xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 35267cea)
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/internal.h> /* Ugh! */
72 #include <asm/pvclock.h>
73 #include <asm/div64.h>
74 #include <asm/irq_remapping.h>
75 #include <asm/mshyperv.h>
76 #include <asm/hypervisor.h>
77 #include <asm/tlbflush.h>
78 #include <asm/intel_pt.h>
79 #include <asm/emulate_prefix.h>
80 #include <asm/sgx.h>
81 #include <clocksource/hyperv_timer.h>
82 
83 #define CREATE_TRACE_POINTS
84 #include "trace.h"
85 
86 #define MAX_IO_MSRS 256
87 #define KVM_MAX_MCE_BANKS 32
88 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
89 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
90 
91 #define emul_to_vcpu(ctxt) \
92 	((struct kvm_vcpu *)(ctxt)->vcpu)
93 
94 /* EFER defaults:
95  * - enable syscall per default because its emulated by KVM
96  * - enable LME and LMA per default on 64 bit KVM
97  */
98 #ifdef CONFIG_X86_64
99 static
100 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
101 #else
102 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
103 #endif
104 
105 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
106 
107 #define KVM_EXIT_HYPERCALL_VALID_MASK (1 << KVM_HC_MAP_GPA_RANGE)
108 
109 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
110                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
111 
112 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
113 static void process_nmi(struct kvm_vcpu *vcpu);
114 static void process_smi(struct kvm_vcpu *vcpu);
115 static void enter_smm(struct kvm_vcpu *vcpu);
116 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
117 static void store_regs(struct kvm_vcpu *vcpu);
118 static int sync_regs(struct kvm_vcpu *vcpu);
119 
120 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
121 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
122 
123 struct kvm_x86_ops kvm_x86_ops __read_mostly;
124 EXPORT_SYMBOL_GPL(kvm_x86_ops);
125 
126 #define KVM_X86_OP(func)					     \
127 	DEFINE_STATIC_CALL_NULL(kvm_x86_##func,			     \
128 				*(((struct kvm_x86_ops *)0)->func));
129 #define KVM_X86_OP_NULL KVM_X86_OP
130 #include <asm/kvm-x86-ops.h>
131 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
132 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
133 EXPORT_STATIC_CALL_GPL(kvm_x86_tlb_flush_current);
134 
135 static bool __read_mostly ignore_msrs = 0;
136 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
137 
138 bool __read_mostly report_ignored_msrs = true;
139 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
140 EXPORT_SYMBOL_GPL(report_ignored_msrs);
141 
142 unsigned int min_timer_period_us = 200;
143 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
144 
145 static bool __read_mostly kvmclock_periodic_sync = true;
146 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
147 
148 bool __read_mostly kvm_has_tsc_control;
149 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
150 u32  __read_mostly kvm_max_guest_tsc_khz;
151 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
152 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
153 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
154 u64  __read_mostly kvm_max_tsc_scaling_ratio;
155 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
156 u64 __read_mostly kvm_default_tsc_scaling_ratio;
157 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
158 bool __read_mostly kvm_has_bus_lock_exit;
159 EXPORT_SYMBOL_GPL(kvm_has_bus_lock_exit);
160 
161 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
162 static u32 __read_mostly tsc_tolerance_ppm = 250;
163 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
164 
165 /*
166  * lapic timer advance (tscdeadline mode only) in nanoseconds.  '-1' enables
167  * adaptive tuning starting from default advancement of 1000ns.  '0' disables
168  * advancement entirely.  Any other value is used as-is and disables adaptive
169  * tuning, i.e. allows privileged userspace to set an exact advancement time.
170  */
171 static int __read_mostly lapic_timer_advance_ns = -1;
172 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
173 
174 static bool __read_mostly vector_hashing = true;
175 module_param(vector_hashing, bool, S_IRUGO);
176 
177 bool __read_mostly enable_vmware_backdoor = false;
178 module_param(enable_vmware_backdoor, bool, S_IRUGO);
179 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
180 
181 static bool __read_mostly force_emulation_prefix = false;
182 module_param(force_emulation_prefix, bool, S_IRUGO);
183 
184 int __read_mostly pi_inject_timer = -1;
185 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
186 
187 /*
188  * Restoring the host value for MSRs that are only consumed when running in
189  * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
190  * returns to userspace, i.e. the kernel can run with the guest's value.
191  */
192 #define KVM_MAX_NR_USER_RETURN_MSRS 16
193 
194 struct kvm_user_return_msrs {
195 	struct user_return_notifier urn;
196 	bool registered;
197 	struct kvm_user_return_msr_values {
198 		u64 host;
199 		u64 curr;
200 	} values[KVM_MAX_NR_USER_RETURN_MSRS];
201 };
202 
203 u32 __read_mostly kvm_nr_uret_msrs;
204 EXPORT_SYMBOL_GPL(kvm_nr_uret_msrs);
205 static u32 __read_mostly kvm_uret_msrs_list[KVM_MAX_NR_USER_RETURN_MSRS];
206 static struct kvm_user_return_msrs __percpu *user_return_msrs;
207 
208 #define KVM_SUPPORTED_XCR0     (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
209 				| XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
210 				| XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
211 				| XFEATURE_MASK_PKRU)
212 
213 u64 __read_mostly host_efer;
214 EXPORT_SYMBOL_GPL(host_efer);
215 
216 bool __read_mostly allow_smaller_maxphyaddr = 0;
217 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
218 
219 bool __read_mostly enable_apicv = true;
220 EXPORT_SYMBOL_GPL(enable_apicv);
221 
222 u64 __read_mostly host_xss;
223 EXPORT_SYMBOL_GPL(host_xss);
224 u64 __read_mostly supported_xss;
225 EXPORT_SYMBOL_GPL(supported_xss);
226 
227 const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
228 	KVM_GENERIC_VM_STATS(),
229 	STATS_DESC_COUNTER(VM, mmu_shadow_zapped),
230 	STATS_DESC_COUNTER(VM, mmu_pte_write),
231 	STATS_DESC_COUNTER(VM, mmu_pde_zapped),
232 	STATS_DESC_COUNTER(VM, mmu_flooded),
233 	STATS_DESC_COUNTER(VM, mmu_recycled),
234 	STATS_DESC_COUNTER(VM, mmu_cache_miss),
235 	STATS_DESC_ICOUNTER(VM, mmu_unsync),
236 	STATS_DESC_ICOUNTER(VM, lpages),
237 	STATS_DESC_ICOUNTER(VM, nx_lpage_splits),
238 	STATS_DESC_PCOUNTER(VM, max_mmu_page_hash_collisions)
239 };
240 static_assert(ARRAY_SIZE(kvm_vm_stats_desc) ==
241 		sizeof(struct kvm_vm_stat) / sizeof(u64));
242 
243 const struct kvm_stats_header kvm_vm_stats_header = {
244 	.name_size = KVM_STATS_NAME_SIZE,
245 	.num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
246 	.id_offset = sizeof(struct kvm_stats_header),
247 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
248 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
249 		       sizeof(kvm_vm_stats_desc),
250 };
251 
252 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
253 	KVM_GENERIC_VCPU_STATS(),
254 	STATS_DESC_COUNTER(VCPU, pf_fixed),
255 	STATS_DESC_COUNTER(VCPU, pf_guest),
256 	STATS_DESC_COUNTER(VCPU, tlb_flush),
257 	STATS_DESC_COUNTER(VCPU, invlpg),
258 	STATS_DESC_COUNTER(VCPU, exits),
259 	STATS_DESC_COUNTER(VCPU, io_exits),
260 	STATS_DESC_COUNTER(VCPU, mmio_exits),
261 	STATS_DESC_COUNTER(VCPU, signal_exits),
262 	STATS_DESC_COUNTER(VCPU, irq_window_exits),
263 	STATS_DESC_COUNTER(VCPU, nmi_window_exits),
264 	STATS_DESC_COUNTER(VCPU, l1d_flush),
265 	STATS_DESC_COUNTER(VCPU, halt_exits),
266 	STATS_DESC_COUNTER(VCPU, request_irq_exits),
267 	STATS_DESC_COUNTER(VCPU, irq_exits),
268 	STATS_DESC_COUNTER(VCPU, host_state_reload),
269 	STATS_DESC_COUNTER(VCPU, fpu_reload),
270 	STATS_DESC_COUNTER(VCPU, insn_emulation),
271 	STATS_DESC_COUNTER(VCPU, insn_emulation_fail),
272 	STATS_DESC_COUNTER(VCPU, hypercalls),
273 	STATS_DESC_COUNTER(VCPU, irq_injections),
274 	STATS_DESC_COUNTER(VCPU, nmi_injections),
275 	STATS_DESC_COUNTER(VCPU, req_event),
276 	STATS_DESC_COUNTER(VCPU, nested_run),
277 	STATS_DESC_COUNTER(VCPU, directed_yield_attempted),
278 	STATS_DESC_COUNTER(VCPU, directed_yield_successful),
279 	STATS_DESC_ICOUNTER(VCPU, guest_mode)
280 };
281 static_assert(ARRAY_SIZE(kvm_vcpu_stats_desc) ==
282 		sizeof(struct kvm_vcpu_stat) / sizeof(u64));
283 
284 const struct kvm_stats_header kvm_vcpu_stats_header = {
285 	.name_size = KVM_STATS_NAME_SIZE,
286 	.num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
287 	.id_offset = sizeof(struct kvm_stats_header),
288 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
289 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
290 		       sizeof(kvm_vcpu_stats_desc),
291 };
292 
293 u64 __read_mostly host_xcr0;
294 u64 __read_mostly supported_xcr0;
295 EXPORT_SYMBOL_GPL(supported_xcr0);
296 
297 static struct kmem_cache *x86_fpu_cache;
298 
299 static struct kmem_cache *x86_emulator_cache;
300 
301 /*
302  * When called, it means the previous get/set msr reached an invalid msr.
303  * Return true if we want to ignore/silent this failed msr access.
304  */
305 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
306 {
307 	const char *op = write ? "wrmsr" : "rdmsr";
308 
309 	if (ignore_msrs) {
310 		if (report_ignored_msrs)
311 			kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
312 				      op, msr, data);
313 		/* Mask the error */
314 		return true;
315 	} else {
316 		kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
317 				      op, msr, data);
318 		return false;
319 	}
320 }
321 
322 static struct kmem_cache *kvm_alloc_emulator_cache(void)
323 {
324 	unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
325 	unsigned int size = sizeof(struct x86_emulate_ctxt);
326 
327 	return kmem_cache_create_usercopy("x86_emulator", size,
328 					  __alignof__(struct x86_emulate_ctxt),
329 					  SLAB_ACCOUNT, useroffset,
330 					  size - useroffset, NULL);
331 }
332 
333 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
334 
335 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
336 {
337 	int i;
338 	for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
339 		vcpu->arch.apf.gfns[i] = ~0;
340 }
341 
342 static void kvm_on_user_return(struct user_return_notifier *urn)
343 {
344 	unsigned slot;
345 	struct kvm_user_return_msrs *msrs
346 		= container_of(urn, struct kvm_user_return_msrs, urn);
347 	struct kvm_user_return_msr_values *values;
348 	unsigned long flags;
349 
350 	/*
351 	 * Disabling irqs at this point since the following code could be
352 	 * interrupted and executed through kvm_arch_hardware_disable()
353 	 */
354 	local_irq_save(flags);
355 	if (msrs->registered) {
356 		msrs->registered = false;
357 		user_return_notifier_unregister(urn);
358 	}
359 	local_irq_restore(flags);
360 	for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
361 		values = &msrs->values[slot];
362 		if (values->host != values->curr) {
363 			wrmsrl(kvm_uret_msrs_list[slot], values->host);
364 			values->curr = values->host;
365 		}
366 	}
367 }
368 
369 static int kvm_probe_user_return_msr(u32 msr)
370 {
371 	u64 val;
372 	int ret;
373 
374 	preempt_disable();
375 	ret = rdmsrl_safe(msr, &val);
376 	if (ret)
377 		goto out;
378 	ret = wrmsrl_safe(msr, val);
379 out:
380 	preempt_enable();
381 	return ret;
382 }
383 
384 int kvm_add_user_return_msr(u32 msr)
385 {
386 	BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
387 
388 	if (kvm_probe_user_return_msr(msr))
389 		return -1;
390 
391 	kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
392 	return kvm_nr_uret_msrs++;
393 }
394 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
395 
396 int kvm_find_user_return_msr(u32 msr)
397 {
398 	int i;
399 
400 	for (i = 0; i < kvm_nr_uret_msrs; ++i) {
401 		if (kvm_uret_msrs_list[i] == msr)
402 			return i;
403 	}
404 	return -1;
405 }
406 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
407 
408 static void kvm_user_return_msr_cpu_online(void)
409 {
410 	unsigned int cpu = smp_processor_id();
411 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
412 	u64 value;
413 	int i;
414 
415 	for (i = 0; i < kvm_nr_uret_msrs; ++i) {
416 		rdmsrl_safe(kvm_uret_msrs_list[i], &value);
417 		msrs->values[i].host = value;
418 		msrs->values[i].curr = value;
419 	}
420 }
421 
422 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
423 {
424 	unsigned int cpu = smp_processor_id();
425 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
426 	int err;
427 
428 	value = (value & mask) | (msrs->values[slot].host & ~mask);
429 	if (value == msrs->values[slot].curr)
430 		return 0;
431 	err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
432 	if (err)
433 		return 1;
434 
435 	msrs->values[slot].curr = value;
436 	if (!msrs->registered) {
437 		msrs->urn.on_user_return = kvm_on_user_return;
438 		user_return_notifier_register(&msrs->urn);
439 		msrs->registered = true;
440 	}
441 	return 0;
442 }
443 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
444 
445 static void drop_user_return_notifiers(void)
446 {
447 	unsigned int cpu = smp_processor_id();
448 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
449 
450 	if (msrs->registered)
451 		kvm_on_user_return(&msrs->urn);
452 }
453 
454 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
455 {
456 	return vcpu->arch.apic_base;
457 }
458 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
459 
460 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
461 {
462 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
463 }
464 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
465 
466 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
467 {
468 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
469 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
470 	u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
471 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
472 
473 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
474 		return 1;
475 	if (!msr_info->host_initiated) {
476 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
477 			return 1;
478 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
479 			return 1;
480 	}
481 
482 	kvm_lapic_set_base(vcpu, msr_info->data);
483 	kvm_recalculate_apic_map(vcpu->kvm);
484 	return 0;
485 }
486 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
487 
488 asmlinkage __visible noinstr void kvm_spurious_fault(void)
489 {
490 	/* Fault while not rebooting.  We want the trace. */
491 	BUG_ON(!kvm_rebooting);
492 }
493 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
494 
495 #define EXCPT_BENIGN		0
496 #define EXCPT_CONTRIBUTORY	1
497 #define EXCPT_PF		2
498 
499 static int exception_class(int vector)
500 {
501 	switch (vector) {
502 	case PF_VECTOR:
503 		return EXCPT_PF;
504 	case DE_VECTOR:
505 	case TS_VECTOR:
506 	case NP_VECTOR:
507 	case SS_VECTOR:
508 	case GP_VECTOR:
509 		return EXCPT_CONTRIBUTORY;
510 	default:
511 		break;
512 	}
513 	return EXCPT_BENIGN;
514 }
515 
516 #define EXCPT_FAULT		0
517 #define EXCPT_TRAP		1
518 #define EXCPT_ABORT		2
519 #define EXCPT_INTERRUPT		3
520 
521 static int exception_type(int vector)
522 {
523 	unsigned int mask;
524 
525 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
526 		return EXCPT_INTERRUPT;
527 
528 	mask = 1 << vector;
529 
530 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
531 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
532 		return EXCPT_TRAP;
533 
534 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
535 		return EXCPT_ABORT;
536 
537 	/* Reserved exceptions will result in fault */
538 	return EXCPT_FAULT;
539 }
540 
541 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
542 {
543 	unsigned nr = vcpu->arch.exception.nr;
544 	bool has_payload = vcpu->arch.exception.has_payload;
545 	unsigned long payload = vcpu->arch.exception.payload;
546 
547 	if (!has_payload)
548 		return;
549 
550 	switch (nr) {
551 	case DB_VECTOR:
552 		/*
553 		 * "Certain debug exceptions may clear bit 0-3.  The
554 		 * remaining contents of the DR6 register are never
555 		 * cleared by the processor".
556 		 */
557 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
558 		/*
559 		 * In order to reflect the #DB exception payload in guest
560 		 * dr6, three components need to be considered: active low
561 		 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
562 		 * DR6_BS and DR6_BT)
563 		 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
564 		 * In the target guest dr6:
565 		 * FIXED_1 bits should always be set.
566 		 * Active low bits should be cleared if 1-setting in payload.
567 		 * Active high bits should be set if 1-setting in payload.
568 		 *
569 		 * Note, the payload is compatible with the pending debug
570 		 * exceptions/exit qualification under VMX, that active_low bits
571 		 * are active high in payload.
572 		 * So they need to be flipped for DR6.
573 		 */
574 		vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
575 		vcpu->arch.dr6 |= payload;
576 		vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
577 
578 		/*
579 		 * The #DB payload is defined as compatible with the 'pending
580 		 * debug exceptions' field under VMX, not DR6. While bit 12 is
581 		 * defined in the 'pending debug exceptions' field (enabled
582 		 * breakpoint), it is reserved and must be zero in DR6.
583 		 */
584 		vcpu->arch.dr6 &= ~BIT(12);
585 		break;
586 	case PF_VECTOR:
587 		vcpu->arch.cr2 = payload;
588 		break;
589 	}
590 
591 	vcpu->arch.exception.has_payload = false;
592 	vcpu->arch.exception.payload = 0;
593 }
594 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
595 
596 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
597 		unsigned nr, bool has_error, u32 error_code,
598 	        bool has_payload, unsigned long payload, bool reinject)
599 {
600 	u32 prev_nr;
601 	int class1, class2;
602 
603 	kvm_make_request(KVM_REQ_EVENT, vcpu);
604 
605 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
606 	queue:
607 		if (reinject) {
608 			/*
609 			 * On vmentry, vcpu->arch.exception.pending is only
610 			 * true if an event injection was blocked by
611 			 * nested_run_pending.  In that case, however,
612 			 * vcpu_enter_guest requests an immediate exit,
613 			 * and the guest shouldn't proceed far enough to
614 			 * need reinjection.
615 			 */
616 			WARN_ON_ONCE(vcpu->arch.exception.pending);
617 			vcpu->arch.exception.injected = true;
618 			if (WARN_ON_ONCE(has_payload)) {
619 				/*
620 				 * A reinjected event has already
621 				 * delivered its payload.
622 				 */
623 				has_payload = false;
624 				payload = 0;
625 			}
626 		} else {
627 			vcpu->arch.exception.pending = true;
628 			vcpu->arch.exception.injected = false;
629 		}
630 		vcpu->arch.exception.has_error_code = has_error;
631 		vcpu->arch.exception.nr = nr;
632 		vcpu->arch.exception.error_code = error_code;
633 		vcpu->arch.exception.has_payload = has_payload;
634 		vcpu->arch.exception.payload = payload;
635 		if (!is_guest_mode(vcpu))
636 			kvm_deliver_exception_payload(vcpu);
637 		return;
638 	}
639 
640 	/* to check exception */
641 	prev_nr = vcpu->arch.exception.nr;
642 	if (prev_nr == DF_VECTOR) {
643 		/* triple fault -> shutdown */
644 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
645 		return;
646 	}
647 	class1 = exception_class(prev_nr);
648 	class2 = exception_class(nr);
649 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
650 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
651 		/*
652 		 * Generate double fault per SDM Table 5-5.  Set
653 		 * exception.pending = true so that the double fault
654 		 * can trigger a nested vmexit.
655 		 */
656 		vcpu->arch.exception.pending = true;
657 		vcpu->arch.exception.injected = false;
658 		vcpu->arch.exception.has_error_code = true;
659 		vcpu->arch.exception.nr = DF_VECTOR;
660 		vcpu->arch.exception.error_code = 0;
661 		vcpu->arch.exception.has_payload = false;
662 		vcpu->arch.exception.payload = 0;
663 	} else
664 		/* replace previous exception with a new one in a hope
665 		   that instruction re-execution will regenerate lost
666 		   exception */
667 		goto queue;
668 }
669 
670 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
671 {
672 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
673 }
674 EXPORT_SYMBOL_GPL(kvm_queue_exception);
675 
676 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
677 {
678 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
679 }
680 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
681 
682 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
683 			   unsigned long payload)
684 {
685 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
686 }
687 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
688 
689 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
690 				    u32 error_code, unsigned long payload)
691 {
692 	kvm_multiple_exception(vcpu, nr, true, error_code,
693 			       true, payload, false);
694 }
695 
696 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
697 {
698 	if (err)
699 		kvm_inject_gp(vcpu, 0);
700 	else
701 		return kvm_skip_emulated_instruction(vcpu);
702 
703 	return 1;
704 }
705 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
706 
707 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
708 {
709 	++vcpu->stat.pf_guest;
710 	vcpu->arch.exception.nested_apf =
711 		is_guest_mode(vcpu) && fault->async_page_fault;
712 	if (vcpu->arch.exception.nested_apf) {
713 		vcpu->arch.apf.nested_apf_token = fault->address;
714 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
715 	} else {
716 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
717 					fault->address);
718 	}
719 }
720 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
721 
722 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
723 				    struct x86_exception *fault)
724 {
725 	struct kvm_mmu *fault_mmu;
726 	WARN_ON_ONCE(fault->vector != PF_VECTOR);
727 
728 	fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
729 					       vcpu->arch.walk_mmu;
730 
731 	/*
732 	 * Invalidate the TLB entry for the faulting address, if it exists,
733 	 * else the access will fault indefinitely (and to emulate hardware).
734 	 */
735 	if ((fault->error_code & PFERR_PRESENT_MASK) &&
736 	    !(fault->error_code & PFERR_RSVD_MASK))
737 		kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
738 				       fault_mmu->root_hpa);
739 
740 	fault_mmu->inject_page_fault(vcpu, fault);
741 	return fault->nested_page_fault;
742 }
743 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
744 
745 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
746 {
747 	atomic_inc(&vcpu->arch.nmi_queued);
748 	kvm_make_request(KVM_REQ_NMI, vcpu);
749 }
750 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
751 
752 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
753 {
754 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
755 }
756 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
757 
758 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
759 {
760 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
761 }
762 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
763 
764 /*
765  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
766  * a #GP and return false.
767  */
768 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
769 {
770 	if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
771 		return true;
772 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
773 	return false;
774 }
775 EXPORT_SYMBOL_GPL(kvm_require_cpl);
776 
777 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
778 {
779 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
780 		return true;
781 
782 	kvm_queue_exception(vcpu, UD_VECTOR);
783 	return false;
784 }
785 EXPORT_SYMBOL_GPL(kvm_require_dr);
786 
787 /*
788  * This function will be used to read from the physical memory of the currently
789  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
790  * can read from guest physical or from the guest's guest physical memory.
791  */
792 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
793 			    gfn_t ngfn, void *data, int offset, int len,
794 			    u32 access)
795 {
796 	struct x86_exception exception;
797 	gfn_t real_gfn;
798 	gpa_t ngpa;
799 
800 	ngpa     = gfn_to_gpa(ngfn);
801 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
802 	if (real_gfn == UNMAPPED_GVA)
803 		return -EFAULT;
804 
805 	real_gfn = gpa_to_gfn(real_gfn);
806 
807 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
808 }
809 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
810 
811 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
812 {
813 	return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
814 }
815 
816 /*
817  * Load the pae pdptrs.  Return 1 if they are all valid, 0 otherwise.
818  */
819 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
820 {
821 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
822 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
823 	int i;
824 	int ret;
825 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
826 
827 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
828 				      offset * sizeof(u64), sizeof(pdpte),
829 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
830 	if (ret < 0) {
831 		ret = 0;
832 		goto out;
833 	}
834 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
835 		if ((pdpte[i] & PT_PRESENT_MASK) &&
836 		    (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
837 			ret = 0;
838 			goto out;
839 		}
840 	}
841 	ret = 1;
842 
843 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
844 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
845 	vcpu->arch.pdptrs_from_userspace = false;
846 
847 out:
848 
849 	return ret;
850 }
851 EXPORT_SYMBOL_GPL(load_pdptrs);
852 
853 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
854 {
855 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
856 		kvm_clear_async_pf_completion_queue(vcpu);
857 		kvm_async_pf_hash_reset(vcpu);
858 	}
859 
860 	if ((cr0 ^ old_cr0) & KVM_MMU_CR0_ROLE_BITS)
861 		kvm_mmu_reset_context(vcpu);
862 
863 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
864 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
865 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
866 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
867 }
868 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
869 
870 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
871 {
872 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
873 	unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
874 
875 	cr0 |= X86_CR0_ET;
876 
877 #ifdef CONFIG_X86_64
878 	if (cr0 & 0xffffffff00000000UL)
879 		return 1;
880 #endif
881 
882 	cr0 &= ~CR0_RESERVED_BITS;
883 
884 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
885 		return 1;
886 
887 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
888 		return 1;
889 
890 #ifdef CONFIG_X86_64
891 	if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
892 	    (cr0 & X86_CR0_PG)) {
893 		int cs_db, cs_l;
894 
895 		if (!is_pae(vcpu))
896 			return 1;
897 		static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
898 		if (cs_l)
899 			return 1;
900 	}
901 #endif
902 	if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
903 	    is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
904 	    !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
905 		return 1;
906 
907 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
908 		return 1;
909 
910 	static_call(kvm_x86_set_cr0)(vcpu, cr0);
911 
912 	kvm_post_set_cr0(vcpu, old_cr0, cr0);
913 
914 	return 0;
915 }
916 EXPORT_SYMBOL_GPL(kvm_set_cr0);
917 
918 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
919 {
920 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
921 }
922 EXPORT_SYMBOL_GPL(kvm_lmsw);
923 
924 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
925 {
926 	if (vcpu->arch.guest_state_protected)
927 		return;
928 
929 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
930 
931 		if (vcpu->arch.xcr0 != host_xcr0)
932 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
933 
934 		if (vcpu->arch.xsaves_enabled &&
935 		    vcpu->arch.ia32_xss != host_xss)
936 			wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
937 	}
938 
939 	if (static_cpu_has(X86_FEATURE_PKU) &&
940 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
941 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
942 	    vcpu->arch.pkru != vcpu->arch.host_pkru)
943 		write_pkru(vcpu->arch.pkru);
944 }
945 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
946 
947 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
948 {
949 	if (vcpu->arch.guest_state_protected)
950 		return;
951 
952 	if (static_cpu_has(X86_FEATURE_PKU) &&
953 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
954 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
955 		vcpu->arch.pkru = rdpkru();
956 		if (vcpu->arch.pkru != vcpu->arch.host_pkru)
957 			write_pkru(vcpu->arch.host_pkru);
958 	}
959 
960 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
961 
962 		if (vcpu->arch.xcr0 != host_xcr0)
963 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
964 
965 		if (vcpu->arch.xsaves_enabled &&
966 		    vcpu->arch.ia32_xss != host_xss)
967 			wrmsrl(MSR_IA32_XSS, host_xss);
968 	}
969 
970 }
971 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
972 
973 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
974 {
975 	u64 xcr0 = xcr;
976 	u64 old_xcr0 = vcpu->arch.xcr0;
977 	u64 valid_bits;
978 
979 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
980 	if (index != XCR_XFEATURE_ENABLED_MASK)
981 		return 1;
982 	if (!(xcr0 & XFEATURE_MASK_FP))
983 		return 1;
984 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
985 		return 1;
986 
987 	/*
988 	 * Do not allow the guest to set bits that we do not support
989 	 * saving.  However, xcr0 bit 0 is always set, even if the
990 	 * emulated CPU does not support XSAVE (see fx_init).
991 	 */
992 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
993 	if (xcr0 & ~valid_bits)
994 		return 1;
995 
996 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
997 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
998 		return 1;
999 
1000 	if (xcr0 & XFEATURE_MASK_AVX512) {
1001 		if (!(xcr0 & XFEATURE_MASK_YMM))
1002 			return 1;
1003 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
1004 			return 1;
1005 	}
1006 	vcpu->arch.xcr0 = xcr0;
1007 
1008 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
1009 		kvm_update_cpuid_runtime(vcpu);
1010 	return 0;
1011 }
1012 
1013 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
1014 {
1015 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
1016 	    __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
1017 		kvm_inject_gp(vcpu, 0);
1018 		return 1;
1019 	}
1020 
1021 	return kvm_skip_emulated_instruction(vcpu);
1022 }
1023 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1024 
1025 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1026 {
1027 	if (cr4 & cr4_reserved_bits)
1028 		return false;
1029 
1030 	if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1031 		return false;
1032 
1033 	return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1034 }
1035 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
1036 
1037 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1038 {
1039 	if (((cr4 ^ old_cr4) & KVM_MMU_CR4_ROLE_BITS) ||
1040 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1041 		kvm_mmu_reset_context(vcpu);
1042 }
1043 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1044 
1045 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1046 {
1047 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
1048 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1049 				   X86_CR4_SMEP;
1050 
1051 	if (!kvm_is_valid_cr4(vcpu, cr4))
1052 		return 1;
1053 
1054 	if (is_long_mode(vcpu)) {
1055 		if (!(cr4 & X86_CR4_PAE))
1056 			return 1;
1057 		if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1058 			return 1;
1059 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1060 		   && ((cr4 ^ old_cr4) & pdptr_bits)
1061 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1062 				   kvm_read_cr3(vcpu)))
1063 		return 1;
1064 
1065 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1066 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1067 			return 1;
1068 
1069 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1070 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1071 			return 1;
1072 	}
1073 
1074 	static_call(kvm_x86_set_cr4)(vcpu, cr4);
1075 
1076 	kvm_post_set_cr4(vcpu, old_cr4, cr4);
1077 
1078 	return 0;
1079 }
1080 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1081 
1082 static void kvm_invalidate_pcid(struct kvm_vcpu *vcpu, unsigned long pcid)
1083 {
1084 	struct kvm_mmu *mmu = vcpu->arch.mmu;
1085 	unsigned long roots_to_free = 0;
1086 	int i;
1087 
1088 	/*
1089 	 * If neither the current CR3 nor any of the prev_roots use the given
1090 	 * PCID, then nothing needs to be done here because a resync will
1091 	 * happen anyway before switching to any other CR3.
1092 	 */
1093 	if (kvm_get_active_pcid(vcpu) == pcid) {
1094 		kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
1095 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1096 	}
1097 
1098 	for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
1099 		if (kvm_get_pcid(vcpu, mmu->prev_roots[i].pgd) == pcid)
1100 			roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
1101 
1102 	kvm_mmu_free_roots(vcpu, mmu, roots_to_free);
1103 }
1104 
1105 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1106 {
1107 	bool skip_tlb_flush = false;
1108 	unsigned long pcid = 0;
1109 #ifdef CONFIG_X86_64
1110 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1111 
1112 	if (pcid_enabled) {
1113 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1114 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
1115 		pcid = cr3 & X86_CR3_PCID_MASK;
1116 	}
1117 #endif
1118 
1119 	/* PDPTRs are always reloaded for PAE paging. */
1120 	if (cr3 == kvm_read_cr3(vcpu) && !is_pae_paging(vcpu))
1121 		goto handle_tlb_flush;
1122 
1123 	/*
1124 	 * Do not condition the GPA check on long mode, this helper is used to
1125 	 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1126 	 * the current vCPU mode is accurate.
1127 	 */
1128 	if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1129 		return 1;
1130 
1131 	if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1132 		return 1;
1133 
1134 	if (cr3 != kvm_read_cr3(vcpu))
1135 		kvm_mmu_new_pgd(vcpu, cr3);
1136 
1137 	vcpu->arch.cr3 = cr3;
1138 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1139 
1140 handle_tlb_flush:
1141 	/*
1142 	 * A load of CR3 that flushes the TLB flushes only the current PCID,
1143 	 * even if PCID is disabled, in which case PCID=0 is flushed.  It's a
1144 	 * moot point in the end because _disabling_ PCID will flush all PCIDs,
1145 	 * and it's impossible to use a non-zero PCID when PCID is disabled,
1146 	 * i.e. only PCID=0 can be relevant.
1147 	 */
1148 	if (!skip_tlb_flush)
1149 		kvm_invalidate_pcid(vcpu, pcid);
1150 
1151 	return 0;
1152 }
1153 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1154 
1155 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1156 {
1157 	if (cr8 & CR8_RESERVED_BITS)
1158 		return 1;
1159 	if (lapic_in_kernel(vcpu))
1160 		kvm_lapic_set_tpr(vcpu, cr8);
1161 	else
1162 		vcpu->arch.cr8 = cr8;
1163 	return 0;
1164 }
1165 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1166 
1167 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1168 {
1169 	if (lapic_in_kernel(vcpu))
1170 		return kvm_lapic_get_cr8(vcpu);
1171 	else
1172 		return vcpu->arch.cr8;
1173 }
1174 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1175 
1176 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1177 {
1178 	int i;
1179 
1180 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1181 		for (i = 0; i < KVM_NR_DB_REGS; i++)
1182 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1183 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1184 	}
1185 }
1186 
1187 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1188 {
1189 	unsigned long dr7;
1190 
1191 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1192 		dr7 = vcpu->arch.guest_debug_dr7;
1193 	else
1194 		dr7 = vcpu->arch.dr7;
1195 	static_call(kvm_x86_set_dr7)(vcpu, dr7);
1196 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1197 	if (dr7 & DR7_BP_EN_MASK)
1198 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1199 }
1200 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1201 
1202 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1203 {
1204 	u64 fixed = DR6_FIXED_1;
1205 
1206 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1207 		fixed |= DR6_RTM;
1208 
1209 	if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1210 		fixed |= DR6_BUS_LOCK;
1211 	return fixed;
1212 }
1213 
1214 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1215 {
1216 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1217 
1218 	switch (dr) {
1219 	case 0 ... 3:
1220 		vcpu->arch.db[array_index_nospec(dr, size)] = val;
1221 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1222 			vcpu->arch.eff_db[dr] = val;
1223 		break;
1224 	case 4:
1225 	case 6:
1226 		if (!kvm_dr6_valid(val))
1227 			return 1; /* #GP */
1228 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1229 		break;
1230 	case 5:
1231 	default: /* 7 */
1232 		if (!kvm_dr7_valid(val))
1233 			return 1; /* #GP */
1234 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1235 		kvm_update_dr7(vcpu);
1236 		break;
1237 	}
1238 
1239 	return 0;
1240 }
1241 EXPORT_SYMBOL_GPL(kvm_set_dr);
1242 
1243 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1244 {
1245 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1246 
1247 	switch (dr) {
1248 	case 0 ... 3:
1249 		*val = vcpu->arch.db[array_index_nospec(dr, size)];
1250 		break;
1251 	case 4:
1252 	case 6:
1253 		*val = vcpu->arch.dr6;
1254 		break;
1255 	case 5:
1256 	default: /* 7 */
1257 		*val = vcpu->arch.dr7;
1258 		break;
1259 	}
1260 }
1261 EXPORT_SYMBOL_GPL(kvm_get_dr);
1262 
1263 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1264 {
1265 	u32 ecx = kvm_rcx_read(vcpu);
1266 	u64 data;
1267 
1268 	if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1269 		kvm_inject_gp(vcpu, 0);
1270 		return 1;
1271 	}
1272 
1273 	kvm_rax_write(vcpu, (u32)data);
1274 	kvm_rdx_write(vcpu, data >> 32);
1275 	return kvm_skip_emulated_instruction(vcpu);
1276 }
1277 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1278 
1279 /*
1280  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1281  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1282  *
1283  * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1284  * extract the supported MSRs from the related const lists.
1285  * msrs_to_save is selected from the msrs_to_save_all to reflect the
1286  * capabilities of the host cpu. This capabilities test skips MSRs that are
1287  * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1288  * may depend on host virtualization features rather than host cpu features.
1289  */
1290 
1291 static const u32 msrs_to_save_all[] = {
1292 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1293 	MSR_STAR,
1294 #ifdef CONFIG_X86_64
1295 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1296 #endif
1297 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1298 	MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1299 	MSR_IA32_SPEC_CTRL,
1300 	MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1301 	MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1302 	MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1303 	MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1304 	MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1305 	MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1306 	MSR_IA32_UMWAIT_CONTROL,
1307 
1308 	MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1309 	MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1310 	MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1311 	MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1312 	MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1313 	MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1314 	MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1315 	MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1316 	MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1317 	MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1318 	MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1319 	MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1320 	MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1321 	MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1322 	MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1323 	MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1324 	MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1325 	MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1326 	MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1327 	MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1328 	MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1329 	MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1330 };
1331 
1332 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1333 static unsigned num_msrs_to_save;
1334 
1335 static const u32 emulated_msrs_all[] = {
1336 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1337 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1338 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1339 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1340 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1341 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1342 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1343 	HV_X64_MSR_RESET,
1344 	HV_X64_MSR_VP_INDEX,
1345 	HV_X64_MSR_VP_RUNTIME,
1346 	HV_X64_MSR_SCONTROL,
1347 	HV_X64_MSR_STIMER0_CONFIG,
1348 	HV_X64_MSR_VP_ASSIST_PAGE,
1349 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1350 	HV_X64_MSR_TSC_EMULATION_STATUS,
1351 	HV_X64_MSR_SYNDBG_OPTIONS,
1352 	HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1353 	HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1354 	HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1355 
1356 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1357 	MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1358 
1359 	MSR_IA32_TSC_ADJUST,
1360 	MSR_IA32_TSC_DEADLINE,
1361 	MSR_IA32_ARCH_CAPABILITIES,
1362 	MSR_IA32_PERF_CAPABILITIES,
1363 	MSR_IA32_MISC_ENABLE,
1364 	MSR_IA32_MCG_STATUS,
1365 	MSR_IA32_MCG_CTL,
1366 	MSR_IA32_MCG_EXT_CTL,
1367 	MSR_IA32_SMBASE,
1368 	MSR_SMI_COUNT,
1369 	MSR_PLATFORM_INFO,
1370 	MSR_MISC_FEATURES_ENABLES,
1371 	MSR_AMD64_VIRT_SPEC_CTRL,
1372 	MSR_IA32_POWER_CTL,
1373 	MSR_IA32_UCODE_REV,
1374 
1375 	/*
1376 	 * The following list leaves out MSRs whose values are determined
1377 	 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1378 	 * We always support the "true" VMX control MSRs, even if the host
1379 	 * processor does not, so I am putting these registers here rather
1380 	 * than in msrs_to_save_all.
1381 	 */
1382 	MSR_IA32_VMX_BASIC,
1383 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1384 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1385 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1386 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1387 	MSR_IA32_VMX_MISC,
1388 	MSR_IA32_VMX_CR0_FIXED0,
1389 	MSR_IA32_VMX_CR4_FIXED0,
1390 	MSR_IA32_VMX_VMCS_ENUM,
1391 	MSR_IA32_VMX_PROCBASED_CTLS2,
1392 	MSR_IA32_VMX_EPT_VPID_CAP,
1393 	MSR_IA32_VMX_VMFUNC,
1394 
1395 	MSR_K7_HWCR,
1396 	MSR_KVM_POLL_CONTROL,
1397 };
1398 
1399 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1400 static unsigned num_emulated_msrs;
1401 
1402 /*
1403  * List of msr numbers which are used to expose MSR-based features that
1404  * can be used by a hypervisor to validate requested CPU features.
1405  */
1406 static const u32 msr_based_features_all[] = {
1407 	MSR_IA32_VMX_BASIC,
1408 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1409 	MSR_IA32_VMX_PINBASED_CTLS,
1410 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1411 	MSR_IA32_VMX_PROCBASED_CTLS,
1412 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1413 	MSR_IA32_VMX_EXIT_CTLS,
1414 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1415 	MSR_IA32_VMX_ENTRY_CTLS,
1416 	MSR_IA32_VMX_MISC,
1417 	MSR_IA32_VMX_CR0_FIXED0,
1418 	MSR_IA32_VMX_CR0_FIXED1,
1419 	MSR_IA32_VMX_CR4_FIXED0,
1420 	MSR_IA32_VMX_CR4_FIXED1,
1421 	MSR_IA32_VMX_VMCS_ENUM,
1422 	MSR_IA32_VMX_PROCBASED_CTLS2,
1423 	MSR_IA32_VMX_EPT_VPID_CAP,
1424 	MSR_IA32_VMX_VMFUNC,
1425 
1426 	MSR_F10H_DECFG,
1427 	MSR_IA32_UCODE_REV,
1428 	MSR_IA32_ARCH_CAPABILITIES,
1429 	MSR_IA32_PERF_CAPABILITIES,
1430 };
1431 
1432 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1433 static unsigned int num_msr_based_features;
1434 
1435 static u64 kvm_get_arch_capabilities(void)
1436 {
1437 	u64 data = 0;
1438 
1439 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1440 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1441 
1442 	/*
1443 	 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1444 	 * the nested hypervisor runs with NX huge pages.  If it is not,
1445 	 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1446 	 * L1 guests, so it need not worry about its own (L2) guests.
1447 	 */
1448 	data |= ARCH_CAP_PSCHANGE_MC_NO;
1449 
1450 	/*
1451 	 * If we're doing cache flushes (either "always" or "cond")
1452 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1453 	 * If an outer hypervisor is doing the cache flush for us
1454 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1455 	 * capability to the guest too, and if EPT is disabled we're not
1456 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1457 	 * require a nested hypervisor to do a flush of its own.
1458 	 */
1459 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1460 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1461 
1462 	if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1463 		data |= ARCH_CAP_RDCL_NO;
1464 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1465 		data |= ARCH_CAP_SSB_NO;
1466 	if (!boot_cpu_has_bug(X86_BUG_MDS))
1467 		data |= ARCH_CAP_MDS_NO;
1468 
1469 	if (!boot_cpu_has(X86_FEATURE_RTM)) {
1470 		/*
1471 		 * If RTM=0 because the kernel has disabled TSX, the host might
1472 		 * have TAA_NO or TSX_CTRL.  Clear TAA_NO (the guest sees RTM=0
1473 		 * and therefore knows that there cannot be TAA) but keep
1474 		 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1475 		 * and we want to allow migrating those guests to tsx=off hosts.
1476 		 */
1477 		data &= ~ARCH_CAP_TAA_NO;
1478 	} else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1479 		data |= ARCH_CAP_TAA_NO;
1480 	} else {
1481 		/*
1482 		 * Nothing to do here; we emulate TSX_CTRL if present on the
1483 		 * host so the guest can choose between disabling TSX or
1484 		 * using VERW to clear CPU buffers.
1485 		 */
1486 	}
1487 
1488 	return data;
1489 }
1490 
1491 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1492 {
1493 	switch (msr->index) {
1494 	case MSR_IA32_ARCH_CAPABILITIES:
1495 		msr->data = kvm_get_arch_capabilities();
1496 		break;
1497 	case MSR_IA32_UCODE_REV:
1498 		rdmsrl_safe(msr->index, &msr->data);
1499 		break;
1500 	default:
1501 		return static_call(kvm_x86_get_msr_feature)(msr);
1502 	}
1503 	return 0;
1504 }
1505 
1506 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1507 {
1508 	struct kvm_msr_entry msr;
1509 	int r;
1510 
1511 	msr.index = index;
1512 	r = kvm_get_msr_feature(&msr);
1513 
1514 	if (r == KVM_MSR_RET_INVALID) {
1515 		/* Unconditionally clear the output for simplicity */
1516 		*data = 0;
1517 		if (kvm_msr_ignored_check(index, 0, false))
1518 			r = 0;
1519 	}
1520 
1521 	if (r)
1522 		return r;
1523 
1524 	*data = msr.data;
1525 
1526 	return 0;
1527 }
1528 
1529 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1530 {
1531 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1532 		return false;
1533 
1534 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1535 		return false;
1536 
1537 	if (efer & (EFER_LME | EFER_LMA) &&
1538 	    !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1539 		return false;
1540 
1541 	if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1542 		return false;
1543 
1544 	return true;
1545 
1546 }
1547 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1548 {
1549 	if (efer & efer_reserved_bits)
1550 		return false;
1551 
1552 	return __kvm_valid_efer(vcpu, efer);
1553 }
1554 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1555 
1556 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1557 {
1558 	u64 old_efer = vcpu->arch.efer;
1559 	u64 efer = msr_info->data;
1560 	int r;
1561 
1562 	if (efer & efer_reserved_bits)
1563 		return 1;
1564 
1565 	if (!msr_info->host_initiated) {
1566 		if (!__kvm_valid_efer(vcpu, efer))
1567 			return 1;
1568 
1569 		if (is_paging(vcpu) &&
1570 		    (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1571 			return 1;
1572 	}
1573 
1574 	efer &= ~EFER_LMA;
1575 	efer |= vcpu->arch.efer & EFER_LMA;
1576 
1577 	r = static_call(kvm_x86_set_efer)(vcpu, efer);
1578 	if (r) {
1579 		WARN_ON(r > 0);
1580 		return r;
1581 	}
1582 
1583 	/* Update reserved bits */
1584 	if ((efer ^ old_efer) & EFER_NX)
1585 		kvm_mmu_reset_context(vcpu);
1586 
1587 	return 0;
1588 }
1589 
1590 void kvm_enable_efer_bits(u64 mask)
1591 {
1592        efer_reserved_bits &= ~mask;
1593 }
1594 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1595 
1596 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1597 {
1598 	struct kvm_x86_msr_filter *msr_filter;
1599 	struct msr_bitmap_range *ranges;
1600 	struct kvm *kvm = vcpu->kvm;
1601 	bool allowed;
1602 	int idx;
1603 	u32 i;
1604 
1605 	/* x2APIC MSRs do not support filtering. */
1606 	if (index >= 0x800 && index <= 0x8ff)
1607 		return true;
1608 
1609 	idx = srcu_read_lock(&kvm->srcu);
1610 
1611 	msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1612 	if (!msr_filter) {
1613 		allowed = true;
1614 		goto out;
1615 	}
1616 
1617 	allowed = msr_filter->default_allow;
1618 	ranges = msr_filter->ranges;
1619 
1620 	for (i = 0; i < msr_filter->count; i++) {
1621 		u32 start = ranges[i].base;
1622 		u32 end = start + ranges[i].nmsrs;
1623 		u32 flags = ranges[i].flags;
1624 		unsigned long *bitmap = ranges[i].bitmap;
1625 
1626 		if ((index >= start) && (index < end) && (flags & type)) {
1627 			allowed = !!test_bit(index - start, bitmap);
1628 			break;
1629 		}
1630 	}
1631 
1632 out:
1633 	srcu_read_unlock(&kvm->srcu, idx);
1634 
1635 	return allowed;
1636 }
1637 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1638 
1639 /*
1640  * Write @data into the MSR specified by @index.  Select MSR specific fault
1641  * checks are bypassed if @host_initiated is %true.
1642  * Returns 0 on success, non-0 otherwise.
1643  * Assumes vcpu_load() was already called.
1644  */
1645 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1646 			 bool host_initiated)
1647 {
1648 	struct msr_data msr;
1649 
1650 	if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1651 		return KVM_MSR_RET_FILTERED;
1652 
1653 	switch (index) {
1654 	case MSR_FS_BASE:
1655 	case MSR_GS_BASE:
1656 	case MSR_KERNEL_GS_BASE:
1657 	case MSR_CSTAR:
1658 	case MSR_LSTAR:
1659 		if (is_noncanonical_address(data, vcpu))
1660 			return 1;
1661 		break;
1662 	case MSR_IA32_SYSENTER_EIP:
1663 	case MSR_IA32_SYSENTER_ESP:
1664 		/*
1665 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1666 		 * non-canonical address is written on Intel but not on
1667 		 * AMD (which ignores the top 32-bits, because it does
1668 		 * not implement 64-bit SYSENTER).
1669 		 *
1670 		 * 64-bit code should hence be able to write a non-canonical
1671 		 * value on AMD.  Making the address canonical ensures that
1672 		 * vmentry does not fail on Intel after writing a non-canonical
1673 		 * value, and that something deterministic happens if the guest
1674 		 * invokes 64-bit SYSENTER.
1675 		 */
1676 		data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1677 		break;
1678 	case MSR_TSC_AUX:
1679 		if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1680 			return 1;
1681 
1682 		if (!host_initiated &&
1683 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1684 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1685 			return 1;
1686 
1687 		/*
1688 		 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1689 		 * incomplete and conflicting architectural behavior.  Current
1690 		 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1691 		 * reserved and always read as zeros.  Enforce Intel's reserved
1692 		 * bits check if and only if the guest CPU is Intel, and clear
1693 		 * the bits in all other cases.  This ensures cross-vendor
1694 		 * migration will provide consistent behavior for the guest.
1695 		 */
1696 		if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1697 			return 1;
1698 
1699 		data = (u32)data;
1700 		break;
1701 	}
1702 
1703 	msr.data = data;
1704 	msr.index = index;
1705 	msr.host_initiated = host_initiated;
1706 
1707 	return static_call(kvm_x86_set_msr)(vcpu, &msr);
1708 }
1709 
1710 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1711 				     u32 index, u64 data, bool host_initiated)
1712 {
1713 	int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1714 
1715 	if (ret == KVM_MSR_RET_INVALID)
1716 		if (kvm_msr_ignored_check(index, data, true))
1717 			ret = 0;
1718 
1719 	return ret;
1720 }
1721 
1722 /*
1723  * Read the MSR specified by @index into @data.  Select MSR specific fault
1724  * checks are bypassed if @host_initiated is %true.
1725  * Returns 0 on success, non-0 otherwise.
1726  * Assumes vcpu_load() was already called.
1727  */
1728 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1729 		  bool host_initiated)
1730 {
1731 	struct msr_data msr;
1732 	int ret;
1733 
1734 	if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1735 		return KVM_MSR_RET_FILTERED;
1736 
1737 	switch (index) {
1738 	case MSR_TSC_AUX:
1739 		if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1740 			return 1;
1741 
1742 		if (!host_initiated &&
1743 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1744 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1745 			return 1;
1746 		break;
1747 	}
1748 
1749 	msr.index = index;
1750 	msr.host_initiated = host_initiated;
1751 
1752 	ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1753 	if (!ret)
1754 		*data = msr.data;
1755 	return ret;
1756 }
1757 
1758 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1759 				     u32 index, u64 *data, bool host_initiated)
1760 {
1761 	int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1762 
1763 	if (ret == KVM_MSR_RET_INVALID) {
1764 		/* Unconditionally clear *data for simplicity */
1765 		*data = 0;
1766 		if (kvm_msr_ignored_check(index, 0, false))
1767 			ret = 0;
1768 	}
1769 
1770 	return ret;
1771 }
1772 
1773 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1774 {
1775 	return kvm_get_msr_ignored_check(vcpu, index, data, false);
1776 }
1777 EXPORT_SYMBOL_GPL(kvm_get_msr);
1778 
1779 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1780 {
1781 	return kvm_set_msr_ignored_check(vcpu, index, data, false);
1782 }
1783 EXPORT_SYMBOL_GPL(kvm_set_msr);
1784 
1785 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1786 {
1787 	int err = vcpu->run->msr.error;
1788 	if (!err) {
1789 		kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1790 		kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1791 	}
1792 
1793 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, err);
1794 }
1795 
1796 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1797 {
1798 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1799 }
1800 
1801 static u64 kvm_msr_reason(int r)
1802 {
1803 	switch (r) {
1804 	case KVM_MSR_RET_INVALID:
1805 		return KVM_MSR_EXIT_REASON_UNKNOWN;
1806 	case KVM_MSR_RET_FILTERED:
1807 		return KVM_MSR_EXIT_REASON_FILTER;
1808 	default:
1809 		return KVM_MSR_EXIT_REASON_INVAL;
1810 	}
1811 }
1812 
1813 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1814 			      u32 exit_reason, u64 data,
1815 			      int (*completion)(struct kvm_vcpu *vcpu),
1816 			      int r)
1817 {
1818 	u64 msr_reason = kvm_msr_reason(r);
1819 
1820 	/* Check if the user wanted to know about this MSR fault */
1821 	if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1822 		return 0;
1823 
1824 	vcpu->run->exit_reason = exit_reason;
1825 	vcpu->run->msr.error = 0;
1826 	memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1827 	vcpu->run->msr.reason = msr_reason;
1828 	vcpu->run->msr.index = index;
1829 	vcpu->run->msr.data = data;
1830 	vcpu->arch.complete_userspace_io = completion;
1831 
1832 	return 1;
1833 }
1834 
1835 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1836 {
1837 	return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1838 				   complete_emulated_rdmsr, r);
1839 }
1840 
1841 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1842 {
1843 	return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1844 				   complete_emulated_wrmsr, r);
1845 }
1846 
1847 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1848 {
1849 	u32 ecx = kvm_rcx_read(vcpu);
1850 	u64 data;
1851 	int r;
1852 
1853 	r = kvm_get_msr(vcpu, ecx, &data);
1854 
1855 	/* MSR read failed? See if we should ask user space */
1856 	if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1857 		/* Bounce to user space */
1858 		return 0;
1859 	}
1860 
1861 	if (!r) {
1862 		trace_kvm_msr_read(ecx, data);
1863 
1864 		kvm_rax_write(vcpu, data & -1u);
1865 		kvm_rdx_write(vcpu, (data >> 32) & -1u);
1866 	} else {
1867 		trace_kvm_msr_read_ex(ecx);
1868 	}
1869 
1870 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1871 }
1872 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1873 
1874 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1875 {
1876 	u32 ecx = kvm_rcx_read(vcpu);
1877 	u64 data = kvm_read_edx_eax(vcpu);
1878 	int r;
1879 
1880 	r = kvm_set_msr(vcpu, ecx, data);
1881 
1882 	/* MSR write failed? See if we should ask user space */
1883 	if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1884 		/* Bounce to user space */
1885 		return 0;
1886 
1887 	/* Signal all other negative errors to userspace */
1888 	if (r < 0)
1889 		return r;
1890 
1891 	if (!r)
1892 		trace_kvm_msr_write(ecx, data);
1893 	else
1894 		trace_kvm_msr_write_ex(ecx, data);
1895 
1896 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1897 }
1898 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1899 
1900 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
1901 {
1902 	return kvm_skip_emulated_instruction(vcpu);
1903 }
1904 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
1905 
1906 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
1907 {
1908 	/* Treat an INVD instruction as a NOP and just skip it. */
1909 	return kvm_emulate_as_nop(vcpu);
1910 }
1911 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
1912 
1913 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
1914 {
1915 	pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
1916 	return kvm_emulate_as_nop(vcpu);
1917 }
1918 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
1919 
1920 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
1921 {
1922 	kvm_queue_exception(vcpu, UD_VECTOR);
1923 	return 1;
1924 }
1925 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
1926 
1927 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
1928 {
1929 	pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
1930 	return kvm_emulate_as_nop(vcpu);
1931 }
1932 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
1933 
1934 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1935 {
1936 	xfer_to_guest_mode_prepare();
1937 	return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1938 		xfer_to_guest_mode_work_pending();
1939 }
1940 
1941 /*
1942  * The fast path for frequent and performance sensitive wrmsr emulation,
1943  * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1944  * the latency of virtual IPI by avoiding the expensive bits of transitioning
1945  * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1946  * other cases which must be called after interrupts are enabled on the host.
1947  */
1948 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1949 {
1950 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1951 		return 1;
1952 
1953 	if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1954 		((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1955 		((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1956 		((u32)(data >> 32) != X2APIC_BROADCAST)) {
1957 
1958 		data &= ~(1 << 12);
1959 		kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1960 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1961 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1962 		trace_kvm_apic_write(APIC_ICR, (u32)data);
1963 		return 0;
1964 	}
1965 
1966 	return 1;
1967 }
1968 
1969 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1970 {
1971 	if (!kvm_can_use_hv_timer(vcpu))
1972 		return 1;
1973 
1974 	kvm_set_lapic_tscdeadline_msr(vcpu, data);
1975 	return 0;
1976 }
1977 
1978 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1979 {
1980 	u32 msr = kvm_rcx_read(vcpu);
1981 	u64 data;
1982 	fastpath_t ret = EXIT_FASTPATH_NONE;
1983 
1984 	switch (msr) {
1985 	case APIC_BASE_MSR + (APIC_ICR >> 4):
1986 		data = kvm_read_edx_eax(vcpu);
1987 		if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1988 			kvm_skip_emulated_instruction(vcpu);
1989 			ret = EXIT_FASTPATH_EXIT_HANDLED;
1990 		}
1991 		break;
1992 	case MSR_IA32_TSC_DEADLINE:
1993 		data = kvm_read_edx_eax(vcpu);
1994 		if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1995 			kvm_skip_emulated_instruction(vcpu);
1996 			ret = EXIT_FASTPATH_REENTER_GUEST;
1997 		}
1998 		break;
1999 	default:
2000 		break;
2001 	}
2002 
2003 	if (ret != EXIT_FASTPATH_NONE)
2004 		trace_kvm_msr_write(msr, data);
2005 
2006 	return ret;
2007 }
2008 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
2009 
2010 /*
2011  * Adapt set_msr() to msr_io()'s calling convention
2012  */
2013 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2014 {
2015 	return kvm_get_msr_ignored_check(vcpu, index, data, true);
2016 }
2017 
2018 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2019 {
2020 	return kvm_set_msr_ignored_check(vcpu, index, *data, true);
2021 }
2022 
2023 #ifdef CONFIG_X86_64
2024 struct pvclock_clock {
2025 	int vclock_mode;
2026 	u64 cycle_last;
2027 	u64 mask;
2028 	u32 mult;
2029 	u32 shift;
2030 	u64 base_cycles;
2031 	u64 offset;
2032 };
2033 
2034 struct pvclock_gtod_data {
2035 	seqcount_t	seq;
2036 
2037 	struct pvclock_clock clock; /* extract of a clocksource struct */
2038 	struct pvclock_clock raw_clock; /* extract of a clocksource struct */
2039 
2040 	ktime_t		offs_boot;
2041 	u64		wall_time_sec;
2042 };
2043 
2044 static struct pvclock_gtod_data pvclock_gtod_data;
2045 
2046 static void update_pvclock_gtod(struct timekeeper *tk)
2047 {
2048 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2049 
2050 	write_seqcount_begin(&vdata->seq);
2051 
2052 	/* copy pvclock gtod data */
2053 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->vdso_clock_mode;
2054 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
2055 	vdata->clock.mask		= tk->tkr_mono.mask;
2056 	vdata->clock.mult		= tk->tkr_mono.mult;
2057 	vdata->clock.shift		= tk->tkr_mono.shift;
2058 	vdata->clock.base_cycles	= tk->tkr_mono.xtime_nsec;
2059 	vdata->clock.offset		= tk->tkr_mono.base;
2060 
2061 	vdata->raw_clock.vclock_mode	= tk->tkr_raw.clock->vdso_clock_mode;
2062 	vdata->raw_clock.cycle_last	= tk->tkr_raw.cycle_last;
2063 	vdata->raw_clock.mask		= tk->tkr_raw.mask;
2064 	vdata->raw_clock.mult		= tk->tkr_raw.mult;
2065 	vdata->raw_clock.shift		= tk->tkr_raw.shift;
2066 	vdata->raw_clock.base_cycles	= tk->tkr_raw.xtime_nsec;
2067 	vdata->raw_clock.offset		= tk->tkr_raw.base;
2068 
2069 	vdata->wall_time_sec            = tk->xtime_sec;
2070 
2071 	vdata->offs_boot		= tk->offs_boot;
2072 
2073 	write_seqcount_end(&vdata->seq);
2074 }
2075 
2076 static s64 get_kvmclock_base_ns(void)
2077 {
2078 	/* Count up from boot time, but with the frequency of the raw clock.  */
2079 	return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2080 }
2081 #else
2082 static s64 get_kvmclock_base_ns(void)
2083 {
2084 	/* Master clock not used, so we can just use CLOCK_BOOTTIME.  */
2085 	return ktime_get_boottime_ns();
2086 }
2087 #endif
2088 
2089 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2090 {
2091 	int version;
2092 	int r;
2093 	struct pvclock_wall_clock wc;
2094 	u32 wc_sec_hi;
2095 	u64 wall_nsec;
2096 
2097 	if (!wall_clock)
2098 		return;
2099 
2100 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2101 	if (r)
2102 		return;
2103 
2104 	if (version & 1)
2105 		++version;  /* first time write, random junk */
2106 
2107 	++version;
2108 
2109 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2110 		return;
2111 
2112 	/*
2113 	 * The guest calculates current wall clock time by adding
2114 	 * system time (updated by kvm_guest_time_update below) to the
2115 	 * wall clock specified here.  We do the reverse here.
2116 	 */
2117 	wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2118 
2119 	wc.nsec = do_div(wall_nsec, 1000000000);
2120 	wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2121 	wc.version = version;
2122 
2123 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2124 
2125 	if (sec_hi_ofs) {
2126 		wc_sec_hi = wall_nsec >> 32;
2127 		kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2128 				&wc_sec_hi, sizeof(wc_sec_hi));
2129 	}
2130 
2131 	version++;
2132 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2133 }
2134 
2135 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2136 				  bool old_msr, bool host_initiated)
2137 {
2138 	struct kvm_arch *ka = &vcpu->kvm->arch;
2139 
2140 	if (vcpu->vcpu_id == 0 && !host_initiated) {
2141 		if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2142 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2143 
2144 		ka->boot_vcpu_runs_old_kvmclock = old_msr;
2145 	}
2146 
2147 	vcpu->arch.time = system_time;
2148 	kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2149 
2150 	/* we verify if the enable bit is set... */
2151 	vcpu->arch.pv_time_enabled = false;
2152 	if (!(system_time & 1))
2153 		return;
2154 
2155 	if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2156 				       &vcpu->arch.pv_time, system_time & ~1ULL,
2157 				       sizeof(struct pvclock_vcpu_time_info)))
2158 		vcpu->arch.pv_time_enabled = true;
2159 
2160 	return;
2161 }
2162 
2163 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2164 {
2165 	do_shl32_div32(dividend, divisor);
2166 	return dividend;
2167 }
2168 
2169 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2170 			       s8 *pshift, u32 *pmultiplier)
2171 {
2172 	uint64_t scaled64;
2173 	int32_t  shift = 0;
2174 	uint64_t tps64;
2175 	uint32_t tps32;
2176 
2177 	tps64 = base_hz;
2178 	scaled64 = scaled_hz;
2179 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2180 		tps64 >>= 1;
2181 		shift--;
2182 	}
2183 
2184 	tps32 = (uint32_t)tps64;
2185 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2186 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2187 			scaled64 >>= 1;
2188 		else
2189 			tps32 <<= 1;
2190 		shift++;
2191 	}
2192 
2193 	*pshift = shift;
2194 	*pmultiplier = div_frac(scaled64, tps32);
2195 }
2196 
2197 #ifdef CONFIG_X86_64
2198 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2199 #endif
2200 
2201 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2202 static unsigned long max_tsc_khz;
2203 
2204 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2205 {
2206 	u64 v = (u64)khz * (1000000 + ppm);
2207 	do_div(v, 1000000);
2208 	return v;
2209 }
2210 
2211 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier);
2212 
2213 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2214 {
2215 	u64 ratio;
2216 
2217 	/* Guest TSC same frequency as host TSC? */
2218 	if (!scale) {
2219 		kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2220 		return 0;
2221 	}
2222 
2223 	/* TSC scaling supported? */
2224 	if (!kvm_has_tsc_control) {
2225 		if (user_tsc_khz > tsc_khz) {
2226 			vcpu->arch.tsc_catchup = 1;
2227 			vcpu->arch.tsc_always_catchup = 1;
2228 			return 0;
2229 		} else {
2230 			pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2231 			return -1;
2232 		}
2233 	}
2234 
2235 	/* TSC scaling required  - calculate ratio */
2236 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2237 				user_tsc_khz, tsc_khz);
2238 
2239 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2240 		pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2241 			            user_tsc_khz);
2242 		return -1;
2243 	}
2244 
2245 	kvm_vcpu_write_tsc_multiplier(vcpu, ratio);
2246 	return 0;
2247 }
2248 
2249 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2250 {
2251 	u32 thresh_lo, thresh_hi;
2252 	int use_scaling = 0;
2253 
2254 	/* tsc_khz can be zero if TSC calibration fails */
2255 	if (user_tsc_khz == 0) {
2256 		/* set tsc_scaling_ratio to a safe value */
2257 		kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2258 		return -1;
2259 	}
2260 
2261 	/* Compute a scale to convert nanoseconds in TSC cycles */
2262 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2263 			   &vcpu->arch.virtual_tsc_shift,
2264 			   &vcpu->arch.virtual_tsc_mult);
2265 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2266 
2267 	/*
2268 	 * Compute the variation in TSC rate which is acceptable
2269 	 * within the range of tolerance and decide if the
2270 	 * rate being applied is within that bounds of the hardware
2271 	 * rate.  If so, no scaling or compensation need be done.
2272 	 */
2273 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2274 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2275 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2276 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2277 		use_scaling = 1;
2278 	}
2279 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2280 }
2281 
2282 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2283 {
2284 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2285 				      vcpu->arch.virtual_tsc_mult,
2286 				      vcpu->arch.virtual_tsc_shift);
2287 	tsc += vcpu->arch.this_tsc_write;
2288 	return tsc;
2289 }
2290 
2291 static inline int gtod_is_based_on_tsc(int mode)
2292 {
2293 	return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2294 }
2295 
2296 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2297 {
2298 #ifdef CONFIG_X86_64
2299 	bool vcpus_matched;
2300 	struct kvm_arch *ka = &vcpu->kvm->arch;
2301 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2302 
2303 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2304 			 atomic_read(&vcpu->kvm->online_vcpus));
2305 
2306 	/*
2307 	 * Once the masterclock is enabled, always perform request in
2308 	 * order to update it.
2309 	 *
2310 	 * In order to enable masterclock, the host clocksource must be TSC
2311 	 * and the vcpus need to have matched TSCs.  When that happens,
2312 	 * perform request to enable masterclock.
2313 	 */
2314 	if (ka->use_master_clock ||
2315 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2316 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2317 
2318 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2319 			    atomic_read(&vcpu->kvm->online_vcpus),
2320 		            ka->use_master_clock, gtod->clock.vclock_mode);
2321 #endif
2322 }
2323 
2324 /*
2325  * Multiply tsc by a fixed point number represented by ratio.
2326  *
2327  * The most significant 64-N bits (mult) of ratio represent the
2328  * integral part of the fixed point number; the remaining N bits
2329  * (frac) represent the fractional part, ie. ratio represents a fixed
2330  * point number (mult + frac * 2^(-N)).
2331  *
2332  * N equals to kvm_tsc_scaling_ratio_frac_bits.
2333  */
2334 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2335 {
2336 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2337 }
2338 
2339 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc, u64 ratio)
2340 {
2341 	u64 _tsc = tsc;
2342 
2343 	if (ratio != kvm_default_tsc_scaling_ratio)
2344 		_tsc = __scale_tsc(ratio, tsc);
2345 
2346 	return _tsc;
2347 }
2348 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2349 
2350 static u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2351 {
2352 	u64 tsc;
2353 
2354 	tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio);
2355 
2356 	return target_tsc - tsc;
2357 }
2358 
2359 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2360 {
2361 	return vcpu->arch.l1_tsc_offset +
2362 		kvm_scale_tsc(vcpu, host_tsc, vcpu->arch.l1_tsc_scaling_ratio);
2363 }
2364 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2365 
2366 u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier)
2367 {
2368 	u64 nested_offset;
2369 
2370 	if (l2_multiplier == kvm_default_tsc_scaling_ratio)
2371 		nested_offset = l1_offset;
2372 	else
2373 		nested_offset = mul_s64_u64_shr((s64) l1_offset, l2_multiplier,
2374 						kvm_tsc_scaling_ratio_frac_bits);
2375 
2376 	nested_offset += l2_offset;
2377 	return nested_offset;
2378 }
2379 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_offset);
2380 
2381 u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier)
2382 {
2383 	if (l2_multiplier != kvm_default_tsc_scaling_ratio)
2384 		return mul_u64_u64_shr(l1_multiplier, l2_multiplier,
2385 				       kvm_tsc_scaling_ratio_frac_bits);
2386 
2387 	return l1_multiplier;
2388 }
2389 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_multiplier);
2390 
2391 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset)
2392 {
2393 	trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2394 				   vcpu->arch.l1_tsc_offset,
2395 				   l1_offset);
2396 
2397 	vcpu->arch.l1_tsc_offset = l1_offset;
2398 
2399 	/*
2400 	 * If we are here because L1 chose not to trap WRMSR to TSC then
2401 	 * according to the spec this should set L1's TSC (as opposed to
2402 	 * setting L1's offset for L2).
2403 	 */
2404 	if (is_guest_mode(vcpu))
2405 		vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2406 			l1_offset,
2407 			static_call(kvm_x86_get_l2_tsc_offset)(vcpu),
2408 			static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2409 	else
2410 		vcpu->arch.tsc_offset = l1_offset;
2411 
2412 	static_call(kvm_x86_write_tsc_offset)(vcpu, vcpu->arch.tsc_offset);
2413 }
2414 
2415 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier)
2416 {
2417 	vcpu->arch.l1_tsc_scaling_ratio = l1_multiplier;
2418 
2419 	/* Userspace is changing the multiplier while L2 is active */
2420 	if (is_guest_mode(vcpu))
2421 		vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2422 			l1_multiplier,
2423 			static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2424 	else
2425 		vcpu->arch.tsc_scaling_ratio = l1_multiplier;
2426 
2427 	if (kvm_has_tsc_control)
2428 		static_call(kvm_x86_write_tsc_multiplier)(
2429 			vcpu, vcpu->arch.tsc_scaling_ratio);
2430 }
2431 
2432 static inline bool kvm_check_tsc_unstable(void)
2433 {
2434 #ifdef CONFIG_X86_64
2435 	/*
2436 	 * TSC is marked unstable when we're running on Hyper-V,
2437 	 * 'TSC page' clocksource is good.
2438 	 */
2439 	if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2440 		return false;
2441 #endif
2442 	return check_tsc_unstable();
2443 }
2444 
2445 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2446 {
2447 	struct kvm *kvm = vcpu->kvm;
2448 	u64 offset, ns, elapsed;
2449 	unsigned long flags;
2450 	bool matched;
2451 	bool already_matched;
2452 	bool synchronizing = false;
2453 
2454 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2455 	offset = kvm_compute_l1_tsc_offset(vcpu, data);
2456 	ns = get_kvmclock_base_ns();
2457 	elapsed = ns - kvm->arch.last_tsc_nsec;
2458 
2459 	if (vcpu->arch.virtual_tsc_khz) {
2460 		if (data == 0) {
2461 			/*
2462 			 * detection of vcpu initialization -- need to sync
2463 			 * with other vCPUs. This particularly helps to keep
2464 			 * kvm_clock stable after CPU hotplug
2465 			 */
2466 			synchronizing = true;
2467 		} else {
2468 			u64 tsc_exp = kvm->arch.last_tsc_write +
2469 						nsec_to_cycles(vcpu, elapsed);
2470 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2471 			/*
2472 			 * Special case: TSC write with a small delta (1 second)
2473 			 * of virtual cycle time against real time is
2474 			 * interpreted as an attempt to synchronize the CPU.
2475 			 */
2476 			synchronizing = data < tsc_exp + tsc_hz &&
2477 					data + tsc_hz > tsc_exp;
2478 		}
2479 	}
2480 
2481 	/*
2482 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
2483 	 * TSC, we add elapsed time in this computation.  We could let the
2484 	 * compensation code attempt to catch up if we fall behind, but
2485 	 * it's better to try to match offsets from the beginning.
2486          */
2487 	if (synchronizing &&
2488 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2489 		if (!kvm_check_tsc_unstable()) {
2490 			offset = kvm->arch.cur_tsc_offset;
2491 		} else {
2492 			u64 delta = nsec_to_cycles(vcpu, elapsed);
2493 			data += delta;
2494 			offset = kvm_compute_l1_tsc_offset(vcpu, data);
2495 		}
2496 		matched = true;
2497 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2498 	} else {
2499 		/*
2500 		 * We split periods of matched TSC writes into generations.
2501 		 * For each generation, we track the original measured
2502 		 * nanosecond time, offset, and write, so if TSCs are in
2503 		 * sync, we can match exact offset, and if not, we can match
2504 		 * exact software computation in compute_guest_tsc()
2505 		 *
2506 		 * These values are tracked in kvm->arch.cur_xxx variables.
2507 		 */
2508 		kvm->arch.cur_tsc_generation++;
2509 		kvm->arch.cur_tsc_nsec = ns;
2510 		kvm->arch.cur_tsc_write = data;
2511 		kvm->arch.cur_tsc_offset = offset;
2512 		matched = false;
2513 	}
2514 
2515 	/*
2516 	 * We also track th most recent recorded KHZ, write and time to
2517 	 * allow the matching interval to be extended at each write.
2518 	 */
2519 	kvm->arch.last_tsc_nsec = ns;
2520 	kvm->arch.last_tsc_write = data;
2521 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2522 
2523 	vcpu->arch.last_guest_tsc = data;
2524 
2525 	/* Keep track of which generation this VCPU has synchronized to */
2526 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2527 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2528 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2529 
2530 	kvm_vcpu_write_tsc_offset(vcpu, offset);
2531 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2532 
2533 	spin_lock_irqsave(&kvm->arch.pvclock_gtod_sync_lock, flags);
2534 	if (!matched) {
2535 		kvm->arch.nr_vcpus_matched_tsc = 0;
2536 	} else if (!already_matched) {
2537 		kvm->arch.nr_vcpus_matched_tsc++;
2538 	}
2539 
2540 	kvm_track_tsc_matching(vcpu);
2541 	spin_unlock_irqrestore(&kvm->arch.pvclock_gtod_sync_lock, flags);
2542 }
2543 
2544 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2545 					   s64 adjustment)
2546 {
2547 	u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2548 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2549 }
2550 
2551 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2552 {
2553 	if (vcpu->arch.l1_tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2554 		WARN_ON(adjustment < 0);
2555 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment,
2556 				   vcpu->arch.l1_tsc_scaling_ratio);
2557 	adjust_tsc_offset_guest(vcpu, adjustment);
2558 }
2559 
2560 #ifdef CONFIG_X86_64
2561 
2562 static u64 read_tsc(void)
2563 {
2564 	u64 ret = (u64)rdtsc_ordered();
2565 	u64 last = pvclock_gtod_data.clock.cycle_last;
2566 
2567 	if (likely(ret >= last))
2568 		return ret;
2569 
2570 	/*
2571 	 * GCC likes to generate cmov here, but this branch is extremely
2572 	 * predictable (it's just a function of time and the likely is
2573 	 * very likely) and there's a data dependence, so force GCC
2574 	 * to generate a branch instead.  I don't barrier() because
2575 	 * we don't actually need a barrier, and if this function
2576 	 * ever gets inlined it will generate worse code.
2577 	 */
2578 	asm volatile ("");
2579 	return last;
2580 }
2581 
2582 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2583 			  int *mode)
2584 {
2585 	long v;
2586 	u64 tsc_pg_val;
2587 
2588 	switch (clock->vclock_mode) {
2589 	case VDSO_CLOCKMODE_HVCLOCK:
2590 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2591 						  tsc_timestamp);
2592 		if (tsc_pg_val != U64_MAX) {
2593 			/* TSC page valid */
2594 			*mode = VDSO_CLOCKMODE_HVCLOCK;
2595 			v = (tsc_pg_val - clock->cycle_last) &
2596 				clock->mask;
2597 		} else {
2598 			/* TSC page invalid */
2599 			*mode = VDSO_CLOCKMODE_NONE;
2600 		}
2601 		break;
2602 	case VDSO_CLOCKMODE_TSC:
2603 		*mode = VDSO_CLOCKMODE_TSC;
2604 		*tsc_timestamp = read_tsc();
2605 		v = (*tsc_timestamp - clock->cycle_last) &
2606 			clock->mask;
2607 		break;
2608 	default:
2609 		*mode = VDSO_CLOCKMODE_NONE;
2610 	}
2611 
2612 	if (*mode == VDSO_CLOCKMODE_NONE)
2613 		*tsc_timestamp = v = 0;
2614 
2615 	return v * clock->mult;
2616 }
2617 
2618 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2619 {
2620 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2621 	unsigned long seq;
2622 	int mode;
2623 	u64 ns;
2624 
2625 	do {
2626 		seq = read_seqcount_begin(&gtod->seq);
2627 		ns = gtod->raw_clock.base_cycles;
2628 		ns += vgettsc(&gtod->raw_clock, tsc_timestamp, &mode);
2629 		ns >>= gtod->raw_clock.shift;
2630 		ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2631 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2632 	*t = ns;
2633 
2634 	return mode;
2635 }
2636 
2637 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2638 {
2639 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2640 	unsigned long seq;
2641 	int mode;
2642 	u64 ns;
2643 
2644 	do {
2645 		seq = read_seqcount_begin(&gtod->seq);
2646 		ts->tv_sec = gtod->wall_time_sec;
2647 		ns = gtod->clock.base_cycles;
2648 		ns += vgettsc(&gtod->clock, tsc_timestamp, &mode);
2649 		ns >>= gtod->clock.shift;
2650 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2651 
2652 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2653 	ts->tv_nsec = ns;
2654 
2655 	return mode;
2656 }
2657 
2658 /* returns true if host is using TSC based clocksource */
2659 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2660 {
2661 	/* checked again under seqlock below */
2662 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2663 		return false;
2664 
2665 	return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2666 						      tsc_timestamp));
2667 }
2668 
2669 /* returns true if host is using TSC based clocksource */
2670 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2671 					   u64 *tsc_timestamp)
2672 {
2673 	/* checked again under seqlock below */
2674 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2675 		return false;
2676 
2677 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2678 }
2679 #endif
2680 
2681 /*
2682  *
2683  * Assuming a stable TSC across physical CPUS, and a stable TSC
2684  * across virtual CPUs, the following condition is possible.
2685  * Each numbered line represents an event visible to both
2686  * CPUs at the next numbered event.
2687  *
2688  * "timespecX" represents host monotonic time. "tscX" represents
2689  * RDTSC value.
2690  *
2691  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
2692  *
2693  * 1.  read timespec0,tsc0
2694  * 2.					| timespec1 = timespec0 + N
2695  * 					| tsc1 = tsc0 + M
2696  * 3. transition to guest		| transition to guest
2697  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2698  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
2699  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2700  *
2701  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2702  *
2703  * 	- ret0 < ret1
2704  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2705  *		...
2706  *	- 0 < N - M => M < N
2707  *
2708  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2709  * always the case (the difference between two distinct xtime instances
2710  * might be smaller then the difference between corresponding TSC reads,
2711  * when updating guest vcpus pvclock areas).
2712  *
2713  * To avoid that problem, do not allow visibility of distinct
2714  * system_timestamp/tsc_timestamp values simultaneously: use a master
2715  * copy of host monotonic time values. Update that master copy
2716  * in lockstep.
2717  *
2718  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2719  *
2720  */
2721 
2722 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2723 {
2724 #ifdef CONFIG_X86_64
2725 	struct kvm_arch *ka = &kvm->arch;
2726 	int vclock_mode;
2727 	bool host_tsc_clocksource, vcpus_matched;
2728 
2729 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2730 			atomic_read(&kvm->online_vcpus));
2731 
2732 	/*
2733 	 * If the host uses TSC clock, then passthrough TSC as stable
2734 	 * to the guest.
2735 	 */
2736 	host_tsc_clocksource = kvm_get_time_and_clockread(
2737 					&ka->master_kernel_ns,
2738 					&ka->master_cycle_now);
2739 
2740 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2741 				&& !ka->backwards_tsc_observed
2742 				&& !ka->boot_vcpu_runs_old_kvmclock;
2743 
2744 	if (ka->use_master_clock)
2745 		atomic_set(&kvm_guest_has_master_clock, 1);
2746 
2747 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2748 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2749 					vcpus_matched);
2750 #endif
2751 }
2752 
2753 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2754 {
2755 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2756 }
2757 
2758 static void kvm_gen_update_masterclock(struct kvm *kvm)
2759 {
2760 #ifdef CONFIG_X86_64
2761 	int i;
2762 	struct kvm_vcpu *vcpu;
2763 	struct kvm_arch *ka = &kvm->arch;
2764 	unsigned long flags;
2765 
2766 	kvm_hv_invalidate_tsc_page(kvm);
2767 
2768 	kvm_make_mclock_inprogress_request(kvm);
2769 
2770 	/* no guest entries from this point */
2771 	spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2772 	pvclock_update_vm_gtod_copy(kvm);
2773 	spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2774 
2775 	kvm_for_each_vcpu(i, vcpu, kvm)
2776 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2777 
2778 	/* guest entries allowed */
2779 	kvm_for_each_vcpu(i, vcpu, kvm)
2780 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2781 #endif
2782 }
2783 
2784 u64 get_kvmclock_ns(struct kvm *kvm)
2785 {
2786 	struct kvm_arch *ka = &kvm->arch;
2787 	struct pvclock_vcpu_time_info hv_clock;
2788 	unsigned long flags;
2789 	u64 ret;
2790 
2791 	spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2792 	if (!ka->use_master_clock) {
2793 		spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2794 		return get_kvmclock_base_ns() + ka->kvmclock_offset;
2795 	}
2796 
2797 	hv_clock.tsc_timestamp = ka->master_cycle_now;
2798 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2799 	spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2800 
2801 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2802 	get_cpu();
2803 
2804 	if (__this_cpu_read(cpu_tsc_khz)) {
2805 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2806 				   &hv_clock.tsc_shift,
2807 				   &hv_clock.tsc_to_system_mul);
2808 		ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2809 	} else
2810 		ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2811 
2812 	put_cpu();
2813 
2814 	return ret;
2815 }
2816 
2817 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2818 				   struct gfn_to_hva_cache *cache,
2819 				   unsigned int offset)
2820 {
2821 	struct kvm_vcpu_arch *vcpu = &v->arch;
2822 	struct pvclock_vcpu_time_info guest_hv_clock;
2823 
2824 	if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2825 		&guest_hv_clock, offset, sizeof(guest_hv_clock))))
2826 		return;
2827 
2828 	/* This VCPU is paused, but it's legal for a guest to read another
2829 	 * VCPU's kvmclock, so we really have to follow the specification where
2830 	 * it says that version is odd if data is being modified, and even after
2831 	 * it is consistent.
2832 	 *
2833 	 * Version field updates must be kept separate.  This is because
2834 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2835 	 * writes within a string instruction are weakly ordered.  So there
2836 	 * are three writes overall.
2837 	 *
2838 	 * As a small optimization, only write the version field in the first
2839 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2840 	 * version field is the first in the struct.
2841 	 */
2842 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2843 
2844 	if (guest_hv_clock.version & 1)
2845 		++guest_hv_clock.version;  /* first time write, random junk */
2846 
2847 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2848 	kvm_write_guest_offset_cached(v->kvm, cache,
2849 				      &vcpu->hv_clock, offset,
2850 				      sizeof(vcpu->hv_clock.version));
2851 
2852 	smp_wmb();
2853 
2854 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2855 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2856 
2857 	if (vcpu->pvclock_set_guest_stopped_request) {
2858 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2859 		vcpu->pvclock_set_guest_stopped_request = false;
2860 	}
2861 
2862 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2863 
2864 	kvm_write_guest_offset_cached(v->kvm, cache,
2865 				      &vcpu->hv_clock, offset,
2866 				      sizeof(vcpu->hv_clock));
2867 
2868 	smp_wmb();
2869 
2870 	vcpu->hv_clock.version++;
2871 	kvm_write_guest_offset_cached(v->kvm, cache,
2872 				     &vcpu->hv_clock, offset,
2873 				     sizeof(vcpu->hv_clock.version));
2874 }
2875 
2876 static int kvm_guest_time_update(struct kvm_vcpu *v)
2877 {
2878 	unsigned long flags, tgt_tsc_khz;
2879 	struct kvm_vcpu_arch *vcpu = &v->arch;
2880 	struct kvm_arch *ka = &v->kvm->arch;
2881 	s64 kernel_ns;
2882 	u64 tsc_timestamp, host_tsc;
2883 	u8 pvclock_flags;
2884 	bool use_master_clock;
2885 
2886 	kernel_ns = 0;
2887 	host_tsc = 0;
2888 
2889 	/*
2890 	 * If the host uses TSC clock, then passthrough TSC as stable
2891 	 * to the guest.
2892 	 */
2893 	spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2894 	use_master_clock = ka->use_master_clock;
2895 	if (use_master_clock) {
2896 		host_tsc = ka->master_cycle_now;
2897 		kernel_ns = ka->master_kernel_ns;
2898 	}
2899 	spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2900 
2901 	/* Keep irq disabled to prevent changes to the clock */
2902 	local_irq_save(flags);
2903 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2904 	if (unlikely(tgt_tsc_khz == 0)) {
2905 		local_irq_restore(flags);
2906 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2907 		return 1;
2908 	}
2909 	if (!use_master_clock) {
2910 		host_tsc = rdtsc();
2911 		kernel_ns = get_kvmclock_base_ns();
2912 	}
2913 
2914 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2915 
2916 	/*
2917 	 * We may have to catch up the TSC to match elapsed wall clock
2918 	 * time for two reasons, even if kvmclock is used.
2919 	 *   1) CPU could have been running below the maximum TSC rate
2920 	 *   2) Broken TSC compensation resets the base at each VCPU
2921 	 *      entry to avoid unknown leaps of TSC even when running
2922 	 *      again on the same CPU.  This may cause apparent elapsed
2923 	 *      time to disappear, and the guest to stand still or run
2924 	 *	very slowly.
2925 	 */
2926 	if (vcpu->tsc_catchup) {
2927 		u64 tsc = compute_guest_tsc(v, kernel_ns);
2928 		if (tsc > tsc_timestamp) {
2929 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2930 			tsc_timestamp = tsc;
2931 		}
2932 	}
2933 
2934 	local_irq_restore(flags);
2935 
2936 	/* With all the info we got, fill in the values */
2937 
2938 	if (kvm_has_tsc_control)
2939 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz,
2940 					    v->arch.l1_tsc_scaling_ratio);
2941 
2942 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2943 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2944 				   &vcpu->hv_clock.tsc_shift,
2945 				   &vcpu->hv_clock.tsc_to_system_mul);
2946 		vcpu->hw_tsc_khz = tgt_tsc_khz;
2947 	}
2948 
2949 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2950 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2951 	vcpu->last_guest_tsc = tsc_timestamp;
2952 
2953 	/* If the host uses TSC clocksource, then it is stable */
2954 	pvclock_flags = 0;
2955 	if (use_master_clock)
2956 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2957 
2958 	vcpu->hv_clock.flags = pvclock_flags;
2959 
2960 	if (vcpu->pv_time_enabled)
2961 		kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
2962 	if (vcpu->xen.vcpu_info_set)
2963 		kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
2964 				       offsetof(struct compat_vcpu_info, time));
2965 	if (vcpu->xen.vcpu_time_info_set)
2966 		kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
2967 	if (v == kvm_get_vcpu(v->kvm, 0))
2968 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2969 	return 0;
2970 }
2971 
2972 /*
2973  * kvmclock updates which are isolated to a given vcpu, such as
2974  * vcpu->cpu migration, should not allow system_timestamp from
2975  * the rest of the vcpus to remain static. Otherwise ntp frequency
2976  * correction applies to one vcpu's system_timestamp but not
2977  * the others.
2978  *
2979  * So in those cases, request a kvmclock update for all vcpus.
2980  * We need to rate-limit these requests though, as they can
2981  * considerably slow guests that have a large number of vcpus.
2982  * The time for a remote vcpu to update its kvmclock is bound
2983  * by the delay we use to rate-limit the updates.
2984  */
2985 
2986 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2987 
2988 static void kvmclock_update_fn(struct work_struct *work)
2989 {
2990 	int i;
2991 	struct delayed_work *dwork = to_delayed_work(work);
2992 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2993 					   kvmclock_update_work);
2994 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2995 	struct kvm_vcpu *vcpu;
2996 
2997 	kvm_for_each_vcpu(i, vcpu, kvm) {
2998 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2999 		kvm_vcpu_kick(vcpu);
3000 	}
3001 }
3002 
3003 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
3004 {
3005 	struct kvm *kvm = v->kvm;
3006 
3007 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3008 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
3009 					KVMCLOCK_UPDATE_DELAY);
3010 }
3011 
3012 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
3013 
3014 static void kvmclock_sync_fn(struct work_struct *work)
3015 {
3016 	struct delayed_work *dwork = to_delayed_work(work);
3017 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3018 					   kvmclock_sync_work);
3019 	struct kvm *kvm = container_of(ka, struct kvm, arch);
3020 
3021 	if (!kvmclock_periodic_sync)
3022 		return;
3023 
3024 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
3025 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
3026 					KVMCLOCK_SYNC_PERIOD);
3027 }
3028 
3029 /*
3030  * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
3031  */
3032 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
3033 {
3034 	/* McStatusWrEn enabled? */
3035 	if (guest_cpuid_is_amd_or_hygon(vcpu))
3036 		return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
3037 
3038 	return false;
3039 }
3040 
3041 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3042 {
3043 	u64 mcg_cap = vcpu->arch.mcg_cap;
3044 	unsigned bank_num = mcg_cap & 0xff;
3045 	u32 msr = msr_info->index;
3046 	u64 data = msr_info->data;
3047 
3048 	switch (msr) {
3049 	case MSR_IA32_MCG_STATUS:
3050 		vcpu->arch.mcg_status = data;
3051 		break;
3052 	case MSR_IA32_MCG_CTL:
3053 		if (!(mcg_cap & MCG_CTL_P) &&
3054 		    (data || !msr_info->host_initiated))
3055 			return 1;
3056 		if (data != 0 && data != ~(u64)0)
3057 			return 1;
3058 		vcpu->arch.mcg_ctl = data;
3059 		break;
3060 	default:
3061 		if (msr >= MSR_IA32_MC0_CTL &&
3062 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
3063 			u32 offset = array_index_nospec(
3064 				msr - MSR_IA32_MC0_CTL,
3065 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3066 
3067 			/* only 0 or all 1s can be written to IA32_MCi_CTL
3068 			 * some Linux kernels though clear bit 10 in bank 4 to
3069 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
3070 			 * this to avoid an uncatched #GP in the guest
3071 			 */
3072 			if ((offset & 0x3) == 0 &&
3073 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
3074 				return -1;
3075 
3076 			/* MCi_STATUS */
3077 			if (!msr_info->host_initiated &&
3078 			    (offset & 0x3) == 1 && data != 0) {
3079 				if (!can_set_mci_status(vcpu))
3080 					return -1;
3081 			}
3082 
3083 			vcpu->arch.mce_banks[offset] = data;
3084 			break;
3085 		}
3086 		return 1;
3087 	}
3088 	return 0;
3089 }
3090 
3091 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
3092 {
3093 	u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3094 
3095 	return (vcpu->arch.apf.msr_en_val & mask) == mask;
3096 }
3097 
3098 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3099 {
3100 	gpa_t gpa = data & ~0x3f;
3101 
3102 	/* Bits 4:5 are reserved, Should be zero */
3103 	if (data & 0x30)
3104 		return 1;
3105 
3106 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3107 	    (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3108 		return 1;
3109 
3110 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3111 	    (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3112 		return 1;
3113 
3114 	if (!lapic_in_kernel(vcpu))
3115 		return data ? 1 : 0;
3116 
3117 	vcpu->arch.apf.msr_en_val = data;
3118 
3119 	if (!kvm_pv_async_pf_enabled(vcpu)) {
3120 		kvm_clear_async_pf_completion_queue(vcpu);
3121 		kvm_async_pf_hash_reset(vcpu);
3122 		return 0;
3123 	}
3124 
3125 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3126 					sizeof(u64)))
3127 		return 1;
3128 
3129 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3130 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3131 
3132 	kvm_async_pf_wakeup_all(vcpu);
3133 
3134 	return 0;
3135 }
3136 
3137 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3138 {
3139 	/* Bits 8-63 are reserved */
3140 	if (data >> 8)
3141 		return 1;
3142 
3143 	if (!lapic_in_kernel(vcpu))
3144 		return 1;
3145 
3146 	vcpu->arch.apf.msr_int_val = data;
3147 
3148 	vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3149 
3150 	return 0;
3151 }
3152 
3153 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3154 {
3155 	vcpu->arch.pv_time_enabled = false;
3156 	vcpu->arch.time = 0;
3157 }
3158 
3159 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3160 {
3161 	++vcpu->stat.tlb_flush;
3162 	static_call(kvm_x86_tlb_flush_all)(vcpu);
3163 }
3164 
3165 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3166 {
3167 	++vcpu->stat.tlb_flush;
3168 
3169 	if (!tdp_enabled) {
3170                /*
3171 		 * A TLB flush on behalf of the guest is equivalent to
3172 		 * INVPCID(all), toggling CR4.PGE, etc., which requires
3173 		 * a forced sync of the shadow page tables.  Unload the
3174 		 * entire MMU here and the subsequent load will sync the
3175 		 * shadow page tables, and also flush the TLB.
3176 		 */
3177 		kvm_mmu_unload(vcpu);
3178 		return;
3179 	}
3180 
3181 	static_call(kvm_x86_tlb_flush_guest)(vcpu);
3182 }
3183 
3184 static void record_steal_time(struct kvm_vcpu *vcpu)
3185 {
3186 	struct kvm_host_map map;
3187 	struct kvm_steal_time *st;
3188 
3189 	if (kvm_xen_msr_enabled(vcpu->kvm)) {
3190 		kvm_xen_runstate_set_running(vcpu);
3191 		return;
3192 	}
3193 
3194 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3195 		return;
3196 
3197 	/* -EAGAIN is returned in atomic context so we can just return. */
3198 	if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
3199 			&map, &vcpu->arch.st.cache, false))
3200 		return;
3201 
3202 	st = map.hva +
3203 		offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3204 
3205 	/*
3206 	 * Doing a TLB flush here, on the guest's behalf, can avoid
3207 	 * expensive IPIs.
3208 	 */
3209 	if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3210 		u8 st_preempted = xchg(&st->preempted, 0);
3211 
3212 		trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3213 				       st_preempted & KVM_VCPU_FLUSH_TLB);
3214 		if (st_preempted & KVM_VCPU_FLUSH_TLB)
3215 			kvm_vcpu_flush_tlb_guest(vcpu);
3216 	} else {
3217 		st->preempted = 0;
3218 	}
3219 
3220 	vcpu->arch.st.preempted = 0;
3221 
3222 	if (st->version & 1)
3223 		st->version += 1;  /* first time write, random junk */
3224 
3225 	st->version += 1;
3226 
3227 	smp_wmb();
3228 
3229 	st->steal += current->sched_info.run_delay -
3230 		vcpu->arch.st.last_steal;
3231 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
3232 
3233 	smp_wmb();
3234 
3235 	st->version += 1;
3236 
3237 	kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
3238 }
3239 
3240 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3241 {
3242 	bool pr = false;
3243 	u32 msr = msr_info->index;
3244 	u64 data = msr_info->data;
3245 
3246 	if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3247 		return kvm_xen_write_hypercall_page(vcpu, data);
3248 
3249 	switch (msr) {
3250 	case MSR_AMD64_NB_CFG:
3251 	case MSR_IA32_UCODE_WRITE:
3252 	case MSR_VM_HSAVE_PA:
3253 	case MSR_AMD64_PATCH_LOADER:
3254 	case MSR_AMD64_BU_CFG2:
3255 	case MSR_AMD64_DC_CFG:
3256 	case MSR_F15H_EX_CFG:
3257 		break;
3258 
3259 	case MSR_IA32_UCODE_REV:
3260 		if (msr_info->host_initiated)
3261 			vcpu->arch.microcode_version = data;
3262 		break;
3263 	case MSR_IA32_ARCH_CAPABILITIES:
3264 		if (!msr_info->host_initiated)
3265 			return 1;
3266 		vcpu->arch.arch_capabilities = data;
3267 		break;
3268 	case MSR_IA32_PERF_CAPABILITIES: {
3269 		struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3270 
3271 		if (!msr_info->host_initiated)
3272 			return 1;
3273 		if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3274 			return 1;
3275 		if (data & ~msr_ent.data)
3276 			return 1;
3277 
3278 		vcpu->arch.perf_capabilities = data;
3279 
3280 		return 0;
3281 		}
3282 	case MSR_EFER:
3283 		return set_efer(vcpu, msr_info);
3284 	case MSR_K7_HWCR:
3285 		data &= ~(u64)0x40;	/* ignore flush filter disable */
3286 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
3287 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
3288 
3289 		/* Handle McStatusWrEn */
3290 		if (data == BIT_ULL(18)) {
3291 			vcpu->arch.msr_hwcr = data;
3292 		} else if (data != 0) {
3293 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3294 				    data);
3295 			return 1;
3296 		}
3297 		break;
3298 	case MSR_FAM10H_MMIO_CONF_BASE:
3299 		if (data != 0) {
3300 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3301 				    "0x%llx\n", data);
3302 			return 1;
3303 		}
3304 		break;
3305 	case 0x200 ... 0x2ff:
3306 		return kvm_mtrr_set_msr(vcpu, msr, data);
3307 	case MSR_IA32_APICBASE:
3308 		return kvm_set_apic_base(vcpu, msr_info);
3309 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3310 		return kvm_x2apic_msr_write(vcpu, msr, data);
3311 	case MSR_IA32_TSC_DEADLINE:
3312 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
3313 		break;
3314 	case MSR_IA32_TSC_ADJUST:
3315 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3316 			if (!msr_info->host_initiated) {
3317 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3318 				adjust_tsc_offset_guest(vcpu, adj);
3319 			}
3320 			vcpu->arch.ia32_tsc_adjust_msr = data;
3321 		}
3322 		break;
3323 	case MSR_IA32_MISC_ENABLE:
3324 		if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3325 		    ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3326 			if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3327 				return 1;
3328 			vcpu->arch.ia32_misc_enable_msr = data;
3329 			kvm_update_cpuid_runtime(vcpu);
3330 		} else {
3331 			vcpu->arch.ia32_misc_enable_msr = data;
3332 		}
3333 		break;
3334 	case MSR_IA32_SMBASE:
3335 		if (!msr_info->host_initiated)
3336 			return 1;
3337 		vcpu->arch.smbase = data;
3338 		break;
3339 	case MSR_IA32_POWER_CTL:
3340 		vcpu->arch.msr_ia32_power_ctl = data;
3341 		break;
3342 	case MSR_IA32_TSC:
3343 		if (msr_info->host_initiated) {
3344 			kvm_synchronize_tsc(vcpu, data);
3345 		} else {
3346 			u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3347 			adjust_tsc_offset_guest(vcpu, adj);
3348 			vcpu->arch.ia32_tsc_adjust_msr += adj;
3349 		}
3350 		break;
3351 	case MSR_IA32_XSS:
3352 		if (!msr_info->host_initiated &&
3353 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3354 			return 1;
3355 		/*
3356 		 * KVM supports exposing PT to the guest, but does not support
3357 		 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3358 		 * XSAVES/XRSTORS to save/restore PT MSRs.
3359 		 */
3360 		if (data & ~supported_xss)
3361 			return 1;
3362 		vcpu->arch.ia32_xss = data;
3363 		break;
3364 	case MSR_SMI_COUNT:
3365 		if (!msr_info->host_initiated)
3366 			return 1;
3367 		vcpu->arch.smi_count = data;
3368 		break;
3369 	case MSR_KVM_WALL_CLOCK_NEW:
3370 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3371 			return 1;
3372 
3373 		vcpu->kvm->arch.wall_clock = data;
3374 		kvm_write_wall_clock(vcpu->kvm, data, 0);
3375 		break;
3376 	case MSR_KVM_WALL_CLOCK:
3377 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3378 			return 1;
3379 
3380 		vcpu->kvm->arch.wall_clock = data;
3381 		kvm_write_wall_clock(vcpu->kvm, data, 0);
3382 		break;
3383 	case MSR_KVM_SYSTEM_TIME_NEW:
3384 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3385 			return 1;
3386 
3387 		kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3388 		break;
3389 	case MSR_KVM_SYSTEM_TIME:
3390 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3391 			return 1;
3392 
3393 		kvm_write_system_time(vcpu, data, true,  msr_info->host_initiated);
3394 		break;
3395 	case MSR_KVM_ASYNC_PF_EN:
3396 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3397 			return 1;
3398 
3399 		if (kvm_pv_enable_async_pf(vcpu, data))
3400 			return 1;
3401 		break;
3402 	case MSR_KVM_ASYNC_PF_INT:
3403 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3404 			return 1;
3405 
3406 		if (kvm_pv_enable_async_pf_int(vcpu, data))
3407 			return 1;
3408 		break;
3409 	case MSR_KVM_ASYNC_PF_ACK:
3410 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3411 			return 1;
3412 		if (data & 0x1) {
3413 			vcpu->arch.apf.pageready_pending = false;
3414 			kvm_check_async_pf_completion(vcpu);
3415 		}
3416 		break;
3417 	case MSR_KVM_STEAL_TIME:
3418 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3419 			return 1;
3420 
3421 		if (unlikely(!sched_info_on()))
3422 			return 1;
3423 
3424 		if (data & KVM_STEAL_RESERVED_MASK)
3425 			return 1;
3426 
3427 		vcpu->arch.st.msr_val = data;
3428 
3429 		if (!(data & KVM_MSR_ENABLED))
3430 			break;
3431 
3432 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3433 
3434 		break;
3435 	case MSR_KVM_PV_EOI_EN:
3436 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3437 			return 1;
3438 
3439 		if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3440 			return 1;
3441 		break;
3442 
3443 	case MSR_KVM_POLL_CONTROL:
3444 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3445 			return 1;
3446 
3447 		/* only enable bit supported */
3448 		if (data & (-1ULL << 1))
3449 			return 1;
3450 
3451 		vcpu->arch.msr_kvm_poll_control = data;
3452 		break;
3453 
3454 	case MSR_IA32_MCG_CTL:
3455 	case MSR_IA32_MCG_STATUS:
3456 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3457 		return set_msr_mce(vcpu, msr_info);
3458 
3459 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3460 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3461 		pr = true;
3462 		fallthrough;
3463 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3464 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3465 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3466 			return kvm_pmu_set_msr(vcpu, msr_info);
3467 
3468 		if (pr || data != 0)
3469 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3470 				    "0x%x data 0x%llx\n", msr, data);
3471 		break;
3472 	case MSR_K7_CLK_CTL:
3473 		/*
3474 		 * Ignore all writes to this no longer documented MSR.
3475 		 * Writes are only relevant for old K7 processors,
3476 		 * all pre-dating SVM, but a recommended workaround from
3477 		 * AMD for these chips. It is possible to specify the
3478 		 * affected processor models on the command line, hence
3479 		 * the need to ignore the workaround.
3480 		 */
3481 		break;
3482 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3483 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3484 	case HV_X64_MSR_SYNDBG_OPTIONS:
3485 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3486 	case HV_X64_MSR_CRASH_CTL:
3487 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3488 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3489 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3490 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3491 		return kvm_hv_set_msr_common(vcpu, msr, data,
3492 					     msr_info->host_initiated);
3493 	case MSR_IA32_BBL_CR_CTL3:
3494 		/* Drop writes to this legacy MSR -- see rdmsr
3495 		 * counterpart for further detail.
3496 		 */
3497 		if (report_ignored_msrs)
3498 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3499 				msr, data);
3500 		break;
3501 	case MSR_AMD64_OSVW_ID_LENGTH:
3502 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3503 			return 1;
3504 		vcpu->arch.osvw.length = data;
3505 		break;
3506 	case MSR_AMD64_OSVW_STATUS:
3507 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3508 			return 1;
3509 		vcpu->arch.osvw.status = data;
3510 		break;
3511 	case MSR_PLATFORM_INFO:
3512 		if (!msr_info->host_initiated ||
3513 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3514 		     cpuid_fault_enabled(vcpu)))
3515 			return 1;
3516 		vcpu->arch.msr_platform_info = data;
3517 		break;
3518 	case MSR_MISC_FEATURES_ENABLES:
3519 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3520 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3521 		     !supports_cpuid_fault(vcpu)))
3522 			return 1;
3523 		vcpu->arch.msr_misc_features_enables = data;
3524 		break;
3525 	default:
3526 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3527 			return kvm_pmu_set_msr(vcpu, msr_info);
3528 		return KVM_MSR_RET_INVALID;
3529 	}
3530 	return 0;
3531 }
3532 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3533 
3534 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3535 {
3536 	u64 data;
3537 	u64 mcg_cap = vcpu->arch.mcg_cap;
3538 	unsigned bank_num = mcg_cap & 0xff;
3539 
3540 	switch (msr) {
3541 	case MSR_IA32_P5_MC_ADDR:
3542 	case MSR_IA32_P5_MC_TYPE:
3543 		data = 0;
3544 		break;
3545 	case MSR_IA32_MCG_CAP:
3546 		data = vcpu->arch.mcg_cap;
3547 		break;
3548 	case MSR_IA32_MCG_CTL:
3549 		if (!(mcg_cap & MCG_CTL_P) && !host)
3550 			return 1;
3551 		data = vcpu->arch.mcg_ctl;
3552 		break;
3553 	case MSR_IA32_MCG_STATUS:
3554 		data = vcpu->arch.mcg_status;
3555 		break;
3556 	default:
3557 		if (msr >= MSR_IA32_MC0_CTL &&
3558 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
3559 			u32 offset = array_index_nospec(
3560 				msr - MSR_IA32_MC0_CTL,
3561 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3562 
3563 			data = vcpu->arch.mce_banks[offset];
3564 			break;
3565 		}
3566 		return 1;
3567 	}
3568 	*pdata = data;
3569 	return 0;
3570 }
3571 
3572 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3573 {
3574 	switch (msr_info->index) {
3575 	case MSR_IA32_PLATFORM_ID:
3576 	case MSR_IA32_EBL_CR_POWERON:
3577 	case MSR_IA32_LASTBRANCHFROMIP:
3578 	case MSR_IA32_LASTBRANCHTOIP:
3579 	case MSR_IA32_LASTINTFROMIP:
3580 	case MSR_IA32_LASTINTTOIP:
3581 	case MSR_AMD64_SYSCFG:
3582 	case MSR_K8_TSEG_ADDR:
3583 	case MSR_K8_TSEG_MASK:
3584 	case MSR_VM_HSAVE_PA:
3585 	case MSR_K8_INT_PENDING_MSG:
3586 	case MSR_AMD64_NB_CFG:
3587 	case MSR_FAM10H_MMIO_CONF_BASE:
3588 	case MSR_AMD64_BU_CFG2:
3589 	case MSR_IA32_PERF_CTL:
3590 	case MSR_AMD64_DC_CFG:
3591 	case MSR_F15H_EX_CFG:
3592 	/*
3593 	 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3594 	 * limit) MSRs. Just return 0, as we do not want to expose the host
3595 	 * data here. Do not conditionalize this on CPUID, as KVM does not do
3596 	 * so for existing CPU-specific MSRs.
3597 	 */
3598 	case MSR_RAPL_POWER_UNIT:
3599 	case MSR_PP0_ENERGY_STATUS:	/* Power plane 0 (core) */
3600 	case MSR_PP1_ENERGY_STATUS:	/* Power plane 1 (graphics uncore) */
3601 	case MSR_PKG_ENERGY_STATUS:	/* Total package */
3602 	case MSR_DRAM_ENERGY_STATUS:	/* DRAM controller */
3603 		msr_info->data = 0;
3604 		break;
3605 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3606 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3607 			return kvm_pmu_get_msr(vcpu, msr_info);
3608 		if (!msr_info->host_initiated)
3609 			return 1;
3610 		msr_info->data = 0;
3611 		break;
3612 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3613 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3614 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3615 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3616 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3617 			return kvm_pmu_get_msr(vcpu, msr_info);
3618 		msr_info->data = 0;
3619 		break;
3620 	case MSR_IA32_UCODE_REV:
3621 		msr_info->data = vcpu->arch.microcode_version;
3622 		break;
3623 	case MSR_IA32_ARCH_CAPABILITIES:
3624 		if (!msr_info->host_initiated &&
3625 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3626 			return 1;
3627 		msr_info->data = vcpu->arch.arch_capabilities;
3628 		break;
3629 	case MSR_IA32_PERF_CAPABILITIES:
3630 		if (!msr_info->host_initiated &&
3631 		    !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3632 			return 1;
3633 		msr_info->data = vcpu->arch.perf_capabilities;
3634 		break;
3635 	case MSR_IA32_POWER_CTL:
3636 		msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3637 		break;
3638 	case MSR_IA32_TSC: {
3639 		/*
3640 		 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3641 		 * even when not intercepted. AMD manual doesn't explicitly
3642 		 * state this but appears to behave the same.
3643 		 *
3644 		 * On userspace reads and writes, however, we unconditionally
3645 		 * return L1's TSC value to ensure backwards-compatible
3646 		 * behavior for migration.
3647 		 */
3648 		u64 offset, ratio;
3649 
3650 		if (msr_info->host_initiated) {
3651 			offset = vcpu->arch.l1_tsc_offset;
3652 			ratio = vcpu->arch.l1_tsc_scaling_ratio;
3653 		} else {
3654 			offset = vcpu->arch.tsc_offset;
3655 			ratio = vcpu->arch.tsc_scaling_ratio;
3656 		}
3657 
3658 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc(), ratio) + offset;
3659 		break;
3660 	}
3661 	case MSR_MTRRcap:
3662 	case 0x200 ... 0x2ff:
3663 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3664 	case 0xcd: /* fsb frequency */
3665 		msr_info->data = 3;
3666 		break;
3667 		/*
3668 		 * MSR_EBC_FREQUENCY_ID
3669 		 * Conservative value valid for even the basic CPU models.
3670 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3671 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3672 		 * and 266MHz for model 3, or 4. Set Core Clock
3673 		 * Frequency to System Bus Frequency Ratio to 1 (bits
3674 		 * 31:24) even though these are only valid for CPU
3675 		 * models > 2, however guests may end up dividing or
3676 		 * multiplying by zero otherwise.
3677 		 */
3678 	case MSR_EBC_FREQUENCY_ID:
3679 		msr_info->data = 1 << 24;
3680 		break;
3681 	case MSR_IA32_APICBASE:
3682 		msr_info->data = kvm_get_apic_base(vcpu);
3683 		break;
3684 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3685 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3686 	case MSR_IA32_TSC_DEADLINE:
3687 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3688 		break;
3689 	case MSR_IA32_TSC_ADJUST:
3690 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3691 		break;
3692 	case MSR_IA32_MISC_ENABLE:
3693 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3694 		break;
3695 	case MSR_IA32_SMBASE:
3696 		if (!msr_info->host_initiated)
3697 			return 1;
3698 		msr_info->data = vcpu->arch.smbase;
3699 		break;
3700 	case MSR_SMI_COUNT:
3701 		msr_info->data = vcpu->arch.smi_count;
3702 		break;
3703 	case MSR_IA32_PERF_STATUS:
3704 		/* TSC increment by tick */
3705 		msr_info->data = 1000ULL;
3706 		/* CPU multiplier */
3707 		msr_info->data |= (((uint64_t)4ULL) << 40);
3708 		break;
3709 	case MSR_EFER:
3710 		msr_info->data = vcpu->arch.efer;
3711 		break;
3712 	case MSR_KVM_WALL_CLOCK:
3713 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3714 			return 1;
3715 
3716 		msr_info->data = vcpu->kvm->arch.wall_clock;
3717 		break;
3718 	case MSR_KVM_WALL_CLOCK_NEW:
3719 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3720 			return 1;
3721 
3722 		msr_info->data = vcpu->kvm->arch.wall_clock;
3723 		break;
3724 	case MSR_KVM_SYSTEM_TIME:
3725 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3726 			return 1;
3727 
3728 		msr_info->data = vcpu->arch.time;
3729 		break;
3730 	case MSR_KVM_SYSTEM_TIME_NEW:
3731 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3732 			return 1;
3733 
3734 		msr_info->data = vcpu->arch.time;
3735 		break;
3736 	case MSR_KVM_ASYNC_PF_EN:
3737 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3738 			return 1;
3739 
3740 		msr_info->data = vcpu->arch.apf.msr_en_val;
3741 		break;
3742 	case MSR_KVM_ASYNC_PF_INT:
3743 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3744 			return 1;
3745 
3746 		msr_info->data = vcpu->arch.apf.msr_int_val;
3747 		break;
3748 	case MSR_KVM_ASYNC_PF_ACK:
3749 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3750 			return 1;
3751 
3752 		msr_info->data = 0;
3753 		break;
3754 	case MSR_KVM_STEAL_TIME:
3755 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3756 			return 1;
3757 
3758 		msr_info->data = vcpu->arch.st.msr_val;
3759 		break;
3760 	case MSR_KVM_PV_EOI_EN:
3761 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3762 			return 1;
3763 
3764 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
3765 		break;
3766 	case MSR_KVM_POLL_CONTROL:
3767 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3768 			return 1;
3769 
3770 		msr_info->data = vcpu->arch.msr_kvm_poll_control;
3771 		break;
3772 	case MSR_IA32_P5_MC_ADDR:
3773 	case MSR_IA32_P5_MC_TYPE:
3774 	case MSR_IA32_MCG_CAP:
3775 	case MSR_IA32_MCG_CTL:
3776 	case MSR_IA32_MCG_STATUS:
3777 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3778 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3779 				   msr_info->host_initiated);
3780 	case MSR_IA32_XSS:
3781 		if (!msr_info->host_initiated &&
3782 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3783 			return 1;
3784 		msr_info->data = vcpu->arch.ia32_xss;
3785 		break;
3786 	case MSR_K7_CLK_CTL:
3787 		/*
3788 		 * Provide expected ramp-up count for K7. All other
3789 		 * are set to zero, indicating minimum divisors for
3790 		 * every field.
3791 		 *
3792 		 * This prevents guest kernels on AMD host with CPU
3793 		 * type 6, model 8 and higher from exploding due to
3794 		 * the rdmsr failing.
3795 		 */
3796 		msr_info->data = 0x20000000;
3797 		break;
3798 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3799 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3800 	case HV_X64_MSR_SYNDBG_OPTIONS:
3801 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3802 	case HV_X64_MSR_CRASH_CTL:
3803 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3804 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3805 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3806 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3807 		return kvm_hv_get_msr_common(vcpu,
3808 					     msr_info->index, &msr_info->data,
3809 					     msr_info->host_initiated);
3810 	case MSR_IA32_BBL_CR_CTL3:
3811 		/* This legacy MSR exists but isn't fully documented in current
3812 		 * silicon.  It is however accessed by winxp in very narrow
3813 		 * scenarios where it sets bit #19, itself documented as
3814 		 * a "reserved" bit.  Best effort attempt to source coherent
3815 		 * read data here should the balance of the register be
3816 		 * interpreted by the guest:
3817 		 *
3818 		 * L2 cache control register 3: 64GB range, 256KB size,
3819 		 * enabled, latency 0x1, configured
3820 		 */
3821 		msr_info->data = 0xbe702111;
3822 		break;
3823 	case MSR_AMD64_OSVW_ID_LENGTH:
3824 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3825 			return 1;
3826 		msr_info->data = vcpu->arch.osvw.length;
3827 		break;
3828 	case MSR_AMD64_OSVW_STATUS:
3829 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3830 			return 1;
3831 		msr_info->data = vcpu->arch.osvw.status;
3832 		break;
3833 	case MSR_PLATFORM_INFO:
3834 		if (!msr_info->host_initiated &&
3835 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3836 			return 1;
3837 		msr_info->data = vcpu->arch.msr_platform_info;
3838 		break;
3839 	case MSR_MISC_FEATURES_ENABLES:
3840 		msr_info->data = vcpu->arch.msr_misc_features_enables;
3841 		break;
3842 	case MSR_K7_HWCR:
3843 		msr_info->data = vcpu->arch.msr_hwcr;
3844 		break;
3845 	default:
3846 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3847 			return kvm_pmu_get_msr(vcpu, msr_info);
3848 		return KVM_MSR_RET_INVALID;
3849 	}
3850 	return 0;
3851 }
3852 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3853 
3854 /*
3855  * Read or write a bunch of msrs. All parameters are kernel addresses.
3856  *
3857  * @return number of msrs set successfully.
3858  */
3859 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3860 		    struct kvm_msr_entry *entries,
3861 		    int (*do_msr)(struct kvm_vcpu *vcpu,
3862 				  unsigned index, u64 *data))
3863 {
3864 	int i;
3865 
3866 	for (i = 0; i < msrs->nmsrs; ++i)
3867 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
3868 			break;
3869 
3870 	return i;
3871 }
3872 
3873 /*
3874  * Read or write a bunch of msrs. Parameters are user addresses.
3875  *
3876  * @return number of msrs set successfully.
3877  */
3878 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3879 		  int (*do_msr)(struct kvm_vcpu *vcpu,
3880 				unsigned index, u64 *data),
3881 		  int writeback)
3882 {
3883 	struct kvm_msrs msrs;
3884 	struct kvm_msr_entry *entries;
3885 	int r, n;
3886 	unsigned size;
3887 
3888 	r = -EFAULT;
3889 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3890 		goto out;
3891 
3892 	r = -E2BIG;
3893 	if (msrs.nmsrs >= MAX_IO_MSRS)
3894 		goto out;
3895 
3896 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3897 	entries = memdup_user(user_msrs->entries, size);
3898 	if (IS_ERR(entries)) {
3899 		r = PTR_ERR(entries);
3900 		goto out;
3901 	}
3902 
3903 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3904 	if (r < 0)
3905 		goto out_free;
3906 
3907 	r = -EFAULT;
3908 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
3909 		goto out_free;
3910 
3911 	r = n;
3912 
3913 out_free:
3914 	kfree(entries);
3915 out:
3916 	return r;
3917 }
3918 
3919 static inline bool kvm_can_mwait_in_guest(void)
3920 {
3921 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
3922 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
3923 		boot_cpu_has(X86_FEATURE_ARAT);
3924 }
3925 
3926 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
3927 					    struct kvm_cpuid2 __user *cpuid_arg)
3928 {
3929 	struct kvm_cpuid2 cpuid;
3930 	int r;
3931 
3932 	r = -EFAULT;
3933 	if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3934 		return r;
3935 
3936 	r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3937 	if (r)
3938 		return r;
3939 
3940 	r = -EFAULT;
3941 	if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3942 		return r;
3943 
3944 	return 0;
3945 }
3946 
3947 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3948 {
3949 	int r = 0;
3950 
3951 	switch (ext) {
3952 	case KVM_CAP_IRQCHIP:
3953 	case KVM_CAP_HLT:
3954 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3955 	case KVM_CAP_SET_TSS_ADDR:
3956 	case KVM_CAP_EXT_CPUID:
3957 	case KVM_CAP_EXT_EMUL_CPUID:
3958 	case KVM_CAP_CLOCKSOURCE:
3959 	case KVM_CAP_PIT:
3960 	case KVM_CAP_NOP_IO_DELAY:
3961 	case KVM_CAP_MP_STATE:
3962 	case KVM_CAP_SYNC_MMU:
3963 	case KVM_CAP_USER_NMI:
3964 	case KVM_CAP_REINJECT_CONTROL:
3965 	case KVM_CAP_IRQ_INJECT_STATUS:
3966 	case KVM_CAP_IOEVENTFD:
3967 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
3968 	case KVM_CAP_PIT2:
3969 	case KVM_CAP_PIT_STATE2:
3970 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3971 	case KVM_CAP_VCPU_EVENTS:
3972 	case KVM_CAP_HYPERV:
3973 	case KVM_CAP_HYPERV_VAPIC:
3974 	case KVM_CAP_HYPERV_SPIN:
3975 	case KVM_CAP_HYPERV_SYNIC:
3976 	case KVM_CAP_HYPERV_SYNIC2:
3977 	case KVM_CAP_HYPERV_VP_INDEX:
3978 	case KVM_CAP_HYPERV_EVENTFD:
3979 	case KVM_CAP_HYPERV_TLBFLUSH:
3980 	case KVM_CAP_HYPERV_SEND_IPI:
3981 	case KVM_CAP_HYPERV_CPUID:
3982 	case KVM_CAP_HYPERV_ENFORCE_CPUID:
3983 	case KVM_CAP_SYS_HYPERV_CPUID:
3984 	case KVM_CAP_PCI_SEGMENT:
3985 	case KVM_CAP_DEBUGREGS:
3986 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
3987 	case KVM_CAP_XSAVE:
3988 	case KVM_CAP_ASYNC_PF:
3989 	case KVM_CAP_ASYNC_PF_INT:
3990 	case KVM_CAP_GET_TSC_KHZ:
3991 	case KVM_CAP_KVMCLOCK_CTRL:
3992 	case KVM_CAP_READONLY_MEM:
3993 	case KVM_CAP_HYPERV_TIME:
3994 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3995 	case KVM_CAP_TSC_DEADLINE_TIMER:
3996 	case KVM_CAP_DISABLE_QUIRKS:
3997 	case KVM_CAP_SET_BOOT_CPU_ID:
3998  	case KVM_CAP_SPLIT_IRQCHIP:
3999 	case KVM_CAP_IMMEDIATE_EXIT:
4000 	case KVM_CAP_PMU_EVENT_FILTER:
4001 	case KVM_CAP_GET_MSR_FEATURES:
4002 	case KVM_CAP_MSR_PLATFORM_INFO:
4003 	case KVM_CAP_EXCEPTION_PAYLOAD:
4004 	case KVM_CAP_SET_GUEST_DEBUG:
4005 	case KVM_CAP_LAST_CPU:
4006 	case KVM_CAP_X86_USER_SPACE_MSR:
4007 	case KVM_CAP_X86_MSR_FILTER:
4008 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4009 #ifdef CONFIG_X86_SGX_KVM
4010 	case KVM_CAP_SGX_ATTRIBUTE:
4011 #endif
4012 	case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
4013 	case KVM_CAP_SREGS2:
4014 	case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
4015 		r = 1;
4016 		break;
4017 	case KVM_CAP_EXIT_HYPERCALL:
4018 		r = KVM_EXIT_HYPERCALL_VALID_MASK;
4019 		break;
4020 	case KVM_CAP_SET_GUEST_DEBUG2:
4021 		return KVM_GUESTDBG_VALID_MASK;
4022 #ifdef CONFIG_KVM_XEN
4023 	case KVM_CAP_XEN_HVM:
4024 		r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
4025 		    KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
4026 		    KVM_XEN_HVM_CONFIG_SHARED_INFO;
4027 		if (sched_info_on())
4028 			r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
4029 		break;
4030 #endif
4031 	case KVM_CAP_SYNC_REGS:
4032 		r = KVM_SYNC_X86_VALID_FIELDS;
4033 		break;
4034 	case KVM_CAP_ADJUST_CLOCK:
4035 		r = KVM_CLOCK_TSC_STABLE;
4036 		break;
4037 	case KVM_CAP_X86_DISABLE_EXITS:
4038 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
4039 		      KVM_X86_DISABLE_EXITS_CSTATE;
4040 		if(kvm_can_mwait_in_guest())
4041 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
4042 		break;
4043 	case KVM_CAP_X86_SMM:
4044 		/* SMBASE is usually relocated above 1M on modern chipsets,
4045 		 * and SMM handlers might indeed rely on 4G segment limits,
4046 		 * so do not report SMM to be available if real mode is
4047 		 * emulated via vm86 mode.  Still, do not go to great lengths
4048 		 * to avoid userspace's usage of the feature, because it is a
4049 		 * fringe case that is not enabled except via specific settings
4050 		 * of the module parameters.
4051 		 */
4052 		r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
4053 		break;
4054 	case KVM_CAP_VAPIC:
4055 		r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
4056 		break;
4057 	case KVM_CAP_NR_VCPUS:
4058 		r = KVM_SOFT_MAX_VCPUS;
4059 		break;
4060 	case KVM_CAP_MAX_VCPUS:
4061 		r = KVM_MAX_VCPUS;
4062 		break;
4063 	case KVM_CAP_MAX_VCPU_ID:
4064 		r = KVM_MAX_VCPU_ID;
4065 		break;
4066 	case KVM_CAP_PV_MMU:	/* obsolete */
4067 		r = 0;
4068 		break;
4069 	case KVM_CAP_MCE:
4070 		r = KVM_MAX_MCE_BANKS;
4071 		break;
4072 	case KVM_CAP_XCRS:
4073 		r = boot_cpu_has(X86_FEATURE_XSAVE);
4074 		break;
4075 	case KVM_CAP_TSC_CONTROL:
4076 		r = kvm_has_tsc_control;
4077 		break;
4078 	case KVM_CAP_X2APIC_API:
4079 		r = KVM_X2APIC_API_VALID_FLAGS;
4080 		break;
4081 	case KVM_CAP_NESTED_STATE:
4082 		r = kvm_x86_ops.nested_ops->get_state ?
4083 			kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
4084 		break;
4085 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4086 		r = kvm_x86_ops.enable_direct_tlbflush != NULL;
4087 		break;
4088 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4089 		r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
4090 		break;
4091 	case KVM_CAP_SMALLER_MAXPHYADDR:
4092 		r = (int) allow_smaller_maxphyaddr;
4093 		break;
4094 	case KVM_CAP_STEAL_TIME:
4095 		r = sched_info_on();
4096 		break;
4097 	case KVM_CAP_X86_BUS_LOCK_EXIT:
4098 		if (kvm_has_bus_lock_exit)
4099 			r = KVM_BUS_LOCK_DETECTION_OFF |
4100 			    KVM_BUS_LOCK_DETECTION_EXIT;
4101 		else
4102 			r = 0;
4103 		break;
4104 	default:
4105 		break;
4106 	}
4107 	return r;
4108 
4109 }
4110 
4111 long kvm_arch_dev_ioctl(struct file *filp,
4112 			unsigned int ioctl, unsigned long arg)
4113 {
4114 	void __user *argp = (void __user *)arg;
4115 	long r;
4116 
4117 	switch (ioctl) {
4118 	case KVM_GET_MSR_INDEX_LIST: {
4119 		struct kvm_msr_list __user *user_msr_list = argp;
4120 		struct kvm_msr_list msr_list;
4121 		unsigned n;
4122 
4123 		r = -EFAULT;
4124 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4125 			goto out;
4126 		n = msr_list.nmsrs;
4127 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4128 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4129 			goto out;
4130 		r = -E2BIG;
4131 		if (n < msr_list.nmsrs)
4132 			goto out;
4133 		r = -EFAULT;
4134 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4135 				 num_msrs_to_save * sizeof(u32)))
4136 			goto out;
4137 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4138 				 &emulated_msrs,
4139 				 num_emulated_msrs * sizeof(u32)))
4140 			goto out;
4141 		r = 0;
4142 		break;
4143 	}
4144 	case KVM_GET_SUPPORTED_CPUID:
4145 	case KVM_GET_EMULATED_CPUID: {
4146 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4147 		struct kvm_cpuid2 cpuid;
4148 
4149 		r = -EFAULT;
4150 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4151 			goto out;
4152 
4153 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4154 					    ioctl);
4155 		if (r)
4156 			goto out;
4157 
4158 		r = -EFAULT;
4159 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4160 			goto out;
4161 		r = 0;
4162 		break;
4163 	}
4164 	case KVM_X86_GET_MCE_CAP_SUPPORTED:
4165 		r = -EFAULT;
4166 		if (copy_to_user(argp, &kvm_mce_cap_supported,
4167 				 sizeof(kvm_mce_cap_supported)))
4168 			goto out;
4169 		r = 0;
4170 		break;
4171 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4172 		struct kvm_msr_list __user *user_msr_list = argp;
4173 		struct kvm_msr_list msr_list;
4174 		unsigned int n;
4175 
4176 		r = -EFAULT;
4177 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4178 			goto out;
4179 		n = msr_list.nmsrs;
4180 		msr_list.nmsrs = num_msr_based_features;
4181 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4182 			goto out;
4183 		r = -E2BIG;
4184 		if (n < msr_list.nmsrs)
4185 			goto out;
4186 		r = -EFAULT;
4187 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
4188 				 num_msr_based_features * sizeof(u32)))
4189 			goto out;
4190 		r = 0;
4191 		break;
4192 	}
4193 	case KVM_GET_MSRS:
4194 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
4195 		break;
4196 	case KVM_GET_SUPPORTED_HV_CPUID:
4197 		r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4198 		break;
4199 	default:
4200 		r = -EINVAL;
4201 		break;
4202 	}
4203 out:
4204 	return r;
4205 }
4206 
4207 static void wbinvd_ipi(void *garbage)
4208 {
4209 	wbinvd();
4210 }
4211 
4212 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4213 {
4214 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4215 }
4216 
4217 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4218 {
4219 	/* Address WBINVD may be executed by guest */
4220 	if (need_emulate_wbinvd(vcpu)) {
4221 		if (static_call(kvm_x86_has_wbinvd_exit)())
4222 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4223 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4224 			smp_call_function_single(vcpu->cpu,
4225 					wbinvd_ipi, NULL, 1);
4226 	}
4227 
4228 	static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4229 
4230 	/* Save host pkru register if supported */
4231 	vcpu->arch.host_pkru = read_pkru();
4232 
4233 	/* Apply any externally detected TSC adjustments (due to suspend) */
4234 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4235 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4236 		vcpu->arch.tsc_offset_adjustment = 0;
4237 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4238 	}
4239 
4240 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4241 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4242 				rdtsc() - vcpu->arch.last_host_tsc;
4243 		if (tsc_delta < 0)
4244 			mark_tsc_unstable("KVM discovered backwards TSC");
4245 
4246 		if (kvm_check_tsc_unstable()) {
4247 			u64 offset = kvm_compute_l1_tsc_offset(vcpu,
4248 						vcpu->arch.last_guest_tsc);
4249 			kvm_vcpu_write_tsc_offset(vcpu, offset);
4250 			vcpu->arch.tsc_catchup = 1;
4251 		}
4252 
4253 		if (kvm_lapic_hv_timer_in_use(vcpu))
4254 			kvm_lapic_restart_hv_timer(vcpu);
4255 
4256 		/*
4257 		 * On a host with synchronized TSC, there is no need to update
4258 		 * kvmclock on vcpu->cpu migration
4259 		 */
4260 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4261 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4262 		if (vcpu->cpu != cpu)
4263 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4264 		vcpu->cpu = cpu;
4265 	}
4266 
4267 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4268 }
4269 
4270 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4271 {
4272 	struct kvm_host_map map;
4273 	struct kvm_steal_time *st;
4274 
4275 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4276 		return;
4277 
4278 	if (vcpu->arch.st.preempted)
4279 		return;
4280 
4281 	if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
4282 			&vcpu->arch.st.cache, true))
4283 		return;
4284 
4285 	st = map.hva +
4286 		offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
4287 
4288 	st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4289 
4290 	kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
4291 }
4292 
4293 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4294 {
4295 	int idx;
4296 
4297 	if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4298 		vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4299 
4300 	/*
4301 	 * Take the srcu lock as memslots will be accessed to check the gfn
4302 	 * cache generation against the memslots generation.
4303 	 */
4304 	idx = srcu_read_lock(&vcpu->kvm->srcu);
4305 	if (kvm_xen_msr_enabled(vcpu->kvm))
4306 		kvm_xen_runstate_set_preempted(vcpu);
4307 	else
4308 		kvm_steal_time_set_preempted(vcpu);
4309 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
4310 
4311 	static_call(kvm_x86_vcpu_put)(vcpu);
4312 	vcpu->arch.last_host_tsc = rdtsc();
4313 	/*
4314 	 * If userspace has set any breakpoints or watchpoints, dr6 is restored
4315 	 * on every vmexit, but if not, we might have a stale dr6 from the
4316 	 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
4317 	 */
4318 	set_debugreg(0, 6);
4319 }
4320 
4321 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4322 				    struct kvm_lapic_state *s)
4323 {
4324 	if (vcpu->arch.apicv_active)
4325 		static_call(kvm_x86_sync_pir_to_irr)(vcpu);
4326 
4327 	return kvm_apic_get_state(vcpu, s);
4328 }
4329 
4330 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4331 				    struct kvm_lapic_state *s)
4332 {
4333 	int r;
4334 
4335 	r = kvm_apic_set_state(vcpu, s);
4336 	if (r)
4337 		return r;
4338 	update_cr8_intercept(vcpu);
4339 
4340 	return 0;
4341 }
4342 
4343 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4344 {
4345 	/*
4346 	 * We can accept userspace's request for interrupt injection
4347 	 * as long as we have a place to store the interrupt number.
4348 	 * The actual injection will happen when the CPU is able to
4349 	 * deliver the interrupt.
4350 	 */
4351 	if (kvm_cpu_has_extint(vcpu))
4352 		return false;
4353 
4354 	/* Acknowledging ExtINT does not happen if LINT0 is masked.  */
4355 	return (!lapic_in_kernel(vcpu) ||
4356 		kvm_apic_accept_pic_intr(vcpu));
4357 }
4358 
4359 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4360 {
4361 	return kvm_arch_interrupt_allowed(vcpu) &&
4362 		kvm_cpu_accept_dm_intr(vcpu);
4363 }
4364 
4365 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4366 				    struct kvm_interrupt *irq)
4367 {
4368 	if (irq->irq >= KVM_NR_INTERRUPTS)
4369 		return -EINVAL;
4370 
4371 	if (!irqchip_in_kernel(vcpu->kvm)) {
4372 		kvm_queue_interrupt(vcpu, irq->irq, false);
4373 		kvm_make_request(KVM_REQ_EVENT, vcpu);
4374 		return 0;
4375 	}
4376 
4377 	/*
4378 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4379 	 * fail for in-kernel 8259.
4380 	 */
4381 	if (pic_in_kernel(vcpu->kvm))
4382 		return -ENXIO;
4383 
4384 	if (vcpu->arch.pending_external_vector != -1)
4385 		return -EEXIST;
4386 
4387 	vcpu->arch.pending_external_vector = irq->irq;
4388 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4389 	return 0;
4390 }
4391 
4392 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4393 {
4394 	kvm_inject_nmi(vcpu);
4395 
4396 	return 0;
4397 }
4398 
4399 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4400 {
4401 	kvm_make_request(KVM_REQ_SMI, vcpu);
4402 
4403 	return 0;
4404 }
4405 
4406 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4407 					   struct kvm_tpr_access_ctl *tac)
4408 {
4409 	if (tac->flags)
4410 		return -EINVAL;
4411 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
4412 	return 0;
4413 }
4414 
4415 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4416 					u64 mcg_cap)
4417 {
4418 	int r;
4419 	unsigned bank_num = mcg_cap & 0xff, bank;
4420 
4421 	r = -EINVAL;
4422 	if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4423 		goto out;
4424 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4425 		goto out;
4426 	r = 0;
4427 	vcpu->arch.mcg_cap = mcg_cap;
4428 	/* Init IA32_MCG_CTL to all 1s */
4429 	if (mcg_cap & MCG_CTL_P)
4430 		vcpu->arch.mcg_ctl = ~(u64)0;
4431 	/* Init IA32_MCi_CTL to all 1s */
4432 	for (bank = 0; bank < bank_num; bank++)
4433 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4434 
4435 	static_call(kvm_x86_setup_mce)(vcpu);
4436 out:
4437 	return r;
4438 }
4439 
4440 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4441 				      struct kvm_x86_mce *mce)
4442 {
4443 	u64 mcg_cap = vcpu->arch.mcg_cap;
4444 	unsigned bank_num = mcg_cap & 0xff;
4445 	u64 *banks = vcpu->arch.mce_banks;
4446 
4447 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4448 		return -EINVAL;
4449 	/*
4450 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4451 	 * reporting is disabled
4452 	 */
4453 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4454 	    vcpu->arch.mcg_ctl != ~(u64)0)
4455 		return 0;
4456 	banks += 4 * mce->bank;
4457 	/*
4458 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4459 	 * reporting is disabled for the bank
4460 	 */
4461 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4462 		return 0;
4463 	if (mce->status & MCI_STATUS_UC) {
4464 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4465 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4466 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4467 			return 0;
4468 		}
4469 		if (banks[1] & MCI_STATUS_VAL)
4470 			mce->status |= MCI_STATUS_OVER;
4471 		banks[2] = mce->addr;
4472 		banks[3] = mce->misc;
4473 		vcpu->arch.mcg_status = mce->mcg_status;
4474 		banks[1] = mce->status;
4475 		kvm_queue_exception(vcpu, MC_VECTOR);
4476 	} else if (!(banks[1] & MCI_STATUS_VAL)
4477 		   || !(banks[1] & MCI_STATUS_UC)) {
4478 		if (banks[1] & MCI_STATUS_VAL)
4479 			mce->status |= MCI_STATUS_OVER;
4480 		banks[2] = mce->addr;
4481 		banks[3] = mce->misc;
4482 		banks[1] = mce->status;
4483 	} else
4484 		banks[1] |= MCI_STATUS_OVER;
4485 	return 0;
4486 }
4487 
4488 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4489 					       struct kvm_vcpu_events *events)
4490 {
4491 	process_nmi(vcpu);
4492 
4493 	if (kvm_check_request(KVM_REQ_SMI, vcpu))
4494 		process_smi(vcpu);
4495 
4496 	/*
4497 	 * In guest mode, payload delivery should be deferred,
4498 	 * so that the L1 hypervisor can intercept #PF before
4499 	 * CR2 is modified (or intercept #DB before DR6 is
4500 	 * modified under nVMX). Unless the per-VM capability,
4501 	 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4502 	 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4503 	 * opportunistically defer the exception payload, deliver it if the
4504 	 * capability hasn't been requested before processing a
4505 	 * KVM_GET_VCPU_EVENTS.
4506 	 */
4507 	if (!vcpu->kvm->arch.exception_payload_enabled &&
4508 	    vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4509 		kvm_deliver_exception_payload(vcpu);
4510 
4511 	/*
4512 	 * The API doesn't provide the instruction length for software
4513 	 * exceptions, so don't report them. As long as the guest RIP
4514 	 * isn't advanced, we should expect to encounter the exception
4515 	 * again.
4516 	 */
4517 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4518 		events->exception.injected = 0;
4519 		events->exception.pending = 0;
4520 	} else {
4521 		events->exception.injected = vcpu->arch.exception.injected;
4522 		events->exception.pending = vcpu->arch.exception.pending;
4523 		/*
4524 		 * For ABI compatibility, deliberately conflate
4525 		 * pending and injected exceptions when
4526 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4527 		 */
4528 		if (!vcpu->kvm->arch.exception_payload_enabled)
4529 			events->exception.injected |=
4530 				vcpu->arch.exception.pending;
4531 	}
4532 	events->exception.nr = vcpu->arch.exception.nr;
4533 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4534 	events->exception.error_code = vcpu->arch.exception.error_code;
4535 	events->exception_has_payload = vcpu->arch.exception.has_payload;
4536 	events->exception_payload = vcpu->arch.exception.payload;
4537 
4538 	events->interrupt.injected =
4539 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4540 	events->interrupt.nr = vcpu->arch.interrupt.nr;
4541 	events->interrupt.soft = 0;
4542 	events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4543 
4544 	events->nmi.injected = vcpu->arch.nmi_injected;
4545 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
4546 	events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4547 	events->nmi.pad = 0;
4548 
4549 	events->sipi_vector = 0; /* never valid when reporting to user space */
4550 
4551 	events->smi.smm = is_smm(vcpu);
4552 	events->smi.pending = vcpu->arch.smi_pending;
4553 	events->smi.smm_inside_nmi =
4554 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4555 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4556 
4557 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4558 			 | KVM_VCPUEVENT_VALID_SHADOW
4559 			 | KVM_VCPUEVENT_VALID_SMM);
4560 	if (vcpu->kvm->arch.exception_payload_enabled)
4561 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4562 
4563 	memset(&events->reserved, 0, sizeof(events->reserved));
4564 }
4565 
4566 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm);
4567 
4568 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4569 					      struct kvm_vcpu_events *events)
4570 {
4571 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4572 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4573 			      | KVM_VCPUEVENT_VALID_SHADOW
4574 			      | KVM_VCPUEVENT_VALID_SMM
4575 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
4576 		return -EINVAL;
4577 
4578 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4579 		if (!vcpu->kvm->arch.exception_payload_enabled)
4580 			return -EINVAL;
4581 		if (events->exception.pending)
4582 			events->exception.injected = 0;
4583 		else
4584 			events->exception_has_payload = 0;
4585 	} else {
4586 		events->exception.pending = 0;
4587 		events->exception_has_payload = 0;
4588 	}
4589 
4590 	if ((events->exception.injected || events->exception.pending) &&
4591 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4592 		return -EINVAL;
4593 
4594 	/* INITs are latched while in SMM */
4595 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4596 	    (events->smi.smm || events->smi.pending) &&
4597 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4598 		return -EINVAL;
4599 
4600 	process_nmi(vcpu);
4601 	vcpu->arch.exception.injected = events->exception.injected;
4602 	vcpu->arch.exception.pending = events->exception.pending;
4603 	vcpu->arch.exception.nr = events->exception.nr;
4604 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4605 	vcpu->arch.exception.error_code = events->exception.error_code;
4606 	vcpu->arch.exception.has_payload = events->exception_has_payload;
4607 	vcpu->arch.exception.payload = events->exception_payload;
4608 
4609 	vcpu->arch.interrupt.injected = events->interrupt.injected;
4610 	vcpu->arch.interrupt.nr = events->interrupt.nr;
4611 	vcpu->arch.interrupt.soft = events->interrupt.soft;
4612 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4613 		static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4614 						events->interrupt.shadow);
4615 
4616 	vcpu->arch.nmi_injected = events->nmi.injected;
4617 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4618 		vcpu->arch.nmi_pending = events->nmi.pending;
4619 	static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4620 
4621 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4622 	    lapic_in_kernel(vcpu))
4623 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
4624 
4625 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4626 		if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm)
4627 			kvm_smm_changed(vcpu, events->smi.smm);
4628 
4629 		vcpu->arch.smi_pending = events->smi.pending;
4630 
4631 		if (events->smi.smm) {
4632 			if (events->smi.smm_inside_nmi)
4633 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4634 			else
4635 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4636 		}
4637 
4638 		if (lapic_in_kernel(vcpu)) {
4639 			if (events->smi.latched_init)
4640 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4641 			else
4642 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4643 		}
4644 	}
4645 
4646 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4647 
4648 	return 0;
4649 }
4650 
4651 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4652 					     struct kvm_debugregs *dbgregs)
4653 {
4654 	unsigned long val;
4655 
4656 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4657 	kvm_get_dr(vcpu, 6, &val);
4658 	dbgregs->dr6 = val;
4659 	dbgregs->dr7 = vcpu->arch.dr7;
4660 	dbgregs->flags = 0;
4661 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4662 }
4663 
4664 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4665 					    struct kvm_debugregs *dbgregs)
4666 {
4667 	if (dbgregs->flags)
4668 		return -EINVAL;
4669 
4670 	if (!kvm_dr6_valid(dbgregs->dr6))
4671 		return -EINVAL;
4672 	if (!kvm_dr7_valid(dbgregs->dr7))
4673 		return -EINVAL;
4674 
4675 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4676 	kvm_update_dr0123(vcpu);
4677 	vcpu->arch.dr6 = dbgregs->dr6;
4678 	vcpu->arch.dr7 = dbgregs->dr7;
4679 	kvm_update_dr7(vcpu);
4680 
4681 	return 0;
4682 }
4683 
4684 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4685 
4686 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4687 {
4688 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4689 	u64 xstate_bv = xsave->header.xfeatures;
4690 	u64 valid;
4691 
4692 	/*
4693 	 * Copy legacy XSAVE area, to avoid complications with CPUID
4694 	 * leaves 0 and 1 in the loop below.
4695 	 */
4696 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4697 
4698 	/* Set XSTATE_BV */
4699 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4700 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4701 
4702 	/*
4703 	 * Copy each region from the possibly compacted offset to the
4704 	 * non-compacted offset.
4705 	 */
4706 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4707 	while (valid) {
4708 		u32 size, offset, ecx, edx;
4709 		u64 xfeature_mask = valid & -valid;
4710 		int xfeature_nr = fls64(xfeature_mask) - 1;
4711 		void *src;
4712 
4713 		cpuid_count(XSTATE_CPUID, xfeature_nr,
4714 			    &size, &offset, &ecx, &edx);
4715 
4716 		if (xfeature_nr == XFEATURE_PKRU) {
4717 			memcpy(dest + offset, &vcpu->arch.pkru,
4718 			       sizeof(vcpu->arch.pkru));
4719 		} else {
4720 			src = get_xsave_addr(xsave, xfeature_nr);
4721 			if (src)
4722 				memcpy(dest + offset, src, size);
4723 		}
4724 
4725 		valid -= xfeature_mask;
4726 	}
4727 }
4728 
4729 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4730 {
4731 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4732 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4733 	u64 valid;
4734 
4735 	/*
4736 	 * Copy legacy XSAVE area, to avoid complications with CPUID
4737 	 * leaves 0 and 1 in the loop below.
4738 	 */
4739 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
4740 
4741 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
4742 	xsave->header.xfeatures = xstate_bv;
4743 	if (boot_cpu_has(X86_FEATURE_XSAVES))
4744 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4745 
4746 	/*
4747 	 * Copy each region from the non-compacted offset to the
4748 	 * possibly compacted offset.
4749 	 */
4750 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4751 	while (valid) {
4752 		u32 size, offset, ecx, edx;
4753 		u64 xfeature_mask = valid & -valid;
4754 		int xfeature_nr = fls64(xfeature_mask) - 1;
4755 
4756 		cpuid_count(XSTATE_CPUID, xfeature_nr,
4757 			    &size, &offset, &ecx, &edx);
4758 
4759 		if (xfeature_nr == XFEATURE_PKRU) {
4760 			memcpy(&vcpu->arch.pkru, src + offset,
4761 			       sizeof(vcpu->arch.pkru));
4762 		} else {
4763 			void *dest = get_xsave_addr(xsave, xfeature_nr);
4764 
4765 			if (dest)
4766 				memcpy(dest, src + offset, size);
4767 		}
4768 
4769 		valid -= xfeature_mask;
4770 	}
4771 }
4772 
4773 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4774 					 struct kvm_xsave *guest_xsave)
4775 {
4776 	if (!vcpu->arch.guest_fpu)
4777 		return;
4778 
4779 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4780 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4781 		fill_xsave((u8 *) guest_xsave->region, vcpu);
4782 	} else {
4783 		memcpy(guest_xsave->region,
4784 			&vcpu->arch.guest_fpu->state.fxsave,
4785 			sizeof(struct fxregs_state));
4786 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4787 			XFEATURE_MASK_FPSSE;
4788 	}
4789 }
4790 
4791 #define XSAVE_MXCSR_OFFSET 24
4792 
4793 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4794 					struct kvm_xsave *guest_xsave)
4795 {
4796 	u64 xstate_bv;
4797 	u32 mxcsr;
4798 
4799 	if (!vcpu->arch.guest_fpu)
4800 		return 0;
4801 
4802 	xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4803 	mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4804 
4805 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4806 		/*
4807 		 * Here we allow setting states that are not present in
4808 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
4809 		 * with old userspace.
4810 		 */
4811 		if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4812 			return -EINVAL;
4813 		load_xsave(vcpu, (u8 *)guest_xsave->region);
4814 	} else {
4815 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4816 			mxcsr & ~mxcsr_feature_mask)
4817 			return -EINVAL;
4818 		memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4819 			guest_xsave->region, sizeof(struct fxregs_state));
4820 	}
4821 	return 0;
4822 }
4823 
4824 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4825 					struct kvm_xcrs *guest_xcrs)
4826 {
4827 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4828 		guest_xcrs->nr_xcrs = 0;
4829 		return;
4830 	}
4831 
4832 	guest_xcrs->nr_xcrs = 1;
4833 	guest_xcrs->flags = 0;
4834 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4835 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4836 }
4837 
4838 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4839 				       struct kvm_xcrs *guest_xcrs)
4840 {
4841 	int i, r = 0;
4842 
4843 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
4844 		return -EINVAL;
4845 
4846 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4847 		return -EINVAL;
4848 
4849 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4850 		/* Only support XCR0 currently */
4851 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4852 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4853 				guest_xcrs->xcrs[i].value);
4854 			break;
4855 		}
4856 	if (r)
4857 		r = -EINVAL;
4858 	return r;
4859 }
4860 
4861 /*
4862  * kvm_set_guest_paused() indicates to the guest kernel that it has been
4863  * stopped by the hypervisor.  This function will be called from the host only.
4864  * EINVAL is returned when the host attempts to set the flag for a guest that
4865  * does not support pv clocks.
4866  */
4867 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4868 {
4869 	if (!vcpu->arch.pv_time_enabled)
4870 		return -EINVAL;
4871 	vcpu->arch.pvclock_set_guest_stopped_request = true;
4872 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4873 	return 0;
4874 }
4875 
4876 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4877 				     struct kvm_enable_cap *cap)
4878 {
4879 	int r;
4880 	uint16_t vmcs_version;
4881 	void __user *user_ptr;
4882 
4883 	if (cap->flags)
4884 		return -EINVAL;
4885 
4886 	switch (cap->cap) {
4887 	case KVM_CAP_HYPERV_SYNIC2:
4888 		if (cap->args[0])
4889 			return -EINVAL;
4890 		fallthrough;
4891 
4892 	case KVM_CAP_HYPERV_SYNIC:
4893 		if (!irqchip_in_kernel(vcpu->kvm))
4894 			return -EINVAL;
4895 		return kvm_hv_activate_synic(vcpu, cap->cap ==
4896 					     KVM_CAP_HYPERV_SYNIC2);
4897 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4898 		if (!kvm_x86_ops.nested_ops->enable_evmcs)
4899 			return -ENOTTY;
4900 		r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4901 		if (!r) {
4902 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
4903 			if (copy_to_user(user_ptr, &vmcs_version,
4904 					 sizeof(vmcs_version)))
4905 				r = -EFAULT;
4906 		}
4907 		return r;
4908 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4909 		if (!kvm_x86_ops.enable_direct_tlbflush)
4910 			return -ENOTTY;
4911 
4912 		return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
4913 
4914 	case KVM_CAP_HYPERV_ENFORCE_CPUID:
4915 		return kvm_hv_set_enforce_cpuid(vcpu, cap->args[0]);
4916 
4917 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4918 		vcpu->arch.pv_cpuid.enforce = cap->args[0];
4919 		if (vcpu->arch.pv_cpuid.enforce)
4920 			kvm_update_pv_runtime(vcpu);
4921 
4922 		return 0;
4923 	default:
4924 		return -EINVAL;
4925 	}
4926 }
4927 
4928 long kvm_arch_vcpu_ioctl(struct file *filp,
4929 			 unsigned int ioctl, unsigned long arg)
4930 {
4931 	struct kvm_vcpu *vcpu = filp->private_data;
4932 	void __user *argp = (void __user *)arg;
4933 	int r;
4934 	union {
4935 		struct kvm_sregs2 *sregs2;
4936 		struct kvm_lapic_state *lapic;
4937 		struct kvm_xsave *xsave;
4938 		struct kvm_xcrs *xcrs;
4939 		void *buffer;
4940 	} u;
4941 
4942 	vcpu_load(vcpu);
4943 
4944 	u.buffer = NULL;
4945 	switch (ioctl) {
4946 	case KVM_GET_LAPIC: {
4947 		r = -EINVAL;
4948 		if (!lapic_in_kernel(vcpu))
4949 			goto out;
4950 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4951 				GFP_KERNEL_ACCOUNT);
4952 
4953 		r = -ENOMEM;
4954 		if (!u.lapic)
4955 			goto out;
4956 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4957 		if (r)
4958 			goto out;
4959 		r = -EFAULT;
4960 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4961 			goto out;
4962 		r = 0;
4963 		break;
4964 	}
4965 	case KVM_SET_LAPIC: {
4966 		r = -EINVAL;
4967 		if (!lapic_in_kernel(vcpu))
4968 			goto out;
4969 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
4970 		if (IS_ERR(u.lapic)) {
4971 			r = PTR_ERR(u.lapic);
4972 			goto out_nofree;
4973 		}
4974 
4975 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4976 		break;
4977 	}
4978 	case KVM_INTERRUPT: {
4979 		struct kvm_interrupt irq;
4980 
4981 		r = -EFAULT;
4982 		if (copy_from_user(&irq, argp, sizeof(irq)))
4983 			goto out;
4984 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4985 		break;
4986 	}
4987 	case KVM_NMI: {
4988 		r = kvm_vcpu_ioctl_nmi(vcpu);
4989 		break;
4990 	}
4991 	case KVM_SMI: {
4992 		r = kvm_vcpu_ioctl_smi(vcpu);
4993 		break;
4994 	}
4995 	case KVM_SET_CPUID: {
4996 		struct kvm_cpuid __user *cpuid_arg = argp;
4997 		struct kvm_cpuid cpuid;
4998 
4999 		r = -EFAULT;
5000 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5001 			goto out;
5002 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
5003 		break;
5004 	}
5005 	case KVM_SET_CPUID2: {
5006 		struct kvm_cpuid2 __user *cpuid_arg = argp;
5007 		struct kvm_cpuid2 cpuid;
5008 
5009 		r = -EFAULT;
5010 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5011 			goto out;
5012 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
5013 					      cpuid_arg->entries);
5014 		break;
5015 	}
5016 	case KVM_GET_CPUID2: {
5017 		struct kvm_cpuid2 __user *cpuid_arg = argp;
5018 		struct kvm_cpuid2 cpuid;
5019 
5020 		r = -EFAULT;
5021 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5022 			goto out;
5023 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
5024 					      cpuid_arg->entries);
5025 		if (r)
5026 			goto out;
5027 		r = -EFAULT;
5028 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
5029 			goto out;
5030 		r = 0;
5031 		break;
5032 	}
5033 	case KVM_GET_MSRS: {
5034 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
5035 		r = msr_io(vcpu, argp, do_get_msr, 1);
5036 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5037 		break;
5038 	}
5039 	case KVM_SET_MSRS: {
5040 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
5041 		r = msr_io(vcpu, argp, do_set_msr, 0);
5042 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5043 		break;
5044 	}
5045 	case KVM_TPR_ACCESS_REPORTING: {
5046 		struct kvm_tpr_access_ctl tac;
5047 
5048 		r = -EFAULT;
5049 		if (copy_from_user(&tac, argp, sizeof(tac)))
5050 			goto out;
5051 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
5052 		if (r)
5053 			goto out;
5054 		r = -EFAULT;
5055 		if (copy_to_user(argp, &tac, sizeof(tac)))
5056 			goto out;
5057 		r = 0;
5058 		break;
5059 	};
5060 	case KVM_SET_VAPIC_ADDR: {
5061 		struct kvm_vapic_addr va;
5062 		int idx;
5063 
5064 		r = -EINVAL;
5065 		if (!lapic_in_kernel(vcpu))
5066 			goto out;
5067 		r = -EFAULT;
5068 		if (copy_from_user(&va, argp, sizeof(va)))
5069 			goto out;
5070 		idx = srcu_read_lock(&vcpu->kvm->srcu);
5071 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
5072 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5073 		break;
5074 	}
5075 	case KVM_X86_SETUP_MCE: {
5076 		u64 mcg_cap;
5077 
5078 		r = -EFAULT;
5079 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
5080 			goto out;
5081 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
5082 		break;
5083 	}
5084 	case KVM_X86_SET_MCE: {
5085 		struct kvm_x86_mce mce;
5086 
5087 		r = -EFAULT;
5088 		if (copy_from_user(&mce, argp, sizeof(mce)))
5089 			goto out;
5090 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
5091 		break;
5092 	}
5093 	case KVM_GET_VCPU_EVENTS: {
5094 		struct kvm_vcpu_events events;
5095 
5096 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
5097 
5098 		r = -EFAULT;
5099 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
5100 			break;
5101 		r = 0;
5102 		break;
5103 	}
5104 	case KVM_SET_VCPU_EVENTS: {
5105 		struct kvm_vcpu_events events;
5106 
5107 		r = -EFAULT;
5108 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5109 			break;
5110 
5111 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5112 		break;
5113 	}
5114 	case KVM_GET_DEBUGREGS: {
5115 		struct kvm_debugregs dbgregs;
5116 
5117 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5118 
5119 		r = -EFAULT;
5120 		if (copy_to_user(argp, &dbgregs,
5121 				 sizeof(struct kvm_debugregs)))
5122 			break;
5123 		r = 0;
5124 		break;
5125 	}
5126 	case KVM_SET_DEBUGREGS: {
5127 		struct kvm_debugregs dbgregs;
5128 
5129 		r = -EFAULT;
5130 		if (copy_from_user(&dbgregs, argp,
5131 				   sizeof(struct kvm_debugregs)))
5132 			break;
5133 
5134 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5135 		break;
5136 	}
5137 	case KVM_GET_XSAVE: {
5138 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5139 		r = -ENOMEM;
5140 		if (!u.xsave)
5141 			break;
5142 
5143 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5144 
5145 		r = -EFAULT;
5146 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5147 			break;
5148 		r = 0;
5149 		break;
5150 	}
5151 	case KVM_SET_XSAVE: {
5152 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
5153 		if (IS_ERR(u.xsave)) {
5154 			r = PTR_ERR(u.xsave);
5155 			goto out_nofree;
5156 		}
5157 
5158 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5159 		break;
5160 	}
5161 	case KVM_GET_XCRS: {
5162 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5163 		r = -ENOMEM;
5164 		if (!u.xcrs)
5165 			break;
5166 
5167 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5168 
5169 		r = -EFAULT;
5170 		if (copy_to_user(argp, u.xcrs,
5171 				 sizeof(struct kvm_xcrs)))
5172 			break;
5173 		r = 0;
5174 		break;
5175 	}
5176 	case KVM_SET_XCRS: {
5177 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5178 		if (IS_ERR(u.xcrs)) {
5179 			r = PTR_ERR(u.xcrs);
5180 			goto out_nofree;
5181 		}
5182 
5183 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5184 		break;
5185 	}
5186 	case KVM_SET_TSC_KHZ: {
5187 		u32 user_tsc_khz;
5188 
5189 		r = -EINVAL;
5190 		user_tsc_khz = (u32)arg;
5191 
5192 		if (kvm_has_tsc_control &&
5193 		    user_tsc_khz >= kvm_max_guest_tsc_khz)
5194 			goto out;
5195 
5196 		if (user_tsc_khz == 0)
5197 			user_tsc_khz = tsc_khz;
5198 
5199 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5200 			r = 0;
5201 
5202 		goto out;
5203 	}
5204 	case KVM_GET_TSC_KHZ: {
5205 		r = vcpu->arch.virtual_tsc_khz;
5206 		goto out;
5207 	}
5208 	case KVM_KVMCLOCK_CTRL: {
5209 		r = kvm_set_guest_paused(vcpu);
5210 		goto out;
5211 	}
5212 	case KVM_ENABLE_CAP: {
5213 		struct kvm_enable_cap cap;
5214 
5215 		r = -EFAULT;
5216 		if (copy_from_user(&cap, argp, sizeof(cap)))
5217 			goto out;
5218 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5219 		break;
5220 	}
5221 	case KVM_GET_NESTED_STATE: {
5222 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5223 		u32 user_data_size;
5224 
5225 		r = -EINVAL;
5226 		if (!kvm_x86_ops.nested_ops->get_state)
5227 			break;
5228 
5229 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5230 		r = -EFAULT;
5231 		if (get_user(user_data_size, &user_kvm_nested_state->size))
5232 			break;
5233 
5234 		r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5235 						     user_data_size);
5236 		if (r < 0)
5237 			break;
5238 
5239 		if (r > user_data_size) {
5240 			if (put_user(r, &user_kvm_nested_state->size))
5241 				r = -EFAULT;
5242 			else
5243 				r = -E2BIG;
5244 			break;
5245 		}
5246 
5247 		r = 0;
5248 		break;
5249 	}
5250 	case KVM_SET_NESTED_STATE: {
5251 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5252 		struct kvm_nested_state kvm_state;
5253 		int idx;
5254 
5255 		r = -EINVAL;
5256 		if (!kvm_x86_ops.nested_ops->set_state)
5257 			break;
5258 
5259 		r = -EFAULT;
5260 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5261 			break;
5262 
5263 		r = -EINVAL;
5264 		if (kvm_state.size < sizeof(kvm_state))
5265 			break;
5266 
5267 		if (kvm_state.flags &
5268 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5269 		      | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5270 		      | KVM_STATE_NESTED_GIF_SET))
5271 			break;
5272 
5273 		/* nested_run_pending implies guest_mode.  */
5274 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5275 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5276 			break;
5277 
5278 		idx = srcu_read_lock(&vcpu->kvm->srcu);
5279 		r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5280 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5281 		break;
5282 	}
5283 	case KVM_GET_SUPPORTED_HV_CPUID:
5284 		r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5285 		break;
5286 #ifdef CONFIG_KVM_XEN
5287 	case KVM_XEN_VCPU_GET_ATTR: {
5288 		struct kvm_xen_vcpu_attr xva;
5289 
5290 		r = -EFAULT;
5291 		if (copy_from_user(&xva, argp, sizeof(xva)))
5292 			goto out;
5293 		r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5294 		if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5295 			r = -EFAULT;
5296 		break;
5297 	}
5298 	case KVM_XEN_VCPU_SET_ATTR: {
5299 		struct kvm_xen_vcpu_attr xva;
5300 
5301 		r = -EFAULT;
5302 		if (copy_from_user(&xva, argp, sizeof(xva)))
5303 			goto out;
5304 		r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5305 		break;
5306 	}
5307 #endif
5308 	case KVM_GET_SREGS2: {
5309 		u.sregs2 = kzalloc(sizeof(struct kvm_sregs2), GFP_KERNEL);
5310 		r = -ENOMEM;
5311 		if (!u.sregs2)
5312 			goto out;
5313 		__get_sregs2(vcpu, u.sregs2);
5314 		r = -EFAULT;
5315 		if (copy_to_user(argp, u.sregs2, sizeof(struct kvm_sregs2)))
5316 			goto out;
5317 		r = 0;
5318 		break;
5319 	}
5320 	case KVM_SET_SREGS2: {
5321 		u.sregs2 = memdup_user(argp, sizeof(struct kvm_sregs2));
5322 		if (IS_ERR(u.sregs2)) {
5323 			r = PTR_ERR(u.sregs2);
5324 			u.sregs2 = NULL;
5325 			goto out;
5326 		}
5327 		r = __set_sregs2(vcpu, u.sregs2);
5328 		break;
5329 	}
5330 	default:
5331 		r = -EINVAL;
5332 	}
5333 out:
5334 	kfree(u.buffer);
5335 out_nofree:
5336 	vcpu_put(vcpu);
5337 	return r;
5338 }
5339 
5340 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5341 {
5342 	return VM_FAULT_SIGBUS;
5343 }
5344 
5345 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5346 {
5347 	int ret;
5348 
5349 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
5350 		return -EINVAL;
5351 	ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5352 	return ret;
5353 }
5354 
5355 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5356 					      u64 ident_addr)
5357 {
5358 	return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5359 }
5360 
5361 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5362 					 unsigned long kvm_nr_mmu_pages)
5363 {
5364 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5365 		return -EINVAL;
5366 
5367 	mutex_lock(&kvm->slots_lock);
5368 
5369 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5370 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5371 
5372 	mutex_unlock(&kvm->slots_lock);
5373 	return 0;
5374 }
5375 
5376 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5377 {
5378 	return kvm->arch.n_max_mmu_pages;
5379 }
5380 
5381 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5382 {
5383 	struct kvm_pic *pic = kvm->arch.vpic;
5384 	int r;
5385 
5386 	r = 0;
5387 	switch (chip->chip_id) {
5388 	case KVM_IRQCHIP_PIC_MASTER:
5389 		memcpy(&chip->chip.pic, &pic->pics[0],
5390 			sizeof(struct kvm_pic_state));
5391 		break;
5392 	case KVM_IRQCHIP_PIC_SLAVE:
5393 		memcpy(&chip->chip.pic, &pic->pics[1],
5394 			sizeof(struct kvm_pic_state));
5395 		break;
5396 	case KVM_IRQCHIP_IOAPIC:
5397 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
5398 		break;
5399 	default:
5400 		r = -EINVAL;
5401 		break;
5402 	}
5403 	return r;
5404 }
5405 
5406 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5407 {
5408 	struct kvm_pic *pic = kvm->arch.vpic;
5409 	int r;
5410 
5411 	r = 0;
5412 	switch (chip->chip_id) {
5413 	case KVM_IRQCHIP_PIC_MASTER:
5414 		spin_lock(&pic->lock);
5415 		memcpy(&pic->pics[0], &chip->chip.pic,
5416 			sizeof(struct kvm_pic_state));
5417 		spin_unlock(&pic->lock);
5418 		break;
5419 	case KVM_IRQCHIP_PIC_SLAVE:
5420 		spin_lock(&pic->lock);
5421 		memcpy(&pic->pics[1], &chip->chip.pic,
5422 			sizeof(struct kvm_pic_state));
5423 		spin_unlock(&pic->lock);
5424 		break;
5425 	case KVM_IRQCHIP_IOAPIC:
5426 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
5427 		break;
5428 	default:
5429 		r = -EINVAL;
5430 		break;
5431 	}
5432 	kvm_pic_update_irq(pic);
5433 	return r;
5434 }
5435 
5436 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5437 {
5438 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5439 
5440 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5441 
5442 	mutex_lock(&kps->lock);
5443 	memcpy(ps, &kps->channels, sizeof(*ps));
5444 	mutex_unlock(&kps->lock);
5445 	return 0;
5446 }
5447 
5448 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5449 {
5450 	int i;
5451 	struct kvm_pit *pit = kvm->arch.vpit;
5452 
5453 	mutex_lock(&pit->pit_state.lock);
5454 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5455 	for (i = 0; i < 3; i++)
5456 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5457 	mutex_unlock(&pit->pit_state.lock);
5458 	return 0;
5459 }
5460 
5461 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5462 {
5463 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
5464 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5465 		sizeof(ps->channels));
5466 	ps->flags = kvm->arch.vpit->pit_state.flags;
5467 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5468 	memset(&ps->reserved, 0, sizeof(ps->reserved));
5469 	return 0;
5470 }
5471 
5472 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5473 {
5474 	int start = 0;
5475 	int i;
5476 	u32 prev_legacy, cur_legacy;
5477 	struct kvm_pit *pit = kvm->arch.vpit;
5478 
5479 	mutex_lock(&pit->pit_state.lock);
5480 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5481 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5482 	if (!prev_legacy && cur_legacy)
5483 		start = 1;
5484 	memcpy(&pit->pit_state.channels, &ps->channels,
5485 	       sizeof(pit->pit_state.channels));
5486 	pit->pit_state.flags = ps->flags;
5487 	for (i = 0; i < 3; i++)
5488 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5489 				   start && i == 0);
5490 	mutex_unlock(&pit->pit_state.lock);
5491 	return 0;
5492 }
5493 
5494 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5495 				 struct kvm_reinject_control *control)
5496 {
5497 	struct kvm_pit *pit = kvm->arch.vpit;
5498 
5499 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
5500 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5501 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
5502 	 */
5503 	mutex_lock(&pit->pit_state.lock);
5504 	kvm_pit_set_reinject(pit, control->pit_reinject);
5505 	mutex_unlock(&pit->pit_state.lock);
5506 
5507 	return 0;
5508 }
5509 
5510 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5511 {
5512 
5513 	/*
5514 	 * Flush all CPUs' dirty log buffers to the  dirty_bitmap.  Called
5515 	 * before reporting dirty_bitmap to userspace.  KVM flushes the buffers
5516 	 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5517 	 * VM-Exit.
5518 	 */
5519 	struct kvm_vcpu *vcpu;
5520 	int i;
5521 
5522 	kvm_for_each_vcpu(i, vcpu, kvm)
5523 		kvm_vcpu_kick(vcpu);
5524 }
5525 
5526 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5527 			bool line_status)
5528 {
5529 	if (!irqchip_in_kernel(kvm))
5530 		return -ENXIO;
5531 
5532 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5533 					irq_event->irq, irq_event->level,
5534 					line_status);
5535 	return 0;
5536 }
5537 
5538 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5539 			    struct kvm_enable_cap *cap)
5540 {
5541 	int r;
5542 
5543 	if (cap->flags)
5544 		return -EINVAL;
5545 
5546 	switch (cap->cap) {
5547 	case KVM_CAP_DISABLE_QUIRKS:
5548 		kvm->arch.disabled_quirks = cap->args[0];
5549 		r = 0;
5550 		break;
5551 	case KVM_CAP_SPLIT_IRQCHIP: {
5552 		mutex_lock(&kvm->lock);
5553 		r = -EINVAL;
5554 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5555 			goto split_irqchip_unlock;
5556 		r = -EEXIST;
5557 		if (irqchip_in_kernel(kvm))
5558 			goto split_irqchip_unlock;
5559 		if (kvm->created_vcpus)
5560 			goto split_irqchip_unlock;
5561 		r = kvm_setup_empty_irq_routing(kvm);
5562 		if (r)
5563 			goto split_irqchip_unlock;
5564 		/* Pairs with irqchip_in_kernel. */
5565 		smp_wmb();
5566 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5567 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5568 		r = 0;
5569 split_irqchip_unlock:
5570 		mutex_unlock(&kvm->lock);
5571 		break;
5572 	}
5573 	case KVM_CAP_X2APIC_API:
5574 		r = -EINVAL;
5575 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5576 			break;
5577 
5578 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5579 			kvm->arch.x2apic_format = true;
5580 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5581 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
5582 
5583 		r = 0;
5584 		break;
5585 	case KVM_CAP_X86_DISABLE_EXITS:
5586 		r = -EINVAL;
5587 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5588 			break;
5589 
5590 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5591 			kvm_can_mwait_in_guest())
5592 			kvm->arch.mwait_in_guest = true;
5593 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5594 			kvm->arch.hlt_in_guest = true;
5595 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5596 			kvm->arch.pause_in_guest = true;
5597 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5598 			kvm->arch.cstate_in_guest = true;
5599 		r = 0;
5600 		break;
5601 	case KVM_CAP_MSR_PLATFORM_INFO:
5602 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5603 		r = 0;
5604 		break;
5605 	case KVM_CAP_EXCEPTION_PAYLOAD:
5606 		kvm->arch.exception_payload_enabled = cap->args[0];
5607 		r = 0;
5608 		break;
5609 	case KVM_CAP_X86_USER_SPACE_MSR:
5610 		kvm->arch.user_space_msr_mask = cap->args[0];
5611 		r = 0;
5612 		break;
5613 	case KVM_CAP_X86_BUS_LOCK_EXIT:
5614 		r = -EINVAL;
5615 		if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5616 			break;
5617 
5618 		if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5619 		    (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5620 			break;
5621 
5622 		if (kvm_has_bus_lock_exit &&
5623 		    cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5624 			kvm->arch.bus_lock_detection_enabled = true;
5625 		r = 0;
5626 		break;
5627 #ifdef CONFIG_X86_SGX_KVM
5628 	case KVM_CAP_SGX_ATTRIBUTE: {
5629 		unsigned long allowed_attributes = 0;
5630 
5631 		r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5632 		if (r)
5633 			break;
5634 
5635 		/* KVM only supports the PROVISIONKEY privileged attribute. */
5636 		if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5637 		    !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5638 			kvm->arch.sgx_provisioning_allowed = true;
5639 		else
5640 			r = -EINVAL;
5641 		break;
5642 	}
5643 #endif
5644 	case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5645 		r = -EINVAL;
5646 		if (kvm_x86_ops.vm_copy_enc_context_from)
5647 			r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5648 		return r;
5649 	case KVM_CAP_EXIT_HYPERCALL:
5650 		if (cap->args[0] & ~KVM_EXIT_HYPERCALL_VALID_MASK) {
5651 			r = -EINVAL;
5652 			break;
5653 		}
5654 		kvm->arch.hypercall_exit_enabled = cap->args[0];
5655 		r = 0;
5656 		break;
5657 	case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
5658 		r = -EINVAL;
5659 		if (cap->args[0] & ~1)
5660 			break;
5661 		kvm->arch.exit_on_emulation_error = cap->args[0];
5662 		r = 0;
5663 		break;
5664 	default:
5665 		r = -EINVAL;
5666 		break;
5667 	}
5668 	return r;
5669 }
5670 
5671 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5672 {
5673 	struct kvm_x86_msr_filter *msr_filter;
5674 
5675 	msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5676 	if (!msr_filter)
5677 		return NULL;
5678 
5679 	msr_filter->default_allow = default_allow;
5680 	return msr_filter;
5681 }
5682 
5683 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5684 {
5685 	u32 i;
5686 
5687 	if (!msr_filter)
5688 		return;
5689 
5690 	for (i = 0; i < msr_filter->count; i++)
5691 		kfree(msr_filter->ranges[i].bitmap);
5692 
5693 	kfree(msr_filter);
5694 }
5695 
5696 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5697 			      struct kvm_msr_filter_range *user_range)
5698 {
5699 	unsigned long *bitmap = NULL;
5700 	size_t bitmap_size;
5701 
5702 	if (!user_range->nmsrs)
5703 		return 0;
5704 
5705 	if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
5706 		return -EINVAL;
5707 
5708 	if (!user_range->flags)
5709 		return -EINVAL;
5710 
5711 	bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5712 	if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5713 		return -EINVAL;
5714 
5715 	bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5716 	if (IS_ERR(bitmap))
5717 		return PTR_ERR(bitmap);
5718 
5719 	msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
5720 		.flags = user_range->flags,
5721 		.base = user_range->base,
5722 		.nmsrs = user_range->nmsrs,
5723 		.bitmap = bitmap,
5724 	};
5725 
5726 	msr_filter->count++;
5727 	return 0;
5728 }
5729 
5730 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5731 {
5732 	struct kvm_msr_filter __user *user_msr_filter = argp;
5733 	struct kvm_x86_msr_filter *new_filter, *old_filter;
5734 	struct kvm_msr_filter filter;
5735 	bool default_allow;
5736 	bool empty = true;
5737 	int r = 0;
5738 	u32 i;
5739 
5740 	if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5741 		return -EFAULT;
5742 
5743 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5744 		empty &= !filter.ranges[i].nmsrs;
5745 
5746 	default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5747 	if (empty && !default_allow)
5748 		return -EINVAL;
5749 
5750 	new_filter = kvm_alloc_msr_filter(default_allow);
5751 	if (!new_filter)
5752 		return -ENOMEM;
5753 
5754 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5755 		r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
5756 		if (r) {
5757 			kvm_free_msr_filter(new_filter);
5758 			return r;
5759 		}
5760 	}
5761 
5762 	mutex_lock(&kvm->lock);
5763 
5764 	/* The per-VM filter is protected by kvm->lock... */
5765 	old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
5766 
5767 	rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
5768 	synchronize_srcu(&kvm->srcu);
5769 
5770 	kvm_free_msr_filter(old_filter);
5771 
5772 	kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5773 	mutex_unlock(&kvm->lock);
5774 
5775 	return 0;
5776 }
5777 
5778 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
5779 static int kvm_arch_suspend_notifier(struct kvm *kvm)
5780 {
5781 	struct kvm_vcpu *vcpu;
5782 	int i, ret = 0;
5783 
5784 	mutex_lock(&kvm->lock);
5785 	kvm_for_each_vcpu(i, vcpu, kvm) {
5786 		if (!vcpu->arch.pv_time_enabled)
5787 			continue;
5788 
5789 		ret = kvm_set_guest_paused(vcpu);
5790 		if (ret) {
5791 			kvm_err("Failed to pause guest VCPU%d: %d\n",
5792 				vcpu->vcpu_id, ret);
5793 			break;
5794 		}
5795 	}
5796 	mutex_unlock(&kvm->lock);
5797 
5798 	return ret ? NOTIFY_BAD : NOTIFY_DONE;
5799 }
5800 
5801 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state)
5802 {
5803 	switch (state) {
5804 	case PM_HIBERNATION_PREPARE:
5805 	case PM_SUSPEND_PREPARE:
5806 		return kvm_arch_suspend_notifier(kvm);
5807 	}
5808 
5809 	return NOTIFY_DONE;
5810 }
5811 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
5812 
5813 long kvm_arch_vm_ioctl(struct file *filp,
5814 		       unsigned int ioctl, unsigned long arg)
5815 {
5816 	struct kvm *kvm = filp->private_data;
5817 	void __user *argp = (void __user *)arg;
5818 	int r = -ENOTTY;
5819 	/*
5820 	 * This union makes it completely explicit to gcc-3.x
5821 	 * that these two variables' stack usage should be
5822 	 * combined, not added together.
5823 	 */
5824 	union {
5825 		struct kvm_pit_state ps;
5826 		struct kvm_pit_state2 ps2;
5827 		struct kvm_pit_config pit_config;
5828 	} u;
5829 
5830 	switch (ioctl) {
5831 	case KVM_SET_TSS_ADDR:
5832 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5833 		break;
5834 	case KVM_SET_IDENTITY_MAP_ADDR: {
5835 		u64 ident_addr;
5836 
5837 		mutex_lock(&kvm->lock);
5838 		r = -EINVAL;
5839 		if (kvm->created_vcpus)
5840 			goto set_identity_unlock;
5841 		r = -EFAULT;
5842 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5843 			goto set_identity_unlock;
5844 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5845 set_identity_unlock:
5846 		mutex_unlock(&kvm->lock);
5847 		break;
5848 	}
5849 	case KVM_SET_NR_MMU_PAGES:
5850 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5851 		break;
5852 	case KVM_GET_NR_MMU_PAGES:
5853 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5854 		break;
5855 	case KVM_CREATE_IRQCHIP: {
5856 		mutex_lock(&kvm->lock);
5857 
5858 		r = -EEXIST;
5859 		if (irqchip_in_kernel(kvm))
5860 			goto create_irqchip_unlock;
5861 
5862 		r = -EINVAL;
5863 		if (kvm->created_vcpus)
5864 			goto create_irqchip_unlock;
5865 
5866 		r = kvm_pic_init(kvm);
5867 		if (r)
5868 			goto create_irqchip_unlock;
5869 
5870 		r = kvm_ioapic_init(kvm);
5871 		if (r) {
5872 			kvm_pic_destroy(kvm);
5873 			goto create_irqchip_unlock;
5874 		}
5875 
5876 		r = kvm_setup_default_irq_routing(kvm);
5877 		if (r) {
5878 			kvm_ioapic_destroy(kvm);
5879 			kvm_pic_destroy(kvm);
5880 			goto create_irqchip_unlock;
5881 		}
5882 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5883 		smp_wmb();
5884 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5885 	create_irqchip_unlock:
5886 		mutex_unlock(&kvm->lock);
5887 		break;
5888 	}
5889 	case KVM_CREATE_PIT:
5890 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5891 		goto create_pit;
5892 	case KVM_CREATE_PIT2:
5893 		r = -EFAULT;
5894 		if (copy_from_user(&u.pit_config, argp,
5895 				   sizeof(struct kvm_pit_config)))
5896 			goto out;
5897 	create_pit:
5898 		mutex_lock(&kvm->lock);
5899 		r = -EEXIST;
5900 		if (kvm->arch.vpit)
5901 			goto create_pit_unlock;
5902 		r = -ENOMEM;
5903 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5904 		if (kvm->arch.vpit)
5905 			r = 0;
5906 	create_pit_unlock:
5907 		mutex_unlock(&kvm->lock);
5908 		break;
5909 	case KVM_GET_IRQCHIP: {
5910 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5911 		struct kvm_irqchip *chip;
5912 
5913 		chip = memdup_user(argp, sizeof(*chip));
5914 		if (IS_ERR(chip)) {
5915 			r = PTR_ERR(chip);
5916 			goto out;
5917 		}
5918 
5919 		r = -ENXIO;
5920 		if (!irqchip_kernel(kvm))
5921 			goto get_irqchip_out;
5922 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5923 		if (r)
5924 			goto get_irqchip_out;
5925 		r = -EFAULT;
5926 		if (copy_to_user(argp, chip, sizeof(*chip)))
5927 			goto get_irqchip_out;
5928 		r = 0;
5929 	get_irqchip_out:
5930 		kfree(chip);
5931 		break;
5932 	}
5933 	case KVM_SET_IRQCHIP: {
5934 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5935 		struct kvm_irqchip *chip;
5936 
5937 		chip = memdup_user(argp, sizeof(*chip));
5938 		if (IS_ERR(chip)) {
5939 			r = PTR_ERR(chip);
5940 			goto out;
5941 		}
5942 
5943 		r = -ENXIO;
5944 		if (!irqchip_kernel(kvm))
5945 			goto set_irqchip_out;
5946 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5947 	set_irqchip_out:
5948 		kfree(chip);
5949 		break;
5950 	}
5951 	case KVM_GET_PIT: {
5952 		r = -EFAULT;
5953 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5954 			goto out;
5955 		r = -ENXIO;
5956 		if (!kvm->arch.vpit)
5957 			goto out;
5958 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5959 		if (r)
5960 			goto out;
5961 		r = -EFAULT;
5962 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5963 			goto out;
5964 		r = 0;
5965 		break;
5966 	}
5967 	case KVM_SET_PIT: {
5968 		r = -EFAULT;
5969 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5970 			goto out;
5971 		mutex_lock(&kvm->lock);
5972 		r = -ENXIO;
5973 		if (!kvm->arch.vpit)
5974 			goto set_pit_out;
5975 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5976 set_pit_out:
5977 		mutex_unlock(&kvm->lock);
5978 		break;
5979 	}
5980 	case KVM_GET_PIT2: {
5981 		r = -ENXIO;
5982 		if (!kvm->arch.vpit)
5983 			goto out;
5984 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5985 		if (r)
5986 			goto out;
5987 		r = -EFAULT;
5988 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5989 			goto out;
5990 		r = 0;
5991 		break;
5992 	}
5993 	case KVM_SET_PIT2: {
5994 		r = -EFAULT;
5995 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5996 			goto out;
5997 		mutex_lock(&kvm->lock);
5998 		r = -ENXIO;
5999 		if (!kvm->arch.vpit)
6000 			goto set_pit2_out;
6001 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
6002 set_pit2_out:
6003 		mutex_unlock(&kvm->lock);
6004 		break;
6005 	}
6006 	case KVM_REINJECT_CONTROL: {
6007 		struct kvm_reinject_control control;
6008 		r =  -EFAULT;
6009 		if (copy_from_user(&control, argp, sizeof(control)))
6010 			goto out;
6011 		r = -ENXIO;
6012 		if (!kvm->arch.vpit)
6013 			goto out;
6014 		r = kvm_vm_ioctl_reinject(kvm, &control);
6015 		break;
6016 	}
6017 	case KVM_SET_BOOT_CPU_ID:
6018 		r = 0;
6019 		mutex_lock(&kvm->lock);
6020 		if (kvm->created_vcpus)
6021 			r = -EBUSY;
6022 		else
6023 			kvm->arch.bsp_vcpu_id = arg;
6024 		mutex_unlock(&kvm->lock);
6025 		break;
6026 #ifdef CONFIG_KVM_XEN
6027 	case KVM_XEN_HVM_CONFIG: {
6028 		struct kvm_xen_hvm_config xhc;
6029 		r = -EFAULT;
6030 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
6031 			goto out;
6032 		r = kvm_xen_hvm_config(kvm, &xhc);
6033 		break;
6034 	}
6035 	case KVM_XEN_HVM_GET_ATTR: {
6036 		struct kvm_xen_hvm_attr xha;
6037 
6038 		r = -EFAULT;
6039 		if (copy_from_user(&xha, argp, sizeof(xha)))
6040 			goto out;
6041 		r = kvm_xen_hvm_get_attr(kvm, &xha);
6042 		if (!r && copy_to_user(argp, &xha, sizeof(xha)))
6043 			r = -EFAULT;
6044 		break;
6045 	}
6046 	case KVM_XEN_HVM_SET_ATTR: {
6047 		struct kvm_xen_hvm_attr xha;
6048 
6049 		r = -EFAULT;
6050 		if (copy_from_user(&xha, argp, sizeof(xha)))
6051 			goto out;
6052 		r = kvm_xen_hvm_set_attr(kvm, &xha);
6053 		break;
6054 	}
6055 #endif
6056 	case KVM_SET_CLOCK: {
6057 		struct kvm_arch *ka = &kvm->arch;
6058 		struct kvm_clock_data user_ns;
6059 		u64 now_ns;
6060 
6061 		r = -EFAULT;
6062 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
6063 			goto out;
6064 
6065 		r = -EINVAL;
6066 		if (user_ns.flags)
6067 			goto out;
6068 
6069 		r = 0;
6070 		/*
6071 		 * TODO: userspace has to take care of races with VCPU_RUN, so
6072 		 * kvm_gen_update_masterclock() can be cut down to locked
6073 		 * pvclock_update_vm_gtod_copy().
6074 		 */
6075 		kvm_gen_update_masterclock(kvm);
6076 
6077 		/*
6078 		 * This pairs with kvm_guest_time_update(): when masterclock is
6079 		 * in use, we use master_kernel_ns + kvmclock_offset to set
6080 		 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
6081 		 * is slightly ahead) here we risk going negative on unsigned
6082 		 * 'system_time' when 'user_ns.clock' is very small.
6083 		 */
6084 		spin_lock_irq(&ka->pvclock_gtod_sync_lock);
6085 		if (kvm->arch.use_master_clock)
6086 			now_ns = ka->master_kernel_ns;
6087 		else
6088 			now_ns = get_kvmclock_base_ns();
6089 		ka->kvmclock_offset = user_ns.clock - now_ns;
6090 		spin_unlock_irq(&ka->pvclock_gtod_sync_lock);
6091 
6092 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
6093 		break;
6094 	}
6095 	case KVM_GET_CLOCK: {
6096 		struct kvm_clock_data user_ns;
6097 		u64 now_ns;
6098 
6099 		now_ns = get_kvmclock_ns(kvm);
6100 		user_ns.clock = now_ns;
6101 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
6102 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
6103 
6104 		r = -EFAULT;
6105 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
6106 			goto out;
6107 		r = 0;
6108 		break;
6109 	}
6110 	case KVM_MEMORY_ENCRYPT_OP: {
6111 		r = -ENOTTY;
6112 		if (kvm_x86_ops.mem_enc_op)
6113 			r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
6114 		break;
6115 	}
6116 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
6117 		struct kvm_enc_region region;
6118 
6119 		r = -EFAULT;
6120 		if (copy_from_user(&region, argp, sizeof(region)))
6121 			goto out;
6122 
6123 		r = -ENOTTY;
6124 		if (kvm_x86_ops.mem_enc_reg_region)
6125 			r = static_call(kvm_x86_mem_enc_reg_region)(kvm, &region);
6126 		break;
6127 	}
6128 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
6129 		struct kvm_enc_region region;
6130 
6131 		r = -EFAULT;
6132 		if (copy_from_user(&region, argp, sizeof(region)))
6133 			goto out;
6134 
6135 		r = -ENOTTY;
6136 		if (kvm_x86_ops.mem_enc_unreg_region)
6137 			r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, &region);
6138 		break;
6139 	}
6140 	case KVM_HYPERV_EVENTFD: {
6141 		struct kvm_hyperv_eventfd hvevfd;
6142 
6143 		r = -EFAULT;
6144 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
6145 			goto out;
6146 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
6147 		break;
6148 	}
6149 	case KVM_SET_PMU_EVENT_FILTER:
6150 		r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
6151 		break;
6152 	case KVM_X86_SET_MSR_FILTER:
6153 		r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
6154 		break;
6155 	default:
6156 		r = -ENOTTY;
6157 	}
6158 out:
6159 	return r;
6160 }
6161 
6162 static void kvm_init_msr_list(void)
6163 {
6164 	struct x86_pmu_capability x86_pmu;
6165 	u32 dummy[2];
6166 	unsigned i;
6167 
6168 	BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
6169 			 "Please update the fixed PMCs in msrs_to_saved_all[]");
6170 
6171 	perf_get_x86_pmu_capability(&x86_pmu);
6172 
6173 	num_msrs_to_save = 0;
6174 	num_emulated_msrs = 0;
6175 	num_msr_based_features = 0;
6176 
6177 	for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
6178 		if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
6179 			continue;
6180 
6181 		/*
6182 		 * Even MSRs that are valid in the host may not be exposed
6183 		 * to the guests in some cases.
6184 		 */
6185 		switch (msrs_to_save_all[i]) {
6186 		case MSR_IA32_BNDCFGS:
6187 			if (!kvm_mpx_supported())
6188 				continue;
6189 			break;
6190 		case MSR_TSC_AUX:
6191 			if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
6192 			    !kvm_cpu_cap_has(X86_FEATURE_RDPID))
6193 				continue;
6194 			break;
6195 		case MSR_IA32_UMWAIT_CONTROL:
6196 			if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
6197 				continue;
6198 			break;
6199 		case MSR_IA32_RTIT_CTL:
6200 		case MSR_IA32_RTIT_STATUS:
6201 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
6202 				continue;
6203 			break;
6204 		case MSR_IA32_RTIT_CR3_MATCH:
6205 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6206 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
6207 				continue;
6208 			break;
6209 		case MSR_IA32_RTIT_OUTPUT_BASE:
6210 		case MSR_IA32_RTIT_OUTPUT_MASK:
6211 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6212 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
6213 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
6214 				continue;
6215 			break;
6216 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
6217 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6218 				msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
6219 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
6220 				continue;
6221 			break;
6222 		case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
6223 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
6224 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6225 				continue;
6226 			break;
6227 		case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
6228 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
6229 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6230 				continue;
6231 			break;
6232 		default:
6233 			break;
6234 		}
6235 
6236 		msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
6237 	}
6238 
6239 	for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
6240 		if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
6241 			continue;
6242 
6243 		emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
6244 	}
6245 
6246 	for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
6247 		struct kvm_msr_entry msr;
6248 
6249 		msr.index = msr_based_features_all[i];
6250 		if (kvm_get_msr_feature(&msr))
6251 			continue;
6252 
6253 		msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6254 	}
6255 }
6256 
6257 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6258 			   const void *v)
6259 {
6260 	int handled = 0;
6261 	int n;
6262 
6263 	do {
6264 		n = min(len, 8);
6265 		if (!(lapic_in_kernel(vcpu) &&
6266 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6267 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6268 			break;
6269 		handled += n;
6270 		addr += n;
6271 		len -= n;
6272 		v += n;
6273 	} while (len);
6274 
6275 	return handled;
6276 }
6277 
6278 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6279 {
6280 	int handled = 0;
6281 	int n;
6282 
6283 	do {
6284 		n = min(len, 8);
6285 		if (!(lapic_in_kernel(vcpu) &&
6286 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6287 					 addr, n, v))
6288 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6289 			break;
6290 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6291 		handled += n;
6292 		addr += n;
6293 		len -= n;
6294 		v += n;
6295 	} while (len);
6296 
6297 	return handled;
6298 }
6299 
6300 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6301 			struct kvm_segment *var, int seg)
6302 {
6303 	static_call(kvm_x86_set_segment)(vcpu, var, seg);
6304 }
6305 
6306 void kvm_get_segment(struct kvm_vcpu *vcpu,
6307 		     struct kvm_segment *var, int seg)
6308 {
6309 	static_call(kvm_x86_get_segment)(vcpu, var, seg);
6310 }
6311 
6312 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6313 			   struct x86_exception *exception)
6314 {
6315 	gpa_t t_gpa;
6316 
6317 	BUG_ON(!mmu_is_nested(vcpu));
6318 
6319 	/* NPT walks are always user-walks */
6320 	access |= PFERR_USER_MASK;
6321 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
6322 
6323 	return t_gpa;
6324 }
6325 
6326 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6327 			      struct x86_exception *exception)
6328 {
6329 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6330 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6331 }
6332 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6333 
6334  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6335 				struct x86_exception *exception)
6336 {
6337 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6338 	access |= PFERR_FETCH_MASK;
6339 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6340 }
6341 
6342 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6343 			       struct x86_exception *exception)
6344 {
6345 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6346 	access |= PFERR_WRITE_MASK;
6347 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6348 }
6349 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6350 
6351 /* uses this to access any guest's mapped memory without checking CPL */
6352 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6353 				struct x86_exception *exception)
6354 {
6355 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
6356 }
6357 
6358 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6359 				      struct kvm_vcpu *vcpu, u32 access,
6360 				      struct x86_exception *exception)
6361 {
6362 	void *data = val;
6363 	int r = X86EMUL_CONTINUE;
6364 
6365 	while (bytes) {
6366 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
6367 							    exception);
6368 		unsigned offset = addr & (PAGE_SIZE-1);
6369 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6370 		int ret;
6371 
6372 		if (gpa == UNMAPPED_GVA)
6373 			return X86EMUL_PROPAGATE_FAULT;
6374 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6375 					       offset, toread);
6376 		if (ret < 0) {
6377 			r = X86EMUL_IO_NEEDED;
6378 			goto out;
6379 		}
6380 
6381 		bytes -= toread;
6382 		data += toread;
6383 		addr += toread;
6384 	}
6385 out:
6386 	return r;
6387 }
6388 
6389 /* used for instruction fetching */
6390 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6391 				gva_t addr, void *val, unsigned int bytes,
6392 				struct x86_exception *exception)
6393 {
6394 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6395 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6396 	unsigned offset;
6397 	int ret;
6398 
6399 	/* Inline kvm_read_guest_virt_helper for speed.  */
6400 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6401 						    exception);
6402 	if (unlikely(gpa == UNMAPPED_GVA))
6403 		return X86EMUL_PROPAGATE_FAULT;
6404 
6405 	offset = addr & (PAGE_SIZE-1);
6406 	if (WARN_ON(offset + bytes > PAGE_SIZE))
6407 		bytes = (unsigned)PAGE_SIZE - offset;
6408 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6409 				       offset, bytes);
6410 	if (unlikely(ret < 0))
6411 		return X86EMUL_IO_NEEDED;
6412 
6413 	return X86EMUL_CONTINUE;
6414 }
6415 
6416 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6417 			       gva_t addr, void *val, unsigned int bytes,
6418 			       struct x86_exception *exception)
6419 {
6420 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6421 
6422 	/*
6423 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6424 	 * is returned, but our callers are not ready for that and they blindly
6425 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
6426 	 * uninitialized kernel stack memory into cr2 and error code.
6427 	 */
6428 	memset(exception, 0, sizeof(*exception));
6429 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6430 					  exception);
6431 }
6432 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6433 
6434 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6435 			     gva_t addr, void *val, unsigned int bytes,
6436 			     struct x86_exception *exception, bool system)
6437 {
6438 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6439 	u32 access = 0;
6440 
6441 	if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6442 		access |= PFERR_USER_MASK;
6443 
6444 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6445 }
6446 
6447 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6448 		unsigned long addr, void *val, unsigned int bytes)
6449 {
6450 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6451 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6452 
6453 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6454 }
6455 
6456 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6457 				      struct kvm_vcpu *vcpu, u32 access,
6458 				      struct x86_exception *exception)
6459 {
6460 	void *data = val;
6461 	int r = X86EMUL_CONTINUE;
6462 
6463 	while (bytes) {
6464 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6465 							     access,
6466 							     exception);
6467 		unsigned offset = addr & (PAGE_SIZE-1);
6468 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6469 		int ret;
6470 
6471 		if (gpa == UNMAPPED_GVA)
6472 			return X86EMUL_PROPAGATE_FAULT;
6473 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6474 		if (ret < 0) {
6475 			r = X86EMUL_IO_NEEDED;
6476 			goto out;
6477 		}
6478 
6479 		bytes -= towrite;
6480 		data += towrite;
6481 		addr += towrite;
6482 	}
6483 out:
6484 	return r;
6485 }
6486 
6487 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6488 			      unsigned int bytes, struct x86_exception *exception,
6489 			      bool system)
6490 {
6491 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6492 	u32 access = PFERR_WRITE_MASK;
6493 
6494 	if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6495 		access |= PFERR_USER_MASK;
6496 
6497 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6498 					   access, exception);
6499 }
6500 
6501 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6502 				unsigned int bytes, struct x86_exception *exception)
6503 {
6504 	/* kvm_write_guest_virt_system can pull in tons of pages. */
6505 	vcpu->arch.l1tf_flush_l1d = true;
6506 
6507 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6508 					   PFERR_WRITE_MASK, exception);
6509 }
6510 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6511 
6512 int handle_ud(struct kvm_vcpu *vcpu)
6513 {
6514 	static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6515 	int emul_type = EMULTYPE_TRAP_UD;
6516 	char sig[5]; /* ud2; .ascii "kvm" */
6517 	struct x86_exception e;
6518 
6519 	if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6520 		return 1;
6521 
6522 	if (force_emulation_prefix &&
6523 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6524 				sig, sizeof(sig), &e) == 0 &&
6525 	    memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6526 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6527 		emul_type = EMULTYPE_TRAP_UD_FORCED;
6528 	}
6529 
6530 	return kvm_emulate_instruction(vcpu, emul_type);
6531 }
6532 EXPORT_SYMBOL_GPL(handle_ud);
6533 
6534 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6535 			    gpa_t gpa, bool write)
6536 {
6537 	/* For APIC access vmexit */
6538 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6539 		return 1;
6540 
6541 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6542 		trace_vcpu_match_mmio(gva, gpa, write, true);
6543 		return 1;
6544 	}
6545 
6546 	return 0;
6547 }
6548 
6549 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6550 				gpa_t *gpa, struct x86_exception *exception,
6551 				bool write)
6552 {
6553 	u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6554 		| (write ? PFERR_WRITE_MASK : 0);
6555 
6556 	/*
6557 	 * currently PKRU is only applied to ept enabled guest so
6558 	 * there is no pkey in EPT page table for L1 guest or EPT
6559 	 * shadow page table for L2 guest.
6560 	 */
6561 	if (vcpu_match_mmio_gva(vcpu, gva)
6562 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
6563 				 vcpu->arch.mmio_access, 0, access)) {
6564 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6565 					(gva & (PAGE_SIZE - 1));
6566 		trace_vcpu_match_mmio(gva, *gpa, write, false);
6567 		return 1;
6568 	}
6569 
6570 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6571 
6572 	if (*gpa == UNMAPPED_GVA)
6573 		return -1;
6574 
6575 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6576 }
6577 
6578 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6579 			const void *val, int bytes)
6580 {
6581 	int ret;
6582 
6583 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6584 	if (ret < 0)
6585 		return 0;
6586 	kvm_page_track_write(vcpu, gpa, val, bytes);
6587 	return 1;
6588 }
6589 
6590 struct read_write_emulator_ops {
6591 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6592 				  int bytes);
6593 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6594 				  void *val, int bytes);
6595 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6596 			       int bytes, void *val);
6597 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6598 				    void *val, int bytes);
6599 	bool write;
6600 };
6601 
6602 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6603 {
6604 	if (vcpu->mmio_read_completed) {
6605 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6606 			       vcpu->mmio_fragments[0].gpa, val);
6607 		vcpu->mmio_read_completed = 0;
6608 		return 1;
6609 	}
6610 
6611 	return 0;
6612 }
6613 
6614 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6615 			void *val, int bytes)
6616 {
6617 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6618 }
6619 
6620 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6621 			 void *val, int bytes)
6622 {
6623 	return emulator_write_phys(vcpu, gpa, val, bytes);
6624 }
6625 
6626 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6627 {
6628 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6629 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
6630 }
6631 
6632 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6633 			  void *val, int bytes)
6634 {
6635 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6636 	return X86EMUL_IO_NEEDED;
6637 }
6638 
6639 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6640 			   void *val, int bytes)
6641 {
6642 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6643 
6644 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6645 	return X86EMUL_CONTINUE;
6646 }
6647 
6648 static const struct read_write_emulator_ops read_emultor = {
6649 	.read_write_prepare = read_prepare,
6650 	.read_write_emulate = read_emulate,
6651 	.read_write_mmio = vcpu_mmio_read,
6652 	.read_write_exit_mmio = read_exit_mmio,
6653 };
6654 
6655 static const struct read_write_emulator_ops write_emultor = {
6656 	.read_write_emulate = write_emulate,
6657 	.read_write_mmio = write_mmio,
6658 	.read_write_exit_mmio = write_exit_mmio,
6659 	.write = true,
6660 };
6661 
6662 static int emulator_read_write_onepage(unsigned long addr, void *val,
6663 				       unsigned int bytes,
6664 				       struct x86_exception *exception,
6665 				       struct kvm_vcpu *vcpu,
6666 				       const struct read_write_emulator_ops *ops)
6667 {
6668 	gpa_t gpa;
6669 	int handled, ret;
6670 	bool write = ops->write;
6671 	struct kvm_mmio_fragment *frag;
6672 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6673 
6674 	/*
6675 	 * If the exit was due to a NPF we may already have a GPA.
6676 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6677 	 * Note, this cannot be used on string operations since string
6678 	 * operation using rep will only have the initial GPA from the NPF
6679 	 * occurred.
6680 	 */
6681 	if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6682 	    (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6683 		gpa = ctxt->gpa_val;
6684 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6685 	} else {
6686 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6687 		if (ret < 0)
6688 			return X86EMUL_PROPAGATE_FAULT;
6689 	}
6690 
6691 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6692 		return X86EMUL_CONTINUE;
6693 
6694 	/*
6695 	 * Is this MMIO handled locally?
6696 	 */
6697 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6698 	if (handled == bytes)
6699 		return X86EMUL_CONTINUE;
6700 
6701 	gpa += handled;
6702 	bytes -= handled;
6703 	val += handled;
6704 
6705 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6706 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6707 	frag->gpa = gpa;
6708 	frag->data = val;
6709 	frag->len = bytes;
6710 	return X86EMUL_CONTINUE;
6711 }
6712 
6713 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6714 			unsigned long addr,
6715 			void *val, unsigned int bytes,
6716 			struct x86_exception *exception,
6717 			const struct read_write_emulator_ops *ops)
6718 {
6719 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6720 	gpa_t gpa;
6721 	int rc;
6722 
6723 	if (ops->read_write_prepare &&
6724 		  ops->read_write_prepare(vcpu, val, bytes))
6725 		return X86EMUL_CONTINUE;
6726 
6727 	vcpu->mmio_nr_fragments = 0;
6728 
6729 	/* Crossing a page boundary? */
6730 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6731 		int now;
6732 
6733 		now = -addr & ~PAGE_MASK;
6734 		rc = emulator_read_write_onepage(addr, val, now, exception,
6735 						 vcpu, ops);
6736 
6737 		if (rc != X86EMUL_CONTINUE)
6738 			return rc;
6739 		addr += now;
6740 		if (ctxt->mode != X86EMUL_MODE_PROT64)
6741 			addr = (u32)addr;
6742 		val += now;
6743 		bytes -= now;
6744 	}
6745 
6746 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
6747 					 vcpu, ops);
6748 	if (rc != X86EMUL_CONTINUE)
6749 		return rc;
6750 
6751 	if (!vcpu->mmio_nr_fragments)
6752 		return rc;
6753 
6754 	gpa = vcpu->mmio_fragments[0].gpa;
6755 
6756 	vcpu->mmio_needed = 1;
6757 	vcpu->mmio_cur_fragment = 0;
6758 
6759 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6760 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6761 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
6762 	vcpu->run->mmio.phys_addr = gpa;
6763 
6764 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6765 }
6766 
6767 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6768 				  unsigned long addr,
6769 				  void *val,
6770 				  unsigned int bytes,
6771 				  struct x86_exception *exception)
6772 {
6773 	return emulator_read_write(ctxt, addr, val, bytes,
6774 				   exception, &read_emultor);
6775 }
6776 
6777 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6778 			    unsigned long addr,
6779 			    const void *val,
6780 			    unsigned int bytes,
6781 			    struct x86_exception *exception)
6782 {
6783 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
6784 				   exception, &write_emultor);
6785 }
6786 
6787 #define CMPXCHG_TYPE(t, ptr, old, new) \
6788 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6789 
6790 #ifdef CONFIG_X86_64
6791 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6792 #else
6793 #  define CMPXCHG64(ptr, old, new) \
6794 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6795 #endif
6796 
6797 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6798 				     unsigned long addr,
6799 				     const void *old,
6800 				     const void *new,
6801 				     unsigned int bytes,
6802 				     struct x86_exception *exception)
6803 {
6804 	struct kvm_host_map map;
6805 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6806 	u64 page_line_mask;
6807 	gpa_t gpa;
6808 	char *kaddr;
6809 	bool exchanged;
6810 
6811 	/* guests cmpxchg8b have to be emulated atomically */
6812 	if (bytes > 8 || (bytes & (bytes - 1)))
6813 		goto emul_write;
6814 
6815 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6816 
6817 	if (gpa == UNMAPPED_GVA ||
6818 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6819 		goto emul_write;
6820 
6821 	/*
6822 	 * Emulate the atomic as a straight write to avoid #AC if SLD is
6823 	 * enabled in the host and the access splits a cache line.
6824 	 */
6825 	if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6826 		page_line_mask = ~(cache_line_size() - 1);
6827 	else
6828 		page_line_mask = PAGE_MASK;
6829 
6830 	if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6831 		goto emul_write;
6832 
6833 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6834 		goto emul_write;
6835 
6836 	kaddr = map.hva + offset_in_page(gpa);
6837 
6838 	switch (bytes) {
6839 	case 1:
6840 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6841 		break;
6842 	case 2:
6843 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6844 		break;
6845 	case 4:
6846 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6847 		break;
6848 	case 8:
6849 		exchanged = CMPXCHG64(kaddr, old, new);
6850 		break;
6851 	default:
6852 		BUG();
6853 	}
6854 
6855 	kvm_vcpu_unmap(vcpu, &map, true);
6856 
6857 	if (!exchanged)
6858 		return X86EMUL_CMPXCHG_FAILED;
6859 
6860 	kvm_page_track_write(vcpu, gpa, new, bytes);
6861 
6862 	return X86EMUL_CONTINUE;
6863 
6864 emul_write:
6865 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6866 
6867 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6868 }
6869 
6870 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6871 {
6872 	int r = 0, i;
6873 
6874 	for (i = 0; i < vcpu->arch.pio.count; i++) {
6875 		if (vcpu->arch.pio.in)
6876 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6877 					    vcpu->arch.pio.size, pd);
6878 		else
6879 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6880 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
6881 					     pd);
6882 		if (r)
6883 			break;
6884 		pd += vcpu->arch.pio.size;
6885 	}
6886 	return r;
6887 }
6888 
6889 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6890 			       unsigned short port, void *val,
6891 			       unsigned int count, bool in)
6892 {
6893 	vcpu->arch.pio.port = port;
6894 	vcpu->arch.pio.in = in;
6895 	vcpu->arch.pio.count  = count;
6896 	vcpu->arch.pio.size = size;
6897 
6898 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6899 		vcpu->arch.pio.count = 0;
6900 		return 1;
6901 	}
6902 
6903 	vcpu->run->exit_reason = KVM_EXIT_IO;
6904 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6905 	vcpu->run->io.size = size;
6906 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6907 	vcpu->run->io.count = count;
6908 	vcpu->run->io.port = port;
6909 
6910 	return 0;
6911 }
6912 
6913 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6914 			   unsigned short port, void *val, unsigned int count)
6915 {
6916 	int ret;
6917 
6918 	if (vcpu->arch.pio.count)
6919 		goto data_avail;
6920 
6921 	memset(vcpu->arch.pio_data, 0, size * count);
6922 
6923 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6924 	if (ret) {
6925 data_avail:
6926 		memcpy(val, vcpu->arch.pio_data, size * count);
6927 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6928 		vcpu->arch.pio.count = 0;
6929 		return 1;
6930 	}
6931 
6932 	return 0;
6933 }
6934 
6935 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6936 				    int size, unsigned short port, void *val,
6937 				    unsigned int count)
6938 {
6939 	return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6940 
6941 }
6942 
6943 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6944 			    unsigned short port, const void *val,
6945 			    unsigned int count)
6946 {
6947 	memcpy(vcpu->arch.pio_data, val, size * count);
6948 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6949 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6950 }
6951 
6952 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6953 				     int size, unsigned short port,
6954 				     const void *val, unsigned int count)
6955 {
6956 	return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6957 }
6958 
6959 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6960 {
6961 	return static_call(kvm_x86_get_segment_base)(vcpu, seg);
6962 }
6963 
6964 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6965 {
6966 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6967 }
6968 
6969 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6970 {
6971 	if (!need_emulate_wbinvd(vcpu))
6972 		return X86EMUL_CONTINUE;
6973 
6974 	if (static_call(kvm_x86_has_wbinvd_exit)()) {
6975 		int cpu = get_cpu();
6976 
6977 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6978 		on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
6979 				wbinvd_ipi, NULL, 1);
6980 		put_cpu();
6981 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6982 	} else
6983 		wbinvd();
6984 	return X86EMUL_CONTINUE;
6985 }
6986 
6987 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6988 {
6989 	kvm_emulate_wbinvd_noskip(vcpu);
6990 	return kvm_skip_emulated_instruction(vcpu);
6991 }
6992 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6993 
6994 
6995 
6996 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6997 {
6998 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6999 }
7000 
7001 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
7002 			    unsigned long *dest)
7003 {
7004 	kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
7005 }
7006 
7007 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
7008 			   unsigned long value)
7009 {
7010 
7011 	return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
7012 }
7013 
7014 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
7015 {
7016 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
7017 }
7018 
7019 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
7020 {
7021 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7022 	unsigned long value;
7023 
7024 	switch (cr) {
7025 	case 0:
7026 		value = kvm_read_cr0(vcpu);
7027 		break;
7028 	case 2:
7029 		value = vcpu->arch.cr2;
7030 		break;
7031 	case 3:
7032 		value = kvm_read_cr3(vcpu);
7033 		break;
7034 	case 4:
7035 		value = kvm_read_cr4(vcpu);
7036 		break;
7037 	case 8:
7038 		value = kvm_get_cr8(vcpu);
7039 		break;
7040 	default:
7041 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
7042 		return 0;
7043 	}
7044 
7045 	return value;
7046 }
7047 
7048 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
7049 {
7050 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7051 	int res = 0;
7052 
7053 	switch (cr) {
7054 	case 0:
7055 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
7056 		break;
7057 	case 2:
7058 		vcpu->arch.cr2 = val;
7059 		break;
7060 	case 3:
7061 		res = kvm_set_cr3(vcpu, val);
7062 		break;
7063 	case 4:
7064 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
7065 		break;
7066 	case 8:
7067 		res = kvm_set_cr8(vcpu, val);
7068 		break;
7069 	default:
7070 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
7071 		res = -1;
7072 	}
7073 
7074 	return res;
7075 }
7076 
7077 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
7078 {
7079 	return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
7080 }
7081 
7082 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7083 {
7084 	static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
7085 }
7086 
7087 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7088 {
7089 	static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
7090 }
7091 
7092 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7093 {
7094 	static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
7095 }
7096 
7097 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7098 {
7099 	static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
7100 }
7101 
7102 static unsigned long emulator_get_cached_segment_base(
7103 	struct x86_emulate_ctxt *ctxt, int seg)
7104 {
7105 	return get_segment_base(emul_to_vcpu(ctxt), seg);
7106 }
7107 
7108 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
7109 				 struct desc_struct *desc, u32 *base3,
7110 				 int seg)
7111 {
7112 	struct kvm_segment var;
7113 
7114 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
7115 	*selector = var.selector;
7116 
7117 	if (var.unusable) {
7118 		memset(desc, 0, sizeof(*desc));
7119 		if (base3)
7120 			*base3 = 0;
7121 		return false;
7122 	}
7123 
7124 	if (var.g)
7125 		var.limit >>= 12;
7126 	set_desc_limit(desc, var.limit);
7127 	set_desc_base(desc, (unsigned long)var.base);
7128 #ifdef CONFIG_X86_64
7129 	if (base3)
7130 		*base3 = var.base >> 32;
7131 #endif
7132 	desc->type = var.type;
7133 	desc->s = var.s;
7134 	desc->dpl = var.dpl;
7135 	desc->p = var.present;
7136 	desc->avl = var.avl;
7137 	desc->l = var.l;
7138 	desc->d = var.db;
7139 	desc->g = var.g;
7140 
7141 	return true;
7142 }
7143 
7144 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
7145 				 struct desc_struct *desc, u32 base3,
7146 				 int seg)
7147 {
7148 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7149 	struct kvm_segment var;
7150 
7151 	var.selector = selector;
7152 	var.base = get_desc_base(desc);
7153 #ifdef CONFIG_X86_64
7154 	var.base |= ((u64)base3) << 32;
7155 #endif
7156 	var.limit = get_desc_limit(desc);
7157 	if (desc->g)
7158 		var.limit = (var.limit << 12) | 0xfff;
7159 	var.type = desc->type;
7160 	var.dpl = desc->dpl;
7161 	var.db = desc->d;
7162 	var.s = desc->s;
7163 	var.l = desc->l;
7164 	var.g = desc->g;
7165 	var.avl = desc->avl;
7166 	var.present = desc->p;
7167 	var.unusable = !var.present;
7168 	var.padding = 0;
7169 
7170 	kvm_set_segment(vcpu, &var, seg);
7171 	return;
7172 }
7173 
7174 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
7175 			    u32 msr_index, u64 *pdata)
7176 {
7177 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7178 	int r;
7179 
7180 	r = kvm_get_msr(vcpu, msr_index, pdata);
7181 
7182 	if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
7183 		/* Bounce to user space */
7184 		return X86EMUL_IO_NEEDED;
7185 	}
7186 
7187 	return r;
7188 }
7189 
7190 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
7191 			    u32 msr_index, u64 data)
7192 {
7193 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7194 	int r;
7195 
7196 	r = kvm_set_msr(vcpu, msr_index, data);
7197 
7198 	if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
7199 		/* Bounce to user space */
7200 		return X86EMUL_IO_NEEDED;
7201 	}
7202 
7203 	return r;
7204 }
7205 
7206 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
7207 {
7208 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7209 
7210 	return vcpu->arch.smbase;
7211 }
7212 
7213 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
7214 {
7215 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7216 
7217 	vcpu->arch.smbase = smbase;
7218 }
7219 
7220 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
7221 			      u32 pmc)
7222 {
7223 	return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
7224 }
7225 
7226 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
7227 			     u32 pmc, u64 *pdata)
7228 {
7229 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
7230 }
7231 
7232 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
7233 {
7234 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
7235 }
7236 
7237 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
7238 			      struct x86_instruction_info *info,
7239 			      enum x86_intercept_stage stage)
7240 {
7241 	return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
7242 					    &ctxt->exception);
7243 }
7244 
7245 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
7246 			      u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
7247 			      bool exact_only)
7248 {
7249 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7250 }
7251 
7252 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7253 {
7254 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7255 }
7256 
7257 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7258 {
7259 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7260 }
7261 
7262 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7263 {
7264 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7265 }
7266 
7267 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7268 {
7269 	return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7270 }
7271 
7272 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7273 {
7274 	kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7275 }
7276 
7277 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7278 {
7279 	static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7280 }
7281 
7282 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7283 {
7284 	return emul_to_vcpu(ctxt)->arch.hflags;
7285 }
7286 
7287 static void emulator_exiting_smm(struct x86_emulate_ctxt *ctxt)
7288 {
7289 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7290 
7291 	kvm_smm_changed(vcpu, false);
7292 }
7293 
7294 static int emulator_leave_smm(struct x86_emulate_ctxt *ctxt,
7295 				  const char *smstate)
7296 {
7297 	return static_call(kvm_x86_leave_smm)(emul_to_vcpu(ctxt), smstate);
7298 }
7299 
7300 static void emulator_triple_fault(struct x86_emulate_ctxt *ctxt)
7301 {
7302 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, emul_to_vcpu(ctxt));
7303 }
7304 
7305 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7306 {
7307 	return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7308 }
7309 
7310 static const struct x86_emulate_ops emulate_ops = {
7311 	.read_gpr            = emulator_read_gpr,
7312 	.write_gpr           = emulator_write_gpr,
7313 	.read_std            = emulator_read_std,
7314 	.write_std           = emulator_write_std,
7315 	.read_phys           = kvm_read_guest_phys_system,
7316 	.fetch               = kvm_fetch_guest_virt,
7317 	.read_emulated       = emulator_read_emulated,
7318 	.write_emulated      = emulator_write_emulated,
7319 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
7320 	.invlpg              = emulator_invlpg,
7321 	.pio_in_emulated     = emulator_pio_in_emulated,
7322 	.pio_out_emulated    = emulator_pio_out_emulated,
7323 	.get_segment         = emulator_get_segment,
7324 	.set_segment         = emulator_set_segment,
7325 	.get_cached_segment_base = emulator_get_cached_segment_base,
7326 	.get_gdt             = emulator_get_gdt,
7327 	.get_idt	     = emulator_get_idt,
7328 	.set_gdt             = emulator_set_gdt,
7329 	.set_idt	     = emulator_set_idt,
7330 	.get_cr              = emulator_get_cr,
7331 	.set_cr              = emulator_set_cr,
7332 	.cpl                 = emulator_get_cpl,
7333 	.get_dr              = emulator_get_dr,
7334 	.set_dr              = emulator_set_dr,
7335 	.get_smbase          = emulator_get_smbase,
7336 	.set_smbase          = emulator_set_smbase,
7337 	.set_msr             = emulator_set_msr,
7338 	.get_msr             = emulator_get_msr,
7339 	.check_pmc	     = emulator_check_pmc,
7340 	.read_pmc            = emulator_read_pmc,
7341 	.halt                = emulator_halt,
7342 	.wbinvd              = emulator_wbinvd,
7343 	.fix_hypercall       = emulator_fix_hypercall,
7344 	.intercept           = emulator_intercept,
7345 	.get_cpuid           = emulator_get_cpuid,
7346 	.guest_has_long_mode = emulator_guest_has_long_mode,
7347 	.guest_has_movbe     = emulator_guest_has_movbe,
7348 	.guest_has_fxsr      = emulator_guest_has_fxsr,
7349 	.set_nmi_mask        = emulator_set_nmi_mask,
7350 	.get_hflags          = emulator_get_hflags,
7351 	.exiting_smm         = emulator_exiting_smm,
7352 	.leave_smm           = emulator_leave_smm,
7353 	.triple_fault        = emulator_triple_fault,
7354 	.set_xcr             = emulator_set_xcr,
7355 };
7356 
7357 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7358 {
7359 	u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7360 	/*
7361 	 * an sti; sti; sequence only disable interrupts for the first
7362 	 * instruction. So, if the last instruction, be it emulated or
7363 	 * not, left the system with the INT_STI flag enabled, it
7364 	 * means that the last instruction is an sti. We should not
7365 	 * leave the flag on in this case. The same goes for mov ss
7366 	 */
7367 	if (int_shadow & mask)
7368 		mask = 0;
7369 	if (unlikely(int_shadow || mask)) {
7370 		static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7371 		if (!mask)
7372 			kvm_make_request(KVM_REQ_EVENT, vcpu);
7373 	}
7374 }
7375 
7376 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7377 {
7378 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7379 	if (ctxt->exception.vector == PF_VECTOR)
7380 		return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7381 
7382 	if (ctxt->exception.error_code_valid)
7383 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7384 				      ctxt->exception.error_code);
7385 	else
7386 		kvm_queue_exception(vcpu, ctxt->exception.vector);
7387 	return false;
7388 }
7389 
7390 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7391 {
7392 	struct x86_emulate_ctxt *ctxt;
7393 
7394 	ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7395 	if (!ctxt) {
7396 		pr_err("kvm: failed to allocate vcpu's emulator\n");
7397 		return NULL;
7398 	}
7399 
7400 	ctxt->vcpu = vcpu;
7401 	ctxt->ops = &emulate_ops;
7402 	vcpu->arch.emulate_ctxt = ctxt;
7403 
7404 	return ctxt;
7405 }
7406 
7407 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7408 {
7409 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7410 	int cs_db, cs_l;
7411 
7412 	static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7413 
7414 	ctxt->gpa_available = false;
7415 	ctxt->eflags = kvm_get_rflags(vcpu);
7416 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7417 
7418 	ctxt->eip = kvm_rip_read(vcpu);
7419 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
7420 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
7421 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
7422 		     cs_db				? X86EMUL_MODE_PROT32 :
7423 							  X86EMUL_MODE_PROT16;
7424 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7425 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7426 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7427 
7428 	ctxt->interruptibility = 0;
7429 	ctxt->have_exception = false;
7430 	ctxt->exception.vector = -1;
7431 	ctxt->perm_ok = false;
7432 
7433 	init_decode_cache(ctxt);
7434 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7435 }
7436 
7437 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7438 {
7439 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7440 	int ret;
7441 
7442 	init_emulate_ctxt(vcpu);
7443 
7444 	ctxt->op_bytes = 2;
7445 	ctxt->ad_bytes = 2;
7446 	ctxt->_eip = ctxt->eip + inc_eip;
7447 	ret = emulate_int_real(ctxt, irq);
7448 
7449 	if (ret != X86EMUL_CONTINUE) {
7450 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7451 	} else {
7452 		ctxt->eip = ctxt->_eip;
7453 		kvm_rip_write(vcpu, ctxt->eip);
7454 		kvm_set_rflags(vcpu, ctxt->eflags);
7455 	}
7456 }
7457 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7458 
7459 static void prepare_emulation_failure_exit(struct kvm_vcpu *vcpu)
7460 {
7461 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7462 	u32 insn_size = ctxt->fetch.end - ctxt->fetch.data;
7463 	struct kvm_run *run = vcpu->run;
7464 
7465 	run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7466 	run->emulation_failure.suberror = KVM_INTERNAL_ERROR_EMULATION;
7467 	run->emulation_failure.ndata = 0;
7468 	run->emulation_failure.flags = 0;
7469 
7470 	if (insn_size) {
7471 		run->emulation_failure.ndata = 3;
7472 		run->emulation_failure.flags |=
7473 			KVM_INTERNAL_ERROR_EMULATION_FLAG_INSTRUCTION_BYTES;
7474 		run->emulation_failure.insn_size = insn_size;
7475 		memset(run->emulation_failure.insn_bytes, 0x90,
7476 		       sizeof(run->emulation_failure.insn_bytes));
7477 		memcpy(run->emulation_failure.insn_bytes,
7478 		       ctxt->fetch.data, insn_size);
7479 	}
7480 }
7481 
7482 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7483 {
7484 	struct kvm *kvm = vcpu->kvm;
7485 
7486 	++vcpu->stat.insn_emulation_fail;
7487 	trace_kvm_emulate_insn_failed(vcpu);
7488 
7489 	if (emulation_type & EMULTYPE_VMWARE_GP) {
7490 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7491 		return 1;
7492 	}
7493 
7494 	if (kvm->arch.exit_on_emulation_error ||
7495 	    (emulation_type & EMULTYPE_SKIP)) {
7496 		prepare_emulation_failure_exit(vcpu);
7497 		return 0;
7498 	}
7499 
7500 	kvm_queue_exception(vcpu, UD_VECTOR);
7501 
7502 	if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7503 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7504 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7505 		vcpu->run->internal.ndata = 0;
7506 		return 0;
7507 	}
7508 
7509 	return 1;
7510 }
7511 
7512 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7513 				  bool write_fault_to_shadow_pgtable,
7514 				  int emulation_type)
7515 {
7516 	gpa_t gpa = cr2_or_gpa;
7517 	kvm_pfn_t pfn;
7518 
7519 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7520 		return false;
7521 
7522 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7523 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7524 		return false;
7525 
7526 	if (!vcpu->arch.mmu->direct_map) {
7527 		/*
7528 		 * Write permission should be allowed since only
7529 		 * write access need to be emulated.
7530 		 */
7531 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7532 
7533 		/*
7534 		 * If the mapping is invalid in guest, let cpu retry
7535 		 * it to generate fault.
7536 		 */
7537 		if (gpa == UNMAPPED_GVA)
7538 			return true;
7539 	}
7540 
7541 	/*
7542 	 * Do not retry the unhandleable instruction if it faults on the
7543 	 * readonly host memory, otherwise it will goto a infinite loop:
7544 	 * retry instruction -> write #PF -> emulation fail -> retry
7545 	 * instruction -> ...
7546 	 */
7547 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7548 
7549 	/*
7550 	 * If the instruction failed on the error pfn, it can not be fixed,
7551 	 * report the error to userspace.
7552 	 */
7553 	if (is_error_noslot_pfn(pfn))
7554 		return false;
7555 
7556 	kvm_release_pfn_clean(pfn);
7557 
7558 	/* The instructions are well-emulated on direct mmu. */
7559 	if (vcpu->arch.mmu->direct_map) {
7560 		unsigned int indirect_shadow_pages;
7561 
7562 		write_lock(&vcpu->kvm->mmu_lock);
7563 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7564 		write_unlock(&vcpu->kvm->mmu_lock);
7565 
7566 		if (indirect_shadow_pages)
7567 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7568 
7569 		return true;
7570 	}
7571 
7572 	/*
7573 	 * if emulation was due to access to shadowed page table
7574 	 * and it failed try to unshadow page and re-enter the
7575 	 * guest to let CPU execute the instruction.
7576 	 */
7577 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7578 
7579 	/*
7580 	 * If the access faults on its page table, it can not
7581 	 * be fixed by unprotecting shadow page and it should
7582 	 * be reported to userspace.
7583 	 */
7584 	return !write_fault_to_shadow_pgtable;
7585 }
7586 
7587 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7588 			      gpa_t cr2_or_gpa,  int emulation_type)
7589 {
7590 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7591 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7592 
7593 	last_retry_eip = vcpu->arch.last_retry_eip;
7594 	last_retry_addr = vcpu->arch.last_retry_addr;
7595 
7596 	/*
7597 	 * If the emulation is caused by #PF and it is non-page_table
7598 	 * writing instruction, it means the VM-EXIT is caused by shadow
7599 	 * page protected, we can zap the shadow page and retry this
7600 	 * instruction directly.
7601 	 *
7602 	 * Note: if the guest uses a non-page-table modifying instruction
7603 	 * on the PDE that points to the instruction, then we will unmap
7604 	 * the instruction and go to an infinite loop. So, we cache the
7605 	 * last retried eip and the last fault address, if we meet the eip
7606 	 * and the address again, we can break out of the potential infinite
7607 	 * loop.
7608 	 */
7609 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7610 
7611 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7612 		return false;
7613 
7614 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7615 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7616 		return false;
7617 
7618 	if (x86_page_table_writing_insn(ctxt))
7619 		return false;
7620 
7621 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7622 		return false;
7623 
7624 	vcpu->arch.last_retry_eip = ctxt->eip;
7625 	vcpu->arch.last_retry_addr = cr2_or_gpa;
7626 
7627 	if (!vcpu->arch.mmu->direct_map)
7628 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7629 
7630 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7631 
7632 	return true;
7633 }
7634 
7635 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7636 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7637 
7638 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm)
7639 {
7640 	trace_kvm_smm_transition(vcpu->vcpu_id, vcpu->arch.smbase, entering_smm);
7641 
7642 	if (entering_smm) {
7643 		vcpu->arch.hflags |= HF_SMM_MASK;
7644 	} else {
7645 		vcpu->arch.hflags &= ~(HF_SMM_MASK | HF_SMM_INSIDE_NMI_MASK);
7646 
7647 		/* Process a latched INIT or SMI, if any.  */
7648 		kvm_make_request(KVM_REQ_EVENT, vcpu);
7649 	}
7650 
7651 	kvm_mmu_reset_context(vcpu);
7652 }
7653 
7654 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7655 				unsigned long *db)
7656 {
7657 	u32 dr6 = 0;
7658 	int i;
7659 	u32 enable, rwlen;
7660 
7661 	enable = dr7;
7662 	rwlen = dr7 >> 16;
7663 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7664 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7665 			dr6 |= (1 << i);
7666 	return dr6;
7667 }
7668 
7669 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7670 {
7671 	struct kvm_run *kvm_run = vcpu->run;
7672 
7673 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7674 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7675 		kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7676 		kvm_run->debug.arch.exception = DB_VECTOR;
7677 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
7678 		return 0;
7679 	}
7680 	kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7681 	return 1;
7682 }
7683 
7684 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7685 {
7686 	unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7687 	int r;
7688 
7689 	r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7690 	if (unlikely(!r))
7691 		return 0;
7692 
7693 	/*
7694 	 * rflags is the old, "raw" value of the flags.  The new value has
7695 	 * not been saved yet.
7696 	 *
7697 	 * This is correct even for TF set by the guest, because "the
7698 	 * processor will not generate this exception after the instruction
7699 	 * that sets the TF flag".
7700 	 */
7701 	if (unlikely(rflags & X86_EFLAGS_TF))
7702 		r = kvm_vcpu_do_singlestep(vcpu);
7703 	return r;
7704 }
7705 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7706 
7707 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7708 {
7709 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7710 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7711 		struct kvm_run *kvm_run = vcpu->run;
7712 		unsigned long eip = kvm_get_linear_rip(vcpu);
7713 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7714 					   vcpu->arch.guest_debug_dr7,
7715 					   vcpu->arch.eff_db);
7716 
7717 		if (dr6 != 0) {
7718 			kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
7719 			kvm_run->debug.arch.pc = eip;
7720 			kvm_run->debug.arch.exception = DB_VECTOR;
7721 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
7722 			*r = 0;
7723 			return true;
7724 		}
7725 	}
7726 
7727 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7728 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7729 		unsigned long eip = kvm_get_linear_rip(vcpu);
7730 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7731 					   vcpu->arch.dr7,
7732 					   vcpu->arch.db);
7733 
7734 		if (dr6 != 0) {
7735 			kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7736 			*r = 1;
7737 			return true;
7738 		}
7739 	}
7740 
7741 	return false;
7742 }
7743 
7744 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7745 {
7746 	switch (ctxt->opcode_len) {
7747 	case 1:
7748 		switch (ctxt->b) {
7749 		case 0xe4:	/* IN */
7750 		case 0xe5:
7751 		case 0xec:
7752 		case 0xed:
7753 		case 0xe6:	/* OUT */
7754 		case 0xe7:
7755 		case 0xee:
7756 		case 0xef:
7757 		case 0x6c:	/* INS */
7758 		case 0x6d:
7759 		case 0x6e:	/* OUTS */
7760 		case 0x6f:
7761 			return true;
7762 		}
7763 		break;
7764 	case 2:
7765 		switch (ctxt->b) {
7766 		case 0x33:	/* RDPMC */
7767 			return true;
7768 		}
7769 		break;
7770 	}
7771 
7772 	return false;
7773 }
7774 
7775 /*
7776  * Decode to be emulated instruction. Return EMULATION_OK if success.
7777  */
7778 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
7779 				    void *insn, int insn_len)
7780 {
7781 	int r = EMULATION_OK;
7782 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7783 
7784 	init_emulate_ctxt(vcpu);
7785 
7786 	/*
7787 	 * We will reenter on the same instruction since we do not set
7788 	 * complete_userspace_io. This does not handle watchpoints yet,
7789 	 * those would be handled in the emulate_ops.
7790 	 */
7791 	if (!(emulation_type & EMULTYPE_SKIP) &&
7792 	    kvm_vcpu_check_breakpoint(vcpu, &r))
7793 		return r;
7794 
7795 	r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
7796 
7797 	trace_kvm_emulate_insn_start(vcpu);
7798 	++vcpu->stat.insn_emulation;
7799 
7800 	return r;
7801 }
7802 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
7803 
7804 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7805 			    int emulation_type, void *insn, int insn_len)
7806 {
7807 	int r;
7808 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7809 	bool writeback = true;
7810 	bool write_fault_to_spt;
7811 
7812 	if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
7813 		return 1;
7814 
7815 	vcpu->arch.l1tf_flush_l1d = true;
7816 
7817 	/*
7818 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
7819 	 * never reused.
7820 	 */
7821 	write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7822 	vcpu->arch.write_fault_to_shadow_pgtable = false;
7823 
7824 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7825 		kvm_clear_exception_queue(vcpu);
7826 
7827 		r = x86_decode_emulated_instruction(vcpu, emulation_type,
7828 						    insn, insn_len);
7829 		if (r != EMULATION_OK)  {
7830 			if ((emulation_type & EMULTYPE_TRAP_UD) ||
7831 			    (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7832 				kvm_queue_exception(vcpu, UD_VECTOR);
7833 				return 1;
7834 			}
7835 			if (reexecute_instruction(vcpu, cr2_or_gpa,
7836 						  write_fault_to_spt,
7837 						  emulation_type))
7838 				return 1;
7839 			if (ctxt->have_exception) {
7840 				/*
7841 				 * #UD should result in just EMULATION_FAILED, and trap-like
7842 				 * exception should not be encountered during decode.
7843 				 */
7844 				WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7845 					     exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7846 				inject_emulated_exception(vcpu);
7847 				return 1;
7848 			}
7849 			return handle_emulation_failure(vcpu, emulation_type);
7850 		}
7851 	}
7852 
7853 	if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7854 	    !is_vmware_backdoor_opcode(ctxt)) {
7855 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7856 		return 1;
7857 	}
7858 
7859 	/*
7860 	 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7861 	 * for kvm_skip_emulated_instruction().  The caller is responsible for
7862 	 * updating interruptibility state and injecting single-step #DBs.
7863 	 */
7864 	if (emulation_type & EMULTYPE_SKIP) {
7865 		kvm_rip_write(vcpu, ctxt->_eip);
7866 		if (ctxt->eflags & X86_EFLAGS_RF)
7867 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7868 		return 1;
7869 	}
7870 
7871 	if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7872 		return 1;
7873 
7874 	/* this is needed for vmware backdoor interface to work since it
7875 	   changes registers values  during IO operation */
7876 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7877 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7878 		emulator_invalidate_register_cache(ctxt);
7879 	}
7880 
7881 restart:
7882 	if (emulation_type & EMULTYPE_PF) {
7883 		/* Save the faulting GPA (cr2) in the address field */
7884 		ctxt->exception.address = cr2_or_gpa;
7885 
7886 		/* With shadow page tables, cr2 contains a GVA or nGPA. */
7887 		if (vcpu->arch.mmu->direct_map) {
7888 			ctxt->gpa_available = true;
7889 			ctxt->gpa_val = cr2_or_gpa;
7890 		}
7891 	} else {
7892 		/* Sanitize the address out of an abundance of paranoia. */
7893 		ctxt->exception.address = 0;
7894 	}
7895 
7896 	r = x86_emulate_insn(ctxt);
7897 
7898 	if (r == EMULATION_INTERCEPTED)
7899 		return 1;
7900 
7901 	if (r == EMULATION_FAILED) {
7902 		if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7903 					emulation_type))
7904 			return 1;
7905 
7906 		return handle_emulation_failure(vcpu, emulation_type);
7907 	}
7908 
7909 	if (ctxt->have_exception) {
7910 		r = 1;
7911 		if (inject_emulated_exception(vcpu))
7912 			return r;
7913 	} else if (vcpu->arch.pio.count) {
7914 		if (!vcpu->arch.pio.in) {
7915 			/* FIXME: return into emulator if single-stepping.  */
7916 			vcpu->arch.pio.count = 0;
7917 		} else {
7918 			writeback = false;
7919 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
7920 		}
7921 		r = 0;
7922 	} else if (vcpu->mmio_needed) {
7923 		++vcpu->stat.mmio_exits;
7924 
7925 		if (!vcpu->mmio_is_write)
7926 			writeback = false;
7927 		r = 0;
7928 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7929 	} else if (r == EMULATION_RESTART)
7930 		goto restart;
7931 	else
7932 		r = 1;
7933 
7934 	if (writeback) {
7935 		unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7936 		toggle_interruptibility(vcpu, ctxt->interruptibility);
7937 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7938 		if (!ctxt->have_exception ||
7939 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7940 			kvm_rip_write(vcpu, ctxt->eip);
7941 			if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7942 				r = kvm_vcpu_do_singlestep(vcpu);
7943 			if (kvm_x86_ops.update_emulated_instruction)
7944 				static_call(kvm_x86_update_emulated_instruction)(vcpu);
7945 			__kvm_set_rflags(vcpu, ctxt->eflags);
7946 		}
7947 
7948 		/*
7949 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7950 		 * do nothing, and it will be requested again as soon as
7951 		 * the shadow expires.  But we still need to check here,
7952 		 * because POPF has no interrupt shadow.
7953 		 */
7954 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7955 			kvm_make_request(KVM_REQ_EVENT, vcpu);
7956 	} else
7957 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7958 
7959 	return r;
7960 }
7961 
7962 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7963 {
7964 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7965 }
7966 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7967 
7968 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7969 					void *insn, int insn_len)
7970 {
7971 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7972 }
7973 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7974 
7975 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7976 {
7977 	vcpu->arch.pio.count = 0;
7978 	return 1;
7979 }
7980 
7981 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7982 {
7983 	vcpu->arch.pio.count = 0;
7984 
7985 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7986 		return 1;
7987 
7988 	return kvm_skip_emulated_instruction(vcpu);
7989 }
7990 
7991 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7992 			    unsigned short port)
7993 {
7994 	unsigned long val = kvm_rax_read(vcpu);
7995 	int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7996 
7997 	if (ret)
7998 		return ret;
7999 
8000 	/*
8001 	 * Workaround userspace that relies on old KVM behavior of %rip being
8002 	 * incremented prior to exiting to userspace to handle "OUT 0x7e".
8003 	 */
8004 	if (port == 0x7e &&
8005 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
8006 		vcpu->arch.complete_userspace_io =
8007 			complete_fast_pio_out_port_0x7e;
8008 		kvm_skip_emulated_instruction(vcpu);
8009 	} else {
8010 		vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8011 		vcpu->arch.complete_userspace_io = complete_fast_pio_out;
8012 	}
8013 	return 0;
8014 }
8015 
8016 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
8017 {
8018 	unsigned long val;
8019 
8020 	/* We should only ever be called with arch.pio.count equal to 1 */
8021 	BUG_ON(vcpu->arch.pio.count != 1);
8022 
8023 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
8024 		vcpu->arch.pio.count = 0;
8025 		return 1;
8026 	}
8027 
8028 	/* For size less than 4 we merge, else we zero extend */
8029 	val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
8030 
8031 	/*
8032 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
8033 	 * the copy and tracing
8034 	 */
8035 	emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
8036 	kvm_rax_write(vcpu, val);
8037 
8038 	return kvm_skip_emulated_instruction(vcpu);
8039 }
8040 
8041 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
8042 			   unsigned short port)
8043 {
8044 	unsigned long val;
8045 	int ret;
8046 
8047 	/* For size less than 4 we merge, else we zero extend */
8048 	val = (size < 4) ? kvm_rax_read(vcpu) : 0;
8049 
8050 	ret = emulator_pio_in(vcpu, size, port, &val, 1);
8051 	if (ret) {
8052 		kvm_rax_write(vcpu, val);
8053 		return ret;
8054 	}
8055 
8056 	vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8057 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
8058 
8059 	return 0;
8060 }
8061 
8062 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
8063 {
8064 	int ret;
8065 
8066 	if (in)
8067 		ret = kvm_fast_pio_in(vcpu, size, port);
8068 	else
8069 		ret = kvm_fast_pio_out(vcpu, size, port);
8070 	return ret && kvm_skip_emulated_instruction(vcpu);
8071 }
8072 EXPORT_SYMBOL_GPL(kvm_fast_pio);
8073 
8074 static int kvmclock_cpu_down_prep(unsigned int cpu)
8075 {
8076 	__this_cpu_write(cpu_tsc_khz, 0);
8077 	return 0;
8078 }
8079 
8080 static void tsc_khz_changed(void *data)
8081 {
8082 	struct cpufreq_freqs *freq = data;
8083 	unsigned long khz = 0;
8084 
8085 	if (data)
8086 		khz = freq->new;
8087 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8088 		khz = cpufreq_quick_get(raw_smp_processor_id());
8089 	if (!khz)
8090 		khz = tsc_khz;
8091 	__this_cpu_write(cpu_tsc_khz, khz);
8092 }
8093 
8094 #ifdef CONFIG_X86_64
8095 static void kvm_hyperv_tsc_notifier(void)
8096 {
8097 	struct kvm *kvm;
8098 	struct kvm_vcpu *vcpu;
8099 	int cpu;
8100 	unsigned long flags;
8101 
8102 	mutex_lock(&kvm_lock);
8103 	list_for_each_entry(kvm, &vm_list, vm_list)
8104 		kvm_make_mclock_inprogress_request(kvm);
8105 
8106 	hyperv_stop_tsc_emulation();
8107 
8108 	/* TSC frequency always matches when on Hyper-V */
8109 	for_each_present_cpu(cpu)
8110 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
8111 	kvm_max_guest_tsc_khz = tsc_khz;
8112 
8113 	list_for_each_entry(kvm, &vm_list, vm_list) {
8114 		struct kvm_arch *ka = &kvm->arch;
8115 
8116 		spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
8117 		pvclock_update_vm_gtod_copy(kvm);
8118 		spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
8119 
8120 		kvm_for_each_vcpu(cpu, vcpu, kvm)
8121 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8122 
8123 		kvm_for_each_vcpu(cpu, vcpu, kvm)
8124 			kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
8125 	}
8126 	mutex_unlock(&kvm_lock);
8127 }
8128 #endif
8129 
8130 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
8131 {
8132 	struct kvm *kvm;
8133 	struct kvm_vcpu *vcpu;
8134 	int i, send_ipi = 0;
8135 
8136 	/*
8137 	 * We allow guests to temporarily run on slowing clocks,
8138 	 * provided we notify them after, or to run on accelerating
8139 	 * clocks, provided we notify them before.  Thus time never
8140 	 * goes backwards.
8141 	 *
8142 	 * However, we have a problem.  We can't atomically update
8143 	 * the frequency of a given CPU from this function; it is
8144 	 * merely a notifier, which can be called from any CPU.
8145 	 * Changing the TSC frequency at arbitrary points in time
8146 	 * requires a recomputation of local variables related to
8147 	 * the TSC for each VCPU.  We must flag these local variables
8148 	 * to be updated and be sure the update takes place with the
8149 	 * new frequency before any guests proceed.
8150 	 *
8151 	 * Unfortunately, the combination of hotplug CPU and frequency
8152 	 * change creates an intractable locking scenario; the order
8153 	 * of when these callouts happen is undefined with respect to
8154 	 * CPU hotplug, and they can race with each other.  As such,
8155 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
8156 	 * undefined; you can actually have a CPU frequency change take
8157 	 * place in between the computation of X and the setting of the
8158 	 * variable.  To protect against this problem, all updates of
8159 	 * the per_cpu tsc_khz variable are done in an interrupt
8160 	 * protected IPI, and all callers wishing to update the value
8161 	 * must wait for a synchronous IPI to complete (which is trivial
8162 	 * if the caller is on the CPU already).  This establishes the
8163 	 * necessary total order on variable updates.
8164 	 *
8165 	 * Note that because a guest time update may take place
8166 	 * anytime after the setting of the VCPU's request bit, the
8167 	 * correct TSC value must be set before the request.  However,
8168 	 * to ensure the update actually makes it to any guest which
8169 	 * starts running in hardware virtualization between the set
8170 	 * and the acquisition of the spinlock, we must also ping the
8171 	 * CPU after setting the request bit.
8172 	 *
8173 	 */
8174 
8175 	smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8176 
8177 	mutex_lock(&kvm_lock);
8178 	list_for_each_entry(kvm, &vm_list, vm_list) {
8179 		kvm_for_each_vcpu(i, vcpu, kvm) {
8180 			if (vcpu->cpu != cpu)
8181 				continue;
8182 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8183 			if (vcpu->cpu != raw_smp_processor_id())
8184 				send_ipi = 1;
8185 		}
8186 	}
8187 	mutex_unlock(&kvm_lock);
8188 
8189 	if (freq->old < freq->new && send_ipi) {
8190 		/*
8191 		 * We upscale the frequency.  Must make the guest
8192 		 * doesn't see old kvmclock values while running with
8193 		 * the new frequency, otherwise we risk the guest sees
8194 		 * time go backwards.
8195 		 *
8196 		 * In case we update the frequency for another cpu
8197 		 * (which might be in guest context) send an interrupt
8198 		 * to kick the cpu out of guest context.  Next time
8199 		 * guest context is entered kvmclock will be updated,
8200 		 * so the guest will not see stale values.
8201 		 */
8202 		smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8203 	}
8204 }
8205 
8206 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
8207 				     void *data)
8208 {
8209 	struct cpufreq_freqs *freq = data;
8210 	int cpu;
8211 
8212 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
8213 		return 0;
8214 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
8215 		return 0;
8216 
8217 	for_each_cpu(cpu, freq->policy->cpus)
8218 		__kvmclock_cpufreq_notifier(freq, cpu);
8219 
8220 	return 0;
8221 }
8222 
8223 static struct notifier_block kvmclock_cpufreq_notifier_block = {
8224 	.notifier_call  = kvmclock_cpufreq_notifier
8225 };
8226 
8227 static int kvmclock_cpu_online(unsigned int cpu)
8228 {
8229 	tsc_khz_changed(NULL);
8230 	return 0;
8231 }
8232 
8233 static void kvm_timer_init(void)
8234 {
8235 	max_tsc_khz = tsc_khz;
8236 
8237 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
8238 #ifdef CONFIG_CPU_FREQ
8239 		struct cpufreq_policy *policy;
8240 		int cpu;
8241 
8242 		cpu = get_cpu();
8243 		policy = cpufreq_cpu_get(cpu);
8244 		if (policy) {
8245 			if (policy->cpuinfo.max_freq)
8246 				max_tsc_khz = policy->cpuinfo.max_freq;
8247 			cpufreq_cpu_put(policy);
8248 		}
8249 		put_cpu();
8250 #endif
8251 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
8252 					  CPUFREQ_TRANSITION_NOTIFIER);
8253 	}
8254 
8255 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
8256 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
8257 }
8258 
8259 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
8260 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
8261 
8262 int kvm_is_in_guest(void)
8263 {
8264 	return __this_cpu_read(current_vcpu) != NULL;
8265 }
8266 
8267 static int kvm_is_user_mode(void)
8268 {
8269 	int user_mode = 3;
8270 
8271 	if (__this_cpu_read(current_vcpu))
8272 		user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
8273 
8274 	return user_mode != 0;
8275 }
8276 
8277 static unsigned long kvm_get_guest_ip(void)
8278 {
8279 	unsigned long ip = 0;
8280 
8281 	if (__this_cpu_read(current_vcpu))
8282 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
8283 
8284 	return ip;
8285 }
8286 
8287 static void kvm_handle_intel_pt_intr(void)
8288 {
8289 	struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
8290 
8291 	kvm_make_request(KVM_REQ_PMI, vcpu);
8292 	__set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8293 			(unsigned long *)&vcpu->arch.pmu.global_status);
8294 }
8295 
8296 static struct perf_guest_info_callbacks kvm_guest_cbs = {
8297 	.is_in_guest		= kvm_is_in_guest,
8298 	.is_user_mode		= kvm_is_user_mode,
8299 	.get_guest_ip		= kvm_get_guest_ip,
8300 	.handle_intel_pt_intr	= kvm_handle_intel_pt_intr,
8301 };
8302 
8303 #ifdef CONFIG_X86_64
8304 static void pvclock_gtod_update_fn(struct work_struct *work)
8305 {
8306 	struct kvm *kvm;
8307 
8308 	struct kvm_vcpu *vcpu;
8309 	int i;
8310 
8311 	mutex_lock(&kvm_lock);
8312 	list_for_each_entry(kvm, &vm_list, vm_list)
8313 		kvm_for_each_vcpu(i, vcpu, kvm)
8314 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8315 	atomic_set(&kvm_guest_has_master_clock, 0);
8316 	mutex_unlock(&kvm_lock);
8317 }
8318 
8319 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8320 
8321 /*
8322  * Indirection to move queue_work() out of the tk_core.seq write held
8323  * region to prevent possible deadlocks against time accessors which
8324  * are invoked with work related locks held.
8325  */
8326 static void pvclock_irq_work_fn(struct irq_work *w)
8327 {
8328 	queue_work(system_long_wq, &pvclock_gtod_work);
8329 }
8330 
8331 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
8332 
8333 /*
8334  * Notification about pvclock gtod data update.
8335  */
8336 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8337 			       void *priv)
8338 {
8339 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8340 	struct timekeeper *tk = priv;
8341 
8342 	update_pvclock_gtod(tk);
8343 
8344 	/*
8345 	 * Disable master clock if host does not trust, or does not use,
8346 	 * TSC based clocksource. Delegate queue_work() to irq_work as
8347 	 * this is invoked with tk_core.seq write held.
8348 	 */
8349 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8350 	    atomic_read(&kvm_guest_has_master_clock) != 0)
8351 		irq_work_queue(&pvclock_irq_work);
8352 	return 0;
8353 }
8354 
8355 static struct notifier_block pvclock_gtod_notifier = {
8356 	.notifier_call = pvclock_gtod_notify,
8357 };
8358 #endif
8359 
8360 int kvm_arch_init(void *opaque)
8361 {
8362 	struct kvm_x86_init_ops *ops = opaque;
8363 	int r;
8364 
8365 	if (kvm_x86_ops.hardware_enable) {
8366 		printk(KERN_ERR "kvm: already loaded the other module\n");
8367 		r = -EEXIST;
8368 		goto out;
8369 	}
8370 
8371 	if (!ops->cpu_has_kvm_support()) {
8372 		pr_err_ratelimited("kvm: no hardware support\n");
8373 		r = -EOPNOTSUPP;
8374 		goto out;
8375 	}
8376 	if (ops->disabled_by_bios()) {
8377 		pr_err_ratelimited("kvm: disabled by bios\n");
8378 		r = -EOPNOTSUPP;
8379 		goto out;
8380 	}
8381 
8382 	/*
8383 	 * KVM explicitly assumes that the guest has an FPU and
8384 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8385 	 * vCPU's FPU state as a fxregs_state struct.
8386 	 */
8387 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8388 		printk(KERN_ERR "kvm: inadequate fpu\n");
8389 		r = -EOPNOTSUPP;
8390 		goto out;
8391 	}
8392 
8393 	r = -ENOMEM;
8394 	x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
8395 					  __alignof__(struct fpu), SLAB_ACCOUNT,
8396 					  NULL);
8397 	if (!x86_fpu_cache) {
8398 		printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
8399 		goto out;
8400 	}
8401 
8402 	x86_emulator_cache = kvm_alloc_emulator_cache();
8403 	if (!x86_emulator_cache) {
8404 		pr_err("kvm: failed to allocate cache for x86 emulator\n");
8405 		goto out_free_x86_fpu_cache;
8406 	}
8407 
8408 	user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8409 	if (!user_return_msrs) {
8410 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8411 		goto out_free_x86_emulator_cache;
8412 	}
8413 	kvm_nr_uret_msrs = 0;
8414 
8415 	r = kvm_mmu_module_init();
8416 	if (r)
8417 		goto out_free_percpu;
8418 
8419 	kvm_timer_init();
8420 
8421 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
8422 
8423 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8424 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8425 		supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8426 	}
8427 
8428 	if (pi_inject_timer == -1)
8429 		pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8430 #ifdef CONFIG_X86_64
8431 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8432 
8433 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8434 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8435 #endif
8436 
8437 	return 0;
8438 
8439 out_free_percpu:
8440 	free_percpu(user_return_msrs);
8441 out_free_x86_emulator_cache:
8442 	kmem_cache_destroy(x86_emulator_cache);
8443 out_free_x86_fpu_cache:
8444 	kmem_cache_destroy(x86_fpu_cache);
8445 out:
8446 	return r;
8447 }
8448 
8449 void kvm_arch_exit(void)
8450 {
8451 #ifdef CONFIG_X86_64
8452 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8453 		clear_hv_tscchange_cb();
8454 #endif
8455 	kvm_lapic_exit();
8456 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8457 
8458 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8459 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8460 					    CPUFREQ_TRANSITION_NOTIFIER);
8461 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8462 #ifdef CONFIG_X86_64
8463 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8464 	irq_work_sync(&pvclock_irq_work);
8465 	cancel_work_sync(&pvclock_gtod_work);
8466 #endif
8467 	kvm_x86_ops.hardware_enable = NULL;
8468 	kvm_mmu_module_exit();
8469 	free_percpu(user_return_msrs);
8470 	kmem_cache_destroy(x86_emulator_cache);
8471 	kmem_cache_destroy(x86_fpu_cache);
8472 #ifdef CONFIG_KVM_XEN
8473 	static_key_deferred_flush(&kvm_xen_enabled);
8474 	WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8475 #endif
8476 }
8477 
8478 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8479 {
8480 	++vcpu->stat.halt_exits;
8481 	if (lapic_in_kernel(vcpu)) {
8482 		vcpu->arch.mp_state = state;
8483 		return 1;
8484 	} else {
8485 		vcpu->run->exit_reason = reason;
8486 		return 0;
8487 	}
8488 }
8489 
8490 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8491 {
8492 	return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8493 }
8494 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8495 
8496 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8497 {
8498 	int ret = kvm_skip_emulated_instruction(vcpu);
8499 	/*
8500 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8501 	 * KVM_EXIT_DEBUG here.
8502 	 */
8503 	return kvm_vcpu_halt(vcpu) && ret;
8504 }
8505 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8506 
8507 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8508 {
8509 	int ret = kvm_skip_emulated_instruction(vcpu);
8510 
8511 	return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8512 }
8513 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8514 
8515 #ifdef CONFIG_X86_64
8516 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8517 			        unsigned long clock_type)
8518 {
8519 	struct kvm_clock_pairing clock_pairing;
8520 	struct timespec64 ts;
8521 	u64 cycle;
8522 	int ret;
8523 
8524 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8525 		return -KVM_EOPNOTSUPP;
8526 
8527 	if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8528 		return -KVM_EOPNOTSUPP;
8529 
8530 	clock_pairing.sec = ts.tv_sec;
8531 	clock_pairing.nsec = ts.tv_nsec;
8532 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8533 	clock_pairing.flags = 0;
8534 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8535 
8536 	ret = 0;
8537 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8538 			    sizeof(struct kvm_clock_pairing)))
8539 		ret = -KVM_EFAULT;
8540 
8541 	return ret;
8542 }
8543 #endif
8544 
8545 /*
8546  * kvm_pv_kick_cpu_op:  Kick a vcpu.
8547  *
8548  * @apicid - apicid of vcpu to be kicked.
8549  */
8550 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8551 {
8552 	struct kvm_lapic_irq lapic_irq;
8553 
8554 	lapic_irq.shorthand = APIC_DEST_NOSHORT;
8555 	lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8556 	lapic_irq.level = 0;
8557 	lapic_irq.dest_id = apicid;
8558 	lapic_irq.msi_redir_hint = false;
8559 
8560 	lapic_irq.delivery_mode = APIC_DM_REMRD;
8561 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8562 }
8563 
8564 bool kvm_apicv_activated(struct kvm *kvm)
8565 {
8566 	return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8567 }
8568 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8569 
8570 static void kvm_apicv_init(struct kvm *kvm)
8571 {
8572 	if (enable_apicv)
8573 		clear_bit(APICV_INHIBIT_REASON_DISABLE,
8574 			  &kvm->arch.apicv_inhibit_reasons);
8575 	else
8576 		set_bit(APICV_INHIBIT_REASON_DISABLE,
8577 			&kvm->arch.apicv_inhibit_reasons);
8578 }
8579 
8580 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8581 {
8582 	struct kvm_vcpu *target = NULL;
8583 	struct kvm_apic_map *map;
8584 
8585 	vcpu->stat.directed_yield_attempted++;
8586 
8587 	if (single_task_running())
8588 		goto no_yield;
8589 
8590 	rcu_read_lock();
8591 	map = rcu_dereference(vcpu->kvm->arch.apic_map);
8592 
8593 	if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8594 		target = map->phys_map[dest_id]->vcpu;
8595 
8596 	rcu_read_unlock();
8597 
8598 	if (!target || !READ_ONCE(target->ready))
8599 		goto no_yield;
8600 
8601 	/* Ignore requests to yield to self */
8602 	if (vcpu == target)
8603 		goto no_yield;
8604 
8605 	if (kvm_vcpu_yield_to(target) <= 0)
8606 		goto no_yield;
8607 
8608 	vcpu->stat.directed_yield_successful++;
8609 
8610 no_yield:
8611 	return;
8612 }
8613 
8614 static int complete_hypercall_exit(struct kvm_vcpu *vcpu)
8615 {
8616 	u64 ret = vcpu->run->hypercall.ret;
8617 
8618 	if (!is_64_bit_mode(vcpu))
8619 		ret = (u32)ret;
8620 	kvm_rax_write(vcpu, ret);
8621 	++vcpu->stat.hypercalls;
8622 	return kvm_skip_emulated_instruction(vcpu);
8623 }
8624 
8625 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8626 {
8627 	unsigned long nr, a0, a1, a2, a3, ret;
8628 	int op_64_bit;
8629 
8630 	if (kvm_xen_hypercall_enabled(vcpu->kvm))
8631 		return kvm_xen_hypercall(vcpu);
8632 
8633 	if (kvm_hv_hypercall_enabled(vcpu))
8634 		return kvm_hv_hypercall(vcpu);
8635 
8636 	nr = kvm_rax_read(vcpu);
8637 	a0 = kvm_rbx_read(vcpu);
8638 	a1 = kvm_rcx_read(vcpu);
8639 	a2 = kvm_rdx_read(vcpu);
8640 	a3 = kvm_rsi_read(vcpu);
8641 
8642 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
8643 
8644 	op_64_bit = is_64_bit_mode(vcpu);
8645 	if (!op_64_bit) {
8646 		nr &= 0xFFFFFFFF;
8647 		a0 &= 0xFFFFFFFF;
8648 		a1 &= 0xFFFFFFFF;
8649 		a2 &= 0xFFFFFFFF;
8650 		a3 &= 0xFFFFFFFF;
8651 	}
8652 
8653 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8654 		ret = -KVM_EPERM;
8655 		goto out;
8656 	}
8657 
8658 	ret = -KVM_ENOSYS;
8659 
8660 	switch (nr) {
8661 	case KVM_HC_VAPIC_POLL_IRQ:
8662 		ret = 0;
8663 		break;
8664 	case KVM_HC_KICK_CPU:
8665 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8666 			break;
8667 
8668 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8669 		kvm_sched_yield(vcpu, a1);
8670 		ret = 0;
8671 		break;
8672 #ifdef CONFIG_X86_64
8673 	case KVM_HC_CLOCK_PAIRING:
8674 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8675 		break;
8676 #endif
8677 	case KVM_HC_SEND_IPI:
8678 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8679 			break;
8680 
8681 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8682 		break;
8683 	case KVM_HC_SCHED_YIELD:
8684 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8685 			break;
8686 
8687 		kvm_sched_yield(vcpu, a0);
8688 		ret = 0;
8689 		break;
8690 	case KVM_HC_MAP_GPA_RANGE: {
8691 		u64 gpa = a0, npages = a1, attrs = a2;
8692 
8693 		ret = -KVM_ENOSYS;
8694 		if (!(vcpu->kvm->arch.hypercall_exit_enabled & (1 << KVM_HC_MAP_GPA_RANGE)))
8695 			break;
8696 
8697 		if (!PAGE_ALIGNED(gpa) || !npages ||
8698 		    gpa_to_gfn(gpa) + npages <= gpa_to_gfn(gpa)) {
8699 			ret = -KVM_EINVAL;
8700 			break;
8701 		}
8702 
8703 		vcpu->run->exit_reason        = KVM_EXIT_HYPERCALL;
8704 		vcpu->run->hypercall.nr       = KVM_HC_MAP_GPA_RANGE;
8705 		vcpu->run->hypercall.args[0]  = gpa;
8706 		vcpu->run->hypercall.args[1]  = npages;
8707 		vcpu->run->hypercall.args[2]  = attrs;
8708 		vcpu->run->hypercall.longmode = op_64_bit;
8709 		vcpu->arch.complete_userspace_io = complete_hypercall_exit;
8710 		return 0;
8711 	}
8712 	default:
8713 		ret = -KVM_ENOSYS;
8714 		break;
8715 	}
8716 out:
8717 	if (!op_64_bit)
8718 		ret = (u32)ret;
8719 	kvm_rax_write(vcpu, ret);
8720 
8721 	++vcpu->stat.hypercalls;
8722 	return kvm_skip_emulated_instruction(vcpu);
8723 }
8724 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8725 
8726 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8727 {
8728 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8729 	char instruction[3];
8730 	unsigned long rip = kvm_rip_read(vcpu);
8731 
8732 	static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
8733 
8734 	return emulator_write_emulated(ctxt, rip, instruction, 3,
8735 		&ctxt->exception);
8736 }
8737 
8738 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8739 {
8740 	return vcpu->run->request_interrupt_window &&
8741 		likely(!pic_in_kernel(vcpu->kvm));
8742 }
8743 
8744 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8745 {
8746 	struct kvm_run *kvm_run = vcpu->run;
8747 
8748 	/*
8749 	 * if_flag is obsolete and useless, so do not bother
8750 	 * setting it for SEV-ES guests.  Userspace can just
8751 	 * use kvm_run->ready_for_interrupt_injection.
8752 	 */
8753 	kvm_run->if_flag = !vcpu->arch.guest_state_protected
8754 		&& (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8755 
8756 	kvm_run->cr8 = kvm_get_cr8(vcpu);
8757 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
8758 	kvm_run->ready_for_interrupt_injection =
8759 		pic_in_kernel(vcpu->kvm) ||
8760 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
8761 
8762 	if (is_smm(vcpu))
8763 		kvm_run->flags |= KVM_RUN_X86_SMM;
8764 }
8765 
8766 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8767 {
8768 	int max_irr, tpr;
8769 
8770 	if (!kvm_x86_ops.update_cr8_intercept)
8771 		return;
8772 
8773 	if (!lapic_in_kernel(vcpu))
8774 		return;
8775 
8776 	if (vcpu->arch.apicv_active)
8777 		return;
8778 
8779 	if (!vcpu->arch.apic->vapic_addr)
8780 		max_irr = kvm_lapic_find_highest_irr(vcpu);
8781 	else
8782 		max_irr = -1;
8783 
8784 	if (max_irr != -1)
8785 		max_irr >>= 4;
8786 
8787 	tpr = kvm_lapic_get_cr8(vcpu);
8788 
8789 	static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
8790 }
8791 
8792 
8793 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
8794 {
8795 	if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8796 		kvm_x86_ops.nested_ops->triple_fault(vcpu);
8797 		return 1;
8798 	}
8799 
8800 	return kvm_x86_ops.nested_ops->check_events(vcpu);
8801 }
8802 
8803 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
8804 {
8805 	if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
8806 		vcpu->arch.exception.error_code = false;
8807 	static_call(kvm_x86_queue_exception)(vcpu);
8808 }
8809 
8810 static int inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8811 {
8812 	int r;
8813 	bool can_inject = true;
8814 
8815 	/* try to reinject previous events if any */
8816 
8817 	if (vcpu->arch.exception.injected) {
8818 		kvm_inject_exception(vcpu);
8819 		can_inject = false;
8820 	}
8821 	/*
8822 	 * Do not inject an NMI or interrupt if there is a pending
8823 	 * exception.  Exceptions and interrupts are recognized at
8824 	 * instruction boundaries, i.e. the start of an instruction.
8825 	 * Trap-like exceptions, e.g. #DB, have higher priority than
8826 	 * NMIs and interrupts, i.e. traps are recognized before an
8827 	 * NMI/interrupt that's pending on the same instruction.
8828 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8829 	 * priority, but are only generated (pended) during instruction
8830 	 * execution, i.e. a pending fault-like exception means the
8831 	 * fault occurred on the *previous* instruction and must be
8832 	 * serviced prior to recognizing any new events in order to
8833 	 * fully complete the previous instruction.
8834 	 */
8835 	else if (!vcpu->arch.exception.pending) {
8836 		if (vcpu->arch.nmi_injected) {
8837 			static_call(kvm_x86_set_nmi)(vcpu);
8838 			can_inject = false;
8839 		} else if (vcpu->arch.interrupt.injected) {
8840 			static_call(kvm_x86_set_irq)(vcpu);
8841 			can_inject = false;
8842 		}
8843 	}
8844 
8845 	WARN_ON_ONCE(vcpu->arch.exception.injected &&
8846 		     vcpu->arch.exception.pending);
8847 
8848 	/*
8849 	 * Call check_nested_events() even if we reinjected a previous event
8850 	 * in order for caller to determine if it should require immediate-exit
8851 	 * from L2 to L1 due to pending L1 events which require exit
8852 	 * from L2 to L1.
8853 	 */
8854 	if (is_guest_mode(vcpu)) {
8855 		r = kvm_check_nested_events(vcpu);
8856 		if (r < 0)
8857 			goto out;
8858 	}
8859 
8860 	/* try to inject new event if pending */
8861 	if (vcpu->arch.exception.pending) {
8862 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
8863 					vcpu->arch.exception.has_error_code,
8864 					vcpu->arch.exception.error_code);
8865 
8866 		vcpu->arch.exception.pending = false;
8867 		vcpu->arch.exception.injected = true;
8868 
8869 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8870 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8871 					     X86_EFLAGS_RF);
8872 
8873 		if (vcpu->arch.exception.nr == DB_VECTOR) {
8874 			kvm_deliver_exception_payload(vcpu);
8875 			if (vcpu->arch.dr7 & DR7_GD) {
8876 				vcpu->arch.dr7 &= ~DR7_GD;
8877 				kvm_update_dr7(vcpu);
8878 			}
8879 		}
8880 
8881 		kvm_inject_exception(vcpu);
8882 		can_inject = false;
8883 	}
8884 
8885 	/*
8886 	 * Finally, inject interrupt events.  If an event cannot be injected
8887 	 * due to architectural conditions (e.g. IF=0) a window-open exit
8888 	 * will re-request KVM_REQ_EVENT.  Sometimes however an event is pending
8889 	 * and can architecturally be injected, but we cannot do it right now:
8890 	 * an interrupt could have arrived just now and we have to inject it
8891 	 * as a vmexit, or there could already an event in the queue, which is
8892 	 * indicated by can_inject.  In that case we request an immediate exit
8893 	 * in order to make progress and get back here for another iteration.
8894 	 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8895 	 */
8896 	if (vcpu->arch.smi_pending) {
8897 		r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
8898 		if (r < 0)
8899 			goto out;
8900 		if (r) {
8901 			vcpu->arch.smi_pending = false;
8902 			++vcpu->arch.smi_count;
8903 			enter_smm(vcpu);
8904 			can_inject = false;
8905 		} else
8906 			static_call(kvm_x86_enable_smi_window)(vcpu);
8907 	}
8908 
8909 	if (vcpu->arch.nmi_pending) {
8910 		r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
8911 		if (r < 0)
8912 			goto out;
8913 		if (r) {
8914 			--vcpu->arch.nmi_pending;
8915 			vcpu->arch.nmi_injected = true;
8916 			static_call(kvm_x86_set_nmi)(vcpu);
8917 			can_inject = false;
8918 			WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
8919 		}
8920 		if (vcpu->arch.nmi_pending)
8921 			static_call(kvm_x86_enable_nmi_window)(vcpu);
8922 	}
8923 
8924 	if (kvm_cpu_has_injectable_intr(vcpu)) {
8925 		r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
8926 		if (r < 0)
8927 			goto out;
8928 		if (r) {
8929 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8930 			static_call(kvm_x86_set_irq)(vcpu);
8931 			WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
8932 		}
8933 		if (kvm_cpu_has_injectable_intr(vcpu))
8934 			static_call(kvm_x86_enable_irq_window)(vcpu);
8935 	}
8936 
8937 	if (is_guest_mode(vcpu) &&
8938 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
8939 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8940 		*req_immediate_exit = true;
8941 
8942 	WARN_ON(vcpu->arch.exception.pending);
8943 	return 0;
8944 
8945 out:
8946 	if (r == -EBUSY) {
8947 		*req_immediate_exit = true;
8948 		r = 0;
8949 	}
8950 	return r;
8951 }
8952 
8953 static void process_nmi(struct kvm_vcpu *vcpu)
8954 {
8955 	unsigned limit = 2;
8956 
8957 	/*
8958 	 * x86 is limited to one NMI running, and one NMI pending after it.
8959 	 * If an NMI is already in progress, limit further NMIs to just one.
8960 	 * Otherwise, allow two (and we'll inject the first one immediately).
8961 	 */
8962 	if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
8963 		limit = 1;
8964 
8965 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8966 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8967 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8968 }
8969 
8970 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8971 {
8972 	u32 flags = 0;
8973 	flags |= seg->g       << 23;
8974 	flags |= seg->db      << 22;
8975 	flags |= seg->l       << 21;
8976 	flags |= seg->avl     << 20;
8977 	flags |= seg->present << 15;
8978 	flags |= seg->dpl     << 13;
8979 	flags |= seg->s       << 12;
8980 	flags |= seg->type    << 8;
8981 	return flags;
8982 }
8983 
8984 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8985 {
8986 	struct kvm_segment seg;
8987 	int offset;
8988 
8989 	kvm_get_segment(vcpu, &seg, n);
8990 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
8991 
8992 	if (n < 3)
8993 		offset = 0x7f84 + n * 12;
8994 	else
8995 		offset = 0x7f2c + (n - 3) * 12;
8996 
8997 	put_smstate(u32, buf, offset + 8, seg.base);
8998 	put_smstate(u32, buf, offset + 4, seg.limit);
8999 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
9000 }
9001 
9002 #ifdef CONFIG_X86_64
9003 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
9004 {
9005 	struct kvm_segment seg;
9006 	int offset;
9007 	u16 flags;
9008 
9009 	kvm_get_segment(vcpu, &seg, n);
9010 	offset = 0x7e00 + n * 16;
9011 
9012 	flags = enter_smm_get_segment_flags(&seg) >> 8;
9013 	put_smstate(u16, buf, offset, seg.selector);
9014 	put_smstate(u16, buf, offset + 2, flags);
9015 	put_smstate(u32, buf, offset + 4, seg.limit);
9016 	put_smstate(u64, buf, offset + 8, seg.base);
9017 }
9018 #endif
9019 
9020 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
9021 {
9022 	struct desc_ptr dt;
9023 	struct kvm_segment seg;
9024 	unsigned long val;
9025 	int i;
9026 
9027 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
9028 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
9029 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
9030 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
9031 
9032 	for (i = 0; i < 8; i++)
9033 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
9034 
9035 	kvm_get_dr(vcpu, 6, &val);
9036 	put_smstate(u32, buf, 0x7fcc, (u32)val);
9037 	kvm_get_dr(vcpu, 7, &val);
9038 	put_smstate(u32, buf, 0x7fc8, (u32)val);
9039 
9040 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9041 	put_smstate(u32, buf, 0x7fc4, seg.selector);
9042 	put_smstate(u32, buf, 0x7f64, seg.base);
9043 	put_smstate(u32, buf, 0x7f60, seg.limit);
9044 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
9045 
9046 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9047 	put_smstate(u32, buf, 0x7fc0, seg.selector);
9048 	put_smstate(u32, buf, 0x7f80, seg.base);
9049 	put_smstate(u32, buf, 0x7f7c, seg.limit);
9050 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
9051 
9052 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
9053 	put_smstate(u32, buf, 0x7f74, dt.address);
9054 	put_smstate(u32, buf, 0x7f70, dt.size);
9055 
9056 	static_call(kvm_x86_get_idt)(vcpu, &dt);
9057 	put_smstate(u32, buf, 0x7f58, dt.address);
9058 	put_smstate(u32, buf, 0x7f54, dt.size);
9059 
9060 	for (i = 0; i < 6; i++)
9061 		enter_smm_save_seg_32(vcpu, buf, i);
9062 
9063 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
9064 
9065 	/* revision id */
9066 	put_smstate(u32, buf, 0x7efc, 0x00020000);
9067 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
9068 }
9069 
9070 #ifdef CONFIG_X86_64
9071 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
9072 {
9073 	struct desc_ptr dt;
9074 	struct kvm_segment seg;
9075 	unsigned long val;
9076 	int i;
9077 
9078 	for (i = 0; i < 16; i++)
9079 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
9080 
9081 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
9082 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
9083 
9084 	kvm_get_dr(vcpu, 6, &val);
9085 	put_smstate(u64, buf, 0x7f68, val);
9086 	kvm_get_dr(vcpu, 7, &val);
9087 	put_smstate(u64, buf, 0x7f60, val);
9088 
9089 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
9090 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
9091 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
9092 
9093 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
9094 
9095 	/* revision id */
9096 	put_smstate(u32, buf, 0x7efc, 0x00020064);
9097 
9098 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
9099 
9100 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9101 	put_smstate(u16, buf, 0x7e90, seg.selector);
9102 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
9103 	put_smstate(u32, buf, 0x7e94, seg.limit);
9104 	put_smstate(u64, buf, 0x7e98, seg.base);
9105 
9106 	static_call(kvm_x86_get_idt)(vcpu, &dt);
9107 	put_smstate(u32, buf, 0x7e84, dt.size);
9108 	put_smstate(u64, buf, 0x7e88, dt.address);
9109 
9110 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9111 	put_smstate(u16, buf, 0x7e70, seg.selector);
9112 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
9113 	put_smstate(u32, buf, 0x7e74, seg.limit);
9114 	put_smstate(u64, buf, 0x7e78, seg.base);
9115 
9116 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
9117 	put_smstate(u32, buf, 0x7e64, dt.size);
9118 	put_smstate(u64, buf, 0x7e68, dt.address);
9119 
9120 	for (i = 0; i < 6; i++)
9121 		enter_smm_save_seg_64(vcpu, buf, i);
9122 }
9123 #endif
9124 
9125 static void enter_smm(struct kvm_vcpu *vcpu)
9126 {
9127 	struct kvm_segment cs, ds;
9128 	struct desc_ptr dt;
9129 	unsigned long cr0;
9130 	char buf[512];
9131 
9132 	memset(buf, 0, 512);
9133 #ifdef CONFIG_X86_64
9134 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9135 		enter_smm_save_state_64(vcpu, buf);
9136 	else
9137 #endif
9138 		enter_smm_save_state_32(vcpu, buf);
9139 
9140 	/*
9141 	 * Give enter_smm() a chance to make ISA-specific changes to the vCPU
9142 	 * state (e.g. leave guest mode) after we've saved the state into the
9143 	 * SMM state-save area.
9144 	 */
9145 	static_call(kvm_x86_enter_smm)(vcpu, buf);
9146 
9147 	kvm_smm_changed(vcpu, true);
9148 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
9149 
9150 	if (static_call(kvm_x86_get_nmi_mask)(vcpu))
9151 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
9152 	else
9153 		static_call(kvm_x86_set_nmi_mask)(vcpu, true);
9154 
9155 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
9156 	kvm_rip_write(vcpu, 0x8000);
9157 
9158 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
9159 	static_call(kvm_x86_set_cr0)(vcpu, cr0);
9160 	vcpu->arch.cr0 = cr0;
9161 
9162 	static_call(kvm_x86_set_cr4)(vcpu, 0);
9163 
9164 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
9165 	dt.address = dt.size = 0;
9166 	static_call(kvm_x86_set_idt)(vcpu, &dt);
9167 
9168 	kvm_set_dr(vcpu, 7, DR7_FIXED_1);
9169 
9170 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
9171 	cs.base = vcpu->arch.smbase;
9172 
9173 	ds.selector = 0;
9174 	ds.base = 0;
9175 
9176 	cs.limit    = ds.limit = 0xffffffff;
9177 	cs.type     = ds.type = 0x3;
9178 	cs.dpl      = ds.dpl = 0;
9179 	cs.db       = ds.db = 0;
9180 	cs.s        = ds.s = 1;
9181 	cs.l        = ds.l = 0;
9182 	cs.g        = ds.g = 1;
9183 	cs.avl      = ds.avl = 0;
9184 	cs.present  = ds.present = 1;
9185 	cs.unusable = ds.unusable = 0;
9186 	cs.padding  = ds.padding = 0;
9187 
9188 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9189 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
9190 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
9191 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
9192 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
9193 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
9194 
9195 #ifdef CONFIG_X86_64
9196 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9197 		static_call(kvm_x86_set_efer)(vcpu, 0);
9198 #endif
9199 
9200 	kvm_update_cpuid_runtime(vcpu);
9201 	kvm_mmu_reset_context(vcpu);
9202 }
9203 
9204 static void process_smi(struct kvm_vcpu *vcpu)
9205 {
9206 	vcpu->arch.smi_pending = true;
9207 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9208 }
9209 
9210 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
9211 				       unsigned long *vcpu_bitmap)
9212 {
9213 	cpumask_var_t cpus;
9214 
9215 	zalloc_cpumask_var(&cpus, GFP_ATOMIC);
9216 
9217 	kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
9218 				    NULL, vcpu_bitmap, cpus);
9219 
9220 	free_cpumask_var(cpus);
9221 }
9222 
9223 void kvm_make_scan_ioapic_request(struct kvm *kvm)
9224 {
9225 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
9226 }
9227 
9228 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
9229 {
9230 	if (!lapic_in_kernel(vcpu))
9231 		return;
9232 
9233 	vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
9234 	kvm_apic_update_apicv(vcpu);
9235 	static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
9236 
9237 	/*
9238 	 * When APICv gets disabled, we may still have injected interrupts
9239 	 * pending. At the same time, KVM_REQ_EVENT may not be set as APICv was
9240 	 * still active when the interrupt got accepted. Make sure
9241 	 * inject_pending_event() is called to check for that.
9242 	 */
9243 	if (!vcpu->arch.apicv_active)
9244 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9245 }
9246 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
9247 
9248 /*
9249  * NOTE: Do not hold any lock prior to calling this.
9250  *
9251  * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
9252  * locked, because it calls __x86_set_memory_region() which does
9253  * synchronize_srcu(&kvm->srcu).
9254  */
9255 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9256 {
9257 	struct kvm_vcpu *except;
9258 	unsigned long old, new, expected;
9259 
9260 	if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
9261 	    !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
9262 		return;
9263 
9264 	old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
9265 	do {
9266 		expected = new = old;
9267 		if (activate)
9268 			__clear_bit(bit, &new);
9269 		else
9270 			__set_bit(bit, &new);
9271 		if (new == old)
9272 			break;
9273 		old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
9274 	} while (old != expected);
9275 
9276 	if (!!old == !!new)
9277 		return;
9278 
9279 	trace_kvm_apicv_update_request(activate, bit);
9280 	if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
9281 		static_call(kvm_x86_pre_update_apicv_exec_ctrl)(kvm, activate);
9282 
9283 	/*
9284 	 * Sending request to update APICV for all other vcpus,
9285 	 * while update the calling vcpu immediately instead of
9286 	 * waiting for another #VMEXIT to handle the request.
9287 	 */
9288 	except = kvm_get_running_vcpu();
9289 	kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
9290 					 except);
9291 	if (except)
9292 		kvm_vcpu_update_apicv(except);
9293 }
9294 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
9295 
9296 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
9297 {
9298 	if (!kvm_apic_present(vcpu))
9299 		return;
9300 
9301 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
9302 
9303 	if (irqchip_split(vcpu->kvm))
9304 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
9305 	else {
9306 		if (vcpu->arch.apicv_active)
9307 			static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9308 		if (ioapic_in_kernel(vcpu->kvm))
9309 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
9310 	}
9311 
9312 	if (is_guest_mode(vcpu))
9313 		vcpu->arch.load_eoi_exitmap_pending = true;
9314 	else
9315 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
9316 }
9317 
9318 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
9319 {
9320 	u64 eoi_exit_bitmap[4];
9321 
9322 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
9323 		return;
9324 
9325 	if (to_hv_vcpu(vcpu))
9326 		bitmap_or((ulong *)eoi_exit_bitmap,
9327 			  vcpu->arch.ioapic_handled_vectors,
9328 			  to_hv_synic(vcpu)->vec_bitmap, 256);
9329 
9330 	static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
9331 }
9332 
9333 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
9334 					    unsigned long start, unsigned long end)
9335 {
9336 	unsigned long apic_address;
9337 
9338 	/*
9339 	 * The physical address of apic access page is stored in the VMCS.
9340 	 * Update it when it becomes invalid.
9341 	 */
9342 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9343 	if (start <= apic_address && apic_address < end)
9344 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9345 }
9346 
9347 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9348 {
9349 	if (!lapic_in_kernel(vcpu))
9350 		return;
9351 
9352 	if (!kvm_x86_ops.set_apic_access_page_addr)
9353 		return;
9354 
9355 	static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9356 }
9357 
9358 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9359 {
9360 	smp_send_reschedule(vcpu->cpu);
9361 }
9362 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9363 
9364 /*
9365  * Returns 1 to let vcpu_run() continue the guest execution loop without
9366  * exiting to the userspace.  Otherwise, the value will be returned to the
9367  * userspace.
9368  */
9369 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9370 {
9371 	int r;
9372 	bool req_int_win =
9373 		dm_request_for_irq_injection(vcpu) &&
9374 		kvm_cpu_accept_dm_intr(vcpu);
9375 	fastpath_t exit_fastpath;
9376 
9377 	bool req_immediate_exit = false;
9378 
9379 	/* Forbid vmenter if vcpu dirty ring is soft-full */
9380 	if (unlikely(vcpu->kvm->dirty_ring_size &&
9381 		     kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9382 		vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9383 		trace_kvm_dirty_ring_exit(vcpu);
9384 		r = 0;
9385 		goto out;
9386 	}
9387 
9388 	if (kvm_request_pending(vcpu)) {
9389 		if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9390 			if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9391 				r = 0;
9392 				goto out;
9393 			}
9394 		}
9395 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9396 			kvm_mmu_unload(vcpu);
9397 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9398 			__kvm_migrate_timers(vcpu);
9399 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9400 			kvm_gen_update_masterclock(vcpu->kvm);
9401 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9402 			kvm_gen_kvmclock_update(vcpu);
9403 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9404 			r = kvm_guest_time_update(vcpu);
9405 			if (unlikely(r))
9406 				goto out;
9407 		}
9408 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9409 			kvm_mmu_sync_roots(vcpu);
9410 		if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9411 			kvm_mmu_load_pgd(vcpu);
9412 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9413 			kvm_vcpu_flush_tlb_all(vcpu);
9414 
9415 			/* Flushing all ASIDs flushes the current ASID... */
9416 			kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9417 		}
9418 		if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
9419 			kvm_vcpu_flush_tlb_current(vcpu);
9420 		if (kvm_check_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu))
9421 			kvm_vcpu_flush_tlb_guest(vcpu);
9422 
9423 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9424 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9425 			r = 0;
9426 			goto out;
9427 		}
9428 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9429 			if (is_guest_mode(vcpu)) {
9430 				kvm_x86_ops.nested_ops->triple_fault(vcpu);
9431 			} else {
9432 				vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9433 				vcpu->mmio_needed = 0;
9434 				r = 0;
9435 				goto out;
9436 			}
9437 		}
9438 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9439 			/* Page is swapped out. Do synthetic halt */
9440 			vcpu->arch.apf.halted = true;
9441 			r = 1;
9442 			goto out;
9443 		}
9444 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9445 			record_steal_time(vcpu);
9446 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
9447 			process_smi(vcpu);
9448 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
9449 			process_nmi(vcpu);
9450 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
9451 			kvm_pmu_handle_event(vcpu);
9452 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
9453 			kvm_pmu_deliver_pmi(vcpu);
9454 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9455 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9456 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
9457 				     vcpu->arch.ioapic_handled_vectors)) {
9458 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9459 				vcpu->run->eoi.vector =
9460 						vcpu->arch.pending_ioapic_eoi;
9461 				r = 0;
9462 				goto out;
9463 			}
9464 		}
9465 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9466 			vcpu_scan_ioapic(vcpu);
9467 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9468 			vcpu_load_eoi_exitmap(vcpu);
9469 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9470 			kvm_vcpu_reload_apic_access_page(vcpu);
9471 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9472 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9473 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9474 			r = 0;
9475 			goto out;
9476 		}
9477 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9478 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9479 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9480 			r = 0;
9481 			goto out;
9482 		}
9483 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9484 			struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9485 
9486 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9487 			vcpu->run->hyperv = hv_vcpu->exit;
9488 			r = 0;
9489 			goto out;
9490 		}
9491 
9492 		/*
9493 		 * KVM_REQ_HV_STIMER has to be processed after
9494 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9495 		 * depend on the guest clock being up-to-date
9496 		 */
9497 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9498 			kvm_hv_process_stimers(vcpu);
9499 		if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9500 			kvm_vcpu_update_apicv(vcpu);
9501 		if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9502 			kvm_check_async_pf_completion(vcpu);
9503 		if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9504 			static_call(kvm_x86_msr_filter_changed)(vcpu);
9505 
9506 		if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9507 			static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9508 	}
9509 
9510 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9511 	    kvm_xen_has_interrupt(vcpu)) {
9512 		++vcpu->stat.req_event;
9513 		r = kvm_apic_accept_events(vcpu);
9514 		if (r < 0) {
9515 			r = 0;
9516 			goto out;
9517 		}
9518 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9519 			r = 1;
9520 			goto out;
9521 		}
9522 
9523 		r = inject_pending_event(vcpu, &req_immediate_exit);
9524 		if (r < 0) {
9525 			r = 0;
9526 			goto out;
9527 		}
9528 		if (req_int_win)
9529 			static_call(kvm_x86_enable_irq_window)(vcpu);
9530 
9531 		if (kvm_lapic_enabled(vcpu)) {
9532 			update_cr8_intercept(vcpu);
9533 			kvm_lapic_sync_to_vapic(vcpu);
9534 		}
9535 	}
9536 
9537 	r = kvm_mmu_reload(vcpu);
9538 	if (unlikely(r)) {
9539 		goto cancel_injection;
9540 	}
9541 
9542 	preempt_disable();
9543 
9544 	static_call(kvm_x86_prepare_guest_switch)(vcpu);
9545 
9546 	/*
9547 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
9548 	 * IPI are then delayed after guest entry, which ensures that they
9549 	 * result in virtual interrupt delivery.
9550 	 */
9551 	local_irq_disable();
9552 	vcpu->mode = IN_GUEST_MODE;
9553 
9554 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9555 
9556 	/*
9557 	 * 1) We should set ->mode before checking ->requests.  Please see
9558 	 * the comment in kvm_vcpu_exiting_guest_mode().
9559 	 *
9560 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
9561 	 * pairs with the memory barrier implicit in pi_test_and_set_on
9562 	 * (see vmx_deliver_posted_interrupt).
9563 	 *
9564 	 * 3) This also orders the write to mode from any reads to the page
9565 	 * tables done while the VCPU is running.  Please see the comment
9566 	 * in kvm_flush_remote_tlbs.
9567 	 */
9568 	smp_mb__after_srcu_read_unlock();
9569 
9570 	/*
9571 	 * This handles the case where a posted interrupt was
9572 	 * notified with kvm_vcpu_kick.
9573 	 */
9574 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
9575 		static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9576 
9577 	if (kvm_vcpu_exit_request(vcpu)) {
9578 		vcpu->mode = OUTSIDE_GUEST_MODE;
9579 		smp_wmb();
9580 		local_irq_enable();
9581 		preempt_enable();
9582 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9583 		r = 1;
9584 		goto cancel_injection;
9585 	}
9586 
9587 	if (req_immediate_exit) {
9588 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9589 		static_call(kvm_x86_request_immediate_exit)(vcpu);
9590 	}
9591 
9592 	fpregs_assert_state_consistent();
9593 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
9594 		switch_fpu_return();
9595 
9596 	if (unlikely(vcpu->arch.switch_db_regs)) {
9597 		set_debugreg(0, 7);
9598 		set_debugreg(vcpu->arch.eff_db[0], 0);
9599 		set_debugreg(vcpu->arch.eff_db[1], 1);
9600 		set_debugreg(vcpu->arch.eff_db[2], 2);
9601 		set_debugreg(vcpu->arch.eff_db[3], 3);
9602 		set_debugreg(vcpu->arch.dr6, 6);
9603 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9604 	} else if (unlikely(hw_breakpoint_active())) {
9605 		set_debugreg(0, 7);
9606 	}
9607 
9608 	for (;;) {
9609 		exit_fastpath = static_call(kvm_x86_run)(vcpu);
9610 		if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9611 			break;
9612 
9613                 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9614 			exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9615 			break;
9616 		}
9617 
9618 		if (vcpu->arch.apicv_active)
9619 			static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9620         }
9621 
9622 	/*
9623 	 * Do this here before restoring debug registers on the host.  And
9624 	 * since we do this before handling the vmexit, a DR access vmexit
9625 	 * can (a) read the correct value of the debug registers, (b) set
9626 	 * KVM_DEBUGREG_WONT_EXIT again.
9627 	 */
9628 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9629 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9630 		static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9631 		kvm_update_dr0123(vcpu);
9632 		kvm_update_dr7(vcpu);
9633 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9634 	}
9635 
9636 	/*
9637 	 * If the guest has used debug registers, at least dr7
9638 	 * will be disabled while returning to the host.
9639 	 * If we don't have active breakpoints in the host, we don't
9640 	 * care about the messed up debug address registers. But if
9641 	 * we have some of them active, restore the old state.
9642 	 */
9643 	if (hw_breakpoint_active())
9644 		hw_breakpoint_restore();
9645 
9646 	vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9647 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9648 
9649 	vcpu->mode = OUTSIDE_GUEST_MODE;
9650 	smp_wmb();
9651 
9652 	static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9653 
9654 	/*
9655 	 * Consume any pending interrupts, including the possible source of
9656 	 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9657 	 * An instruction is required after local_irq_enable() to fully unblock
9658 	 * interrupts on processors that implement an interrupt shadow, the
9659 	 * stat.exits increment will do nicely.
9660 	 */
9661 	kvm_before_interrupt(vcpu);
9662 	local_irq_enable();
9663 	++vcpu->stat.exits;
9664 	local_irq_disable();
9665 	kvm_after_interrupt(vcpu);
9666 
9667 	/*
9668 	 * Wait until after servicing IRQs to account guest time so that any
9669 	 * ticks that occurred while running the guest are properly accounted
9670 	 * to the guest.  Waiting until IRQs are enabled degrades the accuracy
9671 	 * of accounting via context tracking, but the loss of accuracy is
9672 	 * acceptable for all known use cases.
9673 	 */
9674 	vtime_account_guest_exit();
9675 
9676 	if (lapic_in_kernel(vcpu)) {
9677 		s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9678 		if (delta != S64_MIN) {
9679 			trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9680 			vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9681 		}
9682 	}
9683 
9684 	local_irq_enable();
9685 	preempt_enable();
9686 
9687 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9688 
9689 	/*
9690 	 * Profile KVM exit RIPs:
9691 	 */
9692 	if (unlikely(prof_on == KVM_PROFILING)) {
9693 		unsigned long rip = kvm_rip_read(vcpu);
9694 		profile_hit(KVM_PROFILING, (void *)rip);
9695 	}
9696 
9697 	if (unlikely(vcpu->arch.tsc_always_catchup))
9698 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9699 
9700 	if (vcpu->arch.apic_attention)
9701 		kvm_lapic_sync_from_vapic(vcpu);
9702 
9703 	r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
9704 	return r;
9705 
9706 cancel_injection:
9707 	if (req_immediate_exit)
9708 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9709 	static_call(kvm_x86_cancel_injection)(vcpu);
9710 	if (unlikely(vcpu->arch.apic_attention))
9711 		kvm_lapic_sync_from_vapic(vcpu);
9712 out:
9713 	return r;
9714 }
9715 
9716 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9717 {
9718 	if (!kvm_arch_vcpu_runnable(vcpu) &&
9719 	    (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
9720 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9721 		kvm_vcpu_block(vcpu);
9722 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9723 
9724 		if (kvm_x86_ops.post_block)
9725 			static_call(kvm_x86_post_block)(vcpu);
9726 
9727 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9728 			return 1;
9729 	}
9730 
9731 	if (kvm_apic_accept_events(vcpu) < 0)
9732 		return 0;
9733 	switch(vcpu->arch.mp_state) {
9734 	case KVM_MP_STATE_HALTED:
9735 	case KVM_MP_STATE_AP_RESET_HOLD:
9736 		vcpu->arch.pv.pv_unhalted = false;
9737 		vcpu->arch.mp_state =
9738 			KVM_MP_STATE_RUNNABLE;
9739 		fallthrough;
9740 	case KVM_MP_STATE_RUNNABLE:
9741 		vcpu->arch.apf.halted = false;
9742 		break;
9743 	case KVM_MP_STATE_INIT_RECEIVED:
9744 		break;
9745 	default:
9746 		return -EINTR;
9747 	}
9748 	return 1;
9749 }
9750 
9751 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9752 {
9753 	if (is_guest_mode(vcpu))
9754 		kvm_check_nested_events(vcpu);
9755 
9756 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9757 		!vcpu->arch.apf.halted);
9758 }
9759 
9760 static int vcpu_run(struct kvm_vcpu *vcpu)
9761 {
9762 	int r;
9763 	struct kvm *kvm = vcpu->kvm;
9764 
9765 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9766 	vcpu->arch.l1tf_flush_l1d = true;
9767 
9768 	for (;;) {
9769 		if (kvm_vcpu_running(vcpu)) {
9770 			r = vcpu_enter_guest(vcpu);
9771 		} else {
9772 			r = vcpu_block(kvm, vcpu);
9773 		}
9774 
9775 		if (r <= 0)
9776 			break;
9777 
9778 		kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
9779 		if (kvm_cpu_has_pending_timer(vcpu))
9780 			kvm_inject_pending_timer_irqs(vcpu);
9781 
9782 		if (dm_request_for_irq_injection(vcpu) &&
9783 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9784 			r = 0;
9785 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9786 			++vcpu->stat.request_irq_exits;
9787 			break;
9788 		}
9789 
9790 		if (__xfer_to_guest_mode_work_pending()) {
9791 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9792 			r = xfer_to_guest_mode_handle_work(vcpu);
9793 			if (r)
9794 				return r;
9795 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9796 		}
9797 	}
9798 
9799 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9800 
9801 	return r;
9802 }
9803 
9804 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9805 {
9806 	int r;
9807 
9808 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9809 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9810 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9811 	return r;
9812 }
9813 
9814 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9815 {
9816 	BUG_ON(!vcpu->arch.pio.count);
9817 
9818 	return complete_emulated_io(vcpu);
9819 }
9820 
9821 /*
9822  * Implements the following, as a state machine:
9823  *
9824  * read:
9825  *   for each fragment
9826  *     for each mmio piece in the fragment
9827  *       write gpa, len
9828  *       exit
9829  *       copy data
9830  *   execute insn
9831  *
9832  * write:
9833  *   for each fragment
9834  *     for each mmio piece in the fragment
9835  *       write gpa, len
9836  *       copy data
9837  *       exit
9838  */
9839 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9840 {
9841 	struct kvm_run *run = vcpu->run;
9842 	struct kvm_mmio_fragment *frag;
9843 	unsigned len;
9844 
9845 	BUG_ON(!vcpu->mmio_needed);
9846 
9847 	/* Complete previous fragment */
9848 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9849 	len = min(8u, frag->len);
9850 	if (!vcpu->mmio_is_write)
9851 		memcpy(frag->data, run->mmio.data, len);
9852 
9853 	if (frag->len <= 8) {
9854 		/* Switch to the next fragment. */
9855 		frag++;
9856 		vcpu->mmio_cur_fragment++;
9857 	} else {
9858 		/* Go forward to the next mmio piece. */
9859 		frag->data += len;
9860 		frag->gpa += len;
9861 		frag->len -= len;
9862 	}
9863 
9864 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9865 		vcpu->mmio_needed = 0;
9866 
9867 		/* FIXME: return into emulator if single-stepping.  */
9868 		if (vcpu->mmio_is_write)
9869 			return 1;
9870 		vcpu->mmio_read_completed = 1;
9871 		return complete_emulated_io(vcpu);
9872 	}
9873 
9874 	run->exit_reason = KVM_EXIT_MMIO;
9875 	run->mmio.phys_addr = frag->gpa;
9876 	if (vcpu->mmio_is_write)
9877 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9878 	run->mmio.len = min(8u, frag->len);
9879 	run->mmio.is_write = vcpu->mmio_is_write;
9880 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9881 	return 0;
9882 }
9883 
9884 static void kvm_save_current_fpu(struct fpu *fpu)
9885 {
9886 	/*
9887 	 * If the target FPU state is not resident in the CPU registers, just
9888 	 * memcpy() from current, else save CPU state directly to the target.
9889 	 */
9890 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
9891 		memcpy(&fpu->state, &current->thread.fpu.state,
9892 		       fpu_kernel_xstate_size);
9893 	else
9894 		save_fpregs_to_fpstate(fpu);
9895 }
9896 
9897 /* Swap (qemu) user FPU context for the guest FPU context. */
9898 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9899 {
9900 	fpregs_lock();
9901 
9902 	kvm_save_current_fpu(vcpu->arch.user_fpu);
9903 
9904 	/*
9905 	 * Guests with protected state can't have it set by the hypervisor,
9906 	 * so skip trying to set it.
9907 	 */
9908 	if (vcpu->arch.guest_fpu)
9909 		/* PKRU is separately restored in kvm_x86_ops.run. */
9910 		__restore_fpregs_from_fpstate(&vcpu->arch.guest_fpu->state,
9911 					~XFEATURE_MASK_PKRU);
9912 
9913 	fpregs_mark_activate();
9914 	fpregs_unlock();
9915 
9916 	trace_kvm_fpu(1);
9917 }
9918 
9919 /* When vcpu_run ends, restore user space FPU context. */
9920 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9921 {
9922 	fpregs_lock();
9923 
9924 	/*
9925 	 * Guests with protected state can't have it read by the hypervisor,
9926 	 * so skip trying to save it.
9927 	 */
9928 	if (vcpu->arch.guest_fpu)
9929 		kvm_save_current_fpu(vcpu->arch.guest_fpu);
9930 
9931 	restore_fpregs_from_fpstate(&vcpu->arch.user_fpu->state);
9932 
9933 	fpregs_mark_activate();
9934 	fpregs_unlock();
9935 
9936 	++vcpu->stat.fpu_reload;
9937 	trace_kvm_fpu(0);
9938 }
9939 
9940 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9941 {
9942 	struct kvm_run *kvm_run = vcpu->run;
9943 	int r;
9944 
9945 	vcpu_load(vcpu);
9946 	kvm_sigset_activate(vcpu);
9947 	kvm_run->flags = 0;
9948 	kvm_load_guest_fpu(vcpu);
9949 
9950 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9951 		if (kvm_run->immediate_exit) {
9952 			r = -EINTR;
9953 			goto out;
9954 		}
9955 		kvm_vcpu_block(vcpu);
9956 		if (kvm_apic_accept_events(vcpu) < 0) {
9957 			r = 0;
9958 			goto out;
9959 		}
9960 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9961 		r = -EAGAIN;
9962 		if (signal_pending(current)) {
9963 			r = -EINTR;
9964 			kvm_run->exit_reason = KVM_EXIT_INTR;
9965 			++vcpu->stat.signal_exits;
9966 		}
9967 		goto out;
9968 	}
9969 
9970 	if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
9971 		r = -EINVAL;
9972 		goto out;
9973 	}
9974 
9975 	if (kvm_run->kvm_dirty_regs) {
9976 		r = sync_regs(vcpu);
9977 		if (r != 0)
9978 			goto out;
9979 	}
9980 
9981 	/* re-sync apic's tpr */
9982 	if (!lapic_in_kernel(vcpu)) {
9983 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
9984 			r = -EINVAL;
9985 			goto out;
9986 		}
9987 	}
9988 
9989 	if (unlikely(vcpu->arch.complete_userspace_io)) {
9990 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
9991 		vcpu->arch.complete_userspace_io = NULL;
9992 		r = cui(vcpu);
9993 		if (r <= 0)
9994 			goto out;
9995 	} else
9996 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
9997 
9998 	if (kvm_run->immediate_exit)
9999 		r = -EINTR;
10000 	else
10001 		r = vcpu_run(vcpu);
10002 
10003 out:
10004 	kvm_put_guest_fpu(vcpu);
10005 	if (kvm_run->kvm_valid_regs)
10006 		store_regs(vcpu);
10007 	post_kvm_run_save(vcpu);
10008 	kvm_sigset_deactivate(vcpu);
10009 
10010 	vcpu_put(vcpu);
10011 	return r;
10012 }
10013 
10014 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10015 {
10016 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
10017 		/*
10018 		 * We are here if userspace calls get_regs() in the middle of
10019 		 * instruction emulation. Registers state needs to be copied
10020 		 * back from emulation context to vcpu. Userspace shouldn't do
10021 		 * that usually, but some bad designed PV devices (vmware
10022 		 * backdoor interface) need this to work
10023 		 */
10024 		emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
10025 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10026 	}
10027 	regs->rax = kvm_rax_read(vcpu);
10028 	regs->rbx = kvm_rbx_read(vcpu);
10029 	regs->rcx = kvm_rcx_read(vcpu);
10030 	regs->rdx = kvm_rdx_read(vcpu);
10031 	regs->rsi = kvm_rsi_read(vcpu);
10032 	regs->rdi = kvm_rdi_read(vcpu);
10033 	regs->rsp = kvm_rsp_read(vcpu);
10034 	regs->rbp = kvm_rbp_read(vcpu);
10035 #ifdef CONFIG_X86_64
10036 	regs->r8 = kvm_r8_read(vcpu);
10037 	regs->r9 = kvm_r9_read(vcpu);
10038 	regs->r10 = kvm_r10_read(vcpu);
10039 	regs->r11 = kvm_r11_read(vcpu);
10040 	regs->r12 = kvm_r12_read(vcpu);
10041 	regs->r13 = kvm_r13_read(vcpu);
10042 	regs->r14 = kvm_r14_read(vcpu);
10043 	regs->r15 = kvm_r15_read(vcpu);
10044 #endif
10045 
10046 	regs->rip = kvm_rip_read(vcpu);
10047 	regs->rflags = kvm_get_rflags(vcpu);
10048 }
10049 
10050 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10051 {
10052 	vcpu_load(vcpu);
10053 	__get_regs(vcpu, regs);
10054 	vcpu_put(vcpu);
10055 	return 0;
10056 }
10057 
10058 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10059 {
10060 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
10061 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10062 
10063 	kvm_rax_write(vcpu, regs->rax);
10064 	kvm_rbx_write(vcpu, regs->rbx);
10065 	kvm_rcx_write(vcpu, regs->rcx);
10066 	kvm_rdx_write(vcpu, regs->rdx);
10067 	kvm_rsi_write(vcpu, regs->rsi);
10068 	kvm_rdi_write(vcpu, regs->rdi);
10069 	kvm_rsp_write(vcpu, regs->rsp);
10070 	kvm_rbp_write(vcpu, regs->rbp);
10071 #ifdef CONFIG_X86_64
10072 	kvm_r8_write(vcpu, regs->r8);
10073 	kvm_r9_write(vcpu, regs->r9);
10074 	kvm_r10_write(vcpu, regs->r10);
10075 	kvm_r11_write(vcpu, regs->r11);
10076 	kvm_r12_write(vcpu, regs->r12);
10077 	kvm_r13_write(vcpu, regs->r13);
10078 	kvm_r14_write(vcpu, regs->r14);
10079 	kvm_r15_write(vcpu, regs->r15);
10080 #endif
10081 
10082 	kvm_rip_write(vcpu, regs->rip);
10083 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
10084 
10085 	vcpu->arch.exception.pending = false;
10086 
10087 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10088 }
10089 
10090 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10091 {
10092 	vcpu_load(vcpu);
10093 	__set_regs(vcpu, regs);
10094 	vcpu_put(vcpu);
10095 	return 0;
10096 }
10097 
10098 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
10099 {
10100 	struct kvm_segment cs;
10101 
10102 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10103 	*db = cs.db;
10104 	*l = cs.l;
10105 }
10106 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
10107 
10108 static void __get_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10109 {
10110 	struct desc_ptr dt;
10111 
10112 	if (vcpu->arch.guest_state_protected)
10113 		goto skip_protected_regs;
10114 
10115 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10116 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10117 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10118 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10119 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10120 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10121 
10122 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10123 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10124 
10125 	static_call(kvm_x86_get_idt)(vcpu, &dt);
10126 	sregs->idt.limit = dt.size;
10127 	sregs->idt.base = dt.address;
10128 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
10129 	sregs->gdt.limit = dt.size;
10130 	sregs->gdt.base = dt.address;
10131 
10132 	sregs->cr2 = vcpu->arch.cr2;
10133 	sregs->cr3 = kvm_read_cr3(vcpu);
10134 
10135 skip_protected_regs:
10136 	sregs->cr0 = kvm_read_cr0(vcpu);
10137 	sregs->cr4 = kvm_read_cr4(vcpu);
10138 	sregs->cr8 = kvm_get_cr8(vcpu);
10139 	sregs->efer = vcpu->arch.efer;
10140 	sregs->apic_base = kvm_get_apic_base(vcpu);
10141 }
10142 
10143 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10144 {
10145 	__get_sregs_common(vcpu, sregs);
10146 
10147 	if (vcpu->arch.guest_state_protected)
10148 		return;
10149 
10150 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
10151 		set_bit(vcpu->arch.interrupt.nr,
10152 			(unsigned long *)sregs->interrupt_bitmap);
10153 }
10154 
10155 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10156 {
10157 	int i;
10158 
10159 	__get_sregs_common(vcpu, (struct kvm_sregs *)sregs2);
10160 
10161 	if (vcpu->arch.guest_state_protected)
10162 		return;
10163 
10164 	if (is_pae_paging(vcpu)) {
10165 		for (i = 0 ; i < 4 ; i++)
10166 			sregs2->pdptrs[i] = kvm_pdptr_read(vcpu, i);
10167 		sregs2->flags |= KVM_SREGS2_FLAGS_PDPTRS_VALID;
10168 	}
10169 }
10170 
10171 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
10172 				  struct kvm_sregs *sregs)
10173 {
10174 	vcpu_load(vcpu);
10175 	__get_sregs(vcpu, sregs);
10176 	vcpu_put(vcpu);
10177 	return 0;
10178 }
10179 
10180 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
10181 				    struct kvm_mp_state *mp_state)
10182 {
10183 	int r;
10184 
10185 	vcpu_load(vcpu);
10186 	if (kvm_mpx_supported())
10187 		kvm_load_guest_fpu(vcpu);
10188 
10189 	r = kvm_apic_accept_events(vcpu);
10190 	if (r < 0)
10191 		goto out;
10192 	r = 0;
10193 
10194 	if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
10195 	     vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
10196 	    vcpu->arch.pv.pv_unhalted)
10197 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
10198 	else
10199 		mp_state->mp_state = vcpu->arch.mp_state;
10200 
10201 out:
10202 	if (kvm_mpx_supported())
10203 		kvm_put_guest_fpu(vcpu);
10204 	vcpu_put(vcpu);
10205 	return r;
10206 }
10207 
10208 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
10209 				    struct kvm_mp_state *mp_state)
10210 {
10211 	int ret = -EINVAL;
10212 
10213 	vcpu_load(vcpu);
10214 
10215 	if (!lapic_in_kernel(vcpu) &&
10216 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
10217 		goto out;
10218 
10219 	/*
10220 	 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
10221 	 * INIT state; latched init should be reported using
10222 	 * KVM_SET_VCPU_EVENTS, so reject it here.
10223 	 */
10224 	if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
10225 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
10226 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
10227 		goto out;
10228 
10229 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
10230 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
10231 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
10232 	} else
10233 		vcpu->arch.mp_state = mp_state->mp_state;
10234 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10235 
10236 	ret = 0;
10237 out:
10238 	vcpu_put(vcpu);
10239 	return ret;
10240 }
10241 
10242 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
10243 		    int reason, bool has_error_code, u32 error_code)
10244 {
10245 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
10246 	int ret;
10247 
10248 	init_emulate_ctxt(vcpu);
10249 
10250 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
10251 				   has_error_code, error_code);
10252 	if (ret) {
10253 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
10254 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
10255 		vcpu->run->internal.ndata = 0;
10256 		return 0;
10257 	}
10258 
10259 	kvm_rip_write(vcpu, ctxt->eip);
10260 	kvm_set_rflags(vcpu, ctxt->eflags);
10261 	return 1;
10262 }
10263 EXPORT_SYMBOL_GPL(kvm_task_switch);
10264 
10265 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10266 {
10267 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
10268 		/*
10269 		 * When EFER.LME and CR0.PG are set, the processor is in
10270 		 * 64-bit mode (though maybe in a 32-bit code segment).
10271 		 * CR4.PAE and EFER.LMA must be set.
10272 		 */
10273 		if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
10274 			return false;
10275 		if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
10276 			return false;
10277 	} else {
10278 		/*
10279 		 * Not in 64-bit mode: EFER.LMA is clear and the code
10280 		 * segment cannot be 64-bit.
10281 		 */
10282 		if (sregs->efer & EFER_LMA || sregs->cs.l)
10283 			return false;
10284 	}
10285 
10286 	return kvm_is_valid_cr4(vcpu, sregs->cr4);
10287 }
10288 
10289 static int __set_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs,
10290 		int *mmu_reset_needed, bool update_pdptrs)
10291 {
10292 	struct msr_data apic_base_msr;
10293 	int idx;
10294 	struct desc_ptr dt;
10295 
10296 	if (!kvm_is_valid_sregs(vcpu, sregs))
10297 		return -EINVAL;
10298 
10299 	apic_base_msr.data = sregs->apic_base;
10300 	apic_base_msr.host_initiated = true;
10301 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
10302 		return -EINVAL;
10303 
10304 	if (vcpu->arch.guest_state_protected)
10305 		return 0;
10306 
10307 	dt.size = sregs->idt.limit;
10308 	dt.address = sregs->idt.base;
10309 	static_call(kvm_x86_set_idt)(vcpu, &dt);
10310 	dt.size = sregs->gdt.limit;
10311 	dt.address = sregs->gdt.base;
10312 	static_call(kvm_x86_set_gdt)(vcpu, &dt);
10313 
10314 	vcpu->arch.cr2 = sregs->cr2;
10315 	*mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
10316 	vcpu->arch.cr3 = sregs->cr3;
10317 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
10318 
10319 	kvm_set_cr8(vcpu, sregs->cr8);
10320 
10321 	*mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
10322 	static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
10323 
10324 	*mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
10325 	static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
10326 	vcpu->arch.cr0 = sregs->cr0;
10327 
10328 	*mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
10329 	static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
10330 
10331 	if (update_pdptrs) {
10332 		idx = srcu_read_lock(&vcpu->kvm->srcu);
10333 		if (is_pae_paging(vcpu)) {
10334 			load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
10335 			*mmu_reset_needed = 1;
10336 		}
10337 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
10338 	}
10339 
10340 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10341 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10342 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10343 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10344 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10345 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10346 
10347 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10348 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10349 
10350 	update_cr8_intercept(vcpu);
10351 
10352 	/* Older userspace won't unhalt the vcpu on reset. */
10353 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
10354 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
10355 	    !is_protmode(vcpu))
10356 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10357 
10358 	return 0;
10359 }
10360 
10361 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10362 {
10363 	int pending_vec, max_bits;
10364 	int mmu_reset_needed = 0;
10365 	int ret = __set_sregs_common(vcpu, sregs, &mmu_reset_needed, true);
10366 
10367 	if (ret)
10368 		return ret;
10369 
10370 	if (mmu_reset_needed)
10371 		kvm_mmu_reset_context(vcpu);
10372 
10373 	max_bits = KVM_NR_INTERRUPTS;
10374 	pending_vec = find_first_bit(
10375 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
10376 
10377 	if (pending_vec < max_bits) {
10378 		kvm_queue_interrupt(vcpu, pending_vec, false);
10379 		pr_debug("Set back pending irq %d\n", pending_vec);
10380 		kvm_make_request(KVM_REQ_EVENT, vcpu);
10381 	}
10382 	return 0;
10383 }
10384 
10385 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10386 {
10387 	int mmu_reset_needed = 0;
10388 	bool valid_pdptrs = sregs2->flags & KVM_SREGS2_FLAGS_PDPTRS_VALID;
10389 	bool pae = (sregs2->cr0 & X86_CR0_PG) && (sregs2->cr4 & X86_CR4_PAE) &&
10390 		!(sregs2->efer & EFER_LMA);
10391 	int i, ret;
10392 
10393 	if (sregs2->flags & ~KVM_SREGS2_FLAGS_PDPTRS_VALID)
10394 		return -EINVAL;
10395 
10396 	if (valid_pdptrs && (!pae || vcpu->arch.guest_state_protected))
10397 		return -EINVAL;
10398 
10399 	ret = __set_sregs_common(vcpu, (struct kvm_sregs *)sregs2,
10400 				 &mmu_reset_needed, !valid_pdptrs);
10401 	if (ret)
10402 		return ret;
10403 
10404 	if (valid_pdptrs) {
10405 		for (i = 0; i < 4 ; i++)
10406 			kvm_pdptr_write(vcpu, i, sregs2->pdptrs[i]);
10407 
10408 		kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
10409 		mmu_reset_needed = 1;
10410 		vcpu->arch.pdptrs_from_userspace = true;
10411 	}
10412 	if (mmu_reset_needed)
10413 		kvm_mmu_reset_context(vcpu);
10414 	return 0;
10415 }
10416 
10417 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
10418 				  struct kvm_sregs *sregs)
10419 {
10420 	int ret;
10421 
10422 	vcpu_load(vcpu);
10423 	ret = __set_sregs(vcpu, sregs);
10424 	vcpu_put(vcpu);
10425 	return ret;
10426 }
10427 
10428 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
10429 					struct kvm_guest_debug *dbg)
10430 {
10431 	unsigned long rflags;
10432 	int i, r;
10433 
10434 	if (vcpu->arch.guest_state_protected)
10435 		return -EINVAL;
10436 
10437 	vcpu_load(vcpu);
10438 
10439 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10440 		r = -EBUSY;
10441 		if (vcpu->arch.exception.pending)
10442 			goto out;
10443 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10444 			kvm_queue_exception(vcpu, DB_VECTOR);
10445 		else
10446 			kvm_queue_exception(vcpu, BP_VECTOR);
10447 	}
10448 
10449 	/*
10450 	 * Read rflags as long as potentially injected trace flags are still
10451 	 * filtered out.
10452 	 */
10453 	rflags = kvm_get_rflags(vcpu);
10454 
10455 	vcpu->guest_debug = dbg->control;
10456 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10457 		vcpu->guest_debug = 0;
10458 
10459 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10460 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
10461 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10462 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10463 	} else {
10464 		for (i = 0; i < KVM_NR_DB_REGS; i++)
10465 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10466 	}
10467 	kvm_update_dr7(vcpu);
10468 
10469 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10470 		vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
10471 
10472 	/*
10473 	 * Trigger an rflags update that will inject or remove the trace
10474 	 * flags.
10475 	 */
10476 	kvm_set_rflags(vcpu, rflags);
10477 
10478 	static_call(kvm_x86_update_exception_bitmap)(vcpu);
10479 
10480 	r = 0;
10481 
10482 out:
10483 	vcpu_put(vcpu);
10484 	return r;
10485 }
10486 
10487 /*
10488  * Translate a guest virtual address to a guest physical address.
10489  */
10490 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10491 				    struct kvm_translation *tr)
10492 {
10493 	unsigned long vaddr = tr->linear_address;
10494 	gpa_t gpa;
10495 	int idx;
10496 
10497 	vcpu_load(vcpu);
10498 
10499 	idx = srcu_read_lock(&vcpu->kvm->srcu);
10500 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10501 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
10502 	tr->physical_address = gpa;
10503 	tr->valid = gpa != UNMAPPED_GVA;
10504 	tr->writeable = 1;
10505 	tr->usermode = 0;
10506 
10507 	vcpu_put(vcpu);
10508 	return 0;
10509 }
10510 
10511 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10512 {
10513 	struct fxregs_state *fxsave;
10514 
10515 	if (!vcpu->arch.guest_fpu)
10516 		return 0;
10517 
10518 	vcpu_load(vcpu);
10519 
10520 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10521 	memcpy(fpu->fpr, fxsave->st_space, 128);
10522 	fpu->fcw = fxsave->cwd;
10523 	fpu->fsw = fxsave->swd;
10524 	fpu->ftwx = fxsave->twd;
10525 	fpu->last_opcode = fxsave->fop;
10526 	fpu->last_ip = fxsave->rip;
10527 	fpu->last_dp = fxsave->rdp;
10528 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10529 
10530 	vcpu_put(vcpu);
10531 	return 0;
10532 }
10533 
10534 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10535 {
10536 	struct fxregs_state *fxsave;
10537 
10538 	if (!vcpu->arch.guest_fpu)
10539 		return 0;
10540 
10541 	vcpu_load(vcpu);
10542 
10543 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10544 
10545 	memcpy(fxsave->st_space, fpu->fpr, 128);
10546 	fxsave->cwd = fpu->fcw;
10547 	fxsave->swd = fpu->fsw;
10548 	fxsave->twd = fpu->ftwx;
10549 	fxsave->fop = fpu->last_opcode;
10550 	fxsave->rip = fpu->last_ip;
10551 	fxsave->rdp = fpu->last_dp;
10552 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10553 
10554 	vcpu_put(vcpu);
10555 	return 0;
10556 }
10557 
10558 static void store_regs(struct kvm_vcpu *vcpu)
10559 {
10560 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10561 
10562 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10563 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
10564 
10565 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10566 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10567 
10568 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10569 		kvm_vcpu_ioctl_x86_get_vcpu_events(
10570 				vcpu, &vcpu->run->s.regs.events);
10571 }
10572 
10573 static int sync_regs(struct kvm_vcpu *vcpu)
10574 {
10575 	if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
10576 		return -EINVAL;
10577 
10578 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10579 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
10580 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10581 	}
10582 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10583 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10584 			return -EINVAL;
10585 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10586 	}
10587 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10588 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10589 				vcpu, &vcpu->run->s.regs.events))
10590 			return -EINVAL;
10591 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10592 	}
10593 
10594 	return 0;
10595 }
10596 
10597 static void fx_init(struct kvm_vcpu *vcpu)
10598 {
10599 	if (!vcpu->arch.guest_fpu)
10600 		return;
10601 
10602 	fpstate_init(&vcpu->arch.guest_fpu->state);
10603 	if (boot_cpu_has(X86_FEATURE_XSAVES))
10604 		vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
10605 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
10606 
10607 	/*
10608 	 * Ensure guest xcr0 is valid for loading
10609 	 */
10610 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10611 
10612 	vcpu->arch.cr0 |= X86_CR0_ET;
10613 }
10614 
10615 void kvm_free_guest_fpu(struct kvm_vcpu *vcpu)
10616 {
10617 	if (vcpu->arch.guest_fpu) {
10618 		kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
10619 		vcpu->arch.guest_fpu = NULL;
10620 	}
10621 }
10622 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu);
10623 
10624 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10625 {
10626 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10627 		pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10628 			     "guest TSC will not be reliable\n");
10629 
10630 	return 0;
10631 }
10632 
10633 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10634 {
10635 	struct page *page;
10636 	int r;
10637 
10638 	vcpu->arch.last_vmentry_cpu = -1;
10639 
10640 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10641 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10642 	else
10643 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10644 
10645 	r = kvm_mmu_create(vcpu);
10646 	if (r < 0)
10647 		return r;
10648 
10649 	if (irqchip_in_kernel(vcpu->kvm)) {
10650 		r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10651 		if (r < 0)
10652 			goto fail_mmu_destroy;
10653 		if (kvm_apicv_activated(vcpu->kvm))
10654 			vcpu->arch.apicv_active = true;
10655 	} else
10656 		static_branch_inc(&kvm_has_noapic_vcpu);
10657 
10658 	r = -ENOMEM;
10659 
10660 	page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10661 	if (!page)
10662 		goto fail_free_lapic;
10663 	vcpu->arch.pio_data = page_address(page);
10664 
10665 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10666 				       GFP_KERNEL_ACCOUNT);
10667 	if (!vcpu->arch.mce_banks)
10668 		goto fail_free_pio_data;
10669 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
10670 
10671 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
10672 				GFP_KERNEL_ACCOUNT))
10673 		goto fail_free_mce_banks;
10674 
10675 	if (!alloc_emulate_ctxt(vcpu))
10676 		goto free_wbinvd_dirty_mask;
10677 
10678 	vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
10679 						GFP_KERNEL_ACCOUNT);
10680 	if (!vcpu->arch.user_fpu) {
10681 		pr_err("kvm: failed to allocate userspace's fpu\n");
10682 		goto free_emulate_ctxt;
10683 	}
10684 
10685 	vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
10686 						 GFP_KERNEL_ACCOUNT);
10687 	if (!vcpu->arch.guest_fpu) {
10688 		pr_err("kvm: failed to allocate vcpu's fpu\n");
10689 		goto free_user_fpu;
10690 	}
10691 	fx_init(vcpu);
10692 
10693 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10694 	vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
10695 
10696 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10697 
10698 	kvm_async_pf_hash_reset(vcpu);
10699 	kvm_pmu_init(vcpu);
10700 
10701 	vcpu->arch.pending_external_vector = -1;
10702 	vcpu->arch.preempted_in_kernel = false;
10703 
10704 #if IS_ENABLED(CONFIG_HYPERV)
10705 	vcpu->arch.hv_root_tdp = INVALID_PAGE;
10706 #endif
10707 
10708 	r = static_call(kvm_x86_vcpu_create)(vcpu);
10709 	if (r)
10710 		goto free_guest_fpu;
10711 
10712 	vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10713 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10714 	kvm_vcpu_mtrr_init(vcpu);
10715 	vcpu_load(vcpu);
10716 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
10717 	kvm_vcpu_reset(vcpu, false);
10718 	kvm_init_mmu(vcpu);
10719 	vcpu_put(vcpu);
10720 	return 0;
10721 
10722 free_guest_fpu:
10723 	kvm_free_guest_fpu(vcpu);
10724 free_user_fpu:
10725 	kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10726 free_emulate_ctxt:
10727 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10728 free_wbinvd_dirty_mask:
10729 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10730 fail_free_mce_banks:
10731 	kfree(vcpu->arch.mce_banks);
10732 fail_free_pio_data:
10733 	free_page((unsigned long)vcpu->arch.pio_data);
10734 fail_free_lapic:
10735 	kvm_free_lapic(vcpu);
10736 fail_mmu_destroy:
10737 	kvm_mmu_destroy(vcpu);
10738 	return r;
10739 }
10740 
10741 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10742 {
10743 	struct kvm *kvm = vcpu->kvm;
10744 
10745 	if (mutex_lock_killable(&vcpu->mutex))
10746 		return;
10747 	vcpu_load(vcpu);
10748 	kvm_synchronize_tsc(vcpu, 0);
10749 	vcpu_put(vcpu);
10750 
10751 	/* poll control enabled by default */
10752 	vcpu->arch.msr_kvm_poll_control = 1;
10753 
10754 	mutex_unlock(&vcpu->mutex);
10755 
10756 	if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10757 		schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10758 						KVMCLOCK_SYNC_PERIOD);
10759 }
10760 
10761 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10762 {
10763 	struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
10764 	int idx;
10765 
10766 	kvm_release_pfn(cache->pfn, cache->dirty, cache);
10767 
10768 	kvmclock_reset(vcpu);
10769 
10770 	static_call(kvm_x86_vcpu_free)(vcpu);
10771 
10772 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10773 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10774 	kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10775 	kvm_free_guest_fpu(vcpu);
10776 
10777 	kvm_hv_vcpu_uninit(vcpu);
10778 	kvm_pmu_destroy(vcpu);
10779 	kfree(vcpu->arch.mce_banks);
10780 	kvm_free_lapic(vcpu);
10781 	idx = srcu_read_lock(&vcpu->kvm->srcu);
10782 	kvm_mmu_destroy(vcpu);
10783 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
10784 	free_page((unsigned long)vcpu->arch.pio_data);
10785 	kvfree(vcpu->arch.cpuid_entries);
10786 	if (!lapic_in_kernel(vcpu))
10787 		static_branch_dec(&kvm_has_noapic_vcpu);
10788 }
10789 
10790 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10791 {
10792 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
10793 
10794 	kvm_lapic_reset(vcpu, init_event);
10795 
10796 	vcpu->arch.hflags = 0;
10797 
10798 	vcpu->arch.smi_pending = 0;
10799 	vcpu->arch.smi_count = 0;
10800 	atomic_set(&vcpu->arch.nmi_queued, 0);
10801 	vcpu->arch.nmi_pending = 0;
10802 	vcpu->arch.nmi_injected = false;
10803 	kvm_clear_interrupt_queue(vcpu);
10804 	kvm_clear_exception_queue(vcpu);
10805 
10806 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10807 	kvm_update_dr0123(vcpu);
10808 	vcpu->arch.dr6 = DR6_ACTIVE_LOW;
10809 	vcpu->arch.dr7 = DR7_FIXED_1;
10810 	kvm_update_dr7(vcpu);
10811 
10812 	vcpu->arch.cr2 = 0;
10813 
10814 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10815 	vcpu->arch.apf.msr_en_val = 0;
10816 	vcpu->arch.apf.msr_int_val = 0;
10817 	vcpu->arch.st.msr_val = 0;
10818 
10819 	kvmclock_reset(vcpu);
10820 
10821 	kvm_clear_async_pf_completion_queue(vcpu);
10822 	kvm_async_pf_hash_reset(vcpu);
10823 	vcpu->arch.apf.halted = false;
10824 
10825 	if (vcpu->arch.guest_fpu && kvm_mpx_supported()) {
10826 		void *mpx_state_buffer;
10827 
10828 		/*
10829 		 * To avoid have the INIT path from kvm_apic_has_events() that be
10830 		 * called with loaded FPU and does not let userspace fix the state.
10831 		 */
10832 		if (init_event)
10833 			kvm_put_guest_fpu(vcpu);
10834 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10835 					XFEATURE_BNDREGS);
10836 		if (mpx_state_buffer)
10837 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
10838 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10839 					XFEATURE_BNDCSR);
10840 		if (mpx_state_buffer)
10841 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
10842 		if (init_event)
10843 			kvm_load_guest_fpu(vcpu);
10844 	}
10845 
10846 	if (!init_event) {
10847 		kvm_pmu_reset(vcpu);
10848 		vcpu->arch.smbase = 0x30000;
10849 
10850 		vcpu->arch.msr_misc_features_enables = 0;
10851 
10852 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10853 	}
10854 
10855 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10856 	vcpu->arch.regs_avail = ~0;
10857 	vcpu->arch.regs_dirty = ~0;
10858 
10859 	vcpu->arch.ia32_xss = 0;
10860 
10861 	static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
10862 
10863 	/*
10864 	 * Reset the MMU context if paging was enabled prior to INIT (which is
10865 	 * implied if CR0.PG=1 as CR0 will be '0' prior to RESET).  Unlike the
10866 	 * standard CR0/CR4/EFER modification paths, only CR0.PG needs to be
10867 	 * checked because it is unconditionally cleared on INIT and all other
10868 	 * paging related bits are ignored if paging is disabled, i.e. CR0.WP,
10869 	 * CR4, and EFER changes are all irrelevant if CR0.PG was '0'.
10870 	 */
10871 	if (old_cr0 & X86_CR0_PG)
10872 		kvm_mmu_reset_context(vcpu);
10873 }
10874 
10875 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10876 {
10877 	struct kvm_segment cs;
10878 
10879 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10880 	cs.selector = vector << 8;
10881 	cs.base = vector << 12;
10882 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10883 	kvm_rip_write(vcpu, 0);
10884 }
10885 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
10886 
10887 int kvm_arch_hardware_enable(void)
10888 {
10889 	struct kvm *kvm;
10890 	struct kvm_vcpu *vcpu;
10891 	int i;
10892 	int ret;
10893 	u64 local_tsc;
10894 	u64 max_tsc = 0;
10895 	bool stable, backwards_tsc = false;
10896 
10897 	kvm_user_return_msr_cpu_online();
10898 	ret = static_call(kvm_x86_hardware_enable)();
10899 	if (ret != 0)
10900 		return ret;
10901 
10902 	local_tsc = rdtsc();
10903 	stable = !kvm_check_tsc_unstable();
10904 	list_for_each_entry(kvm, &vm_list, vm_list) {
10905 		kvm_for_each_vcpu(i, vcpu, kvm) {
10906 			if (!stable && vcpu->cpu == smp_processor_id())
10907 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10908 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10909 				backwards_tsc = true;
10910 				if (vcpu->arch.last_host_tsc > max_tsc)
10911 					max_tsc = vcpu->arch.last_host_tsc;
10912 			}
10913 		}
10914 	}
10915 
10916 	/*
10917 	 * Sometimes, even reliable TSCs go backwards.  This happens on
10918 	 * platforms that reset TSC during suspend or hibernate actions, but
10919 	 * maintain synchronization.  We must compensate.  Fortunately, we can
10920 	 * detect that condition here, which happens early in CPU bringup,
10921 	 * before any KVM threads can be running.  Unfortunately, we can't
10922 	 * bring the TSCs fully up to date with real time, as we aren't yet far
10923 	 * enough into CPU bringup that we know how much real time has actually
10924 	 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10925 	 * variables that haven't been updated yet.
10926 	 *
10927 	 * So we simply find the maximum observed TSC above, then record the
10928 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
10929 	 * the adjustment will be applied.  Note that we accumulate
10930 	 * adjustments, in case multiple suspend cycles happen before some VCPU
10931 	 * gets a chance to run again.  In the event that no KVM threads get a
10932 	 * chance to run, we will miss the entire elapsed period, as we'll have
10933 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10934 	 * loose cycle time.  This isn't too big a deal, since the loss will be
10935 	 * uniform across all VCPUs (not to mention the scenario is extremely
10936 	 * unlikely). It is possible that a second hibernate recovery happens
10937 	 * much faster than a first, causing the observed TSC here to be
10938 	 * smaller; this would require additional padding adjustment, which is
10939 	 * why we set last_host_tsc to the local tsc observed here.
10940 	 *
10941 	 * N.B. - this code below runs only on platforms with reliable TSC,
10942 	 * as that is the only way backwards_tsc is set above.  Also note
10943 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10944 	 * have the same delta_cyc adjustment applied if backwards_tsc
10945 	 * is detected.  Note further, this adjustment is only done once,
10946 	 * as we reset last_host_tsc on all VCPUs to stop this from being
10947 	 * called multiple times (one for each physical CPU bringup).
10948 	 *
10949 	 * Platforms with unreliable TSCs don't have to deal with this, they
10950 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
10951 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
10952 	 * guarantee that they stay in perfect synchronization.
10953 	 */
10954 	if (backwards_tsc) {
10955 		u64 delta_cyc = max_tsc - local_tsc;
10956 		list_for_each_entry(kvm, &vm_list, vm_list) {
10957 			kvm->arch.backwards_tsc_observed = true;
10958 			kvm_for_each_vcpu(i, vcpu, kvm) {
10959 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
10960 				vcpu->arch.last_host_tsc = local_tsc;
10961 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
10962 			}
10963 
10964 			/*
10965 			 * We have to disable TSC offset matching.. if you were
10966 			 * booting a VM while issuing an S4 host suspend....
10967 			 * you may have some problem.  Solving this issue is
10968 			 * left as an exercise to the reader.
10969 			 */
10970 			kvm->arch.last_tsc_nsec = 0;
10971 			kvm->arch.last_tsc_write = 0;
10972 		}
10973 
10974 	}
10975 	return 0;
10976 }
10977 
10978 void kvm_arch_hardware_disable(void)
10979 {
10980 	static_call(kvm_x86_hardware_disable)();
10981 	drop_user_return_notifiers();
10982 }
10983 
10984 int kvm_arch_hardware_setup(void *opaque)
10985 {
10986 	struct kvm_x86_init_ops *ops = opaque;
10987 	int r;
10988 
10989 	rdmsrl_safe(MSR_EFER, &host_efer);
10990 
10991 	if (boot_cpu_has(X86_FEATURE_XSAVES))
10992 		rdmsrl(MSR_IA32_XSS, host_xss);
10993 
10994 	r = ops->hardware_setup();
10995 	if (r != 0)
10996 		return r;
10997 
10998 	memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
10999 	kvm_ops_static_call_update();
11000 
11001 	if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
11002 		supported_xss = 0;
11003 
11004 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
11005 	cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
11006 #undef __kvm_cpu_cap_has
11007 
11008 	if (kvm_has_tsc_control) {
11009 		/*
11010 		 * Make sure the user can only configure tsc_khz values that
11011 		 * fit into a signed integer.
11012 		 * A min value is not calculated because it will always
11013 		 * be 1 on all machines.
11014 		 */
11015 		u64 max = min(0x7fffffffULL,
11016 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
11017 		kvm_max_guest_tsc_khz = max;
11018 
11019 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
11020 	}
11021 
11022 	kvm_init_msr_list();
11023 	return 0;
11024 }
11025 
11026 void kvm_arch_hardware_unsetup(void)
11027 {
11028 	static_call(kvm_x86_hardware_unsetup)();
11029 }
11030 
11031 int kvm_arch_check_processor_compat(void *opaque)
11032 {
11033 	struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
11034 	struct kvm_x86_init_ops *ops = opaque;
11035 
11036 	WARN_ON(!irqs_disabled());
11037 
11038 	if (__cr4_reserved_bits(cpu_has, c) !=
11039 	    __cr4_reserved_bits(cpu_has, &boot_cpu_data))
11040 		return -EIO;
11041 
11042 	return ops->check_processor_compatibility();
11043 }
11044 
11045 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
11046 {
11047 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
11048 }
11049 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
11050 
11051 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
11052 {
11053 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
11054 }
11055 
11056 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
11057 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
11058 
11059 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
11060 {
11061 	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
11062 
11063 	vcpu->arch.l1tf_flush_l1d = true;
11064 	if (pmu->version && unlikely(pmu->event_count)) {
11065 		pmu->need_cleanup = true;
11066 		kvm_make_request(KVM_REQ_PMU, vcpu);
11067 	}
11068 	static_call(kvm_x86_sched_in)(vcpu, cpu);
11069 }
11070 
11071 void kvm_arch_free_vm(struct kvm *kvm)
11072 {
11073 	kfree(to_kvm_hv(kvm)->hv_pa_pg);
11074 	vfree(kvm);
11075 }
11076 
11077 
11078 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
11079 {
11080 	if (type)
11081 		return -EINVAL;
11082 
11083 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
11084 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
11085 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
11086 	INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
11087 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
11088 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
11089 
11090 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
11091 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
11092 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
11093 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
11094 		&kvm->arch.irq_sources_bitmap);
11095 
11096 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
11097 	mutex_init(&kvm->arch.apic_map_lock);
11098 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
11099 
11100 	kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
11101 	pvclock_update_vm_gtod_copy(kvm);
11102 
11103 	kvm->arch.guest_can_read_msr_platform_info = true;
11104 
11105 #if IS_ENABLED(CONFIG_HYPERV)
11106 	spin_lock_init(&kvm->arch.hv_root_tdp_lock);
11107 	kvm->arch.hv_root_tdp = INVALID_PAGE;
11108 #endif
11109 
11110 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
11111 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
11112 
11113 	kvm_apicv_init(kvm);
11114 	kvm_hv_init_vm(kvm);
11115 	kvm_page_track_init(kvm);
11116 	kvm_mmu_init_vm(kvm);
11117 
11118 	return static_call(kvm_x86_vm_init)(kvm);
11119 }
11120 
11121 int kvm_arch_post_init_vm(struct kvm *kvm)
11122 {
11123 	return kvm_mmu_post_init_vm(kvm);
11124 }
11125 
11126 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
11127 {
11128 	vcpu_load(vcpu);
11129 	kvm_mmu_unload(vcpu);
11130 	vcpu_put(vcpu);
11131 }
11132 
11133 static void kvm_free_vcpus(struct kvm *kvm)
11134 {
11135 	unsigned int i;
11136 	struct kvm_vcpu *vcpu;
11137 
11138 	/*
11139 	 * Unpin any mmu pages first.
11140 	 */
11141 	kvm_for_each_vcpu(i, vcpu, kvm) {
11142 		kvm_clear_async_pf_completion_queue(vcpu);
11143 		kvm_unload_vcpu_mmu(vcpu);
11144 	}
11145 	kvm_for_each_vcpu(i, vcpu, kvm)
11146 		kvm_vcpu_destroy(vcpu);
11147 
11148 	mutex_lock(&kvm->lock);
11149 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
11150 		kvm->vcpus[i] = NULL;
11151 
11152 	atomic_set(&kvm->online_vcpus, 0);
11153 	mutex_unlock(&kvm->lock);
11154 }
11155 
11156 void kvm_arch_sync_events(struct kvm *kvm)
11157 {
11158 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
11159 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
11160 	kvm_free_pit(kvm);
11161 }
11162 
11163 #define  ERR_PTR_USR(e)  ((void __user *)ERR_PTR(e))
11164 
11165 /**
11166  * __x86_set_memory_region: Setup KVM internal memory slot
11167  *
11168  * @kvm: the kvm pointer to the VM.
11169  * @id: the slot ID to setup.
11170  * @gpa: the GPA to install the slot (unused when @size == 0).
11171  * @size: the size of the slot. Set to zero to uninstall a slot.
11172  *
11173  * This function helps to setup a KVM internal memory slot.  Specify
11174  * @size > 0 to install a new slot, while @size == 0 to uninstall a
11175  * slot.  The return code can be one of the following:
11176  *
11177  *   HVA:           on success (uninstall will return a bogus HVA)
11178  *   -errno:        on error
11179  *
11180  * The caller should always use IS_ERR() to check the return value
11181  * before use.  Note, the KVM internal memory slots are guaranteed to
11182  * remain valid and unchanged until the VM is destroyed, i.e., the
11183  * GPA->HVA translation will not change.  However, the HVA is a user
11184  * address, i.e. its accessibility is not guaranteed, and must be
11185  * accessed via __copy_{to,from}_user().
11186  */
11187 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
11188 				      u32 size)
11189 {
11190 	int i, r;
11191 	unsigned long hva, old_npages;
11192 	struct kvm_memslots *slots = kvm_memslots(kvm);
11193 	struct kvm_memory_slot *slot;
11194 
11195 	/* Called with kvm->slots_lock held.  */
11196 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
11197 		return ERR_PTR_USR(-EINVAL);
11198 
11199 	slot = id_to_memslot(slots, id);
11200 	if (size) {
11201 		if (slot && slot->npages)
11202 			return ERR_PTR_USR(-EEXIST);
11203 
11204 		/*
11205 		 * MAP_SHARED to prevent internal slot pages from being moved
11206 		 * by fork()/COW.
11207 		 */
11208 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
11209 			      MAP_SHARED | MAP_ANONYMOUS, 0);
11210 		if (IS_ERR((void *)hva))
11211 			return (void __user *)hva;
11212 	} else {
11213 		if (!slot || !slot->npages)
11214 			return NULL;
11215 
11216 		old_npages = slot->npages;
11217 		hva = slot->userspace_addr;
11218 	}
11219 
11220 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11221 		struct kvm_userspace_memory_region m;
11222 
11223 		m.slot = id | (i << 16);
11224 		m.flags = 0;
11225 		m.guest_phys_addr = gpa;
11226 		m.userspace_addr = hva;
11227 		m.memory_size = size;
11228 		r = __kvm_set_memory_region(kvm, &m);
11229 		if (r < 0)
11230 			return ERR_PTR_USR(r);
11231 	}
11232 
11233 	if (!size)
11234 		vm_munmap(hva, old_npages * PAGE_SIZE);
11235 
11236 	return (void __user *)hva;
11237 }
11238 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
11239 
11240 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
11241 {
11242 	kvm_mmu_pre_destroy_vm(kvm);
11243 }
11244 
11245 void kvm_arch_destroy_vm(struct kvm *kvm)
11246 {
11247 	if (current->mm == kvm->mm) {
11248 		/*
11249 		 * Free memory regions allocated on behalf of userspace,
11250 		 * unless the the memory map has changed due to process exit
11251 		 * or fd copying.
11252 		 */
11253 		mutex_lock(&kvm->slots_lock);
11254 		__x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
11255 					0, 0);
11256 		__x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
11257 					0, 0);
11258 		__x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
11259 		mutex_unlock(&kvm->slots_lock);
11260 	}
11261 	static_call_cond(kvm_x86_vm_destroy)(kvm);
11262 	kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
11263 	kvm_pic_destroy(kvm);
11264 	kvm_ioapic_destroy(kvm);
11265 	kvm_free_vcpus(kvm);
11266 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
11267 	kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
11268 	kvm_mmu_uninit_vm(kvm);
11269 	kvm_page_track_cleanup(kvm);
11270 	kvm_xen_destroy_vm(kvm);
11271 	kvm_hv_destroy_vm(kvm);
11272 }
11273 
11274 static void memslot_rmap_free(struct kvm_memory_slot *slot)
11275 {
11276 	int i;
11277 
11278 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11279 		kvfree(slot->arch.rmap[i]);
11280 		slot->arch.rmap[i] = NULL;
11281 	}
11282 }
11283 
11284 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
11285 {
11286 	int i;
11287 
11288 	memslot_rmap_free(slot);
11289 
11290 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11291 		kvfree(slot->arch.lpage_info[i - 1]);
11292 		slot->arch.lpage_info[i - 1] = NULL;
11293 	}
11294 
11295 	kvm_page_track_free_memslot(slot);
11296 }
11297 
11298 static int memslot_rmap_alloc(struct kvm_memory_slot *slot,
11299 			      unsigned long npages)
11300 {
11301 	const int sz = sizeof(*slot->arch.rmap[0]);
11302 	int i;
11303 
11304 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11305 		int level = i + 1;
11306 		int lpages = gfn_to_index(slot->base_gfn + npages - 1,
11307 					  slot->base_gfn, level) + 1;
11308 
11309 		WARN_ON(slot->arch.rmap[i]);
11310 
11311 		slot->arch.rmap[i] = kvcalloc(lpages, sz, GFP_KERNEL_ACCOUNT);
11312 		if (!slot->arch.rmap[i]) {
11313 			memslot_rmap_free(slot);
11314 			return -ENOMEM;
11315 		}
11316 	}
11317 
11318 	return 0;
11319 }
11320 
11321 int alloc_all_memslots_rmaps(struct kvm *kvm)
11322 {
11323 	struct kvm_memslots *slots;
11324 	struct kvm_memory_slot *slot;
11325 	int r, i;
11326 
11327 	/*
11328 	 * Check if memslots alreday have rmaps early before acquiring
11329 	 * the slots_arch_lock below.
11330 	 */
11331 	if (kvm_memslots_have_rmaps(kvm))
11332 		return 0;
11333 
11334 	mutex_lock(&kvm->slots_arch_lock);
11335 
11336 	/*
11337 	 * Read memslots_have_rmaps again, under the slots arch lock,
11338 	 * before allocating the rmaps
11339 	 */
11340 	if (kvm_memslots_have_rmaps(kvm)) {
11341 		mutex_unlock(&kvm->slots_arch_lock);
11342 		return 0;
11343 	}
11344 
11345 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11346 		slots = __kvm_memslots(kvm, i);
11347 		kvm_for_each_memslot(slot, slots) {
11348 			r = memslot_rmap_alloc(slot, slot->npages);
11349 			if (r) {
11350 				mutex_unlock(&kvm->slots_arch_lock);
11351 				return r;
11352 			}
11353 		}
11354 	}
11355 
11356 	/*
11357 	 * Ensure that memslots_have_rmaps becomes true strictly after
11358 	 * all the rmap pointers are set.
11359 	 */
11360 	smp_store_release(&kvm->arch.memslots_have_rmaps, true);
11361 	mutex_unlock(&kvm->slots_arch_lock);
11362 	return 0;
11363 }
11364 
11365 static int kvm_alloc_memslot_metadata(struct kvm *kvm,
11366 				      struct kvm_memory_slot *slot,
11367 				      unsigned long npages)
11368 {
11369 	int i, r;
11370 
11371 	/*
11372 	 * Clear out the previous array pointers for the KVM_MR_MOVE case.  The
11373 	 * old arrays will be freed by __kvm_set_memory_region() if installing
11374 	 * the new memslot is successful.
11375 	 */
11376 	memset(&slot->arch, 0, sizeof(slot->arch));
11377 
11378 	if (kvm_memslots_have_rmaps(kvm)) {
11379 		r = memslot_rmap_alloc(slot, npages);
11380 		if (r)
11381 			return r;
11382 	}
11383 
11384 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11385 		struct kvm_lpage_info *linfo;
11386 		unsigned long ugfn;
11387 		int lpages;
11388 		int level = i + 1;
11389 
11390 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
11391 				      slot->base_gfn, level) + 1;
11392 
11393 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
11394 		if (!linfo)
11395 			goto out_free;
11396 
11397 		slot->arch.lpage_info[i - 1] = linfo;
11398 
11399 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
11400 			linfo[0].disallow_lpage = 1;
11401 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
11402 			linfo[lpages - 1].disallow_lpage = 1;
11403 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
11404 		/*
11405 		 * If the gfn and userspace address are not aligned wrt each
11406 		 * other, disable large page support for this slot.
11407 		 */
11408 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
11409 			unsigned long j;
11410 
11411 			for (j = 0; j < lpages; ++j)
11412 				linfo[j].disallow_lpage = 1;
11413 		}
11414 	}
11415 
11416 	if (kvm_page_track_create_memslot(slot, npages))
11417 		goto out_free;
11418 
11419 	return 0;
11420 
11421 out_free:
11422 	memslot_rmap_free(slot);
11423 
11424 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11425 		kvfree(slot->arch.lpage_info[i - 1]);
11426 		slot->arch.lpage_info[i - 1] = NULL;
11427 	}
11428 	return -ENOMEM;
11429 }
11430 
11431 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
11432 {
11433 	struct kvm_vcpu *vcpu;
11434 	int i;
11435 
11436 	/*
11437 	 * memslots->generation has been incremented.
11438 	 * mmio generation may have reached its maximum value.
11439 	 */
11440 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
11441 
11442 	/* Force re-initialization of steal_time cache */
11443 	kvm_for_each_vcpu(i, vcpu, kvm)
11444 		kvm_vcpu_kick(vcpu);
11445 }
11446 
11447 int kvm_arch_prepare_memory_region(struct kvm *kvm,
11448 				struct kvm_memory_slot *memslot,
11449 				const struct kvm_userspace_memory_region *mem,
11450 				enum kvm_mr_change change)
11451 {
11452 	if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
11453 		return kvm_alloc_memslot_metadata(kvm, memslot,
11454 						  mem->memory_size >> PAGE_SHIFT);
11455 	return 0;
11456 }
11457 
11458 
11459 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
11460 {
11461 	struct kvm_arch *ka = &kvm->arch;
11462 
11463 	if (!kvm_x86_ops.cpu_dirty_log_size)
11464 		return;
11465 
11466 	if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
11467 	    (!enable && --ka->cpu_dirty_logging_count == 0))
11468 		kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
11469 
11470 	WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
11471 }
11472 
11473 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
11474 				     struct kvm_memory_slot *old,
11475 				     struct kvm_memory_slot *new,
11476 				     enum kvm_mr_change change)
11477 {
11478 	bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES;
11479 
11480 	/*
11481 	 * Update CPU dirty logging if dirty logging is being toggled.  This
11482 	 * applies to all operations.
11483 	 */
11484 	if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)
11485 		kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
11486 
11487 	/*
11488 	 * Nothing more to do for RO slots (which can't be dirtied and can't be
11489 	 * made writable) or CREATE/MOVE/DELETE of a slot.
11490 	 *
11491 	 * For a memslot with dirty logging disabled:
11492 	 * CREATE:      No dirty mappings will already exist.
11493 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
11494 	 *		kvm_arch_flush_shadow_memslot()
11495 	 *
11496 	 * For a memslot with dirty logging enabled:
11497 	 * CREATE:      No shadow pages exist, thus nothing to write-protect
11498 	 *		and no dirty bits to clear.
11499 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
11500 	 *		kvm_arch_flush_shadow_memslot().
11501 	 */
11502 	if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
11503 		return;
11504 
11505 	/*
11506 	 * READONLY and non-flags changes were filtered out above, and the only
11507 	 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
11508 	 * logging isn't being toggled on or off.
11509 	 */
11510 	if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)))
11511 		return;
11512 
11513 	if (!log_dirty_pages) {
11514 		/*
11515 		 * Dirty logging tracks sptes in 4k granularity, meaning that
11516 		 * large sptes have to be split.  If live migration succeeds,
11517 		 * the guest in the source machine will be destroyed and large
11518 		 * sptes will be created in the destination.  However, if the
11519 		 * guest continues to run in the source machine (for example if
11520 		 * live migration fails), small sptes will remain around and
11521 		 * cause bad performance.
11522 		 *
11523 		 * Scan sptes if dirty logging has been stopped, dropping those
11524 		 * which can be collapsed into a single large-page spte.  Later
11525 		 * page faults will create the large-page sptes.
11526 		 */
11527 		kvm_mmu_zap_collapsible_sptes(kvm, new);
11528 	} else {
11529 		/*
11530 		 * Initially-all-set does not require write protecting any page,
11531 		 * because they're all assumed to be dirty.
11532 		 */
11533 		if (kvm_dirty_log_manual_protect_and_init_set(kvm))
11534 			return;
11535 
11536 		if (kvm_x86_ops.cpu_dirty_log_size) {
11537 			kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11538 			kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M);
11539 		} else {
11540 			kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
11541 		}
11542 	}
11543 }
11544 
11545 void kvm_arch_commit_memory_region(struct kvm *kvm,
11546 				const struct kvm_userspace_memory_region *mem,
11547 				struct kvm_memory_slot *old,
11548 				const struct kvm_memory_slot *new,
11549 				enum kvm_mr_change change)
11550 {
11551 	if (!kvm->arch.n_requested_mmu_pages)
11552 		kvm_mmu_change_mmu_pages(kvm,
11553 				kvm_mmu_calculate_default_mmu_pages(kvm));
11554 
11555 	/*
11556 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
11557 	 */
11558 	kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change);
11559 
11560 	/* Free the arrays associated with the old memslot. */
11561 	if (change == KVM_MR_MOVE)
11562 		kvm_arch_free_memslot(kvm, old);
11563 }
11564 
11565 void kvm_arch_flush_shadow_all(struct kvm *kvm)
11566 {
11567 	kvm_mmu_zap_all(kvm);
11568 }
11569 
11570 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
11571 				   struct kvm_memory_slot *slot)
11572 {
11573 	kvm_page_track_flush_slot(kvm, slot);
11574 }
11575 
11576 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
11577 {
11578 	return (is_guest_mode(vcpu) &&
11579 			kvm_x86_ops.guest_apic_has_interrupt &&
11580 			static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
11581 }
11582 
11583 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
11584 {
11585 	if (!list_empty_careful(&vcpu->async_pf.done))
11586 		return true;
11587 
11588 	if (kvm_apic_has_events(vcpu))
11589 		return true;
11590 
11591 	if (vcpu->arch.pv.pv_unhalted)
11592 		return true;
11593 
11594 	if (vcpu->arch.exception.pending)
11595 		return true;
11596 
11597 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11598 	    (vcpu->arch.nmi_pending &&
11599 	     static_call(kvm_x86_nmi_allowed)(vcpu, false)))
11600 		return true;
11601 
11602 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
11603 	    (vcpu->arch.smi_pending &&
11604 	     static_call(kvm_x86_smi_allowed)(vcpu, false)))
11605 		return true;
11606 
11607 	if (kvm_arch_interrupt_allowed(vcpu) &&
11608 	    (kvm_cpu_has_interrupt(vcpu) ||
11609 	    kvm_guest_apic_has_interrupt(vcpu)))
11610 		return true;
11611 
11612 	if (kvm_hv_has_stimer_pending(vcpu))
11613 		return true;
11614 
11615 	if (is_guest_mode(vcpu) &&
11616 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
11617 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
11618 		return true;
11619 
11620 	return false;
11621 }
11622 
11623 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
11624 {
11625 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
11626 }
11627 
11628 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
11629 {
11630 	if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11631 		return true;
11632 
11633 	return false;
11634 }
11635 
11636 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
11637 {
11638 	if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
11639 		return true;
11640 
11641 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11642 		kvm_test_request(KVM_REQ_SMI, vcpu) ||
11643 		 kvm_test_request(KVM_REQ_EVENT, vcpu))
11644 		return true;
11645 
11646 	return kvm_arch_dy_has_pending_interrupt(vcpu);
11647 }
11648 
11649 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
11650 {
11651 	if (vcpu->arch.guest_state_protected)
11652 		return true;
11653 
11654 	return vcpu->arch.preempted_in_kernel;
11655 }
11656 
11657 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
11658 {
11659 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
11660 }
11661 
11662 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
11663 {
11664 	return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
11665 }
11666 
11667 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
11668 {
11669 	/* Can't read the RIP when guest state is protected, just return 0 */
11670 	if (vcpu->arch.guest_state_protected)
11671 		return 0;
11672 
11673 	if (is_64_bit_mode(vcpu))
11674 		return kvm_rip_read(vcpu);
11675 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
11676 		     kvm_rip_read(vcpu));
11677 }
11678 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
11679 
11680 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
11681 {
11682 	return kvm_get_linear_rip(vcpu) == linear_rip;
11683 }
11684 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
11685 
11686 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
11687 {
11688 	unsigned long rflags;
11689 
11690 	rflags = static_call(kvm_x86_get_rflags)(vcpu);
11691 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11692 		rflags &= ~X86_EFLAGS_TF;
11693 	return rflags;
11694 }
11695 EXPORT_SYMBOL_GPL(kvm_get_rflags);
11696 
11697 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11698 {
11699 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
11700 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
11701 		rflags |= X86_EFLAGS_TF;
11702 	static_call(kvm_x86_set_rflags)(vcpu, rflags);
11703 }
11704 
11705 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11706 {
11707 	__kvm_set_rflags(vcpu, rflags);
11708 	kvm_make_request(KVM_REQ_EVENT, vcpu);
11709 }
11710 EXPORT_SYMBOL_GPL(kvm_set_rflags);
11711 
11712 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
11713 {
11714 	int r;
11715 
11716 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
11717 	      work->wakeup_all)
11718 		return;
11719 
11720 	r = kvm_mmu_reload(vcpu);
11721 	if (unlikely(r))
11722 		return;
11723 
11724 	if (!vcpu->arch.mmu->direct_map &&
11725 	      work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
11726 		return;
11727 
11728 	kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
11729 }
11730 
11731 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
11732 {
11733 	BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
11734 
11735 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
11736 }
11737 
11738 static inline u32 kvm_async_pf_next_probe(u32 key)
11739 {
11740 	return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
11741 }
11742 
11743 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11744 {
11745 	u32 key = kvm_async_pf_hash_fn(gfn);
11746 
11747 	while (vcpu->arch.apf.gfns[key] != ~0)
11748 		key = kvm_async_pf_next_probe(key);
11749 
11750 	vcpu->arch.apf.gfns[key] = gfn;
11751 }
11752 
11753 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
11754 {
11755 	int i;
11756 	u32 key = kvm_async_pf_hash_fn(gfn);
11757 
11758 	for (i = 0; i < ASYNC_PF_PER_VCPU &&
11759 		     (vcpu->arch.apf.gfns[key] != gfn &&
11760 		      vcpu->arch.apf.gfns[key] != ~0); i++)
11761 		key = kvm_async_pf_next_probe(key);
11762 
11763 	return key;
11764 }
11765 
11766 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11767 {
11768 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
11769 }
11770 
11771 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11772 {
11773 	u32 i, j, k;
11774 
11775 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
11776 
11777 	if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
11778 		return;
11779 
11780 	while (true) {
11781 		vcpu->arch.apf.gfns[i] = ~0;
11782 		do {
11783 			j = kvm_async_pf_next_probe(j);
11784 			if (vcpu->arch.apf.gfns[j] == ~0)
11785 				return;
11786 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
11787 			/*
11788 			 * k lies cyclically in ]i,j]
11789 			 * |    i.k.j |
11790 			 * |....j i.k.| or  |.k..j i...|
11791 			 */
11792 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
11793 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11794 		i = j;
11795 	}
11796 }
11797 
11798 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11799 {
11800 	u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11801 
11802 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11803 				      sizeof(reason));
11804 }
11805 
11806 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11807 {
11808 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11809 
11810 	return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11811 					     &token, offset, sizeof(token));
11812 }
11813 
11814 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
11815 {
11816 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11817 	u32 val;
11818 
11819 	if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11820 					 &val, offset, sizeof(val)))
11821 		return false;
11822 
11823 	return !val;
11824 }
11825 
11826 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
11827 {
11828 	if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
11829 		return false;
11830 
11831 	if (!kvm_pv_async_pf_enabled(vcpu) ||
11832 	    (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
11833 		return false;
11834 
11835 	return true;
11836 }
11837 
11838 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
11839 {
11840 	if (unlikely(!lapic_in_kernel(vcpu) ||
11841 		     kvm_event_needs_reinjection(vcpu) ||
11842 		     vcpu->arch.exception.pending))
11843 		return false;
11844 
11845 	if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
11846 		return false;
11847 
11848 	/*
11849 	 * If interrupts are off we cannot even use an artificial
11850 	 * halt state.
11851 	 */
11852 	return kvm_arch_interrupt_allowed(vcpu);
11853 }
11854 
11855 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
11856 				     struct kvm_async_pf *work)
11857 {
11858 	struct x86_exception fault;
11859 
11860 	trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
11861 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
11862 
11863 	if (kvm_can_deliver_async_pf(vcpu) &&
11864 	    !apf_put_user_notpresent(vcpu)) {
11865 		fault.vector = PF_VECTOR;
11866 		fault.error_code_valid = true;
11867 		fault.error_code = 0;
11868 		fault.nested_page_fault = false;
11869 		fault.address = work->arch.token;
11870 		fault.async_page_fault = true;
11871 		kvm_inject_page_fault(vcpu, &fault);
11872 		return true;
11873 	} else {
11874 		/*
11875 		 * It is not possible to deliver a paravirtualized asynchronous
11876 		 * page fault, but putting the guest in an artificial halt state
11877 		 * can be beneficial nevertheless: if an interrupt arrives, we
11878 		 * can deliver it timely and perhaps the guest will schedule
11879 		 * another process.  When the instruction that triggered a page
11880 		 * fault is retried, hopefully the page will be ready in the host.
11881 		 */
11882 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
11883 		return false;
11884 	}
11885 }
11886 
11887 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
11888 				 struct kvm_async_pf *work)
11889 {
11890 	struct kvm_lapic_irq irq = {
11891 		.delivery_mode = APIC_DM_FIXED,
11892 		.vector = vcpu->arch.apf.vec
11893 	};
11894 
11895 	if (work->wakeup_all)
11896 		work->arch.token = ~0; /* broadcast wakeup */
11897 	else
11898 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
11899 	trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
11900 
11901 	if ((work->wakeup_all || work->notpresent_injected) &&
11902 	    kvm_pv_async_pf_enabled(vcpu) &&
11903 	    !apf_put_user_ready(vcpu, work->arch.token)) {
11904 		vcpu->arch.apf.pageready_pending = true;
11905 		kvm_apic_set_irq(vcpu, &irq, NULL);
11906 	}
11907 
11908 	vcpu->arch.apf.halted = false;
11909 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11910 }
11911 
11912 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
11913 {
11914 	kvm_make_request(KVM_REQ_APF_READY, vcpu);
11915 	if (!vcpu->arch.apf.pageready_pending)
11916 		kvm_vcpu_kick(vcpu);
11917 }
11918 
11919 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
11920 {
11921 	if (!kvm_pv_async_pf_enabled(vcpu))
11922 		return true;
11923 	else
11924 		return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
11925 }
11926 
11927 void kvm_arch_start_assignment(struct kvm *kvm)
11928 {
11929 	if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
11930 		static_call_cond(kvm_x86_start_assignment)(kvm);
11931 }
11932 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
11933 
11934 void kvm_arch_end_assignment(struct kvm *kvm)
11935 {
11936 	atomic_dec(&kvm->arch.assigned_device_count);
11937 }
11938 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
11939 
11940 bool kvm_arch_has_assigned_device(struct kvm *kvm)
11941 {
11942 	return atomic_read(&kvm->arch.assigned_device_count);
11943 }
11944 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
11945 
11946 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
11947 {
11948 	atomic_inc(&kvm->arch.noncoherent_dma_count);
11949 }
11950 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
11951 
11952 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
11953 {
11954 	atomic_dec(&kvm->arch.noncoherent_dma_count);
11955 }
11956 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
11957 
11958 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
11959 {
11960 	return atomic_read(&kvm->arch.noncoherent_dma_count);
11961 }
11962 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
11963 
11964 bool kvm_arch_has_irq_bypass(void)
11965 {
11966 	return true;
11967 }
11968 
11969 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
11970 				      struct irq_bypass_producer *prod)
11971 {
11972 	struct kvm_kernel_irqfd *irqfd =
11973 		container_of(cons, struct kvm_kernel_irqfd, consumer);
11974 	int ret;
11975 
11976 	irqfd->producer = prod;
11977 	kvm_arch_start_assignment(irqfd->kvm);
11978 	ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
11979 					 prod->irq, irqfd->gsi, 1);
11980 
11981 	if (ret)
11982 		kvm_arch_end_assignment(irqfd->kvm);
11983 
11984 	return ret;
11985 }
11986 
11987 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
11988 				      struct irq_bypass_producer *prod)
11989 {
11990 	int ret;
11991 	struct kvm_kernel_irqfd *irqfd =
11992 		container_of(cons, struct kvm_kernel_irqfd, consumer);
11993 
11994 	WARN_ON(irqfd->producer != prod);
11995 	irqfd->producer = NULL;
11996 
11997 	/*
11998 	 * When producer of consumer is unregistered, we change back to
11999 	 * remapped mode, so we can re-use the current implementation
12000 	 * when the irq is masked/disabled or the consumer side (KVM
12001 	 * int this case doesn't want to receive the interrupts.
12002 	*/
12003 	ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
12004 	if (ret)
12005 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
12006 		       " fails: %d\n", irqfd->consumer.token, ret);
12007 
12008 	kvm_arch_end_assignment(irqfd->kvm);
12009 }
12010 
12011 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
12012 				   uint32_t guest_irq, bool set)
12013 {
12014 	return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
12015 }
12016 
12017 bool kvm_vector_hashing_enabled(void)
12018 {
12019 	return vector_hashing;
12020 }
12021 
12022 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
12023 {
12024 	return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
12025 }
12026 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
12027 
12028 
12029 int kvm_spec_ctrl_test_value(u64 value)
12030 {
12031 	/*
12032 	 * test that setting IA32_SPEC_CTRL to given value
12033 	 * is allowed by the host processor
12034 	 */
12035 
12036 	u64 saved_value;
12037 	unsigned long flags;
12038 	int ret = 0;
12039 
12040 	local_irq_save(flags);
12041 
12042 	if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
12043 		ret = 1;
12044 	else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
12045 		ret = 1;
12046 	else
12047 		wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
12048 
12049 	local_irq_restore(flags);
12050 
12051 	return ret;
12052 }
12053 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
12054 
12055 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
12056 {
12057 	struct x86_exception fault;
12058 	u32 access = error_code &
12059 		(PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
12060 
12061 	if (!(error_code & PFERR_PRESENT_MASK) ||
12062 	    vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
12063 		/*
12064 		 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
12065 		 * tables probably do not match the TLB.  Just proceed
12066 		 * with the error code that the processor gave.
12067 		 */
12068 		fault.vector = PF_VECTOR;
12069 		fault.error_code_valid = true;
12070 		fault.error_code = error_code;
12071 		fault.nested_page_fault = false;
12072 		fault.address = gva;
12073 	}
12074 	vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
12075 }
12076 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
12077 
12078 /*
12079  * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
12080  * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
12081  * indicates whether exit to userspace is needed.
12082  */
12083 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
12084 			      struct x86_exception *e)
12085 {
12086 	if (r == X86EMUL_PROPAGATE_FAULT) {
12087 		kvm_inject_emulated_page_fault(vcpu, e);
12088 		return 1;
12089 	}
12090 
12091 	/*
12092 	 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
12093 	 * while handling a VMX instruction KVM could've handled the request
12094 	 * correctly by exiting to userspace and performing I/O but there
12095 	 * doesn't seem to be a real use-case behind such requests, just return
12096 	 * KVM_EXIT_INTERNAL_ERROR for now.
12097 	 */
12098 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
12099 	vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
12100 	vcpu->run->internal.ndata = 0;
12101 
12102 	return 0;
12103 }
12104 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
12105 
12106 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
12107 {
12108 	bool pcid_enabled;
12109 	struct x86_exception e;
12110 	struct {
12111 		u64 pcid;
12112 		u64 gla;
12113 	} operand;
12114 	int r;
12115 
12116 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
12117 	if (r != X86EMUL_CONTINUE)
12118 		return kvm_handle_memory_failure(vcpu, r, &e);
12119 
12120 	if (operand.pcid >> 12 != 0) {
12121 		kvm_inject_gp(vcpu, 0);
12122 		return 1;
12123 	}
12124 
12125 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
12126 
12127 	switch (type) {
12128 	case INVPCID_TYPE_INDIV_ADDR:
12129 		if ((!pcid_enabled && (operand.pcid != 0)) ||
12130 		    is_noncanonical_address(operand.gla, vcpu)) {
12131 			kvm_inject_gp(vcpu, 0);
12132 			return 1;
12133 		}
12134 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
12135 		return kvm_skip_emulated_instruction(vcpu);
12136 
12137 	case INVPCID_TYPE_SINGLE_CTXT:
12138 		if (!pcid_enabled && (operand.pcid != 0)) {
12139 			kvm_inject_gp(vcpu, 0);
12140 			return 1;
12141 		}
12142 
12143 		kvm_invalidate_pcid(vcpu, operand.pcid);
12144 		return kvm_skip_emulated_instruction(vcpu);
12145 
12146 	case INVPCID_TYPE_ALL_NON_GLOBAL:
12147 		/*
12148 		 * Currently, KVM doesn't mark global entries in the shadow
12149 		 * page tables, so a non-global flush just degenerates to a
12150 		 * global flush. If needed, we could optimize this later by
12151 		 * keeping track of global entries in shadow page tables.
12152 		 */
12153 
12154 		fallthrough;
12155 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
12156 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
12157 		return kvm_skip_emulated_instruction(vcpu);
12158 
12159 	default:
12160 		BUG(); /* We have already checked above that type <= 3 */
12161 	}
12162 }
12163 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
12164 
12165 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
12166 {
12167 	struct kvm_run *run = vcpu->run;
12168 	struct kvm_mmio_fragment *frag;
12169 	unsigned int len;
12170 
12171 	BUG_ON(!vcpu->mmio_needed);
12172 
12173 	/* Complete previous fragment */
12174 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
12175 	len = min(8u, frag->len);
12176 	if (!vcpu->mmio_is_write)
12177 		memcpy(frag->data, run->mmio.data, len);
12178 
12179 	if (frag->len <= 8) {
12180 		/* Switch to the next fragment. */
12181 		frag++;
12182 		vcpu->mmio_cur_fragment++;
12183 	} else {
12184 		/* Go forward to the next mmio piece. */
12185 		frag->data += len;
12186 		frag->gpa += len;
12187 		frag->len -= len;
12188 	}
12189 
12190 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
12191 		vcpu->mmio_needed = 0;
12192 
12193 		// VMG change, at this point, we're always done
12194 		// RIP has already been advanced
12195 		return 1;
12196 	}
12197 
12198 	// More MMIO is needed
12199 	run->mmio.phys_addr = frag->gpa;
12200 	run->mmio.len = min(8u, frag->len);
12201 	run->mmio.is_write = vcpu->mmio_is_write;
12202 	if (run->mmio.is_write)
12203 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
12204 	run->exit_reason = KVM_EXIT_MMIO;
12205 
12206 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12207 
12208 	return 0;
12209 }
12210 
12211 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12212 			  void *data)
12213 {
12214 	int handled;
12215 	struct kvm_mmio_fragment *frag;
12216 
12217 	if (!data)
12218 		return -EINVAL;
12219 
12220 	handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12221 	if (handled == bytes)
12222 		return 1;
12223 
12224 	bytes -= handled;
12225 	gpa += handled;
12226 	data += handled;
12227 
12228 	/*TODO: Check if need to increment number of frags */
12229 	frag = vcpu->mmio_fragments;
12230 	vcpu->mmio_nr_fragments = 1;
12231 	frag->len = bytes;
12232 	frag->gpa = gpa;
12233 	frag->data = data;
12234 
12235 	vcpu->mmio_needed = 1;
12236 	vcpu->mmio_cur_fragment = 0;
12237 
12238 	vcpu->run->mmio.phys_addr = gpa;
12239 	vcpu->run->mmio.len = min(8u, frag->len);
12240 	vcpu->run->mmio.is_write = 1;
12241 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
12242 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
12243 
12244 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12245 
12246 	return 0;
12247 }
12248 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
12249 
12250 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12251 			 void *data)
12252 {
12253 	int handled;
12254 	struct kvm_mmio_fragment *frag;
12255 
12256 	if (!data)
12257 		return -EINVAL;
12258 
12259 	handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12260 	if (handled == bytes)
12261 		return 1;
12262 
12263 	bytes -= handled;
12264 	gpa += handled;
12265 	data += handled;
12266 
12267 	/*TODO: Check if need to increment number of frags */
12268 	frag = vcpu->mmio_fragments;
12269 	vcpu->mmio_nr_fragments = 1;
12270 	frag->len = bytes;
12271 	frag->gpa = gpa;
12272 	frag->data = data;
12273 
12274 	vcpu->mmio_needed = 1;
12275 	vcpu->mmio_cur_fragment = 0;
12276 
12277 	vcpu->run->mmio.phys_addr = gpa;
12278 	vcpu->run->mmio.len = min(8u, frag->len);
12279 	vcpu->run->mmio.is_write = 0;
12280 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
12281 
12282 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12283 
12284 	return 0;
12285 }
12286 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
12287 
12288 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12289 {
12290 	memcpy(vcpu->arch.guest_ins_data, vcpu->arch.pio_data,
12291 	       vcpu->arch.pio.count * vcpu->arch.pio.size);
12292 	vcpu->arch.pio.count = 0;
12293 
12294 	return 1;
12295 }
12296 
12297 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12298 			   unsigned int port, void *data,  unsigned int count)
12299 {
12300 	int ret;
12301 
12302 	ret = emulator_pio_out_emulated(vcpu->arch.emulate_ctxt, size, port,
12303 					data, count);
12304 	if (ret)
12305 		return ret;
12306 
12307 	vcpu->arch.pio.count = 0;
12308 
12309 	return 0;
12310 }
12311 
12312 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12313 			  unsigned int port, void *data, unsigned int count)
12314 {
12315 	int ret;
12316 
12317 	ret = emulator_pio_in_emulated(vcpu->arch.emulate_ctxt, size, port,
12318 				       data, count);
12319 	if (ret) {
12320 		vcpu->arch.pio.count = 0;
12321 	} else {
12322 		vcpu->arch.guest_ins_data = data;
12323 		vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
12324 	}
12325 
12326 	return 0;
12327 }
12328 
12329 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
12330 			 unsigned int port, void *data,  unsigned int count,
12331 			 int in)
12332 {
12333 	return in ? kvm_sev_es_ins(vcpu, size, port, data, count)
12334 		  : kvm_sev_es_outs(vcpu, size, port, data, count);
12335 }
12336 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
12337 
12338 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
12339 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
12340 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
12341 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
12342 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
12343 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
12344 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
12345 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
12346 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
12347 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
12348 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
12349 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
12350 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
12351 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
12352 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
12353 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
12354 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
12355 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
12356 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
12357 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
12358 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
12359 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
12360 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
12361 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
12362 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
12363 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
12364 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);
12365