xref: /linux/arch/x86/kvm/x86.c (revision 41fb0cf1bced59c1fe178cf6cc9f716b5da9e40e)
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 #include "xen.h"
33 
34 #include <linux/clocksource.h>
35 #include <linux/interrupt.h>
36 #include <linux/kvm.h>
37 #include <linux/fs.h>
38 #include <linux/vmalloc.h>
39 #include <linux/export.h>
40 #include <linux/moduleparam.h>
41 #include <linux/mman.h>
42 #include <linux/highmem.h>
43 #include <linux/iommu.h>
44 #include <linux/intel-iommu.h>
45 #include <linux/cpufreq.h>
46 #include <linux/user-return-notifier.h>
47 #include <linux/srcu.h>
48 #include <linux/slab.h>
49 #include <linux/perf_event.h>
50 #include <linux/uaccess.h>
51 #include <linux/hash.h>
52 #include <linux/pci.h>
53 #include <linux/timekeeper_internal.h>
54 #include <linux/pvclock_gtod.h>
55 #include <linux/kvm_irqfd.h>
56 #include <linux/irqbypass.h>
57 #include <linux/sched/stat.h>
58 #include <linux/sched/isolation.h>
59 #include <linux/mem_encrypt.h>
60 #include <linux/entry-kvm.h>
61 #include <linux/suspend.h>
62 
63 #include <trace/events/kvm.h>
64 
65 #include <asm/debugreg.h>
66 #include <asm/msr.h>
67 #include <asm/desc.h>
68 #include <asm/mce.h>
69 #include <asm/pkru.h>
70 #include <linux/kernel_stat.h>
71 #include <asm/fpu/api.h>
72 #include <asm/fpu/xcr.h>
73 #include <asm/fpu/xstate.h>
74 #include <asm/pvclock.h>
75 #include <asm/div64.h>
76 #include <asm/irq_remapping.h>
77 #include <asm/mshyperv.h>
78 #include <asm/hypervisor.h>
79 #include <asm/tlbflush.h>
80 #include <asm/intel_pt.h>
81 #include <asm/emulate_prefix.h>
82 #include <asm/sgx.h>
83 #include <clocksource/hyperv_timer.h>
84 
85 #define CREATE_TRACE_POINTS
86 #include "trace.h"
87 
88 #define MAX_IO_MSRS 256
89 #define KVM_MAX_MCE_BANKS 32
90 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
91 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
92 
93 #define emul_to_vcpu(ctxt) \
94 	((struct kvm_vcpu *)(ctxt)->vcpu)
95 
96 /* EFER defaults:
97  * - enable syscall per default because its emulated by KVM
98  * - enable LME and LMA per default on 64 bit KVM
99  */
100 #ifdef CONFIG_X86_64
101 static
102 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
103 #else
104 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
105 #endif
106 
107 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
108 
109 #define KVM_EXIT_HYPERCALL_VALID_MASK (1 << KVM_HC_MAP_GPA_RANGE)
110 
111 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
112                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
113 
114 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
115 static void process_nmi(struct kvm_vcpu *vcpu);
116 static void process_smi(struct kvm_vcpu *vcpu);
117 static void enter_smm(struct kvm_vcpu *vcpu);
118 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
119 static void store_regs(struct kvm_vcpu *vcpu);
120 static int sync_regs(struct kvm_vcpu *vcpu);
121 
122 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
123 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
124 
125 struct kvm_x86_ops kvm_x86_ops __read_mostly;
126 EXPORT_SYMBOL_GPL(kvm_x86_ops);
127 
128 #define KVM_X86_OP(func)					     \
129 	DEFINE_STATIC_CALL_NULL(kvm_x86_##func,			     \
130 				*(((struct kvm_x86_ops *)0)->func));
131 #define KVM_X86_OP_NULL KVM_X86_OP
132 #include <asm/kvm-x86-ops.h>
133 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
134 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
135 EXPORT_STATIC_CALL_GPL(kvm_x86_tlb_flush_current);
136 
137 static bool __read_mostly ignore_msrs = 0;
138 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
139 
140 bool __read_mostly report_ignored_msrs = true;
141 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
142 EXPORT_SYMBOL_GPL(report_ignored_msrs);
143 
144 unsigned int min_timer_period_us = 200;
145 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
146 
147 static bool __read_mostly kvmclock_periodic_sync = true;
148 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
149 
150 bool __read_mostly kvm_has_tsc_control;
151 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
152 u32  __read_mostly kvm_max_guest_tsc_khz;
153 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
154 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
155 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
156 u64  __read_mostly kvm_max_tsc_scaling_ratio;
157 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
158 u64 __read_mostly kvm_default_tsc_scaling_ratio;
159 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
160 bool __read_mostly kvm_has_bus_lock_exit;
161 EXPORT_SYMBOL_GPL(kvm_has_bus_lock_exit);
162 
163 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
164 static u32 __read_mostly tsc_tolerance_ppm = 250;
165 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
166 
167 /*
168  * lapic timer advance (tscdeadline mode only) in nanoseconds.  '-1' enables
169  * adaptive tuning starting from default advancement of 1000ns.  '0' disables
170  * advancement entirely.  Any other value is used as-is and disables adaptive
171  * tuning, i.e. allows privileged userspace to set an exact advancement time.
172  */
173 static int __read_mostly lapic_timer_advance_ns = -1;
174 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
175 
176 static bool __read_mostly vector_hashing = true;
177 module_param(vector_hashing, bool, S_IRUGO);
178 
179 bool __read_mostly enable_vmware_backdoor = false;
180 module_param(enable_vmware_backdoor, bool, S_IRUGO);
181 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
182 
183 static bool __read_mostly force_emulation_prefix = false;
184 module_param(force_emulation_prefix, bool, S_IRUGO);
185 
186 int __read_mostly pi_inject_timer = -1;
187 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
188 
189 /*
190  * Restoring the host value for MSRs that are only consumed when running in
191  * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
192  * returns to userspace, i.e. the kernel can run with the guest's value.
193  */
194 #define KVM_MAX_NR_USER_RETURN_MSRS 16
195 
196 struct kvm_user_return_msrs {
197 	struct user_return_notifier urn;
198 	bool registered;
199 	struct kvm_user_return_msr_values {
200 		u64 host;
201 		u64 curr;
202 	} values[KVM_MAX_NR_USER_RETURN_MSRS];
203 };
204 
205 u32 __read_mostly kvm_nr_uret_msrs;
206 EXPORT_SYMBOL_GPL(kvm_nr_uret_msrs);
207 static u32 __read_mostly kvm_uret_msrs_list[KVM_MAX_NR_USER_RETURN_MSRS];
208 static struct kvm_user_return_msrs __percpu *user_return_msrs;
209 
210 #define KVM_SUPPORTED_XCR0     (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
211 				| XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
212 				| XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
213 				| XFEATURE_MASK_PKRU)
214 
215 u64 __read_mostly host_efer;
216 EXPORT_SYMBOL_GPL(host_efer);
217 
218 bool __read_mostly allow_smaller_maxphyaddr = 0;
219 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
220 
221 bool __read_mostly enable_apicv = true;
222 EXPORT_SYMBOL_GPL(enable_apicv);
223 
224 u64 __read_mostly host_xss;
225 EXPORT_SYMBOL_GPL(host_xss);
226 u64 __read_mostly supported_xss;
227 EXPORT_SYMBOL_GPL(supported_xss);
228 
229 const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
230 	KVM_GENERIC_VM_STATS(),
231 	STATS_DESC_COUNTER(VM, mmu_shadow_zapped),
232 	STATS_DESC_COUNTER(VM, mmu_pte_write),
233 	STATS_DESC_COUNTER(VM, mmu_pde_zapped),
234 	STATS_DESC_COUNTER(VM, mmu_flooded),
235 	STATS_DESC_COUNTER(VM, mmu_recycled),
236 	STATS_DESC_COUNTER(VM, mmu_cache_miss),
237 	STATS_DESC_ICOUNTER(VM, mmu_unsync),
238 	STATS_DESC_ICOUNTER(VM, pages_4k),
239 	STATS_DESC_ICOUNTER(VM, pages_2m),
240 	STATS_DESC_ICOUNTER(VM, pages_1g),
241 	STATS_DESC_ICOUNTER(VM, nx_lpage_splits),
242 	STATS_DESC_PCOUNTER(VM, max_mmu_rmap_size),
243 	STATS_DESC_PCOUNTER(VM, max_mmu_page_hash_collisions)
244 };
245 
246 const struct kvm_stats_header kvm_vm_stats_header = {
247 	.name_size = KVM_STATS_NAME_SIZE,
248 	.num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
249 	.id_offset = sizeof(struct kvm_stats_header),
250 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
251 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
252 		       sizeof(kvm_vm_stats_desc),
253 };
254 
255 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
256 	KVM_GENERIC_VCPU_STATS(),
257 	STATS_DESC_COUNTER(VCPU, pf_fixed),
258 	STATS_DESC_COUNTER(VCPU, pf_guest),
259 	STATS_DESC_COUNTER(VCPU, tlb_flush),
260 	STATS_DESC_COUNTER(VCPU, invlpg),
261 	STATS_DESC_COUNTER(VCPU, exits),
262 	STATS_DESC_COUNTER(VCPU, io_exits),
263 	STATS_DESC_COUNTER(VCPU, mmio_exits),
264 	STATS_DESC_COUNTER(VCPU, signal_exits),
265 	STATS_DESC_COUNTER(VCPU, irq_window_exits),
266 	STATS_DESC_COUNTER(VCPU, nmi_window_exits),
267 	STATS_DESC_COUNTER(VCPU, l1d_flush),
268 	STATS_DESC_COUNTER(VCPU, halt_exits),
269 	STATS_DESC_COUNTER(VCPU, request_irq_exits),
270 	STATS_DESC_COUNTER(VCPU, irq_exits),
271 	STATS_DESC_COUNTER(VCPU, host_state_reload),
272 	STATS_DESC_COUNTER(VCPU, fpu_reload),
273 	STATS_DESC_COUNTER(VCPU, insn_emulation),
274 	STATS_DESC_COUNTER(VCPU, insn_emulation_fail),
275 	STATS_DESC_COUNTER(VCPU, hypercalls),
276 	STATS_DESC_COUNTER(VCPU, irq_injections),
277 	STATS_DESC_COUNTER(VCPU, nmi_injections),
278 	STATS_DESC_COUNTER(VCPU, req_event),
279 	STATS_DESC_COUNTER(VCPU, nested_run),
280 	STATS_DESC_COUNTER(VCPU, directed_yield_attempted),
281 	STATS_DESC_COUNTER(VCPU, directed_yield_successful),
282 	STATS_DESC_ICOUNTER(VCPU, guest_mode)
283 };
284 
285 const struct kvm_stats_header kvm_vcpu_stats_header = {
286 	.name_size = KVM_STATS_NAME_SIZE,
287 	.num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
288 	.id_offset = sizeof(struct kvm_stats_header),
289 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
290 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
291 		       sizeof(kvm_vcpu_stats_desc),
292 };
293 
294 u64 __read_mostly host_xcr0;
295 u64 __read_mostly supported_xcr0;
296 EXPORT_SYMBOL_GPL(supported_xcr0);
297 
298 static struct kmem_cache *x86_emulator_cache;
299 
300 /*
301  * When called, it means the previous get/set msr reached an invalid msr.
302  * Return true if we want to ignore/silent this failed msr access.
303  */
304 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
305 {
306 	const char *op = write ? "wrmsr" : "rdmsr";
307 
308 	if (ignore_msrs) {
309 		if (report_ignored_msrs)
310 			kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
311 				      op, msr, data);
312 		/* Mask the error */
313 		return true;
314 	} else {
315 		kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
316 				      op, msr, data);
317 		return false;
318 	}
319 }
320 
321 static struct kmem_cache *kvm_alloc_emulator_cache(void)
322 {
323 	unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
324 	unsigned int size = sizeof(struct x86_emulate_ctxt);
325 
326 	return kmem_cache_create_usercopy("x86_emulator", size,
327 					  __alignof__(struct x86_emulate_ctxt),
328 					  SLAB_ACCOUNT, useroffset,
329 					  size - useroffset, NULL);
330 }
331 
332 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
333 
334 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
335 {
336 	int i;
337 	for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
338 		vcpu->arch.apf.gfns[i] = ~0;
339 }
340 
341 static void kvm_on_user_return(struct user_return_notifier *urn)
342 {
343 	unsigned slot;
344 	struct kvm_user_return_msrs *msrs
345 		= container_of(urn, struct kvm_user_return_msrs, urn);
346 	struct kvm_user_return_msr_values *values;
347 	unsigned long flags;
348 
349 	/*
350 	 * Disabling irqs at this point since the following code could be
351 	 * interrupted and executed through kvm_arch_hardware_disable()
352 	 */
353 	local_irq_save(flags);
354 	if (msrs->registered) {
355 		msrs->registered = false;
356 		user_return_notifier_unregister(urn);
357 	}
358 	local_irq_restore(flags);
359 	for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
360 		values = &msrs->values[slot];
361 		if (values->host != values->curr) {
362 			wrmsrl(kvm_uret_msrs_list[slot], values->host);
363 			values->curr = values->host;
364 		}
365 	}
366 }
367 
368 static int kvm_probe_user_return_msr(u32 msr)
369 {
370 	u64 val;
371 	int ret;
372 
373 	preempt_disable();
374 	ret = rdmsrl_safe(msr, &val);
375 	if (ret)
376 		goto out;
377 	ret = wrmsrl_safe(msr, val);
378 out:
379 	preempt_enable();
380 	return ret;
381 }
382 
383 int kvm_add_user_return_msr(u32 msr)
384 {
385 	BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
386 
387 	if (kvm_probe_user_return_msr(msr))
388 		return -1;
389 
390 	kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
391 	return kvm_nr_uret_msrs++;
392 }
393 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
394 
395 int kvm_find_user_return_msr(u32 msr)
396 {
397 	int i;
398 
399 	for (i = 0; i < kvm_nr_uret_msrs; ++i) {
400 		if (kvm_uret_msrs_list[i] == msr)
401 			return i;
402 	}
403 	return -1;
404 }
405 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
406 
407 static void kvm_user_return_msr_cpu_online(void)
408 {
409 	unsigned int cpu = smp_processor_id();
410 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
411 	u64 value;
412 	int i;
413 
414 	for (i = 0; i < kvm_nr_uret_msrs; ++i) {
415 		rdmsrl_safe(kvm_uret_msrs_list[i], &value);
416 		msrs->values[i].host = value;
417 		msrs->values[i].curr = value;
418 	}
419 }
420 
421 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
422 {
423 	unsigned int cpu = smp_processor_id();
424 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
425 	int err;
426 
427 	value = (value & mask) | (msrs->values[slot].host & ~mask);
428 	if (value == msrs->values[slot].curr)
429 		return 0;
430 	err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
431 	if (err)
432 		return 1;
433 
434 	msrs->values[slot].curr = value;
435 	if (!msrs->registered) {
436 		msrs->urn.on_user_return = kvm_on_user_return;
437 		user_return_notifier_register(&msrs->urn);
438 		msrs->registered = true;
439 	}
440 	return 0;
441 }
442 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
443 
444 static void drop_user_return_notifiers(void)
445 {
446 	unsigned int cpu = smp_processor_id();
447 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
448 
449 	if (msrs->registered)
450 		kvm_on_user_return(&msrs->urn);
451 }
452 
453 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
454 {
455 	return vcpu->arch.apic_base;
456 }
457 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
458 
459 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
460 {
461 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
462 }
463 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
464 
465 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
466 {
467 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
468 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
469 	u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
470 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
471 
472 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
473 		return 1;
474 	if (!msr_info->host_initiated) {
475 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
476 			return 1;
477 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
478 			return 1;
479 	}
480 
481 	kvm_lapic_set_base(vcpu, msr_info->data);
482 	kvm_recalculate_apic_map(vcpu->kvm);
483 	return 0;
484 }
485 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
486 
487 /*
488  * Handle a fault on a hardware virtualization (VMX or SVM) instruction.
489  *
490  * Hardware virtualization extension instructions may fault if a reboot turns
491  * off virtualization while processes are running.  Usually after catching the
492  * fault we just panic; during reboot instead the instruction is ignored.
493  */
494 noinstr void kvm_spurious_fault(void)
495 {
496 	/* Fault while not rebooting.  We want the trace. */
497 	BUG_ON(!kvm_rebooting);
498 }
499 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
500 
501 #define EXCPT_BENIGN		0
502 #define EXCPT_CONTRIBUTORY	1
503 #define EXCPT_PF		2
504 
505 static int exception_class(int vector)
506 {
507 	switch (vector) {
508 	case PF_VECTOR:
509 		return EXCPT_PF;
510 	case DE_VECTOR:
511 	case TS_VECTOR:
512 	case NP_VECTOR:
513 	case SS_VECTOR:
514 	case GP_VECTOR:
515 		return EXCPT_CONTRIBUTORY;
516 	default:
517 		break;
518 	}
519 	return EXCPT_BENIGN;
520 }
521 
522 #define EXCPT_FAULT		0
523 #define EXCPT_TRAP		1
524 #define EXCPT_ABORT		2
525 #define EXCPT_INTERRUPT		3
526 
527 static int exception_type(int vector)
528 {
529 	unsigned int mask;
530 
531 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
532 		return EXCPT_INTERRUPT;
533 
534 	mask = 1 << vector;
535 
536 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
537 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
538 		return EXCPT_TRAP;
539 
540 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
541 		return EXCPT_ABORT;
542 
543 	/* Reserved exceptions will result in fault */
544 	return EXCPT_FAULT;
545 }
546 
547 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
548 {
549 	unsigned nr = vcpu->arch.exception.nr;
550 	bool has_payload = vcpu->arch.exception.has_payload;
551 	unsigned long payload = vcpu->arch.exception.payload;
552 
553 	if (!has_payload)
554 		return;
555 
556 	switch (nr) {
557 	case DB_VECTOR:
558 		/*
559 		 * "Certain debug exceptions may clear bit 0-3.  The
560 		 * remaining contents of the DR6 register are never
561 		 * cleared by the processor".
562 		 */
563 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
564 		/*
565 		 * In order to reflect the #DB exception payload in guest
566 		 * dr6, three components need to be considered: active low
567 		 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
568 		 * DR6_BS and DR6_BT)
569 		 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
570 		 * In the target guest dr6:
571 		 * FIXED_1 bits should always be set.
572 		 * Active low bits should be cleared if 1-setting in payload.
573 		 * Active high bits should be set if 1-setting in payload.
574 		 *
575 		 * Note, the payload is compatible with the pending debug
576 		 * exceptions/exit qualification under VMX, that active_low bits
577 		 * are active high in payload.
578 		 * So they need to be flipped for DR6.
579 		 */
580 		vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
581 		vcpu->arch.dr6 |= payload;
582 		vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
583 
584 		/*
585 		 * The #DB payload is defined as compatible with the 'pending
586 		 * debug exceptions' field under VMX, not DR6. While bit 12 is
587 		 * defined in the 'pending debug exceptions' field (enabled
588 		 * breakpoint), it is reserved and must be zero in DR6.
589 		 */
590 		vcpu->arch.dr6 &= ~BIT(12);
591 		break;
592 	case PF_VECTOR:
593 		vcpu->arch.cr2 = payload;
594 		break;
595 	}
596 
597 	vcpu->arch.exception.has_payload = false;
598 	vcpu->arch.exception.payload = 0;
599 }
600 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
601 
602 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
603 		unsigned nr, bool has_error, u32 error_code,
604 	        bool has_payload, unsigned long payload, bool reinject)
605 {
606 	u32 prev_nr;
607 	int class1, class2;
608 
609 	kvm_make_request(KVM_REQ_EVENT, vcpu);
610 
611 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
612 	queue:
613 		if (reinject) {
614 			/*
615 			 * On vmentry, vcpu->arch.exception.pending is only
616 			 * true if an event injection was blocked by
617 			 * nested_run_pending.  In that case, however,
618 			 * vcpu_enter_guest requests an immediate exit,
619 			 * and the guest shouldn't proceed far enough to
620 			 * need reinjection.
621 			 */
622 			WARN_ON_ONCE(vcpu->arch.exception.pending);
623 			vcpu->arch.exception.injected = true;
624 			if (WARN_ON_ONCE(has_payload)) {
625 				/*
626 				 * A reinjected event has already
627 				 * delivered its payload.
628 				 */
629 				has_payload = false;
630 				payload = 0;
631 			}
632 		} else {
633 			vcpu->arch.exception.pending = true;
634 			vcpu->arch.exception.injected = false;
635 		}
636 		vcpu->arch.exception.has_error_code = has_error;
637 		vcpu->arch.exception.nr = nr;
638 		vcpu->arch.exception.error_code = error_code;
639 		vcpu->arch.exception.has_payload = has_payload;
640 		vcpu->arch.exception.payload = payload;
641 		if (!is_guest_mode(vcpu))
642 			kvm_deliver_exception_payload(vcpu);
643 		return;
644 	}
645 
646 	/* to check exception */
647 	prev_nr = vcpu->arch.exception.nr;
648 	if (prev_nr == DF_VECTOR) {
649 		/* triple fault -> shutdown */
650 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
651 		return;
652 	}
653 	class1 = exception_class(prev_nr);
654 	class2 = exception_class(nr);
655 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
656 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
657 		/*
658 		 * Generate double fault per SDM Table 5-5.  Set
659 		 * exception.pending = true so that the double fault
660 		 * can trigger a nested vmexit.
661 		 */
662 		vcpu->arch.exception.pending = true;
663 		vcpu->arch.exception.injected = false;
664 		vcpu->arch.exception.has_error_code = true;
665 		vcpu->arch.exception.nr = DF_VECTOR;
666 		vcpu->arch.exception.error_code = 0;
667 		vcpu->arch.exception.has_payload = false;
668 		vcpu->arch.exception.payload = 0;
669 	} else
670 		/* replace previous exception with a new one in a hope
671 		   that instruction re-execution will regenerate lost
672 		   exception */
673 		goto queue;
674 }
675 
676 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
677 {
678 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
679 }
680 EXPORT_SYMBOL_GPL(kvm_queue_exception);
681 
682 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
683 {
684 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
685 }
686 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
687 
688 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
689 			   unsigned long payload)
690 {
691 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
692 }
693 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
694 
695 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
696 				    u32 error_code, unsigned long payload)
697 {
698 	kvm_multiple_exception(vcpu, nr, true, error_code,
699 			       true, payload, false);
700 }
701 
702 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
703 {
704 	if (err)
705 		kvm_inject_gp(vcpu, 0);
706 	else
707 		return kvm_skip_emulated_instruction(vcpu);
708 
709 	return 1;
710 }
711 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
712 
713 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
714 {
715 	++vcpu->stat.pf_guest;
716 	vcpu->arch.exception.nested_apf =
717 		is_guest_mode(vcpu) && fault->async_page_fault;
718 	if (vcpu->arch.exception.nested_apf) {
719 		vcpu->arch.apf.nested_apf_token = fault->address;
720 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
721 	} else {
722 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
723 					fault->address);
724 	}
725 }
726 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
727 
728 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
729 				    struct x86_exception *fault)
730 {
731 	struct kvm_mmu *fault_mmu;
732 	WARN_ON_ONCE(fault->vector != PF_VECTOR);
733 
734 	fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
735 					       vcpu->arch.walk_mmu;
736 
737 	/*
738 	 * Invalidate the TLB entry for the faulting address, if it exists,
739 	 * else the access will fault indefinitely (and to emulate hardware).
740 	 */
741 	if ((fault->error_code & PFERR_PRESENT_MASK) &&
742 	    !(fault->error_code & PFERR_RSVD_MASK))
743 		kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
744 				       fault_mmu->root_hpa);
745 
746 	fault_mmu->inject_page_fault(vcpu, fault);
747 	return fault->nested_page_fault;
748 }
749 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
750 
751 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
752 {
753 	atomic_inc(&vcpu->arch.nmi_queued);
754 	kvm_make_request(KVM_REQ_NMI, vcpu);
755 }
756 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
757 
758 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
759 {
760 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
761 }
762 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
763 
764 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
765 {
766 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
767 }
768 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
769 
770 /*
771  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
772  * a #GP and return false.
773  */
774 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
775 {
776 	if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
777 		return true;
778 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
779 	return false;
780 }
781 EXPORT_SYMBOL_GPL(kvm_require_cpl);
782 
783 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
784 {
785 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
786 		return true;
787 
788 	kvm_queue_exception(vcpu, UD_VECTOR);
789 	return false;
790 }
791 EXPORT_SYMBOL_GPL(kvm_require_dr);
792 
793 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
794 {
795 	return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
796 }
797 
798 /*
799  * Load the pae pdptrs.  Return 1 if they are all valid, 0 otherwise.
800  */
801 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
802 {
803 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
804 	gpa_t real_gpa;
805 	int i;
806 	int ret;
807 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
808 
809 	/*
810 	 * If the MMU is nested, CR3 holds an L2 GPA and needs to be translated
811 	 * to an L1 GPA.
812 	 */
813 	real_gpa = mmu->translate_gpa(vcpu, gfn_to_gpa(pdpt_gfn),
814 				      PFERR_USER_MASK | PFERR_WRITE_MASK, NULL);
815 	if (real_gpa == UNMAPPED_GVA)
816 		return 0;
817 
818 	/* Note the offset, PDPTRs are 32 byte aligned when using PAE paging. */
819 	ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(real_gpa), pdpte,
820 				       cr3 & GENMASK(11, 5), sizeof(pdpte));
821 	if (ret < 0)
822 		return 0;
823 
824 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
825 		if ((pdpte[i] & PT_PRESENT_MASK) &&
826 		    (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
827 			return 0;
828 		}
829 	}
830 
831 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
832 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
833 	vcpu->arch.pdptrs_from_userspace = false;
834 
835 	return 1;
836 }
837 EXPORT_SYMBOL_GPL(load_pdptrs);
838 
839 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
840 {
841 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
842 		kvm_clear_async_pf_completion_queue(vcpu);
843 		kvm_async_pf_hash_reset(vcpu);
844 	}
845 
846 	if ((cr0 ^ old_cr0) & KVM_MMU_CR0_ROLE_BITS)
847 		kvm_mmu_reset_context(vcpu);
848 
849 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
850 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
851 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
852 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
853 }
854 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
855 
856 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
857 {
858 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
859 	unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
860 
861 	cr0 |= X86_CR0_ET;
862 
863 #ifdef CONFIG_X86_64
864 	if (cr0 & 0xffffffff00000000UL)
865 		return 1;
866 #endif
867 
868 	cr0 &= ~CR0_RESERVED_BITS;
869 
870 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
871 		return 1;
872 
873 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
874 		return 1;
875 
876 #ifdef CONFIG_X86_64
877 	if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
878 	    (cr0 & X86_CR0_PG)) {
879 		int cs_db, cs_l;
880 
881 		if (!is_pae(vcpu))
882 			return 1;
883 		static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
884 		if (cs_l)
885 			return 1;
886 	}
887 #endif
888 	if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
889 	    is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
890 	    !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
891 		return 1;
892 
893 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
894 		return 1;
895 
896 	static_call(kvm_x86_set_cr0)(vcpu, cr0);
897 
898 	kvm_post_set_cr0(vcpu, old_cr0, cr0);
899 
900 	return 0;
901 }
902 EXPORT_SYMBOL_GPL(kvm_set_cr0);
903 
904 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
905 {
906 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
907 }
908 EXPORT_SYMBOL_GPL(kvm_lmsw);
909 
910 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
911 {
912 	if (vcpu->arch.guest_state_protected)
913 		return;
914 
915 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
916 
917 		if (vcpu->arch.xcr0 != host_xcr0)
918 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
919 
920 		if (vcpu->arch.xsaves_enabled &&
921 		    vcpu->arch.ia32_xss != host_xss)
922 			wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
923 	}
924 
925 	if (static_cpu_has(X86_FEATURE_PKU) &&
926 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
927 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
928 	    vcpu->arch.pkru != vcpu->arch.host_pkru)
929 		write_pkru(vcpu->arch.pkru);
930 }
931 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
932 
933 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
934 {
935 	if (vcpu->arch.guest_state_protected)
936 		return;
937 
938 	if (static_cpu_has(X86_FEATURE_PKU) &&
939 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
940 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
941 		vcpu->arch.pkru = rdpkru();
942 		if (vcpu->arch.pkru != vcpu->arch.host_pkru)
943 			write_pkru(vcpu->arch.host_pkru);
944 	}
945 
946 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
947 
948 		if (vcpu->arch.xcr0 != host_xcr0)
949 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
950 
951 		if (vcpu->arch.xsaves_enabled &&
952 		    vcpu->arch.ia32_xss != host_xss)
953 			wrmsrl(MSR_IA32_XSS, host_xss);
954 	}
955 
956 }
957 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
958 
959 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
960 {
961 	u64 xcr0 = xcr;
962 	u64 old_xcr0 = vcpu->arch.xcr0;
963 	u64 valid_bits;
964 
965 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
966 	if (index != XCR_XFEATURE_ENABLED_MASK)
967 		return 1;
968 	if (!(xcr0 & XFEATURE_MASK_FP))
969 		return 1;
970 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
971 		return 1;
972 
973 	/*
974 	 * Do not allow the guest to set bits that we do not support
975 	 * saving.  However, xcr0 bit 0 is always set, even if the
976 	 * emulated CPU does not support XSAVE (see kvm_vcpu_reset()).
977 	 */
978 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
979 	if (xcr0 & ~valid_bits)
980 		return 1;
981 
982 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
983 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
984 		return 1;
985 
986 	if (xcr0 & XFEATURE_MASK_AVX512) {
987 		if (!(xcr0 & XFEATURE_MASK_YMM))
988 			return 1;
989 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
990 			return 1;
991 	}
992 	vcpu->arch.xcr0 = xcr0;
993 
994 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
995 		kvm_update_cpuid_runtime(vcpu);
996 	return 0;
997 }
998 
999 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
1000 {
1001 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
1002 	    __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
1003 		kvm_inject_gp(vcpu, 0);
1004 		return 1;
1005 	}
1006 
1007 	return kvm_skip_emulated_instruction(vcpu);
1008 }
1009 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1010 
1011 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1012 {
1013 	if (cr4 & cr4_reserved_bits)
1014 		return false;
1015 
1016 	if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1017 		return false;
1018 
1019 	return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1020 }
1021 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
1022 
1023 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1024 {
1025 	/*
1026 	 * If any role bit is changed, the MMU needs to be reset.
1027 	 *
1028 	 * If CR4.PCIDE is changed 1 -> 0, the guest TLB must be flushed.
1029 	 * If CR4.PCIDE is changed 0 -> 1, there is no need to flush the TLB
1030 	 * according to the SDM; however, stale prev_roots could be reused
1031 	 * incorrectly in the future after a MOV to CR3 with NOFLUSH=1, so we
1032 	 * free them all.  KVM_REQ_MMU_RELOAD is fit for the both cases; it
1033 	 * is slow, but changing CR4.PCIDE is a rare case.
1034 	 *
1035 	 * If CR4.PGE is changed, the guest TLB must be flushed.
1036 	 *
1037 	 * Note: resetting MMU is a superset of KVM_REQ_MMU_RELOAD and
1038 	 * KVM_REQ_MMU_RELOAD is a superset of KVM_REQ_TLB_FLUSH_GUEST, hence
1039 	 * the usage of "else if".
1040 	 */
1041 	if ((cr4 ^ old_cr4) & KVM_MMU_CR4_ROLE_BITS)
1042 		kvm_mmu_reset_context(vcpu);
1043 	else if ((cr4 ^ old_cr4) & X86_CR4_PCIDE)
1044 		kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
1045 	else if ((cr4 ^ old_cr4) & X86_CR4_PGE)
1046 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1047 }
1048 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1049 
1050 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1051 {
1052 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
1053 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1054 				   X86_CR4_SMEP;
1055 
1056 	if (!kvm_is_valid_cr4(vcpu, cr4))
1057 		return 1;
1058 
1059 	if (is_long_mode(vcpu)) {
1060 		if (!(cr4 & X86_CR4_PAE))
1061 			return 1;
1062 		if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1063 			return 1;
1064 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1065 		   && ((cr4 ^ old_cr4) & pdptr_bits)
1066 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1067 				   kvm_read_cr3(vcpu)))
1068 		return 1;
1069 
1070 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1071 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1072 			return 1;
1073 
1074 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1075 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1076 			return 1;
1077 	}
1078 
1079 	static_call(kvm_x86_set_cr4)(vcpu, cr4);
1080 
1081 	kvm_post_set_cr4(vcpu, old_cr4, cr4);
1082 
1083 	return 0;
1084 }
1085 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1086 
1087 static void kvm_invalidate_pcid(struct kvm_vcpu *vcpu, unsigned long pcid)
1088 {
1089 	struct kvm_mmu *mmu = vcpu->arch.mmu;
1090 	unsigned long roots_to_free = 0;
1091 	int i;
1092 
1093 	/*
1094 	 * MOV CR3 and INVPCID are usually not intercepted when using TDP, but
1095 	 * this is reachable when running EPT=1 and unrestricted_guest=0,  and
1096 	 * also via the emulator.  KVM's TDP page tables are not in the scope of
1097 	 * the invalidation, but the guest's TLB entries need to be flushed as
1098 	 * the CPU may have cached entries in its TLB for the target PCID.
1099 	 */
1100 	if (unlikely(tdp_enabled)) {
1101 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1102 		return;
1103 	}
1104 
1105 	/*
1106 	 * If neither the current CR3 nor any of the prev_roots use the given
1107 	 * PCID, then nothing needs to be done here because a resync will
1108 	 * happen anyway before switching to any other CR3.
1109 	 */
1110 	if (kvm_get_active_pcid(vcpu) == pcid) {
1111 		kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
1112 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1113 	}
1114 
1115 	/*
1116 	 * If PCID is disabled, there is no need to free prev_roots even if the
1117 	 * PCIDs for them are also 0, because MOV to CR3 always flushes the TLB
1118 	 * with PCIDE=0.
1119 	 */
1120 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
1121 		return;
1122 
1123 	for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
1124 		if (kvm_get_pcid(vcpu, mmu->prev_roots[i].pgd) == pcid)
1125 			roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
1126 
1127 	kvm_mmu_free_roots(vcpu, mmu, roots_to_free);
1128 }
1129 
1130 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1131 {
1132 	bool skip_tlb_flush = false;
1133 	unsigned long pcid = 0;
1134 #ifdef CONFIG_X86_64
1135 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1136 
1137 	if (pcid_enabled) {
1138 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1139 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
1140 		pcid = cr3 & X86_CR3_PCID_MASK;
1141 	}
1142 #endif
1143 
1144 	/* PDPTRs are always reloaded for PAE paging. */
1145 	if (cr3 == kvm_read_cr3(vcpu) && !is_pae_paging(vcpu))
1146 		goto handle_tlb_flush;
1147 
1148 	/*
1149 	 * Do not condition the GPA check on long mode, this helper is used to
1150 	 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1151 	 * the current vCPU mode is accurate.
1152 	 */
1153 	if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1154 		return 1;
1155 
1156 	if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1157 		return 1;
1158 
1159 	if (cr3 != kvm_read_cr3(vcpu))
1160 		kvm_mmu_new_pgd(vcpu, cr3);
1161 
1162 	vcpu->arch.cr3 = cr3;
1163 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1164 
1165 handle_tlb_flush:
1166 	/*
1167 	 * A load of CR3 that flushes the TLB flushes only the current PCID,
1168 	 * even if PCID is disabled, in which case PCID=0 is flushed.  It's a
1169 	 * moot point in the end because _disabling_ PCID will flush all PCIDs,
1170 	 * and it's impossible to use a non-zero PCID when PCID is disabled,
1171 	 * i.e. only PCID=0 can be relevant.
1172 	 */
1173 	if (!skip_tlb_flush)
1174 		kvm_invalidate_pcid(vcpu, pcid);
1175 
1176 	return 0;
1177 }
1178 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1179 
1180 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1181 {
1182 	if (cr8 & CR8_RESERVED_BITS)
1183 		return 1;
1184 	if (lapic_in_kernel(vcpu))
1185 		kvm_lapic_set_tpr(vcpu, cr8);
1186 	else
1187 		vcpu->arch.cr8 = cr8;
1188 	return 0;
1189 }
1190 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1191 
1192 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1193 {
1194 	if (lapic_in_kernel(vcpu))
1195 		return kvm_lapic_get_cr8(vcpu);
1196 	else
1197 		return vcpu->arch.cr8;
1198 }
1199 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1200 
1201 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1202 {
1203 	int i;
1204 
1205 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1206 		for (i = 0; i < KVM_NR_DB_REGS; i++)
1207 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1208 	}
1209 }
1210 
1211 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1212 {
1213 	unsigned long dr7;
1214 
1215 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1216 		dr7 = vcpu->arch.guest_debug_dr7;
1217 	else
1218 		dr7 = vcpu->arch.dr7;
1219 	static_call(kvm_x86_set_dr7)(vcpu, dr7);
1220 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1221 	if (dr7 & DR7_BP_EN_MASK)
1222 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1223 }
1224 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1225 
1226 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1227 {
1228 	u64 fixed = DR6_FIXED_1;
1229 
1230 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1231 		fixed |= DR6_RTM;
1232 
1233 	if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1234 		fixed |= DR6_BUS_LOCK;
1235 	return fixed;
1236 }
1237 
1238 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1239 {
1240 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1241 
1242 	switch (dr) {
1243 	case 0 ... 3:
1244 		vcpu->arch.db[array_index_nospec(dr, size)] = val;
1245 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1246 			vcpu->arch.eff_db[dr] = val;
1247 		break;
1248 	case 4:
1249 	case 6:
1250 		if (!kvm_dr6_valid(val))
1251 			return 1; /* #GP */
1252 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1253 		break;
1254 	case 5:
1255 	default: /* 7 */
1256 		if (!kvm_dr7_valid(val))
1257 			return 1; /* #GP */
1258 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1259 		kvm_update_dr7(vcpu);
1260 		break;
1261 	}
1262 
1263 	return 0;
1264 }
1265 EXPORT_SYMBOL_GPL(kvm_set_dr);
1266 
1267 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1268 {
1269 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1270 
1271 	switch (dr) {
1272 	case 0 ... 3:
1273 		*val = vcpu->arch.db[array_index_nospec(dr, size)];
1274 		break;
1275 	case 4:
1276 	case 6:
1277 		*val = vcpu->arch.dr6;
1278 		break;
1279 	case 5:
1280 	default: /* 7 */
1281 		*val = vcpu->arch.dr7;
1282 		break;
1283 	}
1284 }
1285 EXPORT_SYMBOL_GPL(kvm_get_dr);
1286 
1287 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1288 {
1289 	u32 ecx = kvm_rcx_read(vcpu);
1290 	u64 data;
1291 
1292 	if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1293 		kvm_inject_gp(vcpu, 0);
1294 		return 1;
1295 	}
1296 
1297 	kvm_rax_write(vcpu, (u32)data);
1298 	kvm_rdx_write(vcpu, data >> 32);
1299 	return kvm_skip_emulated_instruction(vcpu);
1300 }
1301 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1302 
1303 /*
1304  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1305  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1306  *
1307  * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1308  * extract the supported MSRs from the related const lists.
1309  * msrs_to_save is selected from the msrs_to_save_all to reflect the
1310  * capabilities of the host cpu. This capabilities test skips MSRs that are
1311  * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1312  * may depend on host virtualization features rather than host cpu features.
1313  */
1314 
1315 static const u32 msrs_to_save_all[] = {
1316 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1317 	MSR_STAR,
1318 #ifdef CONFIG_X86_64
1319 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1320 #endif
1321 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1322 	MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1323 	MSR_IA32_SPEC_CTRL,
1324 	MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1325 	MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1326 	MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1327 	MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1328 	MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1329 	MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1330 	MSR_IA32_UMWAIT_CONTROL,
1331 
1332 	MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1333 	MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1334 	MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1335 	MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1336 	MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1337 	MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1338 	MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1339 	MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1340 	MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1341 	MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1342 	MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1343 	MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1344 	MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1345 	MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1346 	MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1347 	MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1348 	MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1349 	MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1350 	MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1351 	MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1352 	MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1353 	MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1354 
1355 	MSR_K7_EVNTSEL0, MSR_K7_EVNTSEL1, MSR_K7_EVNTSEL2, MSR_K7_EVNTSEL3,
1356 	MSR_K7_PERFCTR0, MSR_K7_PERFCTR1, MSR_K7_PERFCTR2, MSR_K7_PERFCTR3,
1357 	MSR_F15H_PERF_CTL0, MSR_F15H_PERF_CTL1, MSR_F15H_PERF_CTL2,
1358 	MSR_F15H_PERF_CTL3, MSR_F15H_PERF_CTL4, MSR_F15H_PERF_CTL5,
1359 	MSR_F15H_PERF_CTR0, MSR_F15H_PERF_CTR1, MSR_F15H_PERF_CTR2,
1360 	MSR_F15H_PERF_CTR3, MSR_F15H_PERF_CTR4, MSR_F15H_PERF_CTR5,
1361 };
1362 
1363 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1364 static unsigned num_msrs_to_save;
1365 
1366 static const u32 emulated_msrs_all[] = {
1367 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1368 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1369 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1370 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1371 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1372 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1373 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1374 	HV_X64_MSR_RESET,
1375 	HV_X64_MSR_VP_INDEX,
1376 	HV_X64_MSR_VP_RUNTIME,
1377 	HV_X64_MSR_SCONTROL,
1378 	HV_X64_MSR_STIMER0_CONFIG,
1379 	HV_X64_MSR_VP_ASSIST_PAGE,
1380 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1381 	HV_X64_MSR_TSC_EMULATION_STATUS,
1382 	HV_X64_MSR_SYNDBG_OPTIONS,
1383 	HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1384 	HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1385 	HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1386 
1387 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1388 	MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1389 
1390 	MSR_IA32_TSC_ADJUST,
1391 	MSR_IA32_TSC_DEADLINE,
1392 	MSR_IA32_ARCH_CAPABILITIES,
1393 	MSR_IA32_PERF_CAPABILITIES,
1394 	MSR_IA32_MISC_ENABLE,
1395 	MSR_IA32_MCG_STATUS,
1396 	MSR_IA32_MCG_CTL,
1397 	MSR_IA32_MCG_EXT_CTL,
1398 	MSR_IA32_SMBASE,
1399 	MSR_SMI_COUNT,
1400 	MSR_PLATFORM_INFO,
1401 	MSR_MISC_FEATURES_ENABLES,
1402 	MSR_AMD64_VIRT_SPEC_CTRL,
1403 	MSR_AMD64_TSC_RATIO,
1404 	MSR_IA32_POWER_CTL,
1405 	MSR_IA32_UCODE_REV,
1406 
1407 	/*
1408 	 * The following list leaves out MSRs whose values are determined
1409 	 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1410 	 * We always support the "true" VMX control MSRs, even if the host
1411 	 * processor does not, so I am putting these registers here rather
1412 	 * than in msrs_to_save_all.
1413 	 */
1414 	MSR_IA32_VMX_BASIC,
1415 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1416 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1417 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1418 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1419 	MSR_IA32_VMX_MISC,
1420 	MSR_IA32_VMX_CR0_FIXED0,
1421 	MSR_IA32_VMX_CR4_FIXED0,
1422 	MSR_IA32_VMX_VMCS_ENUM,
1423 	MSR_IA32_VMX_PROCBASED_CTLS2,
1424 	MSR_IA32_VMX_EPT_VPID_CAP,
1425 	MSR_IA32_VMX_VMFUNC,
1426 
1427 	MSR_K7_HWCR,
1428 	MSR_KVM_POLL_CONTROL,
1429 };
1430 
1431 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1432 static unsigned num_emulated_msrs;
1433 
1434 /*
1435  * List of msr numbers which are used to expose MSR-based features that
1436  * can be used by a hypervisor to validate requested CPU features.
1437  */
1438 static const u32 msr_based_features_all[] = {
1439 	MSR_IA32_VMX_BASIC,
1440 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1441 	MSR_IA32_VMX_PINBASED_CTLS,
1442 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1443 	MSR_IA32_VMX_PROCBASED_CTLS,
1444 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1445 	MSR_IA32_VMX_EXIT_CTLS,
1446 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1447 	MSR_IA32_VMX_ENTRY_CTLS,
1448 	MSR_IA32_VMX_MISC,
1449 	MSR_IA32_VMX_CR0_FIXED0,
1450 	MSR_IA32_VMX_CR0_FIXED1,
1451 	MSR_IA32_VMX_CR4_FIXED0,
1452 	MSR_IA32_VMX_CR4_FIXED1,
1453 	MSR_IA32_VMX_VMCS_ENUM,
1454 	MSR_IA32_VMX_PROCBASED_CTLS2,
1455 	MSR_IA32_VMX_EPT_VPID_CAP,
1456 	MSR_IA32_VMX_VMFUNC,
1457 
1458 	MSR_F10H_DECFG,
1459 	MSR_IA32_UCODE_REV,
1460 	MSR_IA32_ARCH_CAPABILITIES,
1461 	MSR_IA32_PERF_CAPABILITIES,
1462 };
1463 
1464 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1465 static unsigned int num_msr_based_features;
1466 
1467 static u64 kvm_get_arch_capabilities(void)
1468 {
1469 	u64 data = 0;
1470 
1471 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1472 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1473 
1474 	/*
1475 	 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1476 	 * the nested hypervisor runs with NX huge pages.  If it is not,
1477 	 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1478 	 * L1 guests, so it need not worry about its own (L2) guests.
1479 	 */
1480 	data |= ARCH_CAP_PSCHANGE_MC_NO;
1481 
1482 	/*
1483 	 * If we're doing cache flushes (either "always" or "cond")
1484 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1485 	 * If an outer hypervisor is doing the cache flush for us
1486 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1487 	 * capability to the guest too, and if EPT is disabled we're not
1488 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1489 	 * require a nested hypervisor to do a flush of its own.
1490 	 */
1491 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1492 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1493 
1494 	if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1495 		data |= ARCH_CAP_RDCL_NO;
1496 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1497 		data |= ARCH_CAP_SSB_NO;
1498 	if (!boot_cpu_has_bug(X86_BUG_MDS))
1499 		data |= ARCH_CAP_MDS_NO;
1500 
1501 	if (!boot_cpu_has(X86_FEATURE_RTM)) {
1502 		/*
1503 		 * If RTM=0 because the kernel has disabled TSX, the host might
1504 		 * have TAA_NO or TSX_CTRL.  Clear TAA_NO (the guest sees RTM=0
1505 		 * and therefore knows that there cannot be TAA) but keep
1506 		 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1507 		 * and we want to allow migrating those guests to tsx=off hosts.
1508 		 */
1509 		data &= ~ARCH_CAP_TAA_NO;
1510 	} else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1511 		data |= ARCH_CAP_TAA_NO;
1512 	} else {
1513 		/*
1514 		 * Nothing to do here; we emulate TSX_CTRL if present on the
1515 		 * host so the guest can choose between disabling TSX or
1516 		 * using VERW to clear CPU buffers.
1517 		 */
1518 	}
1519 
1520 	return data;
1521 }
1522 
1523 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1524 {
1525 	switch (msr->index) {
1526 	case MSR_IA32_ARCH_CAPABILITIES:
1527 		msr->data = kvm_get_arch_capabilities();
1528 		break;
1529 	case MSR_IA32_UCODE_REV:
1530 		rdmsrl_safe(msr->index, &msr->data);
1531 		break;
1532 	default:
1533 		return static_call(kvm_x86_get_msr_feature)(msr);
1534 	}
1535 	return 0;
1536 }
1537 
1538 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1539 {
1540 	struct kvm_msr_entry msr;
1541 	int r;
1542 
1543 	msr.index = index;
1544 	r = kvm_get_msr_feature(&msr);
1545 
1546 	if (r == KVM_MSR_RET_INVALID) {
1547 		/* Unconditionally clear the output for simplicity */
1548 		*data = 0;
1549 		if (kvm_msr_ignored_check(index, 0, false))
1550 			r = 0;
1551 	}
1552 
1553 	if (r)
1554 		return r;
1555 
1556 	*data = msr.data;
1557 
1558 	return 0;
1559 }
1560 
1561 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1562 {
1563 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1564 		return false;
1565 
1566 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1567 		return false;
1568 
1569 	if (efer & (EFER_LME | EFER_LMA) &&
1570 	    !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1571 		return false;
1572 
1573 	if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1574 		return false;
1575 
1576 	return true;
1577 
1578 }
1579 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1580 {
1581 	if (efer & efer_reserved_bits)
1582 		return false;
1583 
1584 	return __kvm_valid_efer(vcpu, efer);
1585 }
1586 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1587 
1588 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1589 {
1590 	u64 old_efer = vcpu->arch.efer;
1591 	u64 efer = msr_info->data;
1592 	int r;
1593 
1594 	if (efer & efer_reserved_bits)
1595 		return 1;
1596 
1597 	if (!msr_info->host_initiated) {
1598 		if (!__kvm_valid_efer(vcpu, efer))
1599 			return 1;
1600 
1601 		if (is_paging(vcpu) &&
1602 		    (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1603 			return 1;
1604 	}
1605 
1606 	efer &= ~EFER_LMA;
1607 	efer |= vcpu->arch.efer & EFER_LMA;
1608 
1609 	r = static_call(kvm_x86_set_efer)(vcpu, efer);
1610 	if (r) {
1611 		WARN_ON(r > 0);
1612 		return r;
1613 	}
1614 
1615 	/* Update reserved bits */
1616 	if ((efer ^ old_efer) & EFER_NX)
1617 		kvm_mmu_reset_context(vcpu);
1618 
1619 	return 0;
1620 }
1621 
1622 void kvm_enable_efer_bits(u64 mask)
1623 {
1624        efer_reserved_bits &= ~mask;
1625 }
1626 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1627 
1628 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1629 {
1630 	struct kvm_x86_msr_filter *msr_filter;
1631 	struct msr_bitmap_range *ranges;
1632 	struct kvm *kvm = vcpu->kvm;
1633 	bool allowed;
1634 	int idx;
1635 	u32 i;
1636 
1637 	/* x2APIC MSRs do not support filtering. */
1638 	if (index >= 0x800 && index <= 0x8ff)
1639 		return true;
1640 
1641 	idx = srcu_read_lock(&kvm->srcu);
1642 
1643 	msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1644 	if (!msr_filter) {
1645 		allowed = true;
1646 		goto out;
1647 	}
1648 
1649 	allowed = msr_filter->default_allow;
1650 	ranges = msr_filter->ranges;
1651 
1652 	for (i = 0; i < msr_filter->count; i++) {
1653 		u32 start = ranges[i].base;
1654 		u32 end = start + ranges[i].nmsrs;
1655 		u32 flags = ranges[i].flags;
1656 		unsigned long *bitmap = ranges[i].bitmap;
1657 
1658 		if ((index >= start) && (index < end) && (flags & type)) {
1659 			allowed = !!test_bit(index - start, bitmap);
1660 			break;
1661 		}
1662 	}
1663 
1664 out:
1665 	srcu_read_unlock(&kvm->srcu, idx);
1666 
1667 	return allowed;
1668 }
1669 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1670 
1671 /*
1672  * Write @data into the MSR specified by @index.  Select MSR specific fault
1673  * checks are bypassed if @host_initiated is %true.
1674  * Returns 0 on success, non-0 otherwise.
1675  * Assumes vcpu_load() was already called.
1676  */
1677 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1678 			 bool host_initiated)
1679 {
1680 	struct msr_data msr;
1681 
1682 	if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1683 		return KVM_MSR_RET_FILTERED;
1684 
1685 	switch (index) {
1686 	case MSR_FS_BASE:
1687 	case MSR_GS_BASE:
1688 	case MSR_KERNEL_GS_BASE:
1689 	case MSR_CSTAR:
1690 	case MSR_LSTAR:
1691 		if (is_noncanonical_address(data, vcpu))
1692 			return 1;
1693 		break;
1694 	case MSR_IA32_SYSENTER_EIP:
1695 	case MSR_IA32_SYSENTER_ESP:
1696 		/*
1697 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1698 		 * non-canonical address is written on Intel but not on
1699 		 * AMD (which ignores the top 32-bits, because it does
1700 		 * not implement 64-bit SYSENTER).
1701 		 *
1702 		 * 64-bit code should hence be able to write a non-canonical
1703 		 * value on AMD.  Making the address canonical ensures that
1704 		 * vmentry does not fail on Intel after writing a non-canonical
1705 		 * value, and that something deterministic happens if the guest
1706 		 * invokes 64-bit SYSENTER.
1707 		 */
1708 		data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1709 		break;
1710 	case MSR_TSC_AUX:
1711 		if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1712 			return 1;
1713 
1714 		if (!host_initiated &&
1715 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1716 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1717 			return 1;
1718 
1719 		/*
1720 		 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1721 		 * incomplete and conflicting architectural behavior.  Current
1722 		 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1723 		 * reserved and always read as zeros.  Enforce Intel's reserved
1724 		 * bits check if and only if the guest CPU is Intel, and clear
1725 		 * the bits in all other cases.  This ensures cross-vendor
1726 		 * migration will provide consistent behavior for the guest.
1727 		 */
1728 		if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1729 			return 1;
1730 
1731 		data = (u32)data;
1732 		break;
1733 	}
1734 
1735 	msr.data = data;
1736 	msr.index = index;
1737 	msr.host_initiated = host_initiated;
1738 
1739 	return static_call(kvm_x86_set_msr)(vcpu, &msr);
1740 }
1741 
1742 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1743 				     u32 index, u64 data, bool host_initiated)
1744 {
1745 	int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1746 
1747 	if (ret == KVM_MSR_RET_INVALID)
1748 		if (kvm_msr_ignored_check(index, data, true))
1749 			ret = 0;
1750 
1751 	return ret;
1752 }
1753 
1754 /*
1755  * Read the MSR specified by @index into @data.  Select MSR specific fault
1756  * checks are bypassed if @host_initiated is %true.
1757  * Returns 0 on success, non-0 otherwise.
1758  * Assumes vcpu_load() was already called.
1759  */
1760 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1761 		  bool host_initiated)
1762 {
1763 	struct msr_data msr;
1764 	int ret;
1765 
1766 	if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1767 		return KVM_MSR_RET_FILTERED;
1768 
1769 	switch (index) {
1770 	case MSR_TSC_AUX:
1771 		if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1772 			return 1;
1773 
1774 		if (!host_initiated &&
1775 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1776 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1777 			return 1;
1778 		break;
1779 	}
1780 
1781 	msr.index = index;
1782 	msr.host_initiated = host_initiated;
1783 
1784 	ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1785 	if (!ret)
1786 		*data = msr.data;
1787 	return ret;
1788 }
1789 
1790 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1791 				     u32 index, u64 *data, bool host_initiated)
1792 {
1793 	int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1794 
1795 	if (ret == KVM_MSR_RET_INVALID) {
1796 		/* Unconditionally clear *data for simplicity */
1797 		*data = 0;
1798 		if (kvm_msr_ignored_check(index, 0, false))
1799 			ret = 0;
1800 	}
1801 
1802 	return ret;
1803 }
1804 
1805 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1806 {
1807 	return kvm_get_msr_ignored_check(vcpu, index, data, false);
1808 }
1809 EXPORT_SYMBOL_GPL(kvm_get_msr);
1810 
1811 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1812 {
1813 	return kvm_set_msr_ignored_check(vcpu, index, data, false);
1814 }
1815 EXPORT_SYMBOL_GPL(kvm_set_msr);
1816 
1817 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1818 {
1819 	int err = vcpu->run->msr.error;
1820 	if (!err) {
1821 		kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1822 		kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1823 	}
1824 
1825 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, err);
1826 }
1827 
1828 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1829 {
1830 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1831 }
1832 
1833 static u64 kvm_msr_reason(int r)
1834 {
1835 	switch (r) {
1836 	case KVM_MSR_RET_INVALID:
1837 		return KVM_MSR_EXIT_REASON_UNKNOWN;
1838 	case KVM_MSR_RET_FILTERED:
1839 		return KVM_MSR_EXIT_REASON_FILTER;
1840 	default:
1841 		return KVM_MSR_EXIT_REASON_INVAL;
1842 	}
1843 }
1844 
1845 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1846 			      u32 exit_reason, u64 data,
1847 			      int (*completion)(struct kvm_vcpu *vcpu),
1848 			      int r)
1849 {
1850 	u64 msr_reason = kvm_msr_reason(r);
1851 
1852 	/* Check if the user wanted to know about this MSR fault */
1853 	if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1854 		return 0;
1855 
1856 	vcpu->run->exit_reason = exit_reason;
1857 	vcpu->run->msr.error = 0;
1858 	memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1859 	vcpu->run->msr.reason = msr_reason;
1860 	vcpu->run->msr.index = index;
1861 	vcpu->run->msr.data = data;
1862 	vcpu->arch.complete_userspace_io = completion;
1863 
1864 	return 1;
1865 }
1866 
1867 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1868 {
1869 	return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1870 				   complete_emulated_rdmsr, r);
1871 }
1872 
1873 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1874 {
1875 	return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1876 				   complete_emulated_wrmsr, r);
1877 }
1878 
1879 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1880 {
1881 	u32 ecx = kvm_rcx_read(vcpu);
1882 	u64 data;
1883 	int r;
1884 
1885 	r = kvm_get_msr(vcpu, ecx, &data);
1886 
1887 	/* MSR read failed? See if we should ask user space */
1888 	if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1889 		/* Bounce to user space */
1890 		return 0;
1891 	}
1892 
1893 	if (!r) {
1894 		trace_kvm_msr_read(ecx, data);
1895 
1896 		kvm_rax_write(vcpu, data & -1u);
1897 		kvm_rdx_write(vcpu, (data >> 32) & -1u);
1898 	} else {
1899 		trace_kvm_msr_read_ex(ecx);
1900 	}
1901 
1902 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1903 }
1904 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1905 
1906 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1907 {
1908 	u32 ecx = kvm_rcx_read(vcpu);
1909 	u64 data = kvm_read_edx_eax(vcpu);
1910 	int r;
1911 
1912 	r = kvm_set_msr(vcpu, ecx, data);
1913 
1914 	/* MSR write failed? See if we should ask user space */
1915 	if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1916 		/* Bounce to user space */
1917 		return 0;
1918 
1919 	/* Signal all other negative errors to userspace */
1920 	if (r < 0)
1921 		return r;
1922 
1923 	if (!r)
1924 		trace_kvm_msr_write(ecx, data);
1925 	else
1926 		trace_kvm_msr_write_ex(ecx, data);
1927 
1928 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1929 }
1930 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1931 
1932 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
1933 {
1934 	return kvm_skip_emulated_instruction(vcpu);
1935 }
1936 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
1937 
1938 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
1939 {
1940 	/* Treat an INVD instruction as a NOP and just skip it. */
1941 	return kvm_emulate_as_nop(vcpu);
1942 }
1943 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
1944 
1945 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
1946 {
1947 	pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
1948 	return kvm_emulate_as_nop(vcpu);
1949 }
1950 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
1951 
1952 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
1953 {
1954 	kvm_queue_exception(vcpu, UD_VECTOR);
1955 	return 1;
1956 }
1957 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
1958 
1959 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
1960 {
1961 	pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
1962 	return kvm_emulate_as_nop(vcpu);
1963 }
1964 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
1965 
1966 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1967 {
1968 	xfer_to_guest_mode_prepare();
1969 	return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1970 		xfer_to_guest_mode_work_pending();
1971 }
1972 
1973 /*
1974  * The fast path for frequent and performance sensitive wrmsr emulation,
1975  * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1976  * the latency of virtual IPI by avoiding the expensive bits of transitioning
1977  * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1978  * other cases which must be called after interrupts are enabled on the host.
1979  */
1980 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1981 {
1982 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1983 		return 1;
1984 
1985 	if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1986 		((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1987 		((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1988 		((u32)(data >> 32) != X2APIC_BROADCAST)) {
1989 
1990 		data &= ~(1 << 12);
1991 		kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1992 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1993 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1994 		trace_kvm_apic_write(APIC_ICR, (u32)data);
1995 		return 0;
1996 	}
1997 
1998 	return 1;
1999 }
2000 
2001 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
2002 {
2003 	if (!kvm_can_use_hv_timer(vcpu))
2004 		return 1;
2005 
2006 	kvm_set_lapic_tscdeadline_msr(vcpu, data);
2007 	return 0;
2008 }
2009 
2010 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
2011 {
2012 	u32 msr = kvm_rcx_read(vcpu);
2013 	u64 data;
2014 	fastpath_t ret = EXIT_FASTPATH_NONE;
2015 
2016 	switch (msr) {
2017 	case APIC_BASE_MSR + (APIC_ICR >> 4):
2018 		data = kvm_read_edx_eax(vcpu);
2019 		if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
2020 			kvm_skip_emulated_instruction(vcpu);
2021 			ret = EXIT_FASTPATH_EXIT_HANDLED;
2022 		}
2023 		break;
2024 	case MSR_IA32_TSC_DEADLINE:
2025 		data = kvm_read_edx_eax(vcpu);
2026 		if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
2027 			kvm_skip_emulated_instruction(vcpu);
2028 			ret = EXIT_FASTPATH_REENTER_GUEST;
2029 		}
2030 		break;
2031 	default:
2032 		break;
2033 	}
2034 
2035 	if (ret != EXIT_FASTPATH_NONE)
2036 		trace_kvm_msr_write(msr, data);
2037 
2038 	return ret;
2039 }
2040 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
2041 
2042 /*
2043  * Adapt set_msr() to msr_io()'s calling convention
2044  */
2045 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2046 {
2047 	return kvm_get_msr_ignored_check(vcpu, index, data, true);
2048 }
2049 
2050 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2051 {
2052 	return kvm_set_msr_ignored_check(vcpu, index, *data, true);
2053 }
2054 
2055 #ifdef CONFIG_X86_64
2056 struct pvclock_clock {
2057 	int vclock_mode;
2058 	u64 cycle_last;
2059 	u64 mask;
2060 	u32 mult;
2061 	u32 shift;
2062 	u64 base_cycles;
2063 	u64 offset;
2064 };
2065 
2066 struct pvclock_gtod_data {
2067 	seqcount_t	seq;
2068 
2069 	struct pvclock_clock clock; /* extract of a clocksource struct */
2070 	struct pvclock_clock raw_clock; /* extract of a clocksource struct */
2071 
2072 	ktime_t		offs_boot;
2073 	u64		wall_time_sec;
2074 };
2075 
2076 static struct pvclock_gtod_data pvclock_gtod_data;
2077 
2078 static void update_pvclock_gtod(struct timekeeper *tk)
2079 {
2080 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2081 
2082 	write_seqcount_begin(&vdata->seq);
2083 
2084 	/* copy pvclock gtod data */
2085 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->vdso_clock_mode;
2086 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
2087 	vdata->clock.mask		= tk->tkr_mono.mask;
2088 	vdata->clock.mult		= tk->tkr_mono.mult;
2089 	vdata->clock.shift		= tk->tkr_mono.shift;
2090 	vdata->clock.base_cycles	= tk->tkr_mono.xtime_nsec;
2091 	vdata->clock.offset		= tk->tkr_mono.base;
2092 
2093 	vdata->raw_clock.vclock_mode	= tk->tkr_raw.clock->vdso_clock_mode;
2094 	vdata->raw_clock.cycle_last	= tk->tkr_raw.cycle_last;
2095 	vdata->raw_clock.mask		= tk->tkr_raw.mask;
2096 	vdata->raw_clock.mult		= tk->tkr_raw.mult;
2097 	vdata->raw_clock.shift		= tk->tkr_raw.shift;
2098 	vdata->raw_clock.base_cycles	= tk->tkr_raw.xtime_nsec;
2099 	vdata->raw_clock.offset		= tk->tkr_raw.base;
2100 
2101 	vdata->wall_time_sec            = tk->xtime_sec;
2102 
2103 	vdata->offs_boot		= tk->offs_boot;
2104 
2105 	write_seqcount_end(&vdata->seq);
2106 }
2107 
2108 static s64 get_kvmclock_base_ns(void)
2109 {
2110 	/* Count up from boot time, but with the frequency of the raw clock.  */
2111 	return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2112 }
2113 #else
2114 static s64 get_kvmclock_base_ns(void)
2115 {
2116 	/* Master clock not used, so we can just use CLOCK_BOOTTIME.  */
2117 	return ktime_get_boottime_ns();
2118 }
2119 #endif
2120 
2121 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2122 {
2123 	int version;
2124 	int r;
2125 	struct pvclock_wall_clock wc;
2126 	u32 wc_sec_hi;
2127 	u64 wall_nsec;
2128 
2129 	if (!wall_clock)
2130 		return;
2131 
2132 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2133 	if (r)
2134 		return;
2135 
2136 	if (version & 1)
2137 		++version;  /* first time write, random junk */
2138 
2139 	++version;
2140 
2141 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2142 		return;
2143 
2144 	/*
2145 	 * The guest calculates current wall clock time by adding
2146 	 * system time (updated by kvm_guest_time_update below) to the
2147 	 * wall clock specified here.  We do the reverse here.
2148 	 */
2149 	wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2150 
2151 	wc.nsec = do_div(wall_nsec, 1000000000);
2152 	wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2153 	wc.version = version;
2154 
2155 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2156 
2157 	if (sec_hi_ofs) {
2158 		wc_sec_hi = wall_nsec >> 32;
2159 		kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2160 				&wc_sec_hi, sizeof(wc_sec_hi));
2161 	}
2162 
2163 	version++;
2164 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2165 }
2166 
2167 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2168 				  bool old_msr, bool host_initiated)
2169 {
2170 	struct kvm_arch *ka = &vcpu->kvm->arch;
2171 
2172 	if (vcpu->vcpu_id == 0 && !host_initiated) {
2173 		if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2174 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2175 
2176 		ka->boot_vcpu_runs_old_kvmclock = old_msr;
2177 	}
2178 
2179 	vcpu->arch.time = system_time;
2180 	kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2181 
2182 	/* we verify if the enable bit is set... */
2183 	vcpu->arch.pv_time_enabled = false;
2184 	if (!(system_time & 1))
2185 		return;
2186 
2187 	if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2188 				       &vcpu->arch.pv_time, system_time & ~1ULL,
2189 				       sizeof(struct pvclock_vcpu_time_info)))
2190 		vcpu->arch.pv_time_enabled = true;
2191 
2192 	return;
2193 }
2194 
2195 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2196 {
2197 	do_shl32_div32(dividend, divisor);
2198 	return dividend;
2199 }
2200 
2201 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2202 			       s8 *pshift, u32 *pmultiplier)
2203 {
2204 	uint64_t scaled64;
2205 	int32_t  shift = 0;
2206 	uint64_t tps64;
2207 	uint32_t tps32;
2208 
2209 	tps64 = base_hz;
2210 	scaled64 = scaled_hz;
2211 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2212 		tps64 >>= 1;
2213 		shift--;
2214 	}
2215 
2216 	tps32 = (uint32_t)tps64;
2217 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2218 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2219 			scaled64 >>= 1;
2220 		else
2221 			tps32 <<= 1;
2222 		shift++;
2223 	}
2224 
2225 	*pshift = shift;
2226 	*pmultiplier = div_frac(scaled64, tps32);
2227 }
2228 
2229 #ifdef CONFIG_X86_64
2230 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2231 #endif
2232 
2233 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2234 static unsigned long max_tsc_khz;
2235 
2236 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2237 {
2238 	u64 v = (u64)khz * (1000000 + ppm);
2239 	do_div(v, 1000000);
2240 	return v;
2241 }
2242 
2243 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier);
2244 
2245 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2246 {
2247 	u64 ratio;
2248 
2249 	/* Guest TSC same frequency as host TSC? */
2250 	if (!scale) {
2251 		kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2252 		return 0;
2253 	}
2254 
2255 	/* TSC scaling supported? */
2256 	if (!kvm_has_tsc_control) {
2257 		if (user_tsc_khz > tsc_khz) {
2258 			vcpu->arch.tsc_catchup = 1;
2259 			vcpu->arch.tsc_always_catchup = 1;
2260 			return 0;
2261 		} else {
2262 			pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2263 			return -1;
2264 		}
2265 	}
2266 
2267 	/* TSC scaling required  - calculate ratio */
2268 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2269 				user_tsc_khz, tsc_khz);
2270 
2271 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2272 		pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2273 			            user_tsc_khz);
2274 		return -1;
2275 	}
2276 
2277 	kvm_vcpu_write_tsc_multiplier(vcpu, ratio);
2278 	return 0;
2279 }
2280 
2281 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2282 {
2283 	u32 thresh_lo, thresh_hi;
2284 	int use_scaling = 0;
2285 
2286 	/* tsc_khz can be zero if TSC calibration fails */
2287 	if (user_tsc_khz == 0) {
2288 		/* set tsc_scaling_ratio to a safe value */
2289 		kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2290 		return -1;
2291 	}
2292 
2293 	/* Compute a scale to convert nanoseconds in TSC cycles */
2294 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2295 			   &vcpu->arch.virtual_tsc_shift,
2296 			   &vcpu->arch.virtual_tsc_mult);
2297 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2298 
2299 	/*
2300 	 * Compute the variation in TSC rate which is acceptable
2301 	 * within the range of tolerance and decide if the
2302 	 * rate being applied is within that bounds of the hardware
2303 	 * rate.  If so, no scaling or compensation need be done.
2304 	 */
2305 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2306 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2307 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2308 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2309 		use_scaling = 1;
2310 	}
2311 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2312 }
2313 
2314 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2315 {
2316 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2317 				      vcpu->arch.virtual_tsc_mult,
2318 				      vcpu->arch.virtual_tsc_shift);
2319 	tsc += vcpu->arch.this_tsc_write;
2320 	return tsc;
2321 }
2322 
2323 static inline int gtod_is_based_on_tsc(int mode)
2324 {
2325 	return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2326 }
2327 
2328 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2329 {
2330 #ifdef CONFIG_X86_64
2331 	bool vcpus_matched;
2332 	struct kvm_arch *ka = &vcpu->kvm->arch;
2333 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2334 
2335 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2336 			 atomic_read(&vcpu->kvm->online_vcpus));
2337 
2338 	/*
2339 	 * Once the masterclock is enabled, always perform request in
2340 	 * order to update it.
2341 	 *
2342 	 * In order to enable masterclock, the host clocksource must be TSC
2343 	 * and the vcpus need to have matched TSCs.  When that happens,
2344 	 * perform request to enable masterclock.
2345 	 */
2346 	if (ka->use_master_clock ||
2347 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2348 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2349 
2350 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2351 			    atomic_read(&vcpu->kvm->online_vcpus),
2352 		            ka->use_master_clock, gtod->clock.vclock_mode);
2353 #endif
2354 }
2355 
2356 /*
2357  * Multiply tsc by a fixed point number represented by ratio.
2358  *
2359  * The most significant 64-N bits (mult) of ratio represent the
2360  * integral part of the fixed point number; the remaining N bits
2361  * (frac) represent the fractional part, ie. ratio represents a fixed
2362  * point number (mult + frac * 2^(-N)).
2363  *
2364  * N equals to kvm_tsc_scaling_ratio_frac_bits.
2365  */
2366 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2367 {
2368 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2369 }
2370 
2371 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc, u64 ratio)
2372 {
2373 	u64 _tsc = tsc;
2374 
2375 	if (ratio != kvm_default_tsc_scaling_ratio)
2376 		_tsc = __scale_tsc(ratio, tsc);
2377 
2378 	return _tsc;
2379 }
2380 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2381 
2382 static u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2383 {
2384 	u64 tsc;
2385 
2386 	tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio);
2387 
2388 	return target_tsc - tsc;
2389 }
2390 
2391 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2392 {
2393 	return vcpu->arch.l1_tsc_offset +
2394 		kvm_scale_tsc(vcpu, host_tsc, vcpu->arch.l1_tsc_scaling_ratio);
2395 }
2396 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2397 
2398 u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier)
2399 {
2400 	u64 nested_offset;
2401 
2402 	if (l2_multiplier == kvm_default_tsc_scaling_ratio)
2403 		nested_offset = l1_offset;
2404 	else
2405 		nested_offset = mul_s64_u64_shr((s64) l1_offset, l2_multiplier,
2406 						kvm_tsc_scaling_ratio_frac_bits);
2407 
2408 	nested_offset += l2_offset;
2409 	return nested_offset;
2410 }
2411 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_offset);
2412 
2413 u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier)
2414 {
2415 	if (l2_multiplier != kvm_default_tsc_scaling_ratio)
2416 		return mul_u64_u64_shr(l1_multiplier, l2_multiplier,
2417 				       kvm_tsc_scaling_ratio_frac_bits);
2418 
2419 	return l1_multiplier;
2420 }
2421 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_multiplier);
2422 
2423 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset)
2424 {
2425 	trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2426 				   vcpu->arch.l1_tsc_offset,
2427 				   l1_offset);
2428 
2429 	vcpu->arch.l1_tsc_offset = l1_offset;
2430 
2431 	/*
2432 	 * If we are here because L1 chose not to trap WRMSR to TSC then
2433 	 * according to the spec this should set L1's TSC (as opposed to
2434 	 * setting L1's offset for L2).
2435 	 */
2436 	if (is_guest_mode(vcpu))
2437 		vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2438 			l1_offset,
2439 			static_call(kvm_x86_get_l2_tsc_offset)(vcpu),
2440 			static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2441 	else
2442 		vcpu->arch.tsc_offset = l1_offset;
2443 
2444 	static_call(kvm_x86_write_tsc_offset)(vcpu, vcpu->arch.tsc_offset);
2445 }
2446 
2447 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier)
2448 {
2449 	vcpu->arch.l1_tsc_scaling_ratio = l1_multiplier;
2450 
2451 	/* Userspace is changing the multiplier while L2 is active */
2452 	if (is_guest_mode(vcpu))
2453 		vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2454 			l1_multiplier,
2455 			static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2456 	else
2457 		vcpu->arch.tsc_scaling_ratio = l1_multiplier;
2458 
2459 	if (kvm_has_tsc_control)
2460 		static_call(kvm_x86_write_tsc_multiplier)(
2461 			vcpu, vcpu->arch.tsc_scaling_ratio);
2462 }
2463 
2464 static inline bool kvm_check_tsc_unstable(void)
2465 {
2466 #ifdef CONFIG_X86_64
2467 	/*
2468 	 * TSC is marked unstable when we're running on Hyper-V,
2469 	 * 'TSC page' clocksource is good.
2470 	 */
2471 	if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2472 		return false;
2473 #endif
2474 	return check_tsc_unstable();
2475 }
2476 
2477 /*
2478  * Infers attempts to synchronize the guest's tsc from host writes. Sets the
2479  * offset for the vcpu and tracks the TSC matching generation that the vcpu
2480  * participates in.
2481  */
2482 static void __kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 offset, u64 tsc,
2483 				  u64 ns, bool matched)
2484 {
2485 	struct kvm *kvm = vcpu->kvm;
2486 
2487 	lockdep_assert_held(&kvm->arch.tsc_write_lock);
2488 
2489 	/*
2490 	 * We also track th most recent recorded KHZ, write and time to
2491 	 * allow the matching interval to be extended at each write.
2492 	 */
2493 	kvm->arch.last_tsc_nsec = ns;
2494 	kvm->arch.last_tsc_write = tsc;
2495 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2496 	kvm->arch.last_tsc_offset = offset;
2497 
2498 	vcpu->arch.last_guest_tsc = tsc;
2499 
2500 	kvm_vcpu_write_tsc_offset(vcpu, offset);
2501 
2502 	if (!matched) {
2503 		/*
2504 		 * We split periods of matched TSC writes into generations.
2505 		 * For each generation, we track the original measured
2506 		 * nanosecond time, offset, and write, so if TSCs are in
2507 		 * sync, we can match exact offset, and if not, we can match
2508 		 * exact software computation in compute_guest_tsc()
2509 		 *
2510 		 * These values are tracked in kvm->arch.cur_xxx variables.
2511 		 */
2512 		kvm->arch.cur_tsc_generation++;
2513 		kvm->arch.cur_tsc_nsec = ns;
2514 		kvm->arch.cur_tsc_write = tsc;
2515 		kvm->arch.cur_tsc_offset = offset;
2516 		kvm->arch.nr_vcpus_matched_tsc = 0;
2517 	} else if (vcpu->arch.this_tsc_generation != kvm->arch.cur_tsc_generation) {
2518 		kvm->arch.nr_vcpus_matched_tsc++;
2519 	}
2520 
2521 	/* Keep track of which generation this VCPU has synchronized to */
2522 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2523 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2524 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2525 
2526 	kvm_track_tsc_matching(vcpu);
2527 }
2528 
2529 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2530 {
2531 	struct kvm *kvm = vcpu->kvm;
2532 	u64 offset, ns, elapsed;
2533 	unsigned long flags;
2534 	bool matched = false;
2535 	bool synchronizing = false;
2536 
2537 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2538 	offset = kvm_compute_l1_tsc_offset(vcpu, data);
2539 	ns = get_kvmclock_base_ns();
2540 	elapsed = ns - kvm->arch.last_tsc_nsec;
2541 
2542 	if (vcpu->arch.virtual_tsc_khz) {
2543 		if (data == 0) {
2544 			/*
2545 			 * detection of vcpu initialization -- need to sync
2546 			 * with other vCPUs. This particularly helps to keep
2547 			 * kvm_clock stable after CPU hotplug
2548 			 */
2549 			synchronizing = true;
2550 		} else {
2551 			u64 tsc_exp = kvm->arch.last_tsc_write +
2552 						nsec_to_cycles(vcpu, elapsed);
2553 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2554 			/*
2555 			 * Special case: TSC write with a small delta (1 second)
2556 			 * of virtual cycle time against real time is
2557 			 * interpreted as an attempt to synchronize the CPU.
2558 			 */
2559 			synchronizing = data < tsc_exp + tsc_hz &&
2560 					data + tsc_hz > tsc_exp;
2561 		}
2562 	}
2563 
2564 	/*
2565 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
2566 	 * TSC, we add elapsed time in this computation.  We could let the
2567 	 * compensation code attempt to catch up if we fall behind, but
2568 	 * it's better to try to match offsets from the beginning.
2569          */
2570 	if (synchronizing &&
2571 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2572 		if (!kvm_check_tsc_unstable()) {
2573 			offset = kvm->arch.cur_tsc_offset;
2574 		} else {
2575 			u64 delta = nsec_to_cycles(vcpu, elapsed);
2576 			data += delta;
2577 			offset = kvm_compute_l1_tsc_offset(vcpu, data);
2578 		}
2579 		matched = true;
2580 	}
2581 
2582 	__kvm_synchronize_tsc(vcpu, offset, data, ns, matched);
2583 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2584 }
2585 
2586 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2587 					   s64 adjustment)
2588 {
2589 	u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2590 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2591 }
2592 
2593 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2594 {
2595 	if (vcpu->arch.l1_tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2596 		WARN_ON(adjustment < 0);
2597 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment,
2598 				   vcpu->arch.l1_tsc_scaling_ratio);
2599 	adjust_tsc_offset_guest(vcpu, adjustment);
2600 }
2601 
2602 #ifdef CONFIG_X86_64
2603 
2604 static u64 read_tsc(void)
2605 {
2606 	u64 ret = (u64)rdtsc_ordered();
2607 	u64 last = pvclock_gtod_data.clock.cycle_last;
2608 
2609 	if (likely(ret >= last))
2610 		return ret;
2611 
2612 	/*
2613 	 * GCC likes to generate cmov here, but this branch is extremely
2614 	 * predictable (it's just a function of time and the likely is
2615 	 * very likely) and there's a data dependence, so force GCC
2616 	 * to generate a branch instead.  I don't barrier() because
2617 	 * we don't actually need a barrier, and if this function
2618 	 * ever gets inlined it will generate worse code.
2619 	 */
2620 	asm volatile ("");
2621 	return last;
2622 }
2623 
2624 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2625 			  int *mode)
2626 {
2627 	long v;
2628 	u64 tsc_pg_val;
2629 
2630 	switch (clock->vclock_mode) {
2631 	case VDSO_CLOCKMODE_HVCLOCK:
2632 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2633 						  tsc_timestamp);
2634 		if (tsc_pg_val != U64_MAX) {
2635 			/* TSC page valid */
2636 			*mode = VDSO_CLOCKMODE_HVCLOCK;
2637 			v = (tsc_pg_val - clock->cycle_last) &
2638 				clock->mask;
2639 		} else {
2640 			/* TSC page invalid */
2641 			*mode = VDSO_CLOCKMODE_NONE;
2642 		}
2643 		break;
2644 	case VDSO_CLOCKMODE_TSC:
2645 		*mode = VDSO_CLOCKMODE_TSC;
2646 		*tsc_timestamp = read_tsc();
2647 		v = (*tsc_timestamp - clock->cycle_last) &
2648 			clock->mask;
2649 		break;
2650 	default:
2651 		*mode = VDSO_CLOCKMODE_NONE;
2652 	}
2653 
2654 	if (*mode == VDSO_CLOCKMODE_NONE)
2655 		*tsc_timestamp = v = 0;
2656 
2657 	return v * clock->mult;
2658 }
2659 
2660 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2661 {
2662 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2663 	unsigned long seq;
2664 	int mode;
2665 	u64 ns;
2666 
2667 	do {
2668 		seq = read_seqcount_begin(&gtod->seq);
2669 		ns = gtod->raw_clock.base_cycles;
2670 		ns += vgettsc(&gtod->raw_clock, tsc_timestamp, &mode);
2671 		ns >>= gtod->raw_clock.shift;
2672 		ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2673 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2674 	*t = ns;
2675 
2676 	return mode;
2677 }
2678 
2679 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2680 {
2681 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2682 	unsigned long seq;
2683 	int mode;
2684 	u64 ns;
2685 
2686 	do {
2687 		seq = read_seqcount_begin(&gtod->seq);
2688 		ts->tv_sec = gtod->wall_time_sec;
2689 		ns = gtod->clock.base_cycles;
2690 		ns += vgettsc(&gtod->clock, tsc_timestamp, &mode);
2691 		ns >>= gtod->clock.shift;
2692 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2693 
2694 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2695 	ts->tv_nsec = ns;
2696 
2697 	return mode;
2698 }
2699 
2700 /* returns true if host is using TSC based clocksource */
2701 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2702 {
2703 	/* checked again under seqlock below */
2704 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2705 		return false;
2706 
2707 	return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2708 						      tsc_timestamp));
2709 }
2710 
2711 /* returns true if host is using TSC based clocksource */
2712 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2713 					   u64 *tsc_timestamp)
2714 {
2715 	/* checked again under seqlock below */
2716 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2717 		return false;
2718 
2719 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2720 }
2721 #endif
2722 
2723 /*
2724  *
2725  * Assuming a stable TSC across physical CPUS, and a stable TSC
2726  * across virtual CPUs, the following condition is possible.
2727  * Each numbered line represents an event visible to both
2728  * CPUs at the next numbered event.
2729  *
2730  * "timespecX" represents host monotonic time. "tscX" represents
2731  * RDTSC value.
2732  *
2733  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
2734  *
2735  * 1.  read timespec0,tsc0
2736  * 2.					| timespec1 = timespec0 + N
2737  * 					| tsc1 = tsc0 + M
2738  * 3. transition to guest		| transition to guest
2739  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2740  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
2741  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2742  *
2743  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2744  *
2745  * 	- ret0 < ret1
2746  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2747  *		...
2748  *	- 0 < N - M => M < N
2749  *
2750  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2751  * always the case (the difference between two distinct xtime instances
2752  * might be smaller then the difference between corresponding TSC reads,
2753  * when updating guest vcpus pvclock areas).
2754  *
2755  * To avoid that problem, do not allow visibility of distinct
2756  * system_timestamp/tsc_timestamp values simultaneously: use a master
2757  * copy of host monotonic time values. Update that master copy
2758  * in lockstep.
2759  *
2760  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2761  *
2762  */
2763 
2764 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2765 {
2766 #ifdef CONFIG_X86_64
2767 	struct kvm_arch *ka = &kvm->arch;
2768 	int vclock_mode;
2769 	bool host_tsc_clocksource, vcpus_matched;
2770 
2771 	lockdep_assert_held(&kvm->arch.tsc_write_lock);
2772 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2773 			atomic_read(&kvm->online_vcpus));
2774 
2775 	/*
2776 	 * If the host uses TSC clock, then passthrough TSC as stable
2777 	 * to the guest.
2778 	 */
2779 	host_tsc_clocksource = kvm_get_time_and_clockread(
2780 					&ka->master_kernel_ns,
2781 					&ka->master_cycle_now);
2782 
2783 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2784 				&& !ka->backwards_tsc_observed
2785 				&& !ka->boot_vcpu_runs_old_kvmclock;
2786 
2787 	if (ka->use_master_clock)
2788 		atomic_set(&kvm_guest_has_master_clock, 1);
2789 
2790 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2791 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2792 					vcpus_matched);
2793 #endif
2794 }
2795 
2796 static void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2797 {
2798 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2799 }
2800 
2801 static void __kvm_start_pvclock_update(struct kvm *kvm)
2802 {
2803 	raw_spin_lock_irq(&kvm->arch.tsc_write_lock);
2804 	write_seqcount_begin(&kvm->arch.pvclock_sc);
2805 }
2806 
2807 static void kvm_start_pvclock_update(struct kvm *kvm)
2808 {
2809 	kvm_make_mclock_inprogress_request(kvm);
2810 
2811 	/* no guest entries from this point */
2812 	__kvm_start_pvclock_update(kvm);
2813 }
2814 
2815 static void kvm_end_pvclock_update(struct kvm *kvm)
2816 {
2817 	struct kvm_arch *ka = &kvm->arch;
2818 	struct kvm_vcpu *vcpu;
2819 	int i;
2820 
2821 	write_seqcount_end(&ka->pvclock_sc);
2822 	raw_spin_unlock_irq(&ka->tsc_write_lock);
2823 	kvm_for_each_vcpu(i, vcpu, kvm)
2824 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2825 
2826 	/* guest entries allowed */
2827 	kvm_for_each_vcpu(i, vcpu, kvm)
2828 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2829 }
2830 
2831 static void kvm_update_masterclock(struct kvm *kvm)
2832 {
2833 	kvm_hv_invalidate_tsc_page(kvm);
2834 	kvm_start_pvclock_update(kvm);
2835 	pvclock_update_vm_gtod_copy(kvm);
2836 	kvm_end_pvclock_update(kvm);
2837 }
2838 
2839 /* Called within read_seqcount_begin/retry for kvm->pvclock_sc.  */
2840 static void __get_kvmclock(struct kvm *kvm, struct kvm_clock_data *data)
2841 {
2842 	struct kvm_arch *ka = &kvm->arch;
2843 	struct pvclock_vcpu_time_info hv_clock;
2844 
2845 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2846 	get_cpu();
2847 
2848 	data->flags = 0;
2849 	if (ka->use_master_clock && __this_cpu_read(cpu_tsc_khz)) {
2850 #ifdef CONFIG_X86_64
2851 		struct timespec64 ts;
2852 
2853 		if (kvm_get_walltime_and_clockread(&ts, &data->host_tsc)) {
2854 			data->realtime = ts.tv_nsec + NSEC_PER_SEC * ts.tv_sec;
2855 			data->flags |= KVM_CLOCK_REALTIME | KVM_CLOCK_HOST_TSC;
2856 		} else
2857 #endif
2858 		data->host_tsc = rdtsc();
2859 
2860 		data->flags |= KVM_CLOCK_TSC_STABLE;
2861 		hv_clock.tsc_timestamp = ka->master_cycle_now;
2862 		hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2863 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2864 				   &hv_clock.tsc_shift,
2865 				   &hv_clock.tsc_to_system_mul);
2866 		data->clock = __pvclock_read_cycles(&hv_clock, data->host_tsc);
2867 	} else {
2868 		data->clock = get_kvmclock_base_ns() + ka->kvmclock_offset;
2869 	}
2870 
2871 	put_cpu();
2872 }
2873 
2874 static void get_kvmclock(struct kvm *kvm, struct kvm_clock_data *data)
2875 {
2876 	struct kvm_arch *ka = &kvm->arch;
2877 	unsigned seq;
2878 
2879 	do {
2880 		seq = read_seqcount_begin(&ka->pvclock_sc);
2881 		__get_kvmclock(kvm, data);
2882 	} while (read_seqcount_retry(&ka->pvclock_sc, seq));
2883 }
2884 
2885 u64 get_kvmclock_ns(struct kvm *kvm)
2886 {
2887 	struct kvm_clock_data data;
2888 
2889 	get_kvmclock(kvm, &data);
2890 	return data.clock;
2891 }
2892 
2893 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2894 				   struct gfn_to_hva_cache *cache,
2895 				   unsigned int offset)
2896 {
2897 	struct kvm_vcpu_arch *vcpu = &v->arch;
2898 	struct pvclock_vcpu_time_info guest_hv_clock;
2899 
2900 	if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2901 		&guest_hv_clock, offset, sizeof(guest_hv_clock))))
2902 		return;
2903 
2904 	/* This VCPU is paused, but it's legal for a guest to read another
2905 	 * VCPU's kvmclock, so we really have to follow the specification where
2906 	 * it says that version is odd if data is being modified, and even after
2907 	 * it is consistent.
2908 	 *
2909 	 * Version field updates must be kept separate.  This is because
2910 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2911 	 * writes within a string instruction are weakly ordered.  So there
2912 	 * are three writes overall.
2913 	 *
2914 	 * As a small optimization, only write the version field in the first
2915 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2916 	 * version field is the first in the struct.
2917 	 */
2918 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2919 
2920 	if (guest_hv_clock.version & 1)
2921 		++guest_hv_clock.version;  /* first time write, random junk */
2922 
2923 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2924 	kvm_write_guest_offset_cached(v->kvm, cache,
2925 				      &vcpu->hv_clock, offset,
2926 				      sizeof(vcpu->hv_clock.version));
2927 
2928 	smp_wmb();
2929 
2930 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2931 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2932 
2933 	if (vcpu->pvclock_set_guest_stopped_request) {
2934 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2935 		vcpu->pvclock_set_guest_stopped_request = false;
2936 	}
2937 
2938 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2939 
2940 	kvm_write_guest_offset_cached(v->kvm, cache,
2941 				      &vcpu->hv_clock, offset,
2942 				      sizeof(vcpu->hv_clock));
2943 
2944 	smp_wmb();
2945 
2946 	vcpu->hv_clock.version++;
2947 	kvm_write_guest_offset_cached(v->kvm, cache,
2948 				     &vcpu->hv_clock, offset,
2949 				     sizeof(vcpu->hv_clock.version));
2950 }
2951 
2952 static int kvm_guest_time_update(struct kvm_vcpu *v)
2953 {
2954 	unsigned long flags, tgt_tsc_khz;
2955 	unsigned seq;
2956 	struct kvm_vcpu_arch *vcpu = &v->arch;
2957 	struct kvm_arch *ka = &v->kvm->arch;
2958 	s64 kernel_ns;
2959 	u64 tsc_timestamp, host_tsc;
2960 	u8 pvclock_flags;
2961 	bool use_master_clock;
2962 
2963 	kernel_ns = 0;
2964 	host_tsc = 0;
2965 
2966 	/*
2967 	 * If the host uses TSC clock, then passthrough TSC as stable
2968 	 * to the guest.
2969 	 */
2970 	do {
2971 		seq = read_seqcount_begin(&ka->pvclock_sc);
2972 		use_master_clock = ka->use_master_clock;
2973 		if (use_master_clock) {
2974 			host_tsc = ka->master_cycle_now;
2975 			kernel_ns = ka->master_kernel_ns;
2976 		}
2977 	} while (read_seqcount_retry(&ka->pvclock_sc, seq));
2978 
2979 	/* Keep irq disabled to prevent changes to the clock */
2980 	local_irq_save(flags);
2981 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2982 	if (unlikely(tgt_tsc_khz == 0)) {
2983 		local_irq_restore(flags);
2984 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2985 		return 1;
2986 	}
2987 	if (!use_master_clock) {
2988 		host_tsc = rdtsc();
2989 		kernel_ns = get_kvmclock_base_ns();
2990 	}
2991 
2992 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2993 
2994 	/*
2995 	 * We may have to catch up the TSC to match elapsed wall clock
2996 	 * time for two reasons, even if kvmclock is used.
2997 	 *   1) CPU could have been running below the maximum TSC rate
2998 	 *   2) Broken TSC compensation resets the base at each VCPU
2999 	 *      entry to avoid unknown leaps of TSC even when running
3000 	 *      again on the same CPU.  This may cause apparent elapsed
3001 	 *      time to disappear, and the guest to stand still or run
3002 	 *	very slowly.
3003 	 */
3004 	if (vcpu->tsc_catchup) {
3005 		u64 tsc = compute_guest_tsc(v, kernel_ns);
3006 		if (tsc > tsc_timestamp) {
3007 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
3008 			tsc_timestamp = tsc;
3009 		}
3010 	}
3011 
3012 	local_irq_restore(flags);
3013 
3014 	/* With all the info we got, fill in the values */
3015 
3016 	if (kvm_has_tsc_control)
3017 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz,
3018 					    v->arch.l1_tsc_scaling_ratio);
3019 
3020 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
3021 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
3022 				   &vcpu->hv_clock.tsc_shift,
3023 				   &vcpu->hv_clock.tsc_to_system_mul);
3024 		vcpu->hw_tsc_khz = tgt_tsc_khz;
3025 	}
3026 
3027 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
3028 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
3029 	vcpu->last_guest_tsc = tsc_timestamp;
3030 
3031 	/* If the host uses TSC clocksource, then it is stable */
3032 	pvclock_flags = 0;
3033 	if (use_master_clock)
3034 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
3035 
3036 	vcpu->hv_clock.flags = pvclock_flags;
3037 
3038 	if (vcpu->pv_time_enabled)
3039 		kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
3040 	if (vcpu->xen.vcpu_info_set)
3041 		kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
3042 				       offsetof(struct compat_vcpu_info, time));
3043 	if (vcpu->xen.vcpu_time_info_set)
3044 		kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
3045 	if (!v->vcpu_idx)
3046 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
3047 	return 0;
3048 }
3049 
3050 /*
3051  * kvmclock updates which are isolated to a given vcpu, such as
3052  * vcpu->cpu migration, should not allow system_timestamp from
3053  * the rest of the vcpus to remain static. Otherwise ntp frequency
3054  * correction applies to one vcpu's system_timestamp but not
3055  * the others.
3056  *
3057  * So in those cases, request a kvmclock update for all vcpus.
3058  * We need to rate-limit these requests though, as they can
3059  * considerably slow guests that have a large number of vcpus.
3060  * The time for a remote vcpu to update its kvmclock is bound
3061  * by the delay we use to rate-limit the updates.
3062  */
3063 
3064 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
3065 
3066 static void kvmclock_update_fn(struct work_struct *work)
3067 {
3068 	int i;
3069 	struct delayed_work *dwork = to_delayed_work(work);
3070 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3071 					   kvmclock_update_work);
3072 	struct kvm *kvm = container_of(ka, struct kvm, arch);
3073 	struct kvm_vcpu *vcpu;
3074 
3075 	kvm_for_each_vcpu(i, vcpu, kvm) {
3076 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3077 		kvm_vcpu_kick(vcpu);
3078 	}
3079 }
3080 
3081 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
3082 {
3083 	struct kvm *kvm = v->kvm;
3084 
3085 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3086 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
3087 					KVMCLOCK_UPDATE_DELAY);
3088 }
3089 
3090 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
3091 
3092 static void kvmclock_sync_fn(struct work_struct *work)
3093 {
3094 	struct delayed_work *dwork = to_delayed_work(work);
3095 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3096 					   kvmclock_sync_work);
3097 	struct kvm *kvm = container_of(ka, struct kvm, arch);
3098 
3099 	if (!kvmclock_periodic_sync)
3100 		return;
3101 
3102 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
3103 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
3104 					KVMCLOCK_SYNC_PERIOD);
3105 }
3106 
3107 /*
3108  * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
3109  */
3110 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
3111 {
3112 	/* McStatusWrEn enabled? */
3113 	if (guest_cpuid_is_amd_or_hygon(vcpu))
3114 		return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
3115 
3116 	return false;
3117 }
3118 
3119 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3120 {
3121 	u64 mcg_cap = vcpu->arch.mcg_cap;
3122 	unsigned bank_num = mcg_cap & 0xff;
3123 	u32 msr = msr_info->index;
3124 	u64 data = msr_info->data;
3125 
3126 	switch (msr) {
3127 	case MSR_IA32_MCG_STATUS:
3128 		vcpu->arch.mcg_status = data;
3129 		break;
3130 	case MSR_IA32_MCG_CTL:
3131 		if (!(mcg_cap & MCG_CTL_P) &&
3132 		    (data || !msr_info->host_initiated))
3133 			return 1;
3134 		if (data != 0 && data != ~(u64)0)
3135 			return 1;
3136 		vcpu->arch.mcg_ctl = data;
3137 		break;
3138 	default:
3139 		if (msr >= MSR_IA32_MC0_CTL &&
3140 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
3141 			u32 offset = array_index_nospec(
3142 				msr - MSR_IA32_MC0_CTL,
3143 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3144 
3145 			/* only 0 or all 1s can be written to IA32_MCi_CTL
3146 			 * some Linux kernels though clear bit 10 in bank 4 to
3147 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
3148 			 * this to avoid an uncatched #GP in the guest
3149 			 */
3150 			if ((offset & 0x3) == 0 &&
3151 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
3152 				return -1;
3153 
3154 			/* MCi_STATUS */
3155 			if (!msr_info->host_initiated &&
3156 			    (offset & 0x3) == 1 && data != 0) {
3157 				if (!can_set_mci_status(vcpu))
3158 					return -1;
3159 			}
3160 
3161 			vcpu->arch.mce_banks[offset] = data;
3162 			break;
3163 		}
3164 		return 1;
3165 	}
3166 	return 0;
3167 }
3168 
3169 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
3170 {
3171 	u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3172 
3173 	return (vcpu->arch.apf.msr_en_val & mask) == mask;
3174 }
3175 
3176 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3177 {
3178 	gpa_t gpa = data & ~0x3f;
3179 
3180 	/* Bits 4:5 are reserved, Should be zero */
3181 	if (data & 0x30)
3182 		return 1;
3183 
3184 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3185 	    (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3186 		return 1;
3187 
3188 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3189 	    (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3190 		return 1;
3191 
3192 	if (!lapic_in_kernel(vcpu))
3193 		return data ? 1 : 0;
3194 
3195 	vcpu->arch.apf.msr_en_val = data;
3196 
3197 	if (!kvm_pv_async_pf_enabled(vcpu)) {
3198 		kvm_clear_async_pf_completion_queue(vcpu);
3199 		kvm_async_pf_hash_reset(vcpu);
3200 		return 0;
3201 	}
3202 
3203 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3204 					sizeof(u64)))
3205 		return 1;
3206 
3207 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3208 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3209 
3210 	kvm_async_pf_wakeup_all(vcpu);
3211 
3212 	return 0;
3213 }
3214 
3215 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3216 {
3217 	/* Bits 8-63 are reserved */
3218 	if (data >> 8)
3219 		return 1;
3220 
3221 	if (!lapic_in_kernel(vcpu))
3222 		return 1;
3223 
3224 	vcpu->arch.apf.msr_int_val = data;
3225 
3226 	vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3227 
3228 	return 0;
3229 }
3230 
3231 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3232 {
3233 	vcpu->arch.pv_time_enabled = false;
3234 	vcpu->arch.time = 0;
3235 }
3236 
3237 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3238 {
3239 	++vcpu->stat.tlb_flush;
3240 	static_call(kvm_x86_tlb_flush_all)(vcpu);
3241 }
3242 
3243 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3244 {
3245 	++vcpu->stat.tlb_flush;
3246 
3247 	if (!tdp_enabled) {
3248 		/*
3249 		 * A TLB flush on behalf of the guest is equivalent to
3250 		 * INVPCID(all), toggling CR4.PGE, etc., which requires
3251 		 * a forced sync of the shadow page tables.  Ensure all the
3252 		 * roots are synced and the guest TLB in hardware is clean.
3253 		 */
3254 		kvm_mmu_sync_roots(vcpu);
3255 		kvm_mmu_sync_prev_roots(vcpu);
3256 	}
3257 
3258 	static_call(kvm_x86_tlb_flush_guest)(vcpu);
3259 }
3260 
3261 
3262 static inline void kvm_vcpu_flush_tlb_current(struct kvm_vcpu *vcpu)
3263 {
3264 	++vcpu->stat.tlb_flush;
3265 	static_call(kvm_x86_tlb_flush_current)(vcpu);
3266 }
3267 
3268 /*
3269  * Service "local" TLB flush requests, which are specific to the current MMU
3270  * context.  In addition to the generic event handling in vcpu_enter_guest(),
3271  * TLB flushes that are targeted at an MMU context also need to be serviced
3272  * prior before nested VM-Enter/VM-Exit.
3273  */
3274 void kvm_service_local_tlb_flush_requests(struct kvm_vcpu *vcpu)
3275 {
3276 	if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
3277 		kvm_vcpu_flush_tlb_current(vcpu);
3278 
3279 	if (kvm_check_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu))
3280 		kvm_vcpu_flush_tlb_guest(vcpu);
3281 }
3282 EXPORT_SYMBOL_GPL(kvm_service_local_tlb_flush_requests);
3283 
3284 static void record_steal_time(struct kvm_vcpu *vcpu)
3285 {
3286 	struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
3287 	struct kvm_steal_time __user *st;
3288 	struct kvm_memslots *slots;
3289 	u64 steal;
3290 	u32 version;
3291 
3292 	if (kvm_xen_msr_enabled(vcpu->kvm)) {
3293 		kvm_xen_runstate_set_running(vcpu);
3294 		return;
3295 	}
3296 
3297 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3298 		return;
3299 
3300 	if (WARN_ON_ONCE(current->mm != vcpu->kvm->mm))
3301 		return;
3302 
3303 	slots = kvm_memslots(vcpu->kvm);
3304 
3305 	if (unlikely(slots->generation != ghc->generation ||
3306 		     kvm_is_error_hva(ghc->hva) || !ghc->memslot)) {
3307 		gfn_t gfn = vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS;
3308 
3309 		/* We rely on the fact that it fits in a single page. */
3310 		BUILD_BUG_ON((sizeof(*st) - 1) & KVM_STEAL_VALID_BITS);
3311 
3312 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, gfn, sizeof(*st)) ||
3313 		    kvm_is_error_hva(ghc->hva) || !ghc->memslot)
3314 			return;
3315 	}
3316 
3317 	st = (struct kvm_steal_time __user *)ghc->hva;
3318 	/*
3319 	 * Doing a TLB flush here, on the guest's behalf, can avoid
3320 	 * expensive IPIs.
3321 	 */
3322 	if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3323 		u8 st_preempted = 0;
3324 		int err = -EFAULT;
3325 
3326 		if (!user_access_begin(st, sizeof(*st)))
3327 			return;
3328 
3329 		asm volatile("1: xchgb %0, %2\n"
3330 			     "xor %1, %1\n"
3331 			     "2:\n"
3332 			     _ASM_EXTABLE_UA(1b, 2b)
3333 			     : "+q" (st_preempted),
3334 			       "+&r" (err),
3335 			       "+m" (st->preempted));
3336 		if (err)
3337 			goto out;
3338 
3339 		user_access_end();
3340 
3341 		vcpu->arch.st.preempted = 0;
3342 
3343 		trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3344 				       st_preempted & KVM_VCPU_FLUSH_TLB);
3345 		if (st_preempted & KVM_VCPU_FLUSH_TLB)
3346 			kvm_vcpu_flush_tlb_guest(vcpu);
3347 
3348 		if (!user_access_begin(st, sizeof(*st)))
3349 			goto dirty;
3350 	} else {
3351 		if (!user_access_begin(st, sizeof(*st)))
3352 			return;
3353 
3354 		unsafe_put_user(0, &st->preempted, out);
3355 		vcpu->arch.st.preempted = 0;
3356 	}
3357 
3358 	unsafe_get_user(version, &st->version, out);
3359 	if (version & 1)
3360 		version += 1;  /* first time write, random junk */
3361 
3362 	version += 1;
3363 	unsafe_put_user(version, &st->version, out);
3364 
3365 	smp_wmb();
3366 
3367 	unsafe_get_user(steal, &st->steal, out);
3368 	steal += current->sched_info.run_delay -
3369 		vcpu->arch.st.last_steal;
3370 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
3371 	unsafe_put_user(steal, &st->steal, out);
3372 
3373 	version += 1;
3374 	unsafe_put_user(version, &st->version, out);
3375 
3376  out:
3377 	user_access_end();
3378  dirty:
3379 	mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
3380 }
3381 
3382 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3383 {
3384 	bool pr = false;
3385 	u32 msr = msr_info->index;
3386 	u64 data = msr_info->data;
3387 
3388 	if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3389 		return kvm_xen_write_hypercall_page(vcpu, data);
3390 
3391 	switch (msr) {
3392 	case MSR_AMD64_NB_CFG:
3393 	case MSR_IA32_UCODE_WRITE:
3394 	case MSR_VM_HSAVE_PA:
3395 	case MSR_AMD64_PATCH_LOADER:
3396 	case MSR_AMD64_BU_CFG2:
3397 	case MSR_AMD64_DC_CFG:
3398 	case MSR_F15H_EX_CFG:
3399 		break;
3400 
3401 	case MSR_IA32_UCODE_REV:
3402 		if (msr_info->host_initiated)
3403 			vcpu->arch.microcode_version = data;
3404 		break;
3405 	case MSR_IA32_ARCH_CAPABILITIES:
3406 		if (!msr_info->host_initiated)
3407 			return 1;
3408 		vcpu->arch.arch_capabilities = data;
3409 		break;
3410 	case MSR_IA32_PERF_CAPABILITIES: {
3411 		struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3412 
3413 		if (!msr_info->host_initiated)
3414 			return 1;
3415 		if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3416 			return 1;
3417 		if (data & ~msr_ent.data)
3418 			return 1;
3419 
3420 		vcpu->arch.perf_capabilities = data;
3421 
3422 		return 0;
3423 		}
3424 	case MSR_EFER:
3425 		return set_efer(vcpu, msr_info);
3426 	case MSR_K7_HWCR:
3427 		data &= ~(u64)0x40;	/* ignore flush filter disable */
3428 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
3429 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
3430 
3431 		/* Handle McStatusWrEn */
3432 		if (data == BIT_ULL(18)) {
3433 			vcpu->arch.msr_hwcr = data;
3434 		} else if (data != 0) {
3435 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3436 				    data);
3437 			return 1;
3438 		}
3439 		break;
3440 	case MSR_FAM10H_MMIO_CONF_BASE:
3441 		if (data != 0) {
3442 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3443 				    "0x%llx\n", data);
3444 			return 1;
3445 		}
3446 		break;
3447 	case 0x200 ... 0x2ff:
3448 		return kvm_mtrr_set_msr(vcpu, msr, data);
3449 	case MSR_IA32_APICBASE:
3450 		return kvm_set_apic_base(vcpu, msr_info);
3451 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3452 		return kvm_x2apic_msr_write(vcpu, msr, data);
3453 	case MSR_IA32_TSC_DEADLINE:
3454 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
3455 		break;
3456 	case MSR_IA32_TSC_ADJUST:
3457 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3458 			if (!msr_info->host_initiated) {
3459 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3460 				adjust_tsc_offset_guest(vcpu, adj);
3461 				/* Before back to guest, tsc_timestamp must be adjusted
3462 				 * as well, otherwise guest's percpu pvclock time could jump.
3463 				 */
3464 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3465 			}
3466 			vcpu->arch.ia32_tsc_adjust_msr = data;
3467 		}
3468 		break;
3469 	case MSR_IA32_MISC_ENABLE:
3470 		if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3471 		    ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3472 			if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3473 				return 1;
3474 			vcpu->arch.ia32_misc_enable_msr = data;
3475 			kvm_update_cpuid_runtime(vcpu);
3476 		} else {
3477 			vcpu->arch.ia32_misc_enable_msr = data;
3478 		}
3479 		break;
3480 	case MSR_IA32_SMBASE:
3481 		if (!msr_info->host_initiated)
3482 			return 1;
3483 		vcpu->arch.smbase = data;
3484 		break;
3485 	case MSR_IA32_POWER_CTL:
3486 		vcpu->arch.msr_ia32_power_ctl = data;
3487 		break;
3488 	case MSR_IA32_TSC:
3489 		if (msr_info->host_initiated) {
3490 			kvm_synchronize_tsc(vcpu, data);
3491 		} else {
3492 			u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3493 			adjust_tsc_offset_guest(vcpu, adj);
3494 			vcpu->arch.ia32_tsc_adjust_msr += adj;
3495 		}
3496 		break;
3497 	case MSR_IA32_XSS:
3498 		if (!msr_info->host_initiated &&
3499 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3500 			return 1;
3501 		/*
3502 		 * KVM supports exposing PT to the guest, but does not support
3503 		 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3504 		 * XSAVES/XRSTORS to save/restore PT MSRs.
3505 		 */
3506 		if (data & ~supported_xss)
3507 			return 1;
3508 		vcpu->arch.ia32_xss = data;
3509 		break;
3510 	case MSR_SMI_COUNT:
3511 		if (!msr_info->host_initiated)
3512 			return 1;
3513 		vcpu->arch.smi_count = data;
3514 		break;
3515 	case MSR_KVM_WALL_CLOCK_NEW:
3516 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3517 			return 1;
3518 
3519 		vcpu->kvm->arch.wall_clock = data;
3520 		kvm_write_wall_clock(vcpu->kvm, data, 0);
3521 		break;
3522 	case MSR_KVM_WALL_CLOCK:
3523 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3524 			return 1;
3525 
3526 		vcpu->kvm->arch.wall_clock = data;
3527 		kvm_write_wall_clock(vcpu->kvm, data, 0);
3528 		break;
3529 	case MSR_KVM_SYSTEM_TIME_NEW:
3530 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3531 			return 1;
3532 
3533 		kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3534 		break;
3535 	case MSR_KVM_SYSTEM_TIME:
3536 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3537 			return 1;
3538 
3539 		kvm_write_system_time(vcpu, data, true,  msr_info->host_initiated);
3540 		break;
3541 	case MSR_KVM_ASYNC_PF_EN:
3542 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3543 			return 1;
3544 
3545 		if (kvm_pv_enable_async_pf(vcpu, data))
3546 			return 1;
3547 		break;
3548 	case MSR_KVM_ASYNC_PF_INT:
3549 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3550 			return 1;
3551 
3552 		if (kvm_pv_enable_async_pf_int(vcpu, data))
3553 			return 1;
3554 		break;
3555 	case MSR_KVM_ASYNC_PF_ACK:
3556 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3557 			return 1;
3558 		if (data & 0x1) {
3559 			vcpu->arch.apf.pageready_pending = false;
3560 			kvm_check_async_pf_completion(vcpu);
3561 		}
3562 		break;
3563 	case MSR_KVM_STEAL_TIME:
3564 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3565 			return 1;
3566 
3567 		if (unlikely(!sched_info_on()))
3568 			return 1;
3569 
3570 		if (data & KVM_STEAL_RESERVED_MASK)
3571 			return 1;
3572 
3573 		vcpu->arch.st.msr_val = data;
3574 
3575 		if (!(data & KVM_MSR_ENABLED))
3576 			break;
3577 
3578 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3579 
3580 		break;
3581 	case MSR_KVM_PV_EOI_EN:
3582 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3583 			return 1;
3584 
3585 		if (kvm_lapic_set_pv_eoi(vcpu, data, sizeof(u8)))
3586 			return 1;
3587 		break;
3588 
3589 	case MSR_KVM_POLL_CONTROL:
3590 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3591 			return 1;
3592 
3593 		/* only enable bit supported */
3594 		if (data & (-1ULL << 1))
3595 			return 1;
3596 
3597 		vcpu->arch.msr_kvm_poll_control = data;
3598 		break;
3599 
3600 	case MSR_IA32_MCG_CTL:
3601 	case MSR_IA32_MCG_STATUS:
3602 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3603 		return set_msr_mce(vcpu, msr_info);
3604 
3605 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3606 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3607 		pr = true;
3608 		fallthrough;
3609 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3610 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3611 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3612 			return kvm_pmu_set_msr(vcpu, msr_info);
3613 
3614 		if (pr || data != 0)
3615 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3616 				    "0x%x data 0x%llx\n", msr, data);
3617 		break;
3618 	case MSR_K7_CLK_CTL:
3619 		/*
3620 		 * Ignore all writes to this no longer documented MSR.
3621 		 * Writes are only relevant for old K7 processors,
3622 		 * all pre-dating SVM, but a recommended workaround from
3623 		 * AMD for these chips. It is possible to specify the
3624 		 * affected processor models on the command line, hence
3625 		 * the need to ignore the workaround.
3626 		 */
3627 		break;
3628 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3629 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3630 	case HV_X64_MSR_SYNDBG_OPTIONS:
3631 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3632 	case HV_X64_MSR_CRASH_CTL:
3633 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3634 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3635 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3636 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3637 		return kvm_hv_set_msr_common(vcpu, msr, data,
3638 					     msr_info->host_initiated);
3639 	case MSR_IA32_BBL_CR_CTL3:
3640 		/* Drop writes to this legacy MSR -- see rdmsr
3641 		 * counterpart for further detail.
3642 		 */
3643 		if (report_ignored_msrs)
3644 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3645 				msr, data);
3646 		break;
3647 	case MSR_AMD64_OSVW_ID_LENGTH:
3648 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3649 			return 1;
3650 		vcpu->arch.osvw.length = data;
3651 		break;
3652 	case MSR_AMD64_OSVW_STATUS:
3653 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3654 			return 1;
3655 		vcpu->arch.osvw.status = data;
3656 		break;
3657 	case MSR_PLATFORM_INFO:
3658 		if (!msr_info->host_initiated ||
3659 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3660 		     cpuid_fault_enabled(vcpu)))
3661 			return 1;
3662 		vcpu->arch.msr_platform_info = data;
3663 		break;
3664 	case MSR_MISC_FEATURES_ENABLES:
3665 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3666 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3667 		     !supports_cpuid_fault(vcpu)))
3668 			return 1;
3669 		vcpu->arch.msr_misc_features_enables = data;
3670 		break;
3671 	default:
3672 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3673 			return kvm_pmu_set_msr(vcpu, msr_info);
3674 		return KVM_MSR_RET_INVALID;
3675 	}
3676 	return 0;
3677 }
3678 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3679 
3680 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3681 {
3682 	u64 data;
3683 	u64 mcg_cap = vcpu->arch.mcg_cap;
3684 	unsigned bank_num = mcg_cap & 0xff;
3685 
3686 	switch (msr) {
3687 	case MSR_IA32_P5_MC_ADDR:
3688 	case MSR_IA32_P5_MC_TYPE:
3689 		data = 0;
3690 		break;
3691 	case MSR_IA32_MCG_CAP:
3692 		data = vcpu->arch.mcg_cap;
3693 		break;
3694 	case MSR_IA32_MCG_CTL:
3695 		if (!(mcg_cap & MCG_CTL_P) && !host)
3696 			return 1;
3697 		data = vcpu->arch.mcg_ctl;
3698 		break;
3699 	case MSR_IA32_MCG_STATUS:
3700 		data = vcpu->arch.mcg_status;
3701 		break;
3702 	default:
3703 		if (msr >= MSR_IA32_MC0_CTL &&
3704 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
3705 			u32 offset = array_index_nospec(
3706 				msr - MSR_IA32_MC0_CTL,
3707 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3708 
3709 			data = vcpu->arch.mce_banks[offset];
3710 			break;
3711 		}
3712 		return 1;
3713 	}
3714 	*pdata = data;
3715 	return 0;
3716 }
3717 
3718 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3719 {
3720 	switch (msr_info->index) {
3721 	case MSR_IA32_PLATFORM_ID:
3722 	case MSR_IA32_EBL_CR_POWERON:
3723 	case MSR_IA32_LASTBRANCHFROMIP:
3724 	case MSR_IA32_LASTBRANCHTOIP:
3725 	case MSR_IA32_LASTINTFROMIP:
3726 	case MSR_IA32_LASTINTTOIP:
3727 	case MSR_AMD64_SYSCFG:
3728 	case MSR_K8_TSEG_ADDR:
3729 	case MSR_K8_TSEG_MASK:
3730 	case MSR_VM_HSAVE_PA:
3731 	case MSR_K8_INT_PENDING_MSG:
3732 	case MSR_AMD64_NB_CFG:
3733 	case MSR_FAM10H_MMIO_CONF_BASE:
3734 	case MSR_AMD64_BU_CFG2:
3735 	case MSR_IA32_PERF_CTL:
3736 	case MSR_AMD64_DC_CFG:
3737 	case MSR_F15H_EX_CFG:
3738 	/*
3739 	 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3740 	 * limit) MSRs. Just return 0, as we do not want to expose the host
3741 	 * data here. Do not conditionalize this on CPUID, as KVM does not do
3742 	 * so for existing CPU-specific MSRs.
3743 	 */
3744 	case MSR_RAPL_POWER_UNIT:
3745 	case MSR_PP0_ENERGY_STATUS:	/* Power plane 0 (core) */
3746 	case MSR_PP1_ENERGY_STATUS:	/* Power plane 1 (graphics uncore) */
3747 	case MSR_PKG_ENERGY_STATUS:	/* Total package */
3748 	case MSR_DRAM_ENERGY_STATUS:	/* DRAM controller */
3749 		msr_info->data = 0;
3750 		break;
3751 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3752 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3753 			return kvm_pmu_get_msr(vcpu, msr_info);
3754 		if (!msr_info->host_initiated)
3755 			return 1;
3756 		msr_info->data = 0;
3757 		break;
3758 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3759 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3760 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3761 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3762 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3763 			return kvm_pmu_get_msr(vcpu, msr_info);
3764 		msr_info->data = 0;
3765 		break;
3766 	case MSR_IA32_UCODE_REV:
3767 		msr_info->data = vcpu->arch.microcode_version;
3768 		break;
3769 	case MSR_IA32_ARCH_CAPABILITIES:
3770 		if (!msr_info->host_initiated &&
3771 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3772 			return 1;
3773 		msr_info->data = vcpu->arch.arch_capabilities;
3774 		break;
3775 	case MSR_IA32_PERF_CAPABILITIES:
3776 		if (!msr_info->host_initiated &&
3777 		    !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3778 			return 1;
3779 		msr_info->data = vcpu->arch.perf_capabilities;
3780 		break;
3781 	case MSR_IA32_POWER_CTL:
3782 		msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3783 		break;
3784 	case MSR_IA32_TSC: {
3785 		/*
3786 		 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3787 		 * even when not intercepted. AMD manual doesn't explicitly
3788 		 * state this but appears to behave the same.
3789 		 *
3790 		 * On userspace reads and writes, however, we unconditionally
3791 		 * return L1's TSC value to ensure backwards-compatible
3792 		 * behavior for migration.
3793 		 */
3794 		u64 offset, ratio;
3795 
3796 		if (msr_info->host_initiated) {
3797 			offset = vcpu->arch.l1_tsc_offset;
3798 			ratio = vcpu->arch.l1_tsc_scaling_ratio;
3799 		} else {
3800 			offset = vcpu->arch.tsc_offset;
3801 			ratio = vcpu->arch.tsc_scaling_ratio;
3802 		}
3803 
3804 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc(), ratio) + offset;
3805 		break;
3806 	}
3807 	case MSR_MTRRcap:
3808 	case 0x200 ... 0x2ff:
3809 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3810 	case 0xcd: /* fsb frequency */
3811 		msr_info->data = 3;
3812 		break;
3813 		/*
3814 		 * MSR_EBC_FREQUENCY_ID
3815 		 * Conservative value valid for even the basic CPU models.
3816 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3817 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3818 		 * and 266MHz for model 3, or 4. Set Core Clock
3819 		 * Frequency to System Bus Frequency Ratio to 1 (bits
3820 		 * 31:24) even though these are only valid for CPU
3821 		 * models > 2, however guests may end up dividing or
3822 		 * multiplying by zero otherwise.
3823 		 */
3824 	case MSR_EBC_FREQUENCY_ID:
3825 		msr_info->data = 1 << 24;
3826 		break;
3827 	case MSR_IA32_APICBASE:
3828 		msr_info->data = kvm_get_apic_base(vcpu);
3829 		break;
3830 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3831 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3832 	case MSR_IA32_TSC_DEADLINE:
3833 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3834 		break;
3835 	case MSR_IA32_TSC_ADJUST:
3836 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3837 		break;
3838 	case MSR_IA32_MISC_ENABLE:
3839 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3840 		break;
3841 	case MSR_IA32_SMBASE:
3842 		if (!msr_info->host_initiated)
3843 			return 1;
3844 		msr_info->data = vcpu->arch.smbase;
3845 		break;
3846 	case MSR_SMI_COUNT:
3847 		msr_info->data = vcpu->arch.smi_count;
3848 		break;
3849 	case MSR_IA32_PERF_STATUS:
3850 		/* TSC increment by tick */
3851 		msr_info->data = 1000ULL;
3852 		/* CPU multiplier */
3853 		msr_info->data |= (((uint64_t)4ULL) << 40);
3854 		break;
3855 	case MSR_EFER:
3856 		msr_info->data = vcpu->arch.efer;
3857 		break;
3858 	case MSR_KVM_WALL_CLOCK:
3859 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3860 			return 1;
3861 
3862 		msr_info->data = vcpu->kvm->arch.wall_clock;
3863 		break;
3864 	case MSR_KVM_WALL_CLOCK_NEW:
3865 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3866 			return 1;
3867 
3868 		msr_info->data = vcpu->kvm->arch.wall_clock;
3869 		break;
3870 	case MSR_KVM_SYSTEM_TIME:
3871 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3872 			return 1;
3873 
3874 		msr_info->data = vcpu->arch.time;
3875 		break;
3876 	case MSR_KVM_SYSTEM_TIME_NEW:
3877 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3878 			return 1;
3879 
3880 		msr_info->data = vcpu->arch.time;
3881 		break;
3882 	case MSR_KVM_ASYNC_PF_EN:
3883 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3884 			return 1;
3885 
3886 		msr_info->data = vcpu->arch.apf.msr_en_val;
3887 		break;
3888 	case MSR_KVM_ASYNC_PF_INT:
3889 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3890 			return 1;
3891 
3892 		msr_info->data = vcpu->arch.apf.msr_int_val;
3893 		break;
3894 	case MSR_KVM_ASYNC_PF_ACK:
3895 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3896 			return 1;
3897 
3898 		msr_info->data = 0;
3899 		break;
3900 	case MSR_KVM_STEAL_TIME:
3901 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3902 			return 1;
3903 
3904 		msr_info->data = vcpu->arch.st.msr_val;
3905 		break;
3906 	case MSR_KVM_PV_EOI_EN:
3907 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3908 			return 1;
3909 
3910 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
3911 		break;
3912 	case MSR_KVM_POLL_CONTROL:
3913 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3914 			return 1;
3915 
3916 		msr_info->data = vcpu->arch.msr_kvm_poll_control;
3917 		break;
3918 	case MSR_IA32_P5_MC_ADDR:
3919 	case MSR_IA32_P5_MC_TYPE:
3920 	case MSR_IA32_MCG_CAP:
3921 	case MSR_IA32_MCG_CTL:
3922 	case MSR_IA32_MCG_STATUS:
3923 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3924 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3925 				   msr_info->host_initiated);
3926 	case MSR_IA32_XSS:
3927 		if (!msr_info->host_initiated &&
3928 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3929 			return 1;
3930 		msr_info->data = vcpu->arch.ia32_xss;
3931 		break;
3932 	case MSR_K7_CLK_CTL:
3933 		/*
3934 		 * Provide expected ramp-up count for K7. All other
3935 		 * are set to zero, indicating minimum divisors for
3936 		 * every field.
3937 		 *
3938 		 * This prevents guest kernels on AMD host with CPU
3939 		 * type 6, model 8 and higher from exploding due to
3940 		 * the rdmsr failing.
3941 		 */
3942 		msr_info->data = 0x20000000;
3943 		break;
3944 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3945 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3946 	case HV_X64_MSR_SYNDBG_OPTIONS:
3947 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3948 	case HV_X64_MSR_CRASH_CTL:
3949 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3950 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3951 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3952 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3953 		return kvm_hv_get_msr_common(vcpu,
3954 					     msr_info->index, &msr_info->data,
3955 					     msr_info->host_initiated);
3956 	case MSR_IA32_BBL_CR_CTL3:
3957 		/* This legacy MSR exists but isn't fully documented in current
3958 		 * silicon.  It is however accessed by winxp in very narrow
3959 		 * scenarios where it sets bit #19, itself documented as
3960 		 * a "reserved" bit.  Best effort attempt to source coherent
3961 		 * read data here should the balance of the register be
3962 		 * interpreted by the guest:
3963 		 *
3964 		 * L2 cache control register 3: 64GB range, 256KB size,
3965 		 * enabled, latency 0x1, configured
3966 		 */
3967 		msr_info->data = 0xbe702111;
3968 		break;
3969 	case MSR_AMD64_OSVW_ID_LENGTH:
3970 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3971 			return 1;
3972 		msr_info->data = vcpu->arch.osvw.length;
3973 		break;
3974 	case MSR_AMD64_OSVW_STATUS:
3975 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3976 			return 1;
3977 		msr_info->data = vcpu->arch.osvw.status;
3978 		break;
3979 	case MSR_PLATFORM_INFO:
3980 		if (!msr_info->host_initiated &&
3981 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3982 			return 1;
3983 		msr_info->data = vcpu->arch.msr_platform_info;
3984 		break;
3985 	case MSR_MISC_FEATURES_ENABLES:
3986 		msr_info->data = vcpu->arch.msr_misc_features_enables;
3987 		break;
3988 	case MSR_K7_HWCR:
3989 		msr_info->data = vcpu->arch.msr_hwcr;
3990 		break;
3991 	default:
3992 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3993 			return kvm_pmu_get_msr(vcpu, msr_info);
3994 		return KVM_MSR_RET_INVALID;
3995 	}
3996 	return 0;
3997 }
3998 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3999 
4000 /*
4001  * Read or write a bunch of msrs. All parameters are kernel addresses.
4002  *
4003  * @return number of msrs set successfully.
4004  */
4005 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
4006 		    struct kvm_msr_entry *entries,
4007 		    int (*do_msr)(struct kvm_vcpu *vcpu,
4008 				  unsigned index, u64 *data))
4009 {
4010 	int i;
4011 
4012 	for (i = 0; i < msrs->nmsrs; ++i)
4013 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
4014 			break;
4015 
4016 	return i;
4017 }
4018 
4019 /*
4020  * Read or write a bunch of msrs. Parameters are user addresses.
4021  *
4022  * @return number of msrs set successfully.
4023  */
4024 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
4025 		  int (*do_msr)(struct kvm_vcpu *vcpu,
4026 				unsigned index, u64 *data),
4027 		  int writeback)
4028 {
4029 	struct kvm_msrs msrs;
4030 	struct kvm_msr_entry *entries;
4031 	int r, n;
4032 	unsigned size;
4033 
4034 	r = -EFAULT;
4035 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
4036 		goto out;
4037 
4038 	r = -E2BIG;
4039 	if (msrs.nmsrs >= MAX_IO_MSRS)
4040 		goto out;
4041 
4042 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
4043 	entries = memdup_user(user_msrs->entries, size);
4044 	if (IS_ERR(entries)) {
4045 		r = PTR_ERR(entries);
4046 		goto out;
4047 	}
4048 
4049 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
4050 	if (r < 0)
4051 		goto out_free;
4052 
4053 	r = -EFAULT;
4054 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
4055 		goto out_free;
4056 
4057 	r = n;
4058 
4059 out_free:
4060 	kfree(entries);
4061 out:
4062 	return r;
4063 }
4064 
4065 static inline bool kvm_can_mwait_in_guest(void)
4066 {
4067 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
4068 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
4069 		boot_cpu_has(X86_FEATURE_ARAT);
4070 }
4071 
4072 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
4073 					    struct kvm_cpuid2 __user *cpuid_arg)
4074 {
4075 	struct kvm_cpuid2 cpuid;
4076 	int r;
4077 
4078 	r = -EFAULT;
4079 	if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4080 		return r;
4081 
4082 	r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4083 	if (r)
4084 		return r;
4085 
4086 	r = -EFAULT;
4087 	if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4088 		return r;
4089 
4090 	return 0;
4091 }
4092 
4093 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
4094 {
4095 	int r = 0;
4096 
4097 	switch (ext) {
4098 	case KVM_CAP_IRQCHIP:
4099 	case KVM_CAP_HLT:
4100 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
4101 	case KVM_CAP_SET_TSS_ADDR:
4102 	case KVM_CAP_EXT_CPUID:
4103 	case KVM_CAP_EXT_EMUL_CPUID:
4104 	case KVM_CAP_CLOCKSOURCE:
4105 	case KVM_CAP_PIT:
4106 	case KVM_CAP_NOP_IO_DELAY:
4107 	case KVM_CAP_MP_STATE:
4108 	case KVM_CAP_SYNC_MMU:
4109 	case KVM_CAP_USER_NMI:
4110 	case KVM_CAP_REINJECT_CONTROL:
4111 	case KVM_CAP_IRQ_INJECT_STATUS:
4112 	case KVM_CAP_IOEVENTFD:
4113 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
4114 	case KVM_CAP_PIT2:
4115 	case KVM_CAP_PIT_STATE2:
4116 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
4117 	case KVM_CAP_VCPU_EVENTS:
4118 	case KVM_CAP_HYPERV:
4119 	case KVM_CAP_HYPERV_VAPIC:
4120 	case KVM_CAP_HYPERV_SPIN:
4121 	case KVM_CAP_HYPERV_SYNIC:
4122 	case KVM_CAP_HYPERV_SYNIC2:
4123 	case KVM_CAP_HYPERV_VP_INDEX:
4124 	case KVM_CAP_HYPERV_EVENTFD:
4125 	case KVM_CAP_HYPERV_TLBFLUSH:
4126 	case KVM_CAP_HYPERV_SEND_IPI:
4127 	case KVM_CAP_HYPERV_CPUID:
4128 	case KVM_CAP_HYPERV_ENFORCE_CPUID:
4129 	case KVM_CAP_SYS_HYPERV_CPUID:
4130 	case KVM_CAP_PCI_SEGMENT:
4131 	case KVM_CAP_DEBUGREGS:
4132 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
4133 	case KVM_CAP_XSAVE:
4134 	case KVM_CAP_ASYNC_PF:
4135 	case KVM_CAP_ASYNC_PF_INT:
4136 	case KVM_CAP_GET_TSC_KHZ:
4137 	case KVM_CAP_KVMCLOCK_CTRL:
4138 	case KVM_CAP_READONLY_MEM:
4139 	case KVM_CAP_HYPERV_TIME:
4140 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
4141 	case KVM_CAP_TSC_DEADLINE_TIMER:
4142 	case KVM_CAP_DISABLE_QUIRKS:
4143 	case KVM_CAP_SET_BOOT_CPU_ID:
4144  	case KVM_CAP_SPLIT_IRQCHIP:
4145 	case KVM_CAP_IMMEDIATE_EXIT:
4146 	case KVM_CAP_PMU_EVENT_FILTER:
4147 	case KVM_CAP_GET_MSR_FEATURES:
4148 	case KVM_CAP_MSR_PLATFORM_INFO:
4149 	case KVM_CAP_EXCEPTION_PAYLOAD:
4150 	case KVM_CAP_SET_GUEST_DEBUG:
4151 	case KVM_CAP_LAST_CPU:
4152 	case KVM_CAP_X86_USER_SPACE_MSR:
4153 	case KVM_CAP_X86_MSR_FILTER:
4154 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4155 #ifdef CONFIG_X86_SGX_KVM
4156 	case KVM_CAP_SGX_ATTRIBUTE:
4157 #endif
4158 	case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
4159 	case KVM_CAP_VM_MOVE_ENC_CONTEXT_FROM:
4160 	case KVM_CAP_SREGS2:
4161 	case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
4162 	case KVM_CAP_VCPU_ATTRIBUTES:
4163 		r = 1;
4164 		break;
4165 	case KVM_CAP_EXIT_HYPERCALL:
4166 		r = KVM_EXIT_HYPERCALL_VALID_MASK;
4167 		break;
4168 	case KVM_CAP_SET_GUEST_DEBUG2:
4169 		return KVM_GUESTDBG_VALID_MASK;
4170 #ifdef CONFIG_KVM_XEN
4171 	case KVM_CAP_XEN_HVM:
4172 		r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
4173 		    KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
4174 		    KVM_XEN_HVM_CONFIG_SHARED_INFO;
4175 		if (sched_info_on())
4176 			r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
4177 		break;
4178 #endif
4179 	case KVM_CAP_SYNC_REGS:
4180 		r = KVM_SYNC_X86_VALID_FIELDS;
4181 		break;
4182 	case KVM_CAP_ADJUST_CLOCK:
4183 		r = KVM_CLOCK_VALID_FLAGS;
4184 		break;
4185 	case KVM_CAP_X86_DISABLE_EXITS:
4186 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
4187 		      KVM_X86_DISABLE_EXITS_CSTATE;
4188 		if(kvm_can_mwait_in_guest())
4189 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
4190 		break;
4191 	case KVM_CAP_X86_SMM:
4192 		/* SMBASE is usually relocated above 1M on modern chipsets,
4193 		 * and SMM handlers might indeed rely on 4G segment limits,
4194 		 * so do not report SMM to be available if real mode is
4195 		 * emulated via vm86 mode.  Still, do not go to great lengths
4196 		 * to avoid userspace's usage of the feature, because it is a
4197 		 * fringe case that is not enabled except via specific settings
4198 		 * of the module parameters.
4199 		 */
4200 		r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
4201 		break;
4202 	case KVM_CAP_VAPIC:
4203 		r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
4204 		break;
4205 	case KVM_CAP_NR_VCPUS:
4206 		r = min_t(unsigned int, num_online_cpus(), KVM_MAX_VCPUS);
4207 		break;
4208 	case KVM_CAP_MAX_VCPUS:
4209 		r = KVM_MAX_VCPUS;
4210 		break;
4211 	case KVM_CAP_MAX_VCPU_ID:
4212 		r = KVM_MAX_VCPU_IDS;
4213 		break;
4214 	case KVM_CAP_PV_MMU:	/* obsolete */
4215 		r = 0;
4216 		break;
4217 	case KVM_CAP_MCE:
4218 		r = KVM_MAX_MCE_BANKS;
4219 		break;
4220 	case KVM_CAP_XCRS:
4221 		r = boot_cpu_has(X86_FEATURE_XSAVE);
4222 		break;
4223 	case KVM_CAP_TSC_CONTROL:
4224 		r = kvm_has_tsc_control;
4225 		break;
4226 	case KVM_CAP_X2APIC_API:
4227 		r = KVM_X2APIC_API_VALID_FLAGS;
4228 		break;
4229 	case KVM_CAP_NESTED_STATE:
4230 		r = kvm_x86_ops.nested_ops->get_state ?
4231 			kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
4232 		break;
4233 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4234 		r = kvm_x86_ops.enable_direct_tlbflush != NULL;
4235 		break;
4236 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4237 		r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
4238 		break;
4239 	case KVM_CAP_SMALLER_MAXPHYADDR:
4240 		r = (int) allow_smaller_maxphyaddr;
4241 		break;
4242 	case KVM_CAP_STEAL_TIME:
4243 		r = sched_info_on();
4244 		break;
4245 	case KVM_CAP_X86_BUS_LOCK_EXIT:
4246 		if (kvm_has_bus_lock_exit)
4247 			r = KVM_BUS_LOCK_DETECTION_OFF |
4248 			    KVM_BUS_LOCK_DETECTION_EXIT;
4249 		else
4250 			r = 0;
4251 		break;
4252 	default:
4253 		break;
4254 	}
4255 	return r;
4256 
4257 }
4258 
4259 long kvm_arch_dev_ioctl(struct file *filp,
4260 			unsigned int ioctl, unsigned long arg)
4261 {
4262 	void __user *argp = (void __user *)arg;
4263 	long r;
4264 
4265 	switch (ioctl) {
4266 	case KVM_GET_MSR_INDEX_LIST: {
4267 		struct kvm_msr_list __user *user_msr_list = argp;
4268 		struct kvm_msr_list msr_list;
4269 		unsigned n;
4270 
4271 		r = -EFAULT;
4272 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4273 			goto out;
4274 		n = msr_list.nmsrs;
4275 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4276 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4277 			goto out;
4278 		r = -E2BIG;
4279 		if (n < msr_list.nmsrs)
4280 			goto out;
4281 		r = -EFAULT;
4282 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4283 				 num_msrs_to_save * sizeof(u32)))
4284 			goto out;
4285 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4286 				 &emulated_msrs,
4287 				 num_emulated_msrs * sizeof(u32)))
4288 			goto out;
4289 		r = 0;
4290 		break;
4291 	}
4292 	case KVM_GET_SUPPORTED_CPUID:
4293 	case KVM_GET_EMULATED_CPUID: {
4294 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4295 		struct kvm_cpuid2 cpuid;
4296 
4297 		r = -EFAULT;
4298 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4299 			goto out;
4300 
4301 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4302 					    ioctl);
4303 		if (r)
4304 			goto out;
4305 
4306 		r = -EFAULT;
4307 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4308 			goto out;
4309 		r = 0;
4310 		break;
4311 	}
4312 	case KVM_X86_GET_MCE_CAP_SUPPORTED:
4313 		r = -EFAULT;
4314 		if (copy_to_user(argp, &kvm_mce_cap_supported,
4315 				 sizeof(kvm_mce_cap_supported)))
4316 			goto out;
4317 		r = 0;
4318 		break;
4319 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4320 		struct kvm_msr_list __user *user_msr_list = argp;
4321 		struct kvm_msr_list msr_list;
4322 		unsigned int n;
4323 
4324 		r = -EFAULT;
4325 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4326 			goto out;
4327 		n = msr_list.nmsrs;
4328 		msr_list.nmsrs = num_msr_based_features;
4329 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4330 			goto out;
4331 		r = -E2BIG;
4332 		if (n < msr_list.nmsrs)
4333 			goto out;
4334 		r = -EFAULT;
4335 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
4336 				 num_msr_based_features * sizeof(u32)))
4337 			goto out;
4338 		r = 0;
4339 		break;
4340 	}
4341 	case KVM_GET_MSRS:
4342 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
4343 		break;
4344 	case KVM_GET_SUPPORTED_HV_CPUID:
4345 		r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4346 		break;
4347 	default:
4348 		r = -EINVAL;
4349 		break;
4350 	}
4351 out:
4352 	return r;
4353 }
4354 
4355 static void wbinvd_ipi(void *garbage)
4356 {
4357 	wbinvd();
4358 }
4359 
4360 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4361 {
4362 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4363 }
4364 
4365 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4366 {
4367 	/* Address WBINVD may be executed by guest */
4368 	if (need_emulate_wbinvd(vcpu)) {
4369 		if (static_call(kvm_x86_has_wbinvd_exit)())
4370 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4371 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4372 			smp_call_function_single(vcpu->cpu,
4373 					wbinvd_ipi, NULL, 1);
4374 	}
4375 
4376 	static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4377 
4378 	/* Save host pkru register if supported */
4379 	vcpu->arch.host_pkru = read_pkru();
4380 
4381 	/* Apply any externally detected TSC adjustments (due to suspend) */
4382 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4383 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4384 		vcpu->arch.tsc_offset_adjustment = 0;
4385 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4386 	}
4387 
4388 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4389 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4390 				rdtsc() - vcpu->arch.last_host_tsc;
4391 		if (tsc_delta < 0)
4392 			mark_tsc_unstable("KVM discovered backwards TSC");
4393 
4394 		if (kvm_check_tsc_unstable()) {
4395 			u64 offset = kvm_compute_l1_tsc_offset(vcpu,
4396 						vcpu->arch.last_guest_tsc);
4397 			kvm_vcpu_write_tsc_offset(vcpu, offset);
4398 			vcpu->arch.tsc_catchup = 1;
4399 		}
4400 
4401 		if (kvm_lapic_hv_timer_in_use(vcpu))
4402 			kvm_lapic_restart_hv_timer(vcpu);
4403 
4404 		/*
4405 		 * On a host with synchronized TSC, there is no need to update
4406 		 * kvmclock on vcpu->cpu migration
4407 		 */
4408 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4409 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4410 		if (vcpu->cpu != cpu)
4411 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4412 		vcpu->cpu = cpu;
4413 	}
4414 
4415 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4416 }
4417 
4418 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4419 {
4420 	struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
4421 	struct kvm_steal_time __user *st;
4422 	struct kvm_memslots *slots;
4423 	static const u8 preempted = KVM_VCPU_PREEMPTED;
4424 
4425 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4426 		return;
4427 
4428 	if (vcpu->arch.st.preempted)
4429 		return;
4430 
4431 	/* This happens on process exit */
4432 	if (unlikely(current->mm != vcpu->kvm->mm))
4433 		return;
4434 
4435 	slots = kvm_memslots(vcpu->kvm);
4436 
4437 	if (unlikely(slots->generation != ghc->generation ||
4438 		     kvm_is_error_hva(ghc->hva) || !ghc->memslot))
4439 		return;
4440 
4441 	st = (struct kvm_steal_time __user *)ghc->hva;
4442 	BUILD_BUG_ON(sizeof(st->preempted) != sizeof(preempted));
4443 
4444 	if (!copy_to_user_nofault(&st->preempted, &preempted, sizeof(preempted)))
4445 		vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4446 
4447 	mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
4448 }
4449 
4450 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4451 {
4452 	int idx;
4453 
4454 	if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4455 		vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4456 
4457 	/*
4458 	 * Take the srcu lock as memslots will be accessed to check the gfn
4459 	 * cache generation against the memslots generation.
4460 	 */
4461 	idx = srcu_read_lock(&vcpu->kvm->srcu);
4462 	if (kvm_xen_msr_enabled(vcpu->kvm))
4463 		kvm_xen_runstate_set_preempted(vcpu);
4464 	else
4465 		kvm_steal_time_set_preempted(vcpu);
4466 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
4467 
4468 	static_call(kvm_x86_vcpu_put)(vcpu);
4469 	vcpu->arch.last_host_tsc = rdtsc();
4470 }
4471 
4472 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4473 				    struct kvm_lapic_state *s)
4474 {
4475 	static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
4476 
4477 	return kvm_apic_get_state(vcpu, s);
4478 }
4479 
4480 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4481 				    struct kvm_lapic_state *s)
4482 {
4483 	int r;
4484 
4485 	r = kvm_apic_set_state(vcpu, s);
4486 	if (r)
4487 		return r;
4488 	update_cr8_intercept(vcpu);
4489 
4490 	return 0;
4491 }
4492 
4493 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4494 {
4495 	/*
4496 	 * We can accept userspace's request for interrupt injection
4497 	 * as long as we have a place to store the interrupt number.
4498 	 * The actual injection will happen when the CPU is able to
4499 	 * deliver the interrupt.
4500 	 */
4501 	if (kvm_cpu_has_extint(vcpu))
4502 		return false;
4503 
4504 	/* Acknowledging ExtINT does not happen if LINT0 is masked.  */
4505 	return (!lapic_in_kernel(vcpu) ||
4506 		kvm_apic_accept_pic_intr(vcpu));
4507 }
4508 
4509 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4510 {
4511 	/*
4512 	 * Do not cause an interrupt window exit if an exception
4513 	 * is pending or an event needs reinjection; userspace
4514 	 * might want to inject the interrupt manually using KVM_SET_REGS
4515 	 * or KVM_SET_SREGS.  For that to work, we must be at an
4516 	 * instruction boundary and with no events half-injected.
4517 	 */
4518 	return (kvm_arch_interrupt_allowed(vcpu) &&
4519 		kvm_cpu_accept_dm_intr(vcpu) &&
4520 		!kvm_event_needs_reinjection(vcpu) &&
4521 		!vcpu->arch.exception.pending);
4522 }
4523 
4524 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4525 				    struct kvm_interrupt *irq)
4526 {
4527 	if (irq->irq >= KVM_NR_INTERRUPTS)
4528 		return -EINVAL;
4529 
4530 	if (!irqchip_in_kernel(vcpu->kvm)) {
4531 		kvm_queue_interrupt(vcpu, irq->irq, false);
4532 		kvm_make_request(KVM_REQ_EVENT, vcpu);
4533 		return 0;
4534 	}
4535 
4536 	/*
4537 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4538 	 * fail for in-kernel 8259.
4539 	 */
4540 	if (pic_in_kernel(vcpu->kvm))
4541 		return -ENXIO;
4542 
4543 	if (vcpu->arch.pending_external_vector != -1)
4544 		return -EEXIST;
4545 
4546 	vcpu->arch.pending_external_vector = irq->irq;
4547 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4548 	return 0;
4549 }
4550 
4551 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4552 {
4553 	kvm_inject_nmi(vcpu);
4554 
4555 	return 0;
4556 }
4557 
4558 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4559 {
4560 	kvm_make_request(KVM_REQ_SMI, vcpu);
4561 
4562 	return 0;
4563 }
4564 
4565 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4566 					   struct kvm_tpr_access_ctl *tac)
4567 {
4568 	if (tac->flags)
4569 		return -EINVAL;
4570 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
4571 	return 0;
4572 }
4573 
4574 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4575 					u64 mcg_cap)
4576 {
4577 	int r;
4578 	unsigned bank_num = mcg_cap & 0xff, bank;
4579 
4580 	r = -EINVAL;
4581 	if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4582 		goto out;
4583 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4584 		goto out;
4585 	r = 0;
4586 	vcpu->arch.mcg_cap = mcg_cap;
4587 	/* Init IA32_MCG_CTL to all 1s */
4588 	if (mcg_cap & MCG_CTL_P)
4589 		vcpu->arch.mcg_ctl = ~(u64)0;
4590 	/* Init IA32_MCi_CTL to all 1s */
4591 	for (bank = 0; bank < bank_num; bank++)
4592 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4593 
4594 	static_call(kvm_x86_setup_mce)(vcpu);
4595 out:
4596 	return r;
4597 }
4598 
4599 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4600 				      struct kvm_x86_mce *mce)
4601 {
4602 	u64 mcg_cap = vcpu->arch.mcg_cap;
4603 	unsigned bank_num = mcg_cap & 0xff;
4604 	u64 *banks = vcpu->arch.mce_banks;
4605 
4606 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4607 		return -EINVAL;
4608 	/*
4609 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4610 	 * reporting is disabled
4611 	 */
4612 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4613 	    vcpu->arch.mcg_ctl != ~(u64)0)
4614 		return 0;
4615 	banks += 4 * mce->bank;
4616 	/*
4617 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4618 	 * reporting is disabled for the bank
4619 	 */
4620 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4621 		return 0;
4622 	if (mce->status & MCI_STATUS_UC) {
4623 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4624 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4625 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4626 			return 0;
4627 		}
4628 		if (banks[1] & MCI_STATUS_VAL)
4629 			mce->status |= MCI_STATUS_OVER;
4630 		banks[2] = mce->addr;
4631 		banks[3] = mce->misc;
4632 		vcpu->arch.mcg_status = mce->mcg_status;
4633 		banks[1] = mce->status;
4634 		kvm_queue_exception(vcpu, MC_VECTOR);
4635 	} else if (!(banks[1] & MCI_STATUS_VAL)
4636 		   || !(banks[1] & MCI_STATUS_UC)) {
4637 		if (banks[1] & MCI_STATUS_VAL)
4638 			mce->status |= MCI_STATUS_OVER;
4639 		banks[2] = mce->addr;
4640 		banks[3] = mce->misc;
4641 		banks[1] = mce->status;
4642 	} else
4643 		banks[1] |= MCI_STATUS_OVER;
4644 	return 0;
4645 }
4646 
4647 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4648 					       struct kvm_vcpu_events *events)
4649 {
4650 	process_nmi(vcpu);
4651 
4652 	if (kvm_check_request(KVM_REQ_SMI, vcpu))
4653 		process_smi(vcpu);
4654 
4655 	/*
4656 	 * In guest mode, payload delivery should be deferred,
4657 	 * so that the L1 hypervisor can intercept #PF before
4658 	 * CR2 is modified (or intercept #DB before DR6 is
4659 	 * modified under nVMX). Unless the per-VM capability,
4660 	 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4661 	 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4662 	 * opportunistically defer the exception payload, deliver it if the
4663 	 * capability hasn't been requested before processing a
4664 	 * KVM_GET_VCPU_EVENTS.
4665 	 */
4666 	if (!vcpu->kvm->arch.exception_payload_enabled &&
4667 	    vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4668 		kvm_deliver_exception_payload(vcpu);
4669 
4670 	/*
4671 	 * The API doesn't provide the instruction length for software
4672 	 * exceptions, so don't report them. As long as the guest RIP
4673 	 * isn't advanced, we should expect to encounter the exception
4674 	 * again.
4675 	 */
4676 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4677 		events->exception.injected = 0;
4678 		events->exception.pending = 0;
4679 	} else {
4680 		events->exception.injected = vcpu->arch.exception.injected;
4681 		events->exception.pending = vcpu->arch.exception.pending;
4682 		/*
4683 		 * For ABI compatibility, deliberately conflate
4684 		 * pending and injected exceptions when
4685 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4686 		 */
4687 		if (!vcpu->kvm->arch.exception_payload_enabled)
4688 			events->exception.injected |=
4689 				vcpu->arch.exception.pending;
4690 	}
4691 	events->exception.nr = vcpu->arch.exception.nr;
4692 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4693 	events->exception.error_code = vcpu->arch.exception.error_code;
4694 	events->exception_has_payload = vcpu->arch.exception.has_payload;
4695 	events->exception_payload = vcpu->arch.exception.payload;
4696 
4697 	events->interrupt.injected =
4698 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4699 	events->interrupt.nr = vcpu->arch.interrupt.nr;
4700 	events->interrupt.soft = 0;
4701 	events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4702 
4703 	events->nmi.injected = vcpu->arch.nmi_injected;
4704 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
4705 	events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4706 	events->nmi.pad = 0;
4707 
4708 	events->sipi_vector = 0; /* never valid when reporting to user space */
4709 
4710 	events->smi.smm = is_smm(vcpu);
4711 	events->smi.pending = vcpu->arch.smi_pending;
4712 	events->smi.smm_inside_nmi =
4713 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4714 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4715 
4716 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4717 			 | KVM_VCPUEVENT_VALID_SHADOW
4718 			 | KVM_VCPUEVENT_VALID_SMM);
4719 	if (vcpu->kvm->arch.exception_payload_enabled)
4720 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4721 
4722 	memset(&events->reserved, 0, sizeof(events->reserved));
4723 }
4724 
4725 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm);
4726 
4727 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4728 					      struct kvm_vcpu_events *events)
4729 {
4730 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4731 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4732 			      | KVM_VCPUEVENT_VALID_SHADOW
4733 			      | KVM_VCPUEVENT_VALID_SMM
4734 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
4735 		return -EINVAL;
4736 
4737 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4738 		if (!vcpu->kvm->arch.exception_payload_enabled)
4739 			return -EINVAL;
4740 		if (events->exception.pending)
4741 			events->exception.injected = 0;
4742 		else
4743 			events->exception_has_payload = 0;
4744 	} else {
4745 		events->exception.pending = 0;
4746 		events->exception_has_payload = 0;
4747 	}
4748 
4749 	if ((events->exception.injected || events->exception.pending) &&
4750 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4751 		return -EINVAL;
4752 
4753 	/* INITs are latched while in SMM */
4754 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4755 	    (events->smi.smm || events->smi.pending) &&
4756 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4757 		return -EINVAL;
4758 
4759 	process_nmi(vcpu);
4760 	vcpu->arch.exception.injected = events->exception.injected;
4761 	vcpu->arch.exception.pending = events->exception.pending;
4762 	vcpu->arch.exception.nr = events->exception.nr;
4763 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4764 	vcpu->arch.exception.error_code = events->exception.error_code;
4765 	vcpu->arch.exception.has_payload = events->exception_has_payload;
4766 	vcpu->arch.exception.payload = events->exception_payload;
4767 
4768 	vcpu->arch.interrupt.injected = events->interrupt.injected;
4769 	vcpu->arch.interrupt.nr = events->interrupt.nr;
4770 	vcpu->arch.interrupt.soft = events->interrupt.soft;
4771 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4772 		static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4773 						events->interrupt.shadow);
4774 
4775 	vcpu->arch.nmi_injected = events->nmi.injected;
4776 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4777 		vcpu->arch.nmi_pending = events->nmi.pending;
4778 	static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4779 
4780 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4781 	    lapic_in_kernel(vcpu))
4782 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
4783 
4784 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4785 		if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm)
4786 			kvm_smm_changed(vcpu, events->smi.smm);
4787 
4788 		vcpu->arch.smi_pending = events->smi.pending;
4789 
4790 		if (events->smi.smm) {
4791 			if (events->smi.smm_inside_nmi)
4792 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4793 			else
4794 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4795 		}
4796 
4797 		if (lapic_in_kernel(vcpu)) {
4798 			if (events->smi.latched_init)
4799 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4800 			else
4801 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4802 		}
4803 	}
4804 
4805 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4806 
4807 	return 0;
4808 }
4809 
4810 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4811 					     struct kvm_debugregs *dbgregs)
4812 {
4813 	unsigned long val;
4814 
4815 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4816 	kvm_get_dr(vcpu, 6, &val);
4817 	dbgregs->dr6 = val;
4818 	dbgregs->dr7 = vcpu->arch.dr7;
4819 	dbgregs->flags = 0;
4820 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4821 }
4822 
4823 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4824 					    struct kvm_debugregs *dbgregs)
4825 {
4826 	if (dbgregs->flags)
4827 		return -EINVAL;
4828 
4829 	if (!kvm_dr6_valid(dbgregs->dr6))
4830 		return -EINVAL;
4831 	if (!kvm_dr7_valid(dbgregs->dr7))
4832 		return -EINVAL;
4833 
4834 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4835 	kvm_update_dr0123(vcpu);
4836 	vcpu->arch.dr6 = dbgregs->dr6;
4837 	vcpu->arch.dr7 = dbgregs->dr7;
4838 	kvm_update_dr7(vcpu);
4839 
4840 	return 0;
4841 }
4842 
4843 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4844 					 struct kvm_xsave *guest_xsave)
4845 {
4846 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
4847 		return;
4848 
4849 	fpu_copy_guest_fpstate_to_uabi(&vcpu->arch.guest_fpu,
4850 				       guest_xsave->region,
4851 				       sizeof(guest_xsave->region),
4852 				       vcpu->arch.pkru);
4853 }
4854 
4855 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4856 					struct kvm_xsave *guest_xsave)
4857 {
4858 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
4859 		return 0;
4860 
4861 	return fpu_copy_uabi_to_guest_fpstate(&vcpu->arch.guest_fpu,
4862 					      guest_xsave->region,
4863 					      supported_xcr0, &vcpu->arch.pkru);
4864 }
4865 
4866 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4867 					struct kvm_xcrs *guest_xcrs)
4868 {
4869 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4870 		guest_xcrs->nr_xcrs = 0;
4871 		return;
4872 	}
4873 
4874 	guest_xcrs->nr_xcrs = 1;
4875 	guest_xcrs->flags = 0;
4876 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4877 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4878 }
4879 
4880 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4881 				       struct kvm_xcrs *guest_xcrs)
4882 {
4883 	int i, r = 0;
4884 
4885 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
4886 		return -EINVAL;
4887 
4888 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4889 		return -EINVAL;
4890 
4891 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4892 		/* Only support XCR0 currently */
4893 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4894 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4895 				guest_xcrs->xcrs[i].value);
4896 			break;
4897 		}
4898 	if (r)
4899 		r = -EINVAL;
4900 	return r;
4901 }
4902 
4903 /*
4904  * kvm_set_guest_paused() indicates to the guest kernel that it has been
4905  * stopped by the hypervisor.  This function will be called from the host only.
4906  * EINVAL is returned when the host attempts to set the flag for a guest that
4907  * does not support pv clocks.
4908  */
4909 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4910 {
4911 	if (!vcpu->arch.pv_time_enabled)
4912 		return -EINVAL;
4913 	vcpu->arch.pvclock_set_guest_stopped_request = true;
4914 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4915 	return 0;
4916 }
4917 
4918 static int kvm_arch_tsc_has_attr(struct kvm_vcpu *vcpu,
4919 				 struct kvm_device_attr *attr)
4920 {
4921 	int r;
4922 
4923 	switch (attr->attr) {
4924 	case KVM_VCPU_TSC_OFFSET:
4925 		r = 0;
4926 		break;
4927 	default:
4928 		r = -ENXIO;
4929 	}
4930 
4931 	return r;
4932 }
4933 
4934 static int kvm_arch_tsc_get_attr(struct kvm_vcpu *vcpu,
4935 				 struct kvm_device_attr *attr)
4936 {
4937 	u64 __user *uaddr = (u64 __user *)(unsigned long)attr->addr;
4938 	int r;
4939 
4940 	if ((u64)(unsigned long)uaddr != attr->addr)
4941 		return -EFAULT;
4942 
4943 	switch (attr->attr) {
4944 	case KVM_VCPU_TSC_OFFSET:
4945 		r = -EFAULT;
4946 		if (put_user(vcpu->arch.l1_tsc_offset, uaddr))
4947 			break;
4948 		r = 0;
4949 		break;
4950 	default:
4951 		r = -ENXIO;
4952 	}
4953 
4954 	return r;
4955 }
4956 
4957 static int kvm_arch_tsc_set_attr(struct kvm_vcpu *vcpu,
4958 				 struct kvm_device_attr *attr)
4959 {
4960 	u64 __user *uaddr = (u64 __user *)(unsigned long)attr->addr;
4961 	struct kvm *kvm = vcpu->kvm;
4962 	int r;
4963 
4964 	if ((u64)(unsigned long)uaddr != attr->addr)
4965 		return -EFAULT;
4966 
4967 	switch (attr->attr) {
4968 	case KVM_VCPU_TSC_OFFSET: {
4969 		u64 offset, tsc, ns;
4970 		unsigned long flags;
4971 		bool matched;
4972 
4973 		r = -EFAULT;
4974 		if (get_user(offset, uaddr))
4975 			break;
4976 
4977 		raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
4978 
4979 		matched = (vcpu->arch.virtual_tsc_khz &&
4980 			   kvm->arch.last_tsc_khz == vcpu->arch.virtual_tsc_khz &&
4981 			   kvm->arch.last_tsc_offset == offset);
4982 
4983 		tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio) + offset;
4984 		ns = get_kvmclock_base_ns();
4985 
4986 		__kvm_synchronize_tsc(vcpu, offset, tsc, ns, matched);
4987 		raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
4988 
4989 		r = 0;
4990 		break;
4991 	}
4992 	default:
4993 		r = -ENXIO;
4994 	}
4995 
4996 	return r;
4997 }
4998 
4999 static int kvm_vcpu_ioctl_device_attr(struct kvm_vcpu *vcpu,
5000 				      unsigned int ioctl,
5001 				      void __user *argp)
5002 {
5003 	struct kvm_device_attr attr;
5004 	int r;
5005 
5006 	if (copy_from_user(&attr, argp, sizeof(attr)))
5007 		return -EFAULT;
5008 
5009 	if (attr.group != KVM_VCPU_TSC_CTRL)
5010 		return -ENXIO;
5011 
5012 	switch (ioctl) {
5013 	case KVM_HAS_DEVICE_ATTR:
5014 		r = kvm_arch_tsc_has_attr(vcpu, &attr);
5015 		break;
5016 	case KVM_GET_DEVICE_ATTR:
5017 		r = kvm_arch_tsc_get_attr(vcpu, &attr);
5018 		break;
5019 	case KVM_SET_DEVICE_ATTR:
5020 		r = kvm_arch_tsc_set_attr(vcpu, &attr);
5021 		break;
5022 	}
5023 
5024 	return r;
5025 }
5026 
5027 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
5028 				     struct kvm_enable_cap *cap)
5029 {
5030 	int r;
5031 	uint16_t vmcs_version;
5032 	void __user *user_ptr;
5033 
5034 	if (cap->flags)
5035 		return -EINVAL;
5036 
5037 	switch (cap->cap) {
5038 	case KVM_CAP_HYPERV_SYNIC2:
5039 		if (cap->args[0])
5040 			return -EINVAL;
5041 		fallthrough;
5042 
5043 	case KVM_CAP_HYPERV_SYNIC:
5044 		if (!irqchip_in_kernel(vcpu->kvm))
5045 			return -EINVAL;
5046 		return kvm_hv_activate_synic(vcpu, cap->cap ==
5047 					     KVM_CAP_HYPERV_SYNIC2);
5048 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
5049 		if (!kvm_x86_ops.nested_ops->enable_evmcs)
5050 			return -ENOTTY;
5051 		r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
5052 		if (!r) {
5053 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
5054 			if (copy_to_user(user_ptr, &vmcs_version,
5055 					 sizeof(vmcs_version)))
5056 				r = -EFAULT;
5057 		}
5058 		return r;
5059 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
5060 		if (!kvm_x86_ops.enable_direct_tlbflush)
5061 			return -ENOTTY;
5062 
5063 		return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
5064 
5065 	case KVM_CAP_HYPERV_ENFORCE_CPUID:
5066 		return kvm_hv_set_enforce_cpuid(vcpu, cap->args[0]);
5067 
5068 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
5069 		vcpu->arch.pv_cpuid.enforce = cap->args[0];
5070 		if (vcpu->arch.pv_cpuid.enforce)
5071 			kvm_update_pv_runtime(vcpu);
5072 
5073 		return 0;
5074 	default:
5075 		return -EINVAL;
5076 	}
5077 }
5078 
5079 long kvm_arch_vcpu_ioctl(struct file *filp,
5080 			 unsigned int ioctl, unsigned long arg)
5081 {
5082 	struct kvm_vcpu *vcpu = filp->private_data;
5083 	void __user *argp = (void __user *)arg;
5084 	int r;
5085 	union {
5086 		struct kvm_sregs2 *sregs2;
5087 		struct kvm_lapic_state *lapic;
5088 		struct kvm_xsave *xsave;
5089 		struct kvm_xcrs *xcrs;
5090 		void *buffer;
5091 	} u;
5092 
5093 	vcpu_load(vcpu);
5094 
5095 	u.buffer = NULL;
5096 	switch (ioctl) {
5097 	case KVM_GET_LAPIC: {
5098 		r = -EINVAL;
5099 		if (!lapic_in_kernel(vcpu))
5100 			goto out;
5101 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
5102 				GFP_KERNEL_ACCOUNT);
5103 
5104 		r = -ENOMEM;
5105 		if (!u.lapic)
5106 			goto out;
5107 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
5108 		if (r)
5109 			goto out;
5110 		r = -EFAULT;
5111 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
5112 			goto out;
5113 		r = 0;
5114 		break;
5115 	}
5116 	case KVM_SET_LAPIC: {
5117 		r = -EINVAL;
5118 		if (!lapic_in_kernel(vcpu))
5119 			goto out;
5120 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
5121 		if (IS_ERR(u.lapic)) {
5122 			r = PTR_ERR(u.lapic);
5123 			goto out_nofree;
5124 		}
5125 
5126 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
5127 		break;
5128 	}
5129 	case KVM_INTERRUPT: {
5130 		struct kvm_interrupt irq;
5131 
5132 		r = -EFAULT;
5133 		if (copy_from_user(&irq, argp, sizeof(irq)))
5134 			goto out;
5135 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
5136 		break;
5137 	}
5138 	case KVM_NMI: {
5139 		r = kvm_vcpu_ioctl_nmi(vcpu);
5140 		break;
5141 	}
5142 	case KVM_SMI: {
5143 		r = kvm_vcpu_ioctl_smi(vcpu);
5144 		break;
5145 	}
5146 	case KVM_SET_CPUID: {
5147 		struct kvm_cpuid __user *cpuid_arg = argp;
5148 		struct kvm_cpuid cpuid;
5149 
5150 		/*
5151 		 * KVM does not correctly handle changing guest CPUID after KVM_RUN, as
5152 		 * MAXPHYADDR, GBPAGES support, AMD reserved bit behavior, etc.. aren't
5153 		 * tracked in kvm_mmu_page_role.  As a result, KVM may miss guest page
5154 		 * faults due to reusing SPs/SPTEs.  In practice no sane VMM mucks with
5155 		 * the core vCPU model on the fly, so fail.
5156 		 */
5157 		r = -EINVAL;
5158 		if (vcpu->arch.last_vmentry_cpu != -1)
5159 			goto out;
5160 
5161 		r = -EFAULT;
5162 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5163 			goto out;
5164 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
5165 		break;
5166 	}
5167 	case KVM_SET_CPUID2: {
5168 		struct kvm_cpuid2 __user *cpuid_arg = argp;
5169 		struct kvm_cpuid2 cpuid;
5170 
5171 		/*
5172 		 * KVM_SET_CPUID{,2} after KVM_RUN is forbidded, see the comment in
5173 		 * KVM_SET_CPUID case above.
5174 		 */
5175 		r = -EINVAL;
5176 		if (vcpu->arch.last_vmentry_cpu != -1)
5177 			goto out;
5178 
5179 		r = -EFAULT;
5180 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5181 			goto out;
5182 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
5183 					      cpuid_arg->entries);
5184 		break;
5185 	}
5186 	case KVM_GET_CPUID2: {
5187 		struct kvm_cpuid2 __user *cpuid_arg = argp;
5188 		struct kvm_cpuid2 cpuid;
5189 
5190 		r = -EFAULT;
5191 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5192 			goto out;
5193 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
5194 					      cpuid_arg->entries);
5195 		if (r)
5196 			goto out;
5197 		r = -EFAULT;
5198 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
5199 			goto out;
5200 		r = 0;
5201 		break;
5202 	}
5203 	case KVM_GET_MSRS: {
5204 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
5205 		r = msr_io(vcpu, argp, do_get_msr, 1);
5206 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5207 		break;
5208 	}
5209 	case KVM_SET_MSRS: {
5210 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
5211 		r = msr_io(vcpu, argp, do_set_msr, 0);
5212 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5213 		break;
5214 	}
5215 	case KVM_TPR_ACCESS_REPORTING: {
5216 		struct kvm_tpr_access_ctl tac;
5217 
5218 		r = -EFAULT;
5219 		if (copy_from_user(&tac, argp, sizeof(tac)))
5220 			goto out;
5221 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
5222 		if (r)
5223 			goto out;
5224 		r = -EFAULT;
5225 		if (copy_to_user(argp, &tac, sizeof(tac)))
5226 			goto out;
5227 		r = 0;
5228 		break;
5229 	};
5230 	case KVM_SET_VAPIC_ADDR: {
5231 		struct kvm_vapic_addr va;
5232 		int idx;
5233 
5234 		r = -EINVAL;
5235 		if (!lapic_in_kernel(vcpu))
5236 			goto out;
5237 		r = -EFAULT;
5238 		if (copy_from_user(&va, argp, sizeof(va)))
5239 			goto out;
5240 		idx = srcu_read_lock(&vcpu->kvm->srcu);
5241 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
5242 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5243 		break;
5244 	}
5245 	case KVM_X86_SETUP_MCE: {
5246 		u64 mcg_cap;
5247 
5248 		r = -EFAULT;
5249 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
5250 			goto out;
5251 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
5252 		break;
5253 	}
5254 	case KVM_X86_SET_MCE: {
5255 		struct kvm_x86_mce mce;
5256 
5257 		r = -EFAULT;
5258 		if (copy_from_user(&mce, argp, sizeof(mce)))
5259 			goto out;
5260 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
5261 		break;
5262 	}
5263 	case KVM_GET_VCPU_EVENTS: {
5264 		struct kvm_vcpu_events events;
5265 
5266 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
5267 
5268 		r = -EFAULT;
5269 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
5270 			break;
5271 		r = 0;
5272 		break;
5273 	}
5274 	case KVM_SET_VCPU_EVENTS: {
5275 		struct kvm_vcpu_events events;
5276 
5277 		r = -EFAULT;
5278 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5279 			break;
5280 
5281 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5282 		break;
5283 	}
5284 	case KVM_GET_DEBUGREGS: {
5285 		struct kvm_debugregs dbgregs;
5286 
5287 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5288 
5289 		r = -EFAULT;
5290 		if (copy_to_user(argp, &dbgregs,
5291 				 sizeof(struct kvm_debugregs)))
5292 			break;
5293 		r = 0;
5294 		break;
5295 	}
5296 	case KVM_SET_DEBUGREGS: {
5297 		struct kvm_debugregs dbgregs;
5298 
5299 		r = -EFAULT;
5300 		if (copy_from_user(&dbgregs, argp,
5301 				   sizeof(struct kvm_debugregs)))
5302 			break;
5303 
5304 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5305 		break;
5306 	}
5307 	case KVM_GET_XSAVE: {
5308 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5309 		r = -ENOMEM;
5310 		if (!u.xsave)
5311 			break;
5312 
5313 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5314 
5315 		r = -EFAULT;
5316 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5317 			break;
5318 		r = 0;
5319 		break;
5320 	}
5321 	case KVM_SET_XSAVE: {
5322 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
5323 		if (IS_ERR(u.xsave)) {
5324 			r = PTR_ERR(u.xsave);
5325 			goto out_nofree;
5326 		}
5327 
5328 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5329 		break;
5330 	}
5331 	case KVM_GET_XCRS: {
5332 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5333 		r = -ENOMEM;
5334 		if (!u.xcrs)
5335 			break;
5336 
5337 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5338 
5339 		r = -EFAULT;
5340 		if (copy_to_user(argp, u.xcrs,
5341 				 sizeof(struct kvm_xcrs)))
5342 			break;
5343 		r = 0;
5344 		break;
5345 	}
5346 	case KVM_SET_XCRS: {
5347 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5348 		if (IS_ERR(u.xcrs)) {
5349 			r = PTR_ERR(u.xcrs);
5350 			goto out_nofree;
5351 		}
5352 
5353 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5354 		break;
5355 	}
5356 	case KVM_SET_TSC_KHZ: {
5357 		u32 user_tsc_khz;
5358 
5359 		r = -EINVAL;
5360 		user_tsc_khz = (u32)arg;
5361 
5362 		if (kvm_has_tsc_control &&
5363 		    user_tsc_khz >= kvm_max_guest_tsc_khz)
5364 			goto out;
5365 
5366 		if (user_tsc_khz == 0)
5367 			user_tsc_khz = tsc_khz;
5368 
5369 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5370 			r = 0;
5371 
5372 		goto out;
5373 	}
5374 	case KVM_GET_TSC_KHZ: {
5375 		r = vcpu->arch.virtual_tsc_khz;
5376 		goto out;
5377 	}
5378 	case KVM_KVMCLOCK_CTRL: {
5379 		r = kvm_set_guest_paused(vcpu);
5380 		goto out;
5381 	}
5382 	case KVM_ENABLE_CAP: {
5383 		struct kvm_enable_cap cap;
5384 
5385 		r = -EFAULT;
5386 		if (copy_from_user(&cap, argp, sizeof(cap)))
5387 			goto out;
5388 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5389 		break;
5390 	}
5391 	case KVM_GET_NESTED_STATE: {
5392 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5393 		u32 user_data_size;
5394 
5395 		r = -EINVAL;
5396 		if (!kvm_x86_ops.nested_ops->get_state)
5397 			break;
5398 
5399 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5400 		r = -EFAULT;
5401 		if (get_user(user_data_size, &user_kvm_nested_state->size))
5402 			break;
5403 
5404 		r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5405 						     user_data_size);
5406 		if (r < 0)
5407 			break;
5408 
5409 		if (r > user_data_size) {
5410 			if (put_user(r, &user_kvm_nested_state->size))
5411 				r = -EFAULT;
5412 			else
5413 				r = -E2BIG;
5414 			break;
5415 		}
5416 
5417 		r = 0;
5418 		break;
5419 	}
5420 	case KVM_SET_NESTED_STATE: {
5421 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5422 		struct kvm_nested_state kvm_state;
5423 		int idx;
5424 
5425 		r = -EINVAL;
5426 		if (!kvm_x86_ops.nested_ops->set_state)
5427 			break;
5428 
5429 		r = -EFAULT;
5430 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5431 			break;
5432 
5433 		r = -EINVAL;
5434 		if (kvm_state.size < sizeof(kvm_state))
5435 			break;
5436 
5437 		if (kvm_state.flags &
5438 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5439 		      | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5440 		      | KVM_STATE_NESTED_GIF_SET))
5441 			break;
5442 
5443 		/* nested_run_pending implies guest_mode.  */
5444 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5445 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5446 			break;
5447 
5448 		idx = srcu_read_lock(&vcpu->kvm->srcu);
5449 		r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5450 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5451 		break;
5452 	}
5453 	case KVM_GET_SUPPORTED_HV_CPUID:
5454 		r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5455 		break;
5456 #ifdef CONFIG_KVM_XEN
5457 	case KVM_XEN_VCPU_GET_ATTR: {
5458 		struct kvm_xen_vcpu_attr xva;
5459 
5460 		r = -EFAULT;
5461 		if (copy_from_user(&xva, argp, sizeof(xva)))
5462 			goto out;
5463 		r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5464 		if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5465 			r = -EFAULT;
5466 		break;
5467 	}
5468 	case KVM_XEN_VCPU_SET_ATTR: {
5469 		struct kvm_xen_vcpu_attr xva;
5470 
5471 		r = -EFAULT;
5472 		if (copy_from_user(&xva, argp, sizeof(xva)))
5473 			goto out;
5474 		r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5475 		break;
5476 	}
5477 #endif
5478 	case KVM_GET_SREGS2: {
5479 		u.sregs2 = kzalloc(sizeof(struct kvm_sregs2), GFP_KERNEL);
5480 		r = -ENOMEM;
5481 		if (!u.sregs2)
5482 			goto out;
5483 		__get_sregs2(vcpu, u.sregs2);
5484 		r = -EFAULT;
5485 		if (copy_to_user(argp, u.sregs2, sizeof(struct kvm_sregs2)))
5486 			goto out;
5487 		r = 0;
5488 		break;
5489 	}
5490 	case KVM_SET_SREGS2: {
5491 		u.sregs2 = memdup_user(argp, sizeof(struct kvm_sregs2));
5492 		if (IS_ERR(u.sregs2)) {
5493 			r = PTR_ERR(u.sregs2);
5494 			u.sregs2 = NULL;
5495 			goto out;
5496 		}
5497 		r = __set_sregs2(vcpu, u.sregs2);
5498 		break;
5499 	}
5500 	case KVM_HAS_DEVICE_ATTR:
5501 	case KVM_GET_DEVICE_ATTR:
5502 	case KVM_SET_DEVICE_ATTR:
5503 		r = kvm_vcpu_ioctl_device_attr(vcpu, ioctl, argp);
5504 		break;
5505 	default:
5506 		r = -EINVAL;
5507 	}
5508 out:
5509 	kfree(u.buffer);
5510 out_nofree:
5511 	vcpu_put(vcpu);
5512 	return r;
5513 }
5514 
5515 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5516 {
5517 	return VM_FAULT_SIGBUS;
5518 }
5519 
5520 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5521 {
5522 	int ret;
5523 
5524 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
5525 		return -EINVAL;
5526 	ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5527 	return ret;
5528 }
5529 
5530 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5531 					      u64 ident_addr)
5532 {
5533 	return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5534 }
5535 
5536 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5537 					 unsigned long kvm_nr_mmu_pages)
5538 {
5539 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5540 		return -EINVAL;
5541 
5542 	mutex_lock(&kvm->slots_lock);
5543 
5544 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5545 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5546 
5547 	mutex_unlock(&kvm->slots_lock);
5548 	return 0;
5549 }
5550 
5551 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5552 {
5553 	return kvm->arch.n_max_mmu_pages;
5554 }
5555 
5556 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5557 {
5558 	struct kvm_pic *pic = kvm->arch.vpic;
5559 	int r;
5560 
5561 	r = 0;
5562 	switch (chip->chip_id) {
5563 	case KVM_IRQCHIP_PIC_MASTER:
5564 		memcpy(&chip->chip.pic, &pic->pics[0],
5565 			sizeof(struct kvm_pic_state));
5566 		break;
5567 	case KVM_IRQCHIP_PIC_SLAVE:
5568 		memcpy(&chip->chip.pic, &pic->pics[1],
5569 			sizeof(struct kvm_pic_state));
5570 		break;
5571 	case KVM_IRQCHIP_IOAPIC:
5572 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
5573 		break;
5574 	default:
5575 		r = -EINVAL;
5576 		break;
5577 	}
5578 	return r;
5579 }
5580 
5581 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5582 {
5583 	struct kvm_pic *pic = kvm->arch.vpic;
5584 	int r;
5585 
5586 	r = 0;
5587 	switch (chip->chip_id) {
5588 	case KVM_IRQCHIP_PIC_MASTER:
5589 		spin_lock(&pic->lock);
5590 		memcpy(&pic->pics[0], &chip->chip.pic,
5591 			sizeof(struct kvm_pic_state));
5592 		spin_unlock(&pic->lock);
5593 		break;
5594 	case KVM_IRQCHIP_PIC_SLAVE:
5595 		spin_lock(&pic->lock);
5596 		memcpy(&pic->pics[1], &chip->chip.pic,
5597 			sizeof(struct kvm_pic_state));
5598 		spin_unlock(&pic->lock);
5599 		break;
5600 	case KVM_IRQCHIP_IOAPIC:
5601 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
5602 		break;
5603 	default:
5604 		r = -EINVAL;
5605 		break;
5606 	}
5607 	kvm_pic_update_irq(pic);
5608 	return r;
5609 }
5610 
5611 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5612 {
5613 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5614 
5615 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5616 
5617 	mutex_lock(&kps->lock);
5618 	memcpy(ps, &kps->channels, sizeof(*ps));
5619 	mutex_unlock(&kps->lock);
5620 	return 0;
5621 }
5622 
5623 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5624 {
5625 	int i;
5626 	struct kvm_pit *pit = kvm->arch.vpit;
5627 
5628 	mutex_lock(&pit->pit_state.lock);
5629 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5630 	for (i = 0; i < 3; i++)
5631 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5632 	mutex_unlock(&pit->pit_state.lock);
5633 	return 0;
5634 }
5635 
5636 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5637 {
5638 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
5639 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5640 		sizeof(ps->channels));
5641 	ps->flags = kvm->arch.vpit->pit_state.flags;
5642 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5643 	memset(&ps->reserved, 0, sizeof(ps->reserved));
5644 	return 0;
5645 }
5646 
5647 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5648 {
5649 	int start = 0;
5650 	int i;
5651 	u32 prev_legacy, cur_legacy;
5652 	struct kvm_pit *pit = kvm->arch.vpit;
5653 
5654 	mutex_lock(&pit->pit_state.lock);
5655 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5656 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5657 	if (!prev_legacy && cur_legacy)
5658 		start = 1;
5659 	memcpy(&pit->pit_state.channels, &ps->channels,
5660 	       sizeof(pit->pit_state.channels));
5661 	pit->pit_state.flags = ps->flags;
5662 	for (i = 0; i < 3; i++)
5663 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5664 				   start && i == 0);
5665 	mutex_unlock(&pit->pit_state.lock);
5666 	return 0;
5667 }
5668 
5669 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5670 				 struct kvm_reinject_control *control)
5671 {
5672 	struct kvm_pit *pit = kvm->arch.vpit;
5673 
5674 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
5675 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5676 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
5677 	 */
5678 	mutex_lock(&pit->pit_state.lock);
5679 	kvm_pit_set_reinject(pit, control->pit_reinject);
5680 	mutex_unlock(&pit->pit_state.lock);
5681 
5682 	return 0;
5683 }
5684 
5685 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5686 {
5687 
5688 	/*
5689 	 * Flush all CPUs' dirty log buffers to the  dirty_bitmap.  Called
5690 	 * before reporting dirty_bitmap to userspace.  KVM flushes the buffers
5691 	 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5692 	 * VM-Exit.
5693 	 */
5694 	struct kvm_vcpu *vcpu;
5695 	int i;
5696 
5697 	kvm_for_each_vcpu(i, vcpu, kvm)
5698 		kvm_vcpu_kick(vcpu);
5699 }
5700 
5701 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5702 			bool line_status)
5703 {
5704 	if (!irqchip_in_kernel(kvm))
5705 		return -ENXIO;
5706 
5707 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5708 					irq_event->irq, irq_event->level,
5709 					line_status);
5710 	return 0;
5711 }
5712 
5713 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5714 			    struct kvm_enable_cap *cap)
5715 {
5716 	int r;
5717 
5718 	if (cap->flags)
5719 		return -EINVAL;
5720 
5721 	switch (cap->cap) {
5722 	case KVM_CAP_DISABLE_QUIRKS:
5723 		kvm->arch.disabled_quirks = cap->args[0];
5724 		r = 0;
5725 		break;
5726 	case KVM_CAP_SPLIT_IRQCHIP: {
5727 		mutex_lock(&kvm->lock);
5728 		r = -EINVAL;
5729 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5730 			goto split_irqchip_unlock;
5731 		r = -EEXIST;
5732 		if (irqchip_in_kernel(kvm))
5733 			goto split_irqchip_unlock;
5734 		if (kvm->created_vcpus)
5735 			goto split_irqchip_unlock;
5736 		r = kvm_setup_empty_irq_routing(kvm);
5737 		if (r)
5738 			goto split_irqchip_unlock;
5739 		/* Pairs with irqchip_in_kernel. */
5740 		smp_wmb();
5741 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5742 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5743 		r = 0;
5744 split_irqchip_unlock:
5745 		mutex_unlock(&kvm->lock);
5746 		break;
5747 	}
5748 	case KVM_CAP_X2APIC_API:
5749 		r = -EINVAL;
5750 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5751 			break;
5752 
5753 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5754 			kvm->arch.x2apic_format = true;
5755 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5756 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
5757 
5758 		r = 0;
5759 		break;
5760 	case KVM_CAP_X86_DISABLE_EXITS:
5761 		r = -EINVAL;
5762 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5763 			break;
5764 
5765 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5766 			kvm_can_mwait_in_guest())
5767 			kvm->arch.mwait_in_guest = true;
5768 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5769 			kvm->arch.hlt_in_guest = true;
5770 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5771 			kvm->arch.pause_in_guest = true;
5772 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5773 			kvm->arch.cstate_in_guest = true;
5774 		r = 0;
5775 		break;
5776 	case KVM_CAP_MSR_PLATFORM_INFO:
5777 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5778 		r = 0;
5779 		break;
5780 	case KVM_CAP_EXCEPTION_PAYLOAD:
5781 		kvm->arch.exception_payload_enabled = cap->args[0];
5782 		r = 0;
5783 		break;
5784 	case KVM_CAP_X86_USER_SPACE_MSR:
5785 		kvm->arch.user_space_msr_mask = cap->args[0];
5786 		r = 0;
5787 		break;
5788 	case KVM_CAP_X86_BUS_LOCK_EXIT:
5789 		r = -EINVAL;
5790 		if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5791 			break;
5792 
5793 		if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5794 		    (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5795 			break;
5796 
5797 		if (kvm_has_bus_lock_exit &&
5798 		    cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5799 			kvm->arch.bus_lock_detection_enabled = true;
5800 		r = 0;
5801 		break;
5802 #ifdef CONFIG_X86_SGX_KVM
5803 	case KVM_CAP_SGX_ATTRIBUTE: {
5804 		unsigned long allowed_attributes = 0;
5805 
5806 		r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5807 		if (r)
5808 			break;
5809 
5810 		/* KVM only supports the PROVISIONKEY privileged attribute. */
5811 		if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5812 		    !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5813 			kvm->arch.sgx_provisioning_allowed = true;
5814 		else
5815 			r = -EINVAL;
5816 		break;
5817 	}
5818 #endif
5819 	case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5820 		r = -EINVAL;
5821 		if (kvm_x86_ops.vm_copy_enc_context_from)
5822 			r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5823 		return r;
5824 	case KVM_CAP_VM_MOVE_ENC_CONTEXT_FROM:
5825 		r = -EINVAL;
5826 		if (kvm_x86_ops.vm_move_enc_context_from)
5827 			r = kvm_x86_ops.vm_move_enc_context_from(
5828 				kvm, cap->args[0]);
5829 		return r;
5830 	case KVM_CAP_EXIT_HYPERCALL:
5831 		if (cap->args[0] & ~KVM_EXIT_HYPERCALL_VALID_MASK) {
5832 			r = -EINVAL;
5833 			break;
5834 		}
5835 		kvm->arch.hypercall_exit_enabled = cap->args[0];
5836 		r = 0;
5837 		break;
5838 	case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
5839 		r = -EINVAL;
5840 		if (cap->args[0] & ~1)
5841 			break;
5842 		kvm->arch.exit_on_emulation_error = cap->args[0];
5843 		r = 0;
5844 		break;
5845 	default:
5846 		r = -EINVAL;
5847 		break;
5848 	}
5849 	return r;
5850 }
5851 
5852 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5853 {
5854 	struct kvm_x86_msr_filter *msr_filter;
5855 
5856 	msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5857 	if (!msr_filter)
5858 		return NULL;
5859 
5860 	msr_filter->default_allow = default_allow;
5861 	return msr_filter;
5862 }
5863 
5864 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5865 {
5866 	u32 i;
5867 
5868 	if (!msr_filter)
5869 		return;
5870 
5871 	for (i = 0; i < msr_filter->count; i++)
5872 		kfree(msr_filter->ranges[i].bitmap);
5873 
5874 	kfree(msr_filter);
5875 }
5876 
5877 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5878 			      struct kvm_msr_filter_range *user_range)
5879 {
5880 	unsigned long *bitmap = NULL;
5881 	size_t bitmap_size;
5882 
5883 	if (!user_range->nmsrs)
5884 		return 0;
5885 
5886 	if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
5887 		return -EINVAL;
5888 
5889 	if (!user_range->flags)
5890 		return -EINVAL;
5891 
5892 	bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5893 	if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5894 		return -EINVAL;
5895 
5896 	bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5897 	if (IS_ERR(bitmap))
5898 		return PTR_ERR(bitmap);
5899 
5900 	msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
5901 		.flags = user_range->flags,
5902 		.base = user_range->base,
5903 		.nmsrs = user_range->nmsrs,
5904 		.bitmap = bitmap,
5905 	};
5906 
5907 	msr_filter->count++;
5908 	return 0;
5909 }
5910 
5911 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5912 {
5913 	struct kvm_msr_filter __user *user_msr_filter = argp;
5914 	struct kvm_x86_msr_filter *new_filter, *old_filter;
5915 	struct kvm_msr_filter filter;
5916 	bool default_allow;
5917 	bool empty = true;
5918 	int r = 0;
5919 	u32 i;
5920 
5921 	if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5922 		return -EFAULT;
5923 
5924 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5925 		empty &= !filter.ranges[i].nmsrs;
5926 
5927 	default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5928 	if (empty && !default_allow)
5929 		return -EINVAL;
5930 
5931 	new_filter = kvm_alloc_msr_filter(default_allow);
5932 	if (!new_filter)
5933 		return -ENOMEM;
5934 
5935 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5936 		r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
5937 		if (r) {
5938 			kvm_free_msr_filter(new_filter);
5939 			return r;
5940 		}
5941 	}
5942 
5943 	mutex_lock(&kvm->lock);
5944 
5945 	/* The per-VM filter is protected by kvm->lock... */
5946 	old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
5947 
5948 	rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
5949 	synchronize_srcu(&kvm->srcu);
5950 
5951 	kvm_free_msr_filter(old_filter);
5952 
5953 	kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5954 	mutex_unlock(&kvm->lock);
5955 
5956 	return 0;
5957 }
5958 
5959 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
5960 static int kvm_arch_suspend_notifier(struct kvm *kvm)
5961 {
5962 	struct kvm_vcpu *vcpu;
5963 	int i, ret = 0;
5964 
5965 	mutex_lock(&kvm->lock);
5966 	kvm_for_each_vcpu(i, vcpu, kvm) {
5967 		if (!vcpu->arch.pv_time_enabled)
5968 			continue;
5969 
5970 		ret = kvm_set_guest_paused(vcpu);
5971 		if (ret) {
5972 			kvm_err("Failed to pause guest VCPU%d: %d\n",
5973 				vcpu->vcpu_id, ret);
5974 			break;
5975 		}
5976 	}
5977 	mutex_unlock(&kvm->lock);
5978 
5979 	return ret ? NOTIFY_BAD : NOTIFY_DONE;
5980 }
5981 
5982 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state)
5983 {
5984 	switch (state) {
5985 	case PM_HIBERNATION_PREPARE:
5986 	case PM_SUSPEND_PREPARE:
5987 		return kvm_arch_suspend_notifier(kvm);
5988 	}
5989 
5990 	return NOTIFY_DONE;
5991 }
5992 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
5993 
5994 static int kvm_vm_ioctl_get_clock(struct kvm *kvm, void __user *argp)
5995 {
5996 	struct kvm_clock_data data = { 0 };
5997 
5998 	get_kvmclock(kvm, &data);
5999 	if (copy_to_user(argp, &data, sizeof(data)))
6000 		return -EFAULT;
6001 
6002 	return 0;
6003 }
6004 
6005 static int kvm_vm_ioctl_set_clock(struct kvm *kvm, void __user *argp)
6006 {
6007 	struct kvm_arch *ka = &kvm->arch;
6008 	struct kvm_clock_data data;
6009 	u64 now_raw_ns;
6010 
6011 	if (copy_from_user(&data, argp, sizeof(data)))
6012 		return -EFAULT;
6013 
6014 	/*
6015 	 * Only KVM_CLOCK_REALTIME is used, but allow passing the
6016 	 * result of KVM_GET_CLOCK back to KVM_SET_CLOCK.
6017 	 */
6018 	if (data.flags & ~KVM_CLOCK_VALID_FLAGS)
6019 		return -EINVAL;
6020 
6021 	kvm_hv_invalidate_tsc_page(kvm);
6022 	kvm_start_pvclock_update(kvm);
6023 	pvclock_update_vm_gtod_copy(kvm);
6024 
6025 	/*
6026 	 * This pairs with kvm_guest_time_update(): when masterclock is
6027 	 * in use, we use master_kernel_ns + kvmclock_offset to set
6028 	 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
6029 	 * is slightly ahead) here we risk going negative on unsigned
6030 	 * 'system_time' when 'data.clock' is very small.
6031 	 */
6032 	if (data.flags & KVM_CLOCK_REALTIME) {
6033 		u64 now_real_ns = ktime_get_real_ns();
6034 
6035 		/*
6036 		 * Avoid stepping the kvmclock backwards.
6037 		 */
6038 		if (now_real_ns > data.realtime)
6039 			data.clock += now_real_ns - data.realtime;
6040 	}
6041 
6042 	if (ka->use_master_clock)
6043 		now_raw_ns = ka->master_kernel_ns;
6044 	else
6045 		now_raw_ns = get_kvmclock_base_ns();
6046 	ka->kvmclock_offset = data.clock - now_raw_ns;
6047 	kvm_end_pvclock_update(kvm);
6048 	return 0;
6049 }
6050 
6051 long kvm_arch_vm_ioctl(struct file *filp,
6052 		       unsigned int ioctl, unsigned long arg)
6053 {
6054 	struct kvm *kvm = filp->private_data;
6055 	void __user *argp = (void __user *)arg;
6056 	int r = -ENOTTY;
6057 	/*
6058 	 * This union makes it completely explicit to gcc-3.x
6059 	 * that these two variables' stack usage should be
6060 	 * combined, not added together.
6061 	 */
6062 	union {
6063 		struct kvm_pit_state ps;
6064 		struct kvm_pit_state2 ps2;
6065 		struct kvm_pit_config pit_config;
6066 	} u;
6067 
6068 	switch (ioctl) {
6069 	case KVM_SET_TSS_ADDR:
6070 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
6071 		break;
6072 	case KVM_SET_IDENTITY_MAP_ADDR: {
6073 		u64 ident_addr;
6074 
6075 		mutex_lock(&kvm->lock);
6076 		r = -EINVAL;
6077 		if (kvm->created_vcpus)
6078 			goto set_identity_unlock;
6079 		r = -EFAULT;
6080 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
6081 			goto set_identity_unlock;
6082 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
6083 set_identity_unlock:
6084 		mutex_unlock(&kvm->lock);
6085 		break;
6086 	}
6087 	case KVM_SET_NR_MMU_PAGES:
6088 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
6089 		break;
6090 	case KVM_GET_NR_MMU_PAGES:
6091 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
6092 		break;
6093 	case KVM_CREATE_IRQCHIP: {
6094 		mutex_lock(&kvm->lock);
6095 
6096 		r = -EEXIST;
6097 		if (irqchip_in_kernel(kvm))
6098 			goto create_irqchip_unlock;
6099 
6100 		r = -EINVAL;
6101 		if (kvm->created_vcpus)
6102 			goto create_irqchip_unlock;
6103 
6104 		r = kvm_pic_init(kvm);
6105 		if (r)
6106 			goto create_irqchip_unlock;
6107 
6108 		r = kvm_ioapic_init(kvm);
6109 		if (r) {
6110 			kvm_pic_destroy(kvm);
6111 			goto create_irqchip_unlock;
6112 		}
6113 
6114 		r = kvm_setup_default_irq_routing(kvm);
6115 		if (r) {
6116 			kvm_ioapic_destroy(kvm);
6117 			kvm_pic_destroy(kvm);
6118 			goto create_irqchip_unlock;
6119 		}
6120 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
6121 		smp_wmb();
6122 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
6123 	create_irqchip_unlock:
6124 		mutex_unlock(&kvm->lock);
6125 		break;
6126 	}
6127 	case KVM_CREATE_PIT:
6128 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
6129 		goto create_pit;
6130 	case KVM_CREATE_PIT2:
6131 		r = -EFAULT;
6132 		if (copy_from_user(&u.pit_config, argp,
6133 				   sizeof(struct kvm_pit_config)))
6134 			goto out;
6135 	create_pit:
6136 		mutex_lock(&kvm->lock);
6137 		r = -EEXIST;
6138 		if (kvm->arch.vpit)
6139 			goto create_pit_unlock;
6140 		r = -ENOMEM;
6141 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
6142 		if (kvm->arch.vpit)
6143 			r = 0;
6144 	create_pit_unlock:
6145 		mutex_unlock(&kvm->lock);
6146 		break;
6147 	case KVM_GET_IRQCHIP: {
6148 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6149 		struct kvm_irqchip *chip;
6150 
6151 		chip = memdup_user(argp, sizeof(*chip));
6152 		if (IS_ERR(chip)) {
6153 			r = PTR_ERR(chip);
6154 			goto out;
6155 		}
6156 
6157 		r = -ENXIO;
6158 		if (!irqchip_kernel(kvm))
6159 			goto get_irqchip_out;
6160 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
6161 		if (r)
6162 			goto get_irqchip_out;
6163 		r = -EFAULT;
6164 		if (copy_to_user(argp, chip, sizeof(*chip)))
6165 			goto get_irqchip_out;
6166 		r = 0;
6167 	get_irqchip_out:
6168 		kfree(chip);
6169 		break;
6170 	}
6171 	case KVM_SET_IRQCHIP: {
6172 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6173 		struct kvm_irqchip *chip;
6174 
6175 		chip = memdup_user(argp, sizeof(*chip));
6176 		if (IS_ERR(chip)) {
6177 			r = PTR_ERR(chip);
6178 			goto out;
6179 		}
6180 
6181 		r = -ENXIO;
6182 		if (!irqchip_kernel(kvm))
6183 			goto set_irqchip_out;
6184 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
6185 	set_irqchip_out:
6186 		kfree(chip);
6187 		break;
6188 	}
6189 	case KVM_GET_PIT: {
6190 		r = -EFAULT;
6191 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
6192 			goto out;
6193 		r = -ENXIO;
6194 		if (!kvm->arch.vpit)
6195 			goto out;
6196 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
6197 		if (r)
6198 			goto out;
6199 		r = -EFAULT;
6200 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
6201 			goto out;
6202 		r = 0;
6203 		break;
6204 	}
6205 	case KVM_SET_PIT: {
6206 		r = -EFAULT;
6207 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
6208 			goto out;
6209 		mutex_lock(&kvm->lock);
6210 		r = -ENXIO;
6211 		if (!kvm->arch.vpit)
6212 			goto set_pit_out;
6213 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
6214 set_pit_out:
6215 		mutex_unlock(&kvm->lock);
6216 		break;
6217 	}
6218 	case KVM_GET_PIT2: {
6219 		r = -ENXIO;
6220 		if (!kvm->arch.vpit)
6221 			goto out;
6222 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
6223 		if (r)
6224 			goto out;
6225 		r = -EFAULT;
6226 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
6227 			goto out;
6228 		r = 0;
6229 		break;
6230 	}
6231 	case KVM_SET_PIT2: {
6232 		r = -EFAULT;
6233 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
6234 			goto out;
6235 		mutex_lock(&kvm->lock);
6236 		r = -ENXIO;
6237 		if (!kvm->arch.vpit)
6238 			goto set_pit2_out;
6239 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
6240 set_pit2_out:
6241 		mutex_unlock(&kvm->lock);
6242 		break;
6243 	}
6244 	case KVM_REINJECT_CONTROL: {
6245 		struct kvm_reinject_control control;
6246 		r =  -EFAULT;
6247 		if (copy_from_user(&control, argp, sizeof(control)))
6248 			goto out;
6249 		r = -ENXIO;
6250 		if (!kvm->arch.vpit)
6251 			goto out;
6252 		r = kvm_vm_ioctl_reinject(kvm, &control);
6253 		break;
6254 	}
6255 	case KVM_SET_BOOT_CPU_ID:
6256 		r = 0;
6257 		mutex_lock(&kvm->lock);
6258 		if (kvm->created_vcpus)
6259 			r = -EBUSY;
6260 		else
6261 			kvm->arch.bsp_vcpu_id = arg;
6262 		mutex_unlock(&kvm->lock);
6263 		break;
6264 #ifdef CONFIG_KVM_XEN
6265 	case KVM_XEN_HVM_CONFIG: {
6266 		struct kvm_xen_hvm_config xhc;
6267 		r = -EFAULT;
6268 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
6269 			goto out;
6270 		r = kvm_xen_hvm_config(kvm, &xhc);
6271 		break;
6272 	}
6273 	case KVM_XEN_HVM_GET_ATTR: {
6274 		struct kvm_xen_hvm_attr xha;
6275 
6276 		r = -EFAULT;
6277 		if (copy_from_user(&xha, argp, sizeof(xha)))
6278 			goto out;
6279 		r = kvm_xen_hvm_get_attr(kvm, &xha);
6280 		if (!r && copy_to_user(argp, &xha, sizeof(xha)))
6281 			r = -EFAULT;
6282 		break;
6283 	}
6284 	case KVM_XEN_HVM_SET_ATTR: {
6285 		struct kvm_xen_hvm_attr xha;
6286 
6287 		r = -EFAULT;
6288 		if (copy_from_user(&xha, argp, sizeof(xha)))
6289 			goto out;
6290 		r = kvm_xen_hvm_set_attr(kvm, &xha);
6291 		break;
6292 	}
6293 #endif
6294 	case KVM_SET_CLOCK:
6295 		r = kvm_vm_ioctl_set_clock(kvm, argp);
6296 		break;
6297 	case KVM_GET_CLOCK:
6298 		r = kvm_vm_ioctl_get_clock(kvm, argp);
6299 		break;
6300 	case KVM_MEMORY_ENCRYPT_OP: {
6301 		r = -ENOTTY;
6302 		if (kvm_x86_ops.mem_enc_op)
6303 			r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
6304 		break;
6305 	}
6306 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
6307 		struct kvm_enc_region region;
6308 
6309 		r = -EFAULT;
6310 		if (copy_from_user(&region, argp, sizeof(region)))
6311 			goto out;
6312 
6313 		r = -ENOTTY;
6314 		if (kvm_x86_ops.mem_enc_reg_region)
6315 			r = static_call(kvm_x86_mem_enc_reg_region)(kvm, &region);
6316 		break;
6317 	}
6318 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
6319 		struct kvm_enc_region region;
6320 
6321 		r = -EFAULT;
6322 		if (copy_from_user(&region, argp, sizeof(region)))
6323 			goto out;
6324 
6325 		r = -ENOTTY;
6326 		if (kvm_x86_ops.mem_enc_unreg_region)
6327 			r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, &region);
6328 		break;
6329 	}
6330 	case KVM_HYPERV_EVENTFD: {
6331 		struct kvm_hyperv_eventfd hvevfd;
6332 
6333 		r = -EFAULT;
6334 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
6335 			goto out;
6336 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
6337 		break;
6338 	}
6339 	case KVM_SET_PMU_EVENT_FILTER:
6340 		r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
6341 		break;
6342 	case KVM_X86_SET_MSR_FILTER:
6343 		r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
6344 		break;
6345 	default:
6346 		r = -ENOTTY;
6347 	}
6348 out:
6349 	return r;
6350 }
6351 
6352 static void kvm_init_msr_list(void)
6353 {
6354 	struct x86_pmu_capability x86_pmu;
6355 	u32 dummy[2];
6356 	unsigned i;
6357 
6358 	BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
6359 			 "Please update the fixed PMCs in msrs_to_saved_all[]");
6360 
6361 	perf_get_x86_pmu_capability(&x86_pmu);
6362 
6363 	num_msrs_to_save = 0;
6364 	num_emulated_msrs = 0;
6365 	num_msr_based_features = 0;
6366 
6367 	for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
6368 		if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
6369 			continue;
6370 
6371 		/*
6372 		 * Even MSRs that are valid in the host may not be exposed
6373 		 * to the guests in some cases.
6374 		 */
6375 		switch (msrs_to_save_all[i]) {
6376 		case MSR_IA32_BNDCFGS:
6377 			if (!kvm_mpx_supported())
6378 				continue;
6379 			break;
6380 		case MSR_TSC_AUX:
6381 			if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
6382 			    !kvm_cpu_cap_has(X86_FEATURE_RDPID))
6383 				continue;
6384 			break;
6385 		case MSR_IA32_UMWAIT_CONTROL:
6386 			if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
6387 				continue;
6388 			break;
6389 		case MSR_IA32_RTIT_CTL:
6390 		case MSR_IA32_RTIT_STATUS:
6391 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
6392 				continue;
6393 			break;
6394 		case MSR_IA32_RTIT_CR3_MATCH:
6395 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6396 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
6397 				continue;
6398 			break;
6399 		case MSR_IA32_RTIT_OUTPUT_BASE:
6400 		case MSR_IA32_RTIT_OUTPUT_MASK:
6401 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6402 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
6403 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
6404 				continue;
6405 			break;
6406 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
6407 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6408 				msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
6409 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
6410 				continue;
6411 			break;
6412 		case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
6413 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
6414 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6415 				continue;
6416 			break;
6417 		case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
6418 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
6419 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6420 				continue;
6421 			break;
6422 		default:
6423 			break;
6424 		}
6425 
6426 		msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
6427 	}
6428 
6429 	for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
6430 		if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
6431 			continue;
6432 
6433 		emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
6434 	}
6435 
6436 	for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
6437 		struct kvm_msr_entry msr;
6438 
6439 		msr.index = msr_based_features_all[i];
6440 		if (kvm_get_msr_feature(&msr))
6441 			continue;
6442 
6443 		msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6444 	}
6445 }
6446 
6447 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6448 			   const void *v)
6449 {
6450 	int handled = 0;
6451 	int n;
6452 
6453 	do {
6454 		n = min(len, 8);
6455 		if (!(lapic_in_kernel(vcpu) &&
6456 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6457 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6458 			break;
6459 		handled += n;
6460 		addr += n;
6461 		len -= n;
6462 		v += n;
6463 	} while (len);
6464 
6465 	return handled;
6466 }
6467 
6468 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6469 {
6470 	int handled = 0;
6471 	int n;
6472 
6473 	do {
6474 		n = min(len, 8);
6475 		if (!(lapic_in_kernel(vcpu) &&
6476 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6477 					 addr, n, v))
6478 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6479 			break;
6480 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6481 		handled += n;
6482 		addr += n;
6483 		len -= n;
6484 		v += n;
6485 	} while (len);
6486 
6487 	return handled;
6488 }
6489 
6490 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6491 			struct kvm_segment *var, int seg)
6492 {
6493 	static_call(kvm_x86_set_segment)(vcpu, var, seg);
6494 }
6495 
6496 void kvm_get_segment(struct kvm_vcpu *vcpu,
6497 		     struct kvm_segment *var, int seg)
6498 {
6499 	static_call(kvm_x86_get_segment)(vcpu, var, seg);
6500 }
6501 
6502 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6503 			   struct x86_exception *exception)
6504 {
6505 	gpa_t t_gpa;
6506 
6507 	BUG_ON(!mmu_is_nested(vcpu));
6508 
6509 	/* NPT walks are always user-walks */
6510 	access |= PFERR_USER_MASK;
6511 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
6512 
6513 	return t_gpa;
6514 }
6515 
6516 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6517 			      struct x86_exception *exception)
6518 {
6519 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6520 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6521 }
6522 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6523 
6524  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6525 				struct x86_exception *exception)
6526 {
6527 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6528 	access |= PFERR_FETCH_MASK;
6529 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6530 }
6531 
6532 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6533 			       struct x86_exception *exception)
6534 {
6535 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6536 	access |= PFERR_WRITE_MASK;
6537 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6538 }
6539 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6540 
6541 /* uses this to access any guest's mapped memory without checking CPL */
6542 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6543 				struct x86_exception *exception)
6544 {
6545 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
6546 }
6547 
6548 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6549 				      struct kvm_vcpu *vcpu, u32 access,
6550 				      struct x86_exception *exception)
6551 {
6552 	void *data = val;
6553 	int r = X86EMUL_CONTINUE;
6554 
6555 	while (bytes) {
6556 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
6557 							    exception);
6558 		unsigned offset = addr & (PAGE_SIZE-1);
6559 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6560 		int ret;
6561 
6562 		if (gpa == UNMAPPED_GVA)
6563 			return X86EMUL_PROPAGATE_FAULT;
6564 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6565 					       offset, toread);
6566 		if (ret < 0) {
6567 			r = X86EMUL_IO_NEEDED;
6568 			goto out;
6569 		}
6570 
6571 		bytes -= toread;
6572 		data += toread;
6573 		addr += toread;
6574 	}
6575 out:
6576 	return r;
6577 }
6578 
6579 /* used for instruction fetching */
6580 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6581 				gva_t addr, void *val, unsigned int bytes,
6582 				struct x86_exception *exception)
6583 {
6584 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6585 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6586 	unsigned offset;
6587 	int ret;
6588 
6589 	/* Inline kvm_read_guest_virt_helper for speed.  */
6590 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6591 						    exception);
6592 	if (unlikely(gpa == UNMAPPED_GVA))
6593 		return X86EMUL_PROPAGATE_FAULT;
6594 
6595 	offset = addr & (PAGE_SIZE-1);
6596 	if (WARN_ON(offset + bytes > PAGE_SIZE))
6597 		bytes = (unsigned)PAGE_SIZE - offset;
6598 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6599 				       offset, bytes);
6600 	if (unlikely(ret < 0))
6601 		return X86EMUL_IO_NEEDED;
6602 
6603 	return X86EMUL_CONTINUE;
6604 }
6605 
6606 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6607 			       gva_t addr, void *val, unsigned int bytes,
6608 			       struct x86_exception *exception)
6609 {
6610 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6611 
6612 	/*
6613 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6614 	 * is returned, but our callers are not ready for that and they blindly
6615 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
6616 	 * uninitialized kernel stack memory into cr2 and error code.
6617 	 */
6618 	memset(exception, 0, sizeof(*exception));
6619 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6620 					  exception);
6621 }
6622 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6623 
6624 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6625 			     gva_t addr, void *val, unsigned int bytes,
6626 			     struct x86_exception *exception, bool system)
6627 {
6628 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6629 	u32 access = 0;
6630 
6631 	if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6632 		access |= PFERR_USER_MASK;
6633 
6634 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6635 }
6636 
6637 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6638 		unsigned long addr, void *val, unsigned int bytes)
6639 {
6640 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6641 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6642 
6643 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6644 }
6645 
6646 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6647 				      struct kvm_vcpu *vcpu, u32 access,
6648 				      struct x86_exception *exception)
6649 {
6650 	void *data = val;
6651 	int r = X86EMUL_CONTINUE;
6652 
6653 	while (bytes) {
6654 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6655 							     access,
6656 							     exception);
6657 		unsigned offset = addr & (PAGE_SIZE-1);
6658 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6659 		int ret;
6660 
6661 		if (gpa == UNMAPPED_GVA)
6662 			return X86EMUL_PROPAGATE_FAULT;
6663 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6664 		if (ret < 0) {
6665 			r = X86EMUL_IO_NEEDED;
6666 			goto out;
6667 		}
6668 
6669 		bytes -= towrite;
6670 		data += towrite;
6671 		addr += towrite;
6672 	}
6673 out:
6674 	return r;
6675 }
6676 
6677 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6678 			      unsigned int bytes, struct x86_exception *exception,
6679 			      bool system)
6680 {
6681 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6682 	u32 access = PFERR_WRITE_MASK;
6683 
6684 	if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6685 		access |= PFERR_USER_MASK;
6686 
6687 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6688 					   access, exception);
6689 }
6690 
6691 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6692 				unsigned int bytes, struct x86_exception *exception)
6693 {
6694 	/* kvm_write_guest_virt_system can pull in tons of pages. */
6695 	vcpu->arch.l1tf_flush_l1d = true;
6696 
6697 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6698 					   PFERR_WRITE_MASK, exception);
6699 }
6700 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6701 
6702 int handle_ud(struct kvm_vcpu *vcpu)
6703 {
6704 	static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6705 	int emul_type = EMULTYPE_TRAP_UD;
6706 	char sig[5]; /* ud2; .ascii "kvm" */
6707 	struct x86_exception e;
6708 
6709 	if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6710 		return 1;
6711 
6712 	if (force_emulation_prefix &&
6713 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6714 				sig, sizeof(sig), &e) == 0 &&
6715 	    memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6716 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6717 		emul_type = EMULTYPE_TRAP_UD_FORCED;
6718 	}
6719 
6720 	return kvm_emulate_instruction(vcpu, emul_type);
6721 }
6722 EXPORT_SYMBOL_GPL(handle_ud);
6723 
6724 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6725 			    gpa_t gpa, bool write)
6726 {
6727 	/* For APIC access vmexit */
6728 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6729 		return 1;
6730 
6731 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6732 		trace_vcpu_match_mmio(gva, gpa, write, true);
6733 		return 1;
6734 	}
6735 
6736 	return 0;
6737 }
6738 
6739 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6740 				gpa_t *gpa, struct x86_exception *exception,
6741 				bool write)
6742 {
6743 	u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6744 		| (write ? PFERR_WRITE_MASK : 0);
6745 
6746 	/*
6747 	 * currently PKRU is only applied to ept enabled guest so
6748 	 * there is no pkey in EPT page table for L1 guest or EPT
6749 	 * shadow page table for L2 guest.
6750 	 */
6751 	if (vcpu_match_mmio_gva(vcpu, gva) && (!is_paging(vcpu) ||
6752 	    !permission_fault(vcpu, vcpu->arch.walk_mmu,
6753 			      vcpu->arch.mmio_access, 0, access))) {
6754 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6755 					(gva & (PAGE_SIZE - 1));
6756 		trace_vcpu_match_mmio(gva, *gpa, write, false);
6757 		return 1;
6758 	}
6759 
6760 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6761 
6762 	if (*gpa == UNMAPPED_GVA)
6763 		return -1;
6764 
6765 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6766 }
6767 
6768 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6769 			const void *val, int bytes)
6770 {
6771 	int ret;
6772 
6773 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6774 	if (ret < 0)
6775 		return 0;
6776 	kvm_page_track_write(vcpu, gpa, val, bytes);
6777 	return 1;
6778 }
6779 
6780 struct read_write_emulator_ops {
6781 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6782 				  int bytes);
6783 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6784 				  void *val, int bytes);
6785 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6786 			       int bytes, void *val);
6787 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6788 				    void *val, int bytes);
6789 	bool write;
6790 };
6791 
6792 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6793 {
6794 	if (vcpu->mmio_read_completed) {
6795 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6796 			       vcpu->mmio_fragments[0].gpa, val);
6797 		vcpu->mmio_read_completed = 0;
6798 		return 1;
6799 	}
6800 
6801 	return 0;
6802 }
6803 
6804 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6805 			void *val, int bytes)
6806 {
6807 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6808 }
6809 
6810 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6811 			 void *val, int bytes)
6812 {
6813 	return emulator_write_phys(vcpu, gpa, val, bytes);
6814 }
6815 
6816 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6817 {
6818 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6819 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
6820 }
6821 
6822 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6823 			  void *val, int bytes)
6824 {
6825 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6826 	return X86EMUL_IO_NEEDED;
6827 }
6828 
6829 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6830 			   void *val, int bytes)
6831 {
6832 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6833 
6834 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6835 	return X86EMUL_CONTINUE;
6836 }
6837 
6838 static const struct read_write_emulator_ops read_emultor = {
6839 	.read_write_prepare = read_prepare,
6840 	.read_write_emulate = read_emulate,
6841 	.read_write_mmio = vcpu_mmio_read,
6842 	.read_write_exit_mmio = read_exit_mmio,
6843 };
6844 
6845 static const struct read_write_emulator_ops write_emultor = {
6846 	.read_write_emulate = write_emulate,
6847 	.read_write_mmio = write_mmio,
6848 	.read_write_exit_mmio = write_exit_mmio,
6849 	.write = true,
6850 };
6851 
6852 static int emulator_read_write_onepage(unsigned long addr, void *val,
6853 				       unsigned int bytes,
6854 				       struct x86_exception *exception,
6855 				       struct kvm_vcpu *vcpu,
6856 				       const struct read_write_emulator_ops *ops)
6857 {
6858 	gpa_t gpa;
6859 	int handled, ret;
6860 	bool write = ops->write;
6861 	struct kvm_mmio_fragment *frag;
6862 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6863 
6864 	/*
6865 	 * If the exit was due to a NPF we may already have a GPA.
6866 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6867 	 * Note, this cannot be used on string operations since string
6868 	 * operation using rep will only have the initial GPA from the NPF
6869 	 * occurred.
6870 	 */
6871 	if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6872 	    (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6873 		gpa = ctxt->gpa_val;
6874 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6875 	} else {
6876 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6877 		if (ret < 0)
6878 			return X86EMUL_PROPAGATE_FAULT;
6879 	}
6880 
6881 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6882 		return X86EMUL_CONTINUE;
6883 
6884 	/*
6885 	 * Is this MMIO handled locally?
6886 	 */
6887 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6888 	if (handled == bytes)
6889 		return X86EMUL_CONTINUE;
6890 
6891 	gpa += handled;
6892 	bytes -= handled;
6893 	val += handled;
6894 
6895 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6896 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6897 	frag->gpa = gpa;
6898 	frag->data = val;
6899 	frag->len = bytes;
6900 	return X86EMUL_CONTINUE;
6901 }
6902 
6903 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6904 			unsigned long addr,
6905 			void *val, unsigned int bytes,
6906 			struct x86_exception *exception,
6907 			const struct read_write_emulator_ops *ops)
6908 {
6909 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6910 	gpa_t gpa;
6911 	int rc;
6912 
6913 	if (ops->read_write_prepare &&
6914 		  ops->read_write_prepare(vcpu, val, bytes))
6915 		return X86EMUL_CONTINUE;
6916 
6917 	vcpu->mmio_nr_fragments = 0;
6918 
6919 	/* Crossing a page boundary? */
6920 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6921 		int now;
6922 
6923 		now = -addr & ~PAGE_MASK;
6924 		rc = emulator_read_write_onepage(addr, val, now, exception,
6925 						 vcpu, ops);
6926 
6927 		if (rc != X86EMUL_CONTINUE)
6928 			return rc;
6929 		addr += now;
6930 		if (ctxt->mode != X86EMUL_MODE_PROT64)
6931 			addr = (u32)addr;
6932 		val += now;
6933 		bytes -= now;
6934 	}
6935 
6936 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
6937 					 vcpu, ops);
6938 	if (rc != X86EMUL_CONTINUE)
6939 		return rc;
6940 
6941 	if (!vcpu->mmio_nr_fragments)
6942 		return rc;
6943 
6944 	gpa = vcpu->mmio_fragments[0].gpa;
6945 
6946 	vcpu->mmio_needed = 1;
6947 	vcpu->mmio_cur_fragment = 0;
6948 
6949 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6950 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6951 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
6952 	vcpu->run->mmio.phys_addr = gpa;
6953 
6954 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6955 }
6956 
6957 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6958 				  unsigned long addr,
6959 				  void *val,
6960 				  unsigned int bytes,
6961 				  struct x86_exception *exception)
6962 {
6963 	return emulator_read_write(ctxt, addr, val, bytes,
6964 				   exception, &read_emultor);
6965 }
6966 
6967 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6968 			    unsigned long addr,
6969 			    const void *val,
6970 			    unsigned int bytes,
6971 			    struct x86_exception *exception)
6972 {
6973 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
6974 				   exception, &write_emultor);
6975 }
6976 
6977 #define CMPXCHG_TYPE(t, ptr, old, new) \
6978 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6979 
6980 #ifdef CONFIG_X86_64
6981 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6982 #else
6983 #  define CMPXCHG64(ptr, old, new) \
6984 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6985 #endif
6986 
6987 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6988 				     unsigned long addr,
6989 				     const void *old,
6990 				     const void *new,
6991 				     unsigned int bytes,
6992 				     struct x86_exception *exception)
6993 {
6994 	struct kvm_host_map map;
6995 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6996 	u64 page_line_mask;
6997 	gpa_t gpa;
6998 	char *kaddr;
6999 	bool exchanged;
7000 
7001 	/* guests cmpxchg8b have to be emulated atomically */
7002 	if (bytes > 8 || (bytes & (bytes - 1)))
7003 		goto emul_write;
7004 
7005 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
7006 
7007 	if (gpa == UNMAPPED_GVA ||
7008 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
7009 		goto emul_write;
7010 
7011 	/*
7012 	 * Emulate the atomic as a straight write to avoid #AC if SLD is
7013 	 * enabled in the host and the access splits a cache line.
7014 	 */
7015 	if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
7016 		page_line_mask = ~(cache_line_size() - 1);
7017 	else
7018 		page_line_mask = PAGE_MASK;
7019 
7020 	if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
7021 		goto emul_write;
7022 
7023 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
7024 		goto emul_write;
7025 
7026 	kaddr = map.hva + offset_in_page(gpa);
7027 
7028 	switch (bytes) {
7029 	case 1:
7030 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
7031 		break;
7032 	case 2:
7033 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
7034 		break;
7035 	case 4:
7036 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
7037 		break;
7038 	case 8:
7039 		exchanged = CMPXCHG64(kaddr, old, new);
7040 		break;
7041 	default:
7042 		BUG();
7043 	}
7044 
7045 	kvm_vcpu_unmap(vcpu, &map, true);
7046 
7047 	if (!exchanged)
7048 		return X86EMUL_CMPXCHG_FAILED;
7049 
7050 	kvm_page_track_write(vcpu, gpa, new, bytes);
7051 
7052 	return X86EMUL_CONTINUE;
7053 
7054 emul_write:
7055 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
7056 
7057 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
7058 }
7059 
7060 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
7061 {
7062 	int r = 0, i;
7063 
7064 	for (i = 0; i < vcpu->arch.pio.count; i++) {
7065 		if (vcpu->arch.pio.in)
7066 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
7067 					    vcpu->arch.pio.size, pd);
7068 		else
7069 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
7070 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
7071 					     pd);
7072 		if (r)
7073 			break;
7074 		pd += vcpu->arch.pio.size;
7075 	}
7076 	return r;
7077 }
7078 
7079 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
7080 			       unsigned short port,
7081 			       unsigned int count, bool in)
7082 {
7083 	vcpu->arch.pio.port = port;
7084 	vcpu->arch.pio.in = in;
7085 	vcpu->arch.pio.count  = count;
7086 	vcpu->arch.pio.size = size;
7087 
7088 	if (!kernel_pio(vcpu, vcpu->arch.pio_data))
7089 		return 1;
7090 
7091 	vcpu->run->exit_reason = KVM_EXIT_IO;
7092 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
7093 	vcpu->run->io.size = size;
7094 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
7095 	vcpu->run->io.count = count;
7096 	vcpu->run->io.port = port;
7097 
7098 	return 0;
7099 }
7100 
7101 static int __emulator_pio_in(struct kvm_vcpu *vcpu, int size,
7102 			     unsigned short port, unsigned int count)
7103 {
7104 	WARN_ON(vcpu->arch.pio.count);
7105 	memset(vcpu->arch.pio_data, 0, size * count);
7106 	return emulator_pio_in_out(vcpu, size, port, count, true);
7107 }
7108 
7109 static void complete_emulator_pio_in(struct kvm_vcpu *vcpu, void *val)
7110 {
7111 	int size = vcpu->arch.pio.size;
7112 	unsigned count = vcpu->arch.pio.count;
7113 	memcpy(val, vcpu->arch.pio_data, size * count);
7114 	trace_kvm_pio(KVM_PIO_IN, vcpu->arch.pio.port, size, count, vcpu->arch.pio_data);
7115 	vcpu->arch.pio.count = 0;
7116 }
7117 
7118 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
7119 			   unsigned short port, void *val, unsigned int count)
7120 {
7121 	if (vcpu->arch.pio.count) {
7122 		/* Complete previous iteration.  */
7123 	} else {
7124 		int r = __emulator_pio_in(vcpu, size, port, count);
7125 		if (!r)
7126 			return r;
7127 
7128 		/* Results already available, fall through.  */
7129 	}
7130 
7131 	WARN_ON(count != vcpu->arch.pio.count);
7132 	complete_emulator_pio_in(vcpu, val);
7133 	return 1;
7134 }
7135 
7136 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
7137 				    int size, unsigned short port, void *val,
7138 				    unsigned int count)
7139 {
7140 	return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
7141 
7142 }
7143 
7144 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
7145 			    unsigned short port, const void *val,
7146 			    unsigned int count)
7147 {
7148 	int ret;
7149 
7150 	memcpy(vcpu->arch.pio_data, val, size * count);
7151 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
7152 	ret = emulator_pio_in_out(vcpu, size, port, count, false);
7153 	if (ret)
7154                 vcpu->arch.pio.count = 0;
7155 
7156         return ret;
7157 }
7158 
7159 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
7160 				     int size, unsigned short port,
7161 				     const void *val, unsigned int count)
7162 {
7163 	return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
7164 }
7165 
7166 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
7167 {
7168 	return static_call(kvm_x86_get_segment_base)(vcpu, seg);
7169 }
7170 
7171 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
7172 {
7173 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
7174 }
7175 
7176 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
7177 {
7178 	if (!need_emulate_wbinvd(vcpu))
7179 		return X86EMUL_CONTINUE;
7180 
7181 	if (static_call(kvm_x86_has_wbinvd_exit)()) {
7182 		int cpu = get_cpu();
7183 
7184 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
7185 		on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
7186 				wbinvd_ipi, NULL, 1);
7187 		put_cpu();
7188 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
7189 	} else
7190 		wbinvd();
7191 	return X86EMUL_CONTINUE;
7192 }
7193 
7194 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
7195 {
7196 	kvm_emulate_wbinvd_noskip(vcpu);
7197 	return kvm_skip_emulated_instruction(vcpu);
7198 }
7199 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
7200 
7201 
7202 
7203 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
7204 {
7205 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
7206 }
7207 
7208 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
7209 			    unsigned long *dest)
7210 {
7211 	kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
7212 }
7213 
7214 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
7215 			   unsigned long value)
7216 {
7217 
7218 	return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
7219 }
7220 
7221 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
7222 {
7223 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
7224 }
7225 
7226 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
7227 {
7228 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7229 	unsigned long value;
7230 
7231 	switch (cr) {
7232 	case 0:
7233 		value = kvm_read_cr0(vcpu);
7234 		break;
7235 	case 2:
7236 		value = vcpu->arch.cr2;
7237 		break;
7238 	case 3:
7239 		value = kvm_read_cr3(vcpu);
7240 		break;
7241 	case 4:
7242 		value = kvm_read_cr4(vcpu);
7243 		break;
7244 	case 8:
7245 		value = kvm_get_cr8(vcpu);
7246 		break;
7247 	default:
7248 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
7249 		return 0;
7250 	}
7251 
7252 	return value;
7253 }
7254 
7255 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
7256 {
7257 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7258 	int res = 0;
7259 
7260 	switch (cr) {
7261 	case 0:
7262 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
7263 		break;
7264 	case 2:
7265 		vcpu->arch.cr2 = val;
7266 		break;
7267 	case 3:
7268 		res = kvm_set_cr3(vcpu, val);
7269 		break;
7270 	case 4:
7271 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
7272 		break;
7273 	case 8:
7274 		res = kvm_set_cr8(vcpu, val);
7275 		break;
7276 	default:
7277 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
7278 		res = -1;
7279 	}
7280 
7281 	return res;
7282 }
7283 
7284 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
7285 {
7286 	return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
7287 }
7288 
7289 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7290 {
7291 	static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
7292 }
7293 
7294 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7295 {
7296 	static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
7297 }
7298 
7299 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7300 {
7301 	static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
7302 }
7303 
7304 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7305 {
7306 	static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
7307 }
7308 
7309 static unsigned long emulator_get_cached_segment_base(
7310 	struct x86_emulate_ctxt *ctxt, int seg)
7311 {
7312 	return get_segment_base(emul_to_vcpu(ctxt), seg);
7313 }
7314 
7315 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
7316 				 struct desc_struct *desc, u32 *base3,
7317 				 int seg)
7318 {
7319 	struct kvm_segment var;
7320 
7321 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
7322 	*selector = var.selector;
7323 
7324 	if (var.unusable) {
7325 		memset(desc, 0, sizeof(*desc));
7326 		if (base3)
7327 			*base3 = 0;
7328 		return false;
7329 	}
7330 
7331 	if (var.g)
7332 		var.limit >>= 12;
7333 	set_desc_limit(desc, var.limit);
7334 	set_desc_base(desc, (unsigned long)var.base);
7335 #ifdef CONFIG_X86_64
7336 	if (base3)
7337 		*base3 = var.base >> 32;
7338 #endif
7339 	desc->type = var.type;
7340 	desc->s = var.s;
7341 	desc->dpl = var.dpl;
7342 	desc->p = var.present;
7343 	desc->avl = var.avl;
7344 	desc->l = var.l;
7345 	desc->d = var.db;
7346 	desc->g = var.g;
7347 
7348 	return true;
7349 }
7350 
7351 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
7352 				 struct desc_struct *desc, u32 base3,
7353 				 int seg)
7354 {
7355 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7356 	struct kvm_segment var;
7357 
7358 	var.selector = selector;
7359 	var.base = get_desc_base(desc);
7360 #ifdef CONFIG_X86_64
7361 	var.base |= ((u64)base3) << 32;
7362 #endif
7363 	var.limit = get_desc_limit(desc);
7364 	if (desc->g)
7365 		var.limit = (var.limit << 12) | 0xfff;
7366 	var.type = desc->type;
7367 	var.dpl = desc->dpl;
7368 	var.db = desc->d;
7369 	var.s = desc->s;
7370 	var.l = desc->l;
7371 	var.g = desc->g;
7372 	var.avl = desc->avl;
7373 	var.present = desc->p;
7374 	var.unusable = !var.present;
7375 	var.padding = 0;
7376 
7377 	kvm_set_segment(vcpu, &var, seg);
7378 	return;
7379 }
7380 
7381 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
7382 			    u32 msr_index, u64 *pdata)
7383 {
7384 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7385 	int r;
7386 
7387 	r = kvm_get_msr(vcpu, msr_index, pdata);
7388 
7389 	if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
7390 		/* Bounce to user space */
7391 		return X86EMUL_IO_NEEDED;
7392 	}
7393 
7394 	return r;
7395 }
7396 
7397 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
7398 			    u32 msr_index, u64 data)
7399 {
7400 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7401 	int r;
7402 
7403 	r = kvm_set_msr(vcpu, msr_index, data);
7404 
7405 	if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
7406 		/* Bounce to user space */
7407 		return X86EMUL_IO_NEEDED;
7408 	}
7409 
7410 	return r;
7411 }
7412 
7413 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
7414 {
7415 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7416 
7417 	return vcpu->arch.smbase;
7418 }
7419 
7420 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
7421 {
7422 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7423 
7424 	vcpu->arch.smbase = smbase;
7425 }
7426 
7427 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
7428 			      u32 pmc)
7429 {
7430 	if (kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc))
7431 		return 0;
7432 	return -EINVAL;
7433 }
7434 
7435 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
7436 			     u32 pmc, u64 *pdata)
7437 {
7438 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
7439 }
7440 
7441 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
7442 {
7443 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
7444 }
7445 
7446 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
7447 			      struct x86_instruction_info *info,
7448 			      enum x86_intercept_stage stage)
7449 {
7450 	return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
7451 					    &ctxt->exception);
7452 }
7453 
7454 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
7455 			      u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
7456 			      bool exact_only)
7457 {
7458 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7459 }
7460 
7461 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7462 {
7463 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7464 }
7465 
7466 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7467 {
7468 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7469 }
7470 
7471 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7472 {
7473 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7474 }
7475 
7476 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7477 {
7478 	return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7479 }
7480 
7481 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7482 {
7483 	kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7484 }
7485 
7486 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7487 {
7488 	static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7489 }
7490 
7491 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7492 {
7493 	return emul_to_vcpu(ctxt)->arch.hflags;
7494 }
7495 
7496 static void emulator_exiting_smm(struct x86_emulate_ctxt *ctxt)
7497 {
7498 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7499 
7500 	kvm_smm_changed(vcpu, false);
7501 }
7502 
7503 static int emulator_leave_smm(struct x86_emulate_ctxt *ctxt,
7504 				  const char *smstate)
7505 {
7506 	return static_call(kvm_x86_leave_smm)(emul_to_vcpu(ctxt), smstate);
7507 }
7508 
7509 static void emulator_triple_fault(struct x86_emulate_ctxt *ctxt)
7510 {
7511 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, emul_to_vcpu(ctxt));
7512 }
7513 
7514 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7515 {
7516 	return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7517 }
7518 
7519 static const struct x86_emulate_ops emulate_ops = {
7520 	.read_gpr            = emulator_read_gpr,
7521 	.write_gpr           = emulator_write_gpr,
7522 	.read_std            = emulator_read_std,
7523 	.write_std           = emulator_write_std,
7524 	.read_phys           = kvm_read_guest_phys_system,
7525 	.fetch               = kvm_fetch_guest_virt,
7526 	.read_emulated       = emulator_read_emulated,
7527 	.write_emulated      = emulator_write_emulated,
7528 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
7529 	.invlpg              = emulator_invlpg,
7530 	.pio_in_emulated     = emulator_pio_in_emulated,
7531 	.pio_out_emulated    = emulator_pio_out_emulated,
7532 	.get_segment         = emulator_get_segment,
7533 	.set_segment         = emulator_set_segment,
7534 	.get_cached_segment_base = emulator_get_cached_segment_base,
7535 	.get_gdt             = emulator_get_gdt,
7536 	.get_idt	     = emulator_get_idt,
7537 	.set_gdt             = emulator_set_gdt,
7538 	.set_idt	     = emulator_set_idt,
7539 	.get_cr              = emulator_get_cr,
7540 	.set_cr              = emulator_set_cr,
7541 	.cpl                 = emulator_get_cpl,
7542 	.get_dr              = emulator_get_dr,
7543 	.set_dr              = emulator_set_dr,
7544 	.get_smbase          = emulator_get_smbase,
7545 	.set_smbase          = emulator_set_smbase,
7546 	.set_msr             = emulator_set_msr,
7547 	.get_msr             = emulator_get_msr,
7548 	.check_pmc	     = emulator_check_pmc,
7549 	.read_pmc            = emulator_read_pmc,
7550 	.halt                = emulator_halt,
7551 	.wbinvd              = emulator_wbinvd,
7552 	.fix_hypercall       = emulator_fix_hypercall,
7553 	.intercept           = emulator_intercept,
7554 	.get_cpuid           = emulator_get_cpuid,
7555 	.guest_has_long_mode = emulator_guest_has_long_mode,
7556 	.guest_has_movbe     = emulator_guest_has_movbe,
7557 	.guest_has_fxsr      = emulator_guest_has_fxsr,
7558 	.set_nmi_mask        = emulator_set_nmi_mask,
7559 	.get_hflags          = emulator_get_hflags,
7560 	.exiting_smm         = emulator_exiting_smm,
7561 	.leave_smm           = emulator_leave_smm,
7562 	.triple_fault        = emulator_triple_fault,
7563 	.set_xcr             = emulator_set_xcr,
7564 };
7565 
7566 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7567 {
7568 	u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7569 	/*
7570 	 * an sti; sti; sequence only disable interrupts for the first
7571 	 * instruction. So, if the last instruction, be it emulated or
7572 	 * not, left the system with the INT_STI flag enabled, it
7573 	 * means that the last instruction is an sti. We should not
7574 	 * leave the flag on in this case. The same goes for mov ss
7575 	 */
7576 	if (int_shadow & mask)
7577 		mask = 0;
7578 	if (unlikely(int_shadow || mask)) {
7579 		static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7580 		if (!mask)
7581 			kvm_make_request(KVM_REQ_EVENT, vcpu);
7582 	}
7583 }
7584 
7585 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7586 {
7587 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7588 	if (ctxt->exception.vector == PF_VECTOR)
7589 		return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7590 
7591 	if (ctxt->exception.error_code_valid)
7592 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7593 				      ctxt->exception.error_code);
7594 	else
7595 		kvm_queue_exception(vcpu, ctxt->exception.vector);
7596 	return false;
7597 }
7598 
7599 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7600 {
7601 	struct x86_emulate_ctxt *ctxt;
7602 
7603 	ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7604 	if (!ctxt) {
7605 		pr_err("kvm: failed to allocate vcpu's emulator\n");
7606 		return NULL;
7607 	}
7608 
7609 	ctxt->vcpu = vcpu;
7610 	ctxt->ops = &emulate_ops;
7611 	vcpu->arch.emulate_ctxt = ctxt;
7612 
7613 	return ctxt;
7614 }
7615 
7616 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7617 {
7618 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7619 	int cs_db, cs_l;
7620 
7621 	static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7622 
7623 	ctxt->gpa_available = false;
7624 	ctxt->eflags = kvm_get_rflags(vcpu);
7625 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7626 
7627 	ctxt->eip = kvm_rip_read(vcpu);
7628 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
7629 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
7630 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
7631 		     cs_db				? X86EMUL_MODE_PROT32 :
7632 							  X86EMUL_MODE_PROT16;
7633 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7634 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7635 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7636 
7637 	ctxt->interruptibility = 0;
7638 	ctxt->have_exception = false;
7639 	ctxt->exception.vector = -1;
7640 	ctxt->perm_ok = false;
7641 
7642 	init_decode_cache(ctxt);
7643 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7644 }
7645 
7646 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7647 {
7648 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7649 	int ret;
7650 
7651 	init_emulate_ctxt(vcpu);
7652 
7653 	ctxt->op_bytes = 2;
7654 	ctxt->ad_bytes = 2;
7655 	ctxt->_eip = ctxt->eip + inc_eip;
7656 	ret = emulate_int_real(ctxt, irq);
7657 
7658 	if (ret != X86EMUL_CONTINUE) {
7659 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7660 	} else {
7661 		ctxt->eip = ctxt->_eip;
7662 		kvm_rip_write(vcpu, ctxt->eip);
7663 		kvm_set_rflags(vcpu, ctxt->eflags);
7664 	}
7665 }
7666 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7667 
7668 static void prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, u64 *data,
7669 					   u8 ndata, u8 *insn_bytes, u8 insn_size)
7670 {
7671 	struct kvm_run *run = vcpu->run;
7672 	u64 info[5];
7673 	u8 info_start;
7674 
7675 	/*
7676 	 * Zero the whole array used to retrieve the exit info, as casting to
7677 	 * u32 for select entries will leave some chunks uninitialized.
7678 	 */
7679 	memset(&info, 0, sizeof(info));
7680 
7681 	static_call(kvm_x86_get_exit_info)(vcpu, (u32 *)&info[0], &info[1],
7682 					   &info[2], (u32 *)&info[3],
7683 					   (u32 *)&info[4]);
7684 
7685 	run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7686 	run->emulation_failure.suberror = KVM_INTERNAL_ERROR_EMULATION;
7687 
7688 	/*
7689 	 * There's currently space for 13 entries, but 5 are used for the exit
7690 	 * reason and info.  Restrict to 4 to reduce the maintenance burden
7691 	 * when expanding kvm_run.emulation_failure in the future.
7692 	 */
7693 	if (WARN_ON_ONCE(ndata > 4))
7694 		ndata = 4;
7695 
7696 	/* Always include the flags as a 'data' entry. */
7697 	info_start = 1;
7698 	run->emulation_failure.flags = 0;
7699 
7700 	if (insn_size) {
7701 		BUILD_BUG_ON((sizeof(run->emulation_failure.insn_size) +
7702 			      sizeof(run->emulation_failure.insn_bytes) != 16));
7703 		info_start += 2;
7704 		run->emulation_failure.flags |=
7705 			KVM_INTERNAL_ERROR_EMULATION_FLAG_INSTRUCTION_BYTES;
7706 		run->emulation_failure.insn_size = insn_size;
7707 		memset(run->emulation_failure.insn_bytes, 0x90,
7708 		       sizeof(run->emulation_failure.insn_bytes));
7709 		memcpy(run->emulation_failure.insn_bytes, insn_bytes, insn_size);
7710 	}
7711 
7712 	memcpy(&run->internal.data[info_start], info, sizeof(info));
7713 	memcpy(&run->internal.data[info_start + ARRAY_SIZE(info)], data,
7714 	       ndata * sizeof(data[0]));
7715 
7716 	run->emulation_failure.ndata = info_start + ARRAY_SIZE(info) + ndata;
7717 }
7718 
7719 static void prepare_emulation_ctxt_failure_exit(struct kvm_vcpu *vcpu)
7720 {
7721 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7722 
7723 	prepare_emulation_failure_exit(vcpu, NULL, 0, ctxt->fetch.data,
7724 				       ctxt->fetch.end - ctxt->fetch.data);
7725 }
7726 
7727 void __kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, u64 *data,
7728 					  u8 ndata)
7729 {
7730 	prepare_emulation_failure_exit(vcpu, data, ndata, NULL, 0);
7731 }
7732 EXPORT_SYMBOL_GPL(__kvm_prepare_emulation_failure_exit);
7733 
7734 void kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu)
7735 {
7736 	__kvm_prepare_emulation_failure_exit(vcpu, NULL, 0);
7737 }
7738 EXPORT_SYMBOL_GPL(kvm_prepare_emulation_failure_exit);
7739 
7740 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7741 {
7742 	struct kvm *kvm = vcpu->kvm;
7743 
7744 	++vcpu->stat.insn_emulation_fail;
7745 	trace_kvm_emulate_insn_failed(vcpu);
7746 
7747 	if (emulation_type & EMULTYPE_VMWARE_GP) {
7748 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7749 		return 1;
7750 	}
7751 
7752 	if (kvm->arch.exit_on_emulation_error ||
7753 	    (emulation_type & EMULTYPE_SKIP)) {
7754 		prepare_emulation_ctxt_failure_exit(vcpu);
7755 		return 0;
7756 	}
7757 
7758 	kvm_queue_exception(vcpu, UD_VECTOR);
7759 
7760 	if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7761 		prepare_emulation_ctxt_failure_exit(vcpu);
7762 		return 0;
7763 	}
7764 
7765 	return 1;
7766 }
7767 
7768 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7769 				  bool write_fault_to_shadow_pgtable,
7770 				  int emulation_type)
7771 {
7772 	gpa_t gpa = cr2_or_gpa;
7773 	kvm_pfn_t pfn;
7774 
7775 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7776 		return false;
7777 
7778 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7779 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7780 		return false;
7781 
7782 	if (!vcpu->arch.mmu->direct_map) {
7783 		/*
7784 		 * Write permission should be allowed since only
7785 		 * write access need to be emulated.
7786 		 */
7787 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7788 
7789 		/*
7790 		 * If the mapping is invalid in guest, let cpu retry
7791 		 * it to generate fault.
7792 		 */
7793 		if (gpa == UNMAPPED_GVA)
7794 			return true;
7795 	}
7796 
7797 	/*
7798 	 * Do not retry the unhandleable instruction if it faults on the
7799 	 * readonly host memory, otherwise it will goto a infinite loop:
7800 	 * retry instruction -> write #PF -> emulation fail -> retry
7801 	 * instruction -> ...
7802 	 */
7803 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7804 
7805 	/*
7806 	 * If the instruction failed on the error pfn, it can not be fixed,
7807 	 * report the error to userspace.
7808 	 */
7809 	if (is_error_noslot_pfn(pfn))
7810 		return false;
7811 
7812 	kvm_release_pfn_clean(pfn);
7813 
7814 	/* The instructions are well-emulated on direct mmu. */
7815 	if (vcpu->arch.mmu->direct_map) {
7816 		unsigned int indirect_shadow_pages;
7817 
7818 		write_lock(&vcpu->kvm->mmu_lock);
7819 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7820 		write_unlock(&vcpu->kvm->mmu_lock);
7821 
7822 		if (indirect_shadow_pages)
7823 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7824 
7825 		return true;
7826 	}
7827 
7828 	/*
7829 	 * if emulation was due to access to shadowed page table
7830 	 * and it failed try to unshadow page and re-enter the
7831 	 * guest to let CPU execute the instruction.
7832 	 */
7833 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7834 
7835 	/*
7836 	 * If the access faults on its page table, it can not
7837 	 * be fixed by unprotecting shadow page and it should
7838 	 * be reported to userspace.
7839 	 */
7840 	return !write_fault_to_shadow_pgtable;
7841 }
7842 
7843 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7844 			      gpa_t cr2_or_gpa,  int emulation_type)
7845 {
7846 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7847 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7848 
7849 	last_retry_eip = vcpu->arch.last_retry_eip;
7850 	last_retry_addr = vcpu->arch.last_retry_addr;
7851 
7852 	/*
7853 	 * If the emulation is caused by #PF and it is non-page_table
7854 	 * writing instruction, it means the VM-EXIT is caused by shadow
7855 	 * page protected, we can zap the shadow page and retry this
7856 	 * instruction directly.
7857 	 *
7858 	 * Note: if the guest uses a non-page-table modifying instruction
7859 	 * on the PDE that points to the instruction, then we will unmap
7860 	 * the instruction and go to an infinite loop. So, we cache the
7861 	 * last retried eip and the last fault address, if we meet the eip
7862 	 * and the address again, we can break out of the potential infinite
7863 	 * loop.
7864 	 */
7865 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7866 
7867 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7868 		return false;
7869 
7870 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7871 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7872 		return false;
7873 
7874 	if (x86_page_table_writing_insn(ctxt))
7875 		return false;
7876 
7877 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7878 		return false;
7879 
7880 	vcpu->arch.last_retry_eip = ctxt->eip;
7881 	vcpu->arch.last_retry_addr = cr2_or_gpa;
7882 
7883 	if (!vcpu->arch.mmu->direct_map)
7884 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7885 
7886 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7887 
7888 	return true;
7889 }
7890 
7891 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7892 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7893 
7894 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm)
7895 {
7896 	trace_kvm_smm_transition(vcpu->vcpu_id, vcpu->arch.smbase, entering_smm);
7897 
7898 	if (entering_smm) {
7899 		vcpu->arch.hflags |= HF_SMM_MASK;
7900 	} else {
7901 		vcpu->arch.hflags &= ~(HF_SMM_MASK | HF_SMM_INSIDE_NMI_MASK);
7902 
7903 		/* Process a latched INIT or SMI, if any.  */
7904 		kvm_make_request(KVM_REQ_EVENT, vcpu);
7905 
7906 		/*
7907 		 * Even if KVM_SET_SREGS2 loaded PDPTRs out of band,
7908 		 * on SMM exit we still need to reload them from
7909 		 * guest memory
7910 		 */
7911 		vcpu->arch.pdptrs_from_userspace = false;
7912 	}
7913 
7914 	kvm_mmu_reset_context(vcpu);
7915 }
7916 
7917 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7918 				unsigned long *db)
7919 {
7920 	u32 dr6 = 0;
7921 	int i;
7922 	u32 enable, rwlen;
7923 
7924 	enable = dr7;
7925 	rwlen = dr7 >> 16;
7926 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7927 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7928 			dr6 |= (1 << i);
7929 	return dr6;
7930 }
7931 
7932 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7933 {
7934 	struct kvm_run *kvm_run = vcpu->run;
7935 
7936 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7937 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7938 		kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7939 		kvm_run->debug.arch.exception = DB_VECTOR;
7940 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
7941 		return 0;
7942 	}
7943 	kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7944 	return 1;
7945 }
7946 
7947 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7948 {
7949 	unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7950 	int r;
7951 
7952 	r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7953 	if (unlikely(!r))
7954 		return 0;
7955 
7956 	/*
7957 	 * rflags is the old, "raw" value of the flags.  The new value has
7958 	 * not been saved yet.
7959 	 *
7960 	 * This is correct even for TF set by the guest, because "the
7961 	 * processor will not generate this exception after the instruction
7962 	 * that sets the TF flag".
7963 	 */
7964 	if (unlikely(rflags & X86_EFLAGS_TF))
7965 		r = kvm_vcpu_do_singlestep(vcpu);
7966 	return r;
7967 }
7968 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7969 
7970 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7971 {
7972 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7973 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7974 		struct kvm_run *kvm_run = vcpu->run;
7975 		unsigned long eip = kvm_get_linear_rip(vcpu);
7976 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7977 					   vcpu->arch.guest_debug_dr7,
7978 					   vcpu->arch.eff_db);
7979 
7980 		if (dr6 != 0) {
7981 			kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
7982 			kvm_run->debug.arch.pc = eip;
7983 			kvm_run->debug.arch.exception = DB_VECTOR;
7984 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
7985 			*r = 0;
7986 			return true;
7987 		}
7988 	}
7989 
7990 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7991 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7992 		unsigned long eip = kvm_get_linear_rip(vcpu);
7993 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7994 					   vcpu->arch.dr7,
7995 					   vcpu->arch.db);
7996 
7997 		if (dr6 != 0) {
7998 			kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7999 			*r = 1;
8000 			return true;
8001 		}
8002 	}
8003 
8004 	return false;
8005 }
8006 
8007 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
8008 {
8009 	switch (ctxt->opcode_len) {
8010 	case 1:
8011 		switch (ctxt->b) {
8012 		case 0xe4:	/* IN */
8013 		case 0xe5:
8014 		case 0xec:
8015 		case 0xed:
8016 		case 0xe6:	/* OUT */
8017 		case 0xe7:
8018 		case 0xee:
8019 		case 0xef:
8020 		case 0x6c:	/* INS */
8021 		case 0x6d:
8022 		case 0x6e:	/* OUTS */
8023 		case 0x6f:
8024 			return true;
8025 		}
8026 		break;
8027 	case 2:
8028 		switch (ctxt->b) {
8029 		case 0x33:	/* RDPMC */
8030 			return true;
8031 		}
8032 		break;
8033 	}
8034 
8035 	return false;
8036 }
8037 
8038 /*
8039  * Decode to be emulated instruction. Return EMULATION_OK if success.
8040  */
8041 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
8042 				    void *insn, int insn_len)
8043 {
8044 	int r = EMULATION_OK;
8045 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8046 
8047 	init_emulate_ctxt(vcpu);
8048 
8049 	/*
8050 	 * We will reenter on the same instruction since we do not set
8051 	 * complete_userspace_io. This does not handle watchpoints yet,
8052 	 * those would be handled in the emulate_ops.
8053 	 */
8054 	if (!(emulation_type & EMULTYPE_SKIP) &&
8055 	    kvm_vcpu_check_breakpoint(vcpu, &r))
8056 		return r;
8057 
8058 	r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
8059 
8060 	trace_kvm_emulate_insn_start(vcpu);
8061 	++vcpu->stat.insn_emulation;
8062 
8063 	return r;
8064 }
8065 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
8066 
8067 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
8068 			    int emulation_type, void *insn, int insn_len)
8069 {
8070 	int r;
8071 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8072 	bool writeback = true;
8073 	bool write_fault_to_spt;
8074 
8075 	if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
8076 		return 1;
8077 
8078 	vcpu->arch.l1tf_flush_l1d = true;
8079 
8080 	/*
8081 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
8082 	 * never reused.
8083 	 */
8084 	write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
8085 	vcpu->arch.write_fault_to_shadow_pgtable = false;
8086 
8087 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
8088 		kvm_clear_exception_queue(vcpu);
8089 
8090 		r = x86_decode_emulated_instruction(vcpu, emulation_type,
8091 						    insn, insn_len);
8092 		if (r != EMULATION_OK)  {
8093 			if ((emulation_type & EMULTYPE_TRAP_UD) ||
8094 			    (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
8095 				kvm_queue_exception(vcpu, UD_VECTOR);
8096 				return 1;
8097 			}
8098 			if (reexecute_instruction(vcpu, cr2_or_gpa,
8099 						  write_fault_to_spt,
8100 						  emulation_type))
8101 				return 1;
8102 			if (ctxt->have_exception) {
8103 				/*
8104 				 * #UD should result in just EMULATION_FAILED, and trap-like
8105 				 * exception should not be encountered during decode.
8106 				 */
8107 				WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
8108 					     exception_type(ctxt->exception.vector) == EXCPT_TRAP);
8109 				inject_emulated_exception(vcpu);
8110 				return 1;
8111 			}
8112 			return handle_emulation_failure(vcpu, emulation_type);
8113 		}
8114 	}
8115 
8116 	if ((emulation_type & EMULTYPE_VMWARE_GP) &&
8117 	    !is_vmware_backdoor_opcode(ctxt)) {
8118 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
8119 		return 1;
8120 	}
8121 
8122 	/*
8123 	 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
8124 	 * for kvm_skip_emulated_instruction().  The caller is responsible for
8125 	 * updating interruptibility state and injecting single-step #DBs.
8126 	 */
8127 	if (emulation_type & EMULTYPE_SKIP) {
8128 		kvm_rip_write(vcpu, ctxt->_eip);
8129 		if (ctxt->eflags & X86_EFLAGS_RF)
8130 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
8131 		return 1;
8132 	}
8133 
8134 	if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
8135 		return 1;
8136 
8137 	/* this is needed for vmware backdoor interface to work since it
8138 	   changes registers values  during IO operation */
8139 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
8140 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8141 		emulator_invalidate_register_cache(ctxt);
8142 	}
8143 
8144 restart:
8145 	if (emulation_type & EMULTYPE_PF) {
8146 		/* Save the faulting GPA (cr2) in the address field */
8147 		ctxt->exception.address = cr2_or_gpa;
8148 
8149 		/* With shadow page tables, cr2 contains a GVA or nGPA. */
8150 		if (vcpu->arch.mmu->direct_map) {
8151 			ctxt->gpa_available = true;
8152 			ctxt->gpa_val = cr2_or_gpa;
8153 		}
8154 	} else {
8155 		/* Sanitize the address out of an abundance of paranoia. */
8156 		ctxt->exception.address = 0;
8157 	}
8158 
8159 	r = x86_emulate_insn(ctxt);
8160 
8161 	if (r == EMULATION_INTERCEPTED)
8162 		return 1;
8163 
8164 	if (r == EMULATION_FAILED) {
8165 		if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
8166 					emulation_type))
8167 			return 1;
8168 
8169 		return handle_emulation_failure(vcpu, emulation_type);
8170 	}
8171 
8172 	if (ctxt->have_exception) {
8173 		r = 1;
8174 		if (inject_emulated_exception(vcpu))
8175 			return r;
8176 	} else if (vcpu->arch.pio.count) {
8177 		if (!vcpu->arch.pio.in) {
8178 			/* FIXME: return into emulator if single-stepping.  */
8179 			vcpu->arch.pio.count = 0;
8180 		} else {
8181 			writeback = false;
8182 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
8183 		}
8184 		r = 0;
8185 	} else if (vcpu->mmio_needed) {
8186 		++vcpu->stat.mmio_exits;
8187 
8188 		if (!vcpu->mmio_is_write)
8189 			writeback = false;
8190 		r = 0;
8191 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8192 	} else if (r == EMULATION_RESTART)
8193 		goto restart;
8194 	else
8195 		r = 1;
8196 
8197 	if (writeback) {
8198 		unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
8199 		toggle_interruptibility(vcpu, ctxt->interruptibility);
8200 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8201 		if (!ctxt->have_exception ||
8202 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
8203 			kvm_rip_write(vcpu, ctxt->eip);
8204 			if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
8205 				r = kvm_vcpu_do_singlestep(vcpu);
8206 			if (kvm_x86_ops.update_emulated_instruction)
8207 				static_call(kvm_x86_update_emulated_instruction)(vcpu);
8208 			__kvm_set_rflags(vcpu, ctxt->eflags);
8209 		}
8210 
8211 		/*
8212 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
8213 		 * do nothing, and it will be requested again as soon as
8214 		 * the shadow expires.  But we still need to check here,
8215 		 * because POPF has no interrupt shadow.
8216 		 */
8217 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
8218 			kvm_make_request(KVM_REQ_EVENT, vcpu);
8219 	} else
8220 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
8221 
8222 	return r;
8223 }
8224 
8225 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
8226 {
8227 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
8228 }
8229 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
8230 
8231 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
8232 					void *insn, int insn_len)
8233 {
8234 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
8235 }
8236 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
8237 
8238 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
8239 {
8240 	vcpu->arch.pio.count = 0;
8241 	return 1;
8242 }
8243 
8244 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
8245 {
8246 	vcpu->arch.pio.count = 0;
8247 
8248 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
8249 		return 1;
8250 
8251 	return kvm_skip_emulated_instruction(vcpu);
8252 }
8253 
8254 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
8255 			    unsigned short port)
8256 {
8257 	unsigned long val = kvm_rax_read(vcpu);
8258 	int ret = emulator_pio_out(vcpu, size, port, &val, 1);
8259 
8260 	if (ret)
8261 		return ret;
8262 
8263 	/*
8264 	 * Workaround userspace that relies on old KVM behavior of %rip being
8265 	 * incremented prior to exiting to userspace to handle "OUT 0x7e".
8266 	 */
8267 	if (port == 0x7e &&
8268 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
8269 		vcpu->arch.complete_userspace_io =
8270 			complete_fast_pio_out_port_0x7e;
8271 		kvm_skip_emulated_instruction(vcpu);
8272 	} else {
8273 		vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8274 		vcpu->arch.complete_userspace_io = complete_fast_pio_out;
8275 	}
8276 	return 0;
8277 }
8278 
8279 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
8280 {
8281 	unsigned long val;
8282 
8283 	/* We should only ever be called with arch.pio.count equal to 1 */
8284 	BUG_ON(vcpu->arch.pio.count != 1);
8285 
8286 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
8287 		vcpu->arch.pio.count = 0;
8288 		return 1;
8289 	}
8290 
8291 	/* For size less than 4 we merge, else we zero extend */
8292 	val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
8293 
8294 	/*
8295 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
8296 	 * the copy and tracing
8297 	 */
8298 	emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
8299 	kvm_rax_write(vcpu, val);
8300 
8301 	return kvm_skip_emulated_instruction(vcpu);
8302 }
8303 
8304 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
8305 			   unsigned short port)
8306 {
8307 	unsigned long val;
8308 	int ret;
8309 
8310 	/* For size less than 4 we merge, else we zero extend */
8311 	val = (size < 4) ? kvm_rax_read(vcpu) : 0;
8312 
8313 	ret = emulator_pio_in(vcpu, size, port, &val, 1);
8314 	if (ret) {
8315 		kvm_rax_write(vcpu, val);
8316 		return ret;
8317 	}
8318 
8319 	vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8320 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
8321 
8322 	return 0;
8323 }
8324 
8325 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
8326 {
8327 	int ret;
8328 
8329 	if (in)
8330 		ret = kvm_fast_pio_in(vcpu, size, port);
8331 	else
8332 		ret = kvm_fast_pio_out(vcpu, size, port);
8333 	return ret && kvm_skip_emulated_instruction(vcpu);
8334 }
8335 EXPORT_SYMBOL_GPL(kvm_fast_pio);
8336 
8337 static int kvmclock_cpu_down_prep(unsigned int cpu)
8338 {
8339 	__this_cpu_write(cpu_tsc_khz, 0);
8340 	return 0;
8341 }
8342 
8343 static void tsc_khz_changed(void *data)
8344 {
8345 	struct cpufreq_freqs *freq = data;
8346 	unsigned long khz = 0;
8347 
8348 	if (data)
8349 		khz = freq->new;
8350 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8351 		khz = cpufreq_quick_get(raw_smp_processor_id());
8352 	if (!khz)
8353 		khz = tsc_khz;
8354 	__this_cpu_write(cpu_tsc_khz, khz);
8355 }
8356 
8357 #ifdef CONFIG_X86_64
8358 static void kvm_hyperv_tsc_notifier(void)
8359 {
8360 	struct kvm *kvm;
8361 	int cpu;
8362 
8363 	mutex_lock(&kvm_lock);
8364 	list_for_each_entry(kvm, &vm_list, vm_list)
8365 		kvm_make_mclock_inprogress_request(kvm);
8366 
8367 	/* no guest entries from this point */
8368 	hyperv_stop_tsc_emulation();
8369 
8370 	/* TSC frequency always matches when on Hyper-V */
8371 	for_each_present_cpu(cpu)
8372 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
8373 	kvm_max_guest_tsc_khz = tsc_khz;
8374 
8375 	list_for_each_entry(kvm, &vm_list, vm_list) {
8376 		__kvm_start_pvclock_update(kvm);
8377 		pvclock_update_vm_gtod_copy(kvm);
8378 		kvm_end_pvclock_update(kvm);
8379 	}
8380 
8381 	mutex_unlock(&kvm_lock);
8382 }
8383 #endif
8384 
8385 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
8386 {
8387 	struct kvm *kvm;
8388 	struct kvm_vcpu *vcpu;
8389 	int i, send_ipi = 0;
8390 
8391 	/*
8392 	 * We allow guests to temporarily run on slowing clocks,
8393 	 * provided we notify them after, or to run on accelerating
8394 	 * clocks, provided we notify them before.  Thus time never
8395 	 * goes backwards.
8396 	 *
8397 	 * However, we have a problem.  We can't atomically update
8398 	 * the frequency of a given CPU from this function; it is
8399 	 * merely a notifier, which can be called from any CPU.
8400 	 * Changing the TSC frequency at arbitrary points in time
8401 	 * requires a recomputation of local variables related to
8402 	 * the TSC for each VCPU.  We must flag these local variables
8403 	 * to be updated and be sure the update takes place with the
8404 	 * new frequency before any guests proceed.
8405 	 *
8406 	 * Unfortunately, the combination of hotplug CPU and frequency
8407 	 * change creates an intractable locking scenario; the order
8408 	 * of when these callouts happen is undefined with respect to
8409 	 * CPU hotplug, and they can race with each other.  As such,
8410 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
8411 	 * undefined; you can actually have a CPU frequency change take
8412 	 * place in between the computation of X and the setting of the
8413 	 * variable.  To protect against this problem, all updates of
8414 	 * the per_cpu tsc_khz variable are done in an interrupt
8415 	 * protected IPI, and all callers wishing to update the value
8416 	 * must wait for a synchronous IPI to complete (which is trivial
8417 	 * if the caller is on the CPU already).  This establishes the
8418 	 * necessary total order on variable updates.
8419 	 *
8420 	 * Note that because a guest time update may take place
8421 	 * anytime after the setting of the VCPU's request bit, the
8422 	 * correct TSC value must be set before the request.  However,
8423 	 * to ensure the update actually makes it to any guest which
8424 	 * starts running in hardware virtualization between the set
8425 	 * and the acquisition of the spinlock, we must also ping the
8426 	 * CPU after setting the request bit.
8427 	 *
8428 	 */
8429 
8430 	smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8431 
8432 	mutex_lock(&kvm_lock);
8433 	list_for_each_entry(kvm, &vm_list, vm_list) {
8434 		kvm_for_each_vcpu(i, vcpu, kvm) {
8435 			if (vcpu->cpu != cpu)
8436 				continue;
8437 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8438 			if (vcpu->cpu != raw_smp_processor_id())
8439 				send_ipi = 1;
8440 		}
8441 	}
8442 	mutex_unlock(&kvm_lock);
8443 
8444 	if (freq->old < freq->new && send_ipi) {
8445 		/*
8446 		 * We upscale the frequency.  Must make the guest
8447 		 * doesn't see old kvmclock values while running with
8448 		 * the new frequency, otherwise we risk the guest sees
8449 		 * time go backwards.
8450 		 *
8451 		 * In case we update the frequency for another cpu
8452 		 * (which might be in guest context) send an interrupt
8453 		 * to kick the cpu out of guest context.  Next time
8454 		 * guest context is entered kvmclock will be updated,
8455 		 * so the guest will not see stale values.
8456 		 */
8457 		smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8458 	}
8459 }
8460 
8461 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
8462 				     void *data)
8463 {
8464 	struct cpufreq_freqs *freq = data;
8465 	int cpu;
8466 
8467 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
8468 		return 0;
8469 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
8470 		return 0;
8471 
8472 	for_each_cpu(cpu, freq->policy->cpus)
8473 		__kvmclock_cpufreq_notifier(freq, cpu);
8474 
8475 	return 0;
8476 }
8477 
8478 static struct notifier_block kvmclock_cpufreq_notifier_block = {
8479 	.notifier_call  = kvmclock_cpufreq_notifier
8480 };
8481 
8482 static int kvmclock_cpu_online(unsigned int cpu)
8483 {
8484 	tsc_khz_changed(NULL);
8485 	return 0;
8486 }
8487 
8488 static void kvm_timer_init(void)
8489 {
8490 	max_tsc_khz = tsc_khz;
8491 
8492 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
8493 #ifdef CONFIG_CPU_FREQ
8494 		struct cpufreq_policy *policy;
8495 		int cpu;
8496 
8497 		cpu = get_cpu();
8498 		policy = cpufreq_cpu_get(cpu);
8499 		if (policy) {
8500 			if (policy->cpuinfo.max_freq)
8501 				max_tsc_khz = policy->cpuinfo.max_freq;
8502 			cpufreq_cpu_put(policy);
8503 		}
8504 		put_cpu();
8505 #endif
8506 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
8507 					  CPUFREQ_TRANSITION_NOTIFIER);
8508 	}
8509 
8510 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
8511 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
8512 }
8513 
8514 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
8515 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
8516 
8517 int kvm_is_in_guest(void)
8518 {
8519 	return __this_cpu_read(current_vcpu) != NULL;
8520 }
8521 
8522 static int kvm_is_user_mode(void)
8523 {
8524 	int user_mode = 3;
8525 
8526 	if (__this_cpu_read(current_vcpu))
8527 		user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
8528 
8529 	return user_mode != 0;
8530 }
8531 
8532 static unsigned long kvm_get_guest_ip(void)
8533 {
8534 	unsigned long ip = 0;
8535 
8536 	if (__this_cpu_read(current_vcpu))
8537 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
8538 
8539 	return ip;
8540 }
8541 
8542 static void kvm_handle_intel_pt_intr(void)
8543 {
8544 	struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
8545 
8546 	kvm_make_request(KVM_REQ_PMI, vcpu);
8547 	__set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8548 			(unsigned long *)&vcpu->arch.pmu.global_status);
8549 }
8550 
8551 static struct perf_guest_info_callbacks kvm_guest_cbs = {
8552 	.is_in_guest		= kvm_is_in_guest,
8553 	.is_user_mode		= kvm_is_user_mode,
8554 	.get_guest_ip		= kvm_get_guest_ip,
8555 	.handle_intel_pt_intr	= kvm_handle_intel_pt_intr,
8556 };
8557 
8558 #ifdef CONFIG_X86_64
8559 static void pvclock_gtod_update_fn(struct work_struct *work)
8560 {
8561 	struct kvm *kvm;
8562 
8563 	struct kvm_vcpu *vcpu;
8564 	int i;
8565 
8566 	mutex_lock(&kvm_lock);
8567 	list_for_each_entry(kvm, &vm_list, vm_list)
8568 		kvm_for_each_vcpu(i, vcpu, kvm)
8569 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8570 	atomic_set(&kvm_guest_has_master_clock, 0);
8571 	mutex_unlock(&kvm_lock);
8572 }
8573 
8574 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8575 
8576 /*
8577  * Indirection to move queue_work() out of the tk_core.seq write held
8578  * region to prevent possible deadlocks against time accessors which
8579  * are invoked with work related locks held.
8580  */
8581 static void pvclock_irq_work_fn(struct irq_work *w)
8582 {
8583 	queue_work(system_long_wq, &pvclock_gtod_work);
8584 }
8585 
8586 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
8587 
8588 /*
8589  * Notification about pvclock gtod data update.
8590  */
8591 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8592 			       void *priv)
8593 {
8594 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8595 	struct timekeeper *tk = priv;
8596 
8597 	update_pvclock_gtod(tk);
8598 
8599 	/*
8600 	 * Disable master clock if host does not trust, or does not use,
8601 	 * TSC based clocksource. Delegate queue_work() to irq_work as
8602 	 * this is invoked with tk_core.seq write held.
8603 	 */
8604 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8605 	    atomic_read(&kvm_guest_has_master_clock) != 0)
8606 		irq_work_queue(&pvclock_irq_work);
8607 	return 0;
8608 }
8609 
8610 static struct notifier_block pvclock_gtod_notifier = {
8611 	.notifier_call = pvclock_gtod_notify,
8612 };
8613 #endif
8614 
8615 int kvm_arch_init(void *opaque)
8616 {
8617 	struct kvm_x86_init_ops *ops = opaque;
8618 	int r;
8619 
8620 	if (kvm_x86_ops.hardware_enable) {
8621 		pr_err("kvm: already loaded vendor module '%s'\n", kvm_x86_ops.name);
8622 		r = -EEXIST;
8623 		goto out;
8624 	}
8625 
8626 	if (!ops->cpu_has_kvm_support()) {
8627 		pr_err_ratelimited("kvm: no hardware support for '%s'\n",
8628 				   ops->runtime_ops->name);
8629 		r = -EOPNOTSUPP;
8630 		goto out;
8631 	}
8632 	if (ops->disabled_by_bios()) {
8633 		pr_err_ratelimited("kvm: support for '%s' disabled by bios\n",
8634 				   ops->runtime_ops->name);
8635 		r = -EOPNOTSUPP;
8636 		goto out;
8637 	}
8638 
8639 	/*
8640 	 * KVM explicitly assumes that the guest has an FPU and
8641 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8642 	 * vCPU's FPU state as a fxregs_state struct.
8643 	 */
8644 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8645 		printk(KERN_ERR "kvm: inadequate fpu\n");
8646 		r = -EOPNOTSUPP;
8647 		goto out;
8648 	}
8649 
8650 	r = -ENOMEM;
8651 
8652 	x86_emulator_cache = kvm_alloc_emulator_cache();
8653 	if (!x86_emulator_cache) {
8654 		pr_err("kvm: failed to allocate cache for x86 emulator\n");
8655 		goto out;
8656 	}
8657 
8658 	user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8659 	if (!user_return_msrs) {
8660 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8661 		goto out_free_x86_emulator_cache;
8662 	}
8663 	kvm_nr_uret_msrs = 0;
8664 
8665 	r = kvm_mmu_module_init();
8666 	if (r)
8667 		goto out_free_percpu;
8668 
8669 	kvm_timer_init();
8670 
8671 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
8672 
8673 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8674 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8675 		supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8676 	}
8677 
8678 	if (pi_inject_timer == -1)
8679 		pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8680 #ifdef CONFIG_X86_64
8681 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8682 
8683 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8684 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8685 #endif
8686 
8687 	return 0;
8688 
8689 out_free_percpu:
8690 	free_percpu(user_return_msrs);
8691 out_free_x86_emulator_cache:
8692 	kmem_cache_destroy(x86_emulator_cache);
8693 out:
8694 	return r;
8695 }
8696 
8697 void kvm_arch_exit(void)
8698 {
8699 #ifdef CONFIG_X86_64
8700 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8701 		clear_hv_tscchange_cb();
8702 #endif
8703 	kvm_lapic_exit();
8704 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8705 
8706 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8707 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8708 					    CPUFREQ_TRANSITION_NOTIFIER);
8709 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8710 #ifdef CONFIG_X86_64
8711 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8712 	irq_work_sync(&pvclock_irq_work);
8713 	cancel_work_sync(&pvclock_gtod_work);
8714 #endif
8715 	kvm_x86_ops.hardware_enable = NULL;
8716 	kvm_mmu_module_exit();
8717 	free_percpu(user_return_msrs);
8718 	kmem_cache_destroy(x86_emulator_cache);
8719 #ifdef CONFIG_KVM_XEN
8720 	static_key_deferred_flush(&kvm_xen_enabled);
8721 	WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8722 #endif
8723 }
8724 
8725 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8726 {
8727 	++vcpu->stat.halt_exits;
8728 	if (lapic_in_kernel(vcpu)) {
8729 		vcpu->arch.mp_state = state;
8730 		return 1;
8731 	} else {
8732 		vcpu->run->exit_reason = reason;
8733 		return 0;
8734 	}
8735 }
8736 
8737 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8738 {
8739 	return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8740 }
8741 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8742 
8743 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8744 {
8745 	int ret = kvm_skip_emulated_instruction(vcpu);
8746 	/*
8747 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8748 	 * KVM_EXIT_DEBUG here.
8749 	 */
8750 	return kvm_vcpu_halt(vcpu) && ret;
8751 }
8752 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8753 
8754 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8755 {
8756 	int ret = kvm_skip_emulated_instruction(vcpu);
8757 
8758 	return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8759 }
8760 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8761 
8762 #ifdef CONFIG_X86_64
8763 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8764 			        unsigned long clock_type)
8765 {
8766 	struct kvm_clock_pairing clock_pairing;
8767 	struct timespec64 ts;
8768 	u64 cycle;
8769 	int ret;
8770 
8771 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8772 		return -KVM_EOPNOTSUPP;
8773 
8774 	if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8775 		return -KVM_EOPNOTSUPP;
8776 
8777 	clock_pairing.sec = ts.tv_sec;
8778 	clock_pairing.nsec = ts.tv_nsec;
8779 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8780 	clock_pairing.flags = 0;
8781 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8782 
8783 	ret = 0;
8784 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8785 			    sizeof(struct kvm_clock_pairing)))
8786 		ret = -KVM_EFAULT;
8787 
8788 	return ret;
8789 }
8790 #endif
8791 
8792 /*
8793  * kvm_pv_kick_cpu_op:  Kick a vcpu.
8794  *
8795  * @apicid - apicid of vcpu to be kicked.
8796  */
8797 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8798 {
8799 	struct kvm_lapic_irq lapic_irq;
8800 
8801 	lapic_irq.shorthand = APIC_DEST_NOSHORT;
8802 	lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8803 	lapic_irq.level = 0;
8804 	lapic_irq.dest_id = apicid;
8805 	lapic_irq.msi_redir_hint = false;
8806 
8807 	lapic_irq.delivery_mode = APIC_DM_REMRD;
8808 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8809 }
8810 
8811 bool kvm_apicv_activated(struct kvm *kvm)
8812 {
8813 	return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8814 }
8815 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8816 
8817 static void kvm_apicv_init(struct kvm *kvm)
8818 {
8819 	init_rwsem(&kvm->arch.apicv_update_lock);
8820 
8821 	if (enable_apicv)
8822 		clear_bit(APICV_INHIBIT_REASON_DISABLE,
8823 			  &kvm->arch.apicv_inhibit_reasons);
8824 	else
8825 		set_bit(APICV_INHIBIT_REASON_DISABLE,
8826 			&kvm->arch.apicv_inhibit_reasons);
8827 }
8828 
8829 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8830 {
8831 	struct kvm_vcpu *target = NULL;
8832 	struct kvm_apic_map *map;
8833 
8834 	vcpu->stat.directed_yield_attempted++;
8835 
8836 	if (single_task_running())
8837 		goto no_yield;
8838 
8839 	rcu_read_lock();
8840 	map = rcu_dereference(vcpu->kvm->arch.apic_map);
8841 
8842 	if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8843 		target = map->phys_map[dest_id]->vcpu;
8844 
8845 	rcu_read_unlock();
8846 
8847 	if (!target || !READ_ONCE(target->ready))
8848 		goto no_yield;
8849 
8850 	/* Ignore requests to yield to self */
8851 	if (vcpu == target)
8852 		goto no_yield;
8853 
8854 	if (kvm_vcpu_yield_to(target) <= 0)
8855 		goto no_yield;
8856 
8857 	vcpu->stat.directed_yield_successful++;
8858 
8859 no_yield:
8860 	return;
8861 }
8862 
8863 static int complete_hypercall_exit(struct kvm_vcpu *vcpu)
8864 {
8865 	u64 ret = vcpu->run->hypercall.ret;
8866 
8867 	if (!is_64_bit_mode(vcpu))
8868 		ret = (u32)ret;
8869 	kvm_rax_write(vcpu, ret);
8870 	++vcpu->stat.hypercalls;
8871 	return kvm_skip_emulated_instruction(vcpu);
8872 }
8873 
8874 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8875 {
8876 	unsigned long nr, a0, a1, a2, a3, ret;
8877 	int op_64_bit;
8878 
8879 	if (kvm_xen_hypercall_enabled(vcpu->kvm))
8880 		return kvm_xen_hypercall(vcpu);
8881 
8882 	if (kvm_hv_hypercall_enabled(vcpu))
8883 		return kvm_hv_hypercall(vcpu);
8884 
8885 	nr = kvm_rax_read(vcpu);
8886 	a0 = kvm_rbx_read(vcpu);
8887 	a1 = kvm_rcx_read(vcpu);
8888 	a2 = kvm_rdx_read(vcpu);
8889 	a3 = kvm_rsi_read(vcpu);
8890 
8891 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
8892 
8893 	op_64_bit = is_64_bit_hypercall(vcpu);
8894 	if (!op_64_bit) {
8895 		nr &= 0xFFFFFFFF;
8896 		a0 &= 0xFFFFFFFF;
8897 		a1 &= 0xFFFFFFFF;
8898 		a2 &= 0xFFFFFFFF;
8899 		a3 &= 0xFFFFFFFF;
8900 	}
8901 
8902 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8903 		ret = -KVM_EPERM;
8904 		goto out;
8905 	}
8906 
8907 	ret = -KVM_ENOSYS;
8908 
8909 	switch (nr) {
8910 	case KVM_HC_VAPIC_POLL_IRQ:
8911 		ret = 0;
8912 		break;
8913 	case KVM_HC_KICK_CPU:
8914 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8915 			break;
8916 
8917 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8918 		kvm_sched_yield(vcpu, a1);
8919 		ret = 0;
8920 		break;
8921 #ifdef CONFIG_X86_64
8922 	case KVM_HC_CLOCK_PAIRING:
8923 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8924 		break;
8925 #endif
8926 	case KVM_HC_SEND_IPI:
8927 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8928 			break;
8929 
8930 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8931 		break;
8932 	case KVM_HC_SCHED_YIELD:
8933 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8934 			break;
8935 
8936 		kvm_sched_yield(vcpu, a0);
8937 		ret = 0;
8938 		break;
8939 	case KVM_HC_MAP_GPA_RANGE: {
8940 		u64 gpa = a0, npages = a1, attrs = a2;
8941 
8942 		ret = -KVM_ENOSYS;
8943 		if (!(vcpu->kvm->arch.hypercall_exit_enabled & (1 << KVM_HC_MAP_GPA_RANGE)))
8944 			break;
8945 
8946 		if (!PAGE_ALIGNED(gpa) || !npages ||
8947 		    gpa_to_gfn(gpa) + npages <= gpa_to_gfn(gpa)) {
8948 			ret = -KVM_EINVAL;
8949 			break;
8950 		}
8951 
8952 		vcpu->run->exit_reason        = KVM_EXIT_HYPERCALL;
8953 		vcpu->run->hypercall.nr       = KVM_HC_MAP_GPA_RANGE;
8954 		vcpu->run->hypercall.args[0]  = gpa;
8955 		vcpu->run->hypercall.args[1]  = npages;
8956 		vcpu->run->hypercall.args[2]  = attrs;
8957 		vcpu->run->hypercall.longmode = op_64_bit;
8958 		vcpu->arch.complete_userspace_io = complete_hypercall_exit;
8959 		return 0;
8960 	}
8961 	default:
8962 		ret = -KVM_ENOSYS;
8963 		break;
8964 	}
8965 out:
8966 	if (!op_64_bit)
8967 		ret = (u32)ret;
8968 	kvm_rax_write(vcpu, ret);
8969 
8970 	++vcpu->stat.hypercalls;
8971 	return kvm_skip_emulated_instruction(vcpu);
8972 }
8973 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8974 
8975 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8976 {
8977 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8978 	char instruction[3];
8979 	unsigned long rip = kvm_rip_read(vcpu);
8980 
8981 	static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
8982 
8983 	return emulator_write_emulated(ctxt, rip, instruction, 3,
8984 		&ctxt->exception);
8985 }
8986 
8987 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8988 {
8989 	return vcpu->run->request_interrupt_window &&
8990 		likely(!pic_in_kernel(vcpu->kvm));
8991 }
8992 
8993 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8994 {
8995 	struct kvm_run *kvm_run = vcpu->run;
8996 
8997 	/*
8998 	 * if_flag is obsolete and useless, so do not bother
8999 	 * setting it for SEV-ES guests.  Userspace can just
9000 	 * use kvm_run->ready_for_interrupt_injection.
9001 	 */
9002 	kvm_run->if_flag = !vcpu->arch.guest_state_protected
9003 		&& (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
9004 
9005 	kvm_run->cr8 = kvm_get_cr8(vcpu);
9006 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
9007 
9008 	/*
9009 	 * The call to kvm_ready_for_interrupt_injection() may end up in
9010 	 * kvm_xen_has_interrupt() which may require the srcu lock to be
9011 	 * held, to protect against changes in the vcpu_info address.
9012 	 */
9013 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9014 	kvm_run->ready_for_interrupt_injection =
9015 		pic_in_kernel(vcpu->kvm) ||
9016 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
9017 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9018 
9019 	if (is_smm(vcpu))
9020 		kvm_run->flags |= KVM_RUN_X86_SMM;
9021 }
9022 
9023 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
9024 {
9025 	int max_irr, tpr;
9026 
9027 	if (!kvm_x86_ops.update_cr8_intercept)
9028 		return;
9029 
9030 	if (!lapic_in_kernel(vcpu))
9031 		return;
9032 
9033 	if (vcpu->arch.apicv_active)
9034 		return;
9035 
9036 	if (!vcpu->arch.apic->vapic_addr)
9037 		max_irr = kvm_lapic_find_highest_irr(vcpu);
9038 	else
9039 		max_irr = -1;
9040 
9041 	if (max_irr != -1)
9042 		max_irr >>= 4;
9043 
9044 	tpr = kvm_lapic_get_cr8(vcpu);
9045 
9046 	static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
9047 }
9048 
9049 
9050 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
9051 {
9052 	if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9053 		kvm_x86_ops.nested_ops->triple_fault(vcpu);
9054 		return 1;
9055 	}
9056 
9057 	return kvm_x86_ops.nested_ops->check_events(vcpu);
9058 }
9059 
9060 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
9061 {
9062 	if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
9063 		vcpu->arch.exception.error_code = false;
9064 	static_call(kvm_x86_queue_exception)(vcpu);
9065 }
9066 
9067 static int inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
9068 {
9069 	int r;
9070 	bool can_inject = true;
9071 
9072 	/* try to reinject previous events if any */
9073 
9074 	if (vcpu->arch.exception.injected) {
9075 		kvm_inject_exception(vcpu);
9076 		can_inject = false;
9077 	}
9078 	/*
9079 	 * Do not inject an NMI or interrupt if there is a pending
9080 	 * exception.  Exceptions and interrupts are recognized at
9081 	 * instruction boundaries, i.e. the start of an instruction.
9082 	 * Trap-like exceptions, e.g. #DB, have higher priority than
9083 	 * NMIs and interrupts, i.e. traps are recognized before an
9084 	 * NMI/interrupt that's pending on the same instruction.
9085 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
9086 	 * priority, but are only generated (pended) during instruction
9087 	 * execution, i.e. a pending fault-like exception means the
9088 	 * fault occurred on the *previous* instruction and must be
9089 	 * serviced prior to recognizing any new events in order to
9090 	 * fully complete the previous instruction.
9091 	 */
9092 	else if (!vcpu->arch.exception.pending) {
9093 		if (vcpu->arch.nmi_injected) {
9094 			static_call(kvm_x86_set_nmi)(vcpu);
9095 			can_inject = false;
9096 		} else if (vcpu->arch.interrupt.injected) {
9097 			static_call(kvm_x86_set_irq)(vcpu);
9098 			can_inject = false;
9099 		}
9100 	}
9101 
9102 	WARN_ON_ONCE(vcpu->arch.exception.injected &&
9103 		     vcpu->arch.exception.pending);
9104 
9105 	/*
9106 	 * Call check_nested_events() even if we reinjected a previous event
9107 	 * in order for caller to determine if it should require immediate-exit
9108 	 * from L2 to L1 due to pending L1 events which require exit
9109 	 * from L2 to L1.
9110 	 */
9111 	if (is_guest_mode(vcpu)) {
9112 		r = kvm_check_nested_events(vcpu);
9113 		if (r < 0)
9114 			goto out;
9115 	}
9116 
9117 	/* try to inject new event if pending */
9118 	if (vcpu->arch.exception.pending) {
9119 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
9120 					vcpu->arch.exception.has_error_code,
9121 					vcpu->arch.exception.error_code);
9122 
9123 		vcpu->arch.exception.pending = false;
9124 		vcpu->arch.exception.injected = true;
9125 
9126 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
9127 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
9128 					     X86_EFLAGS_RF);
9129 
9130 		if (vcpu->arch.exception.nr == DB_VECTOR) {
9131 			kvm_deliver_exception_payload(vcpu);
9132 			if (vcpu->arch.dr7 & DR7_GD) {
9133 				vcpu->arch.dr7 &= ~DR7_GD;
9134 				kvm_update_dr7(vcpu);
9135 			}
9136 		}
9137 
9138 		kvm_inject_exception(vcpu);
9139 		can_inject = false;
9140 	}
9141 
9142 	/* Don't inject interrupts if the user asked to avoid doing so */
9143 	if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ)
9144 		return 0;
9145 
9146 	/*
9147 	 * Finally, inject interrupt events.  If an event cannot be injected
9148 	 * due to architectural conditions (e.g. IF=0) a window-open exit
9149 	 * will re-request KVM_REQ_EVENT.  Sometimes however an event is pending
9150 	 * and can architecturally be injected, but we cannot do it right now:
9151 	 * an interrupt could have arrived just now and we have to inject it
9152 	 * as a vmexit, or there could already an event in the queue, which is
9153 	 * indicated by can_inject.  In that case we request an immediate exit
9154 	 * in order to make progress and get back here for another iteration.
9155 	 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
9156 	 */
9157 	if (vcpu->arch.smi_pending) {
9158 		r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
9159 		if (r < 0)
9160 			goto out;
9161 		if (r) {
9162 			vcpu->arch.smi_pending = false;
9163 			++vcpu->arch.smi_count;
9164 			enter_smm(vcpu);
9165 			can_inject = false;
9166 		} else
9167 			static_call(kvm_x86_enable_smi_window)(vcpu);
9168 	}
9169 
9170 	if (vcpu->arch.nmi_pending) {
9171 		r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
9172 		if (r < 0)
9173 			goto out;
9174 		if (r) {
9175 			--vcpu->arch.nmi_pending;
9176 			vcpu->arch.nmi_injected = true;
9177 			static_call(kvm_x86_set_nmi)(vcpu);
9178 			can_inject = false;
9179 			WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
9180 		}
9181 		if (vcpu->arch.nmi_pending)
9182 			static_call(kvm_x86_enable_nmi_window)(vcpu);
9183 	}
9184 
9185 	if (kvm_cpu_has_injectable_intr(vcpu)) {
9186 		r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
9187 		if (r < 0)
9188 			goto out;
9189 		if (r) {
9190 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
9191 			static_call(kvm_x86_set_irq)(vcpu);
9192 			WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
9193 		}
9194 		if (kvm_cpu_has_injectable_intr(vcpu))
9195 			static_call(kvm_x86_enable_irq_window)(vcpu);
9196 	}
9197 
9198 	if (is_guest_mode(vcpu) &&
9199 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
9200 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
9201 		*req_immediate_exit = true;
9202 
9203 	WARN_ON(vcpu->arch.exception.pending);
9204 	return 0;
9205 
9206 out:
9207 	if (r == -EBUSY) {
9208 		*req_immediate_exit = true;
9209 		r = 0;
9210 	}
9211 	return r;
9212 }
9213 
9214 static void process_nmi(struct kvm_vcpu *vcpu)
9215 {
9216 	unsigned limit = 2;
9217 
9218 	/*
9219 	 * x86 is limited to one NMI running, and one NMI pending after it.
9220 	 * If an NMI is already in progress, limit further NMIs to just one.
9221 	 * Otherwise, allow two (and we'll inject the first one immediately).
9222 	 */
9223 	if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
9224 		limit = 1;
9225 
9226 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
9227 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
9228 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9229 }
9230 
9231 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
9232 {
9233 	u32 flags = 0;
9234 	flags |= seg->g       << 23;
9235 	flags |= seg->db      << 22;
9236 	flags |= seg->l       << 21;
9237 	flags |= seg->avl     << 20;
9238 	flags |= seg->present << 15;
9239 	flags |= seg->dpl     << 13;
9240 	flags |= seg->s       << 12;
9241 	flags |= seg->type    << 8;
9242 	return flags;
9243 }
9244 
9245 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
9246 {
9247 	struct kvm_segment seg;
9248 	int offset;
9249 
9250 	kvm_get_segment(vcpu, &seg, n);
9251 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
9252 
9253 	if (n < 3)
9254 		offset = 0x7f84 + n * 12;
9255 	else
9256 		offset = 0x7f2c + (n - 3) * 12;
9257 
9258 	put_smstate(u32, buf, offset + 8, seg.base);
9259 	put_smstate(u32, buf, offset + 4, seg.limit);
9260 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
9261 }
9262 
9263 #ifdef CONFIG_X86_64
9264 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
9265 {
9266 	struct kvm_segment seg;
9267 	int offset;
9268 	u16 flags;
9269 
9270 	kvm_get_segment(vcpu, &seg, n);
9271 	offset = 0x7e00 + n * 16;
9272 
9273 	flags = enter_smm_get_segment_flags(&seg) >> 8;
9274 	put_smstate(u16, buf, offset, seg.selector);
9275 	put_smstate(u16, buf, offset + 2, flags);
9276 	put_smstate(u32, buf, offset + 4, seg.limit);
9277 	put_smstate(u64, buf, offset + 8, seg.base);
9278 }
9279 #endif
9280 
9281 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
9282 {
9283 	struct desc_ptr dt;
9284 	struct kvm_segment seg;
9285 	unsigned long val;
9286 	int i;
9287 
9288 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
9289 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
9290 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
9291 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
9292 
9293 	for (i = 0; i < 8; i++)
9294 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
9295 
9296 	kvm_get_dr(vcpu, 6, &val);
9297 	put_smstate(u32, buf, 0x7fcc, (u32)val);
9298 	kvm_get_dr(vcpu, 7, &val);
9299 	put_smstate(u32, buf, 0x7fc8, (u32)val);
9300 
9301 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9302 	put_smstate(u32, buf, 0x7fc4, seg.selector);
9303 	put_smstate(u32, buf, 0x7f64, seg.base);
9304 	put_smstate(u32, buf, 0x7f60, seg.limit);
9305 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
9306 
9307 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9308 	put_smstate(u32, buf, 0x7fc0, seg.selector);
9309 	put_smstate(u32, buf, 0x7f80, seg.base);
9310 	put_smstate(u32, buf, 0x7f7c, seg.limit);
9311 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
9312 
9313 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
9314 	put_smstate(u32, buf, 0x7f74, dt.address);
9315 	put_smstate(u32, buf, 0x7f70, dt.size);
9316 
9317 	static_call(kvm_x86_get_idt)(vcpu, &dt);
9318 	put_smstate(u32, buf, 0x7f58, dt.address);
9319 	put_smstate(u32, buf, 0x7f54, dt.size);
9320 
9321 	for (i = 0; i < 6; i++)
9322 		enter_smm_save_seg_32(vcpu, buf, i);
9323 
9324 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
9325 
9326 	/* revision id */
9327 	put_smstate(u32, buf, 0x7efc, 0x00020000);
9328 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
9329 }
9330 
9331 #ifdef CONFIG_X86_64
9332 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
9333 {
9334 	struct desc_ptr dt;
9335 	struct kvm_segment seg;
9336 	unsigned long val;
9337 	int i;
9338 
9339 	for (i = 0; i < 16; i++)
9340 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
9341 
9342 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
9343 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
9344 
9345 	kvm_get_dr(vcpu, 6, &val);
9346 	put_smstate(u64, buf, 0x7f68, val);
9347 	kvm_get_dr(vcpu, 7, &val);
9348 	put_smstate(u64, buf, 0x7f60, val);
9349 
9350 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
9351 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
9352 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
9353 
9354 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
9355 
9356 	/* revision id */
9357 	put_smstate(u32, buf, 0x7efc, 0x00020064);
9358 
9359 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
9360 
9361 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9362 	put_smstate(u16, buf, 0x7e90, seg.selector);
9363 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
9364 	put_smstate(u32, buf, 0x7e94, seg.limit);
9365 	put_smstate(u64, buf, 0x7e98, seg.base);
9366 
9367 	static_call(kvm_x86_get_idt)(vcpu, &dt);
9368 	put_smstate(u32, buf, 0x7e84, dt.size);
9369 	put_smstate(u64, buf, 0x7e88, dt.address);
9370 
9371 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9372 	put_smstate(u16, buf, 0x7e70, seg.selector);
9373 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
9374 	put_smstate(u32, buf, 0x7e74, seg.limit);
9375 	put_smstate(u64, buf, 0x7e78, seg.base);
9376 
9377 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
9378 	put_smstate(u32, buf, 0x7e64, dt.size);
9379 	put_smstate(u64, buf, 0x7e68, dt.address);
9380 
9381 	for (i = 0; i < 6; i++)
9382 		enter_smm_save_seg_64(vcpu, buf, i);
9383 }
9384 #endif
9385 
9386 static void enter_smm(struct kvm_vcpu *vcpu)
9387 {
9388 	struct kvm_segment cs, ds;
9389 	struct desc_ptr dt;
9390 	unsigned long cr0;
9391 	char buf[512];
9392 
9393 	memset(buf, 0, 512);
9394 #ifdef CONFIG_X86_64
9395 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9396 		enter_smm_save_state_64(vcpu, buf);
9397 	else
9398 #endif
9399 		enter_smm_save_state_32(vcpu, buf);
9400 
9401 	/*
9402 	 * Give enter_smm() a chance to make ISA-specific changes to the vCPU
9403 	 * state (e.g. leave guest mode) after we've saved the state into the
9404 	 * SMM state-save area.
9405 	 */
9406 	static_call(kvm_x86_enter_smm)(vcpu, buf);
9407 
9408 	kvm_smm_changed(vcpu, true);
9409 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
9410 
9411 	if (static_call(kvm_x86_get_nmi_mask)(vcpu))
9412 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
9413 	else
9414 		static_call(kvm_x86_set_nmi_mask)(vcpu, true);
9415 
9416 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
9417 	kvm_rip_write(vcpu, 0x8000);
9418 
9419 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
9420 	static_call(kvm_x86_set_cr0)(vcpu, cr0);
9421 	vcpu->arch.cr0 = cr0;
9422 
9423 	static_call(kvm_x86_set_cr4)(vcpu, 0);
9424 
9425 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
9426 	dt.address = dt.size = 0;
9427 	static_call(kvm_x86_set_idt)(vcpu, &dt);
9428 
9429 	kvm_set_dr(vcpu, 7, DR7_FIXED_1);
9430 
9431 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
9432 	cs.base = vcpu->arch.smbase;
9433 
9434 	ds.selector = 0;
9435 	ds.base = 0;
9436 
9437 	cs.limit    = ds.limit = 0xffffffff;
9438 	cs.type     = ds.type = 0x3;
9439 	cs.dpl      = ds.dpl = 0;
9440 	cs.db       = ds.db = 0;
9441 	cs.s        = ds.s = 1;
9442 	cs.l        = ds.l = 0;
9443 	cs.g        = ds.g = 1;
9444 	cs.avl      = ds.avl = 0;
9445 	cs.present  = ds.present = 1;
9446 	cs.unusable = ds.unusable = 0;
9447 	cs.padding  = ds.padding = 0;
9448 
9449 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9450 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
9451 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
9452 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
9453 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
9454 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
9455 
9456 #ifdef CONFIG_X86_64
9457 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9458 		static_call(kvm_x86_set_efer)(vcpu, 0);
9459 #endif
9460 
9461 	kvm_update_cpuid_runtime(vcpu);
9462 	kvm_mmu_reset_context(vcpu);
9463 }
9464 
9465 static void process_smi(struct kvm_vcpu *vcpu)
9466 {
9467 	vcpu->arch.smi_pending = true;
9468 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9469 }
9470 
9471 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
9472 				       unsigned long *vcpu_bitmap)
9473 {
9474 	kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC, vcpu_bitmap);
9475 }
9476 
9477 void kvm_make_scan_ioapic_request(struct kvm *kvm)
9478 {
9479 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
9480 }
9481 
9482 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
9483 {
9484 	bool activate;
9485 
9486 	if (!lapic_in_kernel(vcpu))
9487 		return;
9488 
9489 	down_read(&vcpu->kvm->arch.apicv_update_lock);
9490 
9491 	activate = kvm_apicv_activated(vcpu->kvm);
9492 	if (vcpu->arch.apicv_active == activate)
9493 		goto out;
9494 
9495 	vcpu->arch.apicv_active = activate;
9496 	kvm_apic_update_apicv(vcpu);
9497 	static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
9498 
9499 	/*
9500 	 * When APICv gets disabled, we may still have injected interrupts
9501 	 * pending. At the same time, KVM_REQ_EVENT may not be set as APICv was
9502 	 * still active when the interrupt got accepted. Make sure
9503 	 * inject_pending_event() is called to check for that.
9504 	 */
9505 	if (!vcpu->arch.apicv_active)
9506 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9507 
9508 out:
9509 	up_read(&vcpu->kvm->arch.apicv_update_lock);
9510 }
9511 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
9512 
9513 void __kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9514 {
9515 	unsigned long old, new;
9516 
9517 	lockdep_assert_held_write(&kvm->arch.apicv_update_lock);
9518 
9519 	if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
9520 	    !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
9521 		return;
9522 
9523 	old = new = kvm->arch.apicv_inhibit_reasons;
9524 
9525 	if (activate)
9526 		__clear_bit(bit, &new);
9527 	else
9528 		__set_bit(bit, &new);
9529 
9530 	if (!!old != !!new) {
9531 		trace_kvm_apicv_update_request(activate, bit);
9532 		/*
9533 		 * Kick all vCPUs before setting apicv_inhibit_reasons to avoid
9534 		 * false positives in the sanity check WARN in svm_vcpu_run().
9535 		 * This task will wait for all vCPUs to ack the kick IRQ before
9536 		 * updating apicv_inhibit_reasons, and all other vCPUs will
9537 		 * block on acquiring apicv_update_lock so that vCPUs can't
9538 		 * redo svm_vcpu_run() without seeing the new inhibit state.
9539 		 *
9540 		 * Note, holding apicv_update_lock and taking it in the read
9541 		 * side (handling the request) also prevents other vCPUs from
9542 		 * servicing the request with a stale apicv_inhibit_reasons.
9543 		 */
9544 		kvm_make_all_cpus_request(kvm, KVM_REQ_APICV_UPDATE);
9545 		kvm->arch.apicv_inhibit_reasons = new;
9546 		if (new) {
9547 			unsigned long gfn = gpa_to_gfn(APIC_DEFAULT_PHYS_BASE);
9548 			kvm_zap_gfn_range(kvm, gfn, gfn+1);
9549 		}
9550 	} else
9551 		kvm->arch.apicv_inhibit_reasons = new;
9552 }
9553 EXPORT_SYMBOL_GPL(__kvm_request_apicv_update);
9554 
9555 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9556 {
9557 	down_write(&kvm->arch.apicv_update_lock);
9558 	__kvm_request_apicv_update(kvm, activate, bit);
9559 	up_write(&kvm->arch.apicv_update_lock);
9560 }
9561 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
9562 
9563 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
9564 {
9565 	if (!kvm_apic_present(vcpu))
9566 		return;
9567 
9568 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
9569 
9570 	if (irqchip_split(vcpu->kvm))
9571 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
9572 	else {
9573 		static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
9574 		if (ioapic_in_kernel(vcpu->kvm))
9575 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
9576 	}
9577 
9578 	if (is_guest_mode(vcpu))
9579 		vcpu->arch.load_eoi_exitmap_pending = true;
9580 	else
9581 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
9582 }
9583 
9584 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
9585 {
9586 	u64 eoi_exit_bitmap[4];
9587 
9588 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
9589 		return;
9590 
9591 	if (to_hv_vcpu(vcpu)) {
9592 		bitmap_or((ulong *)eoi_exit_bitmap,
9593 			  vcpu->arch.ioapic_handled_vectors,
9594 			  to_hv_synic(vcpu)->vec_bitmap, 256);
9595 		static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
9596 		return;
9597 	}
9598 
9599 	static_call(kvm_x86_load_eoi_exitmap)(
9600 		vcpu, (u64 *)vcpu->arch.ioapic_handled_vectors);
9601 }
9602 
9603 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
9604 					    unsigned long start, unsigned long end)
9605 {
9606 	unsigned long apic_address;
9607 
9608 	/*
9609 	 * The physical address of apic access page is stored in the VMCS.
9610 	 * Update it when it becomes invalid.
9611 	 */
9612 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9613 	if (start <= apic_address && apic_address < end)
9614 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9615 }
9616 
9617 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9618 {
9619 	if (!lapic_in_kernel(vcpu))
9620 		return;
9621 
9622 	if (!kvm_x86_ops.set_apic_access_page_addr)
9623 		return;
9624 
9625 	static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9626 }
9627 
9628 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9629 {
9630 	smp_send_reschedule(vcpu->cpu);
9631 }
9632 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9633 
9634 /*
9635  * Returns 1 to let vcpu_run() continue the guest execution loop without
9636  * exiting to the userspace.  Otherwise, the value will be returned to the
9637  * userspace.
9638  */
9639 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9640 {
9641 	int r;
9642 	bool req_int_win =
9643 		dm_request_for_irq_injection(vcpu) &&
9644 		kvm_cpu_accept_dm_intr(vcpu);
9645 	fastpath_t exit_fastpath;
9646 
9647 	bool req_immediate_exit = false;
9648 
9649 	/* Forbid vmenter if vcpu dirty ring is soft-full */
9650 	if (unlikely(vcpu->kvm->dirty_ring_size &&
9651 		     kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9652 		vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9653 		trace_kvm_dirty_ring_exit(vcpu);
9654 		r = 0;
9655 		goto out;
9656 	}
9657 
9658 	if (kvm_request_pending(vcpu)) {
9659 		if (kvm_check_request(KVM_REQ_VM_DEAD, vcpu)) {
9660 			r = -EIO;
9661 			goto out;
9662 		}
9663 		if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9664 			if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9665 				r = 0;
9666 				goto out;
9667 			}
9668 		}
9669 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9670 			kvm_mmu_unload(vcpu);
9671 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9672 			__kvm_migrate_timers(vcpu);
9673 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9674 			kvm_update_masterclock(vcpu->kvm);
9675 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9676 			kvm_gen_kvmclock_update(vcpu);
9677 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9678 			r = kvm_guest_time_update(vcpu);
9679 			if (unlikely(r))
9680 				goto out;
9681 		}
9682 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9683 			kvm_mmu_sync_roots(vcpu);
9684 		if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9685 			kvm_mmu_load_pgd(vcpu);
9686 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9687 			kvm_vcpu_flush_tlb_all(vcpu);
9688 
9689 			/* Flushing all ASIDs flushes the current ASID... */
9690 			kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9691 		}
9692 		kvm_service_local_tlb_flush_requests(vcpu);
9693 
9694 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9695 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9696 			r = 0;
9697 			goto out;
9698 		}
9699 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9700 			if (is_guest_mode(vcpu)) {
9701 				kvm_x86_ops.nested_ops->triple_fault(vcpu);
9702 			} else {
9703 				vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9704 				vcpu->mmio_needed = 0;
9705 				r = 0;
9706 				goto out;
9707 			}
9708 		}
9709 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9710 			/* Page is swapped out. Do synthetic halt */
9711 			vcpu->arch.apf.halted = true;
9712 			r = 1;
9713 			goto out;
9714 		}
9715 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9716 			record_steal_time(vcpu);
9717 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
9718 			process_smi(vcpu);
9719 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
9720 			process_nmi(vcpu);
9721 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
9722 			kvm_pmu_handle_event(vcpu);
9723 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
9724 			kvm_pmu_deliver_pmi(vcpu);
9725 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9726 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9727 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
9728 				     vcpu->arch.ioapic_handled_vectors)) {
9729 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9730 				vcpu->run->eoi.vector =
9731 						vcpu->arch.pending_ioapic_eoi;
9732 				r = 0;
9733 				goto out;
9734 			}
9735 		}
9736 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9737 			vcpu_scan_ioapic(vcpu);
9738 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9739 			vcpu_load_eoi_exitmap(vcpu);
9740 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9741 			kvm_vcpu_reload_apic_access_page(vcpu);
9742 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9743 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9744 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9745 			r = 0;
9746 			goto out;
9747 		}
9748 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9749 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9750 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9751 			r = 0;
9752 			goto out;
9753 		}
9754 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9755 			struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9756 
9757 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9758 			vcpu->run->hyperv = hv_vcpu->exit;
9759 			r = 0;
9760 			goto out;
9761 		}
9762 
9763 		/*
9764 		 * KVM_REQ_HV_STIMER has to be processed after
9765 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9766 		 * depend on the guest clock being up-to-date
9767 		 */
9768 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9769 			kvm_hv_process_stimers(vcpu);
9770 		if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9771 			kvm_vcpu_update_apicv(vcpu);
9772 		if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9773 			kvm_check_async_pf_completion(vcpu);
9774 		if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9775 			static_call(kvm_x86_msr_filter_changed)(vcpu);
9776 
9777 		if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9778 			static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9779 	}
9780 
9781 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9782 	    kvm_xen_has_interrupt(vcpu)) {
9783 		++vcpu->stat.req_event;
9784 		r = kvm_apic_accept_events(vcpu);
9785 		if (r < 0) {
9786 			r = 0;
9787 			goto out;
9788 		}
9789 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9790 			r = 1;
9791 			goto out;
9792 		}
9793 
9794 		r = inject_pending_event(vcpu, &req_immediate_exit);
9795 		if (r < 0) {
9796 			r = 0;
9797 			goto out;
9798 		}
9799 		if (req_int_win)
9800 			static_call(kvm_x86_enable_irq_window)(vcpu);
9801 
9802 		if (kvm_lapic_enabled(vcpu)) {
9803 			update_cr8_intercept(vcpu);
9804 			kvm_lapic_sync_to_vapic(vcpu);
9805 		}
9806 	}
9807 
9808 	r = kvm_mmu_reload(vcpu);
9809 	if (unlikely(r)) {
9810 		goto cancel_injection;
9811 	}
9812 
9813 	preempt_disable();
9814 
9815 	static_call(kvm_x86_prepare_guest_switch)(vcpu);
9816 
9817 	/*
9818 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
9819 	 * IPI are then delayed after guest entry, which ensures that they
9820 	 * result in virtual interrupt delivery.
9821 	 */
9822 	local_irq_disable();
9823 	vcpu->mode = IN_GUEST_MODE;
9824 
9825 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9826 
9827 	/*
9828 	 * 1) We should set ->mode before checking ->requests.  Please see
9829 	 * the comment in kvm_vcpu_exiting_guest_mode().
9830 	 *
9831 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
9832 	 * pairs with the memory barrier implicit in pi_test_and_set_on
9833 	 * (see vmx_deliver_posted_interrupt).
9834 	 *
9835 	 * 3) This also orders the write to mode from any reads to the page
9836 	 * tables done while the VCPU is running.  Please see the comment
9837 	 * in kvm_flush_remote_tlbs.
9838 	 */
9839 	smp_mb__after_srcu_read_unlock();
9840 
9841 	/*
9842 	 * This handles the case where a posted interrupt was
9843 	 * notified with kvm_vcpu_kick.  Assigned devices can
9844 	 * use the POSTED_INTR_VECTOR even if APICv is disabled,
9845 	 * so do it even if APICv is disabled on this vCPU.
9846 	 */
9847 	if (kvm_lapic_enabled(vcpu))
9848 		static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
9849 
9850 	if (kvm_vcpu_exit_request(vcpu)) {
9851 		vcpu->mode = OUTSIDE_GUEST_MODE;
9852 		smp_wmb();
9853 		local_irq_enable();
9854 		preempt_enable();
9855 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9856 		r = 1;
9857 		goto cancel_injection;
9858 	}
9859 
9860 	if (req_immediate_exit) {
9861 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9862 		static_call(kvm_x86_request_immediate_exit)(vcpu);
9863 	}
9864 
9865 	fpregs_assert_state_consistent();
9866 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
9867 		switch_fpu_return();
9868 
9869 	if (unlikely(vcpu->arch.switch_db_regs)) {
9870 		set_debugreg(0, 7);
9871 		set_debugreg(vcpu->arch.eff_db[0], 0);
9872 		set_debugreg(vcpu->arch.eff_db[1], 1);
9873 		set_debugreg(vcpu->arch.eff_db[2], 2);
9874 		set_debugreg(vcpu->arch.eff_db[3], 3);
9875 	} else if (unlikely(hw_breakpoint_active())) {
9876 		set_debugreg(0, 7);
9877 	}
9878 
9879 	for (;;) {
9880 		/*
9881 		 * Assert that vCPU vs. VM APICv state is consistent.  An APICv
9882 		 * update must kick and wait for all vCPUs before toggling the
9883 		 * per-VM state, and responsing vCPUs must wait for the update
9884 		 * to complete before servicing KVM_REQ_APICV_UPDATE.
9885 		 */
9886 		WARN_ON_ONCE(kvm_apicv_activated(vcpu->kvm) != kvm_vcpu_apicv_active(vcpu));
9887 
9888 		exit_fastpath = static_call(kvm_x86_run)(vcpu);
9889 		if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9890 			break;
9891 
9892 		if (kvm_lapic_enabled(vcpu))
9893 			static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
9894 
9895 		if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9896 			exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9897 			break;
9898 		}
9899 	}
9900 
9901 	/*
9902 	 * Do this here before restoring debug registers on the host.  And
9903 	 * since we do this before handling the vmexit, a DR access vmexit
9904 	 * can (a) read the correct value of the debug registers, (b) set
9905 	 * KVM_DEBUGREG_WONT_EXIT again.
9906 	 */
9907 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9908 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9909 		static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9910 		kvm_update_dr0123(vcpu);
9911 		kvm_update_dr7(vcpu);
9912 	}
9913 
9914 	/*
9915 	 * If the guest has used debug registers, at least dr7
9916 	 * will be disabled while returning to the host.
9917 	 * If we don't have active breakpoints in the host, we don't
9918 	 * care about the messed up debug address registers. But if
9919 	 * we have some of them active, restore the old state.
9920 	 */
9921 	if (hw_breakpoint_active())
9922 		hw_breakpoint_restore();
9923 
9924 	vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9925 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9926 
9927 	vcpu->mode = OUTSIDE_GUEST_MODE;
9928 	smp_wmb();
9929 
9930 	static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9931 
9932 	/*
9933 	 * Consume any pending interrupts, including the possible source of
9934 	 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9935 	 * An instruction is required after local_irq_enable() to fully unblock
9936 	 * interrupts on processors that implement an interrupt shadow, the
9937 	 * stat.exits increment will do nicely.
9938 	 */
9939 	kvm_before_interrupt(vcpu);
9940 	local_irq_enable();
9941 	++vcpu->stat.exits;
9942 	local_irq_disable();
9943 	kvm_after_interrupt(vcpu);
9944 
9945 	/*
9946 	 * Wait until after servicing IRQs to account guest time so that any
9947 	 * ticks that occurred while running the guest are properly accounted
9948 	 * to the guest.  Waiting until IRQs are enabled degrades the accuracy
9949 	 * of accounting via context tracking, but the loss of accuracy is
9950 	 * acceptable for all known use cases.
9951 	 */
9952 	vtime_account_guest_exit();
9953 
9954 	if (lapic_in_kernel(vcpu)) {
9955 		s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9956 		if (delta != S64_MIN) {
9957 			trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9958 			vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9959 		}
9960 	}
9961 
9962 	local_irq_enable();
9963 	preempt_enable();
9964 
9965 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9966 
9967 	/*
9968 	 * Profile KVM exit RIPs:
9969 	 */
9970 	if (unlikely(prof_on == KVM_PROFILING)) {
9971 		unsigned long rip = kvm_rip_read(vcpu);
9972 		profile_hit(KVM_PROFILING, (void *)rip);
9973 	}
9974 
9975 	if (unlikely(vcpu->arch.tsc_always_catchup))
9976 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9977 
9978 	if (vcpu->arch.apic_attention)
9979 		kvm_lapic_sync_from_vapic(vcpu);
9980 
9981 	r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
9982 	return r;
9983 
9984 cancel_injection:
9985 	if (req_immediate_exit)
9986 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9987 	static_call(kvm_x86_cancel_injection)(vcpu);
9988 	if (unlikely(vcpu->arch.apic_attention))
9989 		kvm_lapic_sync_from_vapic(vcpu);
9990 out:
9991 	return r;
9992 }
9993 
9994 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9995 {
9996 	if (!kvm_arch_vcpu_runnable(vcpu) &&
9997 	    (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
9998 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9999 		kvm_vcpu_block(vcpu);
10000 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
10001 
10002 		if (kvm_x86_ops.post_block)
10003 			static_call(kvm_x86_post_block)(vcpu);
10004 
10005 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
10006 			return 1;
10007 	}
10008 
10009 	if (kvm_apic_accept_events(vcpu) < 0)
10010 		return 0;
10011 	switch(vcpu->arch.mp_state) {
10012 	case KVM_MP_STATE_HALTED:
10013 	case KVM_MP_STATE_AP_RESET_HOLD:
10014 		vcpu->arch.pv.pv_unhalted = false;
10015 		vcpu->arch.mp_state =
10016 			KVM_MP_STATE_RUNNABLE;
10017 		fallthrough;
10018 	case KVM_MP_STATE_RUNNABLE:
10019 		vcpu->arch.apf.halted = false;
10020 		break;
10021 	case KVM_MP_STATE_INIT_RECEIVED:
10022 		break;
10023 	default:
10024 		return -EINTR;
10025 	}
10026 	return 1;
10027 }
10028 
10029 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
10030 {
10031 	if (is_guest_mode(vcpu))
10032 		kvm_check_nested_events(vcpu);
10033 
10034 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
10035 		!vcpu->arch.apf.halted);
10036 }
10037 
10038 static int vcpu_run(struct kvm_vcpu *vcpu)
10039 {
10040 	int r;
10041 	struct kvm *kvm = vcpu->kvm;
10042 
10043 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
10044 	vcpu->arch.l1tf_flush_l1d = true;
10045 
10046 	for (;;) {
10047 		if (kvm_vcpu_running(vcpu)) {
10048 			r = vcpu_enter_guest(vcpu);
10049 		} else {
10050 			r = vcpu_block(kvm, vcpu);
10051 		}
10052 
10053 		if (r <= 0)
10054 			break;
10055 
10056 		kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
10057 		if (kvm_cpu_has_pending_timer(vcpu))
10058 			kvm_inject_pending_timer_irqs(vcpu);
10059 
10060 		if (dm_request_for_irq_injection(vcpu) &&
10061 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
10062 			r = 0;
10063 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
10064 			++vcpu->stat.request_irq_exits;
10065 			break;
10066 		}
10067 
10068 		if (__xfer_to_guest_mode_work_pending()) {
10069 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
10070 			r = xfer_to_guest_mode_handle_work(vcpu);
10071 			if (r)
10072 				return r;
10073 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
10074 		}
10075 	}
10076 
10077 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
10078 
10079 	return r;
10080 }
10081 
10082 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
10083 {
10084 	int r;
10085 
10086 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
10087 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
10088 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
10089 	return r;
10090 }
10091 
10092 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
10093 {
10094 	BUG_ON(!vcpu->arch.pio.count);
10095 
10096 	return complete_emulated_io(vcpu);
10097 }
10098 
10099 /*
10100  * Implements the following, as a state machine:
10101  *
10102  * read:
10103  *   for each fragment
10104  *     for each mmio piece in the fragment
10105  *       write gpa, len
10106  *       exit
10107  *       copy data
10108  *   execute insn
10109  *
10110  * write:
10111  *   for each fragment
10112  *     for each mmio piece in the fragment
10113  *       write gpa, len
10114  *       copy data
10115  *       exit
10116  */
10117 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
10118 {
10119 	struct kvm_run *run = vcpu->run;
10120 	struct kvm_mmio_fragment *frag;
10121 	unsigned len;
10122 
10123 	BUG_ON(!vcpu->mmio_needed);
10124 
10125 	/* Complete previous fragment */
10126 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
10127 	len = min(8u, frag->len);
10128 	if (!vcpu->mmio_is_write)
10129 		memcpy(frag->data, run->mmio.data, len);
10130 
10131 	if (frag->len <= 8) {
10132 		/* Switch to the next fragment. */
10133 		frag++;
10134 		vcpu->mmio_cur_fragment++;
10135 	} else {
10136 		/* Go forward to the next mmio piece. */
10137 		frag->data += len;
10138 		frag->gpa += len;
10139 		frag->len -= len;
10140 	}
10141 
10142 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
10143 		vcpu->mmio_needed = 0;
10144 
10145 		/* FIXME: return into emulator if single-stepping.  */
10146 		if (vcpu->mmio_is_write)
10147 			return 1;
10148 		vcpu->mmio_read_completed = 1;
10149 		return complete_emulated_io(vcpu);
10150 	}
10151 
10152 	run->exit_reason = KVM_EXIT_MMIO;
10153 	run->mmio.phys_addr = frag->gpa;
10154 	if (vcpu->mmio_is_write)
10155 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
10156 	run->mmio.len = min(8u, frag->len);
10157 	run->mmio.is_write = vcpu->mmio_is_write;
10158 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
10159 	return 0;
10160 }
10161 
10162 /* Swap (qemu) user FPU context for the guest FPU context. */
10163 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
10164 {
10165 	/*
10166 	 * Exclude PKRU from restore as restored separately in
10167 	 * kvm_x86_ops.run().
10168 	 */
10169 	fpu_swap_kvm_fpstate(&vcpu->arch.guest_fpu, true);
10170 	trace_kvm_fpu(1);
10171 }
10172 
10173 /* When vcpu_run ends, restore user space FPU context. */
10174 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
10175 {
10176 	fpu_swap_kvm_fpstate(&vcpu->arch.guest_fpu, false);
10177 	++vcpu->stat.fpu_reload;
10178 	trace_kvm_fpu(0);
10179 }
10180 
10181 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
10182 {
10183 	struct kvm_run *kvm_run = vcpu->run;
10184 	int r;
10185 
10186 	vcpu_load(vcpu);
10187 	kvm_sigset_activate(vcpu);
10188 	kvm_run->flags = 0;
10189 	kvm_load_guest_fpu(vcpu);
10190 
10191 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
10192 		if (kvm_run->immediate_exit) {
10193 			r = -EINTR;
10194 			goto out;
10195 		}
10196 		kvm_vcpu_block(vcpu);
10197 		if (kvm_apic_accept_events(vcpu) < 0) {
10198 			r = 0;
10199 			goto out;
10200 		}
10201 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
10202 		r = -EAGAIN;
10203 		if (signal_pending(current)) {
10204 			r = -EINTR;
10205 			kvm_run->exit_reason = KVM_EXIT_INTR;
10206 			++vcpu->stat.signal_exits;
10207 		}
10208 		goto out;
10209 	}
10210 
10211 	if ((kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) ||
10212 	    (kvm_run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)) {
10213 		r = -EINVAL;
10214 		goto out;
10215 	}
10216 
10217 	if (kvm_run->kvm_dirty_regs) {
10218 		r = sync_regs(vcpu);
10219 		if (r != 0)
10220 			goto out;
10221 	}
10222 
10223 	/* re-sync apic's tpr */
10224 	if (!lapic_in_kernel(vcpu)) {
10225 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
10226 			r = -EINVAL;
10227 			goto out;
10228 		}
10229 	}
10230 
10231 	if (unlikely(vcpu->arch.complete_userspace_io)) {
10232 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
10233 		vcpu->arch.complete_userspace_io = NULL;
10234 		r = cui(vcpu);
10235 		if (r <= 0)
10236 			goto out;
10237 	} else
10238 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
10239 
10240 	if (kvm_run->immediate_exit)
10241 		r = -EINTR;
10242 	else
10243 		r = vcpu_run(vcpu);
10244 
10245 out:
10246 	kvm_put_guest_fpu(vcpu);
10247 	if (kvm_run->kvm_valid_regs)
10248 		store_regs(vcpu);
10249 	post_kvm_run_save(vcpu);
10250 	kvm_sigset_deactivate(vcpu);
10251 
10252 	vcpu_put(vcpu);
10253 	return r;
10254 }
10255 
10256 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10257 {
10258 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
10259 		/*
10260 		 * We are here if userspace calls get_regs() in the middle of
10261 		 * instruction emulation. Registers state needs to be copied
10262 		 * back from emulation context to vcpu. Userspace shouldn't do
10263 		 * that usually, but some bad designed PV devices (vmware
10264 		 * backdoor interface) need this to work
10265 		 */
10266 		emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
10267 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10268 	}
10269 	regs->rax = kvm_rax_read(vcpu);
10270 	regs->rbx = kvm_rbx_read(vcpu);
10271 	regs->rcx = kvm_rcx_read(vcpu);
10272 	regs->rdx = kvm_rdx_read(vcpu);
10273 	regs->rsi = kvm_rsi_read(vcpu);
10274 	regs->rdi = kvm_rdi_read(vcpu);
10275 	regs->rsp = kvm_rsp_read(vcpu);
10276 	regs->rbp = kvm_rbp_read(vcpu);
10277 #ifdef CONFIG_X86_64
10278 	regs->r8 = kvm_r8_read(vcpu);
10279 	regs->r9 = kvm_r9_read(vcpu);
10280 	regs->r10 = kvm_r10_read(vcpu);
10281 	regs->r11 = kvm_r11_read(vcpu);
10282 	regs->r12 = kvm_r12_read(vcpu);
10283 	regs->r13 = kvm_r13_read(vcpu);
10284 	regs->r14 = kvm_r14_read(vcpu);
10285 	regs->r15 = kvm_r15_read(vcpu);
10286 #endif
10287 
10288 	regs->rip = kvm_rip_read(vcpu);
10289 	regs->rflags = kvm_get_rflags(vcpu);
10290 }
10291 
10292 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10293 {
10294 	vcpu_load(vcpu);
10295 	__get_regs(vcpu, regs);
10296 	vcpu_put(vcpu);
10297 	return 0;
10298 }
10299 
10300 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10301 {
10302 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
10303 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10304 
10305 	kvm_rax_write(vcpu, regs->rax);
10306 	kvm_rbx_write(vcpu, regs->rbx);
10307 	kvm_rcx_write(vcpu, regs->rcx);
10308 	kvm_rdx_write(vcpu, regs->rdx);
10309 	kvm_rsi_write(vcpu, regs->rsi);
10310 	kvm_rdi_write(vcpu, regs->rdi);
10311 	kvm_rsp_write(vcpu, regs->rsp);
10312 	kvm_rbp_write(vcpu, regs->rbp);
10313 #ifdef CONFIG_X86_64
10314 	kvm_r8_write(vcpu, regs->r8);
10315 	kvm_r9_write(vcpu, regs->r9);
10316 	kvm_r10_write(vcpu, regs->r10);
10317 	kvm_r11_write(vcpu, regs->r11);
10318 	kvm_r12_write(vcpu, regs->r12);
10319 	kvm_r13_write(vcpu, regs->r13);
10320 	kvm_r14_write(vcpu, regs->r14);
10321 	kvm_r15_write(vcpu, regs->r15);
10322 #endif
10323 
10324 	kvm_rip_write(vcpu, regs->rip);
10325 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
10326 
10327 	vcpu->arch.exception.pending = false;
10328 
10329 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10330 }
10331 
10332 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10333 {
10334 	vcpu_load(vcpu);
10335 	__set_regs(vcpu, regs);
10336 	vcpu_put(vcpu);
10337 	return 0;
10338 }
10339 
10340 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
10341 {
10342 	struct kvm_segment cs;
10343 
10344 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10345 	*db = cs.db;
10346 	*l = cs.l;
10347 }
10348 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
10349 
10350 static void __get_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10351 {
10352 	struct desc_ptr dt;
10353 
10354 	if (vcpu->arch.guest_state_protected)
10355 		goto skip_protected_regs;
10356 
10357 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10358 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10359 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10360 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10361 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10362 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10363 
10364 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10365 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10366 
10367 	static_call(kvm_x86_get_idt)(vcpu, &dt);
10368 	sregs->idt.limit = dt.size;
10369 	sregs->idt.base = dt.address;
10370 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
10371 	sregs->gdt.limit = dt.size;
10372 	sregs->gdt.base = dt.address;
10373 
10374 	sregs->cr2 = vcpu->arch.cr2;
10375 	sregs->cr3 = kvm_read_cr3(vcpu);
10376 
10377 skip_protected_regs:
10378 	sregs->cr0 = kvm_read_cr0(vcpu);
10379 	sregs->cr4 = kvm_read_cr4(vcpu);
10380 	sregs->cr8 = kvm_get_cr8(vcpu);
10381 	sregs->efer = vcpu->arch.efer;
10382 	sregs->apic_base = kvm_get_apic_base(vcpu);
10383 }
10384 
10385 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10386 {
10387 	__get_sregs_common(vcpu, sregs);
10388 
10389 	if (vcpu->arch.guest_state_protected)
10390 		return;
10391 
10392 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
10393 		set_bit(vcpu->arch.interrupt.nr,
10394 			(unsigned long *)sregs->interrupt_bitmap);
10395 }
10396 
10397 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10398 {
10399 	int i;
10400 
10401 	__get_sregs_common(vcpu, (struct kvm_sregs *)sregs2);
10402 
10403 	if (vcpu->arch.guest_state_protected)
10404 		return;
10405 
10406 	if (is_pae_paging(vcpu)) {
10407 		for (i = 0 ; i < 4 ; i++)
10408 			sregs2->pdptrs[i] = kvm_pdptr_read(vcpu, i);
10409 		sregs2->flags |= KVM_SREGS2_FLAGS_PDPTRS_VALID;
10410 	}
10411 }
10412 
10413 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
10414 				  struct kvm_sregs *sregs)
10415 {
10416 	vcpu_load(vcpu);
10417 	__get_sregs(vcpu, sregs);
10418 	vcpu_put(vcpu);
10419 	return 0;
10420 }
10421 
10422 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
10423 				    struct kvm_mp_state *mp_state)
10424 {
10425 	int r;
10426 
10427 	vcpu_load(vcpu);
10428 	if (kvm_mpx_supported())
10429 		kvm_load_guest_fpu(vcpu);
10430 
10431 	r = kvm_apic_accept_events(vcpu);
10432 	if (r < 0)
10433 		goto out;
10434 	r = 0;
10435 
10436 	if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
10437 	     vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
10438 	    vcpu->arch.pv.pv_unhalted)
10439 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
10440 	else
10441 		mp_state->mp_state = vcpu->arch.mp_state;
10442 
10443 out:
10444 	if (kvm_mpx_supported())
10445 		kvm_put_guest_fpu(vcpu);
10446 	vcpu_put(vcpu);
10447 	return r;
10448 }
10449 
10450 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
10451 				    struct kvm_mp_state *mp_state)
10452 {
10453 	int ret = -EINVAL;
10454 
10455 	vcpu_load(vcpu);
10456 
10457 	if (!lapic_in_kernel(vcpu) &&
10458 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
10459 		goto out;
10460 
10461 	/*
10462 	 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
10463 	 * INIT state; latched init should be reported using
10464 	 * KVM_SET_VCPU_EVENTS, so reject it here.
10465 	 */
10466 	if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
10467 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
10468 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
10469 		goto out;
10470 
10471 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
10472 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
10473 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
10474 	} else
10475 		vcpu->arch.mp_state = mp_state->mp_state;
10476 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10477 
10478 	ret = 0;
10479 out:
10480 	vcpu_put(vcpu);
10481 	return ret;
10482 }
10483 
10484 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
10485 		    int reason, bool has_error_code, u32 error_code)
10486 {
10487 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
10488 	int ret;
10489 
10490 	init_emulate_ctxt(vcpu);
10491 
10492 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
10493 				   has_error_code, error_code);
10494 	if (ret) {
10495 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
10496 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
10497 		vcpu->run->internal.ndata = 0;
10498 		return 0;
10499 	}
10500 
10501 	kvm_rip_write(vcpu, ctxt->eip);
10502 	kvm_set_rflags(vcpu, ctxt->eflags);
10503 	return 1;
10504 }
10505 EXPORT_SYMBOL_GPL(kvm_task_switch);
10506 
10507 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10508 {
10509 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
10510 		/*
10511 		 * When EFER.LME and CR0.PG are set, the processor is in
10512 		 * 64-bit mode (though maybe in a 32-bit code segment).
10513 		 * CR4.PAE and EFER.LMA must be set.
10514 		 */
10515 		if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
10516 			return false;
10517 		if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
10518 			return false;
10519 	} else {
10520 		/*
10521 		 * Not in 64-bit mode: EFER.LMA is clear and the code
10522 		 * segment cannot be 64-bit.
10523 		 */
10524 		if (sregs->efer & EFER_LMA || sregs->cs.l)
10525 			return false;
10526 	}
10527 
10528 	return kvm_is_valid_cr4(vcpu, sregs->cr4);
10529 }
10530 
10531 static int __set_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs,
10532 		int *mmu_reset_needed, bool update_pdptrs)
10533 {
10534 	struct msr_data apic_base_msr;
10535 	int idx;
10536 	struct desc_ptr dt;
10537 
10538 	if (!kvm_is_valid_sregs(vcpu, sregs))
10539 		return -EINVAL;
10540 
10541 	apic_base_msr.data = sregs->apic_base;
10542 	apic_base_msr.host_initiated = true;
10543 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
10544 		return -EINVAL;
10545 
10546 	if (vcpu->arch.guest_state_protected)
10547 		return 0;
10548 
10549 	dt.size = sregs->idt.limit;
10550 	dt.address = sregs->idt.base;
10551 	static_call(kvm_x86_set_idt)(vcpu, &dt);
10552 	dt.size = sregs->gdt.limit;
10553 	dt.address = sregs->gdt.base;
10554 	static_call(kvm_x86_set_gdt)(vcpu, &dt);
10555 
10556 	vcpu->arch.cr2 = sregs->cr2;
10557 	*mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
10558 	vcpu->arch.cr3 = sregs->cr3;
10559 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
10560 
10561 	kvm_set_cr8(vcpu, sregs->cr8);
10562 
10563 	*mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
10564 	static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
10565 
10566 	*mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
10567 	static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
10568 	vcpu->arch.cr0 = sregs->cr0;
10569 
10570 	*mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
10571 	static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
10572 
10573 	if (update_pdptrs) {
10574 		idx = srcu_read_lock(&vcpu->kvm->srcu);
10575 		if (is_pae_paging(vcpu)) {
10576 			load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
10577 			*mmu_reset_needed = 1;
10578 		}
10579 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
10580 	}
10581 
10582 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10583 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10584 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10585 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10586 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10587 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10588 
10589 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10590 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10591 
10592 	update_cr8_intercept(vcpu);
10593 
10594 	/* Older userspace won't unhalt the vcpu on reset. */
10595 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
10596 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
10597 	    !is_protmode(vcpu))
10598 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10599 
10600 	return 0;
10601 }
10602 
10603 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10604 {
10605 	int pending_vec, max_bits;
10606 	int mmu_reset_needed = 0;
10607 	int ret = __set_sregs_common(vcpu, sregs, &mmu_reset_needed, true);
10608 
10609 	if (ret)
10610 		return ret;
10611 
10612 	if (mmu_reset_needed)
10613 		kvm_mmu_reset_context(vcpu);
10614 
10615 	max_bits = KVM_NR_INTERRUPTS;
10616 	pending_vec = find_first_bit(
10617 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
10618 
10619 	if (pending_vec < max_bits) {
10620 		kvm_queue_interrupt(vcpu, pending_vec, false);
10621 		pr_debug("Set back pending irq %d\n", pending_vec);
10622 		kvm_make_request(KVM_REQ_EVENT, vcpu);
10623 	}
10624 	return 0;
10625 }
10626 
10627 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10628 {
10629 	int mmu_reset_needed = 0;
10630 	bool valid_pdptrs = sregs2->flags & KVM_SREGS2_FLAGS_PDPTRS_VALID;
10631 	bool pae = (sregs2->cr0 & X86_CR0_PG) && (sregs2->cr4 & X86_CR4_PAE) &&
10632 		!(sregs2->efer & EFER_LMA);
10633 	int i, ret;
10634 
10635 	if (sregs2->flags & ~KVM_SREGS2_FLAGS_PDPTRS_VALID)
10636 		return -EINVAL;
10637 
10638 	if (valid_pdptrs && (!pae || vcpu->arch.guest_state_protected))
10639 		return -EINVAL;
10640 
10641 	ret = __set_sregs_common(vcpu, (struct kvm_sregs *)sregs2,
10642 				 &mmu_reset_needed, !valid_pdptrs);
10643 	if (ret)
10644 		return ret;
10645 
10646 	if (valid_pdptrs) {
10647 		for (i = 0; i < 4 ; i++)
10648 			kvm_pdptr_write(vcpu, i, sregs2->pdptrs[i]);
10649 
10650 		kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
10651 		mmu_reset_needed = 1;
10652 		vcpu->arch.pdptrs_from_userspace = true;
10653 	}
10654 	if (mmu_reset_needed)
10655 		kvm_mmu_reset_context(vcpu);
10656 	return 0;
10657 }
10658 
10659 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
10660 				  struct kvm_sregs *sregs)
10661 {
10662 	int ret;
10663 
10664 	vcpu_load(vcpu);
10665 	ret = __set_sregs(vcpu, sregs);
10666 	vcpu_put(vcpu);
10667 	return ret;
10668 }
10669 
10670 static void kvm_arch_vcpu_guestdbg_update_apicv_inhibit(struct kvm *kvm)
10671 {
10672 	bool inhibit = false;
10673 	struct kvm_vcpu *vcpu;
10674 	int i;
10675 
10676 	down_write(&kvm->arch.apicv_update_lock);
10677 
10678 	kvm_for_each_vcpu(i, vcpu, kvm) {
10679 		if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ) {
10680 			inhibit = true;
10681 			break;
10682 		}
10683 	}
10684 	__kvm_request_apicv_update(kvm, !inhibit, APICV_INHIBIT_REASON_BLOCKIRQ);
10685 	up_write(&kvm->arch.apicv_update_lock);
10686 }
10687 
10688 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
10689 					struct kvm_guest_debug *dbg)
10690 {
10691 	unsigned long rflags;
10692 	int i, r;
10693 
10694 	if (vcpu->arch.guest_state_protected)
10695 		return -EINVAL;
10696 
10697 	vcpu_load(vcpu);
10698 
10699 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10700 		r = -EBUSY;
10701 		if (vcpu->arch.exception.pending)
10702 			goto out;
10703 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10704 			kvm_queue_exception(vcpu, DB_VECTOR);
10705 		else
10706 			kvm_queue_exception(vcpu, BP_VECTOR);
10707 	}
10708 
10709 	/*
10710 	 * Read rflags as long as potentially injected trace flags are still
10711 	 * filtered out.
10712 	 */
10713 	rflags = kvm_get_rflags(vcpu);
10714 
10715 	vcpu->guest_debug = dbg->control;
10716 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10717 		vcpu->guest_debug = 0;
10718 
10719 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10720 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
10721 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10722 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10723 	} else {
10724 		for (i = 0; i < KVM_NR_DB_REGS; i++)
10725 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10726 	}
10727 	kvm_update_dr7(vcpu);
10728 
10729 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10730 		vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
10731 
10732 	/*
10733 	 * Trigger an rflags update that will inject or remove the trace
10734 	 * flags.
10735 	 */
10736 	kvm_set_rflags(vcpu, rflags);
10737 
10738 	static_call(kvm_x86_update_exception_bitmap)(vcpu);
10739 
10740 	kvm_arch_vcpu_guestdbg_update_apicv_inhibit(vcpu->kvm);
10741 
10742 	r = 0;
10743 
10744 out:
10745 	vcpu_put(vcpu);
10746 	return r;
10747 }
10748 
10749 /*
10750  * Translate a guest virtual address to a guest physical address.
10751  */
10752 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10753 				    struct kvm_translation *tr)
10754 {
10755 	unsigned long vaddr = tr->linear_address;
10756 	gpa_t gpa;
10757 	int idx;
10758 
10759 	vcpu_load(vcpu);
10760 
10761 	idx = srcu_read_lock(&vcpu->kvm->srcu);
10762 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10763 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
10764 	tr->physical_address = gpa;
10765 	tr->valid = gpa != UNMAPPED_GVA;
10766 	tr->writeable = 1;
10767 	tr->usermode = 0;
10768 
10769 	vcpu_put(vcpu);
10770 	return 0;
10771 }
10772 
10773 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10774 {
10775 	struct fxregs_state *fxsave;
10776 
10777 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
10778 		return 0;
10779 
10780 	vcpu_load(vcpu);
10781 
10782 	fxsave = &vcpu->arch.guest_fpu.fpstate->regs.fxsave;
10783 	memcpy(fpu->fpr, fxsave->st_space, 128);
10784 	fpu->fcw = fxsave->cwd;
10785 	fpu->fsw = fxsave->swd;
10786 	fpu->ftwx = fxsave->twd;
10787 	fpu->last_opcode = fxsave->fop;
10788 	fpu->last_ip = fxsave->rip;
10789 	fpu->last_dp = fxsave->rdp;
10790 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10791 
10792 	vcpu_put(vcpu);
10793 	return 0;
10794 }
10795 
10796 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10797 {
10798 	struct fxregs_state *fxsave;
10799 
10800 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
10801 		return 0;
10802 
10803 	vcpu_load(vcpu);
10804 
10805 	fxsave = &vcpu->arch.guest_fpu.fpstate->regs.fxsave;
10806 
10807 	memcpy(fxsave->st_space, fpu->fpr, 128);
10808 	fxsave->cwd = fpu->fcw;
10809 	fxsave->swd = fpu->fsw;
10810 	fxsave->twd = fpu->ftwx;
10811 	fxsave->fop = fpu->last_opcode;
10812 	fxsave->rip = fpu->last_ip;
10813 	fxsave->rdp = fpu->last_dp;
10814 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10815 
10816 	vcpu_put(vcpu);
10817 	return 0;
10818 }
10819 
10820 static void store_regs(struct kvm_vcpu *vcpu)
10821 {
10822 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10823 
10824 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10825 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
10826 
10827 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10828 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10829 
10830 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10831 		kvm_vcpu_ioctl_x86_get_vcpu_events(
10832 				vcpu, &vcpu->run->s.regs.events);
10833 }
10834 
10835 static int sync_regs(struct kvm_vcpu *vcpu)
10836 {
10837 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10838 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
10839 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10840 	}
10841 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10842 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10843 			return -EINVAL;
10844 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10845 	}
10846 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10847 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10848 				vcpu, &vcpu->run->s.regs.events))
10849 			return -EINVAL;
10850 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10851 	}
10852 
10853 	return 0;
10854 }
10855 
10856 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10857 {
10858 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10859 		pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10860 			     "guest TSC will not be reliable\n");
10861 
10862 	return 0;
10863 }
10864 
10865 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10866 {
10867 	struct page *page;
10868 	int r;
10869 
10870 	vcpu->arch.last_vmentry_cpu = -1;
10871 	vcpu->arch.regs_avail = ~0;
10872 	vcpu->arch.regs_dirty = ~0;
10873 
10874 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10875 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10876 	else
10877 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10878 
10879 	r = kvm_mmu_create(vcpu);
10880 	if (r < 0)
10881 		return r;
10882 
10883 	if (irqchip_in_kernel(vcpu->kvm)) {
10884 		r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10885 		if (r < 0)
10886 			goto fail_mmu_destroy;
10887 		if (kvm_apicv_activated(vcpu->kvm))
10888 			vcpu->arch.apicv_active = true;
10889 	} else
10890 		static_branch_inc(&kvm_has_noapic_vcpu);
10891 
10892 	r = -ENOMEM;
10893 
10894 	page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10895 	if (!page)
10896 		goto fail_free_lapic;
10897 	vcpu->arch.pio_data = page_address(page);
10898 
10899 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10900 				       GFP_KERNEL_ACCOUNT);
10901 	if (!vcpu->arch.mce_banks)
10902 		goto fail_free_pio_data;
10903 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
10904 
10905 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
10906 				GFP_KERNEL_ACCOUNT))
10907 		goto fail_free_mce_banks;
10908 
10909 	if (!alloc_emulate_ctxt(vcpu))
10910 		goto free_wbinvd_dirty_mask;
10911 
10912 	if (!fpu_alloc_guest_fpstate(&vcpu->arch.guest_fpu)) {
10913 		pr_err("kvm: failed to allocate vcpu's fpu\n");
10914 		goto free_emulate_ctxt;
10915 	}
10916 
10917 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10918 	vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
10919 
10920 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10921 
10922 	kvm_async_pf_hash_reset(vcpu);
10923 	kvm_pmu_init(vcpu);
10924 
10925 	vcpu->arch.pending_external_vector = -1;
10926 	vcpu->arch.preempted_in_kernel = false;
10927 
10928 #if IS_ENABLED(CONFIG_HYPERV)
10929 	vcpu->arch.hv_root_tdp = INVALID_PAGE;
10930 #endif
10931 
10932 	r = static_call(kvm_x86_vcpu_create)(vcpu);
10933 	if (r)
10934 		goto free_guest_fpu;
10935 
10936 	vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10937 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10938 	kvm_vcpu_mtrr_init(vcpu);
10939 	vcpu_load(vcpu);
10940 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
10941 	kvm_vcpu_reset(vcpu, false);
10942 	kvm_init_mmu(vcpu);
10943 	vcpu_put(vcpu);
10944 	return 0;
10945 
10946 free_guest_fpu:
10947 	fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
10948 free_emulate_ctxt:
10949 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10950 free_wbinvd_dirty_mask:
10951 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10952 fail_free_mce_banks:
10953 	kfree(vcpu->arch.mce_banks);
10954 fail_free_pio_data:
10955 	free_page((unsigned long)vcpu->arch.pio_data);
10956 fail_free_lapic:
10957 	kvm_free_lapic(vcpu);
10958 fail_mmu_destroy:
10959 	kvm_mmu_destroy(vcpu);
10960 	return r;
10961 }
10962 
10963 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10964 {
10965 	struct kvm *kvm = vcpu->kvm;
10966 
10967 	if (mutex_lock_killable(&vcpu->mutex))
10968 		return;
10969 	vcpu_load(vcpu);
10970 	kvm_synchronize_tsc(vcpu, 0);
10971 	vcpu_put(vcpu);
10972 
10973 	/* poll control enabled by default */
10974 	vcpu->arch.msr_kvm_poll_control = 1;
10975 
10976 	mutex_unlock(&vcpu->mutex);
10977 
10978 	if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10979 		schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10980 						KVMCLOCK_SYNC_PERIOD);
10981 }
10982 
10983 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10984 {
10985 	int idx;
10986 
10987 	kvmclock_reset(vcpu);
10988 
10989 	static_call(kvm_x86_vcpu_free)(vcpu);
10990 
10991 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10992 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10993 	fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
10994 
10995 	kvm_hv_vcpu_uninit(vcpu);
10996 	kvm_pmu_destroy(vcpu);
10997 	kfree(vcpu->arch.mce_banks);
10998 	kvm_free_lapic(vcpu);
10999 	idx = srcu_read_lock(&vcpu->kvm->srcu);
11000 	kvm_mmu_destroy(vcpu);
11001 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
11002 	free_page((unsigned long)vcpu->arch.pio_data);
11003 	kvfree(vcpu->arch.cpuid_entries);
11004 	if (!lapic_in_kernel(vcpu))
11005 		static_branch_dec(&kvm_has_noapic_vcpu);
11006 }
11007 
11008 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
11009 {
11010 	struct kvm_cpuid_entry2 *cpuid_0x1;
11011 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
11012 	unsigned long new_cr0;
11013 
11014 	/*
11015 	 * Several of the "set" flows, e.g. ->set_cr0(), read other registers
11016 	 * to handle side effects.  RESET emulation hits those flows and relies
11017 	 * on emulated/virtualized registers, including those that are loaded
11018 	 * into hardware, to be zeroed at vCPU creation.  Use CRs as a sentinel
11019 	 * to detect improper or missing initialization.
11020 	 */
11021 	WARN_ON_ONCE(!init_event &&
11022 		     (old_cr0 || kvm_read_cr3(vcpu) || kvm_read_cr4(vcpu)));
11023 
11024 	kvm_lapic_reset(vcpu, init_event);
11025 
11026 	vcpu->arch.hflags = 0;
11027 
11028 	vcpu->arch.smi_pending = 0;
11029 	vcpu->arch.smi_count = 0;
11030 	atomic_set(&vcpu->arch.nmi_queued, 0);
11031 	vcpu->arch.nmi_pending = 0;
11032 	vcpu->arch.nmi_injected = false;
11033 	kvm_clear_interrupt_queue(vcpu);
11034 	kvm_clear_exception_queue(vcpu);
11035 
11036 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
11037 	kvm_update_dr0123(vcpu);
11038 	vcpu->arch.dr6 = DR6_ACTIVE_LOW;
11039 	vcpu->arch.dr7 = DR7_FIXED_1;
11040 	kvm_update_dr7(vcpu);
11041 
11042 	vcpu->arch.cr2 = 0;
11043 
11044 	kvm_make_request(KVM_REQ_EVENT, vcpu);
11045 	vcpu->arch.apf.msr_en_val = 0;
11046 	vcpu->arch.apf.msr_int_val = 0;
11047 	vcpu->arch.st.msr_val = 0;
11048 
11049 	kvmclock_reset(vcpu);
11050 
11051 	kvm_clear_async_pf_completion_queue(vcpu);
11052 	kvm_async_pf_hash_reset(vcpu);
11053 	vcpu->arch.apf.halted = false;
11054 
11055 	if (vcpu->arch.guest_fpu.fpstate && kvm_mpx_supported()) {
11056 		struct fpstate *fpstate = vcpu->arch.guest_fpu.fpstate;
11057 
11058 		/*
11059 		 * To avoid have the INIT path from kvm_apic_has_events() that be
11060 		 * called with loaded FPU and does not let userspace fix the state.
11061 		 */
11062 		if (init_event)
11063 			kvm_put_guest_fpu(vcpu);
11064 
11065 		fpstate_clear_xstate_component(fpstate, XFEATURE_BNDREGS);
11066 		fpstate_clear_xstate_component(fpstate, XFEATURE_BNDCSR);
11067 
11068 		if (init_event)
11069 			kvm_load_guest_fpu(vcpu);
11070 	}
11071 
11072 	if (!init_event) {
11073 		kvm_pmu_reset(vcpu);
11074 		vcpu->arch.smbase = 0x30000;
11075 
11076 		vcpu->arch.msr_misc_features_enables = 0;
11077 
11078 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
11079 	}
11080 
11081 	/* All GPRs except RDX (handled below) are zeroed on RESET/INIT. */
11082 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
11083 	kvm_register_mark_dirty(vcpu, VCPU_REGS_RSP);
11084 
11085 	/*
11086 	 * Fall back to KVM's default Family/Model/Stepping of 0x600 (P6/Athlon)
11087 	 * if no CPUID match is found.  Note, it's impossible to get a match at
11088 	 * RESET since KVM emulates RESET before exposing the vCPU to userspace,
11089 	 * i.e. it's impossible for kvm_find_cpuid_entry() to find a valid entry
11090 	 * on RESET.  But, go through the motions in case that's ever remedied.
11091 	 */
11092 	cpuid_0x1 = kvm_find_cpuid_entry(vcpu, 1, 0);
11093 	kvm_rdx_write(vcpu, cpuid_0x1 ? cpuid_0x1->eax : 0x600);
11094 
11095 	vcpu->arch.ia32_xss = 0;
11096 
11097 	static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
11098 
11099 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
11100 	kvm_rip_write(vcpu, 0xfff0);
11101 
11102 	vcpu->arch.cr3 = 0;
11103 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
11104 
11105 	/*
11106 	 * CR0.CD/NW are set on RESET, preserved on INIT.  Note, some versions
11107 	 * of Intel's SDM list CD/NW as being set on INIT, but they contradict
11108 	 * (or qualify) that with a footnote stating that CD/NW are preserved.
11109 	 */
11110 	new_cr0 = X86_CR0_ET;
11111 	if (init_event)
11112 		new_cr0 |= (old_cr0 & (X86_CR0_NW | X86_CR0_CD));
11113 	else
11114 		new_cr0 |= X86_CR0_NW | X86_CR0_CD;
11115 
11116 	static_call(kvm_x86_set_cr0)(vcpu, new_cr0);
11117 	static_call(kvm_x86_set_cr4)(vcpu, 0);
11118 	static_call(kvm_x86_set_efer)(vcpu, 0);
11119 	static_call(kvm_x86_update_exception_bitmap)(vcpu);
11120 
11121 	/*
11122 	 * Reset the MMU context if paging was enabled prior to INIT (which is
11123 	 * implied if CR0.PG=1 as CR0 will be '0' prior to RESET).  Unlike the
11124 	 * standard CR0/CR4/EFER modification paths, only CR0.PG needs to be
11125 	 * checked because it is unconditionally cleared on INIT and all other
11126 	 * paging related bits are ignored if paging is disabled, i.e. CR0.WP,
11127 	 * CR4, and EFER changes are all irrelevant if CR0.PG was '0'.
11128 	 */
11129 	if (old_cr0 & X86_CR0_PG)
11130 		kvm_mmu_reset_context(vcpu);
11131 
11132 	/*
11133 	 * Intel's SDM states that all TLB entries are flushed on INIT.  AMD's
11134 	 * APM states the TLBs are untouched by INIT, but it also states that
11135 	 * the TLBs are flushed on "External initialization of the processor."
11136 	 * Flush the guest TLB regardless of vendor, there is no meaningful
11137 	 * benefit in relying on the guest to flush the TLB immediately after
11138 	 * INIT.  A spurious TLB flush is benign and likely negligible from a
11139 	 * performance perspective.
11140 	 */
11141 	if (init_event)
11142 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
11143 }
11144 EXPORT_SYMBOL_GPL(kvm_vcpu_reset);
11145 
11146 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
11147 {
11148 	struct kvm_segment cs;
11149 
11150 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
11151 	cs.selector = vector << 8;
11152 	cs.base = vector << 12;
11153 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
11154 	kvm_rip_write(vcpu, 0);
11155 }
11156 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
11157 
11158 int kvm_arch_hardware_enable(void)
11159 {
11160 	struct kvm *kvm;
11161 	struct kvm_vcpu *vcpu;
11162 	int i;
11163 	int ret;
11164 	u64 local_tsc;
11165 	u64 max_tsc = 0;
11166 	bool stable, backwards_tsc = false;
11167 
11168 	kvm_user_return_msr_cpu_online();
11169 	ret = static_call(kvm_x86_hardware_enable)();
11170 	if (ret != 0)
11171 		return ret;
11172 
11173 	local_tsc = rdtsc();
11174 	stable = !kvm_check_tsc_unstable();
11175 	list_for_each_entry(kvm, &vm_list, vm_list) {
11176 		kvm_for_each_vcpu(i, vcpu, kvm) {
11177 			if (!stable && vcpu->cpu == smp_processor_id())
11178 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
11179 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
11180 				backwards_tsc = true;
11181 				if (vcpu->arch.last_host_tsc > max_tsc)
11182 					max_tsc = vcpu->arch.last_host_tsc;
11183 			}
11184 		}
11185 	}
11186 
11187 	/*
11188 	 * Sometimes, even reliable TSCs go backwards.  This happens on
11189 	 * platforms that reset TSC during suspend or hibernate actions, but
11190 	 * maintain synchronization.  We must compensate.  Fortunately, we can
11191 	 * detect that condition here, which happens early in CPU bringup,
11192 	 * before any KVM threads can be running.  Unfortunately, we can't
11193 	 * bring the TSCs fully up to date with real time, as we aren't yet far
11194 	 * enough into CPU bringup that we know how much real time has actually
11195 	 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
11196 	 * variables that haven't been updated yet.
11197 	 *
11198 	 * So we simply find the maximum observed TSC above, then record the
11199 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
11200 	 * the adjustment will be applied.  Note that we accumulate
11201 	 * adjustments, in case multiple suspend cycles happen before some VCPU
11202 	 * gets a chance to run again.  In the event that no KVM threads get a
11203 	 * chance to run, we will miss the entire elapsed period, as we'll have
11204 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
11205 	 * loose cycle time.  This isn't too big a deal, since the loss will be
11206 	 * uniform across all VCPUs (not to mention the scenario is extremely
11207 	 * unlikely). It is possible that a second hibernate recovery happens
11208 	 * much faster than a first, causing the observed TSC here to be
11209 	 * smaller; this would require additional padding adjustment, which is
11210 	 * why we set last_host_tsc to the local tsc observed here.
11211 	 *
11212 	 * N.B. - this code below runs only on platforms with reliable TSC,
11213 	 * as that is the only way backwards_tsc is set above.  Also note
11214 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
11215 	 * have the same delta_cyc adjustment applied if backwards_tsc
11216 	 * is detected.  Note further, this adjustment is only done once,
11217 	 * as we reset last_host_tsc on all VCPUs to stop this from being
11218 	 * called multiple times (one for each physical CPU bringup).
11219 	 *
11220 	 * Platforms with unreliable TSCs don't have to deal with this, they
11221 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
11222 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
11223 	 * guarantee that they stay in perfect synchronization.
11224 	 */
11225 	if (backwards_tsc) {
11226 		u64 delta_cyc = max_tsc - local_tsc;
11227 		list_for_each_entry(kvm, &vm_list, vm_list) {
11228 			kvm->arch.backwards_tsc_observed = true;
11229 			kvm_for_each_vcpu(i, vcpu, kvm) {
11230 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
11231 				vcpu->arch.last_host_tsc = local_tsc;
11232 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
11233 			}
11234 
11235 			/*
11236 			 * We have to disable TSC offset matching.. if you were
11237 			 * booting a VM while issuing an S4 host suspend....
11238 			 * you may have some problem.  Solving this issue is
11239 			 * left as an exercise to the reader.
11240 			 */
11241 			kvm->arch.last_tsc_nsec = 0;
11242 			kvm->arch.last_tsc_write = 0;
11243 		}
11244 
11245 	}
11246 	return 0;
11247 }
11248 
11249 void kvm_arch_hardware_disable(void)
11250 {
11251 	static_call(kvm_x86_hardware_disable)();
11252 	drop_user_return_notifiers();
11253 }
11254 
11255 int kvm_arch_hardware_setup(void *opaque)
11256 {
11257 	struct kvm_x86_init_ops *ops = opaque;
11258 	int r;
11259 
11260 	rdmsrl_safe(MSR_EFER, &host_efer);
11261 
11262 	if (boot_cpu_has(X86_FEATURE_XSAVES))
11263 		rdmsrl(MSR_IA32_XSS, host_xss);
11264 
11265 	r = ops->hardware_setup();
11266 	if (r != 0)
11267 		return r;
11268 
11269 	memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
11270 	kvm_ops_static_call_update();
11271 
11272 	if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
11273 		supported_xss = 0;
11274 
11275 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
11276 	cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
11277 #undef __kvm_cpu_cap_has
11278 
11279 	if (kvm_has_tsc_control) {
11280 		/*
11281 		 * Make sure the user can only configure tsc_khz values that
11282 		 * fit into a signed integer.
11283 		 * A min value is not calculated because it will always
11284 		 * be 1 on all machines.
11285 		 */
11286 		u64 max = min(0x7fffffffULL,
11287 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
11288 		kvm_max_guest_tsc_khz = max;
11289 
11290 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
11291 	}
11292 
11293 	kvm_init_msr_list();
11294 	return 0;
11295 }
11296 
11297 void kvm_arch_hardware_unsetup(void)
11298 {
11299 	static_call(kvm_x86_hardware_unsetup)();
11300 }
11301 
11302 int kvm_arch_check_processor_compat(void *opaque)
11303 {
11304 	struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
11305 	struct kvm_x86_init_ops *ops = opaque;
11306 
11307 	WARN_ON(!irqs_disabled());
11308 
11309 	if (__cr4_reserved_bits(cpu_has, c) !=
11310 	    __cr4_reserved_bits(cpu_has, &boot_cpu_data))
11311 		return -EIO;
11312 
11313 	return ops->check_processor_compatibility();
11314 }
11315 
11316 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
11317 {
11318 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
11319 }
11320 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
11321 
11322 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
11323 {
11324 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
11325 }
11326 
11327 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
11328 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
11329 
11330 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
11331 {
11332 	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
11333 
11334 	vcpu->arch.l1tf_flush_l1d = true;
11335 	if (pmu->version && unlikely(pmu->event_count)) {
11336 		pmu->need_cleanup = true;
11337 		kvm_make_request(KVM_REQ_PMU, vcpu);
11338 	}
11339 	static_call(kvm_x86_sched_in)(vcpu, cpu);
11340 }
11341 
11342 void kvm_arch_free_vm(struct kvm *kvm)
11343 {
11344 	kfree(to_kvm_hv(kvm)->hv_pa_pg);
11345 	__kvm_arch_free_vm(kvm);
11346 }
11347 
11348 
11349 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
11350 {
11351 	int ret;
11352 	unsigned long flags;
11353 
11354 	if (type)
11355 		return -EINVAL;
11356 
11357 	ret = kvm_page_track_init(kvm);
11358 	if (ret)
11359 		return ret;
11360 
11361 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
11362 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
11363 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
11364 	INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
11365 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
11366 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
11367 
11368 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
11369 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
11370 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
11371 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
11372 		&kvm->arch.irq_sources_bitmap);
11373 
11374 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
11375 	mutex_init(&kvm->arch.apic_map_lock);
11376 	seqcount_raw_spinlock_init(&kvm->arch.pvclock_sc, &kvm->arch.tsc_write_lock);
11377 	kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
11378 
11379 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
11380 	pvclock_update_vm_gtod_copy(kvm);
11381 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
11382 
11383 	kvm->arch.guest_can_read_msr_platform_info = true;
11384 
11385 #if IS_ENABLED(CONFIG_HYPERV)
11386 	spin_lock_init(&kvm->arch.hv_root_tdp_lock);
11387 	kvm->arch.hv_root_tdp = INVALID_PAGE;
11388 #endif
11389 
11390 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
11391 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
11392 
11393 	kvm_apicv_init(kvm);
11394 	kvm_hv_init_vm(kvm);
11395 	kvm_mmu_init_vm(kvm);
11396 	kvm_xen_init_vm(kvm);
11397 
11398 	return static_call(kvm_x86_vm_init)(kvm);
11399 }
11400 
11401 int kvm_arch_post_init_vm(struct kvm *kvm)
11402 {
11403 	return kvm_mmu_post_init_vm(kvm);
11404 }
11405 
11406 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
11407 {
11408 	vcpu_load(vcpu);
11409 	kvm_mmu_unload(vcpu);
11410 	vcpu_put(vcpu);
11411 }
11412 
11413 static void kvm_free_vcpus(struct kvm *kvm)
11414 {
11415 	unsigned int i;
11416 	struct kvm_vcpu *vcpu;
11417 
11418 	/*
11419 	 * Unpin any mmu pages first.
11420 	 */
11421 	kvm_for_each_vcpu(i, vcpu, kvm) {
11422 		kvm_clear_async_pf_completion_queue(vcpu);
11423 		kvm_unload_vcpu_mmu(vcpu);
11424 	}
11425 	kvm_for_each_vcpu(i, vcpu, kvm)
11426 		kvm_vcpu_destroy(vcpu);
11427 
11428 	mutex_lock(&kvm->lock);
11429 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
11430 		kvm->vcpus[i] = NULL;
11431 
11432 	atomic_set(&kvm->online_vcpus, 0);
11433 	mutex_unlock(&kvm->lock);
11434 }
11435 
11436 void kvm_arch_sync_events(struct kvm *kvm)
11437 {
11438 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
11439 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
11440 	kvm_free_pit(kvm);
11441 }
11442 
11443 #define  ERR_PTR_USR(e)  ((void __user *)ERR_PTR(e))
11444 
11445 /**
11446  * __x86_set_memory_region: Setup KVM internal memory slot
11447  *
11448  * @kvm: the kvm pointer to the VM.
11449  * @id: the slot ID to setup.
11450  * @gpa: the GPA to install the slot (unused when @size == 0).
11451  * @size: the size of the slot. Set to zero to uninstall a slot.
11452  *
11453  * This function helps to setup a KVM internal memory slot.  Specify
11454  * @size > 0 to install a new slot, while @size == 0 to uninstall a
11455  * slot.  The return code can be one of the following:
11456  *
11457  *   HVA:           on success (uninstall will return a bogus HVA)
11458  *   -errno:        on error
11459  *
11460  * The caller should always use IS_ERR() to check the return value
11461  * before use.  Note, the KVM internal memory slots are guaranteed to
11462  * remain valid and unchanged until the VM is destroyed, i.e., the
11463  * GPA->HVA translation will not change.  However, the HVA is a user
11464  * address, i.e. its accessibility is not guaranteed, and must be
11465  * accessed via __copy_{to,from}_user().
11466  */
11467 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
11468 				      u32 size)
11469 {
11470 	int i, r;
11471 	unsigned long hva, old_npages;
11472 	struct kvm_memslots *slots = kvm_memslots(kvm);
11473 	struct kvm_memory_slot *slot;
11474 
11475 	/* Called with kvm->slots_lock held.  */
11476 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
11477 		return ERR_PTR_USR(-EINVAL);
11478 
11479 	slot = id_to_memslot(slots, id);
11480 	if (size) {
11481 		if (slot && slot->npages)
11482 			return ERR_PTR_USR(-EEXIST);
11483 
11484 		/*
11485 		 * MAP_SHARED to prevent internal slot pages from being moved
11486 		 * by fork()/COW.
11487 		 */
11488 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
11489 			      MAP_SHARED | MAP_ANONYMOUS, 0);
11490 		if (IS_ERR((void *)hva))
11491 			return (void __user *)hva;
11492 	} else {
11493 		if (!slot || !slot->npages)
11494 			return NULL;
11495 
11496 		old_npages = slot->npages;
11497 		hva = slot->userspace_addr;
11498 	}
11499 
11500 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11501 		struct kvm_userspace_memory_region m;
11502 
11503 		m.slot = id | (i << 16);
11504 		m.flags = 0;
11505 		m.guest_phys_addr = gpa;
11506 		m.userspace_addr = hva;
11507 		m.memory_size = size;
11508 		r = __kvm_set_memory_region(kvm, &m);
11509 		if (r < 0)
11510 			return ERR_PTR_USR(r);
11511 	}
11512 
11513 	if (!size)
11514 		vm_munmap(hva, old_npages * PAGE_SIZE);
11515 
11516 	return (void __user *)hva;
11517 }
11518 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
11519 
11520 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
11521 {
11522 	kvm_mmu_pre_destroy_vm(kvm);
11523 }
11524 
11525 void kvm_arch_destroy_vm(struct kvm *kvm)
11526 {
11527 	if (current->mm == kvm->mm) {
11528 		/*
11529 		 * Free memory regions allocated on behalf of userspace,
11530 		 * unless the the memory map has changed due to process exit
11531 		 * or fd copying.
11532 		 */
11533 		mutex_lock(&kvm->slots_lock);
11534 		__x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
11535 					0, 0);
11536 		__x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
11537 					0, 0);
11538 		__x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
11539 		mutex_unlock(&kvm->slots_lock);
11540 	}
11541 	static_call_cond(kvm_x86_vm_destroy)(kvm);
11542 	kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
11543 	kvm_pic_destroy(kvm);
11544 	kvm_ioapic_destroy(kvm);
11545 	kvm_free_vcpus(kvm);
11546 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
11547 	kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
11548 	kvm_mmu_uninit_vm(kvm);
11549 	kvm_page_track_cleanup(kvm);
11550 	kvm_xen_destroy_vm(kvm);
11551 	kvm_hv_destroy_vm(kvm);
11552 }
11553 
11554 static void memslot_rmap_free(struct kvm_memory_slot *slot)
11555 {
11556 	int i;
11557 
11558 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11559 		kvfree(slot->arch.rmap[i]);
11560 		slot->arch.rmap[i] = NULL;
11561 	}
11562 }
11563 
11564 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
11565 {
11566 	int i;
11567 
11568 	memslot_rmap_free(slot);
11569 
11570 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11571 		kvfree(slot->arch.lpage_info[i - 1]);
11572 		slot->arch.lpage_info[i - 1] = NULL;
11573 	}
11574 
11575 	kvm_page_track_free_memslot(slot);
11576 }
11577 
11578 int memslot_rmap_alloc(struct kvm_memory_slot *slot, unsigned long npages)
11579 {
11580 	const int sz = sizeof(*slot->arch.rmap[0]);
11581 	int i;
11582 
11583 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11584 		int level = i + 1;
11585 		int lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11586 
11587 		if (slot->arch.rmap[i])
11588 			continue;
11589 
11590 		slot->arch.rmap[i] = kvcalloc(lpages, sz, GFP_KERNEL_ACCOUNT);
11591 		if (!slot->arch.rmap[i]) {
11592 			memslot_rmap_free(slot);
11593 			return -ENOMEM;
11594 		}
11595 	}
11596 
11597 	return 0;
11598 }
11599 
11600 static int kvm_alloc_memslot_metadata(struct kvm *kvm,
11601 				      struct kvm_memory_slot *slot,
11602 				      unsigned long npages)
11603 {
11604 	int i, r;
11605 
11606 	/*
11607 	 * Clear out the previous array pointers for the KVM_MR_MOVE case.  The
11608 	 * old arrays will be freed by __kvm_set_memory_region() if installing
11609 	 * the new memslot is successful.
11610 	 */
11611 	memset(&slot->arch, 0, sizeof(slot->arch));
11612 
11613 	if (kvm_memslots_have_rmaps(kvm)) {
11614 		r = memslot_rmap_alloc(slot, npages);
11615 		if (r)
11616 			return r;
11617 	}
11618 
11619 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11620 		struct kvm_lpage_info *linfo;
11621 		unsigned long ugfn;
11622 		int lpages;
11623 		int level = i + 1;
11624 
11625 		lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11626 
11627 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
11628 		if (!linfo)
11629 			goto out_free;
11630 
11631 		slot->arch.lpage_info[i - 1] = linfo;
11632 
11633 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
11634 			linfo[0].disallow_lpage = 1;
11635 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
11636 			linfo[lpages - 1].disallow_lpage = 1;
11637 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
11638 		/*
11639 		 * If the gfn and userspace address are not aligned wrt each
11640 		 * other, disable large page support for this slot.
11641 		 */
11642 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
11643 			unsigned long j;
11644 
11645 			for (j = 0; j < lpages; ++j)
11646 				linfo[j].disallow_lpage = 1;
11647 		}
11648 	}
11649 
11650 	if (kvm_page_track_create_memslot(kvm, slot, npages))
11651 		goto out_free;
11652 
11653 	return 0;
11654 
11655 out_free:
11656 	memslot_rmap_free(slot);
11657 
11658 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11659 		kvfree(slot->arch.lpage_info[i - 1]);
11660 		slot->arch.lpage_info[i - 1] = NULL;
11661 	}
11662 	return -ENOMEM;
11663 }
11664 
11665 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
11666 {
11667 	struct kvm_vcpu *vcpu;
11668 	int i;
11669 
11670 	/*
11671 	 * memslots->generation has been incremented.
11672 	 * mmio generation may have reached its maximum value.
11673 	 */
11674 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
11675 
11676 	/* Force re-initialization of steal_time cache */
11677 	kvm_for_each_vcpu(i, vcpu, kvm)
11678 		kvm_vcpu_kick(vcpu);
11679 }
11680 
11681 int kvm_arch_prepare_memory_region(struct kvm *kvm,
11682 				struct kvm_memory_slot *memslot,
11683 				const struct kvm_userspace_memory_region *mem,
11684 				enum kvm_mr_change change)
11685 {
11686 	if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
11687 		return kvm_alloc_memslot_metadata(kvm, memslot,
11688 						  mem->memory_size >> PAGE_SHIFT);
11689 	return 0;
11690 }
11691 
11692 
11693 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
11694 {
11695 	struct kvm_arch *ka = &kvm->arch;
11696 
11697 	if (!kvm_x86_ops.cpu_dirty_log_size)
11698 		return;
11699 
11700 	if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
11701 	    (!enable && --ka->cpu_dirty_logging_count == 0))
11702 		kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
11703 
11704 	WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
11705 }
11706 
11707 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
11708 				     struct kvm_memory_slot *old,
11709 				     const struct kvm_memory_slot *new,
11710 				     enum kvm_mr_change change)
11711 {
11712 	bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES;
11713 
11714 	/*
11715 	 * Update CPU dirty logging if dirty logging is being toggled.  This
11716 	 * applies to all operations.
11717 	 */
11718 	if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)
11719 		kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
11720 
11721 	/*
11722 	 * Nothing more to do for RO slots (which can't be dirtied and can't be
11723 	 * made writable) or CREATE/MOVE/DELETE of a slot.
11724 	 *
11725 	 * For a memslot with dirty logging disabled:
11726 	 * CREATE:      No dirty mappings will already exist.
11727 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
11728 	 *		kvm_arch_flush_shadow_memslot()
11729 	 *
11730 	 * For a memslot with dirty logging enabled:
11731 	 * CREATE:      No shadow pages exist, thus nothing to write-protect
11732 	 *		and no dirty bits to clear.
11733 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
11734 	 *		kvm_arch_flush_shadow_memslot().
11735 	 */
11736 	if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
11737 		return;
11738 
11739 	/*
11740 	 * READONLY and non-flags changes were filtered out above, and the only
11741 	 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
11742 	 * logging isn't being toggled on or off.
11743 	 */
11744 	if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)))
11745 		return;
11746 
11747 	if (!log_dirty_pages) {
11748 		/*
11749 		 * Dirty logging tracks sptes in 4k granularity, meaning that
11750 		 * large sptes have to be split.  If live migration succeeds,
11751 		 * the guest in the source machine will be destroyed and large
11752 		 * sptes will be created in the destination.  However, if the
11753 		 * guest continues to run in the source machine (for example if
11754 		 * live migration fails), small sptes will remain around and
11755 		 * cause bad performance.
11756 		 *
11757 		 * Scan sptes if dirty logging has been stopped, dropping those
11758 		 * which can be collapsed into a single large-page spte.  Later
11759 		 * page faults will create the large-page sptes.
11760 		 */
11761 		kvm_mmu_zap_collapsible_sptes(kvm, new);
11762 	} else {
11763 		/*
11764 		 * Initially-all-set does not require write protecting any page,
11765 		 * because they're all assumed to be dirty.
11766 		 */
11767 		if (kvm_dirty_log_manual_protect_and_init_set(kvm))
11768 			return;
11769 
11770 		if (kvm_x86_ops.cpu_dirty_log_size) {
11771 			kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11772 			kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M);
11773 		} else {
11774 			kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
11775 		}
11776 	}
11777 }
11778 
11779 void kvm_arch_commit_memory_region(struct kvm *kvm,
11780 				const struct kvm_userspace_memory_region *mem,
11781 				struct kvm_memory_slot *old,
11782 				const struct kvm_memory_slot *new,
11783 				enum kvm_mr_change change)
11784 {
11785 	if (!kvm->arch.n_requested_mmu_pages)
11786 		kvm_mmu_change_mmu_pages(kvm,
11787 				kvm_mmu_calculate_default_mmu_pages(kvm));
11788 
11789 	kvm_mmu_slot_apply_flags(kvm, old, new, change);
11790 
11791 	/* Free the arrays associated with the old memslot. */
11792 	if (change == KVM_MR_MOVE)
11793 		kvm_arch_free_memslot(kvm, old);
11794 }
11795 
11796 void kvm_arch_flush_shadow_all(struct kvm *kvm)
11797 {
11798 	kvm_mmu_zap_all(kvm);
11799 }
11800 
11801 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
11802 				   struct kvm_memory_slot *slot)
11803 {
11804 	kvm_page_track_flush_slot(kvm, slot);
11805 }
11806 
11807 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
11808 {
11809 	return (is_guest_mode(vcpu) &&
11810 			kvm_x86_ops.guest_apic_has_interrupt &&
11811 			static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
11812 }
11813 
11814 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
11815 {
11816 	if (!list_empty_careful(&vcpu->async_pf.done))
11817 		return true;
11818 
11819 	if (kvm_apic_has_events(vcpu))
11820 		return true;
11821 
11822 	if (vcpu->arch.pv.pv_unhalted)
11823 		return true;
11824 
11825 	if (vcpu->arch.exception.pending)
11826 		return true;
11827 
11828 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11829 	    (vcpu->arch.nmi_pending &&
11830 	     static_call(kvm_x86_nmi_allowed)(vcpu, false)))
11831 		return true;
11832 
11833 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
11834 	    (vcpu->arch.smi_pending &&
11835 	     static_call(kvm_x86_smi_allowed)(vcpu, false)))
11836 		return true;
11837 
11838 	if (kvm_arch_interrupt_allowed(vcpu) &&
11839 	    (kvm_cpu_has_interrupt(vcpu) ||
11840 	    kvm_guest_apic_has_interrupt(vcpu)))
11841 		return true;
11842 
11843 	if (kvm_hv_has_stimer_pending(vcpu))
11844 		return true;
11845 
11846 	if (is_guest_mode(vcpu) &&
11847 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
11848 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
11849 		return true;
11850 
11851 	return false;
11852 }
11853 
11854 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
11855 {
11856 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
11857 }
11858 
11859 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
11860 {
11861 	if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11862 		return true;
11863 
11864 	return false;
11865 }
11866 
11867 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
11868 {
11869 	if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
11870 		return true;
11871 
11872 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11873 		kvm_test_request(KVM_REQ_SMI, vcpu) ||
11874 		 kvm_test_request(KVM_REQ_EVENT, vcpu))
11875 		return true;
11876 
11877 	return kvm_arch_dy_has_pending_interrupt(vcpu);
11878 }
11879 
11880 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
11881 {
11882 	if (vcpu->arch.guest_state_protected)
11883 		return true;
11884 
11885 	return vcpu->arch.preempted_in_kernel;
11886 }
11887 
11888 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
11889 {
11890 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
11891 }
11892 
11893 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
11894 {
11895 	return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
11896 }
11897 
11898 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
11899 {
11900 	/* Can't read the RIP when guest state is protected, just return 0 */
11901 	if (vcpu->arch.guest_state_protected)
11902 		return 0;
11903 
11904 	if (is_64_bit_mode(vcpu))
11905 		return kvm_rip_read(vcpu);
11906 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
11907 		     kvm_rip_read(vcpu));
11908 }
11909 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
11910 
11911 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
11912 {
11913 	return kvm_get_linear_rip(vcpu) == linear_rip;
11914 }
11915 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
11916 
11917 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
11918 {
11919 	unsigned long rflags;
11920 
11921 	rflags = static_call(kvm_x86_get_rflags)(vcpu);
11922 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11923 		rflags &= ~X86_EFLAGS_TF;
11924 	return rflags;
11925 }
11926 EXPORT_SYMBOL_GPL(kvm_get_rflags);
11927 
11928 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11929 {
11930 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
11931 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
11932 		rflags |= X86_EFLAGS_TF;
11933 	static_call(kvm_x86_set_rflags)(vcpu, rflags);
11934 }
11935 
11936 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11937 {
11938 	__kvm_set_rflags(vcpu, rflags);
11939 	kvm_make_request(KVM_REQ_EVENT, vcpu);
11940 }
11941 EXPORT_SYMBOL_GPL(kvm_set_rflags);
11942 
11943 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
11944 {
11945 	int r;
11946 
11947 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
11948 	      work->wakeup_all)
11949 		return;
11950 
11951 	r = kvm_mmu_reload(vcpu);
11952 	if (unlikely(r))
11953 		return;
11954 
11955 	if (!vcpu->arch.mmu->direct_map &&
11956 	      work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
11957 		return;
11958 
11959 	kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
11960 }
11961 
11962 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
11963 {
11964 	BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
11965 
11966 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
11967 }
11968 
11969 static inline u32 kvm_async_pf_next_probe(u32 key)
11970 {
11971 	return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
11972 }
11973 
11974 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11975 {
11976 	u32 key = kvm_async_pf_hash_fn(gfn);
11977 
11978 	while (vcpu->arch.apf.gfns[key] != ~0)
11979 		key = kvm_async_pf_next_probe(key);
11980 
11981 	vcpu->arch.apf.gfns[key] = gfn;
11982 }
11983 
11984 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
11985 {
11986 	int i;
11987 	u32 key = kvm_async_pf_hash_fn(gfn);
11988 
11989 	for (i = 0; i < ASYNC_PF_PER_VCPU &&
11990 		     (vcpu->arch.apf.gfns[key] != gfn &&
11991 		      vcpu->arch.apf.gfns[key] != ~0); i++)
11992 		key = kvm_async_pf_next_probe(key);
11993 
11994 	return key;
11995 }
11996 
11997 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11998 {
11999 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
12000 }
12001 
12002 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12003 {
12004 	u32 i, j, k;
12005 
12006 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
12007 
12008 	if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
12009 		return;
12010 
12011 	while (true) {
12012 		vcpu->arch.apf.gfns[i] = ~0;
12013 		do {
12014 			j = kvm_async_pf_next_probe(j);
12015 			if (vcpu->arch.apf.gfns[j] == ~0)
12016 				return;
12017 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
12018 			/*
12019 			 * k lies cyclically in ]i,j]
12020 			 * |    i.k.j |
12021 			 * |....j i.k.| or  |.k..j i...|
12022 			 */
12023 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
12024 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
12025 		i = j;
12026 	}
12027 }
12028 
12029 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
12030 {
12031 	u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
12032 
12033 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
12034 				      sizeof(reason));
12035 }
12036 
12037 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
12038 {
12039 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
12040 
12041 	return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
12042 					     &token, offset, sizeof(token));
12043 }
12044 
12045 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
12046 {
12047 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
12048 	u32 val;
12049 
12050 	if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
12051 					 &val, offset, sizeof(val)))
12052 		return false;
12053 
12054 	return !val;
12055 }
12056 
12057 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
12058 {
12059 	if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
12060 		return false;
12061 
12062 	if (!kvm_pv_async_pf_enabled(vcpu) ||
12063 	    (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
12064 		return false;
12065 
12066 	return true;
12067 }
12068 
12069 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
12070 {
12071 	if (unlikely(!lapic_in_kernel(vcpu) ||
12072 		     kvm_event_needs_reinjection(vcpu) ||
12073 		     vcpu->arch.exception.pending))
12074 		return false;
12075 
12076 	if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
12077 		return false;
12078 
12079 	/*
12080 	 * If interrupts are off we cannot even use an artificial
12081 	 * halt state.
12082 	 */
12083 	return kvm_arch_interrupt_allowed(vcpu);
12084 }
12085 
12086 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
12087 				     struct kvm_async_pf *work)
12088 {
12089 	struct x86_exception fault;
12090 
12091 	trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
12092 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
12093 
12094 	if (kvm_can_deliver_async_pf(vcpu) &&
12095 	    !apf_put_user_notpresent(vcpu)) {
12096 		fault.vector = PF_VECTOR;
12097 		fault.error_code_valid = true;
12098 		fault.error_code = 0;
12099 		fault.nested_page_fault = false;
12100 		fault.address = work->arch.token;
12101 		fault.async_page_fault = true;
12102 		kvm_inject_page_fault(vcpu, &fault);
12103 		return true;
12104 	} else {
12105 		/*
12106 		 * It is not possible to deliver a paravirtualized asynchronous
12107 		 * page fault, but putting the guest in an artificial halt state
12108 		 * can be beneficial nevertheless: if an interrupt arrives, we
12109 		 * can deliver it timely and perhaps the guest will schedule
12110 		 * another process.  When the instruction that triggered a page
12111 		 * fault is retried, hopefully the page will be ready in the host.
12112 		 */
12113 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
12114 		return false;
12115 	}
12116 }
12117 
12118 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
12119 				 struct kvm_async_pf *work)
12120 {
12121 	struct kvm_lapic_irq irq = {
12122 		.delivery_mode = APIC_DM_FIXED,
12123 		.vector = vcpu->arch.apf.vec
12124 	};
12125 
12126 	if (work->wakeup_all)
12127 		work->arch.token = ~0; /* broadcast wakeup */
12128 	else
12129 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
12130 	trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
12131 
12132 	if ((work->wakeup_all || work->notpresent_injected) &&
12133 	    kvm_pv_async_pf_enabled(vcpu) &&
12134 	    !apf_put_user_ready(vcpu, work->arch.token)) {
12135 		vcpu->arch.apf.pageready_pending = true;
12136 		kvm_apic_set_irq(vcpu, &irq, NULL);
12137 	}
12138 
12139 	vcpu->arch.apf.halted = false;
12140 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
12141 }
12142 
12143 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
12144 {
12145 	kvm_make_request(KVM_REQ_APF_READY, vcpu);
12146 	if (!vcpu->arch.apf.pageready_pending)
12147 		kvm_vcpu_kick(vcpu);
12148 }
12149 
12150 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
12151 {
12152 	if (!kvm_pv_async_pf_enabled(vcpu))
12153 		return true;
12154 	else
12155 		return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
12156 }
12157 
12158 void kvm_arch_start_assignment(struct kvm *kvm)
12159 {
12160 	if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
12161 		static_call_cond(kvm_x86_start_assignment)(kvm);
12162 }
12163 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
12164 
12165 void kvm_arch_end_assignment(struct kvm *kvm)
12166 {
12167 	atomic_dec(&kvm->arch.assigned_device_count);
12168 }
12169 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
12170 
12171 bool kvm_arch_has_assigned_device(struct kvm *kvm)
12172 {
12173 	return atomic_read(&kvm->arch.assigned_device_count);
12174 }
12175 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
12176 
12177 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
12178 {
12179 	atomic_inc(&kvm->arch.noncoherent_dma_count);
12180 }
12181 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
12182 
12183 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
12184 {
12185 	atomic_dec(&kvm->arch.noncoherent_dma_count);
12186 }
12187 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
12188 
12189 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
12190 {
12191 	return atomic_read(&kvm->arch.noncoherent_dma_count);
12192 }
12193 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
12194 
12195 bool kvm_arch_has_irq_bypass(void)
12196 {
12197 	return true;
12198 }
12199 
12200 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
12201 				      struct irq_bypass_producer *prod)
12202 {
12203 	struct kvm_kernel_irqfd *irqfd =
12204 		container_of(cons, struct kvm_kernel_irqfd, consumer);
12205 	int ret;
12206 
12207 	irqfd->producer = prod;
12208 	kvm_arch_start_assignment(irqfd->kvm);
12209 	ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
12210 					 prod->irq, irqfd->gsi, 1);
12211 
12212 	if (ret)
12213 		kvm_arch_end_assignment(irqfd->kvm);
12214 
12215 	return ret;
12216 }
12217 
12218 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
12219 				      struct irq_bypass_producer *prod)
12220 {
12221 	int ret;
12222 	struct kvm_kernel_irqfd *irqfd =
12223 		container_of(cons, struct kvm_kernel_irqfd, consumer);
12224 
12225 	WARN_ON(irqfd->producer != prod);
12226 	irqfd->producer = NULL;
12227 
12228 	/*
12229 	 * When producer of consumer is unregistered, we change back to
12230 	 * remapped mode, so we can re-use the current implementation
12231 	 * when the irq is masked/disabled or the consumer side (KVM
12232 	 * int this case doesn't want to receive the interrupts.
12233 	*/
12234 	ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
12235 	if (ret)
12236 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
12237 		       " fails: %d\n", irqfd->consumer.token, ret);
12238 
12239 	kvm_arch_end_assignment(irqfd->kvm);
12240 }
12241 
12242 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
12243 				   uint32_t guest_irq, bool set)
12244 {
12245 	return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
12246 }
12247 
12248 bool kvm_arch_irqfd_route_changed(struct kvm_kernel_irq_routing_entry *old,
12249 				  struct kvm_kernel_irq_routing_entry *new)
12250 {
12251 	if (new->type != KVM_IRQ_ROUTING_MSI)
12252 		return true;
12253 
12254 	return !!memcmp(&old->msi, &new->msi, sizeof(new->msi));
12255 }
12256 
12257 bool kvm_vector_hashing_enabled(void)
12258 {
12259 	return vector_hashing;
12260 }
12261 
12262 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
12263 {
12264 	return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
12265 }
12266 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
12267 
12268 
12269 int kvm_spec_ctrl_test_value(u64 value)
12270 {
12271 	/*
12272 	 * test that setting IA32_SPEC_CTRL to given value
12273 	 * is allowed by the host processor
12274 	 */
12275 
12276 	u64 saved_value;
12277 	unsigned long flags;
12278 	int ret = 0;
12279 
12280 	local_irq_save(flags);
12281 
12282 	if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
12283 		ret = 1;
12284 	else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
12285 		ret = 1;
12286 	else
12287 		wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
12288 
12289 	local_irq_restore(flags);
12290 
12291 	return ret;
12292 }
12293 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
12294 
12295 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
12296 {
12297 	struct x86_exception fault;
12298 	u32 access = error_code &
12299 		(PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
12300 
12301 	if (!(error_code & PFERR_PRESENT_MASK) ||
12302 	    vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
12303 		/*
12304 		 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
12305 		 * tables probably do not match the TLB.  Just proceed
12306 		 * with the error code that the processor gave.
12307 		 */
12308 		fault.vector = PF_VECTOR;
12309 		fault.error_code_valid = true;
12310 		fault.error_code = error_code;
12311 		fault.nested_page_fault = false;
12312 		fault.address = gva;
12313 	}
12314 	vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
12315 }
12316 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
12317 
12318 /*
12319  * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
12320  * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
12321  * indicates whether exit to userspace is needed.
12322  */
12323 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
12324 			      struct x86_exception *e)
12325 {
12326 	if (r == X86EMUL_PROPAGATE_FAULT) {
12327 		kvm_inject_emulated_page_fault(vcpu, e);
12328 		return 1;
12329 	}
12330 
12331 	/*
12332 	 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
12333 	 * while handling a VMX instruction KVM could've handled the request
12334 	 * correctly by exiting to userspace and performing I/O but there
12335 	 * doesn't seem to be a real use-case behind such requests, just return
12336 	 * KVM_EXIT_INTERNAL_ERROR for now.
12337 	 */
12338 	kvm_prepare_emulation_failure_exit(vcpu);
12339 
12340 	return 0;
12341 }
12342 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
12343 
12344 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
12345 {
12346 	bool pcid_enabled;
12347 	struct x86_exception e;
12348 	struct {
12349 		u64 pcid;
12350 		u64 gla;
12351 	} operand;
12352 	int r;
12353 
12354 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
12355 	if (r != X86EMUL_CONTINUE)
12356 		return kvm_handle_memory_failure(vcpu, r, &e);
12357 
12358 	if (operand.pcid >> 12 != 0) {
12359 		kvm_inject_gp(vcpu, 0);
12360 		return 1;
12361 	}
12362 
12363 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
12364 
12365 	switch (type) {
12366 	case INVPCID_TYPE_INDIV_ADDR:
12367 		if ((!pcid_enabled && (operand.pcid != 0)) ||
12368 		    is_noncanonical_address(operand.gla, vcpu)) {
12369 			kvm_inject_gp(vcpu, 0);
12370 			return 1;
12371 		}
12372 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
12373 		return kvm_skip_emulated_instruction(vcpu);
12374 
12375 	case INVPCID_TYPE_SINGLE_CTXT:
12376 		if (!pcid_enabled && (operand.pcid != 0)) {
12377 			kvm_inject_gp(vcpu, 0);
12378 			return 1;
12379 		}
12380 
12381 		kvm_invalidate_pcid(vcpu, operand.pcid);
12382 		return kvm_skip_emulated_instruction(vcpu);
12383 
12384 	case INVPCID_TYPE_ALL_NON_GLOBAL:
12385 		/*
12386 		 * Currently, KVM doesn't mark global entries in the shadow
12387 		 * page tables, so a non-global flush just degenerates to a
12388 		 * global flush. If needed, we could optimize this later by
12389 		 * keeping track of global entries in shadow page tables.
12390 		 */
12391 
12392 		fallthrough;
12393 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
12394 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
12395 		return kvm_skip_emulated_instruction(vcpu);
12396 
12397 	default:
12398 		kvm_inject_gp(vcpu, 0);
12399 		return 1;
12400 	}
12401 }
12402 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
12403 
12404 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
12405 {
12406 	struct kvm_run *run = vcpu->run;
12407 	struct kvm_mmio_fragment *frag;
12408 	unsigned int len;
12409 
12410 	BUG_ON(!vcpu->mmio_needed);
12411 
12412 	/* Complete previous fragment */
12413 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
12414 	len = min(8u, frag->len);
12415 	if (!vcpu->mmio_is_write)
12416 		memcpy(frag->data, run->mmio.data, len);
12417 
12418 	if (frag->len <= 8) {
12419 		/* Switch to the next fragment. */
12420 		frag++;
12421 		vcpu->mmio_cur_fragment++;
12422 	} else {
12423 		/* Go forward to the next mmio piece. */
12424 		frag->data += len;
12425 		frag->gpa += len;
12426 		frag->len -= len;
12427 	}
12428 
12429 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
12430 		vcpu->mmio_needed = 0;
12431 
12432 		// VMG change, at this point, we're always done
12433 		// RIP has already been advanced
12434 		return 1;
12435 	}
12436 
12437 	// More MMIO is needed
12438 	run->mmio.phys_addr = frag->gpa;
12439 	run->mmio.len = min(8u, frag->len);
12440 	run->mmio.is_write = vcpu->mmio_is_write;
12441 	if (run->mmio.is_write)
12442 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
12443 	run->exit_reason = KVM_EXIT_MMIO;
12444 
12445 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12446 
12447 	return 0;
12448 }
12449 
12450 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12451 			  void *data)
12452 {
12453 	int handled;
12454 	struct kvm_mmio_fragment *frag;
12455 
12456 	if (!data)
12457 		return -EINVAL;
12458 
12459 	handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12460 	if (handled == bytes)
12461 		return 1;
12462 
12463 	bytes -= handled;
12464 	gpa += handled;
12465 	data += handled;
12466 
12467 	/*TODO: Check if need to increment number of frags */
12468 	frag = vcpu->mmio_fragments;
12469 	vcpu->mmio_nr_fragments = 1;
12470 	frag->len = bytes;
12471 	frag->gpa = gpa;
12472 	frag->data = data;
12473 
12474 	vcpu->mmio_needed = 1;
12475 	vcpu->mmio_cur_fragment = 0;
12476 
12477 	vcpu->run->mmio.phys_addr = gpa;
12478 	vcpu->run->mmio.len = min(8u, frag->len);
12479 	vcpu->run->mmio.is_write = 1;
12480 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
12481 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
12482 
12483 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12484 
12485 	return 0;
12486 }
12487 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
12488 
12489 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12490 			 void *data)
12491 {
12492 	int handled;
12493 	struct kvm_mmio_fragment *frag;
12494 
12495 	if (!data)
12496 		return -EINVAL;
12497 
12498 	handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12499 	if (handled == bytes)
12500 		return 1;
12501 
12502 	bytes -= handled;
12503 	gpa += handled;
12504 	data += handled;
12505 
12506 	/*TODO: Check if need to increment number of frags */
12507 	frag = vcpu->mmio_fragments;
12508 	vcpu->mmio_nr_fragments = 1;
12509 	frag->len = bytes;
12510 	frag->gpa = gpa;
12511 	frag->data = data;
12512 
12513 	vcpu->mmio_needed = 1;
12514 	vcpu->mmio_cur_fragment = 0;
12515 
12516 	vcpu->run->mmio.phys_addr = gpa;
12517 	vcpu->run->mmio.len = min(8u, frag->len);
12518 	vcpu->run->mmio.is_write = 0;
12519 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
12520 
12521 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12522 
12523 	return 0;
12524 }
12525 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
12526 
12527 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12528 			   unsigned int port);
12529 
12530 static int complete_sev_es_emulated_outs(struct kvm_vcpu *vcpu)
12531 {
12532 	int size = vcpu->arch.pio.size;
12533 	int port = vcpu->arch.pio.port;
12534 
12535 	vcpu->arch.pio.count = 0;
12536 	if (vcpu->arch.sev_pio_count)
12537 		return kvm_sev_es_outs(vcpu, size, port);
12538 	return 1;
12539 }
12540 
12541 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12542 			   unsigned int port)
12543 {
12544 	for (;;) {
12545 		unsigned int count =
12546 			min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
12547 		int ret = emulator_pio_out(vcpu, size, port, vcpu->arch.sev_pio_data, count);
12548 
12549 		/* memcpy done already by emulator_pio_out.  */
12550 		vcpu->arch.sev_pio_count -= count;
12551 		vcpu->arch.sev_pio_data += count * vcpu->arch.pio.size;
12552 		if (!ret)
12553 			break;
12554 
12555 		/* Emulation done by the kernel.  */
12556 		if (!vcpu->arch.sev_pio_count)
12557 			return 1;
12558 	}
12559 
12560 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_outs;
12561 	return 0;
12562 }
12563 
12564 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12565 			  unsigned int port);
12566 
12567 static void advance_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12568 {
12569 	unsigned count = vcpu->arch.pio.count;
12570 	complete_emulator_pio_in(vcpu, vcpu->arch.sev_pio_data);
12571 	vcpu->arch.sev_pio_count -= count;
12572 	vcpu->arch.sev_pio_data += count * vcpu->arch.pio.size;
12573 }
12574 
12575 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12576 {
12577 	int size = vcpu->arch.pio.size;
12578 	int port = vcpu->arch.pio.port;
12579 
12580 	advance_sev_es_emulated_ins(vcpu);
12581 	if (vcpu->arch.sev_pio_count)
12582 		return kvm_sev_es_ins(vcpu, size, port);
12583 	return 1;
12584 }
12585 
12586 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12587 			  unsigned int port)
12588 {
12589 	for (;;) {
12590 		unsigned int count =
12591 			min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
12592 		if (!__emulator_pio_in(vcpu, size, port, count))
12593 			break;
12594 
12595 		/* Emulation done by the kernel.  */
12596 		advance_sev_es_emulated_ins(vcpu);
12597 		if (!vcpu->arch.sev_pio_count)
12598 			return 1;
12599 	}
12600 
12601 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
12602 	return 0;
12603 }
12604 
12605 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
12606 			 unsigned int port, void *data,  unsigned int count,
12607 			 int in)
12608 {
12609 	vcpu->arch.sev_pio_data = data;
12610 	vcpu->arch.sev_pio_count = count;
12611 	return in ? kvm_sev_es_ins(vcpu, size, port)
12612 		  : kvm_sev_es_outs(vcpu, size, port);
12613 }
12614 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
12615 
12616 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
12617 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
12618 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
12619 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
12620 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
12621 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
12622 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
12623 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
12624 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
12625 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
12626 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
12627 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
12628 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
12629 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
12630 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
12631 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
12632 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
12633 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
12634 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
12635 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
12636 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
12637 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
12638 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
12639 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
12640 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
12641 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
12642 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);
12643