
10799 
10800 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
10801 {
10802 	return (is_guest_mode(vcpu) &&
10803 			kvm_x86_ops.guest_apic_has_interrupt &&
10804 			kvm_x86_ops.guest_apic_has_interrupt(vcpu));
10805 }
10806 
10807 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
10808 {
10809 	if (!list_empty_careful(&vcpu->async_pf.done))
10810 		return true;
10811 
10812 	if (kvm_apic_has_events(vcpu))
10813 		return true;
10814 
10815 	if (vcpu->arch.pv.pv_unhalted)
10816 		return true;
10817 
10818 	if (vcpu->arch.exception.pending)
10819 		return true;
10820 
10821 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10822 	    (vcpu->arch.nmi_pending &&
10823 	     kvm_x86_ops.nmi_allowed(vcpu, false)))
10824 		return true;
10825 
10826 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
10827 	    (vcpu->arch.smi_pending &&
10828 	     kvm_x86_ops.smi_allowed(vcpu, false)))
10829 		return true;
10830 
10831 	if (kvm_arch_interrupt_allowed(vcpu) &&
10832 	    (kvm_cpu_has_interrupt(vcpu) ||
10833 	    kvm_guest_apic_has_interrupt(vcpu)))
10834 		return true;
10835 
10836 	if (kvm_hv_has_stimer_pending(vcpu))
10837 		return true;
10838 
10839 	if (is_guest_mode(vcpu) &&
10840 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
10841 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
10842 		return true;
10843 
10844 	return false;
10845 }
10846 
10847 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
10848 {
10849 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
10850 }
10851 
10852 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
10853 {
10854 	if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
10855 		return true;
10856 
10857 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10858 		kvm_test_request(KVM_REQ_SMI, vcpu) ||
10859 		 kvm_test_request(KVM_REQ_EVENT, vcpu))
10860 		return true;
10861 
10862 	if (vcpu->arch.apicv_active && kvm_x86_ops.dy_apicv_has_pending_interrupt(vcpu))
10863 		return true;
10864 
10865 	return false;
10866 }
10867 
10868 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
10869 {
10870 	return vcpu->arch.preempted_in_kernel;
10871 }
10872 
10873 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
10874 {
10875 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
10876 }
10877 
10878 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
10879 {
10880 	return kvm_x86_ops.interrupt_allowed(vcpu, false);
10881 }
10882 
10883 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
10884 {
10885 	/* Can't read the RIP when guest state is protected, just return 0 */
10886 	if (vcpu->arch.guest_state_protected)
10887 		return 0;
10888 
10889 	if (is_64_bit_mode(vcpu))
10890 		return kvm_rip_read(vcpu);
10891 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
10892 		     kvm_rip_read(vcpu));
10893 }
10894 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
10895 
10896 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
10897 {
10898 	return kvm_get_linear_rip(vcpu) == linear_rip;
10899 }
10900 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
10901 
10902 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
10903 {
10904 	unsigned long rflags;
10905 
10906 	rflags = kvm_x86_ops.get_rflags(vcpu);
10907 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10908 		rflags &= ~X86_EFLAGS_TF;
10909 	return rflags;
10910 }
10911 EXPORT_SYMBOL_GPL(kvm_get_rflags);
10912 
10913 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10914 {
10915 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
10916 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
10917 		rflags |= X86_EFLAGS_TF;
10918 	kvm_x86_ops.set_rflags(vcpu, rflags);
10919 }
10920 
10921 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10922 {
10923 	__kvm_set_rflags(vcpu, rflags);
10924 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10925 }
10926 EXPORT_SYMBOL_GPL(kvm_set_rflags);
10927 
10928 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
10929 {
10930 	int r;
10931 
10932 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
10933 	      work->wakeup_all)
10934 		return;
10935 
10936 	r = kvm_mmu_reload(vcpu);
10937 	if (unlikely(r))
10938 		return;
10939 
10940 	if (!vcpu->arch.mmu->direct_map &&
10941 	      work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
10942 		return;
10943 
10944 	kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
10945 }
10946 
10947 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
10948 {
10949 	BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
10950 
10951 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
10952 }
10953 
10954 static inline u32 kvm_async_pf_next_probe(u32 key)
10955 {
10956 	return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
10957 }
10958 
10959 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10960 {
10961 	u32 key = kvm_async_pf_hash_fn(gfn);
10962 
10963 	while (vcpu->arch.apf.gfns[key] != ~0)
10964 		key = kvm_async_pf_next_probe(key);
10965 
10966 	vcpu->arch.apf.gfns[key] = gfn;
10967 }
10968 
10969 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
10970 {
10971 	int i;
10972 	u32 key = kvm_async_pf_hash_fn(gfn);
10973 
10974 	for (i = 0; i < ASYNC_PF_PER_VCPU &&
10975 		     (vcpu->arch.apf.gfns[key] != gfn &&
10976 		      vcpu->arch.apf.gfns[key] != ~0); i++)
10977 		key = kvm_async_pf_next_probe(key);
10978 
10979 	return key;
10980 }
10981 
10982 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10983 {
10984 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
10985 }
10986 
10987 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10988 {
10989 	u32 i, j, k;
10990 
10991 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
10992 
10993 	if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
10994 		return;
10995 
10996 	while (true) {
10997 		vcpu->arch.apf.gfns[i] = ~0;
10998 		do {
10999 			j = kvm_async_pf_next_probe(j);
11000 			if (vcpu->arch.apf.gfns[j] == ~0)
11001 				return;
11002 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
11003 			/*
11004 			 * k lies cyclically in ]i,j]
11005 			 * |    i.k.j |
11006 			 * |....j i.k.| or  |.k..j i...|
11007 			 */
11008 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
11009 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11010 		i = j;
11011 	}
11012 }
11013 
11014 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11015 {
11016 	u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11017 
11018 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11019 				      sizeof(reason));
11020 }
11021 
11022 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11023 {
11024 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11025 
11026 	return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11027 					     &token, offset, sizeof(token));
11028 }
11029 
11030 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
11031 {
11032 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11033 	u32 val;
11034 
11035 	if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11036 					 &val, offset, sizeof(val)))
11037 		return false;
11038 
11039 	return !val;
11040 }
11041 
11042 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
11043 {
11044 	if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
11045 		return false;
11046 
11047 	if (!kvm_pv_async_pf_enabled(vcpu) ||
11048 	    (vcpu->arch.apf.send_user_only && kvm_x86_ops.get_cpl(vcpu) == 0))
11049 		return false;
11050 
11051 	return true;
11052 }
11053 
11054 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
11055 {
11056 	if (unlikely(!lapic_in_kernel(vcpu) ||
11057 		     kvm_event_needs_reinjection(vcpu) ||
11058 		     vcpu->arch.exception.pending))
11059 		return false;
11060 
11061 	if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
11062 		return false;
11063 
11064 	/*
11065 	 * If interrupts are off we cannot even use an artificial
11066 	 * halt state.
11067 	 */
11068 	return kvm_arch_interrupt_allowed(vcpu);
11069 }
11070 
11071 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
11072 				     struct kvm_async_pf *work)
11073 {
11074 	struct x86_exception fault;
11075 
11076 	trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
11077 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
11078 
11079 	if (kvm_can_deliver_async_pf(vcpu) &&
11080 	    !apf_put_user_notpresent(vcpu)) {
11081 		fault.vector = PF_VECTOR;
11082 		fault.error_code_valid = true;
11083 		fault.error_code = 0;
11084 		fault.nested_page_fault = false;
11085 		fault.address = work->arch.token;
11086 		fault.async_page_fault = true;
11087 		kvm_inject_page_fault(vcpu, &fault);
11088 		return true;
11089 	} else {
11090 		/*
11091 		 * It is not possible to deliver a paravirtualized asynchronous
11092 		 * page fault, but putting the guest in an artificial halt state
11093 		 * can be beneficial nevertheless: if an interrupt arrives, we
11094 		 * can deliver it timely and perhaps the guest will schedule
11095 		 * another process.  When the instruction that triggered a page
11096 		 * fault is retried, hopefully the page will be ready in the host.
11097 		 */
11098 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
11099 		return false;
11100 	}
11101 }
11102 
11103 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
11104 				 struct kvm_async_pf *work)
11105 {
11106 	struct kvm_lapic_irq irq = {
11107 		.delivery_mode = APIC_DM_FIXED,
11108 		.vector = vcpu->arch.apf.vec
11109 	};
11110 
11111 	if (work->wakeup_all)
11112 		work->arch.token = ~0; /* broadcast wakeup */
11113 	else
11114 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
11115 	trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
11116 
11117 	if ((work->wakeup_all || work->notpresent_injected) &&
11118 	    kvm_pv_async_pf_enabled(vcpu) &&
11119 	    !apf_put_user_ready(vcpu, work->arch.token)) {
11120 		vcpu->arch.apf.pageready_pending = true;
11121 		kvm_apic_set_irq(vcpu, &irq, NULL);
11122 	}
11123 
11124 	vcpu->arch.apf.halted = false;
11125 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11126 }
11127 
11128 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
11129 {
11130 	kvm_make_request(KVM_REQ_APF_READY, vcpu);
11131 	if (!vcpu->arch.apf.pageready_pending)
11132 		kvm_vcpu_kick(vcpu);
11133 }
11134 
11135 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
11136 {
11137 	if (!kvm_pv_async_pf_enabled(vcpu))
11138 		return true;
11139 	else
11140 		return apf_pageready_slot_free(vcpu);
11141 }
11142 
11143 void kvm_arch_start_assignment(struct kvm *kvm)
11144 {
11145 	atomic_inc(&kvm->arch.assigned_device_count);
11146 }
11147 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
11148 
11149 void kvm_arch_end_assignment(struct kvm *kvm)
11150 {
11151 	atomic_dec(&kvm->arch.assigned_device_count);
11152 }
11153 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
11154 
11155 bool kvm_arch_has_assigned_device(struct kvm *kvm)
11156 {
11157 	return atomic_read(&kvm->arch.assigned_device_count);
11158 }
11159 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
11160 
11161 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
11162 {
11163 	atomic_inc(&kvm->arch.noncoherent_dma_count);
11164 }
11165 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
11166 
11167 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
11168 {
11169 	atomic_dec(&kvm->arch.noncoherent_dma_count);
11170 }
11171 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
11172 
11173 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
11174 {
11175 	return atomic_read(&kvm->arch.noncoherent_dma_count);
11176 }
11177 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
11178 
11179 bool kvm_arch_has_irq_bypass(void)
11180 {
11181 	return true;
11182 }
11183 
11184 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
11185 				      struct irq_bypass_producer *prod)
11186 {
11187 	struct kvm_kernel_irqfd *irqfd =
11188 		container_of(cons, struct kvm_kernel_irqfd, consumer);
11189 	int ret;
11190 
11191 	irqfd->producer = prod;
11192 	kvm_arch_start_assignment(irqfd->kvm);
11193 	ret = kvm_x86_ops.update_pi_irte(irqfd->kvm,
11194 					 prod->irq, irqfd->gsi, 1);
11195 
11196 	if (ret)
11197 		kvm_arch_end_assignment(irqfd->kvm);
11198 
11199 	return ret;
11200 }
11201 
11202 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
11203 				      struct irq_bypass_producer *prod)
11204 {
11205 	int ret;
11206 	struct kvm_kernel_irqfd *irqfd =
11207 		container_of(cons, struct kvm_kernel_irqfd, consumer);
11208 
11209 	WARN_ON(irqfd->producer != prod);
11210 	irqfd->producer = NULL;
11211 
11212 	/*
11213 	 * When producer of consumer is unregistered, we change back to
11214 	 * remapped mode, so we can re-use the current implementation
11215 	 * when the irq is masked/disabled or the consumer side (KVM
11216 	 * int this case doesn't want to receive the interrupts.
11217 	*/
11218 	ret = kvm_x86_ops.update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
11219 	if (ret)
11220 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
11221 		       " fails: %d\n", irqfd->consumer.token, ret);
11222 
11223 	kvm_arch_end_assignment(irqfd->kvm);
11224 }
11225 
11226 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
11227 				   uint32_t guest_irq, bool set)
11228 {
11229 	return kvm_x86_ops.update_pi_irte(kvm, host_irq, guest_irq, set);
11230 }
11231 
11232 bool kvm_vector_hashing_enabled(void)
11233 {
11234 	return vector_hashing;
11235 }
11236 
11237 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
11238 {
11239 	return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
11240 }
11241 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
11242 
11243 
11244 int kvm_spec_ctrl_test_value(u64 value)
11245 {
11246 	/*
11247 	 * test that setting IA32_SPEC_CTRL to given value
11248 	 * is allowed by the host processor
11249 	 */
11250 
11251 	u64 saved_value;
11252 	unsigned long flags;
11253 	int ret = 0;
11254 
11255 	local_irq_save(flags);
11256 
11257 	if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
11258 		ret = 1;
11259 	else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
11260 		ret = 1;
11261 	else
11262 		wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
11263 
11264 	local_irq_restore(flags);
11265 
11266 	return ret;
11267 }
11268 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
11269 
11270 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
11271 {
11272 	struct x86_exception fault;
11273 	u32 access = error_code &
11274 		(PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
11275 
11276 	if (!(error_code & PFERR_PRESENT_MASK) ||
11277 	    vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
11278 		/*
11279 		 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
11280 		 * tables probably do not match the TLB.  Just proceed
11281 		 * with the error code that the processor gave.
11282 		 */
11283 		fault.vector = PF_VECTOR;
11284 		fault.error_code_valid = true;
11285 		fault.error_code = error_code;
11286 		fault.nested_page_fault = false;
11287 		fault.address = gva;
11288 	}
11289 	vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
11290 }
11291 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
11292 
11293 /*
11294  * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
11295  * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
11296  * indicates whether exit to userspace is needed.
11297  */
11298 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
11299 			      struct x86_exception *e)
11300 {
11301 	if (r == X86EMUL_PROPAGATE_FAULT) {
11302 		kvm_inject_emulated_page_fault(vcpu, e);
11303 		return 1;
11304 	}
11305 
11306 	/*
11307 	 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
11308 	 * while handling a VMX instruction KVM could've handled the request
11309 	 * correctly by exiting to userspace and performing I/O but there
11310 	 * doesn't seem to be a real use-case behind such requests, just return
11311 	 * KVM_EXIT_INTERNAL_ERROR for now.
11312 	 */
11313 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
11314 	vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
11315 	vcpu->run->internal.ndata = 0;
11316 
11317 	return 0;
11318 }
11319 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
11320 
11321 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
11322 {
11323 	bool pcid_enabled;
11324 	struct x86_exception e;
11325 	unsigned i;
11326 	unsigned long roots_to_free = 0;
11327 	struct {
11328 		u64 pcid;
11329 		u64 gla;
11330 	} operand;
11331 	int r;
11332 
11333 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
11334 	if (r != X86EMUL_CONTINUE)
11335 		return kvm_handle_memory_failure(vcpu, r, &e);
11336 
11337 	if (operand.pcid >> 12 != 0) {
11338 		kvm_inject_gp(vcpu, 0);
11339 		return 1;
11340 	}
11341 
11342 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
11343 
11344 	switch (type) {
11345 	case INVPCID_TYPE_INDIV_ADDR:
11346 		if ((!pcid_enabled && (operand.pcid != 0)) ||
11347 		    is_noncanonical_address(operand.gla, vcpu)) {
11348 			kvm_inject_gp(vcpu, 0);
11349 			return 1;
11350 		}
11351 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
11352 		return kvm_skip_emulated_instruction(vcpu);
11353 
11354 	case INVPCID_TYPE_SINGLE_CTXT:
11355 		if (!pcid_enabled && (operand.pcid != 0)) {
11356 			kvm_inject_gp(vcpu, 0);
11357 			return 1;
11358 		}
11359 
11360 		if (kvm_get_active_pcid(vcpu) == operand.pcid) {
11361 			kvm_mmu_sync_roots(vcpu);
11362 			kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
11363 		}
11364 
11365 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
11366 			if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd)
11367 			    == operand.pcid)
11368 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
11369 
11370 		kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
11371 		/*
11372 		 * If neither the current cr3 nor any of the prev_roots use the
11373 		 * given PCID, then nothing needs to be done here because a
11374 		 * resync will happen anyway before switching to any other CR3.
11375 		 */
11376 
11377 		return kvm_skip_emulated_instruction(vcpu);
11378 
11379 	case INVPCID_TYPE_ALL_NON_GLOBAL:
11380 		/*
11381 		 * Currently, KVM doesn't mark global entries in the shadow
11382 		 * page tables, so a non-global flush just degenerates to a
11383 		 * global flush. If needed, we could optimize this later by
11384 		 * keeping track of global entries in shadow page tables.
11385 		 */
11386 
11387 		fallthrough;
11388 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
11389 		kvm_mmu_unload(vcpu);
11390 		return kvm_skip_emulated_instruction(vcpu);
11391 
11392 	default:
11393 		BUG(); /* We have already checked above that type <= 3 */
11394 	}
11395 }
11396 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
11397 
11398 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
11399 {
11400 	struct kvm_run *run = vcpu->run;
11401 	struct kvm_mmio_fragment *frag;
11402 	unsigned int len;
11403 
11404 	BUG_ON(!vcpu->mmio_needed);
11405 
11406 	/* Complete previous fragment */
11407 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
11408 	len = min(8u, frag->len);
11409 	if (!vcpu->mmio_is_write)
11410 		memcpy(frag->data, run->mmio.data, len);
11411 
11412 	if (frag->len <= 8) {
11413 		/* Switch to the next fragment. */
11414 		frag++;
11415 		vcpu->mmio_cur_fragment++;
11416 	} else {
11417 		/* Go forward to the next mmio piece. */
11418 		frag->data += len;
11419 		frag->gpa += len;
11420 		frag->len -= len;
11421 	}
11422 
11423 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
11424 		vcpu->mmio_needed = 0;
11425 
11426 		// VMG change, at this point, we're always done
11427 		// RIP has already been advanced
11428 		return 1;
11429 	}
11430 
11431 	// More MMIO is needed
11432 	run->mmio.phys_addr = frag->gpa;
11433 	run->mmio.len = min(8u, frag->len);
11434 	run->mmio.is_write = vcpu->mmio_is_write;
11435 	if (run->mmio.is_write)
11436 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
11437 	run->exit_reason = KVM_EXIT_MMIO;
11438 
11439 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11440 
11441 	return 0;
11442 }
11443 
11444 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11445 			  void *data)
11446 {
11447 	int handled;
11448 	struct kvm_mmio_fragment *frag;
11449 
11450 	if (!data)
11451 		return -EINVAL;
11452 
11453 	handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11454 	if (handled == bytes)
11455 		return 1;
11456 
11457 	bytes -= handled;
11458 	gpa += handled;
11459 	data += handled;
11460 
11461 	/*TODO: Check if need to increment number of frags */
11462 	frag = vcpu->mmio_fragments;
11463 	vcpu->mmio_nr_fragments = 1;
11464 	frag->len = bytes;
11465 	frag->gpa = gpa;
11466 	frag->data = data;
11467 
11468 	vcpu->mmio_needed = 1;
11469 	vcpu->mmio_cur_fragment = 0;
11470 
11471 	vcpu->run->mmio.phys_addr = gpa;
11472 	vcpu->run->mmio.len = min(8u, frag->len);
11473 	vcpu->run->mmio.is_write = 1;
11474 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
11475 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
11476 
11477 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11478 
11479 	return 0;
11480 }
11481 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
11482 
11483 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11484 			 void *data)
11485 {
11486 	int handled;
11487 	struct kvm_mmio_fragment *frag;
11488 
11489 	if (!data)
11490 		return -EINVAL;
11491 
11492 	handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11493 	if (handled == bytes)
11494 		return 1;
11495 
11496 	bytes -= handled;
11497 	gpa += handled;
11498 	data += handled;
11499 
11500 	/*TODO: Check if need to increment number of frags */
11501 	frag = vcpu->mmio_fragments;
11502 	vcpu->mmio_nr_fragments = 1;
11503 	frag->len = bytes;
11504 	frag->gpa = gpa;
11505 	frag->data = data;
11506 
11507 	vcpu->mmio_needed = 1;
11508 	vcpu->mmio_cur_fragment = 0;
11509 
11510 	vcpu->run->mmio.phys_addr = gpa;
11511 	vcpu->run->mmio.len = min(8u, frag->len);
11512 	vcpu->run->mmio.is_write = 0;
11513 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
11514 
11515 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11516 
11517 	return 0;
11518 }
11519 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
11520 
11521 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
11522 {
11523 	memcpy(vcpu->arch.guest_ins_data, vcpu->arch.pio_data,
11524 	       vcpu->arch.pio.count * vcpu->arch.pio.size);
11525 	vcpu->arch.pio.count = 0;
11526 
11527 	return 1;
11528 }
11529 
11530 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
11531 			   unsigned int port, void *data,  unsigned int count)
11532 {
11533 	int ret;
11534 
11535 	ret = emulator_pio_out_emulated(vcpu->arch.emulate_ctxt, size, port,
11536 					data, count);
11537 	if (ret)
11538 		return ret;
11539 
11540 	vcpu->arch.pio.count = 0;
11541 
11542 	return 0;
11543 }
11544 
11545 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
11546 			  unsigned int port, void *data, unsigned int count)
11547 {
11548 	int ret;
11549 
11550 	ret = emulator_pio_in_emulated(vcpu->arch.emulate_ctxt, size, port,
11551 				       data, count);
11552 	if (ret) {
11553 		vcpu->arch.pio.count = 0;
11554 	} else {
11555 		vcpu->arch.guest_ins_data = data;
11556 		vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
11557 	}
11558 
11559 	return 0;
11560 }
11561 
11562 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
11563 			 unsigned int port, void *data,  unsigned int count,
11564 			 int in)
11565 {
11566 	return in ? kvm_sev_es_ins(vcpu, size, port, data, count)
11567 		  : kvm_sev_es_outs(vcpu, size, port, data, count);
11568 }
11569 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
11570 
11571 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
11572 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
11573 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
11574 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
11575 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
11576 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
11577 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
11578 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
11579 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
11580 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
11581 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
11582 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
11583 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
11584 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
11585 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
11586 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
11587 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
11588 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
11589 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
11590 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
11591 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
11592 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
11593 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
11594 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
11595 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
11596 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
11597 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);
11598