xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 901181b7)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * derived from drivers/kvm/kvm_main.c
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright (C) 2008 Qumranet, Inc.
9  * Copyright IBM Corporation, 2008
10  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11  *
12  * Authors:
13  *   Avi Kivity   <avi@qumranet.com>
14  *   Yaniv Kamay  <yaniv@qumranet.com>
15  *   Amit Shah    <amit.shah@qumranet.com>
16  *   Ben-Ami Yassour <benami@il.ibm.com>
17  */
18 
19 #include <linux/kvm_host.h>
20 #include "irq.h"
21 #include "ioapic.h"
22 #include "mmu.h"
23 #include "i8254.h"
24 #include "tss.h"
25 #include "kvm_cache_regs.h"
26 #include "kvm_emulate.h"
27 #include "x86.h"
28 #include "cpuid.h"
29 #include "pmu.h"
30 #include "hyperv.h"
31 #include "lapic.h"
32 #include "xen.h"
33 
34 #include <linux/clocksource.h>
35 #include <linux/interrupt.h>
36 #include <linux/kvm.h>
37 #include <linux/fs.h>
38 #include <linux/vmalloc.h>
39 #include <linux/export.h>
40 #include <linux/moduleparam.h>
41 #include <linux/mman.h>
42 #include <linux/highmem.h>
43 #include <linux/iommu.h>
44 #include <linux/intel-iommu.h>
45 #include <linux/cpufreq.h>
46 #include <linux/user-return-notifier.h>
47 #include <linux/srcu.h>
48 #include <linux/slab.h>
49 #include <linux/perf_event.h>
50 #include <linux/uaccess.h>
51 #include <linux/hash.h>
52 #include <linux/pci.h>
53 #include <linux/timekeeper_internal.h>
54 #include <linux/pvclock_gtod.h>
55 #include <linux/kvm_irqfd.h>
56 #include <linux/irqbypass.h>
57 #include <linux/sched/stat.h>
58 #include <linux/sched/isolation.h>
59 #include <linux/mem_encrypt.h>
60 #include <linux/entry-kvm.h>
61 #include <linux/suspend.h>
62 
63 #include <trace/events/kvm.h>
64 
65 #include <asm/debugreg.h>
66 #include <asm/msr.h>
67 #include <asm/desc.h>
68 #include <asm/mce.h>
69 #include <asm/pkru.h>
70 #include <linux/kernel_stat.h>
71 #include <asm/fpu/api.h>
72 #include <asm/fpu/xcr.h>
73 #include <asm/fpu/xstate.h>
74 #include <asm/pvclock.h>
75 #include <asm/div64.h>
76 #include <asm/irq_remapping.h>
77 #include <asm/mshyperv.h>
78 #include <asm/hypervisor.h>
79 #include <asm/tlbflush.h>
80 #include <asm/intel_pt.h>
81 #include <asm/emulate_prefix.h>
82 #include <asm/sgx.h>
83 #include <clocksource/hyperv_timer.h>
84 
85 #define CREATE_TRACE_POINTS
86 #include "trace.h"
87 
88 #define MAX_IO_MSRS 256
89 #define KVM_MAX_MCE_BANKS 32
90 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
91 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
92 
93 #define emul_to_vcpu(ctxt) \
94 	((struct kvm_vcpu *)(ctxt)->vcpu)
95 
96 /* EFER defaults:
97  * - enable syscall per default because its emulated by KVM
98  * - enable LME and LMA per default on 64 bit KVM
99  */
100 #ifdef CONFIG_X86_64
101 static
102 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
103 #else
104 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
105 #endif
106 
107 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
108 
109 #define KVM_EXIT_HYPERCALL_VALID_MASK (1 << KVM_HC_MAP_GPA_RANGE)
110 
111 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
112                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
113 
114 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
115 static void process_nmi(struct kvm_vcpu *vcpu);
116 static void process_smi(struct kvm_vcpu *vcpu);
117 static void enter_smm(struct kvm_vcpu *vcpu);
118 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
119 static void store_regs(struct kvm_vcpu *vcpu);
120 static int sync_regs(struct kvm_vcpu *vcpu);
121 
122 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
123 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
124 
125 struct kvm_x86_ops kvm_x86_ops __read_mostly;
126 EXPORT_SYMBOL_GPL(kvm_x86_ops);
127 
128 #define KVM_X86_OP(func)					     \
129 	DEFINE_STATIC_CALL_NULL(kvm_x86_##func,			     \
130 				*(((struct kvm_x86_ops *)0)->func));
131 #define KVM_X86_OP_NULL KVM_X86_OP
132 #include <asm/kvm-x86-ops.h>
133 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
134 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
135 EXPORT_STATIC_CALL_GPL(kvm_x86_tlb_flush_current);
136 
137 static bool __read_mostly ignore_msrs = 0;
138 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
139 
140 bool __read_mostly report_ignored_msrs = true;
141 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
142 EXPORT_SYMBOL_GPL(report_ignored_msrs);
143 
144 unsigned int min_timer_period_us = 200;
145 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
146 
147 static bool __read_mostly kvmclock_periodic_sync = true;
148 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
149 
150 bool __read_mostly kvm_has_tsc_control;
151 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
152 u32  __read_mostly kvm_max_guest_tsc_khz;
153 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
154 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
155 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
156 u64  __read_mostly kvm_max_tsc_scaling_ratio;
157 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
158 u64 __read_mostly kvm_default_tsc_scaling_ratio;
159 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
160 bool __read_mostly kvm_has_bus_lock_exit;
161 EXPORT_SYMBOL_GPL(kvm_has_bus_lock_exit);
162 
163 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
164 static u32 __read_mostly tsc_tolerance_ppm = 250;
165 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
166 
167 /*
168  * lapic timer advance (tscdeadline mode only) in nanoseconds.  '-1' enables
169  * adaptive tuning starting from default advancement of 1000ns.  '0' disables
170  * advancement entirely.  Any other value is used as-is and disables adaptive
171  * tuning, i.e. allows privileged userspace to set an exact advancement time.
172  */
173 static int __read_mostly lapic_timer_advance_ns = -1;
174 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
175 
176 static bool __read_mostly vector_hashing = true;
177 module_param(vector_hashing, bool, S_IRUGO);
178 
179 bool __read_mostly enable_vmware_backdoor = false;
180 module_param(enable_vmware_backdoor, bool, S_IRUGO);
181 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
182 
183 static bool __read_mostly force_emulation_prefix = false;
184 module_param(force_emulation_prefix, bool, S_IRUGO);
185 
186 int __read_mostly pi_inject_timer = -1;
187 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
188 
189 /*
190  * Restoring the host value for MSRs that are only consumed when running in
191  * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
192  * returns to userspace, i.e. the kernel can run with the guest's value.
193  */
194 #define KVM_MAX_NR_USER_RETURN_MSRS 16
195 
196 struct kvm_user_return_msrs {
197 	struct user_return_notifier urn;
198 	bool registered;
199 	struct kvm_user_return_msr_values {
200 		u64 host;
201 		u64 curr;
202 	} values[KVM_MAX_NR_USER_RETURN_MSRS];
203 };
204 
205 u32 __read_mostly kvm_nr_uret_msrs;
206 EXPORT_SYMBOL_GPL(kvm_nr_uret_msrs);
207 static u32 __read_mostly kvm_uret_msrs_list[KVM_MAX_NR_USER_RETURN_MSRS];
208 static struct kvm_user_return_msrs __percpu *user_return_msrs;
209 
210 #define KVM_SUPPORTED_XCR0     (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
211 				| XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
212 				| XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
213 				| XFEATURE_MASK_PKRU)
214 
215 u64 __read_mostly host_efer;
216 EXPORT_SYMBOL_GPL(host_efer);
217 
218 bool __read_mostly allow_smaller_maxphyaddr = 0;
219 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
220 
221 bool __read_mostly enable_apicv = true;
222 EXPORT_SYMBOL_GPL(enable_apicv);
223 
224 u64 __read_mostly host_xss;
225 EXPORT_SYMBOL_GPL(host_xss);
226 u64 __read_mostly supported_xss;
227 EXPORT_SYMBOL_GPL(supported_xss);
228 
229 const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
230 	KVM_GENERIC_VM_STATS(),
231 	STATS_DESC_COUNTER(VM, mmu_shadow_zapped),
232 	STATS_DESC_COUNTER(VM, mmu_pte_write),
233 	STATS_DESC_COUNTER(VM, mmu_pde_zapped),
234 	STATS_DESC_COUNTER(VM, mmu_flooded),
235 	STATS_DESC_COUNTER(VM, mmu_recycled),
236 	STATS_DESC_COUNTER(VM, mmu_cache_miss),
237 	STATS_DESC_ICOUNTER(VM, mmu_unsync),
238 	STATS_DESC_ICOUNTER(VM, pages_4k),
239 	STATS_DESC_ICOUNTER(VM, pages_2m),
240 	STATS_DESC_ICOUNTER(VM, pages_1g),
241 	STATS_DESC_ICOUNTER(VM, nx_lpage_splits),
242 	STATS_DESC_PCOUNTER(VM, max_mmu_rmap_size),
243 	STATS_DESC_PCOUNTER(VM, max_mmu_page_hash_collisions)
244 };
245 
246 const struct kvm_stats_header kvm_vm_stats_header = {
247 	.name_size = KVM_STATS_NAME_SIZE,
248 	.num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
249 	.id_offset = sizeof(struct kvm_stats_header),
250 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
251 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
252 		       sizeof(kvm_vm_stats_desc),
253 };
254 
255 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
256 	KVM_GENERIC_VCPU_STATS(),
257 	STATS_DESC_COUNTER(VCPU, pf_fixed),
258 	STATS_DESC_COUNTER(VCPU, pf_guest),
259 	STATS_DESC_COUNTER(VCPU, tlb_flush),
260 	STATS_DESC_COUNTER(VCPU, invlpg),
261 	STATS_DESC_COUNTER(VCPU, exits),
262 	STATS_DESC_COUNTER(VCPU, io_exits),
263 	STATS_DESC_COUNTER(VCPU, mmio_exits),
264 	STATS_DESC_COUNTER(VCPU, signal_exits),
265 	STATS_DESC_COUNTER(VCPU, irq_window_exits),
266 	STATS_DESC_COUNTER(VCPU, nmi_window_exits),
267 	STATS_DESC_COUNTER(VCPU, l1d_flush),
268 	STATS_DESC_COUNTER(VCPU, halt_exits),
269 	STATS_DESC_COUNTER(VCPU, request_irq_exits),
270 	STATS_DESC_COUNTER(VCPU, irq_exits),
271 	STATS_DESC_COUNTER(VCPU, host_state_reload),
272 	STATS_DESC_COUNTER(VCPU, fpu_reload),
273 	STATS_DESC_COUNTER(VCPU, insn_emulation),
274 	STATS_DESC_COUNTER(VCPU, insn_emulation_fail),
275 	STATS_DESC_COUNTER(VCPU, hypercalls),
276 	STATS_DESC_COUNTER(VCPU, irq_injections),
277 	STATS_DESC_COUNTER(VCPU, nmi_injections),
278 	STATS_DESC_COUNTER(VCPU, req_event),
279 	STATS_DESC_COUNTER(VCPU, nested_run),
280 	STATS_DESC_COUNTER(VCPU, directed_yield_attempted),
281 	STATS_DESC_COUNTER(VCPU, directed_yield_successful),
282 	STATS_DESC_ICOUNTER(VCPU, guest_mode)
283 };
284 
285 const struct kvm_stats_header kvm_vcpu_stats_header = {
286 	.name_size = KVM_STATS_NAME_SIZE,
287 	.num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
288 	.id_offset = sizeof(struct kvm_stats_header),
289 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
290 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
291 		       sizeof(kvm_vcpu_stats_desc),
292 };
293 
294 u64 __read_mostly host_xcr0;
295 u64 __read_mostly supported_xcr0;
296 EXPORT_SYMBOL_GPL(supported_xcr0);
297 
298 static struct kmem_cache *x86_emulator_cache;
299 
300 /*
301  * When called, it means the previous get/set msr reached an invalid msr.
302  * Return true if we want to ignore/silent this failed msr access.
303  */
304 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
305 {
306 	const char *op = write ? "wrmsr" : "rdmsr";
307 
308 	if (ignore_msrs) {
309 		if (report_ignored_msrs)
310 			kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
311 				      op, msr, data);
312 		/* Mask the error */
313 		return true;
314 	} else {
315 		kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
316 				      op, msr, data);
317 		return false;
318 	}
319 }
320 
321 static struct kmem_cache *kvm_alloc_emulator_cache(void)
322 {
323 	unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
324 	unsigned int size = sizeof(struct x86_emulate_ctxt);
325 
326 	return kmem_cache_create_usercopy("x86_emulator", size,
327 					  __alignof__(struct x86_emulate_ctxt),
328 					  SLAB_ACCOUNT, useroffset,
329 					  size - useroffset, NULL);
330 }
331 
332 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
333 
334 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
335 {
336 	int i;
337 	for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
338 		vcpu->arch.apf.gfns[i] = ~0;
339 }
340 
341 static void kvm_on_user_return(struct user_return_notifier *urn)
342 {
343 	unsigned slot;
344 	struct kvm_user_return_msrs *msrs
345 		= container_of(urn, struct kvm_user_return_msrs, urn);
346 	struct kvm_user_return_msr_values *values;
347 	unsigned long flags;
348 
349 	/*
350 	 * Disabling irqs at this point since the following code could be
351 	 * interrupted and executed through kvm_arch_hardware_disable()
352 	 */
353 	local_irq_save(flags);
354 	if (msrs->registered) {
355 		msrs->registered = false;
356 		user_return_notifier_unregister(urn);
357 	}
358 	local_irq_restore(flags);
359 	for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
360 		values = &msrs->values[slot];
361 		if (values->host != values->curr) {
362 			wrmsrl(kvm_uret_msrs_list[slot], values->host);
363 			values->curr = values->host;
364 		}
365 	}
366 }
367 
368 static int kvm_probe_user_return_msr(u32 msr)
369 {
370 	u64 val;
371 	int ret;
372 
373 	preempt_disable();
374 	ret = rdmsrl_safe(msr, &val);
375 	if (ret)
376 		goto out;
377 	ret = wrmsrl_safe(msr, val);
378 out:
379 	preempt_enable();
380 	return ret;
381 }
382 
383 int kvm_add_user_return_msr(u32 msr)
384 {
385 	BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
386 
387 	if (kvm_probe_user_return_msr(msr))
388 		return -1;
389 
390 	kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
391 	return kvm_nr_uret_msrs++;
392 }
393 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
394 
395 int kvm_find_user_return_msr(u32 msr)
396 {
397 	int i;
398 
399 	for (i = 0; i < kvm_nr_uret_msrs; ++i) {
400 		if (kvm_uret_msrs_list[i] == msr)
401 			return i;
402 	}
403 	return -1;
404 }
405 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
406 
407 static void kvm_user_return_msr_cpu_online(void)
408 {
409 	unsigned int cpu = smp_processor_id();
410 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
411 	u64 value;
412 	int i;
413 
414 	for (i = 0; i < kvm_nr_uret_msrs; ++i) {
415 		rdmsrl_safe(kvm_uret_msrs_list[i], &value);
416 		msrs->values[i].host = value;
417 		msrs->values[i].curr = value;
418 	}
419 }
420 
421 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
422 {
423 	unsigned int cpu = smp_processor_id();
424 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
425 	int err;
426 
427 	value = (value & mask) | (msrs->values[slot].host & ~mask);
428 	if (value == msrs->values[slot].curr)
429 		return 0;
430 	err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
431 	if (err)
432 		return 1;
433 
434 	msrs->values[slot].curr = value;
435 	if (!msrs->registered) {
436 		msrs->urn.on_user_return = kvm_on_user_return;
437 		user_return_notifier_register(&msrs->urn);
438 		msrs->registered = true;
439 	}
440 	return 0;
441 }
442 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
443 
444 static void drop_user_return_notifiers(void)
445 {
446 	unsigned int cpu = smp_processor_id();
447 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
448 
449 	if (msrs->registered)
450 		kvm_on_user_return(&msrs->urn);
451 }
452 
453 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
454 {
455 	return vcpu->arch.apic_base;
456 }
457 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
458 
459 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
460 {
461 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
462 }
463 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
464 
465 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
466 {
467 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
468 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
469 	u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
470 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
471 
472 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
473 		return 1;
474 	if (!msr_info->host_initiated) {
475 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
476 			return 1;
477 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
478 			return 1;
479 	}
480 
481 	kvm_lapic_set_base(vcpu, msr_info->data);
482 	kvm_recalculate_apic_map(vcpu->kvm);
483 	return 0;
484 }
485 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
486 
487 /*
488  * Handle a fault on a hardware virtualization (VMX or SVM) instruction.
489  *
490  * Hardware virtualization extension instructions may fault if a reboot turns
491  * off virtualization while processes are running.  Usually after catching the
492  * fault we just panic; during reboot instead the instruction is ignored.
493  */
494 noinstr void kvm_spurious_fault(void)
495 {
496 	/* Fault while not rebooting.  We want the trace. */
497 	BUG_ON(!kvm_rebooting);
498 }
499 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
500 
501 #define EXCPT_BENIGN		0
502 #define EXCPT_CONTRIBUTORY	1
503 #define EXCPT_PF		2
504 
505 static int exception_class(int vector)
506 {
507 	switch (vector) {
508 	case PF_VECTOR:
509 		return EXCPT_PF;
510 	case DE_VECTOR:
511 	case TS_VECTOR:
512 	case NP_VECTOR:
513 	case SS_VECTOR:
514 	case GP_VECTOR:
515 		return EXCPT_CONTRIBUTORY;
516 	default:
517 		break;
518 	}
519 	return EXCPT_BENIGN;
520 }
521 
522 #define EXCPT_FAULT		0
523 #define EXCPT_TRAP		1
524 #define EXCPT_ABORT		2
525 #define EXCPT_INTERRUPT		3
526 
527 static int exception_type(int vector)
528 {
529 	unsigned int mask;
530 
531 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
532 		return EXCPT_INTERRUPT;
533 
534 	mask = 1 << vector;
535 
536 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
537 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
538 		return EXCPT_TRAP;
539 
540 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
541 		return EXCPT_ABORT;
542 
543 	/* Reserved exceptions will result in fault */
544 	return EXCPT_FAULT;
545 }
546 
547 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
548 {
549 	unsigned nr = vcpu->arch.exception.nr;
550 	bool has_payload = vcpu->arch.exception.has_payload;
551 	unsigned long payload = vcpu->arch.exception.payload;
552 
553 	if (!has_payload)
554 		return;
555 
556 	switch (nr) {
557 	case DB_VECTOR:
558 		/*
559 		 * "Certain debug exceptions may clear bit 0-3.  The
560 		 * remaining contents of the DR6 register are never
561 		 * cleared by the processor".
562 		 */
563 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
564 		/*
565 		 * In order to reflect the #DB exception payload in guest
566 		 * dr6, three components need to be considered: active low
567 		 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
568 		 * DR6_BS and DR6_BT)
569 		 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
570 		 * In the target guest dr6:
571 		 * FIXED_1 bits should always be set.
572 		 * Active low bits should be cleared if 1-setting in payload.
573 		 * Active high bits should be set if 1-setting in payload.
574 		 *
575 		 * Note, the payload is compatible with the pending debug
576 		 * exceptions/exit qualification under VMX, that active_low bits
577 		 * are active high in payload.
578 		 * So they need to be flipped for DR6.
579 		 */
580 		vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
581 		vcpu->arch.dr6 |= payload;
582 		vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
583 
584 		/*
585 		 * The #DB payload is defined as compatible with the 'pending
586 		 * debug exceptions' field under VMX, not DR6. While bit 12 is
587 		 * defined in the 'pending debug exceptions' field (enabled
588 		 * breakpoint), it is reserved and must be zero in DR6.
589 		 */
590 		vcpu->arch.dr6 &= ~BIT(12);
591 		break;
592 	case PF_VECTOR:
593 		vcpu->arch.cr2 = payload;
594 		break;
595 	}
596 
597 	vcpu->arch.exception.has_payload = false;
598 	vcpu->arch.exception.payload = 0;
599 }
600 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
601 
602 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
603 		unsigned nr, bool has_error, u32 error_code,
604 	        bool has_payload, unsigned long payload, bool reinject)
605 {
606 	u32 prev_nr;
607 	int class1, class2;
608 
609 	kvm_make_request(KVM_REQ_EVENT, vcpu);
610 
611 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
612 	queue:
613 		if (reinject) {
614 			/*
615 			 * On vmentry, vcpu->arch.exception.pending is only
616 			 * true if an event injection was blocked by
617 			 * nested_run_pending.  In that case, however,
618 			 * vcpu_enter_guest requests an immediate exit,
619 			 * and the guest shouldn't proceed far enough to
620 			 * need reinjection.
621 			 */
622 			WARN_ON_ONCE(vcpu->arch.exception.pending);
623 			vcpu->arch.exception.injected = true;
624 			if (WARN_ON_ONCE(has_payload)) {
625 				/*
626 				 * A reinjected event has already
627 				 * delivered its payload.
628 				 */
629 				has_payload = false;
630 				payload = 0;
631 			}
632 		} else {
633 			vcpu->arch.exception.pending = true;
634 			vcpu->arch.exception.injected = false;
635 		}
636 		vcpu->arch.exception.has_error_code = has_error;
637 		vcpu->arch.exception.nr = nr;
638 		vcpu->arch.exception.error_code = error_code;
639 		vcpu->arch.exception.has_payload = has_payload;
640 		vcpu->arch.exception.payload = payload;
641 		if (!is_guest_mode(vcpu))
642 			kvm_deliver_exception_payload(vcpu);
643 		return;
644 	}
645 
646 	/* to check exception */
647 	prev_nr = vcpu->arch.exception.nr;
648 	if (prev_nr == DF_VECTOR) {
649 		/* triple fault -> shutdown */
650 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
651 		return;
652 	}
653 	class1 = exception_class(prev_nr);
654 	class2 = exception_class(nr);
655 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
656 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
657 		/*
658 		 * Generate double fault per SDM Table 5-5.  Set
659 		 * exception.pending = true so that the double fault
660 		 * can trigger a nested vmexit.
661 		 */
662 		vcpu->arch.exception.pending = true;
663 		vcpu->arch.exception.injected = false;
664 		vcpu->arch.exception.has_error_code = true;
665 		vcpu->arch.exception.nr = DF_VECTOR;
666 		vcpu->arch.exception.error_code = 0;
667 		vcpu->arch.exception.has_payload = false;
668 		vcpu->arch.exception.payload = 0;
669 	} else
670 		/* replace previous exception with a new one in a hope
671 		   that instruction re-execution will regenerate lost
672 		   exception */
673 		goto queue;
674 }
675 
676 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
677 {
678 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
679 }
680 EXPORT_SYMBOL_GPL(kvm_queue_exception);
681 
682 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
683 {
684 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
685 }
686 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
687 
688 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
689 			   unsigned long payload)
690 {
691 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
692 }
693 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
694 
695 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
696 				    u32 error_code, unsigned long payload)
697 {
698 	kvm_multiple_exception(vcpu, nr, true, error_code,
699 			       true, payload, false);
700 }
701 
702 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
703 {
704 	if (err)
705 		kvm_inject_gp(vcpu, 0);
706 	else
707 		return kvm_skip_emulated_instruction(vcpu);
708 
709 	return 1;
710 }
711 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
712 
713 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
714 {
715 	++vcpu->stat.pf_guest;
716 	vcpu->arch.exception.nested_apf =
717 		is_guest_mode(vcpu) && fault->async_page_fault;
718 	if (vcpu->arch.exception.nested_apf) {
719 		vcpu->arch.apf.nested_apf_token = fault->address;
720 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
721 	} else {
722 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
723 					fault->address);
724 	}
725 }
726 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
727 
728 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
729 				    struct x86_exception *fault)
730 {
731 	struct kvm_mmu *fault_mmu;
732 	WARN_ON_ONCE(fault->vector != PF_VECTOR);
733 
734 	fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
735 					       vcpu->arch.walk_mmu;
736 
737 	/*
738 	 * Invalidate the TLB entry for the faulting address, if it exists,
739 	 * else the access will fault indefinitely (and to emulate hardware).
740 	 */
741 	if ((fault->error_code & PFERR_PRESENT_MASK) &&
742 	    !(fault->error_code & PFERR_RSVD_MASK))
743 		kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
744 				       fault_mmu->root_hpa);
745 
746 	fault_mmu->inject_page_fault(vcpu, fault);
747 	return fault->nested_page_fault;
748 }
749 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
750 
751 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
752 {
753 	atomic_inc(&vcpu->arch.nmi_queued);
754 	kvm_make_request(KVM_REQ_NMI, vcpu);
755 }
756 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
757 
758 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
759 {
760 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
761 }
762 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
763 
764 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
765 {
766 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
767 }
768 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
769 
770 /*
771  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
772  * a #GP and return false.
773  */
774 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
775 {
776 	if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
777 		return true;
778 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
779 	return false;
780 }
781 EXPORT_SYMBOL_GPL(kvm_require_cpl);
782 
783 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
784 {
785 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
786 		return true;
787 
788 	kvm_queue_exception(vcpu, UD_VECTOR);
789 	return false;
790 }
791 EXPORT_SYMBOL_GPL(kvm_require_dr);
792 
793 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
794 {
795 	return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
796 }
797 
798 /*
799  * Load the pae pdptrs.  Return 1 if they are all valid, 0 otherwise.
800  */
801 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
802 {
803 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
804 	gpa_t real_gpa;
805 	int i;
806 	int ret;
807 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
808 
809 	/*
810 	 * If the MMU is nested, CR3 holds an L2 GPA and needs to be translated
811 	 * to an L1 GPA.
812 	 */
813 	real_gpa = mmu->translate_gpa(vcpu, gfn_to_gpa(pdpt_gfn),
814 				      PFERR_USER_MASK | PFERR_WRITE_MASK, NULL);
815 	if (real_gpa == UNMAPPED_GVA)
816 		return 0;
817 
818 	/* Note the offset, PDPTRs are 32 byte aligned when using PAE paging. */
819 	ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(real_gpa), pdpte,
820 				       cr3 & GENMASK(11, 5), sizeof(pdpte));
821 	if (ret < 0)
822 		return 0;
823 
824 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
825 		if ((pdpte[i] & PT_PRESENT_MASK) &&
826 		    (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
827 			return 0;
828 		}
829 	}
830 
831 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
832 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
833 	vcpu->arch.pdptrs_from_userspace = false;
834 
835 	return 1;
836 }
837 EXPORT_SYMBOL_GPL(load_pdptrs);
838 
839 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
840 {
841 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
842 		kvm_clear_async_pf_completion_queue(vcpu);
843 		kvm_async_pf_hash_reset(vcpu);
844 	}
845 
846 	if ((cr0 ^ old_cr0) & KVM_MMU_CR0_ROLE_BITS)
847 		kvm_mmu_reset_context(vcpu);
848 
849 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
850 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
851 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
852 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
853 }
854 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
855 
856 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
857 {
858 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
859 	unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
860 
861 	cr0 |= X86_CR0_ET;
862 
863 #ifdef CONFIG_X86_64
864 	if (cr0 & 0xffffffff00000000UL)
865 		return 1;
866 #endif
867 
868 	cr0 &= ~CR0_RESERVED_BITS;
869 
870 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
871 		return 1;
872 
873 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
874 		return 1;
875 
876 #ifdef CONFIG_X86_64
877 	if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
878 	    (cr0 & X86_CR0_PG)) {
879 		int cs_db, cs_l;
880 
881 		if (!is_pae(vcpu))
882 			return 1;
883 		static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
884 		if (cs_l)
885 			return 1;
886 	}
887 #endif
888 	if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
889 	    is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
890 	    !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
891 		return 1;
892 
893 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
894 		return 1;
895 
896 	static_call(kvm_x86_set_cr0)(vcpu, cr0);
897 
898 	kvm_post_set_cr0(vcpu, old_cr0, cr0);
899 
900 	return 0;
901 }
902 EXPORT_SYMBOL_GPL(kvm_set_cr0);
903 
904 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
905 {
906 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
907 }
908 EXPORT_SYMBOL_GPL(kvm_lmsw);
909 
910 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
911 {
912 	if (vcpu->arch.guest_state_protected)
913 		return;
914 
915 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
916 
917 		if (vcpu->arch.xcr0 != host_xcr0)
918 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
919 
920 		if (vcpu->arch.xsaves_enabled &&
921 		    vcpu->arch.ia32_xss != host_xss)
922 			wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
923 	}
924 
925 	if (static_cpu_has(X86_FEATURE_PKU) &&
926 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
927 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
928 	    vcpu->arch.pkru != vcpu->arch.host_pkru)
929 		write_pkru(vcpu->arch.pkru);
930 }
931 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
932 
933 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
934 {
935 	if (vcpu->arch.guest_state_protected)
936 		return;
937 
938 	if (static_cpu_has(X86_FEATURE_PKU) &&
939 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
940 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
941 		vcpu->arch.pkru = rdpkru();
942 		if (vcpu->arch.pkru != vcpu->arch.host_pkru)
943 			write_pkru(vcpu->arch.host_pkru);
944 	}
945 
946 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
947 
948 		if (vcpu->arch.xcr0 != host_xcr0)
949 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
950 
951 		if (vcpu->arch.xsaves_enabled &&
952 		    vcpu->arch.ia32_xss != host_xss)
953 			wrmsrl(MSR_IA32_XSS, host_xss);
954 	}
955 
956 }
957 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
958 
959 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
960 {
961 	u64 xcr0 = xcr;
962 	u64 old_xcr0 = vcpu->arch.xcr0;
963 	u64 valid_bits;
964 
965 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
966 	if (index != XCR_XFEATURE_ENABLED_MASK)
967 		return 1;
968 	if (!(xcr0 & XFEATURE_MASK_FP))
969 		return 1;
970 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
971 		return 1;
972 
973 	/*
974 	 * Do not allow the guest to set bits that we do not support
975 	 * saving.  However, xcr0 bit 0 is always set, even if the
976 	 * emulated CPU does not support XSAVE (see kvm_vcpu_reset()).
977 	 */
978 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
979 	if (xcr0 & ~valid_bits)
980 		return 1;
981 
982 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
983 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
984 		return 1;
985 
986 	if (xcr0 & XFEATURE_MASK_AVX512) {
987 		if (!(xcr0 & XFEATURE_MASK_YMM))
988 			return 1;
989 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
990 			return 1;
991 	}
992 	vcpu->arch.xcr0 = xcr0;
993 
994 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
995 		kvm_update_cpuid_runtime(vcpu);
996 	return 0;
997 }
998 
999 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
1000 {
1001 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
1002 	    __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
1003 		kvm_inject_gp(vcpu, 0);
1004 		return 1;
1005 	}
1006 
1007 	return kvm_skip_emulated_instruction(vcpu);
1008 }
1009 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1010 
1011 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1012 {
1013 	if (cr4 & cr4_reserved_bits)
1014 		return false;
1015 
1016 	if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1017 		return false;
1018 
1019 	return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1020 }
1021 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
1022 
1023 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1024 {
1025 	/*
1026 	 * If any role bit is changed, the MMU needs to be reset.
1027 	 *
1028 	 * If CR4.PCIDE is changed 1 -> 0, the guest TLB must be flushed.
1029 	 * If CR4.PCIDE is changed 0 -> 1, there is no need to flush the TLB
1030 	 * according to the SDM; however, stale prev_roots could be reused
1031 	 * incorrectly in the future after a MOV to CR3 with NOFLUSH=1, so we
1032 	 * free them all.  KVM_REQ_MMU_RELOAD is fit for the both cases; it
1033 	 * is slow, but changing CR4.PCIDE is a rare case.
1034 	 *
1035 	 * If CR4.PGE is changed, the guest TLB must be flushed.
1036 	 *
1037 	 * Note: resetting MMU is a superset of KVM_REQ_MMU_RELOAD and
1038 	 * KVM_REQ_MMU_RELOAD is a superset of KVM_REQ_TLB_FLUSH_GUEST, hence
1039 	 * the usage of "else if".
1040 	 */
1041 	if ((cr4 ^ old_cr4) & KVM_MMU_CR4_ROLE_BITS)
1042 		kvm_mmu_reset_context(vcpu);
1043 	else if ((cr4 ^ old_cr4) & X86_CR4_PCIDE)
1044 		kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
1045 	else if ((cr4 ^ old_cr4) & X86_CR4_PGE)
1046 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1047 }
1048 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1049 
1050 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1051 {
1052 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
1053 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1054 				   X86_CR4_SMEP;
1055 
1056 	if (!kvm_is_valid_cr4(vcpu, cr4))
1057 		return 1;
1058 
1059 	if (is_long_mode(vcpu)) {
1060 		if (!(cr4 & X86_CR4_PAE))
1061 			return 1;
1062 		if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1063 			return 1;
1064 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1065 		   && ((cr4 ^ old_cr4) & pdptr_bits)
1066 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1067 				   kvm_read_cr3(vcpu)))
1068 		return 1;
1069 
1070 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1071 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1072 			return 1;
1073 
1074 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1075 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1076 			return 1;
1077 	}
1078 
1079 	static_call(kvm_x86_set_cr4)(vcpu, cr4);
1080 
1081 	kvm_post_set_cr4(vcpu, old_cr4, cr4);
1082 
1083 	return 0;
1084 }
1085 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1086 
1087 static void kvm_invalidate_pcid(struct kvm_vcpu *vcpu, unsigned long pcid)
1088 {
1089 	struct kvm_mmu *mmu = vcpu->arch.mmu;
1090 	unsigned long roots_to_free = 0;
1091 	int i;
1092 
1093 	/*
1094 	 * MOV CR3 and INVPCID are usually not intercepted when using TDP, but
1095 	 * this is reachable when running EPT=1 and unrestricted_guest=0,  and
1096 	 * also via the emulator.  KVM's TDP page tables are not in the scope of
1097 	 * the invalidation, but the guest's TLB entries need to be flushed as
1098 	 * the CPU may have cached entries in its TLB for the target PCID.
1099 	 */
1100 	if (unlikely(tdp_enabled)) {
1101 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1102 		return;
1103 	}
1104 
1105 	/*
1106 	 * If neither the current CR3 nor any of the prev_roots use the given
1107 	 * PCID, then nothing needs to be done here because a resync will
1108 	 * happen anyway before switching to any other CR3.
1109 	 */
1110 	if (kvm_get_active_pcid(vcpu) == pcid) {
1111 		kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
1112 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1113 	}
1114 
1115 	/*
1116 	 * If PCID is disabled, there is no need to free prev_roots even if the
1117 	 * PCIDs for them are also 0, because MOV to CR3 always flushes the TLB
1118 	 * with PCIDE=0.
1119 	 */
1120 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
1121 		return;
1122 
1123 	for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
1124 		if (kvm_get_pcid(vcpu, mmu->prev_roots[i].pgd) == pcid)
1125 			roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
1126 
1127 	kvm_mmu_free_roots(vcpu, mmu, roots_to_free);
1128 }
1129 
1130 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1131 {
1132 	bool skip_tlb_flush = false;
1133 	unsigned long pcid = 0;
1134 #ifdef CONFIG_X86_64
1135 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1136 
1137 	if (pcid_enabled) {
1138 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1139 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
1140 		pcid = cr3 & X86_CR3_PCID_MASK;
1141 	}
1142 #endif
1143 
1144 	/* PDPTRs are always reloaded for PAE paging. */
1145 	if (cr3 == kvm_read_cr3(vcpu) && !is_pae_paging(vcpu))
1146 		goto handle_tlb_flush;
1147 
1148 	/*
1149 	 * Do not condition the GPA check on long mode, this helper is used to
1150 	 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1151 	 * the current vCPU mode is accurate.
1152 	 */
1153 	if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1154 		return 1;
1155 
1156 	if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1157 		return 1;
1158 
1159 	if (cr3 != kvm_read_cr3(vcpu))
1160 		kvm_mmu_new_pgd(vcpu, cr3);
1161 
1162 	vcpu->arch.cr3 = cr3;
1163 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1164 
1165 handle_tlb_flush:
1166 	/*
1167 	 * A load of CR3 that flushes the TLB flushes only the current PCID,
1168 	 * even if PCID is disabled, in which case PCID=0 is flushed.  It's a
1169 	 * moot point in the end because _disabling_ PCID will flush all PCIDs,
1170 	 * and it's impossible to use a non-zero PCID when PCID is disabled,
1171 	 * i.e. only PCID=0 can be relevant.
1172 	 */
1173 	if (!skip_tlb_flush)
1174 		kvm_invalidate_pcid(vcpu, pcid);
1175 
1176 	return 0;
1177 }
1178 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1179 
1180 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1181 {
1182 	if (cr8 & CR8_RESERVED_BITS)
1183 		return 1;
1184 	if (lapic_in_kernel(vcpu))
1185 		kvm_lapic_set_tpr(vcpu, cr8);
1186 	else
1187 		vcpu->arch.cr8 = cr8;
1188 	return 0;
1189 }
1190 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1191 
1192 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1193 {
1194 	if (lapic_in_kernel(vcpu))
1195 		return kvm_lapic_get_cr8(vcpu);
1196 	else
1197 		return vcpu->arch.cr8;
1198 }
1199 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1200 
1201 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1202 {
1203 	int i;
1204 
1205 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1206 		for (i = 0; i < KVM_NR_DB_REGS; i++)
1207 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1208 	}
1209 }
1210 
1211 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1212 {
1213 	unsigned long dr7;
1214 
1215 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1216 		dr7 = vcpu->arch.guest_debug_dr7;
1217 	else
1218 		dr7 = vcpu->arch.dr7;
1219 	static_call(kvm_x86_set_dr7)(vcpu, dr7);
1220 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1221 	if (dr7 & DR7_BP_EN_MASK)
1222 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1223 }
1224 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1225 
1226 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1227 {
1228 	u64 fixed = DR6_FIXED_1;
1229 
1230 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1231 		fixed |= DR6_RTM;
1232 
1233 	if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1234 		fixed |= DR6_BUS_LOCK;
1235 	return fixed;
1236 }
1237 
1238 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1239 {
1240 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1241 
1242 	switch (dr) {
1243 	case 0 ... 3:
1244 		vcpu->arch.db[array_index_nospec(dr, size)] = val;
1245 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1246 			vcpu->arch.eff_db[dr] = val;
1247 		break;
1248 	case 4:
1249 	case 6:
1250 		if (!kvm_dr6_valid(val))
1251 			return 1; /* #GP */
1252 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1253 		break;
1254 	case 5:
1255 	default: /* 7 */
1256 		if (!kvm_dr7_valid(val))
1257 			return 1; /* #GP */
1258 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1259 		kvm_update_dr7(vcpu);
1260 		break;
1261 	}
1262 
1263 	return 0;
1264 }
1265 EXPORT_SYMBOL_GPL(kvm_set_dr);
1266 
1267 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1268 {
1269 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1270 
1271 	switch (dr) {
1272 	case 0 ... 3:
1273 		*val = vcpu->arch.db[array_index_nospec(dr, size)];
1274 		break;
1275 	case 4:
1276 	case 6:
1277 		*val = vcpu->arch.dr6;
1278 		break;
1279 	case 5:
1280 	default: /* 7 */
1281 		*val = vcpu->arch.dr7;
1282 		break;
1283 	}
1284 }
1285 EXPORT_SYMBOL_GPL(kvm_get_dr);
1286 
1287 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1288 {
1289 	u32 ecx = kvm_rcx_read(vcpu);
1290 	u64 data;
1291 
1292 	if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1293 		kvm_inject_gp(vcpu, 0);
1294 		return 1;
1295 	}
1296 
1297 	kvm_rax_write(vcpu, (u32)data);
1298 	kvm_rdx_write(vcpu, data >> 32);
1299 	return kvm_skip_emulated_instruction(vcpu);
1300 }
1301 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1302 
1303 /*
1304  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1305  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1306  *
1307  * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1308  * extract the supported MSRs from the related const lists.
1309  * msrs_to_save is selected from the msrs_to_save_all to reflect the
1310  * capabilities of the host cpu. This capabilities test skips MSRs that are
1311  * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1312  * may depend on host virtualization features rather than host cpu features.
1313  */
1314 
1315 static const u32 msrs_to_save_all[] = {
1316 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1317 	MSR_STAR,
1318 #ifdef CONFIG_X86_64
1319 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1320 #endif
1321 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1322 	MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1323 	MSR_IA32_SPEC_CTRL,
1324 	MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1325 	MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1326 	MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1327 	MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1328 	MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1329 	MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1330 	MSR_IA32_UMWAIT_CONTROL,
1331 
1332 	MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1333 	MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1334 	MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1335 	MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1336 	MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1337 	MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1338 	MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1339 	MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1340 	MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1341 	MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1342 	MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1343 	MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1344 	MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1345 	MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1346 	MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1347 	MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1348 	MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1349 	MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1350 	MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1351 	MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1352 	MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1353 	MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1354 
1355 	MSR_K7_EVNTSEL0, MSR_K7_EVNTSEL1, MSR_K7_EVNTSEL2, MSR_K7_EVNTSEL3,
1356 	MSR_K7_PERFCTR0, MSR_K7_PERFCTR1, MSR_K7_PERFCTR2, MSR_K7_PERFCTR3,
1357 	MSR_F15H_PERF_CTL0, MSR_F15H_PERF_CTL1, MSR_F15H_PERF_CTL2,
1358 	MSR_F15H_PERF_CTL3, MSR_F15H_PERF_CTL4, MSR_F15H_PERF_CTL5,
1359 	MSR_F15H_PERF_CTR0, MSR_F15H_PERF_CTR1, MSR_F15H_PERF_CTR2,
1360 	MSR_F15H_PERF_CTR3, MSR_F15H_PERF_CTR4, MSR_F15H_PERF_CTR5,
1361 };
1362 
1363 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1364 static unsigned num_msrs_to_save;
1365 
1366 static const u32 emulated_msrs_all[] = {
1367 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1368 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1369 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1370 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1371 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1372 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1373 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1374 	HV_X64_MSR_RESET,
1375 	HV_X64_MSR_VP_INDEX,
1376 	HV_X64_MSR_VP_RUNTIME,
1377 	HV_X64_MSR_SCONTROL,
1378 	HV_X64_MSR_STIMER0_CONFIG,
1379 	HV_X64_MSR_VP_ASSIST_PAGE,
1380 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1381 	HV_X64_MSR_TSC_EMULATION_STATUS,
1382 	HV_X64_MSR_SYNDBG_OPTIONS,
1383 	HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1384 	HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1385 	HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1386 
1387 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1388 	MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1389 
1390 	MSR_IA32_TSC_ADJUST,
1391 	MSR_IA32_TSC_DEADLINE,
1392 	MSR_IA32_ARCH_CAPABILITIES,
1393 	MSR_IA32_PERF_CAPABILITIES,
1394 	MSR_IA32_MISC_ENABLE,
1395 	MSR_IA32_MCG_STATUS,
1396 	MSR_IA32_MCG_CTL,
1397 	MSR_IA32_MCG_EXT_CTL,
1398 	MSR_IA32_SMBASE,
1399 	MSR_SMI_COUNT,
1400 	MSR_PLATFORM_INFO,
1401 	MSR_MISC_FEATURES_ENABLES,
1402 	MSR_AMD64_VIRT_SPEC_CTRL,
1403 	MSR_AMD64_TSC_RATIO,
1404 	MSR_IA32_POWER_CTL,
1405 	MSR_IA32_UCODE_REV,
1406 
1407 	/*
1408 	 * The following list leaves out MSRs whose values are determined
1409 	 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1410 	 * We always support the "true" VMX control MSRs, even if the host
1411 	 * processor does not, so I am putting these registers here rather
1412 	 * than in msrs_to_save_all.
1413 	 */
1414 	MSR_IA32_VMX_BASIC,
1415 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1416 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1417 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1418 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1419 	MSR_IA32_VMX_MISC,
1420 	MSR_IA32_VMX_CR0_FIXED0,
1421 	MSR_IA32_VMX_CR4_FIXED0,
1422 	MSR_IA32_VMX_VMCS_ENUM,
1423 	MSR_IA32_VMX_PROCBASED_CTLS2,
1424 	MSR_IA32_VMX_EPT_VPID_CAP,
1425 	MSR_IA32_VMX_VMFUNC,
1426 
1427 	MSR_K7_HWCR,
1428 	MSR_KVM_POLL_CONTROL,
1429 };
1430 
1431 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1432 static unsigned num_emulated_msrs;
1433 
1434 /*
1435  * List of msr numbers which are used to expose MSR-based features that
1436  * can be used by a hypervisor to validate requested CPU features.
1437  */
1438 static const u32 msr_based_features_all[] = {
1439 	MSR_IA32_VMX_BASIC,
1440 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1441 	MSR_IA32_VMX_PINBASED_CTLS,
1442 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1443 	MSR_IA32_VMX_PROCBASED_CTLS,
1444 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1445 	MSR_IA32_VMX_EXIT_CTLS,
1446 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1447 	MSR_IA32_VMX_ENTRY_CTLS,
1448 	MSR_IA32_VMX_MISC,
1449 	MSR_IA32_VMX_CR0_FIXED0,
1450 	MSR_IA32_VMX_CR0_FIXED1,
1451 	MSR_IA32_VMX_CR4_FIXED0,
1452 	MSR_IA32_VMX_CR4_FIXED1,
1453 	MSR_IA32_VMX_VMCS_ENUM,
1454 	MSR_IA32_VMX_PROCBASED_CTLS2,
1455 	MSR_IA32_VMX_EPT_VPID_CAP,
1456 	MSR_IA32_VMX_VMFUNC,
1457 
1458 	MSR_F10H_DECFG,
1459 	MSR_IA32_UCODE_REV,
1460 	MSR_IA32_ARCH_CAPABILITIES,
1461 	MSR_IA32_PERF_CAPABILITIES,
1462 };
1463 
1464 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1465 static unsigned int num_msr_based_features;
1466 
1467 static u64 kvm_get_arch_capabilities(void)
1468 {
1469 	u64 data = 0;
1470 
1471 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1472 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1473 
1474 	/*
1475 	 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1476 	 * the nested hypervisor runs with NX huge pages.  If it is not,
1477 	 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1478 	 * L1 guests, so it need not worry about its own (L2) guests.
1479 	 */
1480 	data |= ARCH_CAP_PSCHANGE_MC_NO;
1481 
1482 	/*
1483 	 * If we're doing cache flushes (either "always" or "cond")
1484 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1485 	 * If an outer hypervisor is doing the cache flush for us
1486 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1487 	 * capability to the guest too, and if EPT is disabled we're not
1488 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1489 	 * require a nested hypervisor to do a flush of its own.
1490 	 */
1491 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1492 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1493 
1494 	if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1495 		data |= ARCH_CAP_RDCL_NO;
1496 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1497 		data |= ARCH_CAP_SSB_NO;
1498 	if (!boot_cpu_has_bug(X86_BUG_MDS))
1499 		data |= ARCH_CAP_MDS_NO;
1500 
1501 	if (!boot_cpu_has(X86_FEATURE_RTM)) {
1502 		/*
1503 		 * If RTM=0 because the kernel has disabled TSX, the host might
1504 		 * have TAA_NO or TSX_CTRL.  Clear TAA_NO (the guest sees RTM=0
1505 		 * and therefore knows that there cannot be TAA) but keep
1506 		 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1507 		 * and we want to allow migrating those guests to tsx=off hosts.
1508 		 */
1509 		data &= ~ARCH_CAP_TAA_NO;
1510 	} else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1511 		data |= ARCH_CAP_TAA_NO;
1512 	} else {
1513 		/*
1514 		 * Nothing to do here; we emulate TSX_CTRL if present on the
1515 		 * host so the guest can choose between disabling TSX or
1516 		 * using VERW to clear CPU buffers.
1517 		 */
1518 	}
1519 
1520 	return data;
1521 }
1522 
1523 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1524 {
1525 	switch (msr->index) {
1526 	case MSR_IA32_ARCH_CAPABILITIES:
1527 		msr->data = kvm_get_arch_capabilities();
1528 		break;
1529 	case MSR_IA32_UCODE_REV:
1530 		rdmsrl_safe(msr->index, &msr->data);
1531 		break;
1532 	default:
1533 		return static_call(kvm_x86_get_msr_feature)(msr);
1534 	}
1535 	return 0;
1536 }
1537 
1538 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1539 {
1540 	struct kvm_msr_entry msr;
1541 	int r;
1542 
1543 	msr.index = index;
1544 	r = kvm_get_msr_feature(&msr);
1545 
1546 	if (r == KVM_MSR_RET_INVALID) {
1547 		/* Unconditionally clear the output for simplicity */
1548 		*data = 0;
1549 		if (kvm_msr_ignored_check(index, 0, false))
1550 			r = 0;
1551 	}
1552 
1553 	if (r)
1554 		return r;
1555 
1556 	*data = msr.data;
1557 
1558 	return 0;
1559 }
1560 
1561 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1562 {
1563 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1564 		return false;
1565 
1566 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1567 		return false;
1568 
1569 	if (efer & (EFER_LME | EFER_LMA) &&
1570 	    !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1571 		return false;
1572 
1573 	if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1574 		return false;
1575 
1576 	return true;
1577 
1578 }
1579 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1580 {
1581 	if (efer & efer_reserved_bits)
1582 		return false;
1583 
1584 	return __kvm_valid_efer(vcpu, efer);
1585 }
1586 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1587 
1588 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1589 {
1590 	u64 old_efer = vcpu->arch.efer;
1591 	u64 efer = msr_info->data;
1592 	int r;
1593 
1594 	if (efer & efer_reserved_bits)
1595 		return 1;
1596 
1597 	if (!msr_info->host_initiated) {
1598 		if (!__kvm_valid_efer(vcpu, efer))
1599 			return 1;
1600 
1601 		if (is_paging(vcpu) &&
1602 		    (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1603 			return 1;
1604 	}
1605 
1606 	efer &= ~EFER_LMA;
1607 	efer |= vcpu->arch.efer & EFER_LMA;
1608 
1609 	r = static_call(kvm_x86_set_efer)(vcpu, efer);
1610 	if (r) {
1611 		WARN_ON(r > 0);
1612 		return r;
1613 	}
1614 
1615 	/* Update reserved bits */
1616 	if ((efer ^ old_efer) & EFER_NX)
1617 		kvm_mmu_reset_context(vcpu);
1618 
1619 	return 0;
1620 }
1621 
1622 void kvm_enable_efer_bits(u64 mask)
1623 {
1624        efer_reserved_bits &= ~mask;
1625 }
1626 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1627 
1628 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1629 {
1630 	struct kvm_x86_msr_filter *msr_filter;
1631 	struct msr_bitmap_range *ranges;
1632 	struct kvm *kvm = vcpu->kvm;
1633 	bool allowed;
1634 	int idx;
1635 	u32 i;
1636 
1637 	/* x2APIC MSRs do not support filtering. */
1638 	if (index >= 0x800 && index <= 0x8ff)
1639 		return true;
1640 
1641 	idx = srcu_read_lock(&kvm->srcu);
1642 
1643 	msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1644 	if (!msr_filter) {
1645 		allowed = true;
1646 		goto out;
1647 	}
1648 
1649 	allowed = msr_filter->default_allow;
1650 	ranges = msr_filter->ranges;
1651 
1652 	for (i = 0; i < msr_filter->count; i++) {
1653 		u32 start = ranges[i].base;
1654 		u32 end = start + ranges[i].nmsrs;
1655 		u32 flags = ranges[i].flags;
1656 		unsigned long *bitmap = ranges[i].bitmap;
1657 
1658 		if ((index >= start) && (index < end) && (flags & type)) {
1659 			allowed = !!test_bit(index - start, bitmap);
1660 			break;
1661 		}
1662 	}
1663 
1664 out:
1665 	srcu_read_unlock(&kvm->srcu, idx);
1666 
1667 	return allowed;
1668 }
1669 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1670 
1671 /*
1672  * Write @data into the MSR specified by @index.  Select MSR specific fault
1673  * checks are bypassed if @host_initiated is %true.
1674  * Returns 0 on success, non-0 otherwise.
1675  * Assumes vcpu_load() was already called.
1676  */
1677 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1678 			 bool host_initiated)
1679 {
1680 	struct msr_data msr;
1681 
1682 	if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1683 		return KVM_MSR_RET_FILTERED;
1684 
1685 	switch (index) {
1686 	case MSR_FS_BASE:
1687 	case MSR_GS_BASE:
1688 	case MSR_KERNEL_GS_BASE:
1689 	case MSR_CSTAR:
1690 	case MSR_LSTAR:
1691 		if (is_noncanonical_address(data, vcpu))
1692 			return 1;
1693 		break;
1694 	case MSR_IA32_SYSENTER_EIP:
1695 	case MSR_IA32_SYSENTER_ESP:
1696 		/*
1697 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1698 		 * non-canonical address is written on Intel but not on
1699 		 * AMD (which ignores the top 32-bits, because it does
1700 		 * not implement 64-bit SYSENTER).
1701 		 *
1702 		 * 64-bit code should hence be able to write a non-canonical
1703 		 * value on AMD.  Making the address canonical ensures that
1704 		 * vmentry does not fail on Intel after writing a non-canonical
1705 		 * value, and that something deterministic happens if the guest
1706 		 * invokes 64-bit SYSENTER.
1707 		 */
1708 		data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1709 		break;
1710 	case MSR_TSC_AUX:
1711 		if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1712 			return 1;
1713 
1714 		if (!host_initiated &&
1715 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1716 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1717 			return 1;
1718 
1719 		/*
1720 		 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1721 		 * incomplete and conflicting architectural behavior.  Current
1722 		 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1723 		 * reserved and always read as zeros.  Enforce Intel's reserved
1724 		 * bits check if and only if the guest CPU is Intel, and clear
1725 		 * the bits in all other cases.  This ensures cross-vendor
1726 		 * migration will provide consistent behavior for the guest.
1727 		 */
1728 		if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1729 			return 1;
1730 
1731 		data = (u32)data;
1732 		break;
1733 	}
1734 
1735 	msr.data = data;
1736 	msr.index = index;
1737 	msr.host_initiated = host_initiated;
1738 
1739 	return static_call(kvm_x86_set_msr)(vcpu, &msr);
1740 }
1741 
1742 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1743 				     u32 index, u64 data, bool host_initiated)
1744 {
1745 	int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1746 
1747 	if (ret == KVM_MSR_RET_INVALID)
1748 		if (kvm_msr_ignored_check(index, data, true))
1749 			ret = 0;
1750 
1751 	return ret;
1752 }
1753 
1754 /*
1755  * Read the MSR specified by @index into @data.  Select MSR specific fault
1756  * checks are bypassed if @host_initiated is %true.
1757  * Returns 0 on success, non-0 otherwise.
1758  * Assumes vcpu_load() was already called.
1759  */
1760 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1761 		  bool host_initiated)
1762 {
1763 	struct msr_data msr;
1764 	int ret;
1765 
1766 	if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1767 		return KVM_MSR_RET_FILTERED;
1768 
1769 	switch (index) {
1770 	case MSR_TSC_AUX:
1771 		if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1772 			return 1;
1773 
1774 		if (!host_initiated &&
1775 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1776 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1777 			return 1;
1778 		break;
1779 	}
1780 
1781 	msr.index = index;
1782 	msr.host_initiated = host_initiated;
1783 
1784 	ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1785 	if (!ret)
1786 		*data = msr.data;
1787 	return ret;
1788 }
1789 
1790 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1791 				     u32 index, u64 *data, bool host_initiated)
1792 {
1793 	int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1794 
1795 	if (ret == KVM_MSR_RET_INVALID) {
1796 		/* Unconditionally clear *data for simplicity */
1797 		*data = 0;
1798 		if (kvm_msr_ignored_check(index, 0, false))
1799 			ret = 0;
1800 	}
1801 
1802 	return ret;
1803 }
1804 
1805 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1806 {
1807 	return kvm_get_msr_ignored_check(vcpu, index, data, false);
1808 }
1809 EXPORT_SYMBOL_GPL(kvm_get_msr);
1810 
1811 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1812 {
1813 	return kvm_set_msr_ignored_check(vcpu, index, data, false);
1814 }
1815 EXPORT_SYMBOL_GPL(kvm_set_msr);
1816 
1817 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1818 {
1819 	int err = vcpu->run->msr.error;
1820 	if (!err) {
1821 		kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1822 		kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1823 	}
1824 
1825 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, err);
1826 }
1827 
1828 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1829 {
1830 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1831 }
1832 
1833 static u64 kvm_msr_reason(int r)
1834 {
1835 	switch (r) {
1836 	case KVM_MSR_RET_INVALID:
1837 		return KVM_MSR_EXIT_REASON_UNKNOWN;
1838 	case KVM_MSR_RET_FILTERED:
1839 		return KVM_MSR_EXIT_REASON_FILTER;
1840 	default:
1841 		return KVM_MSR_EXIT_REASON_INVAL;
1842 	}
1843 }
1844 
1845 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1846 			      u32 exit_reason, u64 data,
1847 			      int (*completion)(struct kvm_vcpu *vcpu),
1848 			      int r)
1849 {
1850 	u64 msr_reason = kvm_msr_reason(r);
1851 
1852 	/* Check if the user wanted to know about this MSR fault */
1853 	if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1854 		return 0;
1855 
1856 	vcpu->run->exit_reason = exit_reason;
1857 	vcpu->run->msr.error = 0;
1858 	memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1859 	vcpu->run->msr.reason = msr_reason;
1860 	vcpu->run->msr.index = index;
1861 	vcpu->run->msr.data = data;
1862 	vcpu->arch.complete_userspace_io = completion;
1863 
1864 	return 1;
1865 }
1866 
1867 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1868 {
1869 	return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1870 				   complete_emulated_rdmsr, r);
1871 }
1872 
1873 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1874 {
1875 	return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1876 				   complete_emulated_wrmsr, r);
1877 }
1878 
1879 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1880 {
1881 	u32 ecx = kvm_rcx_read(vcpu);
1882 	u64 data;
1883 	int r;
1884 
1885 	r = kvm_get_msr(vcpu, ecx, &data);
1886 
1887 	/* MSR read failed? See if we should ask user space */
1888 	if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1889 		/* Bounce to user space */
1890 		return 0;
1891 	}
1892 
1893 	if (!r) {
1894 		trace_kvm_msr_read(ecx, data);
1895 
1896 		kvm_rax_write(vcpu, data & -1u);
1897 		kvm_rdx_write(vcpu, (data >> 32) & -1u);
1898 	} else {
1899 		trace_kvm_msr_read_ex(ecx);
1900 	}
1901 
1902 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1903 }
1904 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1905 
1906 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1907 {
1908 	u32 ecx = kvm_rcx_read(vcpu);
1909 	u64 data = kvm_read_edx_eax(vcpu);
1910 	int r;
1911 
1912 	r = kvm_set_msr(vcpu, ecx, data);
1913 
1914 	/* MSR write failed? See if we should ask user space */
1915 	if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1916 		/* Bounce to user space */
1917 		return 0;
1918 
1919 	/* Signal all other negative errors to userspace */
1920 	if (r < 0)
1921 		return r;
1922 
1923 	if (!r)
1924 		trace_kvm_msr_write(ecx, data);
1925 	else
1926 		trace_kvm_msr_write_ex(ecx, data);
1927 
1928 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1929 }
1930 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1931 
1932 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
1933 {
1934 	return kvm_skip_emulated_instruction(vcpu);
1935 }
1936 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
1937 
1938 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
1939 {
1940 	/* Treat an INVD instruction as a NOP and just skip it. */
1941 	return kvm_emulate_as_nop(vcpu);
1942 }
1943 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
1944 
1945 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
1946 {
1947 	pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
1948 	return kvm_emulate_as_nop(vcpu);
1949 }
1950 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
1951 
1952 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
1953 {
1954 	kvm_queue_exception(vcpu, UD_VECTOR);
1955 	return 1;
1956 }
1957 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
1958 
1959 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
1960 {
1961 	pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
1962 	return kvm_emulate_as_nop(vcpu);
1963 }
1964 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
1965 
1966 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1967 {
1968 	xfer_to_guest_mode_prepare();
1969 	return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1970 		xfer_to_guest_mode_work_pending();
1971 }
1972 
1973 /*
1974  * The fast path for frequent and performance sensitive wrmsr emulation,
1975  * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1976  * the latency of virtual IPI by avoiding the expensive bits of transitioning
1977  * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1978  * other cases which must be called after interrupts are enabled on the host.
1979  */
1980 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1981 {
1982 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1983 		return 1;
1984 
1985 	if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1986 		((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1987 		((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1988 		((u32)(data >> 32) != X2APIC_BROADCAST)) {
1989 
1990 		data &= ~(1 << 12);
1991 		kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1992 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1993 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1994 		trace_kvm_apic_write(APIC_ICR, (u32)data);
1995 		return 0;
1996 	}
1997 
1998 	return 1;
1999 }
2000 
2001 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
2002 {
2003 	if (!kvm_can_use_hv_timer(vcpu))
2004 		return 1;
2005 
2006 	kvm_set_lapic_tscdeadline_msr(vcpu, data);
2007 	return 0;
2008 }
2009 
2010 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
2011 {
2012 	u32 msr = kvm_rcx_read(vcpu);
2013 	u64 data;
2014 	fastpath_t ret = EXIT_FASTPATH_NONE;
2015 
2016 	switch (msr) {
2017 	case APIC_BASE_MSR + (APIC_ICR >> 4):
2018 		data = kvm_read_edx_eax(vcpu);
2019 		if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
2020 			kvm_skip_emulated_instruction(vcpu);
2021 			ret = EXIT_FASTPATH_EXIT_HANDLED;
2022 		}
2023 		break;
2024 	case MSR_IA32_TSC_DEADLINE:
2025 		data = kvm_read_edx_eax(vcpu);
2026 		if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
2027 			kvm_skip_emulated_instruction(vcpu);
2028 			ret = EXIT_FASTPATH_REENTER_GUEST;
2029 		}
2030 		break;
2031 	default:
2032 		break;
2033 	}
2034 
2035 	if (ret != EXIT_FASTPATH_NONE)
2036 		trace_kvm_msr_write(msr, data);
2037 
2038 	return ret;
2039 }
2040 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
2041 
2042 /*
2043  * Adapt set_msr() to msr_io()'s calling convention
2044  */
2045 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2046 {
2047 	return kvm_get_msr_ignored_check(vcpu, index, data, true);
2048 }
2049 
2050 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2051 {
2052 	return kvm_set_msr_ignored_check(vcpu, index, *data, true);
2053 }
2054 
2055 #ifdef CONFIG_X86_64
2056 struct pvclock_clock {
2057 	int vclock_mode;
2058 	u64 cycle_last;
2059 	u64 mask;
2060 	u32 mult;
2061 	u32 shift;
2062 	u64 base_cycles;
2063 	u64 offset;
2064 };
2065 
2066 struct pvclock_gtod_data {
2067 	seqcount_t	seq;
2068 
2069 	struct pvclock_clock clock; /* extract of a clocksource struct */
2070 	struct pvclock_clock raw_clock; /* extract of a clocksource struct */
2071 
2072 	ktime_t		offs_boot;
2073 	u64		wall_time_sec;
2074 };
2075 
2076 static struct pvclock_gtod_data pvclock_gtod_data;
2077 
2078 static void update_pvclock_gtod(struct timekeeper *tk)
2079 {
2080 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2081 
2082 	write_seqcount_begin(&vdata->seq);
2083 
2084 	/* copy pvclock gtod data */
2085 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->vdso_clock_mode;
2086 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
2087 	vdata->clock.mask		= tk->tkr_mono.mask;
2088 	vdata->clock.mult		= tk->tkr_mono.mult;
2089 	vdata->clock.shift		= tk->tkr_mono.shift;
2090 	vdata->clock.base_cycles	= tk->tkr_mono.xtime_nsec;
2091 	vdata->clock.offset		= tk->tkr_mono.base;
2092 
2093 	vdata->raw_clock.vclock_mode	= tk->tkr_raw.clock->vdso_clock_mode;
2094 	vdata->raw_clock.cycle_last	= tk->tkr_raw.cycle_last;
2095 	vdata->raw_clock.mask		= tk->tkr_raw.mask;
2096 	vdata->raw_clock.mult		= tk->tkr_raw.mult;
2097 	vdata->raw_clock.shift		= tk->tkr_raw.shift;
2098 	vdata->raw_clock.base_cycles	= tk->tkr_raw.xtime_nsec;
2099 	vdata->raw_clock.offset		= tk->tkr_raw.base;
2100 
2101 	vdata->wall_time_sec            = tk->xtime_sec;
2102 
2103 	vdata->offs_boot		= tk->offs_boot;
2104 
2105 	write_seqcount_end(&vdata->seq);
2106 }
2107 
2108 static s64 get_kvmclock_base_ns(void)
2109 {
2110 	/* Count up from boot time, but with the frequency of the raw clock.  */
2111 	return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2112 }
2113 #else
2114 static s64 get_kvmclock_base_ns(void)
2115 {
2116 	/* Master clock not used, so we can just use CLOCK_BOOTTIME.  */
2117 	return ktime_get_boottime_ns();
2118 }
2119 #endif
2120 
2121 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2122 {
2123 	int version;
2124 	int r;
2125 	struct pvclock_wall_clock wc;
2126 	u32 wc_sec_hi;
2127 	u64 wall_nsec;
2128 
2129 	if (!wall_clock)
2130 		return;
2131 
2132 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2133 	if (r)
2134 		return;
2135 
2136 	if (version & 1)
2137 		++version;  /* first time write, random junk */
2138 
2139 	++version;
2140 
2141 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2142 		return;
2143 
2144 	/*
2145 	 * The guest calculates current wall clock time by adding
2146 	 * system time (updated by kvm_guest_time_update below) to the
2147 	 * wall clock specified here.  We do the reverse here.
2148 	 */
2149 	wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2150 
2151 	wc.nsec = do_div(wall_nsec, 1000000000);
2152 	wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2153 	wc.version = version;
2154 
2155 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2156 
2157 	if (sec_hi_ofs) {
2158 		wc_sec_hi = wall_nsec >> 32;
2159 		kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2160 				&wc_sec_hi, sizeof(wc_sec_hi));
2161 	}
2162 
2163 	version++;
2164 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2165 }
2166 
2167 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2168 				  bool old_msr, bool host_initiated)
2169 {
2170 	struct kvm_arch *ka = &vcpu->kvm->arch;
2171 
2172 	if (vcpu->vcpu_id == 0 && !host_initiated) {
2173 		if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2174 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2175 
2176 		ka->boot_vcpu_runs_old_kvmclock = old_msr;
2177 	}
2178 
2179 	vcpu->arch.time = system_time;
2180 	kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2181 
2182 	/* we verify if the enable bit is set... */
2183 	vcpu->arch.pv_time_enabled = false;
2184 	if (!(system_time & 1))
2185 		return;
2186 
2187 	if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2188 				       &vcpu->arch.pv_time, system_time & ~1ULL,
2189 				       sizeof(struct pvclock_vcpu_time_info)))
2190 		vcpu->arch.pv_time_enabled = true;
2191 
2192 	return;
2193 }
2194 
2195 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2196 {
2197 	do_shl32_div32(dividend, divisor);
2198 	return dividend;
2199 }
2200 
2201 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2202 			       s8 *pshift, u32 *pmultiplier)
2203 {
2204 	uint64_t scaled64;
2205 	int32_t  shift = 0;
2206 	uint64_t tps64;
2207 	uint32_t tps32;
2208 
2209 	tps64 = base_hz;
2210 	scaled64 = scaled_hz;
2211 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2212 		tps64 >>= 1;
2213 		shift--;
2214 	}
2215 
2216 	tps32 = (uint32_t)tps64;
2217 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2218 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2219 			scaled64 >>= 1;
2220 		else
2221 			tps32 <<= 1;
2222 		shift++;
2223 	}
2224 
2225 	*pshift = shift;
2226 	*pmultiplier = div_frac(scaled64, tps32);
2227 }
2228 
2229 #ifdef CONFIG_X86_64
2230 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2231 #endif
2232 
2233 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2234 static unsigned long max_tsc_khz;
2235 
2236 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2237 {
2238 	u64 v = (u64)khz * (1000000 + ppm);
2239 	do_div(v, 1000000);
2240 	return v;
2241 }
2242 
2243 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier);
2244 
2245 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2246 {
2247 	u64 ratio;
2248 
2249 	/* Guest TSC same frequency as host TSC? */
2250 	if (!scale) {
2251 		kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2252 		return 0;
2253 	}
2254 
2255 	/* TSC scaling supported? */
2256 	if (!kvm_has_tsc_control) {
2257 		if (user_tsc_khz > tsc_khz) {
2258 			vcpu->arch.tsc_catchup = 1;
2259 			vcpu->arch.tsc_always_catchup = 1;
2260 			return 0;
2261 		} else {
2262 			pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2263 			return -1;
2264 		}
2265 	}
2266 
2267 	/* TSC scaling required  - calculate ratio */
2268 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2269 				user_tsc_khz, tsc_khz);
2270 
2271 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2272 		pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2273 			            user_tsc_khz);
2274 		return -1;
2275 	}
2276 
2277 	kvm_vcpu_write_tsc_multiplier(vcpu, ratio);
2278 	return 0;
2279 }
2280 
2281 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2282 {
2283 	u32 thresh_lo, thresh_hi;
2284 	int use_scaling = 0;
2285 
2286 	/* tsc_khz can be zero if TSC calibration fails */
2287 	if (user_tsc_khz == 0) {
2288 		/* set tsc_scaling_ratio to a safe value */
2289 		kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2290 		return -1;
2291 	}
2292 
2293 	/* Compute a scale to convert nanoseconds in TSC cycles */
2294 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2295 			   &vcpu->arch.virtual_tsc_shift,
2296 			   &vcpu->arch.virtual_tsc_mult);
2297 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2298 
2299 	/*
2300 	 * Compute the variation in TSC rate which is acceptable
2301 	 * within the range of tolerance and decide if the
2302 	 * rate being applied is within that bounds of the hardware
2303 	 * rate.  If so, no scaling or compensation need be done.
2304 	 */
2305 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2306 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2307 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2308 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2309 		use_scaling = 1;
2310 	}
2311 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2312 }
2313 
2314 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2315 {
2316 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2317 				      vcpu->arch.virtual_tsc_mult,
2318 				      vcpu->arch.virtual_tsc_shift);
2319 	tsc += vcpu->arch.this_tsc_write;
2320 	return tsc;
2321 }
2322 
2323 static inline int gtod_is_based_on_tsc(int mode)
2324 {
2325 	return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2326 }
2327 
2328 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2329 {
2330 #ifdef CONFIG_X86_64
2331 	bool vcpus_matched;
2332 	struct kvm_arch *ka = &vcpu->kvm->arch;
2333 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2334 
2335 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2336 			 atomic_read(&vcpu->kvm->online_vcpus));
2337 
2338 	/*
2339 	 * Once the masterclock is enabled, always perform request in
2340 	 * order to update it.
2341 	 *
2342 	 * In order to enable masterclock, the host clocksource must be TSC
2343 	 * and the vcpus need to have matched TSCs.  When that happens,
2344 	 * perform request to enable masterclock.
2345 	 */
2346 	if (ka->use_master_clock ||
2347 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2348 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2349 
2350 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2351 			    atomic_read(&vcpu->kvm->online_vcpus),
2352 		            ka->use_master_clock, gtod->clock.vclock_mode);
2353 #endif
2354 }
2355 
2356 /*
2357  * Multiply tsc by a fixed point number represented by ratio.
2358  *
2359  * The most significant 64-N bits (mult) of ratio represent the
2360  * integral part of the fixed point number; the remaining N bits
2361  * (frac) represent the fractional part, ie. ratio represents a fixed
2362  * point number (mult + frac * 2^(-N)).
2363  *
2364  * N equals to kvm_tsc_scaling_ratio_frac_bits.
2365  */
2366 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2367 {
2368 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2369 }
2370 
2371 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc, u64 ratio)
2372 {
2373 	u64 _tsc = tsc;
2374 
2375 	if (ratio != kvm_default_tsc_scaling_ratio)
2376 		_tsc = __scale_tsc(ratio, tsc);
2377 
2378 	return _tsc;
2379 }
2380 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2381 
2382 static u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2383 {
2384 	u64 tsc;
2385 
2386 	tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio);
2387 
2388 	return target_tsc - tsc;
2389 }
2390 
2391 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2392 {
2393 	return vcpu->arch.l1_tsc_offset +
2394 		kvm_scale_tsc(vcpu, host_tsc, vcpu->arch.l1_tsc_scaling_ratio);
2395 }
2396 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2397 
2398 u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier)
2399 {
2400 	u64 nested_offset;
2401 
2402 	if (l2_multiplier == kvm_default_tsc_scaling_ratio)
2403 		nested_offset = l1_offset;
2404 	else
2405 		nested_offset = mul_s64_u64_shr((s64) l1_offset, l2_multiplier,
2406 						kvm_tsc_scaling_ratio_frac_bits);
2407 
2408 	nested_offset += l2_offset;
2409 	return nested_offset;
2410 }
2411 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_offset);
2412 
2413 u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier)
2414 {
2415 	if (l2_multiplier != kvm_default_tsc_scaling_ratio)
2416 		return mul_u64_u64_shr(l1_multiplier, l2_multiplier,
2417 				       kvm_tsc_scaling_ratio_frac_bits);
2418 
2419 	return l1_multiplier;
2420 }
2421 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_multiplier);
2422 
2423 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset)
2424 {
2425 	trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2426 				   vcpu->arch.l1_tsc_offset,
2427 				   l1_offset);
2428 
2429 	vcpu->arch.l1_tsc_offset = l1_offset;
2430 
2431 	/*
2432 	 * If we are here because L1 chose not to trap WRMSR to TSC then
2433 	 * according to the spec this should set L1's TSC (as opposed to
2434 	 * setting L1's offset for L2).
2435 	 */
2436 	if (is_guest_mode(vcpu))
2437 		vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2438 			l1_offset,
2439 			static_call(kvm_x86_get_l2_tsc_offset)(vcpu),
2440 			static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2441 	else
2442 		vcpu->arch.tsc_offset = l1_offset;
2443 
2444 	static_call(kvm_x86_write_tsc_offset)(vcpu, vcpu->arch.tsc_offset);
2445 }
2446 
2447 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier)
2448 {
2449 	vcpu->arch.l1_tsc_scaling_ratio = l1_multiplier;
2450 
2451 	/* Userspace is changing the multiplier while L2 is active */
2452 	if (is_guest_mode(vcpu))
2453 		vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2454 			l1_multiplier,
2455 			static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2456 	else
2457 		vcpu->arch.tsc_scaling_ratio = l1_multiplier;
2458 
2459 	if (kvm_has_tsc_control)
2460 		static_call(kvm_x86_write_tsc_multiplier)(
2461 			vcpu, vcpu->arch.tsc_scaling_ratio);
2462 }
2463 
2464 static inline bool kvm_check_tsc_unstable(void)
2465 {
2466 #ifdef CONFIG_X86_64
2467 	/*
2468 	 * TSC is marked unstable when we're running on Hyper-V,
2469 	 * 'TSC page' clocksource is good.
2470 	 */
2471 	if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2472 		return false;
2473 #endif
2474 	return check_tsc_unstable();
2475 }
2476 
2477 /*
2478  * Infers attempts to synchronize the guest's tsc from host writes. Sets the
2479  * offset for the vcpu and tracks the TSC matching generation that the vcpu
2480  * participates in.
2481  */
2482 static void __kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 offset, u64 tsc,
2483 				  u64 ns, bool matched)
2484 {
2485 	struct kvm *kvm = vcpu->kvm;
2486 
2487 	lockdep_assert_held(&kvm->arch.tsc_write_lock);
2488 
2489 	/*
2490 	 * We also track th most recent recorded KHZ, write and time to
2491 	 * allow the matching interval to be extended at each write.
2492 	 */
2493 	kvm->arch.last_tsc_nsec = ns;
2494 	kvm->arch.last_tsc_write = tsc;
2495 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2496 	kvm->arch.last_tsc_offset = offset;
2497 
2498 	vcpu->arch.last_guest_tsc = tsc;
2499 
2500 	kvm_vcpu_write_tsc_offset(vcpu, offset);
2501 
2502 	if (!matched) {
2503 		/*
2504 		 * We split periods of matched TSC writes into generations.
2505 		 * For each generation, we track the original measured
2506 		 * nanosecond time, offset, and write, so if TSCs are in
2507 		 * sync, we can match exact offset, and if not, we can match
2508 		 * exact software computation in compute_guest_tsc()
2509 		 *
2510 		 * These values are tracked in kvm->arch.cur_xxx variables.
2511 		 */
2512 		kvm->arch.cur_tsc_generation++;
2513 		kvm->arch.cur_tsc_nsec = ns;
2514 		kvm->arch.cur_tsc_write = tsc;
2515 		kvm->arch.cur_tsc_offset = offset;
2516 		kvm->arch.nr_vcpus_matched_tsc = 0;
2517 	} else if (vcpu->arch.this_tsc_generation != kvm->arch.cur_tsc_generation) {
2518 		kvm->arch.nr_vcpus_matched_tsc++;
2519 	}
2520 
2521 	/* Keep track of which generation this VCPU has synchronized to */
2522 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2523 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2524 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2525 
2526 	kvm_track_tsc_matching(vcpu);
2527 }
2528 
2529 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2530 {
2531 	struct kvm *kvm = vcpu->kvm;
2532 	u64 offset, ns, elapsed;
2533 	unsigned long flags;
2534 	bool matched = false;
2535 	bool synchronizing = false;
2536 
2537 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2538 	offset = kvm_compute_l1_tsc_offset(vcpu, data);
2539 	ns = get_kvmclock_base_ns();
2540 	elapsed = ns - kvm->arch.last_tsc_nsec;
2541 
2542 	if (vcpu->arch.virtual_tsc_khz) {
2543 		if (data == 0) {
2544 			/*
2545 			 * detection of vcpu initialization -- need to sync
2546 			 * with other vCPUs. This particularly helps to keep
2547 			 * kvm_clock stable after CPU hotplug
2548 			 */
2549 			synchronizing = true;
2550 		} else {
2551 			u64 tsc_exp = kvm->arch.last_tsc_write +
2552 						nsec_to_cycles(vcpu, elapsed);
2553 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2554 			/*
2555 			 * Special case: TSC write with a small delta (1 second)
2556 			 * of virtual cycle time against real time is
2557 			 * interpreted as an attempt to synchronize the CPU.
2558 			 */
2559 			synchronizing = data < tsc_exp + tsc_hz &&
2560 					data + tsc_hz > tsc_exp;
2561 		}
2562 	}
2563 
2564 	/*
2565 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
2566 	 * TSC, we add elapsed time in this computation.  We could let the
2567 	 * compensation code attempt to catch up if we fall behind, but
2568 	 * it's better to try to match offsets from the beginning.
2569          */
2570 	if (synchronizing &&
2571 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2572 		if (!kvm_check_tsc_unstable()) {
2573 			offset = kvm->arch.cur_tsc_offset;
2574 		} else {
2575 			u64 delta = nsec_to_cycles(vcpu, elapsed);
2576 			data += delta;
2577 			offset = kvm_compute_l1_tsc_offset(vcpu, data);
2578 		}
2579 		matched = true;
2580 	}
2581 
2582 	__kvm_synchronize_tsc(vcpu, offset, data, ns, matched);
2583 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2584 }
2585 
2586 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2587 					   s64 adjustment)
2588 {
2589 	u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2590 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2591 }
2592 
2593 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2594 {
2595 	if (vcpu->arch.l1_tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2596 		WARN_ON(adjustment < 0);
2597 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment,
2598 				   vcpu->arch.l1_tsc_scaling_ratio);
2599 	adjust_tsc_offset_guest(vcpu, adjustment);
2600 }
2601 
2602 #ifdef CONFIG_X86_64
2603 
2604 static u64 read_tsc(void)
2605 {
2606 	u64 ret = (u64)rdtsc_ordered();
2607 	u64 last = pvclock_gtod_data.clock.cycle_last;
2608 
2609 	if (likely(ret >= last))
2610 		return ret;
2611 
2612 	/*
2613 	 * GCC likes to generate cmov here, but this branch is extremely
2614 	 * predictable (it's just a function of time and the likely is
2615 	 * very likely) and there's a data dependence, so force GCC
2616 	 * to generate a branch instead.  I don't barrier() because
2617 	 * we don't actually need a barrier, and if this function
2618 	 * ever gets inlined it will generate worse code.
2619 	 */
2620 	asm volatile ("");
2621 	return last;
2622 }
2623 
2624 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2625 			  int *mode)
2626 {
2627 	long v;
2628 	u64 tsc_pg_val;
2629 
2630 	switch (clock->vclock_mode) {
2631 	case VDSO_CLOCKMODE_HVCLOCK:
2632 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2633 						  tsc_timestamp);
2634 		if (tsc_pg_val != U64_MAX) {
2635 			/* TSC page valid */
2636 			*mode = VDSO_CLOCKMODE_HVCLOCK;
2637 			v = (tsc_pg_val - clock->cycle_last) &
2638 				clock->mask;
2639 		} else {
2640 			/* TSC page invalid */
2641 			*mode = VDSO_CLOCKMODE_NONE;
2642 		}
2643 		break;
2644 	case VDSO_CLOCKMODE_TSC:
2645 		*mode = VDSO_CLOCKMODE_TSC;
2646 		*tsc_timestamp = read_tsc();
2647 		v = (*tsc_timestamp - clock->cycle_last) &
2648 			clock->mask;
2649 		break;
2650 	default:
2651 		*mode = VDSO_CLOCKMODE_NONE;
2652 	}
2653 
2654 	if (*mode == VDSO_CLOCKMODE_NONE)
2655 		*tsc_timestamp = v = 0;
2656 
2657 	return v * clock->mult;
2658 }
2659 
2660 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2661 {
2662 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2663 	unsigned long seq;
2664 	int mode;
2665 	u64 ns;
2666 
2667 	do {
2668 		seq = read_seqcount_begin(&gtod->seq);
2669 		ns = gtod->raw_clock.base_cycles;
2670 		ns += vgettsc(&gtod->raw_clock, tsc_timestamp, &mode);
2671 		ns >>= gtod->raw_clock.shift;
2672 		ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2673 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2674 	*t = ns;
2675 
2676 	return mode;
2677 }
2678 
2679 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2680 {
2681 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2682 	unsigned long seq;
2683 	int mode;
2684 	u64 ns;
2685 
2686 	do {
2687 		seq = read_seqcount_begin(&gtod->seq);
2688 		ts->tv_sec = gtod->wall_time_sec;
2689 		ns = gtod->clock.base_cycles;
2690 		ns += vgettsc(&gtod->clock, tsc_timestamp, &mode);
2691 		ns >>= gtod->clock.shift;
2692 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2693 
2694 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2695 	ts->tv_nsec = ns;
2696 
2697 	return mode;
2698 }
2699 
2700 /* returns true if host is using TSC based clocksource */
2701 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2702 {
2703 	/* checked again under seqlock below */
2704 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2705 		return false;
2706 
2707 	return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2708 						      tsc_timestamp));
2709 }
2710 
2711 /* returns true if host is using TSC based clocksource */
2712 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2713 					   u64 *tsc_timestamp)
2714 {
2715 	/* checked again under seqlock below */
2716 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2717 		return false;
2718 
2719 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2720 }
2721 #endif
2722 
2723 /*
2724  *
2725  * Assuming a stable TSC across physical CPUS, and a stable TSC
2726  * across virtual CPUs, the following condition is possible.
2727  * Each numbered line represents an event visible to both
2728  * CPUs at the next numbered event.
2729  *
2730  * "timespecX" represents host monotonic time. "tscX" represents
2731  * RDTSC value.
2732  *
2733  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
2734  *
2735  * 1.  read timespec0,tsc0
2736  * 2.					| timespec1 = timespec0 + N
2737  * 					| tsc1 = tsc0 + M
2738  * 3. transition to guest		| transition to guest
2739  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2740  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
2741  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2742  *
2743  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2744  *
2745  * 	- ret0 < ret1
2746  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2747  *		...
2748  *	- 0 < N - M => M < N
2749  *
2750  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2751  * always the case (the difference between two distinct xtime instances
2752  * might be smaller then the difference between corresponding TSC reads,
2753  * when updating guest vcpus pvclock areas).
2754  *
2755  * To avoid that problem, do not allow visibility of distinct
2756  * system_timestamp/tsc_timestamp values simultaneously: use a master
2757  * copy of host monotonic time values. Update that master copy
2758  * in lockstep.
2759  *
2760  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2761  *
2762  */
2763 
2764 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2765 {
2766 #ifdef CONFIG_X86_64
2767 	struct kvm_arch *ka = &kvm->arch;
2768 	int vclock_mode;
2769 	bool host_tsc_clocksource, vcpus_matched;
2770 
2771 	lockdep_assert_held(&kvm->arch.tsc_write_lock);
2772 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2773 			atomic_read(&kvm->online_vcpus));
2774 
2775 	/*
2776 	 * If the host uses TSC clock, then passthrough TSC as stable
2777 	 * to the guest.
2778 	 */
2779 	host_tsc_clocksource = kvm_get_time_and_clockread(
2780 					&ka->master_kernel_ns,
2781 					&ka->master_cycle_now);
2782 
2783 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2784 				&& !ka->backwards_tsc_observed
2785 				&& !ka->boot_vcpu_runs_old_kvmclock;
2786 
2787 	if (ka->use_master_clock)
2788 		atomic_set(&kvm_guest_has_master_clock, 1);
2789 
2790 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2791 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2792 					vcpus_matched);
2793 #endif
2794 }
2795 
2796 static void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2797 {
2798 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2799 }
2800 
2801 static void __kvm_start_pvclock_update(struct kvm *kvm)
2802 {
2803 	raw_spin_lock_irq(&kvm->arch.tsc_write_lock);
2804 	write_seqcount_begin(&kvm->arch.pvclock_sc);
2805 }
2806 
2807 static void kvm_start_pvclock_update(struct kvm *kvm)
2808 {
2809 	kvm_make_mclock_inprogress_request(kvm);
2810 
2811 	/* no guest entries from this point */
2812 	__kvm_start_pvclock_update(kvm);
2813 }
2814 
2815 static void kvm_end_pvclock_update(struct kvm *kvm)
2816 {
2817 	struct kvm_arch *ka = &kvm->arch;
2818 	struct kvm_vcpu *vcpu;
2819 	int i;
2820 
2821 	write_seqcount_end(&ka->pvclock_sc);
2822 	raw_spin_unlock_irq(&ka->tsc_write_lock);
2823 	kvm_for_each_vcpu(i, vcpu, kvm)
2824 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2825 
2826 	/* guest entries allowed */
2827 	kvm_for_each_vcpu(i, vcpu, kvm)
2828 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2829 }
2830 
2831 static void kvm_update_masterclock(struct kvm *kvm)
2832 {
2833 	kvm_hv_invalidate_tsc_page(kvm);
2834 	kvm_start_pvclock_update(kvm);
2835 	pvclock_update_vm_gtod_copy(kvm);
2836 	kvm_end_pvclock_update(kvm);
2837 }
2838 
2839 /* Called within read_seqcount_begin/retry for kvm->pvclock_sc.  */
2840 static void __get_kvmclock(struct kvm *kvm, struct kvm_clock_data *data)
2841 {
2842 	struct kvm_arch *ka = &kvm->arch;
2843 	struct pvclock_vcpu_time_info hv_clock;
2844 
2845 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2846 	get_cpu();
2847 
2848 	data->flags = 0;
2849 	if (ka->use_master_clock && __this_cpu_read(cpu_tsc_khz)) {
2850 #ifdef CONFIG_X86_64
2851 		struct timespec64 ts;
2852 
2853 		if (kvm_get_walltime_and_clockread(&ts, &data->host_tsc)) {
2854 			data->realtime = ts.tv_nsec + NSEC_PER_SEC * ts.tv_sec;
2855 			data->flags |= KVM_CLOCK_REALTIME | KVM_CLOCK_HOST_TSC;
2856 		} else
2857 #endif
2858 		data->host_tsc = rdtsc();
2859 
2860 		data->flags |= KVM_CLOCK_TSC_STABLE;
2861 		hv_clock.tsc_timestamp = ka->master_cycle_now;
2862 		hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2863 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2864 				   &hv_clock.tsc_shift,
2865 				   &hv_clock.tsc_to_system_mul);
2866 		data->clock = __pvclock_read_cycles(&hv_clock, data->host_tsc);
2867 	} else {
2868 		data->clock = get_kvmclock_base_ns() + ka->kvmclock_offset;
2869 	}
2870 
2871 	put_cpu();
2872 }
2873 
2874 static void get_kvmclock(struct kvm *kvm, struct kvm_clock_data *data)
2875 {
2876 	struct kvm_arch *ka = &kvm->arch;
2877 	unsigned seq;
2878 
2879 	do {
2880 		seq = read_seqcount_begin(&ka->pvclock_sc);
2881 		__get_kvmclock(kvm, data);
2882 	} while (read_seqcount_retry(&ka->pvclock_sc, seq));
2883 }
2884 
2885 u64 get_kvmclock_ns(struct kvm *kvm)
2886 {
2887 	struct kvm_clock_data data;
2888 
2889 	get_kvmclock(kvm, &data);
2890 	return data.clock;
2891 }
2892 
2893 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2894 				   struct gfn_to_hva_cache *cache,
2895 				   unsigned int offset)
2896 {
2897 	struct kvm_vcpu_arch *vcpu = &v->arch;
2898 	struct pvclock_vcpu_time_info guest_hv_clock;
2899 
2900 	if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2901 		&guest_hv_clock, offset, sizeof(guest_hv_clock))))
2902 		return;
2903 
2904 	/* This VCPU is paused, but it's legal for a guest to read another
2905 	 * VCPU's kvmclock, so we really have to follow the specification where
2906 	 * it says that version is odd if data is being modified, and even after
2907 	 * it is consistent.
2908 	 *
2909 	 * Version field updates must be kept separate.  This is because
2910 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2911 	 * writes within a string instruction are weakly ordered.  So there
2912 	 * are three writes overall.
2913 	 *
2914 	 * As a small optimization, only write the version field in the first
2915 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2916 	 * version field is the first in the struct.
2917 	 */
2918 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2919 
2920 	if (guest_hv_clock.version & 1)
2921 		++guest_hv_clock.version;  /* first time write, random junk */
2922 
2923 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2924 	kvm_write_guest_offset_cached(v->kvm, cache,
2925 				      &vcpu->hv_clock, offset,
2926 				      sizeof(vcpu->hv_clock.version));
2927 
2928 	smp_wmb();
2929 
2930 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2931 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2932 
2933 	if (vcpu->pvclock_set_guest_stopped_request) {
2934 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2935 		vcpu->pvclock_set_guest_stopped_request = false;
2936 	}
2937 
2938 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2939 
2940 	kvm_write_guest_offset_cached(v->kvm, cache,
2941 				      &vcpu->hv_clock, offset,
2942 				      sizeof(vcpu->hv_clock));
2943 
2944 	smp_wmb();
2945 
2946 	vcpu->hv_clock.version++;
2947 	kvm_write_guest_offset_cached(v->kvm, cache,
2948 				     &vcpu->hv_clock, offset,
2949 				     sizeof(vcpu->hv_clock.version));
2950 }
2951 
2952 static int kvm_guest_time_update(struct kvm_vcpu *v)
2953 {
2954 	unsigned long flags, tgt_tsc_khz;
2955 	unsigned seq;
2956 	struct kvm_vcpu_arch *vcpu = &v->arch;
2957 	struct kvm_arch *ka = &v->kvm->arch;
2958 	s64 kernel_ns;
2959 	u64 tsc_timestamp, host_tsc;
2960 	u8 pvclock_flags;
2961 	bool use_master_clock;
2962 
2963 	kernel_ns = 0;
2964 	host_tsc = 0;
2965 
2966 	/*
2967 	 * If the host uses TSC clock, then passthrough TSC as stable
2968 	 * to the guest.
2969 	 */
2970 	do {
2971 		seq = read_seqcount_begin(&ka->pvclock_sc);
2972 		use_master_clock = ka->use_master_clock;
2973 		if (use_master_clock) {
2974 			host_tsc = ka->master_cycle_now;
2975 			kernel_ns = ka->master_kernel_ns;
2976 		}
2977 	} while (read_seqcount_retry(&ka->pvclock_sc, seq));
2978 
2979 	/* Keep irq disabled to prevent changes to the clock */
2980 	local_irq_save(flags);
2981 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2982 	if (unlikely(tgt_tsc_khz == 0)) {
2983 		local_irq_restore(flags);
2984 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2985 		return 1;
2986 	}
2987 	if (!use_master_clock) {
2988 		host_tsc = rdtsc();
2989 		kernel_ns = get_kvmclock_base_ns();
2990 	}
2991 
2992 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2993 
2994 	/*
2995 	 * We may have to catch up the TSC to match elapsed wall clock
2996 	 * time for two reasons, even if kvmclock is used.
2997 	 *   1) CPU could have been running below the maximum TSC rate
2998 	 *   2) Broken TSC compensation resets the base at each VCPU
2999 	 *      entry to avoid unknown leaps of TSC even when running
3000 	 *      again on the same CPU.  This may cause apparent elapsed
3001 	 *      time to disappear, and the guest to stand still or run
3002 	 *	very slowly.
3003 	 */
3004 	if (vcpu->tsc_catchup) {
3005 		u64 tsc = compute_guest_tsc(v, kernel_ns);
3006 		if (tsc > tsc_timestamp) {
3007 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
3008 			tsc_timestamp = tsc;
3009 		}
3010 	}
3011 
3012 	local_irq_restore(flags);
3013 
3014 	/* With all the info we got, fill in the values */
3015 
3016 	if (kvm_has_tsc_control)
3017 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz,
3018 					    v->arch.l1_tsc_scaling_ratio);
3019 
3020 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
3021 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
3022 				   &vcpu->hv_clock.tsc_shift,
3023 				   &vcpu->hv_clock.tsc_to_system_mul);
3024 		vcpu->hw_tsc_khz = tgt_tsc_khz;
3025 	}
3026 
3027 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
3028 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
3029 	vcpu->last_guest_tsc = tsc_timestamp;
3030 
3031 	/* If the host uses TSC clocksource, then it is stable */
3032 	pvclock_flags = 0;
3033 	if (use_master_clock)
3034 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
3035 
3036 	vcpu->hv_clock.flags = pvclock_flags;
3037 
3038 	if (vcpu->pv_time_enabled)
3039 		kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
3040 	if (vcpu->xen.vcpu_info_set)
3041 		kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
3042 				       offsetof(struct compat_vcpu_info, time));
3043 	if (vcpu->xen.vcpu_time_info_set)
3044 		kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
3045 	if (!v->vcpu_idx)
3046 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
3047 	return 0;
3048 }
3049 
3050 /*
3051  * kvmclock updates which are isolated to a given vcpu, such as
3052  * vcpu->cpu migration, should not allow system_timestamp from
3053  * the rest of the vcpus to remain static. Otherwise ntp frequency
3054  * correction applies to one vcpu's system_timestamp but not
3055  * the others.
3056  *
3057  * So in those cases, request a kvmclock update for all vcpus.
3058  * We need to rate-limit these requests though, as they can
3059  * considerably slow guests that have a large number of vcpus.
3060  * The time for a remote vcpu to update its kvmclock is bound
3061  * by the delay we use to rate-limit the updates.
3062  */
3063 
3064 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
3065 
3066 static void kvmclock_update_fn(struct work_struct *work)
3067 {
3068 	int i;
3069 	struct delayed_work *dwork = to_delayed_work(work);
3070 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3071 					   kvmclock_update_work);
3072 	struct kvm *kvm = container_of(ka, struct kvm, arch);
3073 	struct kvm_vcpu *vcpu;
3074 
3075 	kvm_for_each_vcpu(i, vcpu, kvm) {
3076 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3077 		kvm_vcpu_kick(vcpu);
3078 	}
3079 }
3080 
3081 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
3082 {
3083 	struct kvm *kvm = v->kvm;
3084 
3085 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3086 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
3087 					KVMCLOCK_UPDATE_DELAY);
3088 }
3089 
3090 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
3091 
3092 static void kvmclock_sync_fn(struct work_struct *work)
3093 {
3094 	struct delayed_work *dwork = to_delayed_work(work);
3095 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3096 					   kvmclock_sync_work);
3097 	struct kvm *kvm = container_of(ka, struct kvm, arch);
3098 
3099 	if (!kvmclock_periodic_sync)
3100 		return;
3101 
3102 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
3103 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
3104 					KVMCLOCK_SYNC_PERIOD);
3105 }
3106 
3107 /*
3108  * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
3109  */
3110 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
3111 {
3112 	/* McStatusWrEn enabled? */
3113 	if (guest_cpuid_is_amd_or_hygon(vcpu))
3114 		return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
3115 
3116 	return false;
3117 }
3118 
3119 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3120 {
3121 	u64 mcg_cap = vcpu->arch.mcg_cap;
3122 	unsigned bank_num = mcg_cap & 0xff;
3123 	u32 msr = msr_info->index;
3124 	u64 data = msr_info->data;
3125 
3126 	switch (msr) {
3127 	case MSR_IA32_MCG_STATUS:
3128 		vcpu->arch.mcg_status = data;
3129 		break;
3130 	case MSR_IA32_MCG_CTL:
3131 		if (!(mcg_cap & MCG_CTL_P) &&
3132 		    (data || !msr_info->host_initiated))
3133 			return 1;
3134 		if (data != 0 && data != ~(u64)0)
3135 			return 1;
3136 		vcpu->arch.mcg_ctl = data;
3137 		break;
3138 	default:
3139 		if (msr >= MSR_IA32_MC0_CTL &&
3140 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
3141 			u32 offset = array_index_nospec(
3142 				msr - MSR_IA32_MC0_CTL,
3143 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3144 
3145 			/* only 0 or all 1s can be written to IA32_MCi_CTL
3146 			 * some Linux kernels though clear bit 10 in bank 4 to
3147 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
3148 			 * this to avoid an uncatched #GP in the guest
3149 			 */
3150 			if ((offset & 0x3) == 0 &&
3151 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
3152 				return -1;
3153 
3154 			/* MCi_STATUS */
3155 			if (!msr_info->host_initiated &&
3156 			    (offset & 0x3) == 1 && data != 0) {
3157 				if (!can_set_mci_status(vcpu))
3158 					return -1;
3159 			}
3160 
3161 			vcpu->arch.mce_banks[offset] = data;
3162 			break;
3163 		}
3164 		return 1;
3165 	}
3166 	return 0;
3167 }
3168 
3169 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
3170 {
3171 	u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3172 
3173 	return (vcpu->arch.apf.msr_en_val & mask) == mask;
3174 }
3175 
3176 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3177 {
3178 	gpa_t gpa = data & ~0x3f;
3179 
3180 	/* Bits 4:5 are reserved, Should be zero */
3181 	if (data & 0x30)
3182 		return 1;
3183 
3184 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3185 	    (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3186 		return 1;
3187 
3188 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3189 	    (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3190 		return 1;
3191 
3192 	if (!lapic_in_kernel(vcpu))
3193 		return data ? 1 : 0;
3194 
3195 	vcpu->arch.apf.msr_en_val = data;
3196 
3197 	if (!kvm_pv_async_pf_enabled(vcpu)) {
3198 		kvm_clear_async_pf_completion_queue(vcpu);
3199 		kvm_async_pf_hash_reset(vcpu);
3200 		return 0;
3201 	}
3202 
3203 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3204 					sizeof(u64)))
3205 		return 1;
3206 
3207 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3208 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3209 
3210 	kvm_async_pf_wakeup_all(vcpu);
3211 
3212 	return 0;
3213 }
3214 
3215 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3216 {
3217 	/* Bits 8-63 are reserved */
3218 	if (data >> 8)
3219 		return 1;
3220 
3221 	if (!lapic_in_kernel(vcpu))
3222 		return 1;
3223 
3224 	vcpu->arch.apf.msr_int_val = data;
3225 
3226 	vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3227 
3228 	return 0;
3229 }
3230 
3231 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3232 {
3233 	vcpu->arch.pv_time_enabled = false;
3234 	vcpu->arch.time = 0;
3235 }
3236 
3237 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3238 {
3239 	++vcpu->stat.tlb_flush;
3240 	static_call(kvm_x86_tlb_flush_all)(vcpu);
3241 }
3242 
3243 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3244 {
3245 	++vcpu->stat.tlb_flush;
3246 
3247 	if (!tdp_enabled) {
3248 		/*
3249 		 * A TLB flush on behalf of the guest is equivalent to
3250 		 * INVPCID(all), toggling CR4.PGE, etc., which requires
3251 		 * a forced sync of the shadow page tables.  Ensure all the
3252 		 * roots are synced and the guest TLB in hardware is clean.
3253 		 */
3254 		kvm_mmu_sync_roots(vcpu);
3255 		kvm_mmu_sync_prev_roots(vcpu);
3256 	}
3257 
3258 	static_call(kvm_x86_tlb_flush_guest)(vcpu);
3259 }
3260 
3261 static void record_steal_time(struct kvm_vcpu *vcpu)
3262 {
3263 	struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
3264 	struct kvm_steal_time __user *st;
3265 	struct kvm_memslots *slots;
3266 	u64 steal;
3267 	u32 version;
3268 
3269 	if (kvm_xen_msr_enabled(vcpu->kvm)) {
3270 		kvm_xen_runstate_set_running(vcpu);
3271 		return;
3272 	}
3273 
3274 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3275 		return;
3276 
3277 	if (WARN_ON_ONCE(current->mm != vcpu->kvm->mm))
3278 		return;
3279 
3280 	slots = kvm_memslots(vcpu->kvm);
3281 
3282 	if (unlikely(slots->generation != ghc->generation ||
3283 		     kvm_is_error_hva(ghc->hva) || !ghc->memslot)) {
3284 		gfn_t gfn = vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS;
3285 
3286 		/* We rely on the fact that it fits in a single page. */
3287 		BUILD_BUG_ON((sizeof(*st) - 1) & KVM_STEAL_VALID_BITS);
3288 
3289 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, gfn, sizeof(*st)) ||
3290 		    kvm_is_error_hva(ghc->hva) || !ghc->memslot)
3291 			return;
3292 	}
3293 
3294 	st = (struct kvm_steal_time __user *)ghc->hva;
3295 	/*
3296 	 * Doing a TLB flush here, on the guest's behalf, can avoid
3297 	 * expensive IPIs.
3298 	 */
3299 	if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3300 		u8 st_preempted = 0;
3301 		int err = -EFAULT;
3302 
3303 		if (!user_access_begin(st, sizeof(*st)))
3304 			return;
3305 
3306 		asm volatile("1: xchgb %0, %2\n"
3307 			     "xor %1, %1\n"
3308 			     "2:\n"
3309 			     _ASM_EXTABLE_UA(1b, 2b)
3310 			     : "+r" (st_preempted),
3311 			       "+&r" (err)
3312 			     : "m" (st->preempted));
3313 		if (err)
3314 			goto out;
3315 
3316 		user_access_end();
3317 
3318 		vcpu->arch.st.preempted = 0;
3319 
3320 		trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3321 				       st_preempted & KVM_VCPU_FLUSH_TLB);
3322 		if (st_preempted & KVM_VCPU_FLUSH_TLB)
3323 			kvm_vcpu_flush_tlb_guest(vcpu);
3324 
3325 		if (!user_access_begin(st, sizeof(*st)))
3326 			goto dirty;
3327 	} else {
3328 		if (!user_access_begin(st, sizeof(*st)))
3329 			return;
3330 
3331 		unsafe_put_user(0, &st->preempted, out);
3332 		vcpu->arch.st.preempted = 0;
3333 	}
3334 
3335 	unsafe_get_user(version, &st->version, out);
3336 	if (version & 1)
3337 		version += 1;  /* first time write, random junk */
3338 
3339 	version += 1;
3340 	unsafe_put_user(version, &st->version, out);
3341 
3342 	smp_wmb();
3343 
3344 	unsafe_get_user(steal, &st->steal, out);
3345 	steal += current->sched_info.run_delay -
3346 		vcpu->arch.st.last_steal;
3347 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
3348 	unsafe_put_user(steal, &st->steal, out);
3349 
3350 	version += 1;
3351 	unsafe_put_user(version, &st->version, out);
3352 
3353  out:
3354 	user_access_end();
3355  dirty:
3356 	mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
3357 }
3358 
3359 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3360 {
3361 	bool pr = false;
3362 	u32 msr = msr_info->index;
3363 	u64 data = msr_info->data;
3364 
3365 	if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3366 		return kvm_xen_write_hypercall_page(vcpu, data);
3367 
3368 	switch (msr) {
3369 	case MSR_AMD64_NB_CFG:
3370 	case MSR_IA32_UCODE_WRITE:
3371 	case MSR_VM_HSAVE_PA:
3372 	case MSR_AMD64_PATCH_LOADER:
3373 	case MSR_AMD64_BU_CFG2:
3374 	case MSR_AMD64_DC_CFG:
3375 	case MSR_F15H_EX_CFG:
3376 		break;
3377 
3378 	case MSR_IA32_UCODE_REV:
3379 		if (msr_info->host_initiated)
3380 			vcpu->arch.microcode_version = data;
3381 		break;
3382 	case MSR_IA32_ARCH_CAPABILITIES:
3383 		if (!msr_info->host_initiated)
3384 			return 1;
3385 		vcpu->arch.arch_capabilities = data;
3386 		break;
3387 	case MSR_IA32_PERF_CAPABILITIES: {
3388 		struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3389 
3390 		if (!msr_info->host_initiated)
3391 			return 1;
3392 		if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3393 			return 1;
3394 		if (data & ~msr_ent.data)
3395 			return 1;
3396 
3397 		vcpu->arch.perf_capabilities = data;
3398 
3399 		return 0;
3400 		}
3401 	case MSR_EFER:
3402 		return set_efer(vcpu, msr_info);
3403 	case MSR_K7_HWCR:
3404 		data &= ~(u64)0x40;	/* ignore flush filter disable */
3405 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
3406 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
3407 
3408 		/* Handle McStatusWrEn */
3409 		if (data == BIT_ULL(18)) {
3410 			vcpu->arch.msr_hwcr = data;
3411 		} else if (data != 0) {
3412 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3413 				    data);
3414 			return 1;
3415 		}
3416 		break;
3417 	case MSR_FAM10H_MMIO_CONF_BASE:
3418 		if (data != 0) {
3419 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3420 				    "0x%llx\n", data);
3421 			return 1;
3422 		}
3423 		break;
3424 	case 0x200 ... 0x2ff:
3425 		return kvm_mtrr_set_msr(vcpu, msr, data);
3426 	case MSR_IA32_APICBASE:
3427 		return kvm_set_apic_base(vcpu, msr_info);
3428 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3429 		return kvm_x2apic_msr_write(vcpu, msr, data);
3430 	case MSR_IA32_TSC_DEADLINE:
3431 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
3432 		break;
3433 	case MSR_IA32_TSC_ADJUST:
3434 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3435 			if (!msr_info->host_initiated) {
3436 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3437 				adjust_tsc_offset_guest(vcpu, adj);
3438 				/* Before back to guest, tsc_timestamp must be adjusted
3439 				 * as well, otherwise guest's percpu pvclock time could jump.
3440 				 */
3441 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3442 			}
3443 			vcpu->arch.ia32_tsc_adjust_msr = data;
3444 		}
3445 		break;
3446 	case MSR_IA32_MISC_ENABLE:
3447 		if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3448 		    ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3449 			if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3450 				return 1;
3451 			vcpu->arch.ia32_misc_enable_msr = data;
3452 			kvm_update_cpuid_runtime(vcpu);
3453 		} else {
3454 			vcpu->arch.ia32_misc_enable_msr = data;
3455 		}
3456 		break;
3457 	case MSR_IA32_SMBASE:
3458 		if (!msr_info->host_initiated)
3459 			return 1;
3460 		vcpu->arch.smbase = data;
3461 		break;
3462 	case MSR_IA32_POWER_CTL:
3463 		vcpu->arch.msr_ia32_power_ctl = data;
3464 		break;
3465 	case MSR_IA32_TSC:
3466 		if (msr_info->host_initiated) {
3467 			kvm_synchronize_tsc(vcpu, data);
3468 		} else {
3469 			u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3470 			adjust_tsc_offset_guest(vcpu, adj);
3471 			vcpu->arch.ia32_tsc_adjust_msr += adj;
3472 		}
3473 		break;
3474 	case MSR_IA32_XSS:
3475 		if (!msr_info->host_initiated &&
3476 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3477 			return 1;
3478 		/*
3479 		 * KVM supports exposing PT to the guest, but does not support
3480 		 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3481 		 * XSAVES/XRSTORS to save/restore PT MSRs.
3482 		 */
3483 		if (data & ~supported_xss)
3484 			return 1;
3485 		vcpu->arch.ia32_xss = data;
3486 		break;
3487 	case MSR_SMI_COUNT:
3488 		if (!msr_info->host_initiated)
3489 			return 1;
3490 		vcpu->arch.smi_count = data;
3491 		break;
3492 	case MSR_KVM_WALL_CLOCK_NEW:
3493 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3494 			return 1;
3495 
3496 		vcpu->kvm->arch.wall_clock = data;
3497 		kvm_write_wall_clock(vcpu->kvm, data, 0);
3498 		break;
3499 	case MSR_KVM_WALL_CLOCK:
3500 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3501 			return 1;
3502 
3503 		vcpu->kvm->arch.wall_clock = data;
3504 		kvm_write_wall_clock(vcpu->kvm, data, 0);
3505 		break;
3506 	case MSR_KVM_SYSTEM_TIME_NEW:
3507 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3508 			return 1;
3509 
3510 		kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3511 		break;
3512 	case MSR_KVM_SYSTEM_TIME:
3513 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3514 			return 1;
3515 
3516 		kvm_write_system_time(vcpu, data, true,  msr_info->host_initiated);
3517 		break;
3518 	case MSR_KVM_ASYNC_PF_EN:
3519 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3520 			return 1;
3521 
3522 		if (kvm_pv_enable_async_pf(vcpu, data))
3523 			return 1;
3524 		break;
3525 	case MSR_KVM_ASYNC_PF_INT:
3526 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3527 			return 1;
3528 
3529 		if (kvm_pv_enable_async_pf_int(vcpu, data))
3530 			return 1;
3531 		break;
3532 	case MSR_KVM_ASYNC_PF_ACK:
3533 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3534 			return 1;
3535 		if (data & 0x1) {
3536 			vcpu->arch.apf.pageready_pending = false;
3537 			kvm_check_async_pf_completion(vcpu);
3538 		}
3539 		break;
3540 	case MSR_KVM_STEAL_TIME:
3541 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3542 			return 1;
3543 
3544 		if (unlikely(!sched_info_on()))
3545 			return 1;
3546 
3547 		if (data & KVM_STEAL_RESERVED_MASK)
3548 			return 1;
3549 
3550 		vcpu->arch.st.msr_val = data;
3551 
3552 		if (!(data & KVM_MSR_ENABLED))
3553 			break;
3554 
3555 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3556 
3557 		break;
3558 	case MSR_KVM_PV_EOI_EN:
3559 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3560 			return 1;
3561 
3562 		if (kvm_lapic_set_pv_eoi(vcpu, data, sizeof(u8)))
3563 			return 1;
3564 		break;
3565 
3566 	case MSR_KVM_POLL_CONTROL:
3567 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3568 			return 1;
3569 
3570 		/* only enable bit supported */
3571 		if (data & (-1ULL << 1))
3572 			return 1;
3573 
3574 		vcpu->arch.msr_kvm_poll_control = data;
3575 		break;
3576 
3577 	case MSR_IA32_MCG_CTL:
3578 	case MSR_IA32_MCG_STATUS:
3579 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3580 		return set_msr_mce(vcpu, msr_info);
3581 
3582 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3583 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3584 		pr = true;
3585 		fallthrough;
3586 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3587 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3588 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3589 			return kvm_pmu_set_msr(vcpu, msr_info);
3590 
3591 		if (pr || data != 0)
3592 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3593 				    "0x%x data 0x%llx\n", msr, data);
3594 		break;
3595 	case MSR_K7_CLK_CTL:
3596 		/*
3597 		 * Ignore all writes to this no longer documented MSR.
3598 		 * Writes are only relevant for old K7 processors,
3599 		 * all pre-dating SVM, but a recommended workaround from
3600 		 * AMD for these chips. It is possible to specify the
3601 		 * affected processor models on the command line, hence
3602 		 * the need to ignore the workaround.
3603 		 */
3604 		break;
3605 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3606 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3607 	case HV_X64_MSR_SYNDBG_OPTIONS:
3608 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3609 	case HV_X64_MSR_CRASH_CTL:
3610 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3611 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3612 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3613 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3614 		return kvm_hv_set_msr_common(vcpu, msr, data,
3615 					     msr_info->host_initiated);
3616 	case MSR_IA32_BBL_CR_CTL3:
3617 		/* Drop writes to this legacy MSR -- see rdmsr
3618 		 * counterpart for further detail.
3619 		 */
3620 		if (report_ignored_msrs)
3621 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3622 				msr, data);
3623 		break;
3624 	case MSR_AMD64_OSVW_ID_LENGTH:
3625 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3626 			return 1;
3627 		vcpu->arch.osvw.length = data;
3628 		break;
3629 	case MSR_AMD64_OSVW_STATUS:
3630 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3631 			return 1;
3632 		vcpu->arch.osvw.status = data;
3633 		break;
3634 	case MSR_PLATFORM_INFO:
3635 		if (!msr_info->host_initiated ||
3636 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3637 		     cpuid_fault_enabled(vcpu)))
3638 			return 1;
3639 		vcpu->arch.msr_platform_info = data;
3640 		break;
3641 	case MSR_MISC_FEATURES_ENABLES:
3642 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3643 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3644 		     !supports_cpuid_fault(vcpu)))
3645 			return 1;
3646 		vcpu->arch.msr_misc_features_enables = data;
3647 		break;
3648 	default:
3649 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3650 			return kvm_pmu_set_msr(vcpu, msr_info);
3651 		return KVM_MSR_RET_INVALID;
3652 	}
3653 	return 0;
3654 }
3655 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3656 
3657 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3658 {
3659 	u64 data;
3660 	u64 mcg_cap = vcpu->arch.mcg_cap;
3661 	unsigned bank_num = mcg_cap & 0xff;
3662 
3663 	switch (msr) {
3664 	case MSR_IA32_P5_MC_ADDR:
3665 	case MSR_IA32_P5_MC_TYPE:
3666 		data = 0;
3667 		break;
3668 	case MSR_IA32_MCG_CAP:
3669 		data = vcpu->arch.mcg_cap;
3670 		break;
3671 	case MSR_IA32_MCG_CTL:
3672 		if (!(mcg_cap & MCG_CTL_P) && !host)
3673 			return 1;
3674 		data = vcpu->arch.mcg_ctl;
3675 		break;
3676 	case MSR_IA32_MCG_STATUS:
3677 		data = vcpu->arch.mcg_status;
3678 		break;
3679 	default:
3680 		if (msr >= MSR_IA32_MC0_CTL &&
3681 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
3682 			u32 offset = array_index_nospec(
3683 				msr - MSR_IA32_MC0_CTL,
3684 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3685 
3686 			data = vcpu->arch.mce_banks[offset];
3687 			break;
3688 		}
3689 		return 1;
3690 	}
3691 	*pdata = data;
3692 	return 0;
3693 }
3694 
3695 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3696 {
3697 	switch (msr_info->index) {
3698 	case MSR_IA32_PLATFORM_ID:
3699 	case MSR_IA32_EBL_CR_POWERON:
3700 	case MSR_IA32_LASTBRANCHFROMIP:
3701 	case MSR_IA32_LASTBRANCHTOIP:
3702 	case MSR_IA32_LASTINTFROMIP:
3703 	case MSR_IA32_LASTINTTOIP:
3704 	case MSR_AMD64_SYSCFG:
3705 	case MSR_K8_TSEG_ADDR:
3706 	case MSR_K8_TSEG_MASK:
3707 	case MSR_VM_HSAVE_PA:
3708 	case MSR_K8_INT_PENDING_MSG:
3709 	case MSR_AMD64_NB_CFG:
3710 	case MSR_FAM10H_MMIO_CONF_BASE:
3711 	case MSR_AMD64_BU_CFG2:
3712 	case MSR_IA32_PERF_CTL:
3713 	case MSR_AMD64_DC_CFG:
3714 	case MSR_F15H_EX_CFG:
3715 	/*
3716 	 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3717 	 * limit) MSRs. Just return 0, as we do not want to expose the host
3718 	 * data here. Do not conditionalize this on CPUID, as KVM does not do
3719 	 * so for existing CPU-specific MSRs.
3720 	 */
3721 	case MSR_RAPL_POWER_UNIT:
3722 	case MSR_PP0_ENERGY_STATUS:	/* Power plane 0 (core) */
3723 	case MSR_PP1_ENERGY_STATUS:	/* Power plane 1 (graphics uncore) */
3724 	case MSR_PKG_ENERGY_STATUS:	/* Total package */
3725 	case MSR_DRAM_ENERGY_STATUS:	/* DRAM controller */
3726 		msr_info->data = 0;
3727 		break;
3728 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3729 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3730 			return kvm_pmu_get_msr(vcpu, msr_info);
3731 		if (!msr_info->host_initiated)
3732 			return 1;
3733 		msr_info->data = 0;
3734 		break;
3735 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3736 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3737 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3738 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3739 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3740 			return kvm_pmu_get_msr(vcpu, msr_info);
3741 		msr_info->data = 0;
3742 		break;
3743 	case MSR_IA32_UCODE_REV:
3744 		msr_info->data = vcpu->arch.microcode_version;
3745 		break;
3746 	case MSR_IA32_ARCH_CAPABILITIES:
3747 		if (!msr_info->host_initiated &&
3748 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3749 			return 1;
3750 		msr_info->data = vcpu->arch.arch_capabilities;
3751 		break;
3752 	case MSR_IA32_PERF_CAPABILITIES:
3753 		if (!msr_info->host_initiated &&
3754 		    !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3755 			return 1;
3756 		msr_info->data = vcpu->arch.perf_capabilities;
3757 		break;
3758 	case MSR_IA32_POWER_CTL:
3759 		msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3760 		break;
3761 	case MSR_IA32_TSC: {
3762 		/*
3763 		 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3764 		 * even when not intercepted. AMD manual doesn't explicitly
3765 		 * state this but appears to behave the same.
3766 		 *
3767 		 * On userspace reads and writes, however, we unconditionally
3768 		 * return L1's TSC value to ensure backwards-compatible
3769 		 * behavior for migration.
3770 		 */
3771 		u64 offset, ratio;
3772 
3773 		if (msr_info->host_initiated) {
3774 			offset = vcpu->arch.l1_tsc_offset;
3775 			ratio = vcpu->arch.l1_tsc_scaling_ratio;
3776 		} else {
3777 			offset = vcpu->arch.tsc_offset;
3778 			ratio = vcpu->arch.tsc_scaling_ratio;
3779 		}
3780 
3781 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc(), ratio) + offset;
3782 		break;
3783 	}
3784 	case MSR_MTRRcap:
3785 	case 0x200 ... 0x2ff:
3786 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3787 	case 0xcd: /* fsb frequency */
3788 		msr_info->data = 3;
3789 		break;
3790 		/*
3791 		 * MSR_EBC_FREQUENCY_ID
3792 		 * Conservative value valid for even the basic CPU models.
3793 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3794 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3795 		 * and 266MHz for model 3, or 4. Set Core Clock
3796 		 * Frequency to System Bus Frequency Ratio to 1 (bits
3797 		 * 31:24) even though these are only valid for CPU
3798 		 * models > 2, however guests may end up dividing or
3799 		 * multiplying by zero otherwise.
3800 		 */
3801 	case MSR_EBC_FREQUENCY_ID:
3802 		msr_info->data = 1 << 24;
3803 		break;
3804 	case MSR_IA32_APICBASE:
3805 		msr_info->data = kvm_get_apic_base(vcpu);
3806 		break;
3807 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3808 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3809 	case MSR_IA32_TSC_DEADLINE:
3810 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3811 		break;
3812 	case MSR_IA32_TSC_ADJUST:
3813 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3814 		break;
3815 	case MSR_IA32_MISC_ENABLE:
3816 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3817 		break;
3818 	case MSR_IA32_SMBASE:
3819 		if (!msr_info->host_initiated)
3820 			return 1;
3821 		msr_info->data = vcpu->arch.smbase;
3822 		break;
3823 	case MSR_SMI_COUNT:
3824 		msr_info->data = vcpu->arch.smi_count;
3825 		break;
3826 	case MSR_IA32_PERF_STATUS:
3827 		/* TSC increment by tick */
3828 		msr_info->data = 1000ULL;
3829 		/* CPU multiplier */
3830 		msr_info->data |= (((uint64_t)4ULL) << 40);
3831 		break;
3832 	case MSR_EFER:
3833 		msr_info->data = vcpu->arch.efer;
3834 		break;
3835 	case MSR_KVM_WALL_CLOCK:
3836 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3837 			return 1;
3838 
3839 		msr_info->data = vcpu->kvm->arch.wall_clock;
3840 		break;
3841 	case MSR_KVM_WALL_CLOCK_NEW:
3842 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3843 			return 1;
3844 
3845 		msr_info->data = vcpu->kvm->arch.wall_clock;
3846 		break;
3847 	case MSR_KVM_SYSTEM_TIME:
3848 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3849 			return 1;
3850 
3851 		msr_info->data = vcpu->arch.time;
3852 		break;
3853 	case MSR_KVM_SYSTEM_TIME_NEW:
3854 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3855 			return 1;
3856 
3857 		msr_info->data = vcpu->arch.time;
3858 		break;
3859 	case MSR_KVM_ASYNC_PF_EN:
3860 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3861 			return 1;
3862 
3863 		msr_info->data = vcpu->arch.apf.msr_en_val;
3864 		break;
3865 	case MSR_KVM_ASYNC_PF_INT:
3866 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3867 			return 1;
3868 
3869 		msr_info->data = vcpu->arch.apf.msr_int_val;
3870 		break;
3871 	case MSR_KVM_ASYNC_PF_ACK:
3872 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3873 			return 1;
3874 
3875 		msr_info->data = 0;
3876 		break;
3877 	case MSR_KVM_STEAL_TIME:
3878 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3879 			return 1;
3880 
3881 		msr_info->data = vcpu->arch.st.msr_val;
3882 		break;
3883 	case MSR_KVM_PV_EOI_EN:
3884 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3885 			return 1;
3886 
3887 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
3888 		break;
3889 	case MSR_KVM_POLL_CONTROL:
3890 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3891 			return 1;
3892 
3893 		msr_info->data = vcpu->arch.msr_kvm_poll_control;
3894 		break;
3895 	case MSR_IA32_P5_MC_ADDR:
3896 	case MSR_IA32_P5_MC_TYPE:
3897 	case MSR_IA32_MCG_CAP:
3898 	case MSR_IA32_MCG_CTL:
3899 	case MSR_IA32_MCG_STATUS:
3900 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3901 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3902 				   msr_info->host_initiated);
3903 	case MSR_IA32_XSS:
3904 		if (!msr_info->host_initiated &&
3905 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3906 			return 1;
3907 		msr_info->data = vcpu->arch.ia32_xss;
3908 		break;
3909 	case MSR_K7_CLK_CTL:
3910 		/*
3911 		 * Provide expected ramp-up count for K7. All other
3912 		 * are set to zero, indicating minimum divisors for
3913 		 * every field.
3914 		 *
3915 		 * This prevents guest kernels on AMD host with CPU
3916 		 * type 6, model 8 and higher from exploding due to
3917 		 * the rdmsr failing.
3918 		 */
3919 		msr_info->data = 0x20000000;
3920 		break;
3921 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3922 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3923 	case HV_X64_MSR_SYNDBG_OPTIONS:
3924 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3925 	case HV_X64_MSR_CRASH_CTL:
3926 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3927 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3928 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3929 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3930 		return kvm_hv_get_msr_common(vcpu,
3931 					     msr_info->index, &msr_info->data,
3932 					     msr_info->host_initiated);
3933 	case MSR_IA32_BBL_CR_CTL3:
3934 		/* This legacy MSR exists but isn't fully documented in current
3935 		 * silicon.  It is however accessed by winxp in very narrow
3936 		 * scenarios where it sets bit #19, itself documented as
3937 		 * a "reserved" bit.  Best effort attempt to source coherent
3938 		 * read data here should the balance of the register be
3939 		 * interpreted by the guest:
3940 		 *
3941 		 * L2 cache control register 3: 64GB range, 256KB size,
3942 		 * enabled, latency 0x1, configured
3943 		 */
3944 		msr_info->data = 0xbe702111;
3945 		break;
3946 	case MSR_AMD64_OSVW_ID_LENGTH:
3947 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3948 			return 1;
3949 		msr_info->data = vcpu->arch.osvw.length;
3950 		break;
3951 	case MSR_AMD64_OSVW_STATUS:
3952 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3953 			return 1;
3954 		msr_info->data = vcpu->arch.osvw.status;
3955 		break;
3956 	case MSR_PLATFORM_INFO:
3957 		if (!msr_info->host_initiated &&
3958 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3959 			return 1;
3960 		msr_info->data = vcpu->arch.msr_platform_info;
3961 		break;
3962 	case MSR_MISC_FEATURES_ENABLES:
3963 		msr_info->data = vcpu->arch.msr_misc_features_enables;
3964 		break;
3965 	case MSR_K7_HWCR:
3966 		msr_info->data = vcpu->arch.msr_hwcr;
3967 		break;
3968 	default:
3969 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3970 			return kvm_pmu_get_msr(vcpu, msr_info);
3971 		return KVM_MSR_RET_INVALID;
3972 	}
3973 	return 0;
3974 }
3975 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3976 
3977 /*
3978  * Read or write a bunch of msrs. All parameters are kernel addresses.
3979  *
3980  * @return number of msrs set successfully.
3981  */
3982 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3983 		    struct kvm_msr_entry *entries,
3984 		    int (*do_msr)(struct kvm_vcpu *vcpu,
3985 				  unsigned index, u64 *data))
3986 {
3987 	int i;
3988 
3989 	for (i = 0; i < msrs->nmsrs; ++i)
3990 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
3991 			break;
3992 
3993 	return i;
3994 }
3995 
3996 /*
3997  * Read or write a bunch of msrs. Parameters are user addresses.
3998  *
3999  * @return number of msrs set successfully.
4000  */
4001 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
4002 		  int (*do_msr)(struct kvm_vcpu *vcpu,
4003 				unsigned index, u64 *data),
4004 		  int writeback)
4005 {
4006 	struct kvm_msrs msrs;
4007 	struct kvm_msr_entry *entries;
4008 	int r, n;
4009 	unsigned size;
4010 
4011 	r = -EFAULT;
4012 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
4013 		goto out;
4014 
4015 	r = -E2BIG;
4016 	if (msrs.nmsrs >= MAX_IO_MSRS)
4017 		goto out;
4018 
4019 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
4020 	entries = memdup_user(user_msrs->entries, size);
4021 	if (IS_ERR(entries)) {
4022 		r = PTR_ERR(entries);
4023 		goto out;
4024 	}
4025 
4026 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
4027 	if (r < 0)
4028 		goto out_free;
4029 
4030 	r = -EFAULT;
4031 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
4032 		goto out_free;
4033 
4034 	r = n;
4035 
4036 out_free:
4037 	kfree(entries);
4038 out:
4039 	return r;
4040 }
4041 
4042 static inline bool kvm_can_mwait_in_guest(void)
4043 {
4044 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
4045 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
4046 		boot_cpu_has(X86_FEATURE_ARAT);
4047 }
4048 
4049 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
4050 					    struct kvm_cpuid2 __user *cpuid_arg)
4051 {
4052 	struct kvm_cpuid2 cpuid;
4053 	int r;
4054 
4055 	r = -EFAULT;
4056 	if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4057 		return r;
4058 
4059 	r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4060 	if (r)
4061 		return r;
4062 
4063 	r = -EFAULT;
4064 	if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4065 		return r;
4066 
4067 	return 0;
4068 }
4069 
4070 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
4071 {
4072 	int r = 0;
4073 
4074 	switch (ext) {
4075 	case KVM_CAP_IRQCHIP:
4076 	case KVM_CAP_HLT:
4077 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
4078 	case KVM_CAP_SET_TSS_ADDR:
4079 	case KVM_CAP_EXT_CPUID:
4080 	case KVM_CAP_EXT_EMUL_CPUID:
4081 	case KVM_CAP_CLOCKSOURCE:
4082 	case KVM_CAP_PIT:
4083 	case KVM_CAP_NOP_IO_DELAY:
4084 	case KVM_CAP_MP_STATE:
4085 	case KVM_CAP_SYNC_MMU:
4086 	case KVM_CAP_USER_NMI:
4087 	case KVM_CAP_REINJECT_CONTROL:
4088 	case KVM_CAP_IRQ_INJECT_STATUS:
4089 	case KVM_CAP_IOEVENTFD:
4090 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
4091 	case KVM_CAP_PIT2:
4092 	case KVM_CAP_PIT_STATE2:
4093 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
4094 	case KVM_CAP_VCPU_EVENTS:
4095 	case KVM_CAP_HYPERV:
4096 	case KVM_CAP_HYPERV_VAPIC:
4097 	case KVM_CAP_HYPERV_SPIN:
4098 	case KVM_CAP_HYPERV_SYNIC:
4099 	case KVM_CAP_HYPERV_SYNIC2:
4100 	case KVM_CAP_HYPERV_VP_INDEX:
4101 	case KVM_CAP_HYPERV_EVENTFD:
4102 	case KVM_CAP_HYPERV_TLBFLUSH:
4103 	case KVM_CAP_HYPERV_SEND_IPI:
4104 	case KVM_CAP_HYPERV_CPUID:
4105 	case KVM_CAP_HYPERV_ENFORCE_CPUID:
4106 	case KVM_CAP_SYS_HYPERV_CPUID:
4107 	case KVM_CAP_PCI_SEGMENT:
4108 	case KVM_CAP_DEBUGREGS:
4109 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
4110 	case KVM_CAP_XSAVE:
4111 	case KVM_CAP_ASYNC_PF:
4112 	case KVM_CAP_ASYNC_PF_INT:
4113 	case KVM_CAP_GET_TSC_KHZ:
4114 	case KVM_CAP_KVMCLOCK_CTRL:
4115 	case KVM_CAP_READONLY_MEM:
4116 	case KVM_CAP_HYPERV_TIME:
4117 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
4118 	case KVM_CAP_TSC_DEADLINE_TIMER:
4119 	case KVM_CAP_DISABLE_QUIRKS:
4120 	case KVM_CAP_SET_BOOT_CPU_ID:
4121  	case KVM_CAP_SPLIT_IRQCHIP:
4122 	case KVM_CAP_IMMEDIATE_EXIT:
4123 	case KVM_CAP_PMU_EVENT_FILTER:
4124 	case KVM_CAP_GET_MSR_FEATURES:
4125 	case KVM_CAP_MSR_PLATFORM_INFO:
4126 	case KVM_CAP_EXCEPTION_PAYLOAD:
4127 	case KVM_CAP_SET_GUEST_DEBUG:
4128 	case KVM_CAP_LAST_CPU:
4129 	case KVM_CAP_X86_USER_SPACE_MSR:
4130 	case KVM_CAP_X86_MSR_FILTER:
4131 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4132 #ifdef CONFIG_X86_SGX_KVM
4133 	case KVM_CAP_SGX_ATTRIBUTE:
4134 #endif
4135 	case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
4136 	case KVM_CAP_SREGS2:
4137 	case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
4138 	case KVM_CAP_VCPU_ATTRIBUTES:
4139 		r = 1;
4140 		break;
4141 	case KVM_CAP_EXIT_HYPERCALL:
4142 		r = KVM_EXIT_HYPERCALL_VALID_MASK;
4143 		break;
4144 	case KVM_CAP_SET_GUEST_DEBUG2:
4145 		return KVM_GUESTDBG_VALID_MASK;
4146 #ifdef CONFIG_KVM_XEN
4147 	case KVM_CAP_XEN_HVM:
4148 		r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
4149 		    KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
4150 		    KVM_XEN_HVM_CONFIG_SHARED_INFO;
4151 		if (sched_info_on())
4152 			r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
4153 		break;
4154 #endif
4155 	case KVM_CAP_SYNC_REGS:
4156 		r = KVM_SYNC_X86_VALID_FIELDS;
4157 		break;
4158 	case KVM_CAP_ADJUST_CLOCK:
4159 		r = KVM_CLOCK_VALID_FLAGS;
4160 		break;
4161 	case KVM_CAP_X86_DISABLE_EXITS:
4162 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
4163 		      KVM_X86_DISABLE_EXITS_CSTATE;
4164 		if(kvm_can_mwait_in_guest())
4165 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
4166 		break;
4167 	case KVM_CAP_X86_SMM:
4168 		/* SMBASE is usually relocated above 1M on modern chipsets,
4169 		 * and SMM handlers might indeed rely on 4G segment limits,
4170 		 * so do not report SMM to be available if real mode is
4171 		 * emulated via vm86 mode.  Still, do not go to great lengths
4172 		 * to avoid userspace's usage of the feature, because it is a
4173 		 * fringe case that is not enabled except via specific settings
4174 		 * of the module parameters.
4175 		 */
4176 		r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
4177 		break;
4178 	case KVM_CAP_VAPIC:
4179 		r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
4180 		break;
4181 	case KVM_CAP_NR_VCPUS:
4182 		r = num_online_cpus();
4183 		break;
4184 	case KVM_CAP_MAX_VCPUS:
4185 		r = KVM_MAX_VCPUS;
4186 		break;
4187 	case KVM_CAP_MAX_VCPU_ID:
4188 		r = KVM_MAX_VCPU_IDS;
4189 		break;
4190 	case KVM_CAP_PV_MMU:	/* obsolete */
4191 		r = 0;
4192 		break;
4193 	case KVM_CAP_MCE:
4194 		r = KVM_MAX_MCE_BANKS;
4195 		break;
4196 	case KVM_CAP_XCRS:
4197 		r = boot_cpu_has(X86_FEATURE_XSAVE);
4198 		break;
4199 	case KVM_CAP_TSC_CONTROL:
4200 		r = kvm_has_tsc_control;
4201 		break;
4202 	case KVM_CAP_X2APIC_API:
4203 		r = KVM_X2APIC_API_VALID_FLAGS;
4204 		break;
4205 	case KVM_CAP_NESTED_STATE:
4206 		r = kvm_x86_ops.nested_ops->get_state ?
4207 			kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
4208 		break;
4209 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4210 		r = kvm_x86_ops.enable_direct_tlbflush != NULL;
4211 		break;
4212 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4213 		r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
4214 		break;
4215 	case KVM_CAP_SMALLER_MAXPHYADDR:
4216 		r = (int) allow_smaller_maxphyaddr;
4217 		break;
4218 	case KVM_CAP_STEAL_TIME:
4219 		r = sched_info_on();
4220 		break;
4221 	case KVM_CAP_X86_BUS_LOCK_EXIT:
4222 		if (kvm_has_bus_lock_exit)
4223 			r = KVM_BUS_LOCK_DETECTION_OFF |
4224 			    KVM_BUS_LOCK_DETECTION_EXIT;
4225 		else
4226 			r = 0;
4227 		break;
4228 	default:
4229 		break;
4230 	}
4231 	return r;
4232 
4233 }
4234 
4235 long kvm_arch_dev_ioctl(struct file *filp,
4236 			unsigned int ioctl, unsigned long arg)
4237 {
4238 	void __user *argp = (void __user *)arg;
4239 	long r;
4240 
4241 	switch (ioctl) {
4242 	case KVM_GET_MSR_INDEX_LIST: {
4243 		struct kvm_msr_list __user *user_msr_list = argp;
4244 		struct kvm_msr_list msr_list;
4245 		unsigned n;
4246 
4247 		r = -EFAULT;
4248 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4249 			goto out;
4250 		n = msr_list.nmsrs;
4251 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4252 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4253 			goto out;
4254 		r = -E2BIG;
4255 		if (n < msr_list.nmsrs)
4256 			goto out;
4257 		r = -EFAULT;
4258 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4259 				 num_msrs_to_save * sizeof(u32)))
4260 			goto out;
4261 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4262 				 &emulated_msrs,
4263 				 num_emulated_msrs * sizeof(u32)))
4264 			goto out;
4265 		r = 0;
4266 		break;
4267 	}
4268 	case KVM_GET_SUPPORTED_CPUID:
4269 	case KVM_GET_EMULATED_CPUID: {
4270 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4271 		struct kvm_cpuid2 cpuid;
4272 
4273 		r = -EFAULT;
4274 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4275 			goto out;
4276 
4277 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4278 					    ioctl);
4279 		if (r)
4280 			goto out;
4281 
4282 		r = -EFAULT;
4283 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4284 			goto out;
4285 		r = 0;
4286 		break;
4287 	}
4288 	case KVM_X86_GET_MCE_CAP_SUPPORTED:
4289 		r = -EFAULT;
4290 		if (copy_to_user(argp, &kvm_mce_cap_supported,
4291 				 sizeof(kvm_mce_cap_supported)))
4292 			goto out;
4293 		r = 0;
4294 		break;
4295 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4296 		struct kvm_msr_list __user *user_msr_list = argp;
4297 		struct kvm_msr_list msr_list;
4298 		unsigned int n;
4299 
4300 		r = -EFAULT;
4301 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4302 			goto out;
4303 		n = msr_list.nmsrs;
4304 		msr_list.nmsrs = num_msr_based_features;
4305 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4306 			goto out;
4307 		r = -E2BIG;
4308 		if (n < msr_list.nmsrs)
4309 			goto out;
4310 		r = -EFAULT;
4311 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
4312 				 num_msr_based_features * sizeof(u32)))
4313 			goto out;
4314 		r = 0;
4315 		break;
4316 	}
4317 	case KVM_GET_MSRS:
4318 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
4319 		break;
4320 	case KVM_GET_SUPPORTED_HV_CPUID:
4321 		r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4322 		break;
4323 	default:
4324 		r = -EINVAL;
4325 		break;
4326 	}
4327 out:
4328 	return r;
4329 }
4330 
4331 static void wbinvd_ipi(void *garbage)
4332 {
4333 	wbinvd();
4334 }
4335 
4336 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4337 {
4338 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4339 }
4340 
4341 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4342 {
4343 	/* Address WBINVD may be executed by guest */
4344 	if (need_emulate_wbinvd(vcpu)) {
4345 		if (static_call(kvm_x86_has_wbinvd_exit)())
4346 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4347 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4348 			smp_call_function_single(vcpu->cpu,
4349 					wbinvd_ipi, NULL, 1);
4350 	}
4351 
4352 	static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4353 
4354 	/* Save host pkru register if supported */
4355 	vcpu->arch.host_pkru = read_pkru();
4356 
4357 	/* Apply any externally detected TSC adjustments (due to suspend) */
4358 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4359 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4360 		vcpu->arch.tsc_offset_adjustment = 0;
4361 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4362 	}
4363 
4364 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4365 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4366 				rdtsc() - vcpu->arch.last_host_tsc;
4367 		if (tsc_delta < 0)
4368 			mark_tsc_unstable("KVM discovered backwards TSC");
4369 
4370 		if (kvm_check_tsc_unstable()) {
4371 			u64 offset = kvm_compute_l1_tsc_offset(vcpu,
4372 						vcpu->arch.last_guest_tsc);
4373 			kvm_vcpu_write_tsc_offset(vcpu, offset);
4374 			vcpu->arch.tsc_catchup = 1;
4375 		}
4376 
4377 		if (kvm_lapic_hv_timer_in_use(vcpu))
4378 			kvm_lapic_restart_hv_timer(vcpu);
4379 
4380 		/*
4381 		 * On a host with synchronized TSC, there is no need to update
4382 		 * kvmclock on vcpu->cpu migration
4383 		 */
4384 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4385 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4386 		if (vcpu->cpu != cpu)
4387 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4388 		vcpu->cpu = cpu;
4389 	}
4390 
4391 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4392 }
4393 
4394 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4395 {
4396 	struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
4397 	struct kvm_steal_time __user *st;
4398 	struct kvm_memslots *slots;
4399 	static const u8 preempted = KVM_VCPU_PREEMPTED;
4400 
4401 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4402 		return;
4403 
4404 	if (vcpu->arch.st.preempted)
4405 		return;
4406 
4407 	/* This happens on process exit */
4408 	if (unlikely(current->mm != vcpu->kvm->mm))
4409 		return;
4410 
4411 	slots = kvm_memslots(vcpu->kvm);
4412 
4413 	if (unlikely(slots->generation != ghc->generation ||
4414 		     kvm_is_error_hva(ghc->hva) || !ghc->memslot))
4415 		return;
4416 
4417 	st = (struct kvm_steal_time __user *)ghc->hva;
4418 	BUILD_BUG_ON(sizeof(st->preempted) != sizeof(preempted));
4419 
4420 	if (!copy_to_user_nofault(&st->preempted, &preempted, sizeof(preempted)))
4421 		vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4422 
4423 	mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
4424 }
4425 
4426 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4427 {
4428 	int idx;
4429 
4430 	if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4431 		vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4432 
4433 	/*
4434 	 * Take the srcu lock as memslots will be accessed to check the gfn
4435 	 * cache generation against the memslots generation.
4436 	 */
4437 	idx = srcu_read_lock(&vcpu->kvm->srcu);
4438 	if (kvm_xen_msr_enabled(vcpu->kvm))
4439 		kvm_xen_runstate_set_preempted(vcpu);
4440 	else
4441 		kvm_steal_time_set_preempted(vcpu);
4442 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
4443 
4444 	static_call(kvm_x86_vcpu_put)(vcpu);
4445 	vcpu->arch.last_host_tsc = rdtsc();
4446 }
4447 
4448 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4449 				    struct kvm_lapic_state *s)
4450 {
4451 	if (vcpu->arch.apicv_active)
4452 		static_call(kvm_x86_sync_pir_to_irr)(vcpu);
4453 
4454 	return kvm_apic_get_state(vcpu, s);
4455 }
4456 
4457 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4458 				    struct kvm_lapic_state *s)
4459 {
4460 	int r;
4461 
4462 	r = kvm_apic_set_state(vcpu, s);
4463 	if (r)
4464 		return r;
4465 	update_cr8_intercept(vcpu);
4466 
4467 	return 0;
4468 }
4469 
4470 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4471 {
4472 	/*
4473 	 * We can accept userspace's request for interrupt injection
4474 	 * as long as we have a place to store the interrupt number.
4475 	 * The actual injection will happen when the CPU is able to
4476 	 * deliver the interrupt.
4477 	 */
4478 	if (kvm_cpu_has_extint(vcpu))
4479 		return false;
4480 
4481 	/* Acknowledging ExtINT does not happen if LINT0 is masked.  */
4482 	return (!lapic_in_kernel(vcpu) ||
4483 		kvm_apic_accept_pic_intr(vcpu));
4484 }
4485 
4486 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4487 {
4488 	/*
4489 	 * Do not cause an interrupt window exit if an exception
4490 	 * is pending or an event needs reinjection; userspace
4491 	 * might want to inject the interrupt manually using KVM_SET_REGS
4492 	 * or KVM_SET_SREGS.  For that to work, we must be at an
4493 	 * instruction boundary and with no events half-injected.
4494 	 */
4495 	return (kvm_arch_interrupt_allowed(vcpu) &&
4496 		kvm_cpu_accept_dm_intr(vcpu) &&
4497 		!kvm_event_needs_reinjection(vcpu) &&
4498 		!vcpu->arch.exception.pending);
4499 }
4500 
4501 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4502 				    struct kvm_interrupt *irq)
4503 {
4504 	if (irq->irq >= KVM_NR_INTERRUPTS)
4505 		return -EINVAL;
4506 
4507 	if (!irqchip_in_kernel(vcpu->kvm)) {
4508 		kvm_queue_interrupt(vcpu, irq->irq, false);
4509 		kvm_make_request(KVM_REQ_EVENT, vcpu);
4510 		return 0;
4511 	}
4512 
4513 	/*
4514 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4515 	 * fail for in-kernel 8259.
4516 	 */
4517 	if (pic_in_kernel(vcpu->kvm))
4518 		return -ENXIO;
4519 
4520 	if (vcpu->arch.pending_external_vector != -1)
4521 		return -EEXIST;
4522 
4523 	vcpu->arch.pending_external_vector = irq->irq;
4524 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4525 	return 0;
4526 }
4527 
4528 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4529 {
4530 	kvm_inject_nmi(vcpu);
4531 
4532 	return 0;
4533 }
4534 
4535 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4536 {
4537 	kvm_make_request(KVM_REQ_SMI, vcpu);
4538 
4539 	return 0;
4540 }
4541 
4542 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4543 					   struct kvm_tpr_access_ctl *tac)
4544 {
4545 	if (tac->flags)
4546 		return -EINVAL;
4547 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
4548 	return 0;
4549 }
4550 
4551 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4552 					u64 mcg_cap)
4553 {
4554 	int r;
4555 	unsigned bank_num = mcg_cap & 0xff, bank;
4556 
4557 	r = -EINVAL;
4558 	if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4559 		goto out;
4560 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4561 		goto out;
4562 	r = 0;
4563 	vcpu->arch.mcg_cap = mcg_cap;
4564 	/* Init IA32_MCG_CTL to all 1s */
4565 	if (mcg_cap & MCG_CTL_P)
4566 		vcpu->arch.mcg_ctl = ~(u64)0;
4567 	/* Init IA32_MCi_CTL to all 1s */
4568 	for (bank = 0; bank < bank_num; bank++)
4569 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4570 
4571 	static_call(kvm_x86_setup_mce)(vcpu);
4572 out:
4573 	return r;
4574 }
4575 
4576 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4577 				      struct kvm_x86_mce *mce)
4578 {
4579 	u64 mcg_cap = vcpu->arch.mcg_cap;
4580 	unsigned bank_num = mcg_cap & 0xff;
4581 	u64 *banks = vcpu->arch.mce_banks;
4582 
4583 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4584 		return -EINVAL;
4585 	/*
4586 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4587 	 * reporting is disabled
4588 	 */
4589 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4590 	    vcpu->arch.mcg_ctl != ~(u64)0)
4591 		return 0;
4592 	banks += 4 * mce->bank;
4593 	/*
4594 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4595 	 * reporting is disabled for the bank
4596 	 */
4597 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4598 		return 0;
4599 	if (mce->status & MCI_STATUS_UC) {
4600 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4601 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4602 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4603 			return 0;
4604 		}
4605 		if (banks[1] & MCI_STATUS_VAL)
4606 			mce->status |= MCI_STATUS_OVER;
4607 		banks[2] = mce->addr;
4608 		banks[3] = mce->misc;
4609 		vcpu->arch.mcg_status = mce->mcg_status;
4610 		banks[1] = mce->status;
4611 		kvm_queue_exception(vcpu, MC_VECTOR);
4612 	} else if (!(banks[1] & MCI_STATUS_VAL)
4613 		   || !(banks[1] & MCI_STATUS_UC)) {
4614 		if (banks[1] & MCI_STATUS_VAL)
4615 			mce->status |= MCI_STATUS_OVER;
4616 		banks[2] = mce->addr;
4617 		banks[3] = mce->misc;
4618 		banks[1] = mce->status;
4619 	} else
4620 		banks[1] |= MCI_STATUS_OVER;
4621 	return 0;
4622 }
4623 
4624 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4625 					       struct kvm_vcpu_events *events)
4626 {
4627 	process_nmi(vcpu);
4628 
4629 	if (kvm_check_request(KVM_REQ_SMI, vcpu))
4630 		process_smi(vcpu);
4631 
4632 	/*
4633 	 * In guest mode, payload delivery should be deferred,
4634 	 * so that the L1 hypervisor can intercept #PF before
4635 	 * CR2 is modified (or intercept #DB before DR6 is
4636 	 * modified under nVMX). Unless the per-VM capability,
4637 	 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4638 	 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4639 	 * opportunistically defer the exception payload, deliver it if the
4640 	 * capability hasn't been requested before processing a
4641 	 * KVM_GET_VCPU_EVENTS.
4642 	 */
4643 	if (!vcpu->kvm->arch.exception_payload_enabled &&
4644 	    vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4645 		kvm_deliver_exception_payload(vcpu);
4646 
4647 	/*
4648 	 * The API doesn't provide the instruction length for software
4649 	 * exceptions, so don't report them. As long as the guest RIP
4650 	 * isn't advanced, we should expect to encounter the exception
4651 	 * again.
4652 	 */
4653 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4654 		events->exception.injected = 0;
4655 		events->exception.pending = 0;
4656 	} else {
4657 		events->exception.injected = vcpu->arch.exception.injected;
4658 		events->exception.pending = vcpu->arch.exception.pending;
4659 		/*
4660 		 * For ABI compatibility, deliberately conflate
4661 		 * pending and injected exceptions when
4662 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4663 		 */
4664 		if (!vcpu->kvm->arch.exception_payload_enabled)
4665 			events->exception.injected |=
4666 				vcpu->arch.exception.pending;
4667 	}
4668 	events->exception.nr = vcpu->arch.exception.nr;
4669 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4670 	events->exception.error_code = vcpu->arch.exception.error_code;
4671 	events->exception_has_payload = vcpu->arch.exception.has_payload;
4672 	events->exception_payload = vcpu->arch.exception.payload;
4673 
4674 	events->interrupt.injected =
4675 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4676 	events->interrupt.nr = vcpu->arch.interrupt.nr;
4677 	events->interrupt.soft = 0;
4678 	events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4679 
4680 	events->nmi.injected = vcpu->arch.nmi_injected;
4681 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
4682 	events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4683 	events->nmi.pad = 0;
4684 
4685 	events->sipi_vector = 0; /* never valid when reporting to user space */
4686 
4687 	events->smi.smm = is_smm(vcpu);
4688 	events->smi.pending = vcpu->arch.smi_pending;
4689 	events->smi.smm_inside_nmi =
4690 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4691 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4692 
4693 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4694 			 | KVM_VCPUEVENT_VALID_SHADOW
4695 			 | KVM_VCPUEVENT_VALID_SMM);
4696 	if (vcpu->kvm->arch.exception_payload_enabled)
4697 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4698 
4699 	memset(&events->reserved, 0, sizeof(events->reserved));
4700 }
4701 
4702 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm);
4703 
4704 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4705 					      struct kvm_vcpu_events *events)
4706 {
4707 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4708 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4709 			      | KVM_VCPUEVENT_VALID_SHADOW
4710 			      | KVM_VCPUEVENT_VALID_SMM
4711 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
4712 		return -EINVAL;
4713 
4714 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4715 		if (!vcpu->kvm->arch.exception_payload_enabled)
4716 			return -EINVAL;
4717 		if (events->exception.pending)
4718 			events->exception.injected = 0;
4719 		else
4720 			events->exception_has_payload = 0;
4721 	} else {
4722 		events->exception.pending = 0;
4723 		events->exception_has_payload = 0;
4724 	}
4725 
4726 	if ((events->exception.injected || events->exception.pending) &&
4727 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4728 		return -EINVAL;
4729 
4730 	/* INITs are latched while in SMM */
4731 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4732 	    (events->smi.smm || events->smi.pending) &&
4733 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4734 		return -EINVAL;
4735 
4736 	process_nmi(vcpu);
4737 	vcpu->arch.exception.injected = events->exception.injected;
4738 	vcpu->arch.exception.pending = events->exception.pending;
4739 	vcpu->arch.exception.nr = events->exception.nr;
4740 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4741 	vcpu->arch.exception.error_code = events->exception.error_code;
4742 	vcpu->arch.exception.has_payload = events->exception_has_payload;
4743 	vcpu->arch.exception.payload = events->exception_payload;
4744 
4745 	vcpu->arch.interrupt.injected = events->interrupt.injected;
4746 	vcpu->arch.interrupt.nr = events->interrupt.nr;
4747 	vcpu->arch.interrupt.soft = events->interrupt.soft;
4748 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4749 		static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4750 						events->interrupt.shadow);
4751 
4752 	vcpu->arch.nmi_injected = events->nmi.injected;
4753 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4754 		vcpu->arch.nmi_pending = events->nmi.pending;
4755 	static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4756 
4757 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4758 	    lapic_in_kernel(vcpu))
4759 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
4760 
4761 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4762 		if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm)
4763 			kvm_smm_changed(vcpu, events->smi.smm);
4764 
4765 		vcpu->arch.smi_pending = events->smi.pending;
4766 
4767 		if (events->smi.smm) {
4768 			if (events->smi.smm_inside_nmi)
4769 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4770 			else
4771 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4772 		}
4773 
4774 		if (lapic_in_kernel(vcpu)) {
4775 			if (events->smi.latched_init)
4776 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4777 			else
4778 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4779 		}
4780 	}
4781 
4782 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4783 
4784 	return 0;
4785 }
4786 
4787 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4788 					     struct kvm_debugregs *dbgregs)
4789 {
4790 	unsigned long val;
4791 
4792 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4793 	kvm_get_dr(vcpu, 6, &val);
4794 	dbgregs->dr6 = val;
4795 	dbgregs->dr7 = vcpu->arch.dr7;
4796 	dbgregs->flags = 0;
4797 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4798 }
4799 
4800 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4801 					    struct kvm_debugregs *dbgregs)
4802 {
4803 	if (dbgregs->flags)
4804 		return -EINVAL;
4805 
4806 	if (!kvm_dr6_valid(dbgregs->dr6))
4807 		return -EINVAL;
4808 	if (!kvm_dr7_valid(dbgregs->dr7))
4809 		return -EINVAL;
4810 
4811 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4812 	kvm_update_dr0123(vcpu);
4813 	vcpu->arch.dr6 = dbgregs->dr6;
4814 	vcpu->arch.dr7 = dbgregs->dr7;
4815 	kvm_update_dr7(vcpu);
4816 
4817 	return 0;
4818 }
4819 
4820 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4821 					 struct kvm_xsave *guest_xsave)
4822 {
4823 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
4824 		return;
4825 
4826 	fpu_copy_guest_fpstate_to_uabi(&vcpu->arch.guest_fpu,
4827 				       guest_xsave->region,
4828 				       sizeof(guest_xsave->region),
4829 				       vcpu->arch.pkru);
4830 }
4831 
4832 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4833 					struct kvm_xsave *guest_xsave)
4834 {
4835 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
4836 		return 0;
4837 
4838 	return fpu_copy_uabi_to_guest_fpstate(&vcpu->arch.guest_fpu,
4839 					      guest_xsave->region,
4840 					      supported_xcr0, &vcpu->arch.pkru);
4841 }
4842 
4843 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4844 					struct kvm_xcrs *guest_xcrs)
4845 {
4846 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4847 		guest_xcrs->nr_xcrs = 0;
4848 		return;
4849 	}
4850 
4851 	guest_xcrs->nr_xcrs = 1;
4852 	guest_xcrs->flags = 0;
4853 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4854 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4855 }
4856 
4857 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4858 				       struct kvm_xcrs *guest_xcrs)
4859 {
4860 	int i, r = 0;
4861 
4862 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
4863 		return -EINVAL;
4864 
4865 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4866 		return -EINVAL;
4867 
4868 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4869 		/* Only support XCR0 currently */
4870 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4871 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4872 				guest_xcrs->xcrs[i].value);
4873 			break;
4874 		}
4875 	if (r)
4876 		r = -EINVAL;
4877 	return r;
4878 }
4879 
4880 /*
4881  * kvm_set_guest_paused() indicates to the guest kernel that it has been
4882  * stopped by the hypervisor.  This function will be called from the host only.
4883  * EINVAL is returned when the host attempts to set the flag for a guest that
4884  * does not support pv clocks.
4885  */
4886 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4887 {
4888 	if (!vcpu->arch.pv_time_enabled)
4889 		return -EINVAL;
4890 	vcpu->arch.pvclock_set_guest_stopped_request = true;
4891 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4892 	return 0;
4893 }
4894 
4895 static int kvm_arch_tsc_has_attr(struct kvm_vcpu *vcpu,
4896 				 struct kvm_device_attr *attr)
4897 {
4898 	int r;
4899 
4900 	switch (attr->attr) {
4901 	case KVM_VCPU_TSC_OFFSET:
4902 		r = 0;
4903 		break;
4904 	default:
4905 		r = -ENXIO;
4906 	}
4907 
4908 	return r;
4909 }
4910 
4911 static int kvm_arch_tsc_get_attr(struct kvm_vcpu *vcpu,
4912 				 struct kvm_device_attr *attr)
4913 {
4914 	u64 __user *uaddr = (u64 __user *)(unsigned long)attr->addr;
4915 	int r;
4916 
4917 	if ((u64)(unsigned long)uaddr != attr->addr)
4918 		return -EFAULT;
4919 
4920 	switch (attr->attr) {
4921 	case KVM_VCPU_TSC_OFFSET:
4922 		r = -EFAULT;
4923 		if (put_user(vcpu->arch.l1_tsc_offset, uaddr))
4924 			break;
4925 		r = 0;
4926 		break;
4927 	default:
4928 		r = -ENXIO;
4929 	}
4930 
4931 	return r;
4932 }
4933 
4934 static int kvm_arch_tsc_set_attr(struct kvm_vcpu *vcpu,
4935 				 struct kvm_device_attr *attr)
4936 {
4937 	u64 __user *uaddr = (u64 __user *)(unsigned long)attr->addr;
4938 	struct kvm *kvm = vcpu->kvm;
4939 	int r;
4940 
4941 	if ((u64)(unsigned long)uaddr != attr->addr)
4942 		return -EFAULT;
4943 
4944 	switch (attr->attr) {
4945 	case KVM_VCPU_TSC_OFFSET: {
4946 		u64 offset, tsc, ns;
4947 		unsigned long flags;
4948 		bool matched;
4949 
4950 		r = -EFAULT;
4951 		if (get_user(offset, uaddr))
4952 			break;
4953 
4954 		raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
4955 
4956 		matched = (vcpu->arch.virtual_tsc_khz &&
4957 			   kvm->arch.last_tsc_khz == vcpu->arch.virtual_tsc_khz &&
4958 			   kvm->arch.last_tsc_offset == offset);
4959 
4960 		tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio) + offset;
4961 		ns = get_kvmclock_base_ns();
4962 
4963 		__kvm_synchronize_tsc(vcpu, offset, tsc, ns, matched);
4964 		raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
4965 
4966 		r = 0;
4967 		break;
4968 	}
4969 	default:
4970 		r = -ENXIO;
4971 	}
4972 
4973 	return r;
4974 }
4975 
4976 static int kvm_vcpu_ioctl_device_attr(struct kvm_vcpu *vcpu,
4977 				      unsigned int ioctl,
4978 				      void __user *argp)
4979 {
4980 	struct kvm_device_attr attr;
4981 	int r;
4982 
4983 	if (copy_from_user(&attr, argp, sizeof(attr)))
4984 		return -EFAULT;
4985 
4986 	if (attr.group != KVM_VCPU_TSC_CTRL)
4987 		return -ENXIO;
4988 
4989 	switch (ioctl) {
4990 	case KVM_HAS_DEVICE_ATTR:
4991 		r = kvm_arch_tsc_has_attr(vcpu, &attr);
4992 		break;
4993 	case KVM_GET_DEVICE_ATTR:
4994 		r = kvm_arch_tsc_get_attr(vcpu, &attr);
4995 		break;
4996 	case KVM_SET_DEVICE_ATTR:
4997 		r = kvm_arch_tsc_set_attr(vcpu, &attr);
4998 		break;
4999 	}
5000 
5001 	return r;
5002 }
5003 
5004 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
5005 				     struct kvm_enable_cap *cap)
5006 {
5007 	int r;
5008 	uint16_t vmcs_version;
5009 	void __user *user_ptr;
5010 
5011 	if (cap->flags)
5012 		return -EINVAL;
5013 
5014 	switch (cap->cap) {
5015 	case KVM_CAP_HYPERV_SYNIC2:
5016 		if (cap->args[0])
5017 			return -EINVAL;
5018 		fallthrough;
5019 
5020 	case KVM_CAP_HYPERV_SYNIC:
5021 		if (!irqchip_in_kernel(vcpu->kvm))
5022 			return -EINVAL;
5023 		return kvm_hv_activate_synic(vcpu, cap->cap ==
5024 					     KVM_CAP_HYPERV_SYNIC2);
5025 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
5026 		if (!kvm_x86_ops.nested_ops->enable_evmcs)
5027 			return -ENOTTY;
5028 		r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
5029 		if (!r) {
5030 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
5031 			if (copy_to_user(user_ptr, &vmcs_version,
5032 					 sizeof(vmcs_version)))
5033 				r = -EFAULT;
5034 		}
5035 		return r;
5036 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
5037 		if (!kvm_x86_ops.enable_direct_tlbflush)
5038 			return -ENOTTY;
5039 
5040 		return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
5041 
5042 	case KVM_CAP_HYPERV_ENFORCE_CPUID:
5043 		return kvm_hv_set_enforce_cpuid(vcpu, cap->args[0]);
5044 
5045 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
5046 		vcpu->arch.pv_cpuid.enforce = cap->args[0];
5047 		if (vcpu->arch.pv_cpuid.enforce)
5048 			kvm_update_pv_runtime(vcpu);
5049 
5050 		return 0;
5051 	default:
5052 		return -EINVAL;
5053 	}
5054 }
5055 
5056 long kvm_arch_vcpu_ioctl(struct file *filp,
5057 			 unsigned int ioctl, unsigned long arg)
5058 {
5059 	struct kvm_vcpu *vcpu = filp->private_data;
5060 	void __user *argp = (void __user *)arg;
5061 	int r;
5062 	union {
5063 		struct kvm_sregs2 *sregs2;
5064 		struct kvm_lapic_state *lapic;
5065 		struct kvm_xsave *xsave;
5066 		struct kvm_xcrs *xcrs;
5067 		void *buffer;
5068 	} u;
5069 
5070 	vcpu_load(vcpu);
5071 
5072 	u.buffer = NULL;
5073 	switch (ioctl) {
5074 	case KVM_GET_LAPIC: {
5075 		r = -EINVAL;
5076 		if (!lapic_in_kernel(vcpu))
5077 			goto out;
5078 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
5079 				GFP_KERNEL_ACCOUNT);
5080 
5081 		r = -ENOMEM;
5082 		if (!u.lapic)
5083 			goto out;
5084 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
5085 		if (r)
5086 			goto out;
5087 		r = -EFAULT;
5088 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
5089 			goto out;
5090 		r = 0;
5091 		break;
5092 	}
5093 	case KVM_SET_LAPIC: {
5094 		r = -EINVAL;
5095 		if (!lapic_in_kernel(vcpu))
5096 			goto out;
5097 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
5098 		if (IS_ERR(u.lapic)) {
5099 			r = PTR_ERR(u.lapic);
5100 			goto out_nofree;
5101 		}
5102 
5103 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
5104 		break;
5105 	}
5106 	case KVM_INTERRUPT: {
5107 		struct kvm_interrupt irq;
5108 
5109 		r = -EFAULT;
5110 		if (copy_from_user(&irq, argp, sizeof(irq)))
5111 			goto out;
5112 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
5113 		break;
5114 	}
5115 	case KVM_NMI: {
5116 		r = kvm_vcpu_ioctl_nmi(vcpu);
5117 		break;
5118 	}
5119 	case KVM_SMI: {
5120 		r = kvm_vcpu_ioctl_smi(vcpu);
5121 		break;
5122 	}
5123 	case KVM_SET_CPUID: {
5124 		struct kvm_cpuid __user *cpuid_arg = argp;
5125 		struct kvm_cpuid cpuid;
5126 
5127 		r = -EFAULT;
5128 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5129 			goto out;
5130 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
5131 		break;
5132 	}
5133 	case KVM_SET_CPUID2: {
5134 		struct kvm_cpuid2 __user *cpuid_arg = argp;
5135 		struct kvm_cpuid2 cpuid;
5136 
5137 		r = -EFAULT;
5138 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5139 			goto out;
5140 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
5141 					      cpuid_arg->entries);
5142 		break;
5143 	}
5144 	case KVM_GET_CPUID2: {
5145 		struct kvm_cpuid2 __user *cpuid_arg = argp;
5146 		struct kvm_cpuid2 cpuid;
5147 
5148 		r = -EFAULT;
5149 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5150 			goto out;
5151 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
5152 					      cpuid_arg->entries);
5153 		if (r)
5154 			goto out;
5155 		r = -EFAULT;
5156 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
5157 			goto out;
5158 		r = 0;
5159 		break;
5160 	}
5161 	case KVM_GET_MSRS: {
5162 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
5163 		r = msr_io(vcpu, argp, do_get_msr, 1);
5164 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5165 		break;
5166 	}
5167 	case KVM_SET_MSRS: {
5168 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
5169 		r = msr_io(vcpu, argp, do_set_msr, 0);
5170 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5171 		break;
5172 	}
5173 	case KVM_TPR_ACCESS_REPORTING: {
5174 		struct kvm_tpr_access_ctl tac;
5175 
5176 		r = -EFAULT;
5177 		if (copy_from_user(&tac, argp, sizeof(tac)))
5178 			goto out;
5179 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
5180 		if (r)
5181 			goto out;
5182 		r = -EFAULT;
5183 		if (copy_to_user(argp, &tac, sizeof(tac)))
5184 			goto out;
5185 		r = 0;
5186 		break;
5187 	};
5188 	case KVM_SET_VAPIC_ADDR: {
5189 		struct kvm_vapic_addr va;
5190 		int idx;
5191 
5192 		r = -EINVAL;
5193 		if (!lapic_in_kernel(vcpu))
5194 			goto out;
5195 		r = -EFAULT;
5196 		if (copy_from_user(&va, argp, sizeof(va)))
5197 			goto out;
5198 		idx = srcu_read_lock(&vcpu->kvm->srcu);
5199 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
5200 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5201 		break;
5202 	}
5203 	case KVM_X86_SETUP_MCE: {
5204 		u64 mcg_cap;
5205 
5206 		r = -EFAULT;
5207 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
5208 			goto out;
5209 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
5210 		break;
5211 	}
5212 	case KVM_X86_SET_MCE: {
5213 		struct kvm_x86_mce mce;
5214 
5215 		r = -EFAULT;
5216 		if (copy_from_user(&mce, argp, sizeof(mce)))
5217 			goto out;
5218 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
5219 		break;
5220 	}
5221 	case KVM_GET_VCPU_EVENTS: {
5222 		struct kvm_vcpu_events events;
5223 
5224 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
5225 
5226 		r = -EFAULT;
5227 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
5228 			break;
5229 		r = 0;
5230 		break;
5231 	}
5232 	case KVM_SET_VCPU_EVENTS: {
5233 		struct kvm_vcpu_events events;
5234 
5235 		r = -EFAULT;
5236 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5237 			break;
5238 
5239 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5240 		break;
5241 	}
5242 	case KVM_GET_DEBUGREGS: {
5243 		struct kvm_debugregs dbgregs;
5244 
5245 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5246 
5247 		r = -EFAULT;
5248 		if (copy_to_user(argp, &dbgregs,
5249 				 sizeof(struct kvm_debugregs)))
5250 			break;
5251 		r = 0;
5252 		break;
5253 	}
5254 	case KVM_SET_DEBUGREGS: {
5255 		struct kvm_debugregs dbgregs;
5256 
5257 		r = -EFAULT;
5258 		if (copy_from_user(&dbgregs, argp,
5259 				   sizeof(struct kvm_debugregs)))
5260 			break;
5261 
5262 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5263 		break;
5264 	}
5265 	case KVM_GET_XSAVE: {
5266 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5267 		r = -ENOMEM;
5268 		if (!u.xsave)
5269 			break;
5270 
5271 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5272 
5273 		r = -EFAULT;
5274 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5275 			break;
5276 		r = 0;
5277 		break;
5278 	}
5279 	case KVM_SET_XSAVE: {
5280 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
5281 		if (IS_ERR(u.xsave)) {
5282 			r = PTR_ERR(u.xsave);
5283 			goto out_nofree;
5284 		}
5285 
5286 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5287 		break;
5288 	}
5289 	case KVM_GET_XCRS: {
5290 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5291 		r = -ENOMEM;
5292 		if (!u.xcrs)
5293 			break;
5294 
5295 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5296 
5297 		r = -EFAULT;
5298 		if (copy_to_user(argp, u.xcrs,
5299 				 sizeof(struct kvm_xcrs)))
5300 			break;
5301 		r = 0;
5302 		break;
5303 	}
5304 	case KVM_SET_XCRS: {
5305 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5306 		if (IS_ERR(u.xcrs)) {
5307 			r = PTR_ERR(u.xcrs);
5308 			goto out_nofree;
5309 		}
5310 
5311 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5312 		break;
5313 	}
5314 	case KVM_SET_TSC_KHZ: {
5315 		u32 user_tsc_khz;
5316 
5317 		r = -EINVAL;
5318 		user_tsc_khz = (u32)arg;
5319 
5320 		if (kvm_has_tsc_control &&
5321 		    user_tsc_khz >= kvm_max_guest_tsc_khz)
5322 			goto out;
5323 
5324 		if (user_tsc_khz == 0)
5325 			user_tsc_khz = tsc_khz;
5326 
5327 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5328 			r = 0;
5329 
5330 		goto out;
5331 	}
5332 	case KVM_GET_TSC_KHZ: {
5333 		r = vcpu->arch.virtual_tsc_khz;
5334 		goto out;
5335 	}
5336 	case KVM_KVMCLOCK_CTRL: {
5337 		r = kvm_set_guest_paused(vcpu);
5338 		goto out;
5339 	}
5340 	case KVM_ENABLE_CAP: {
5341 		struct kvm_enable_cap cap;
5342 
5343 		r = -EFAULT;
5344 		if (copy_from_user(&cap, argp, sizeof(cap)))
5345 			goto out;
5346 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5347 		break;
5348 	}
5349 	case KVM_GET_NESTED_STATE: {
5350 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5351 		u32 user_data_size;
5352 
5353 		r = -EINVAL;
5354 		if (!kvm_x86_ops.nested_ops->get_state)
5355 			break;
5356 
5357 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5358 		r = -EFAULT;
5359 		if (get_user(user_data_size, &user_kvm_nested_state->size))
5360 			break;
5361 
5362 		r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5363 						     user_data_size);
5364 		if (r < 0)
5365 			break;
5366 
5367 		if (r > user_data_size) {
5368 			if (put_user(r, &user_kvm_nested_state->size))
5369 				r = -EFAULT;
5370 			else
5371 				r = -E2BIG;
5372 			break;
5373 		}
5374 
5375 		r = 0;
5376 		break;
5377 	}
5378 	case KVM_SET_NESTED_STATE: {
5379 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5380 		struct kvm_nested_state kvm_state;
5381 		int idx;
5382 
5383 		r = -EINVAL;
5384 		if (!kvm_x86_ops.nested_ops->set_state)
5385 			break;
5386 
5387 		r = -EFAULT;
5388 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5389 			break;
5390 
5391 		r = -EINVAL;
5392 		if (kvm_state.size < sizeof(kvm_state))
5393 			break;
5394 
5395 		if (kvm_state.flags &
5396 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5397 		      | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5398 		      | KVM_STATE_NESTED_GIF_SET))
5399 			break;
5400 
5401 		/* nested_run_pending implies guest_mode.  */
5402 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5403 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5404 			break;
5405 
5406 		idx = srcu_read_lock(&vcpu->kvm->srcu);
5407 		r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5408 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5409 		break;
5410 	}
5411 	case KVM_GET_SUPPORTED_HV_CPUID:
5412 		r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5413 		break;
5414 #ifdef CONFIG_KVM_XEN
5415 	case KVM_XEN_VCPU_GET_ATTR: {
5416 		struct kvm_xen_vcpu_attr xva;
5417 
5418 		r = -EFAULT;
5419 		if (copy_from_user(&xva, argp, sizeof(xva)))
5420 			goto out;
5421 		r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5422 		if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5423 			r = -EFAULT;
5424 		break;
5425 	}
5426 	case KVM_XEN_VCPU_SET_ATTR: {
5427 		struct kvm_xen_vcpu_attr xva;
5428 
5429 		r = -EFAULT;
5430 		if (copy_from_user(&xva, argp, sizeof(xva)))
5431 			goto out;
5432 		r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5433 		break;
5434 	}
5435 #endif
5436 	case KVM_GET_SREGS2: {
5437 		u.sregs2 = kzalloc(sizeof(struct kvm_sregs2), GFP_KERNEL);
5438 		r = -ENOMEM;
5439 		if (!u.sregs2)
5440 			goto out;
5441 		__get_sregs2(vcpu, u.sregs2);
5442 		r = -EFAULT;
5443 		if (copy_to_user(argp, u.sregs2, sizeof(struct kvm_sregs2)))
5444 			goto out;
5445 		r = 0;
5446 		break;
5447 	}
5448 	case KVM_SET_SREGS2: {
5449 		u.sregs2 = memdup_user(argp, sizeof(struct kvm_sregs2));
5450 		if (IS_ERR(u.sregs2)) {
5451 			r = PTR_ERR(u.sregs2);
5452 			u.sregs2 = NULL;
5453 			goto out;
5454 		}
5455 		r = __set_sregs2(vcpu, u.sregs2);
5456 		break;
5457 	}
5458 	case KVM_HAS_DEVICE_ATTR:
5459 	case KVM_GET_DEVICE_ATTR:
5460 	case KVM_SET_DEVICE_ATTR:
5461 		r = kvm_vcpu_ioctl_device_attr(vcpu, ioctl, argp);
5462 		break;
5463 	default:
5464 		r = -EINVAL;
5465 	}
5466 out:
5467 	kfree(u.buffer);
5468 out_nofree:
5469 	vcpu_put(vcpu);
5470 	return r;
5471 }
5472 
5473 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5474 {
5475 	return VM_FAULT_SIGBUS;
5476 }
5477 
5478 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5479 {
5480 	int ret;
5481 
5482 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
5483 		return -EINVAL;
5484 	ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5485 	return ret;
5486 }
5487 
5488 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5489 					      u64 ident_addr)
5490 {
5491 	return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5492 }
5493 
5494 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5495 					 unsigned long kvm_nr_mmu_pages)
5496 {
5497 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5498 		return -EINVAL;
5499 
5500 	mutex_lock(&kvm->slots_lock);
5501 
5502 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5503 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5504 
5505 	mutex_unlock(&kvm->slots_lock);
5506 	return 0;
5507 }
5508 
5509 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5510 {
5511 	return kvm->arch.n_max_mmu_pages;
5512 }
5513 
5514 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5515 {
5516 	struct kvm_pic *pic = kvm->arch.vpic;
5517 	int r;
5518 
5519 	r = 0;
5520 	switch (chip->chip_id) {
5521 	case KVM_IRQCHIP_PIC_MASTER:
5522 		memcpy(&chip->chip.pic, &pic->pics[0],
5523 			sizeof(struct kvm_pic_state));
5524 		break;
5525 	case KVM_IRQCHIP_PIC_SLAVE:
5526 		memcpy(&chip->chip.pic, &pic->pics[1],
5527 			sizeof(struct kvm_pic_state));
5528 		break;
5529 	case KVM_IRQCHIP_IOAPIC:
5530 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
5531 		break;
5532 	default:
5533 		r = -EINVAL;
5534 		break;
5535 	}
5536 	return r;
5537 }
5538 
5539 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5540 {
5541 	struct kvm_pic *pic = kvm->arch.vpic;
5542 	int r;
5543 
5544 	r = 0;
5545 	switch (chip->chip_id) {
5546 	case KVM_IRQCHIP_PIC_MASTER:
5547 		spin_lock(&pic->lock);
5548 		memcpy(&pic->pics[0], &chip->chip.pic,
5549 			sizeof(struct kvm_pic_state));
5550 		spin_unlock(&pic->lock);
5551 		break;
5552 	case KVM_IRQCHIP_PIC_SLAVE:
5553 		spin_lock(&pic->lock);
5554 		memcpy(&pic->pics[1], &chip->chip.pic,
5555 			sizeof(struct kvm_pic_state));
5556 		spin_unlock(&pic->lock);
5557 		break;
5558 	case KVM_IRQCHIP_IOAPIC:
5559 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
5560 		break;
5561 	default:
5562 		r = -EINVAL;
5563 		break;
5564 	}
5565 	kvm_pic_update_irq(pic);
5566 	return r;
5567 }
5568 
5569 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5570 {
5571 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5572 
5573 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5574 
5575 	mutex_lock(&kps->lock);
5576 	memcpy(ps, &kps->channels, sizeof(*ps));
5577 	mutex_unlock(&kps->lock);
5578 	return 0;
5579 }
5580 
5581 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5582 {
5583 	int i;
5584 	struct kvm_pit *pit = kvm->arch.vpit;
5585 
5586 	mutex_lock(&pit->pit_state.lock);
5587 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5588 	for (i = 0; i < 3; i++)
5589 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5590 	mutex_unlock(&pit->pit_state.lock);
5591 	return 0;
5592 }
5593 
5594 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5595 {
5596 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
5597 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5598 		sizeof(ps->channels));
5599 	ps->flags = kvm->arch.vpit->pit_state.flags;
5600 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5601 	memset(&ps->reserved, 0, sizeof(ps->reserved));
5602 	return 0;
5603 }
5604 
5605 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5606 {
5607 	int start = 0;
5608 	int i;
5609 	u32 prev_legacy, cur_legacy;
5610 	struct kvm_pit *pit = kvm->arch.vpit;
5611 
5612 	mutex_lock(&pit->pit_state.lock);
5613 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5614 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5615 	if (!prev_legacy && cur_legacy)
5616 		start = 1;
5617 	memcpy(&pit->pit_state.channels, &ps->channels,
5618 	       sizeof(pit->pit_state.channels));
5619 	pit->pit_state.flags = ps->flags;
5620 	for (i = 0; i < 3; i++)
5621 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5622 				   start && i == 0);
5623 	mutex_unlock(&pit->pit_state.lock);
5624 	return 0;
5625 }
5626 
5627 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5628 				 struct kvm_reinject_control *control)
5629 {
5630 	struct kvm_pit *pit = kvm->arch.vpit;
5631 
5632 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
5633 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5634 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
5635 	 */
5636 	mutex_lock(&pit->pit_state.lock);
5637 	kvm_pit_set_reinject(pit, control->pit_reinject);
5638 	mutex_unlock(&pit->pit_state.lock);
5639 
5640 	return 0;
5641 }
5642 
5643 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5644 {
5645 
5646 	/*
5647 	 * Flush all CPUs' dirty log buffers to the  dirty_bitmap.  Called
5648 	 * before reporting dirty_bitmap to userspace.  KVM flushes the buffers
5649 	 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5650 	 * VM-Exit.
5651 	 */
5652 	struct kvm_vcpu *vcpu;
5653 	int i;
5654 
5655 	kvm_for_each_vcpu(i, vcpu, kvm)
5656 		kvm_vcpu_kick(vcpu);
5657 }
5658 
5659 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5660 			bool line_status)
5661 {
5662 	if (!irqchip_in_kernel(kvm))
5663 		return -ENXIO;
5664 
5665 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5666 					irq_event->irq, irq_event->level,
5667 					line_status);
5668 	return 0;
5669 }
5670 
5671 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5672 			    struct kvm_enable_cap *cap)
5673 {
5674 	int r;
5675 
5676 	if (cap->flags)
5677 		return -EINVAL;
5678 
5679 	switch (cap->cap) {
5680 	case KVM_CAP_DISABLE_QUIRKS:
5681 		kvm->arch.disabled_quirks = cap->args[0];
5682 		r = 0;
5683 		break;
5684 	case KVM_CAP_SPLIT_IRQCHIP: {
5685 		mutex_lock(&kvm->lock);
5686 		r = -EINVAL;
5687 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5688 			goto split_irqchip_unlock;
5689 		r = -EEXIST;
5690 		if (irqchip_in_kernel(kvm))
5691 			goto split_irqchip_unlock;
5692 		if (kvm->created_vcpus)
5693 			goto split_irqchip_unlock;
5694 		r = kvm_setup_empty_irq_routing(kvm);
5695 		if (r)
5696 			goto split_irqchip_unlock;
5697 		/* Pairs with irqchip_in_kernel. */
5698 		smp_wmb();
5699 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5700 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5701 		r = 0;
5702 split_irqchip_unlock:
5703 		mutex_unlock(&kvm->lock);
5704 		break;
5705 	}
5706 	case KVM_CAP_X2APIC_API:
5707 		r = -EINVAL;
5708 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5709 			break;
5710 
5711 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5712 			kvm->arch.x2apic_format = true;
5713 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5714 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
5715 
5716 		r = 0;
5717 		break;
5718 	case KVM_CAP_X86_DISABLE_EXITS:
5719 		r = -EINVAL;
5720 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5721 			break;
5722 
5723 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5724 			kvm_can_mwait_in_guest())
5725 			kvm->arch.mwait_in_guest = true;
5726 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5727 			kvm->arch.hlt_in_guest = true;
5728 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5729 			kvm->arch.pause_in_guest = true;
5730 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5731 			kvm->arch.cstate_in_guest = true;
5732 		r = 0;
5733 		break;
5734 	case KVM_CAP_MSR_PLATFORM_INFO:
5735 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5736 		r = 0;
5737 		break;
5738 	case KVM_CAP_EXCEPTION_PAYLOAD:
5739 		kvm->arch.exception_payload_enabled = cap->args[0];
5740 		r = 0;
5741 		break;
5742 	case KVM_CAP_X86_USER_SPACE_MSR:
5743 		kvm->arch.user_space_msr_mask = cap->args[0];
5744 		r = 0;
5745 		break;
5746 	case KVM_CAP_X86_BUS_LOCK_EXIT:
5747 		r = -EINVAL;
5748 		if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5749 			break;
5750 
5751 		if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5752 		    (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5753 			break;
5754 
5755 		if (kvm_has_bus_lock_exit &&
5756 		    cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5757 			kvm->arch.bus_lock_detection_enabled = true;
5758 		r = 0;
5759 		break;
5760 #ifdef CONFIG_X86_SGX_KVM
5761 	case KVM_CAP_SGX_ATTRIBUTE: {
5762 		unsigned long allowed_attributes = 0;
5763 
5764 		r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5765 		if (r)
5766 			break;
5767 
5768 		/* KVM only supports the PROVISIONKEY privileged attribute. */
5769 		if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5770 		    !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5771 			kvm->arch.sgx_provisioning_allowed = true;
5772 		else
5773 			r = -EINVAL;
5774 		break;
5775 	}
5776 #endif
5777 	case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5778 		r = -EINVAL;
5779 		if (kvm_x86_ops.vm_copy_enc_context_from)
5780 			r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5781 		return r;
5782 	case KVM_CAP_VM_MOVE_ENC_CONTEXT_FROM:
5783 		r = -EINVAL;
5784 		if (kvm_x86_ops.vm_move_enc_context_from)
5785 			r = kvm_x86_ops.vm_move_enc_context_from(
5786 				kvm, cap->args[0]);
5787 		return r;
5788 	case KVM_CAP_EXIT_HYPERCALL:
5789 		if (cap->args[0] & ~KVM_EXIT_HYPERCALL_VALID_MASK) {
5790 			r = -EINVAL;
5791 			break;
5792 		}
5793 		kvm->arch.hypercall_exit_enabled = cap->args[0];
5794 		r = 0;
5795 		break;
5796 	case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
5797 		r = -EINVAL;
5798 		if (cap->args[0] & ~1)
5799 			break;
5800 		kvm->arch.exit_on_emulation_error = cap->args[0];
5801 		r = 0;
5802 		break;
5803 	default:
5804 		r = -EINVAL;
5805 		break;
5806 	}
5807 	return r;
5808 }
5809 
5810 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5811 {
5812 	struct kvm_x86_msr_filter *msr_filter;
5813 
5814 	msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5815 	if (!msr_filter)
5816 		return NULL;
5817 
5818 	msr_filter->default_allow = default_allow;
5819 	return msr_filter;
5820 }
5821 
5822 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5823 {
5824 	u32 i;
5825 
5826 	if (!msr_filter)
5827 		return;
5828 
5829 	for (i = 0; i < msr_filter->count; i++)
5830 		kfree(msr_filter->ranges[i].bitmap);
5831 
5832 	kfree(msr_filter);
5833 }
5834 
5835 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5836 			      struct kvm_msr_filter_range *user_range)
5837 {
5838 	unsigned long *bitmap = NULL;
5839 	size_t bitmap_size;
5840 
5841 	if (!user_range->nmsrs)
5842 		return 0;
5843 
5844 	if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
5845 		return -EINVAL;
5846 
5847 	if (!user_range->flags)
5848 		return -EINVAL;
5849 
5850 	bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5851 	if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5852 		return -EINVAL;
5853 
5854 	bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5855 	if (IS_ERR(bitmap))
5856 		return PTR_ERR(bitmap);
5857 
5858 	msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
5859 		.flags = user_range->flags,
5860 		.base = user_range->base,
5861 		.nmsrs = user_range->nmsrs,
5862 		.bitmap = bitmap,
5863 	};
5864 
5865 	msr_filter->count++;
5866 	return 0;
5867 }
5868 
5869 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5870 {
5871 	struct kvm_msr_filter __user *user_msr_filter = argp;
5872 	struct kvm_x86_msr_filter *new_filter, *old_filter;
5873 	struct kvm_msr_filter filter;
5874 	bool default_allow;
5875 	bool empty = true;
5876 	int r = 0;
5877 	u32 i;
5878 
5879 	if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5880 		return -EFAULT;
5881 
5882 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5883 		empty &= !filter.ranges[i].nmsrs;
5884 
5885 	default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5886 	if (empty && !default_allow)
5887 		return -EINVAL;
5888 
5889 	new_filter = kvm_alloc_msr_filter(default_allow);
5890 	if (!new_filter)
5891 		return -ENOMEM;
5892 
5893 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5894 		r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
5895 		if (r) {
5896 			kvm_free_msr_filter(new_filter);
5897 			return r;
5898 		}
5899 	}
5900 
5901 	mutex_lock(&kvm->lock);
5902 
5903 	/* The per-VM filter is protected by kvm->lock... */
5904 	old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
5905 
5906 	rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
5907 	synchronize_srcu(&kvm->srcu);
5908 
5909 	kvm_free_msr_filter(old_filter);
5910 
5911 	kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5912 	mutex_unlock(&kvm->lock);
5913 
5914 	return 0;
5915 }
5916 
5917 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
5918 static int kvm_arch_suspend_notifier(struct kvm *kvm)
5919 {
5920 	struct kvm_vcpu *vcpu;
5921 	int i, ret = 0;
5922 
5923 	mutex_lock(&kvm->lock);
5924 	kvm_for_each_vcpu(i, vcpu, kvm) {
5925 		if (!vcpu->arch.pv_time_enabled)
5926 			continue;
5927 
5928 		ret = kvm_set_guest_paused(vcpu);
5929 		if (ret) {
5930 			kvm_err("Failed to pause guest VCPU%d: %d\n",
5931 				vcpu->vcpu_id, ret);
5932 			break;
5933 		}
5934 	}
5935 	mutex_unlock(&kvm->lock);
5936 
5937 	return ret ? NOTIFY_BAD : NOTIFY_DONE;
5938 }
5939 
5940 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state)
5941 {
5942 	switch (state) {
5943 	case PM_HIBERNATION_PREPARE:
5944 	case PM_SUSPEND_PREPARE:
5945 		return kvm_arch_suspend_notifier(kvm);
5946 	}
5947 
5948 	return NOTIFY_DONE;
5949 }
5950 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
5951 
5952 static int kvm_vm_ioctl_get_clock(struct kvm *kvm, void __user *argp)
5953 {
5954 	struct kvm_clock_data data = { 0 };
5955 
5956 	get_kvmclock(kvm, &data);
5957 	if (copy_to_user(argp, &data, sizeof(data)))
5958 		return -EFAULT;
5959 
5960 	return 0;
5961 }
5962 
5963 static int kvm_vm_ioctl_set_clock(struct kvm *kvm, void __user *argp)
5964 {
5965 	struct kvm_arch *ka = &kvm->arch;
5966 	struct kvm_clock_data data;
5967 	u64 now_raw_ns;
5968 
5969 	if (copy_from_user(&data, argp, sizeof(data)))
5970 		return -EFAULT;
5971 
5972 	/*
5973 	 * Only KVM_CLOCK_REALTIME is used, but allow passing the
5974 	 * result of KVM_GET_CLOCK back to KVM_SET_CLOCK.
5975 	 */
5976 	if (data.flags & ~KVM_CLOCK_VALID_FLAGS)
5977 		return -EINVAL;
5978 
5979 	kvm_hv_invalidate_tsc_page(kvm);
5980 	kvm_start_pvclock_update(kvm);
5981 	pvclock_update_vm_gtod_copy(kvm);
5982 
5983 	/*
5984 	 * This pairs with kvm_guest_time_update(): when masterclock is
5985 	 * in use, we use master_kernel_ns + kvmclock_offset to set
5986 	 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
5987 	 * is slightly ahead) here we risk going negative on unsigned
5988 	 * 'system_time' when 'data.clock' is very small.
5989 	 */
5990 	if (data.flags & KVM_CLOCK_REALTIME) {
5991 		u64 now_real_ns = ktime_get_real_ns();
5992 
5993 		/*
5994 		 * Avoid stepping the kvmclock backwards.
5995 		 */
5996 		if (now_real_ns > data.realtime)
5997 			data.clock += now_real_ns - data.realtime;
5998 	}
5999 
6000 	if (ka->use_master_clock)
6001 		now_raw_ns = ka->master_kernel_ns;
6002 	else
6003 		now_raw_ns = get_kvmclock_base_ns();
6004 	ka->kvmclock_offset = data.clock - now_raw_ns;
6005 	kvm_end_pvclock_update(kvm);
6006 	return 0;
6007 }
6008 
6009 long kvm_arch_vm_ioctl(struct file *filp,
6010 		       unsigned int ioctl, unsigned long arg)
6011 {
6012 	struct kvm *kvm = filp->private_data;
6013 	void __user *argp = (void __user *)arg;
6014 	int r = -ENOTTY;
6015 	/*
6016 	 * This union makes it completely explicit to gcc-3.x
6017 	 * that these two variables' stack usage should be
6018 	 * combined, not added together.
6019 	 */
6020 	union {
6021 		struct kvm_pit_state ps;
6022 		struct kvm_pit_state2 ps2;
6023 		struct kvm_pit_config pit_config;
6024 	} u;
6025 
6026 	switch (ioctl) {
6027 	case KVM_SET_TSS_ADDR:
6028 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
6029 		break;
6030 	case KVM_SET_IDENTITY_MAP_ADDR: {
6031 		u64 ident_addr;
6032 
6033 		mutex_lock(&kvm->lock);
6034 		r = -EINVAL;
6035 		if (kvm->created_vcpus)
6036 			goto set_identity_unlock;
6037 		r = -EFAULT;
6038 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
6039 			goto set_identity_unlock;
6040 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
6041 set_identity_unlock:
6042 		mutex_unlock(&kvm->lock);
6043 		break;
6044 	}
6045 	case KVM_SET_NR_MMU_PAGES:
6046 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
6047 		break;
6048 	case KVM_GET_NR_MMU_PAGES:
6049 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
6050 		break;
6051 	case KVM_CREATE_IRQCHIP: {
6052 		mutex_lock(&kvm->lock);
6053 
6054 		r = -EEXIST;
6055 		if (irqchip_in_kernel(kvm))
6056 			goto create_irqchip_unlock;
6057 
6058 		r = -EINVAL;
6059 		if (kvm->created_vcpus)
6060 			goto create_irqchip_unlock;
6061 
6062 		r = kvm_pic_init(kvm);
6063 		if (r)
6064 			goto create_irqchip_unlock;
6065 
6066 		r = kvm_ioapic_init(kvm);
6067 		if (r) {
6068 			kvm_pic_destroy(kvm);
6069 			goto create_irqchip_unlock;
6070 		}
6071 
6072 		r = kvm_setup_default_irq_routing(kvm);
6073 		if (r) {
6074 			kvm_ioapic_destroy(kvm);
6075 			kvm_pic_destroy(kvm);
6076 			goto create_irqchip_unlock;
6077 		}
6078 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
6079 		smp_wmb();
6080 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
6081 	create_irqchip_unlock:
6082 		mutex_unlock(&kvm->lock);
6083 		break;
6084 	}
6085 	case KVM_CREATE_PIT:
6086 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
6087 		goto create_pit;
6088 	case KVM_CREATE_PIT2:
6089 		r = -EFAULT;
6090 		if (copy_from_user(&u.pit_config, argp,
6091 				   sizeof(struct kvm_pit_config)))
6092 			goto out;
6093 	create_pit:
6094 		mutex_lock(&kvm->lock);
6095 		r = -EEXIST;
6096 		if (kvm->arch.vpit)
6097 			goto create_pit_unlock;
6098 		r = -ENOMEM;
6099 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
6100 		if (kvm->arch.vpit)
6101 			r = 0;
6102 	create_pit_unlock:
6103 		mutex_unlock(&kvm->lock);
6104 		break;
6105 	case KVM_GET_IRQCHIP: {
6106 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6107 		struct kvm_irqchip *chip;
6108 
6109 		chip = memdup_user(argp, sizeof(*chip));
6110 		if (IS_ERR(chip)) {
6111 			r = PTR_ERR(chip);
6112 			goto out;
6113 		}
6114 
6115 		r = -ENXIO;
6116 		if (!irqchip_kernel(kvm))
6117 			goto get_irqchip_out;
6118 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
6119 		if (r)
6120 			goto get_irqchip_out;
6121 		r = -EFAULT;
6122 		if (copy_to_user(argp, chip, sizeof(*chip)))
6123 			goto get_irqchip_out;
6124 		r = 0;
6125 	get_irqchip_out:
6126 		kfree(chip);
6127 		break;
6128 	}
6129 	case KVM_SET_IRQCHIP: {
6130 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6131 		struct kvm_irqchip *chip;
6132 
6133 		chip = memdup_user(argp, sizeof(*chip));
6134 		if (IS_ERR(chip)) {
6135 			r = PTR_ERR(chip);
6136 			goto out;
6137 		}
6138 
6139 		r = -ENXIO;
6140 		if (!irqchip_kernel(kvm))
6141 			goto set_irqchip_out;
6142 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
6143 	set_irqchip_out:
6144 		kfree(chip);
6145 		break;
6146 	}
6147 	case KVM_GET_PIT: {
6148 		r = -EFAULT;
6149 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
6150 			goto out;
6151 		r = -ENXIO;
6152 		if (!kvm->arch.vpit)
6153 			goto out;
6154 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
6155 		if (r)
6156 			goto out;
6157 		r = -EFAULT;
6158 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
6159 			goto out;
6160 		r = 0;
6161 		break;
6162 	}
6163 	case KVM_SET_PIT: {
6164 		r = -EFAULT;
6165 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
6166 			goto out;
6167 		mutex_lock(&kvm->lock);
6168 		r = -ENXIO;
6169 		if (!kvm->arch.vpit)
6170 			goto set_pit_out;
6171 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
6172 set_pit_out:
6173 		mutex_unlock(&kvm->lock);
6174 		break;
6175 	}
6176 	case KVM_GET_PIT2: {
6177 		r = -ENXIO;
6178 		if (!kvm->arch.vpit)
6179 			goto out;
6180 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
6181 		if (r)
6182 			goto out;
6183 		r = -EFAULT;
6184 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
6185 			goto out;
6186 		r = 0;
6187 		break;
6188 	}
6189 	case KVM_SET_PIT2: {
6190 		r = -EFAULT;
6191 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
6192 			goto out;
6193 		mutex_lock(&kvm->lock);
6194 		r = -ENXIO;
6195 		if (!kvm->arch.vpit)
6196 			goto set_pit2_out;
6197 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
6198 set_pit2_out:
6199 		mutex_unlock(&kvm->lock);
6200 		break;
6201 	}
6202 	case KVM_REINJECT_CONTROL: {
6203 		struct kvm_reinject_control control;
6204 		r =  -EFAULT;
6205 		if (copy_from_user(&control, argp, sizeof(control)))
6206 			goto out;
6207 		r = -ENXIO;
6208 		if (!kvm->arch.vpit)
6209 			goto out;
6210 		r = kvm_vm_ioctl_reinject(kvm, &control);
6211 		break;
6212 	}
6213 	case KVM_SET_BOOT_CPU_ID:
6214 		r = 0;
6215 		mutex_lock(&kvm->lock);
6216 		if (kvm->created_vcpus)
6217 			r = -EBUSY;
6218 		else
6219 			kvm->arch.bsp_vcpu_id = arg;
6220 		mutex_unlock(&kvm->lock);
6221 		break;
6222 #ifdef CONFIG_KVM_XEN
6223 	case KVM_XEN_HVM_CONFIG: {
6224 		struct kvm_xen_hvm_config xhc;
6225 		r = -EFAULT;
6226 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
6227 			goto out;
6228 		r = kvm_xen_hvm_config(kvm, &xhc);
6229 		break;
6230 	}
6231 	case KVM_XEN_HVM_GET_ATTR: {
6232 		struct kvm_xen_hvm_attr xha;
6233 
6234 		r = -EFAULT;
6235 		if (copy_from_user(&xha, argp, sizeof(xha)))
6236 			goto out;
6237 		r = kvm_xen_hvm_get_attr(kvm, &xha);
6238 		if (!r && copy_to_user(argp, &xha, sizeof(xha)))
6239 			r = -EFAULT;
6240 		break;
6241 	}
6242 	case KVM_XEN_HVM_SET_ATTR: {
6243 		struct kvm_xen_hvm_attr xha;
6244 
6245 		r = -EFAULT;
6246 		if (copy_from_user(&xha, argp, sizeof(xha)))
6247 			goto out;
6248 		r = kvm_xen_hvm_set_attr(kvm, &xha);
6249 		break;
6250 	}
6251 #endif
6252 	case KVM_SET_CLOCK:
6253 		r = kvm_vm_ioctl_set_clock(kvm, argp);
6254 		break;
6255 	case KVM_GET_CLOCK:
6256 		r = kvm_vm_ioctl_get_clock(kvm, argp);
6257 		break;
6258 	case KVM_MEMORY_ENCRYPT_OP: {
6259 		r = -ENOTTY;
6260 		if (kvm_x86_ops.mem_enc_op)
6261 			r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
6262 		break;
6263 	}
6264 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
6265 		struct kvm_enc_region region;
6266 
6267 		r = -EFAULT;
6268 		if (copy_from_user(&region, argp, sizeof(region)))
6269 			goto out;
6270 
6271 		r = -ENOTTY;
6272 		if (kvm_x86_ops.mem_enc_reg_region)
6273 			r = static_call(kvm_x86_mem_enc_reg_region)(kvm, &region);
6274 		break;
6275 	}
6276 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
6277 		struct kvm_enc_region region;
6278 
6279 		r = -EFAULT;
6280 		if (copy_from_user(&region, argp, sizeof(region)))
6281 			goto out;
6282 
6283 		r = -ENOTTY;
6284 		if (kvm_x86_ops.mem_enc_unreg_region)
6285 			r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, &region);
6286 		break;
6287 	}
6288 	case KVM_HYPERV_EVENTFD: {
6289 		struct kvm_hyperv_eventfd hvevfd;
6290 
6291 		r = -EFAULT;
6292 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
6293 			goto out;
6294 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
6295 		break;
6296 	}
6297 	case KVM_SET_PMU_EVENT_FILTER:
6298 		r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
6299 		break;
6300 	case KVM_X86_SET_MSR_FILTER:
6301 		r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
6302 		break;
6303 	default:
6304 		r = -ENOTTY;
6305 	}
6306 out:
6307 	return r;
6308 }
6309 
6310 static void kvm_init_msr_list(void)
6311 {
6312 	struct x86_pmu_capability x86_pmu;
6313 	u32 dummy[2];
6314 	unsigned i;
6315 
6316 	BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
6317 			 "Please update the fixed PMCs in msrs_to_saved_all[]");
6318 
6319 	perf_get_x86_pmu_capability(&x86_pmu);
6320 
6321 	num_msrs_to_save = 0;
6322 	num_emulated_msrs = 0;
6323 	num_msr_based_features = 0;
6324 
6325 	for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
6326 		if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
6327 			continue;
6328 
6329 		/*
6330 		 * Even MSRs that are valid in the host may not be exposed
6331 		 * to the guests in some cases.
6332 		 */
6333 		switch (msrs_to_save_all[i]) {
6334 		case MSR_IA32_BNDCFGS:
6335 			if (!kvm_mpx_supported())
6336 				continue;
6337 			break;
6338 		case MSR_TSC_AUX:
6339 			if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
6340 			    !kvm_cpu_cap_has(X86_FEATURE_RDPID))
6341 				continue;
6342 			break;
6343 		case MSR_IA32_UMWAIT_CONTROL:
6344 			if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
6345 				continue;
6346 			break;
6347 		case MSR_IA32_RTIT_CTL:
6348 		case MSR_IA32_RTIT_STATUS:
6349 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
6350 				continue;
6351 			break;
6352 		case MSR_IA32_RTIT_CR3_MATCH:
6353 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6354 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
6355 				continue;
6356 			break;
6357 		case MSR_IA32_RTIT_OUTPUT_BASE:
6358 		case MSR_IA32_RTIT_OUTPUT_MASK:
6359 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6360 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
6361 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
6362 				continue;
6363 			break;
6364 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
6365 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6366 				msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
6367 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
6368 				continue;
6369 			break;
6370 		case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
6371 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
6372 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6373 				continue;
6374 			break;
6375 		case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
6376 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
6377 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6378 				continue;
6379 			break;
6380 		default:
6381 			break;
6382 		}
6383 
6384 		msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
6385 	}
6386 
6387 	for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
6388 		if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
6389 			continue;
6390 
6391 		emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
6392 	}
6393 
6394 	for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
6395 		struct kvm_msr_entry msr;
6396 
6397 		msr.index = msr_based_features_all[i];
6398 		if (kvm_get_msr_feature(&msr))
6399 			continue;
6400 
6401 		msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6402 	}
6403 }
6404 
6405 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6406 			   const void *v)
6407 {
6408 	int handled = 0;
6409 	int n;
6410 
6411 	do {
6412 		n = min(len, 8);
6413 		if (!(lapic_in_kernel(vcpu) &&
6414 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6415 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6416 			break;
6417 		handled += n;
6418 		addr += n;
6419 		len -= n;
6420 		v += n;
6421 	} while (len);
6422 
6423 	return handled;
6424 }
6425 
6426 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6427 {
6428 	int handled = 0;
6429 	int n;
6430 
6431 	do {
6432 		n = min(len, 8);
6433 		if (!(lapic_in_kernel(vcpu) &&
6434 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6435 					 addr, n, v))
6436 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6437 			break;
6438 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6439 		handled += n;
6440 		addr += n;
6441 		len -= n;
6442 		v += n;
6443 	} while (len);
6444 
6445 	return handled;
6446 }
6447 
6448 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6449 			struct kvm_segment *var, int seg)
6450 {
6451 	static_call(kvm_x86_set_segment)(vcpu, var, seg);
6452 }
6453 
6454 void kvm_get_segment(struct kvm_vcpu *vcpu,
6455 		     struct kvm_segment *var, int seg)
6456 {
6457 	static_call(kvm_x86_get_segment)(vcpu, var, seg);
6458 }
6459 
6460 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6461 			   struct x86_exception *exception)
6462 {
6463 	gpa_t t_gpa;
6464 
6465 	BUG_ON(!mmu_is_nested(vcpu));
6466 
6467 	/* NPT walks are always user-walks */
6468 	access |= PFERR_USER_MASK;
6469 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
6470 
6471 	return t_gpa;
6472 }
6473 
6474 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6475 			      struct x86_exception *exception)
6476 {
6477 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6478 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6479 }
6480 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6481 
6482  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6483 				struct x86_exception *exception)
6484 {
6485 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6486 	access |= PFERR_FETCH_MASK;
6487 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6488 }
6489 
6490 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6491 			       struct x86_exception *exception)
6492 {
6493 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6494 	access |= PFERR_WRITE_MASK;
6495 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6496 }
6497 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6498 
6499 /* uses this to access any guest's mapped memory without checking CPL */
6500 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6501 				struct x86_exception *exception)
6502 {
6503 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
6504 }
6505 
6506 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6507 				      struct kvm_vcpu *vcpu, u32 access,
6508 				      struct x86_exception *exception)
6509 {
6510 	void *data = val;
6511 	int r = X86EMUL_CONTINUE;
6512 
6513 	while (bytes) {
6514 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
6515 							    exception);
6516 		unsigned offset = addr & (PAGE_SIZE-1);
6517 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6518 		int ret;
6519 
6520 		if (gpa == UNMAPPED_GVA)
6521 			return X86EMUL_PROPAGATE_FAULT;
6522 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6523 					       offset, toread);
6524 		if (ret < 0) {
6525 			r = X86EMUL_IO_NEEDED;
6526 			goto out;
6527 		}
6528 
6529 		bytes -= toread;
6530 		data += toread;
6531 		addr += toread;
6532 	}
6533 out:
6534 	return r;
6535 }
6536 
6537 /* used for instruction fetching */
6538 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6539 				gva_t addr, void *val, unsigned int bytes,
6540 				struct x86_exception *exception)
6541 {
6542 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6543 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6544 	unsigned offset;
6545 	int ret;
6546 
6547 	/* Inline kvm_read_guest_virt_helper for speed.  */
6548 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6549 						    exception);
6550 	if (unlikely(gpa == UNMAPPED_GVA))
6551 		return X86EMUL_PROPAGATE_FAULT;
6552 
6553 	offset = addr & (PAGE_SIZE-1);
6554 	if (WARN_ON(offset + bytes > PAGE_SIZE))
6555 		bytes = (unsigned)PAGE_SIZE - offset;
6556 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6557 				       offset, bytes);
6558 	if (unlikely(ret < 0))
6559 		return X86EMUL_IO_NEEDED;
6560 
6561 	return X86EMUL_CONTINUE;
6562 }
6563 
6564 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6565 			       gva_t addr, void *val, unsigned int bytes,
6566 			       struct x86_exception *exception)
6567 {
6568 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6569 
6570 	/*
6571 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6572 	 * is returned, but our callers are not ready for that and they blindly
6573 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
6574 	 * uninitialized kernel stack memory into cr2 and error code.
6575 	 */
6576 	memset(exception, 0, sizeof(*exception));
6577 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6578 					  exception);
6579 }
6580 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6581 
6582 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6583 			     gva_t addr, void *val, unsigned int bytes,
6584 			     struct x86_exception *exception, bool system)
6585 {
6586 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6587 	u32 access = 0;
6588 
6589 	if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6590 		access |= PFERR_USER_MASK;
6591 
6592 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6593 }
6594 
6595 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6596 		unsigned long addr, void *val, unsigned int bytes)
6597 {
6598 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6599 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6600 
6601 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6602 }
6603 
6604 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6605 				      struct kvm_vcpu *vcpu, u32 access,
6606 				      struct x86_exception *exception)
6607 {
6608 	void *data = val;
6609 	int r = X86EMUL_CONTINUE;
6610 
6611 	while (bytes) {
6612 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6613 							     access,
6614 							     exception);
6615 		unsigned offset = addr & (PAGE_SIZE-1);
6616 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6617 		int ret;
6618 
6619 		if (gpa == UNMAPPED_GVA)
6620 			return X86EMUL_PROPAGATE_FAULT;
6621 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6622 		if (ret < 0) {
6623 			r = X86EMUL_IO_NEEDED;
6624 			goto out;
6625 		}
6626 
6627 		bytes -= towrite;
6628 		data += towrite;
6629 		addr += towrite;
6630 	}
6631 out:
6632 	return r;
6633 }
6634 
6635 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6636 			      unsigned int bytes, struct x86_exception *exception,
6637 			      bool system)
6638 {
6639 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6640 	u32 access = PFERR_WRITE_MASK;
6641 
6642 	if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6643 		access |= PFERR_USER_MASK;
6644 
6645 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6646 					   access, exception);
6647 }
6648 
6649 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6650 				unsigned int bytes, struct x86_exception *exception)
6651 {
6652 	/* kvm_write_guest_virt_system can pull in tons of pages. */
6653 	vcpu->arch.l1tf_flush_l1d = true;
6654 
6655 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6656 					   PFERR_WRITE_MASK, exception);
6657 }
6658 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6659 
6660 int handle_ud(struct kvm_vcpu *vcpu)
6661 {
6662 	static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6663 	int emul_type = EMULTYPE_TRAP_UD;
6664 	char sig[5]; /* ud2; .ascii "kvm" */
6665 	struct x86_exception e;
6666 
6667 	if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6668 		return 1;
6669 
6670 	if (force_emulation_prefix &&
6671 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6672 				sig, sizeof(sig), &e) == 0 &&
6673 	    memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6674 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6675 		emul_type = EMULTYPE_TRAP_UD_FORCED;
6676 	}
6677 
6678 	return kvm_emulate_instruction(vcpu, emul_type);
6679 }
6680 EXPORT_SYMBOL_GPL(handle_ud);
6681 
6682 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6683 			    gpa_t gpa, bool write)
6684 {
6685 	/* For APIC access vmexit */
6686 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6687 		return 1;
6688 
6689 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6690 		trace_vcpu_match_mmio(gva, gpa, write, true);
6691 		return 1;
6692 	}
6693 
6694 	return 0;
6695 }
6696 
6697 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6698 				gpa_t *gpa, struct x86_exception *exception,
6699 				bool write)
6700 {
6701 	u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6702 		| (write ? PFERR_WRITE_MASK : 0);
6703 
6704 	/*
6705 	 * currently PKRU is only applied to ept enabled guest so
6706 	 * there is no pkey in EPT page table for L1 guest or EPT
6707 	 * shadow page table for L2 guest.
6708 	 */
6709 	if (vcpu_match_mmio_gva(vcpu, gva) && (!is_paging(vcpu) ||
6710 	    !permission_fault(vcpu, vcpu->arch.walk_mmu,
6711 			      vcpu->arch.mmio_access, 0, access))) {
6712 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6713 					(gva & (PAGE_SIZE - 1));
6714 		trace_vcpu_match_mmio(gva, *gpa, write, false);
6715 		return 1;
6716 	}
6717 
6718 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6719 
6720 	if (*gpa == UNMAPPED_GVA)
6721 		return -1;
6722 
6723 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6724 }
6725 
6726 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6727 			const void *val, int bytes)
6728 {
6729 	int ret;
6730 
6731 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6732 	if (ret < 0)
6733 		return 0;
6734 	kvm_page_track_write(vcpu, gpa, val, bytes);
6735 	return 1;
6736 }
6737 
6738 struct read_write_emulator_ops {
6739 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6740 				  int bytes);
6741 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6742 				  void *val, int bytes);
6743 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6744 			       int bytes, void *val);
6745 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6746 				    void *val, int bytes);
6747 	bool write;
6748 };
6749 
6750 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6751 {
6752 	if (vcpu->mmio_read_completed) {
6753 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6754 			       vcpu->mmio_fragments[0].gpa, val);
6755 		vcpu->mmio_read_completed = 0;
6756 		return 1;
6757 	}
6758 
6759 	return 0;
6760 }
6761 
6762 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6763 			void *val, int bytes)
6764 {
6765 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6766 }
6767 
6768 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6769 			 void *val, int bytes)
6770 {
6771 	return emulator_write_phys(vcpu, gpa, val, bytes);
6772 }
6773 
6774 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6775 {
6776 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6777 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
6778 }
6779 
6780 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6781 			  void *val, int bytes)
6782 {
6783 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6784 	return X86EMUL_IO_NEEDED;
6785 }
6786 
6787 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6788 			   void *val, int bytes)
6789 {
6790 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6791 
6792 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6793 	return X86EMUL_CONTINUE;
6794 }
6795 
6796 static const struct read_write_emulator_ops read_emultor = {
6797 	.read_write_prepare = read_prepare,
6798 	.read_write_emulate = read_emulate,
6799 	.read_write_mmio = vcpu_mmio_read,
6800 	.read_write_exit_mmio = read_exit_mmio,
6801 };
6802 
6803 static const struct read_write_emulator_ops write_emultor = {
6804 	.read_write_emulate = write_emulate,
6805 	.read_write_mmio = write_mmio,
6806 	.read_write_exit_mmio = write_exit_mmio,
6807 	.write = true,
6808 };
6809 
6810 static int emulator_read_write_onepage(unsigned long addr, void *val,
6811 				       unsigned int bytes,
6812 				       struct x86_exception *exception,
6813 				       struct kvm_vcpu *vcpu,
6814 				       const struct read_write_emulator_ops *ops)
6815 {
6816 	gpa_t gpa;
6817 	int handled, ret;
6818 	bool write = ops->write;
6819 	struct kvm_mmio_fragment *frag;
6820 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6821 
6822 	/*
6823 	 * If the exit was due to a NPF we may already have a GPA.
6824 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6825 	 * Note, this cannot be used on string operations since string
6826 	 * operation using rep will only have the initial GPA from the NPF
6827 	 * occurred.
6828 	 */
6829 	if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6830 	    (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6831 		gpa = ctxt->gpa_val;
6832 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6833 	} else {
6834 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6835 		if (ret < 0)
6836 			return X86EMUL_PROPAGATE_FAULT;
6837 	}
6838 
6839 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6840 		return X86EMUL_CONTINUE;
6841 
6842 	/*
6843 	 * Is this MMIO handled locally?
6844 	 */
6845 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6846 	if (handled == bytes)
6847 		return X86EMUL_CONTINUE;
6848 
6849 	gpa += handled;
6850 	bytes -= handled;
6851 	val += handled;
6852 
6853 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6854 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6855 	frag->gpa = gpa;
6856 	frag->data = val;
6857 	frag->len = bytes;
6858 	return X86EMUL_CONTINUE;
6859 }
6860 
6861 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6862 			unsigned long addr,
6863 			void *val, unsigned int bytes,
6864 			struct x86_exception *exception,
6865 			const struct read_write_emulator_ops *ops)
6866 {
6867 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6868 	gpa_t gpa;
6869 	int rc;
6870 
6871 	if (ops->read_write_prepare &&
6872 		  ops->read_write_prepare(vcpu, val, bytes))
6873 		return X86EMUL_CONTINUE;
6874 
6875 	vcpu->mmio_nr_fragments = 0;
6876 
6877 	/* Crossing a page boundary? */
6878 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6879 		int now;
6880 
6881 		now = -addr & ~PAGE_MASK;
6882 		rc = emulator_read_write_onepage(addr, val, now, exception,
6883 						 vcpu, ops);
6884 
6885 		if (rc != X86EMUL_CONTINUE)
6886 			return rc;
6887 		addr += now;
6888 		if (ctxt->mode != X86EMUL_MODE_PROT64)
6889 			addr = (u32)addr;
6890 		val += now;
6891 		bytes -= now;
6892 	}
6893 
6894 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
6895 					 vcpu, ops);
6896 	if (rc != X86EMUL_CONTINUE)
6897 		return rc;
6898 
6899 	if (!vcpu->mmio_nr_fragments)
6900 		return rc;
6901 
6902 	gpa = vcpu->mmio_fragments[0].gpa;
6903 
6904 	vcpu->mmio_needed = 1;
6905 	vcpu->mmio_cur_fragment = 0;
6906 
6907 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6908 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6909 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
6910 	vcpu->run->mmio.phys_addr = gpa;
6911 
6912 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6913 }
6914 
6915 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6916 				  unsigned long addr,
6917 				  void *val,
6918 				  unsigned int bytes,
6919 				  struct x86_exception *exception)
6920 {
6921 	return emulator_read_write(ctxt, addr, val, bytes,
6922 				   exception, &read_emultor);
6923 }
6924 
6925 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6926 			    unsigned long addr,
6927 			    const void *val,
6928 			    unsigned int bytes,
6929 			    struct x86_exception *exception)
6930 {
6931 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
6932 				   exception, &write_emultor);
6933 }
6934 
6935 #define CMPXCHG_TYPE(t, ptr, old, new) \
6936 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6937 
6938 #ifdef CONFIG_X86_64
6939 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6940 #else
6941 #  define CMPXCHG64(ptr, old, new) \
6942 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6943 #endif
6944 
6945 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6946 				     unsigned long addr,
6947 				     const void *old,
6948 				     const void *new,
6949 				     unsigned int bytes,
6950 				     struct x86_exception *exception)
6951 {
6952 	struct kvm_host_map map;
6953 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6954 	u64 page_line_mask;
6955 	gpa_t gpa;
6956 	char *kaddr;
6957 	bool exchanged;
6958 
6959 	/* guests cmpxchg8b have to be emulated atomically */
6960 	if (bytes > 8 || (bytes & (bytes - 1)))
6961 		goto emul_write;
6962 
6963 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6964 
6965 	if (gpa == UNMAPPED_GVA ||
6966 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6967 		goto emul_write;
6968 
6969 	/*
6970 	 * Emulate the atomic as a straight write to avoid #AC if SLD is
6971 	 * enabled in the host and the access splits a cache line.
6972 	 */
6973 	if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6974 		page_line_mask = ~(cache_line_size() - 1);
6975 	else
6976 		page_line_mask = PAGE_MASK;
6977 
6978 	if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6979 		goto emul_write;
6980 
6981 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6982 		goto emul_write;
6983 
6984 	kaddr = map.hva + offset_in_page(gpa);
6985 
6986 	switch (bytes) {
6987 	case 1:
6988 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6989 		break;
6990 	case 2:
6991 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6992 		break;
6993 	case 4:
6994 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6995 		break;
6996 	case 8:
6997 		exchanged = CMPXCHG64(kaddr, old, new);
6998 		break;
6999 	default:
7000 		BUG();
7001 	}
7002 
7003 	kvm_vcpu_unmap(vcpu, &map, true);
7004 
7005 	if (!exchanged)
7006 		return X86EMUL_CMPXCHG_FAILED;
7007 
7008 	kvm_page_track_write(vcpu, gpa, new, bytes);
7009 
7010 	return X86EMUL_CONTINUE;
7011 
7012 emul_write:
7013 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
7014 
7015 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
7016 }
7017 
7018 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
7019 {
7020 	int r = 0, i;
7021 
7022 	for (i = 0; i < vcpu->arch.pio.count; i++) {
7023 		if (vcpu->arch.pio.in)
7024 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
7025 					    vcpu->arch.pio.size, pd);
7026 		else
7027 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
7028 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
7029 					     pd);
7030 		if (r)
7031 			break;
7032 		pd += vcpu->arch.pio.size;
7033 	}
7034 	return r;
7035 }
7036 
7037 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
7038 			       unsigned short port,
7039 			       unsigned int count, bool in)
7040 {
7041 	vcpu->arch.pio.port = port;
7042 	vcpu->arch.pio.in = in;
7043 	vcpu->arch.pio.count  = count;
7044 	vcpu->arch.pio.size = size;
7045 
7046 	if (!kernel_pio(vcpu, vcpu->arch.pio_data))
7047 		return 1;
7048 
7049 	vcpu->run->exit_reason = KVM_EXIT_IO;
7050 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
7051 	vcpu->run->io.size = size;
7052 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
7053 	vcpu->run->io.count = count;
7054 	vcpu->run->io.port = port;
7055 
7056 	return 0;
7057 }
7058 
7059 static int __emulator_pio_in(struct kvm_vcpu *vcpu, int size,
7060 			     unsigned short port, unsigned int count)
7061 {
7062 	WARN_ON(vcpu->arch.pio.count);
7063 	memset(vcpu->arch.pio_data, 0, size * count);
7064 	return emulator_pio_in_out(vcpu, size, port, count, true);
7065 }
7066 
7067 static void complete_emulator_pio_in(struct kvm_vcpu *vcpu, void *val)
7068 {
7069 	int size = vcpu->arch.pio.size;
7070 	unsigned count = vcpu->arch.pio.count;
7071 	memcpy(val, vcpu->arch.pio_data, size * count);
7072 	trace_kvm_pio(KVM_PIO_IN, vcpu->arch.pio.port, size, count, vcpu->arch.pio_data);
7073 	vcpu->arch.pio.count = 0;
7074 }
7075 
7076 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
7077 			   unsigned short port, void *val, unsigned int count)
7078 {
7079 	if (vcpu->arch.pio.count) {
7080 		/* Complete previous iteration.  */
7081 	} else {
7082 		int r = __emulator_pio_in(vcpu, size, port, count);
7083 		if (!r)
7084 			return r;
7085 
7086 		/* Results already available, fall through.  */
7087 	}
7088 
7089 	WARN_ON(count != vcpu->arch.pio.count);
7090 	complete_emulator_pio_in(vcpu, val);
7091 	return 1;
7092 }
7093 
7094 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
7095 				    int size, unsigned short port, void *val,
7096 				    unsigned int count)
7097 {
7098 	return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
7099 
7100 }
7101 
7102 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
7103 			    unsigned short port, const void *val,
7104 			    unsigned int count)
7105 {
7106 	int ret;
7107 
7108 	memcpy(vcpu->arch.pio_data, val, size * count);
7109 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
7110 	ret = emulator_pio_in_out(vcpu, size, port, count, false);
7111 	if (ret)
7112                 vcpu->arch.pio.count = 0;
7113 
7114         return ret;
7115 }
7116 
7117 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
7118 				     int size, unsigned short port,
7119 				     const void *val, unsigned int count)
7120 {
7121 	return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
7122 }
7123 
7124 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
7125 {
7126 	return static_call(kvm_x86_get_segment_base)(vcpu, seg);
7127 }
7128 
7129 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
7130 {
7131 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
7132 }
7133 
7134 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
7135 {
7136 	if (!need_emulate_wbinvd(vcpu))
7137 		return X86EMUL_CONTINUE;
7138 
7139 	if (static_call(kvm_x86_has_wbinvd_exit)()) {
7140 		int cpu = get_cpu();
7141 
7142 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
7143 		on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
7144 				wbinvd_ipi, NULL, 1);
7145 		put_cpu();
7146 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
7147 	} else
7148 		wbinvd();
7149 	return X86EMUL_CONTINUE;
7150 }
7151 
7152 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
7153 {
7154 	kvm_emulate_wbinvd_noskip(vcpu);
7155 	return kvm_skip_emulated_instruction(vcpu);
7156 }
7157 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
7158 
7159 
7160 
7161 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
7162 {
7163 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
7164 }
7165 
7166 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
7167 			    unsigned long *dest)
7168 {
7169 	kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
7170 }
7171 
7172 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
7173 			   unsigned long value)
7174 {
7175 
7176 	return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
7177 }
7178 
7179 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
7180 {
7181 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
7182 }
7183 
7184 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
7185 {
7186 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7187 	unsigned long value;
7188 
7189 	switch (cr) {
7190 	case 0:
7191 		value = kvm_read_cr0(vcpu);
7192 		break;
7193 	case 2:
7194 		value = vcpu->arch.cr2;
7195 		break;
7196 	case 3:
7197 		value = kvm_read_cr3(vcpu);
7198 		break;
7199 	case 4:
7200 		value = kvm_read_cr4(vcpu);
7201 		break;
7202 	case 8:
7203 		value = kvm_get_cr8(vcpu);
7204 		break;
7205 	default:
7206 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
7207 		return 0;
7208 	}
7209 
7210 	return value;
7211 }
7212 
7213 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
7214 {
7215 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7216 	int res = 0;
7217 
7218 	switch (cr) {
7219 	case 0:
7220 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
7221 		break;
7222 	case 2:
7223 		vcpu->arch.cr2 = val;
7224 		break;
7225 	case 3:
7226 		res = kvm_set_cr3(vcpu, val);
7227 		break;
7228 	case 4:
7229 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
7230 		break;
7231 	case 8:
7232 		res = kvm_set_cr8(vcpu, val);
7233 		break;
7234 	default:
7235 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
7236 		res = -1;
7237 	}
7238 
7239 	return res;
7240 }
7241 
7242 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
7243 {
7244 	return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
7245 }
7246 
7247 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7248 {
7249 	static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
7250 }
7251 
7252 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7253 {
7254 	static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
7255 }
7256 
7257 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7258 {
7259 	static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
7260 }
7261 
7262 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7263 {
7264 	static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
7265 }
7266 
7267 static unsigned long emulator_get_cached_segment_base(
7268 	struct x86_emulate_ctxt *ctxt, int seg)
7269 {
7270 	return get_segment_base(emul_to_vcpu(ctxt), seg);
7271 }
7272 
7273 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
7274 				 struct desc_struct *desc, u32 *base3,
7275 				 int seg)
7276 {
7277 	struct kvm_segment var;
7278 
7279 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
7280 	*selector = var.selector;
7281 
7282 	if (var.unusable) {
7283 		memset(desc, 0, sizeof(*desc));
7284 		if (base3)
7285 			*base3 = 0;
7286 		return false;
7287 	}
7288 
7289 	if (var.g)
7290 		var.limit >>= 12;
7291 	set_desc_limit(desc, var.limit);
7292 	set_desc_base(desc, (unsigned long)var.base);
7293 #ifdef CONFIG_X86_64
7294 	if (base3)
7295 		*base3 = var.base >> 32;
7296 #endif
7297 	desc->type = var.type;
7298 	desc->s = var.s;
7299 	desc->dpl = var.dpl;
7300 	desc->p = var.present;
7301 	desc->avl = var.avl;
7302 	desc->l = var.l;
7303 	desc->d = var.db;
7304 	desc->g = var.g;
7305 
7306 	return true;
7307 }
7308 
7309 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
7310 				 struct desc_struct *desc, u32 base3,
7311 				 int seg)
7312 {
7313 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7314 	struct kvm_segment var;
7315 
7316 	var.selector = selector;
7317 	var.base = get_desc_base(desc);
7318 #ifdef CONFIG_X86_64
7319 	var.base |= ((u64)base3) << 32;
7320 #endif
7321 	var.limit = get_desc_limit(desc);
7322 	if (desc->g)
7323 		var.limit = (var.limit << 12) | 0xfff;
7324 	var.type = desc->type;
7325 	var.dpl = desc->dpl;
7326 	var.db = desc->d;
7327 	var.s = desc->s;
7328 	var.l = desc->l;
7329 	var.g = desc->g;
7330 	var.avl = desc->avl;
7331 	var.present = desc->p;
7332 	var.unusable = !var.present;
7333 	var.padding = 0;
7334 
7335 	kvm_set_segment(vcpu, &var, seg);
7336 	return;
7337 }
7338 
7339 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
7340 			    u32 msr_index, u64 *pdata)
7341 {
7342 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7343 	int r;
7344 
7345 	r = kvm_get_msr(vcpu, msr_index, pdata);
7346 
7347 	if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
7348 		/* Bounce to user space */
7349 		return X86EMUL_IO_NEEDED;
7350 	}
7351 
7352 	return r;
7353 }
7354 
7355 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
7356 			    u32 msr_index, u64 data)
7357 {
7358 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7359 	int r;
7360 
7361 	r = kvm_set_msr(vcpu, msr_index, data);
7362 
7363 	if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
7364 		/* Bounce to user space */
7365 		return X86EMUL_IO_NEEDED;
7366 	}
7367 
7368 	return r;
7369 }
7370 
7371 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
7372 {
7373 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7374 
7375 	return vcpu->arch.smbase;
7376 }
7377 
7378 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
7379 {
7380 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7381 
7382 	vcpu->arch.smbase = smbase;
7383 }
7384 
7385 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
7386 			      u32 pmc)
7387 {
7388 	if (kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc))
7389 		return 0;
7390 	return -EINVAL;
7391 }
7392 
7393 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
7394 			     u32 pmc, u64 *pdata)
7395 {
7396 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
7397 }
7398 
7399 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
7400 {
7401 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
7402 }
7403 
7404 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
7405 			      struct x86_instruction_info *info,
7406 			      enum x86_intercept_stage stage)
7407 {
7408 	return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
7409 					    &ctxt->exception);
7410 }
7411 
7412 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
7413 			      u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
7414 			      bool exact_only)
7415 {
7416 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7417 }
7418 
7419 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7420 {
7421 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7422 }
7423 
7424 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7425 {
7426 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7427 }
7428 
7429 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7430 {
7431 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7432 }
7433 
7434 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7435 {
7436 	return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7437 }
7438 
7439 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7440 {
7441 	kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7442 }
7443 
7444 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7445 {
7446 	static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7447 }
7448 
7449 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7450 {
7451 	return emul_to_vcpu(ctxt)->arch.hflags;
7452 }
7453 
7454 static void emulator_exiting_smm(struct x86_emulate_ctxt *ctxt)
7455 {
7456 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7457 
7458 	kvm_smm_changed(vcpu, false);
7459 }
7460 
7461 static int emulator_leave_smm(struct x86_emulate_ctxt *ctxt,
7462 				  const char *smstate)
7463 {
7464 	return static_call(kvm_x86_leave_smm)(emul_to_vcpu(ctxt), smstate);
7465 }
7466 
7467 static void emulator_triple_fault(struct x86_emulate_ctxt *ctxt)
7468 {
7469 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, emul_to_vcpu(ctxt));
7470 }
7471 
7472 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7473 {
7474 	return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7475 }
7476 
7477 static const struct x86_emulate_ops emulate_ops = {
7478 	.read_gpr            = emulator_read_gpr,
7479 	.write_gpr           = emulator_write_gpr,
7480 	.read_std            = emulator_read_std,
7481 	.write_std           = emulator_write_std,
7482 	.read_phys           = kvm_read_guest_phys_system,
7483 	.fetch               = kvm_fetch_guest_virt,
7484 	.read_emulated       = emulator_read_emulated,
7485 	.write_emulated      = emulator_write_emulated,
7486 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
7487 	.invlpg              = emulator_invlpg,
7488 	.pio_in_emulated     = emulator_pio_in_emulated,
7489 	.pio_out_emulated    = emulator_pio_out_emulated,
7490 	.get_segment         = emulator_get_segment,
7491 	.set_segment         = emulator_set_segment,
7492 	.get_cached_segment_base = emulator_get_cached_segment_base,
7493 	.get_gdt             = emulator_get_gdt,
7494 	.get_idt	     = emulator_get_idt,
7495 	.set_gdt             = emulator_set_gdt,
7496 	.set_idt	     = emulator_set_idt,
7497 	.get_cr              = emulator_get_cr,
7498 	.set_cr              = emulator_set_cr,
7499 	.cpl                 = emulator_get_cpl,
7500 	.get_dr              = emulator_get_dr,
7501 	.set_dr              = emulator_set_dr,
7502 	.get_smbase          = emulator_get_smbase,
7503 	.set_smbase          = emulator_set_smbase,
7504 	.set_msr             = emulator_set_msr,
7505 	.get_msr             = emulator_get_msr,
7506 	.check_pmc	     = emulator_check_pmc,
7507 	.read_pmc            = emulator_read_pmc,
7508 	.halt                = emulator_halt,
7509 	.wbinvd              = emulator_wbinvd,
7510 	.fix_hypercall       = emulator_fix_hypercall,
7511 	.intercept           = emulator_intercept,
7512 	.get_cpuid           = emulator_get_cpuid,
7513 	.guest_has_long_mode = emulator_guest_has_long_mode,
7514 	.guest_has_movbe     = emulator_guest_has_movbe,
7515 	.guest_has_fxsr      = emulator_guest_has_fxsr,
7516 	.set_nmi_mask        = emulator_set_nmi_mask,
7517 	.get_hflags          = emulator_get_hflags,
7518 	.exiting_smm         = emulator_exiting_smm,
7519 	.leave_smm           = emulator_leave_smm,
7520 	.triple_fault        = emulator_triple_fault,
7521 	.set_xcr             = emulator_set_xcr,
7522 };
7523 
7524 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7525 {
7526 	u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7527 	/*
7528 	 * an sti; sti; sequence only disable interrupts for the first
7529 	 * instruction. So, if the last instruction, be it emulated or
7530 	 * not, left the system with the INT_STI flag enabled, it
7531 	 * means that the last instruction is an sti. We should not
7532 	 * leave the flag on in this case. The same goes for mov ss
7533 	 */
7534 	if (int_shadow & mask)
7535 		mask = 0;
7536 	if (unlikely(int_shadow || mask)) {
7537 		static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7538 		if (!mask)
7539 			kvm_make_request(KVM_REQ_EVENT, vcpu);
7540 	}
7541 }
7542 
7543 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7544 {
7545 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7546 	if (ctxt->exception.vector == PF_VECTOR)
7547 		return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7548 
7549 	if (ctxt->exception.error_code_valid)
7550 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7551 				      ctxt->exception.error_code);
7552 	else
7553 		kvm_queue_exception(vcpu, ctxt->exception.vector);
7554 	return false;
7555 }
7556 
7557 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7558 {
7559 	struct x86_emulate_ctxt *ctxt;
7560 
7561 	ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7562 	if (!ctxt) {
7563 		pr_err("kvm: failed to allocate vcpu's emulator\n");
7564 		return NULL;
7565 	}
7566 
7567 	ctxt->vcpu = vcpu;
7568 	ctxt->ops = &emulate_ops;
7569 	vcpu->arch.emulate_ctxt = ctxt;
7570 
7571 	return ctxt;
7572 }
7573 
7574 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7575 {
7576 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7577 	int cs_db, cs_l;
7578 
7579 	static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7580 
7581 	ctxt->gpa_available = false;
7582 	ctxt->eflags = kvm_get_rflags(vcpu);
7583 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7584 
7585 	ctxt->eip = kvm_rip_read(vcpu);
7586 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
7587 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
7588 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
7589 		     cs_db				? X86EMUL_MODE_PROT32 :
7590 							  X86EMUL_MODE_PROT16;
7591 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7592 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7593 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7594 
7595 	ctxt->interruptibility = 0;
7596 	ctxt->have_exception = false;
7597 	ctxt->exception.vector = -1;
7598 	ctxt->perm_ok = false;
7599 
7600 	init_decode_cache(ctxt);
7601 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7602 }
7603 
7604 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7605 {
7606 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7607 	int ret;
7608 
7609 	init_emulate_ctxt(vcpu);
7610 
7611 	ctxt->op_bytes = 2;
7612 	ctxt->ad_bytes = 2;
7613 	ctxt->_eip = ctxt->eip + inc_eip;
7614 	ret = emulate_int_real(ctxt, irq);
7615 
7616 	if (ret != X86EMUL_CONTINUE) {
7617 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7618 	} else {
7619 		ctxt->eip = ctxt->_eip;
7620 		kvm_rip_write(vcpu, ctxt->eip);
7621 		kvm_set_rflags(vcpu, ctxt->eflags);
7622 	}
7623 }
7624 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7625 
7626 static void prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, u64 *data,
7627 					   u8 ndata, u8 *insn_bytes, u8 insn_size)
7628 {
7629 	struct kvm_run *run = vcpu->run;
7630 	u64 info[5];
7631 	u8 info_start;
7632 
7633 	/*
7634 	 * Zero the whole array used to retrieve the exit info, as casting to
7635 	 * u32 for select entries will leave some chunks uninitialized.
7636 	 */
7637 	memset(&info, 0, sizeof(info));
7638 
7639 	static_call(kvm_x86_get_exit_info)(vcpu, (u32 *)&info[0], &info[1],
7640 					   &info[2], (u32 *)&info[3],
7641 					   (u32 *)&info[4]);
7642 
7643 	run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7644 	run->emulation_failure.suberror = KVM_INTERNAL_ERROR_EMULATION;
7645 
7646 	/*
7647 	 * There's currently space for 13 entries, but 5 are used for the exit
7648 	 * reason and info.  Restrict to 4 to reduce the maintenance burden
7649 	 * when expanding kvm_run.emulation_failure in the future.
7650 	 */
7651 	if (WARN_ON_ONCE(ndata > 4))
7652 		ndata = 4;
7653 
7654 	/* Always include the flags as a 'data' entry. */
7655 	info_start = 1;
7656 	run->emulation_failure.flags = 0;
7657 
7658 	if (insn_size) {
7659 		BUILD_BUG_ON((sizeof(run->emulation_failure.insn_size) +
7660 			      sizeof(run->emulation_failure.insn_bytes) != 16));
7661 		info_start += 2;
7662 		run->emulation_failure.flags |=
7663 			KVM_INTERNAL_ERROR_EMULATION_FLAG_INSTRUCTION_BYTES;
7664 		run->emulation_failure.insn_size = insn_size;
7665 		memset(run->emulation_failure.insn_bytes, 0x90,
7666 		       sizeof(run->emulation_failure.insn_bytes));
7667 		memcpy(run->emulation_failure.insn_bytes, insn_bytes, insn_size);
7668 	}
7669 
7670 	memcpy(&run->internal.data[info_start], info, sizeof(info));
7671 	memcpy(&run->internal.data[info_start + ARRAY_SIZE(info)], data,
7672 	       ndata * sizeof(data[0]));
7673 
7674 	run->emulation_failure.ndata = info_start + ARRAY_SIZE(info) + ndata;
7675 }
7676 
7677 static void prepare_emulation_ctxt_failure_exit(struct kvm_vcpu *vcpu)
7678 {
7679 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7680 
7681 	prepare_emulation_failure_exit(vcpu, NULL, 0, ctxt->fetch.data,
7682 				       ctxt->fetch.end - ctxt->fetch.data);
7683 }
7684 
7685 void __kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, u64 *data,
7686 					  u8 ndata)
7687 {
7688 	prepare_emulation_failure_exit(vcpu, data, ndata, NULL, 0);
7689 }
7690 EXPORT_SYMBOL_GPL(__kvm_prepare_emulation_failure_exit);
7691 
7692 void kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu)
7693 {
7694 	__kvm_prepare_emulation_failure_exit(vcpu, NULL, 0);
7695 }
7696 EXPORT_SYMBOL_GPL(kvm_prepare_emulation_failure_exit);
7697 
7698 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7699 {
7700 	struct kvm *kvm = vcpu->kvm;
7701 
7702 	++vcpu->stat.insn_emulation_fail;
7703 	trace_kvm_emulate_insn_failed(vcpu);
7704 
7705 	if (emulation_type & EMULTYPE_VMWARE_GP) {
7706 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7707 		return 1;
7708 	}
7709 
7710 	if (kvm->arch.exit_on_emulation_error ||
7711 	    (emulation_type & EMULTYPE_SKIP)) {
7712 		prepare_emulation_ctxt_failure_exit(vcpu);
7713 		return 0;
7714 	}
7715 
7716 	kvm_queue_exception(vcpu, UD_VECTOR);
7717 
7718 	if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7719 		prepare_emulation_ctxt_failure_exit(vcpu);
7720 		return 0;
7721 	}
7722 
7723 	return 1;
7724 }
7725 
7726 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7727 				  bool write_fault_to_shadow_pgtable,
7728 				  int emulation_type)
7729 {
7730 	gpa_t gpa = cr2_or_gpa;
7731 	kvm_pfn_t pfn;
7732 
7733 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7734 		return false;
7735 
7736 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7737 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7738 		return false;
7739 
7740 	if (!vcpu->arch.mmu->direct_map) {
7741 		/*
7742 		 * Write permission should be allowed since only
7743 		 * write access need to be emulated.
7744 		 */
7745 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7746 
7747 		/*
7748 		 * If the mapping is invalid in guest, let cpu retry
7749 		 * it to generate fault.
7750 		 */
7751 		if (gpa == UNMAPPED_GVA)
7752 			return true;
7753 	}
7754 
7755 	/*
7756 	 * Do not retry the unhandleable instruction if it faults on the
7757 	 * readonly host memory, otherwise it will goto a infinite loop:
7758 	 * retry instruction -> write #PF -> emulation fail -> retry
7759 	 * instruction -> ...
7760 	 */
7761 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7762 
7763 	/*
7764 	 * If the instruction failed on the error pfn, it can not be fixed,
7765 	 * report the error to userspace.
7766 	 */
7767 	if (is_error_noslot_pfn(pfn))
7768 		return false;
7769 
7770 	kvm_release_pfn_clean(pfn);
7771 
7772 	/* The instructions are well-emulated on direct mmu. */
7773 	if (vcpu->arch.mmu->direct_map) {
7774 		unsigned int indirect_shadow_pages;
7775 
7776 		write_lock(&vcpu->kvm->mmu_lock);
7777 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7778 		write_unlock(&vcpu->kvm->mmu_lock);
7779 
7780 		if (indirect_shadow_pages)
7781 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7782 
7783 		return true;
7784 	}
7785 
7786 	/*
7787 	 * if emulation was due to access to shadowed page table
7788 	 * and it failed try to unshadow page and re-enter the
7789 	 * guest to let CPU execute the instruction.
7790 	 */
7791 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7792 
7793 	/*
7794 	 * If the access faults on its page table, it can not
7795 	 * be fixed by unprotecting shadow page and it should
7796 	 * be reported to userspace.
7797 	 */
7798 	return !write_fault_to_shadow_pgtable;
7799 }
7800 
7801 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7802 			      gpa_t cr2_or_gpa,  int emulation_type)
7803 {
7804 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7805 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7806 
7807 	last_retry_eip = vcpu->arch.last_retry_eip;
7808 	last_retry_addr = vcpu->arch.last_retry_addr;
7809 
7810 	/*
7811 	 * If the emulation is caused by #PF and it is non-page_table
7812 	 * writing instruction, it means the VM-EXIT is caused by shadow
7813 	 * page protected, we can zap the shadow page and retry this
7814 	 * instruction directly.
7815 	 *
7816 	 * Note: if the guest uses a non-page-table modifying instruction
7817 	 * on the PDE that points to the instruction, then we will unmap
7818 	 * the instruction and go to an infinite loop. So, we cache the
7819 	 * last retried eip and the last fault address, if we meet the eip
7820 	 * and the address again, we can break out of the potential infinite
7821 	 * loop.
7822 	 */
7823 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7824 
7825 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7826 		return false;
7827 
7828 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7829 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7830 		return false;
7831 
7832 	if (x86_page_table_writing_insn(ctxt))
7833 		return false;
7834 
7835 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7836 		return false;
7837 
7838 	vcpu->arch.last_retry_eip = ctxt->eip;
7839 	vcpu->arch.last_retry_addr = cr2_or_gpa;
7840 
7841 	if (!vcpu->arch.mmu->direct_map)
7842 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7843 
7844 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7845 
7846 	return true;
7847 }
7848 
7849 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7850 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7851 
7852 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm)
7853 {
7854 	trace_kvm_smm_transition(vcpu->vcpu_id, vcpu->arch.smbase, entering_smm);
7855 
7856 	if (entering_smm) {
7857 		vcpu->arch.hflags |= HF_SMM_MASK;
7858 	} else {
7859 		vcpu->arch.hflags &= ~(HF_SMM_MASK | HF_SMM_INSIDE_NMI_MASK);
7860 
7861 		/* Process a latched INIT or SMI, if any.  */
7862 		kvm_make_request(KVM_REQ_EVENT, vcpu);
7863 
7864 		/*
7865 		 * Even if KVM_SET_SREGS2 loaded PDPTRs out of band,
7866 		 * on SMM exit we still need to reload them from
7867 		 * guest memory
7868 		 */
7869 		vcpu->arch.pdptrs_from_userspace = false;
7870 	}
7871 
7872 	kvm_mmu_reset_context(vcpu);
7873 }
7874 
7875 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7876 				unsigned long *db)
7877 {
7878 	u32 dr6 = 0;
7879 	int i;
7880 	u32 enable, rwlen;
7881 
7882 	enable = dr7;
7883 	rwlen = dr7 >> 16;
7884 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7885 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7886 			dr6 |= (1 << i);
7887 	return dr6;
7888 }
7889 
7890 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7891 {
7892 	struct kvm_run *kvm_run = vcpu->run;
7893 
7894 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7895 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7896 		kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7897 		kvm_run->debug.arch.exception = DB_VECTOR;
7898 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
7899 		return 0;
7900 	}
7901 	kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7902 	return 1;
7903 }
7904 
7905 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7906 {
7907 	unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7908 	int r;
7909 
7910 	r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7911 	if (unlikely(!r))
7912 		return 0;
7913 
7914 	/*
7915 	 * rflags is the old, "raw" value of the flags.  The new value has
7916 	 * not been saved yet.
7917 	 *
7918 	 * This is correct even for TF set by the guest, because "the
7919 	 * processor will not generate this exception after the instruction
7920 	 * that sets the TF flag".
7921 	 */
7922 	if (unlikely(rflags & X86_EFLAGS_TF))
7923 		r = kvm_vcpu_do_singlestep(vcpu);
7924 	return r;
7925 }
7926 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7927 
7928 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7929 {
7930 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7931 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7932 		struct kvm_run *kvm_run = vcpu->run;
7933 		unsigned long eip = kvm_get_linear_rip(vcpu);
7934 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7935 					   vcpu->arch.guest_debug_dr7,
7936 					   vcpu->arch.eff_db);
7937 
7938 		if (dr6 != 0) {
7939 			kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
7940 			kvm_run->debug.arch.pc = eip;
7941 			kvm_run->debug.arch.exception = DB_VECTOR;
7942 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
7943 			*r = 0;
7944 			return true;
7945 		}
7946 	}
7947 
7948 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7949 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7950 		unsigned long eip = kvm_get_linear_rip(vcpu);
7951 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7952 					   vcpu->arch.dr7,
7953 					   vcpu->arch.db);
7954 
7955 		if (dr6 != 0) {
7956 			kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7957 			*r = 1;
7958 			return true;
7959 		}
7960 	}
7961 
7962 	return false;
7963 }
7964 
7965 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7966 {
7967 	switch (ctxt->opcode_len) {
7968 	case 1:
7969 		switch (ctxt->b) {
7970 		case 0xe4:	/* IN */
7971 		case 0xe5:
7972 		case 0xec:
7973 		case 0xed:
7974 		case 0xe6:	/* OUT */
7975 		case 0xe7:
7976 		case 0xee:
7977 		case 0xef:
7978 		case 0x6c:	/* INS */
7979 		case 0x6d:
7980 		case 0x6e:	/* OUTS */
7981 		case 0x6f:
7982 			return true;
7983 		}
7984 		break;
7985 	case 2:
7986 		switch (ctxt->b) {
7987 		case 0x33:	/* RDPMC */
7988 			return true;
7989 		}
7990 		break;
7991 	}
7992 
7993 	return false;
7994 }
7995 
7996 /*
7997  * Decode to be emulated instruction. Return EMULATION_OK if success.
7998  */
7999 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
8000 				    void *insn, int insn_len)
8001 {
8002 	int r = EMULATION_OK;
8003 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8004 
8005 	init_emulate_ctxt(vcpu);
8006 
8007 	/*
8008 	 * We will reenter on the same instruction since we do not set
8009 	 * complete_userspace_io. This does not handle watchpoints yet,
8010 	 * those would be handled in the emulate_ops.
8011 	 */
8012 	if (!(emulation_type & EMULTYPE_SKIP) &&
8013 	    kvm_vcpu_check_breakpoint(vcpu, &r))
8014 		return r;
8015 
8016 	r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
8017 
8018 	trace_kvm_emulate_insn_start(vcpu);
8019 	++vcpu->stat.insn_emulation;
8020 
8021 	return r;
8022 }
8023 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
8024 
8025 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
8026 			    int emulation_type, void *insn, int insn_len)
8027 {
8028 	int r;
8029 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8030 	bool writeback = true;
8031 	bool write_fault_to_spt;
8032 
8033 	if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
8034 		return 1;
8035 
8036 	vcpu->arch.l1tf_flush_l1d = true;
8037 
8038 	/*
8039 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
8040 	 * never reused.
8041 	 */
8042 	write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
8043 	vcpu->arch.write_fault_to_shadow_pgtable = false;
8044 
8045 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
8046 		kvm_clear_exception_queue(vcpu);
8047 
8048 		r = x86_decode_emulated_instruction(vcpu, emulation_type,
8049 						    insn, insn_len);
8050 		if (r != EMULATION_OK)  {
8051 			if ((emulation_type & EMULTYPE_TRAP_UD) ||
8052 			    (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
8053 				kvm_queue_exception(vcpu, UD_VECTOR);
8054 				return 1;
8055 			}
8056 			if (reexecute_instruction(vcpu, cr2_or_gpa,
8057 						  write_fault_to_spt,
8058 						  emulation_type))
8059 				return 1;
8060 			if (ctxt->have_exception) {
8061 				/*
8062 				 * #UD should result in just EMULATION_FAILED, and trap-like
8063 				 * exception should not be encountered during decode.
8064 				 */
8065 				WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
8066 					     exception_type(ctxt->exception.vector) == EXCPT_TRAP);
8067 				inject_emulated_exception(vcpu);
8068 				return 1;
8069 			}
8070 			return handle_emulation_failure(vcpu, emulation_type);
8071 		}
8072 	}
8073 
8074 	if ((emulation_type & EMULTYPE_VMWARE_GP) &&
8075 	    !is_vmware_backdoor_opcode(ctxt)) {
8076 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
8077 		return 1;
8078 	}
8079 
8080 	/*
8081 	 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
8082 	 * for kvm_skip_emulated_instruction().  The caller is responsible for
8083 	 * updating interruptibility state and injecting single-step #DBs.
8084 	 */
8085 	if (emulation_type & EMULTYPE_SKIP) {
8086 		kvm_rip_write(vcpu, ctxt->_eip);
8087 		if (ctxt->eflags & X86_EFLAGS_RF)
8088 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
8089 		return 1;
8090 	}
8091 
8092 	if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
8093 		return 1;
8094 
8095 	/* this is needed for vmware backdoor interface to work since it
8096 	   changes registers values  during IO operation */
8097 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
8098 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8099 		emulator_invalidate_register_cache(ctxt);
8100 	}
8101 
8102 restart:
8103 	if (emulation_type & EMULTYPE_PF) {
8104 		/* Save the faulting GPA (cr2) in the address field */
8105 		ctxt->exception.address = cr2_or_gpa;
8106 
8107 		/* With shadow page tables, cr2 contains a GVA or nGPA. */
8108 		if (vcpu->arch.mmu->direct_map) {
8109 			ctxt->gpa_available = true;
8110 			ctxt->gpa_val = cr2_or_gpa;
8111 		}
8112 	} else {
8113 		/* Sanitize the address out of an abundance of paranoia. */
8114 		ctxt->exception.address = 0;
8115 	}
8116 
8117 	r = x86_emulate_insn(ctxt);
8118 
8119 	if (r == EMULATION_INTERCEPTED)
8120 		return 1;
8121 
8122 	if (r == EMULATION_FAILED) {
8123 		if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
8124 					emulation_type))
8125 			return 1;
8126 
8127 		return handle_emulation_failure(vcpu, emulation_type);
8128 	}
8129 
8130 	if (ctxt->have_exception) {
8131 		r = 1;
8132 		if (inject_emulated_exception(vcpu))
8133 			return r;
8134 	} else if (vcpu->arch.pio.count) {
8135 		if (!vcpu->arch.pio.in) {
8136 			/* FIXME: return into emulator if single-stepping.  */
8137 			vcpu->arch.pio.count = 0;
8138 		} else {
8139 			writeback = false;
8140 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
8141 		}
8142 		r = 0;
8143 	} else if (vcpu->mmio_needed) {
8144 		++vcpu->stat.mmio_exits;
8145 
8146 		if (!vcpu->mmio_is_write)
8147 			writeback = false;
8148 		r = 0;
8149 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8150 	} else if (r == EMULATION_RESTART)
8151 		goto restart;
8152 	else
8153 		r = 1;
8154 
8155 	if (writeback) {
8156 		unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
8157 		toggle_interruptibility(vcpu, ctxt->interruptibility);
8158 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8159 		if (!ctxt->have_exception ||
8160 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
8161 			kvm_rip_write(vcpu, ctxt->eip);
8162 			if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
8163 				r = kvm_vcpu_do_singlestep(vcpu);
8164 			if (kvm_x86_ops.update_emulated_instruction)
8165 				static_call(kvm_x86_update_emulated_instruction)(vcpu);
8166 			__kvm_set_rflags(vcpu, ctxt->eflags);
8167 		}
8168 
8169 		/*
8170 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
8171 		 * do nothing, and it will be requested again as soon as
8172 		 * the shadow expires.  But we still need to check here,
8173 		 * because POPF has no interrupt shadow.
8174 		 */
8175 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
8176 			kvm_make_request(KVM_REQ_EVENT, vcpu);
8177 	} else
8178 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
8179 
8180 	return r;
8181 }
8182 
8183 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
8184 {
8185 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
8186 }
8187 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
8188 
8189 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
8190 					void *insn, int insn_len)
8191 {
8192 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
8193 }
8194 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
8195 
8196 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
8197 {
8198 	vcpu->arch.pio.count = 0;
8199 	return 1;
8200 }
8201 
8202 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
8203 {
8204 	vcpu->arch.pio.count = 0;
8205 
8206 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
8207 		return 1;
8208 
8209 	return kvm_skip_emulated_instruction(vcpu);
8210 }
8211 
8212 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
8213 			    unsigned short port)
8214 {
8215 	unsigned long val = kvm_rax_read(vcpu);
8216 	int ret = emulator_pio_out(vcpu, size, port, &val, 1);
8217 
8218 	if (ret)
8219 		return ret;
8220 
8221 	/*
8222 	 * Workaround userspace that relies on old KVM behavior of %rip being
8223 	 * incremented prior to exiting to userspace to handle "OUT 0x7e".
8224 	 */
8225 	if (port == 0x7e &&
8226 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
8227 		vcpu->arch.complete_userspace_io =
8228 			complete_fast_pio_out_port_0x7e;
8229 		kvm_skip_emulated_instruction(vcpu);
8230 	} else {
8231 		vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8232 		vcpu->arch.complete_userspace_io = complete_fast_pio_out;
8233 	}
8234 	return 0;
8235 }
8236 
8237 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
8238 {
8239 	unsigned long val;
8240 
8241 	/* We should only ever be called with arch.pio.count equal to 1 */
8242 	BUG_ON(vcpu->arch.pio.count != 1);
8243 
8244 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
8245 		vcpu->arch.pio.count = 0;
8246 		return 1;
8247 	}
8248 
8249 	/* For size less than 4 we merge, else we zero extend */
8250 	val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
8251 
8252 	/*
8253 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
8254 	 * the copy and tracing
8255 	 */
8256 	emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
8257 	kvm_rax_write(vcpu, val);
8258 
8259 	return kvm_skip_emulated_instruction(vcpu);
8260 }
8261 
8262 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
8263 			   unsigned short port)
8264 {
8265 	unsigned long val;
8266 	int ret;
8267 
8268 	/* For size less than 4 we merge, else we zero extend */
8269 	val = (size < 4) ? kvm_rax_read(vcpu) : 0;
8270 
8271 	ret = emulator_pio_in(vcpu, size, port, &val, 1);
8272 	if (ret) {
8273 		kvm_rax_write(vcpu, val);
8274 		return ret;
8275 	}
8276 
8277 	vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8278 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
8279 
8280 	return 0;
8281 }
8282 
8283 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
8284 {
8285 	int ret;
8286 
8287 	if (in)
8288 		ret = kvm_fast_pio_in(vcpu, size, port);
8289 	else
8290 		ret = kvm_fast_pio_out(vcpu, size, port);
8291 	return ret && kvm_skip_emulated_instruction(vcpu);
8292 }
8293 EXPORT_SYMBOL_GPL(kvm_fast_pio);
8294 
8295 static int kvmclock_cpu_down_prep(unsigned int cpu)
8296 {
8297 	__this_cpu_write(cpu_tsc_khz, 0);
8298 	return 0;
8299 }
8300 
8301 static void tsc_khz_changed(void *data)
8302 {
8303 	struct cpufreq_freqs *freq = data;
8304 	unsigned long khz = 0;
8305 
8306 	if (data)
8307 		khz = freq->new;
8308 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8309 		khz = cpufreq_quick_get(raw_smp_processor_id());
8310 	if (!khz)
8311 		khz = tsc_khz;
8312 	__this_cpu_write(cpu_tsc_khz, khz);
8313 }
8314 
8315 #ifdef CONFIG_X86_64
8316 static void kvm_hyperv_tsc_notifier(void)
8317 {
8318 	struct kvm *kvm;
8319 	int cpu;
8320 
8321 	mutex_lock(&kvm_lock);
8322 	list_for_each_entry(kvm, &vm_list, vm_list)
8323 		kvm_make_mclock_inprogress_request(kvm);
8324 
8325 	/* no guest entries from this point */
8326 	hyperv_stop_tsc_emulation();
8327 
8328 	/* TSC frequency always matches when on Hyper-V */
8329 	for_each_present_cpu(cpu)
8330 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
8331 	kvm_max_guest_tsc_khz = tsc_khz;
8332 
8333 	list_for_each_entry(kvm, &vm_list, vm_list) {
8334 		__kvm_start_pvclock_update(kvm);
8335 		pvclock_update_vm_gtod_copy(kvm);
8336 		kvm_end_pvclock_update(kvm);
8337 	}
8338 
8339 	mutex_unlock(&kvm_lock);
8340 }
8341 #endif
8342 
8343 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
8344 {
8345 	struct kvm *kvm;
8346 	struct kvm_vcpu *vcpu;
8347 	int i, send_ipi = 0;
8348 
8349 	/*
8350 	 * We allow guests to temporarily run on slowing clocks,
8351 	 * provided we notify them after, or to run on accelerating
8352 	 * clocks, provided we notify them before.  Thus time never
8353 	 * goes backwards.
8354 	 *
8355 	 * However, we have a problem.  We can't atomically update
8356 	 * the frequency of a given CPU from this function; it is
8357 	 * merely a notifier, which can be called from any CPU.
8358 	 * Changing the TSC frequency at arbitrary points in time
8359 	 * requires a recomputation of local variables related to
8360 	 * the TSC for each VCPU.  We must flag these local variables
8361 	 * to be updated and be sure the update takes place with the
8362 	 * new frequency before any guests proceed.
8363 	 *
8364 	 * Unfortunately, the combination of hotplug CPU and frequency
8365 	 * change creates an intractable locking scenario; the order
8366 	 * of when these callouts happen is undefined with respect to
8367 	 * CPU hotplug, and they can race with each other.  As such,
8368 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
8369 	 * undefined; you can actually have a CPU frequency change take
8370 	 * place in between the computation of X and the setting of the
8371 	 * variable.  To protect against this problem, all updates of
8372 	 * the per_cpu tsc_khz variable are done in an interrupt
8373 	 * protected IPI, and all callers wishing to update the value
8374 	 * must wait for a synchronous IPI to complete (which is trivial
8375 	 * if the caller is on the CPU already).  This establishes the
8376 	 * necessary total order on variable updates.
8377 	 *
8378 	 * Note that because a guest time update may take place
8379 	 * anytime after the setting of the VCPU's request bit, the
8380 	 * correct TSC value must be set before the request.  However,
8381 	 * to ensure the update actually makes it to any guest which
8382 	 * starts running in hardware virtualization between the set
8383 	 * and the acquisition of the spinlock, we must also ping the
8384 	 * CPU after setting the request bit.
8385 	 *
8386 	 */
8387 
8388 	smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8389 
8390 	mutex_lock(&kvm_lock);
8391 	list_for_each_entry(kvm, &vm_list, vm_list) {
8392 		kvm_for_each_vcpu(i, vcpu, kvm) {
8393 			if (vcpu->cpu != cpu)
8394 				continue;
8395 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8396 			if (vcpu->cpu != raw_smp_processor_id())
8397 				send_ipi = 1;
8398 		}
8399 	}
8400 	mutex_unlock(&kvm_lock);
8401 
8402 	if (freq->old < freq->new && send_ipi) {
8403 		/*
8404 		 * We upscale the frequency.  Must make the guest
8405 		 * doesn't see old kvmclock values while running with
8406 		 * the new frequency, otherwise we risk the guest sees
8407 		 * time go backwards.
8408 		 *
8409 		 * In case we update the frequency for another cpu
8410 		 * (which might be in guest context) send an interrupt
8411 		 * to kick the cpu out of guest context.  Next time
8412 		 * guest context is entered kvmclock will be updated,
8413 		 * so the guest will not see stale values.
8414 		 */
8415 		smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8416 	}
8417 }
8418 
8419 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
8420 				     void *data)
8421 {
8422 	struct cpufreq_freqs *freq = data;
8423 	int cpu;
8424 
8425 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
8426 		return 0;
8427 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
8428 		return 0;
8429 
8430 	for_each_cpu(cpu, freq->policy->cpus)
8431 		__kvmclock_cpufreq_notifier(freq, cpu);
8432 
8433 	return 0;
8434 }
8435 
8436 static struct notifier_block kvmclock_cpufreq_notifier_block = {
8437 	.notifier_call  = kvmclock_cpufreq_notifier
8438 };
8439 
8440 static int kvmclock_cpu_online(unsigned int cpu)
8441 {
8442 	tsc_khz_changed(NULL);
8443 	return 0;
8444 }
8445 
8446 static void kvm_timer_init(void)
8447 {
8448 	max_tsc_khz = tsc_khz;
8449 
8450 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
8451 #ifdef CONFIG_CPU_FREQ
8452 		struct cpufreq_policy *policy;
8453 		int cpu;
8454 
8455 		cpu = get_cpu();
8456 		policy = cpufreq_cpu_get(cpu);
8457 		if (policy) {
8458 			if (policy->cpuinfo.max_freq)
8459 				max_tsc_khz = policy->cpuinfo.max_freq;
8460 			cpufreq_cpu_put(policy);
8461 		}
8462 		put_cpu();
8463 #endif
8464 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
8465 					  CPUFREQ_TRANSITION_NOTIFIER);
8466 	}
8467 
8468 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
8469 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
8470 }
8471 
8472 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
8473 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
8474 
8475 int kvm_is_in_guest(void)
8476 {
8477 	return __this_cpu_read(current_vcpu) != NULL;
8478 }
8479 
8480 static int kvm_is_user_mode(void)
8481 {
8482 	int user_mode = 3;
8483 
8484 	if (__this_cpu_read(current_vcpu))
8485 		user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
8486 
8487 	return user_mode != 0;
8488 }
8489 
8490 static unsigned long kvm_get_guest_ip(void)
8491 {
8492 	unsigned long ip = 0;
8493 
8494 	if (__this_cpu_read(current_vcpu))
8495 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
8496 
8497 	return ip;
8498 }
8499 
8500 static void kvm_handle_intel_pt_intr(void)
8501 {
8502 	struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
8503 
8504 	kvm_make_request(KVM_REQ_PMI, vcpu);
8505 	__set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8506 			(unsigned long *)&vcpu->arch.pmu.global_status);
8507 }
8508 
8509 static struct perf_guest_info_callbacks kvm_guest_cbs = {
8510 	.is_in_guest		= kvm_is_in_guest,
8511 	.is_user_mode		= kvm_is_user_mode,
8512 	.get_guest_ip		= kvm_get_guest_ip,
8513 	.handle_intel_pt_intr	= kvm_handle_intel_pt_intr,
8514 };
8515 
8516 #ifdef CONFIG_X86_64
8517 static void pvclock_gtod_update_fn(struct work_struct *work)
8518 {
8519 	struct kvm *kvm;
8520 
8521 	struct kvm_vcpu *vcpu;
8522 	int i;
8523 
8524 	mutex_lock(&kvm_lock);
8525 	list_for_each_entry(kvm, &vm_list, vm_list)
8526 		kvm_for_each_vcpu(i, vcpu, kvm)
8527 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8528 	atomic_set(&kvm_guest_has_master_clock, 0);
8529 	mutex_unlock(&kvm_lock);
8530 }
8531 
8532 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8533 
8534 /*
8535  * Indirection to move queue_work() out of the tk_core.seq write held
8536  * region to prevent possible deadlocks against time accessors which
8537  * are invoked with work related locks held.
8538  */
8539 static void pvclock_irq_work_fn(struct irq_work *w)
8540 {
8541 	queue_work(system_long_wq, &pvclock_gtod_work);
8542 }
8543 
8544 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
8545 
8546 /*
8547  * Notification about pvclock gtod data update.
8548  */
8549 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8550 			       void *priv)
8551 {
8552 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8553 	struct timekeeper *tk = priv;
8554 
8555 	update_pvclock_gtod(tk);
8556 
8557 	/*
8558 	 * Disable master clock if host does not trust, or does not use,
8559 	 * TSC based clocksource. Delegate queue_work() to irq_work as
8560 	 * this is invoked with tk_core.seq write held.
8561 	 */
8562 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8563 	    atomic_read(&kvm_guest_has_master_clock) != 0)
8564 		irq_work_queue(&pvclock_irq_work);
8565 	return 0;
8566 }
8567 
8568 static struct notifier_block pvclock_gtod_notifier = {
8569 	.notifier_call = pvclock_gtod_notify,
8570 };
8571 #endif
8572 
8573 int kvm_arch_init(void *opaque)
8574 {
8575 	struct kvm_x86_init_ops *ops = opaque;
8576 	int r;
8577 
8578 	if (kvm_x86_ops.hardware_enable) {
8579 		pr_err("kvm: already loaded vendor module '%s'\n", kvm_x86_ops.name);
8580 		r = -EEXIST;
8581 		goto out;
8582 	}
8583 
8584 	if (!ops->cpu_has_kvm_support()) {
8585 		pr_err_ratelimited("kvm: no hardware support for '%s'\n",
8586 				   ops->runtime_ops->name);
8587 		r = -EOPNOTSUPP;
8588 		goto out;
8589 	}
8590 	if (ops->disabled_by_bios()) {
8591 		pr_err_ratelimited("kvm: support for '%s' disabled by bios\n",
8592 				   ops->runtime_ops->name);
8593 		r = -EOPNOTSUPP;
8594 		goto out;
8595 	}
8596 
8597 	/*
8598 	 * KVM explicitly assumes that the guest has an FPU and
8599 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8600 	 * vCPU's FPU state as a fxregs_state struct.
8601 	 */
8602 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8603 		printk(KERN_ERR "kvm: inadequate fpu\n");
8604 		r = -EOPNOTSUPP;
8605 		goto out;
8606 	}
8607 
8608 	r = -ENOMEM;
8609 
8610 	x86_emulator_cache = kvm_alloc_emulator_cache();
8611 	if (!x86_emulator_cache) {
8612 		pr_err("kvm: failed to allocate cache for x86 emulator\n");
8613 		goto out;
8614 	}
8615 
8616 	user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8617 	if (!user_return_msrs) {
8618 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8619 		goto out_free_x86_emulator_cache;
8620 	}
8621 	kvm_nr_uret_msrs = 0;
8622 
8623 	r = kvm_mmu_module_init();
8624 	if (r)
8625 		goto out_free_percpu;
8626 
8627 	kvm_timer_init();
8628 
8629 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
8630 
8631 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8632 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8633 		supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8634 	}
8635 
8636 	if (pi_inject_timer == -1)
8637 		pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8638 #ifdef CONFIG_X86_64
8639 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8640 
8641 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8642 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8643 #endif
8644 
8645 	return 0;
8646 
8647 out_free_percpu:
8648 	free_percpu(user_return_msrs);
8649 out_free_x86_emulator_cache:
8650 	kmem_cache_destroy(x86_emulator_cache);
8651 out:
8652 	return r;
8653 }
8654 
8655 void kvm_arch_exit(void)
8656 {
8657 #ifdef CONFIG_X86_64
8658 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8659 		clear_hv_tscchange_cb();
8660 #endif
8661 	kvm_lapic_exit();
8662 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8663 
8664 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8665 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8666 					    CPUFREQ_TRANSITION_NOTIFIER);
8667 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8668 #ifdef CONFIG_X86_64
8669 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8670 	irq_work_sync(&pvclock_irq_work);
8671 	cancel_work_sync(&pvclock_gtod_work);
8672 #endif
8673 	kvm_x86_ops.hardware_enable = NULL;
8674 	kvm_mmu_module_exit();
8675 	free_percpu(user_return_msrs);
8676 	kmem_cache_destroy(x86_emulator_cache);
8677 #ifdef CONFIG_KVM_XEN
8678 	static_key_deferred_flush(&kvm_xen_enabled);
8679 	WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8680 #endif
8681 }
8682 
8683 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8684 {
8685 	++vcpu->stat.halt_exits;
8686 	if (lapic_in_kernel(vcpu)) {
8687 		vcpu->arch.mp_state = state;
8688 		return 1;
8689 	} else {
8690 		vcpu->run->exit_reason = reason;
8691 		return 0;
8692 	}
8693 }
8694 
8695 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8696 {
8697 	return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8698 }
8699 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8700 
8701 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8702 {
8703 	int ret = kvm_skip_emulated_instruction(vcpu);
8704 	/*
8705 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8706 	 * KVM_EXIT_DEBUG here.
8707 	 */
8708 	return kvm_vcpu_halt(vcpu) && ret;
8709 }
8710 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8711 
8712 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8713 {
8714 	int ret = kvm_skip_emulated_instruction(vcpu);
8715 
8716 	return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8717 }
8718 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8719 
8720 #ifdef CONFIG_X86_64
8721 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8722 			        unsigned long clock_type)
8723 {
8724 	struct kvm_clock_pairing clock_pairing;
8725 	struct timespec64 ts;
8726 	u64 cycle;
8727 	int ret;
8728 
8729 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8730 		return -KVM_EOPNOTSUPP;
8731 
8732 	if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8733 		return -KVM_EOPNOTSUPP;
8734 
8735 	clock_pairing.sec = ts.tv_sec;
8736 	clock_pairing.nsec = ts.tv_nsec;
8737 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8738 	clock_pairing.flags = 0;
8739 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8740 
8741 	ret = 0;
8742 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8743 			    sizeof(struct kvm_clock_pairing)))
8744 		ret = -KVM_EFAULT;
8745 
8746 	return ret;
8747 }
8748 #endif
8749 
8750 /*
8751  * kvm_pv_kick_cpu_op:  Kick a vcpu.
8752  *
8753  * @apicid - apicid of vcpu to be kicked.
8754  */
8755 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8756 {
8757 	struct kvm_lapic_irq lapic_irq;
8758 
8759 	lapic_irq.shorthand = APIC_DEST_NOSHORT;
8760 	lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8761 	lapic_irq.level = 0;
8762 	lapic_irq.dest_id = apicid;
8763 	lapic_irq.msi_redir_hint = false;
8764 
8765 	lapic_irq.delivery_mode = APIC_DM_REMRD;
8766 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8767 }
8768 
8769 bool kvm_apicv_activated(struct kvm *kvm)
8770 {
8771 	return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8772 }
8773 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8774 
8775 static void kvm_apicv_init(struct kvm *kvm)
8776 {
8777 	init_rwsem(&kvm->arch.apicv_update_lock);
8778 
8779 	if (enable_apicv)
8780 		clear_bit(APICV_INHIBIT_REASON_DISABLE,
8781 			  &kvm->arch.apicv_inhibit_reasons);
8782 	else
8783 		set_bit(APICV_INHIBIT_REASON_DISABLE,
8784 			&kvm->arch.apicv_inhibit_reasons);
8785 }
8786 
8787 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8788 {
8789 	struct kvm_vcpu *target = NULL;
8790 	struct kvm_apic_map *map;
8791 
8792 	vcpu->stat.directed_yield_attempted++;
8793 
8794 	if (single_task_running())
8795 		goto no_yield;
8796 
8797 	rcu_read_lock();
8798 	map = rcu_dereference(vcpu->kvm->arch.apic_map);
8799 
8800 	if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8801 		target = map->phys_map[dest_id]->vcpu;
8802 
8803 	rcu_read_unlock();
8804 
8805 	if (!target || !READ_ONCE(target->ready))
8806 		goto no_yield;
8807 
8808 	/* Ignore requests to yield to self */
8809 	if (vcpu == target)
8810 		goto no_yield;
8811 
8812 	if (kvm_vcpu_yield_to(target) <= 0)
8813 		goto no_yield;
8814 
8815 	vcpu->stat.directed_yield_successful++;
8816 
8817 no_yield:
8818 	return;
8819 }
8820 
8821 static int complete_hypercall_exit(struct kvm_vcpu *vcpu)
8822 {
8823 	u64 ret = vcpu->run->hypercall.ret;
8824 
8825 	if (!is_64_bit_mode(vcpu))
8826 		ret = (u32)ret;
8827 	kvm_rax_write(vcpu, ret);
8828 	++vcpu->stat.hypercalls;
8829 	return kvm_skip_emulated_instruction(vcpu);
8830 }
8831 
8832 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8833 {
8834 	unsigned long nr, a0, a1, a2, a3, ret;
8835 	int op_64_bit;
8836 
8837 	if (kvm_xen_hypercall_enabled(vcpu->kvm))
8838 		return kvm_xen_hypercall(vcpu);
8839 
8840 	if (kvm_hv_hypercall_enabled(vcpu))
8841 		return kvm_hv_hypercall(vcpu);
8842 
8843 	nr = kvm_rax_read(vcpu);
8844 	a0 = kvm_rbx_read(vcpu);
8845 	a1 = kvm_rcx_read(vcpu);
8846 	a2 = kvm_rdx_read(vcpu);
8847 	a3 = kvm_rsi_read(vcpu);
8848 
8849 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
8850 
8851 	op_64_bit = is_64_bit_mode(vcpu);
8852 	if (!op_64_bit) {
8853 		nr &= 0xFFFFFFFF;
8854 		a0 &= 0xFFFFFFFF;
8855 		a1 &= 0xFFFFFFFF;
8856 		a2 &= 0xFFFFFFFF;
8857 		a3 &= 0xFFFFFFFF;
8858 	}
8859 
8860 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8861 		ret = -KVM_EPERM;
8862 		goto out;
8863 	}
8864 
8865 	ret = -KVM_ENOSYS;
8866 
8867 	switch (nr) {
8868 	case KVM_HC_VAPIC_POLL_IRQ:
8869 		ret = 0;
8870 		break;
8871 	case KVM_HC_KICK_CPU:
8872 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8873 			break;
8874 
8875 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8876 		kvm_sched_yield(vcpu, a1);
8877 		ret = 0;
8878 		break;
8879 #ifdef CONFIG_X86_64
8880 	case KVM_HC_CLOCK_PAIRING:
8881 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8882 		break;
8883 #endif
8884 	case KVM_HC_SEND_IPI:
8885 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8886 			break;
8887 
8888 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8889 		break;
8890 	case KVM_HC_SCHED_YIELD:
8891 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8892 			break;
8893 
8894 		kvm_sched_yield(vcpu, a0);
8895 		ret = 0;
8896 		break;
8897 	case KVM_HC_MAP_GPA_RANGE: {
8898 		u64 gpa = a0, npages = a1, attrs = a2;
8899 
8900 		ret = -KVM_ENOSYS;
8901 		if (!(vcpu->kvm->arch.hypercall_exit_enabled & (1 << KVM_HC_MAP_GPA_RANGE)))
8902 			break;
8903 
8904 		if (!PAGE_ALIGNED(gpa) || !npages ||
8905 		    gpa_to_gfn(gpa) + npages <= gpa_to_gfn(gpa)) {
8906 			ret = -KVM_EINVAL;
8907 			break;
8908 		}
8909 
8910 		vcpu->run->exit_reason        = KVM_EXIT_HYPERCALL;
8911 		vcpu->run->hypercall.nr       = KVM_HC_MAP_GPA_RANGE;
8912 		vcpu->run->hypercall.args[0]  = gpa;
8913 		vcpu->run->hypercall.args[1]  = npages;
8914 		vcpu->run->hypercall.args[2]  = attrs;
8915 		vcpu->run->hypercall.longmode = op_64_bit;
8916 		vcpu->arch.complete_userspace_io = complete_hypercall_exit;
8917 		return 0;
8918 	}
8919 	default:
8920 		ret = -KVM_ENOSYS;
8921 		break;
8922 	}
8923 out:
8924 	if (!op_64_bit)
8925 		ret = (u32)ret;
8926 	kvm_rax_write(vcpu, ret);
8927 
8928 	++vcpu->stat.hypercalls;
8929 	return kvm_skip_emulated_instruction(vcpu);
8930 }
8931 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8932 
8933 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8934 {
8935 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8936 	char instruction[3];
8937 	unsigned long rip = kvm_rip_read(vcpu);
8938 
8939 	static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
8940 
8941 	return emulator_write_emulated(ctxt, rip, instruction, 3,
8942 		&ctxt->exception);
8943 }
8944 
8945 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8946 {
8947 	return vcpu->run->request_interrupt_window &&
8948 		likely(!pic_in_kernel(vcpu->kvm));
8949 }
8950 
8951 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8952 {
8953 	struct kvm_run *kvm_run = vcpu->run;
8954 
8955 	/*
8956 	 * if_flag is obsolete and useless, so do not bother
8957 	 * setting it for SEV-ES guests.  Userspace can just
8958 	 * use kvm_run->ready_for_interrupt_injection.
8959 	 */
8960 	kvm_run->if_flag = !vcpu->arch.guest_state_protected
8961 		&& (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8962 
8963 	kvm_run->cr8 = kvm_get_cr8(vcpu);
8964 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
8965 
8966 	/*
8967 	 * The call to kvm_ready_for_interrupt_injection() may end up in
8968 	 * kvm_xen_has_interrupt() which may require the srcu lock to be
8969 	 * held, to protect against changes in the vcpu_info address.
8970 	 */
8971 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8972 	kvm_run->ready_for_interrupt_injection =
8973 		pic_in_kernel(vcpu->kvm) ||
8974 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
8975 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8976 
8977 	if (is_smm(vcpu))
8978 		kvm_run->flags |= KVM_RUN_X86_SMM;
8979 }
8980 
8981 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8982 {
8983 	int max_irr, tpr;
8984 
8985 	if (!kvm_x86_ops.update_cr8_intercept)
8986 		return;
8987 
8988 	if (!lapic_in_kernel(vcpu))
8989 		return;
8990 
8991 	if (vcpu->arch.apicv_active)
8992 		return;
8993 
8994 	if (!vcpu->arch.apic->vapic_addr)
8995 		max_irr = kvm_lapic_find_highest_irr(vcpu);
8996 	else
8997 		max_irr = -1;
8998 
8999 	if (max_irr != -1)
9000 		max_irr >>= 4;
9001 
9002 	tpr = kvm_lapic_get_cr8(vcpu);
9003 
9004 	static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
9005 }
9006 
9007 
9008 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
9009 {
9010 	if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9011 		kvm_x86_ops.nested_ops->triple_fault(vcpu);
9012 		return 1;
9013 	}
9014 
9015 	return kvm_x86_ops.nested_ops->check_events(vcpu);
9016 }
9017 
9018 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
9019 {
9020 	if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
9021 		vcpu->arch.exception.error_code = false;
9022 	static_call(kvm_x86_queue_exception)(vcpu);
9023 }
9024 
9025 static int inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
9026 {
9027 	int r;
9028 	bool can_inject = true;
9029 
9030 	/* try to reinject previous events if any */
9031 
9032 	if (vcpu->arch.exception.injected) {
9033 		kvm_inject_exception(vcpu);
9034 		can_inject = false;
9035 	}
9036 	/*
9037 	 * Do not inject an NMI or interrupt if there is a pending
9038 	 * exception.  Exceptions and interrupts are recognized at
9039 	 * instruction boundaries, i.e. the start of an instruction.
9040 	 * Trap-like exceptions, e.g. #DB, have higher priority than
9041 	 * NMIs and interrupts, i.e. traps are recognized before an
9042 	 * NMI/interrupt that's pending on the same instruction.
9043 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
9044 	 * priority, but are only generated (pended) during instruction
9045 	 * execution, i.e. a pending fault-like exception means the
9046 	 * fault occurred on the *previous* instruction and must be
9047 	 * serviced prior to recognizing any new events in order to
9048 	 * fully complete the previous instruction.
9049 	 */
9050 	else if (!vcpu->arch.exception.pending) {
9051 		if (vcpu->arch.nmi_injected) {
9052 			static_call(kvm_x86_set_nmi)(vcpu);
9053 			can_inject = false;
9054 		} else if (vcpu->arch.interrupt.injected) {
9055 			static_call(kvm_x86_set_irq)(vcpu);
9056 			can_inject = false;
9057 		}
9058 	}
9059 
9060 	WARN_ON_ONCE(vcpu->arch.exception.injected &&
9061 		     vcpu->arch.exception.pending);
9062 
9063 	/*
9064 	 * Call check_nested_events() even if we reinjected a previous event
9065 	 * in order for caller to determine if it should require immediate-exit
9066 	 * from L2 to L1 due to pending L1 events which require exit
9067 	 * from L2 to L1.
9068 	 */
9069 	if (is_guest_mode(vcpu)) {
9070 		r = kvm_check_nested_events(vcpu);
9071 		if (r < 0)
9072 			goto out;
9073 	}
9074 
9075 	/* try to inject new event if pending */
9076 	if (vcpu->arch.exception.pending) {
9077 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
9078 					vcpu->arch.exception.has_error_code,
9079 					vcpu->arch.exception.error_code);
9080 
9081 		vcpu->arch.exception.pending = false;
9082 		vcpu->arch.exception.injected = true;
9083 
9084 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
9085 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
9086 					     X86_EFLAGS_RF);
9087 
9088 		if (vcpu->arch.exception.nr == DB_VECTOR) {
9089 			kvm_deliver_exception_payload(vcpu);
9090 			if (vcpu->arch.dr7 & DR7_GD) {
9091 				vcpu->arch.dr7 &= ~DR7_GD;
9092 				kvm_update_dr7(vcpu);
9093 			}
9094 		}
9095 
9096 		kvm_inject_exception(vcpu);
9097 		can_inject = false;
9098 	}
9099 
9100 	/* Don't inject interrupts if the user asked to avoid doing so */
9101 	if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ)
9102 		return 0;
9103 
9104 	/*
9105 	 * Finally, inject interrupt events.  If an event cannot be injected
9106 	 * due to architectural conditions (e.g. IF=0) a window-open exit
9107 	 * will re-request KVM_REQ_EVENT.  Sometimes however an event is pending
9108 	 * and can architecturally be injected, but we cannot do it right now:
9109 	 * an interrupt could have arrived just now and we have to inject it
9110 	 * as a vmexit, or there could already an event in the queue, which is
9111 	 * indicated by can_inject.  In that case we request an immediate exit
9112 	 * in order to make progress and get back here for another iteration.
9113 	 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
9114 	 */
9115 	if (vcpu->arch.smi_pending) {
9116 		r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
9117 		if (r < 0)
9118 			goto out;
9119 		if (r) {
9120 			vcpu->arch.smi_pending = false;
9121 			++vcpu->arch.smi_count;
9122 			enter_smm(vcpu);
9123 			can_inject = false;
9124 		} else
9125 			static_call(kvm_x86_enable_smi_window)(vcpu);
9126 	}
9127 
9128 	if (vcpu->arch.nmi_pending) {
9129 		r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
9130 		if (r < 0)
9131 			goto out;
9132 		if (r) {
9133 			--vcpu->arch.nmi_pending;
9134 			vcpu->arch.nmi_injected = true;
9135 			static_call(kvm_x86_set_nmi)(vcpu);
9136 			can_inject = false;
9137 			WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
9138 		}
9139 		if (vcpu->arch.nmi_pending)
9140 			static_call(kvm_x86_enable_nmi_window)(vcpu);
9141 	}
9142 
9143 	if (kvm_cpu_has_injectable_intr(vcpu)) {
9144 		r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
9145 		if (r < 0)
9146 			goto out;
9147 		if (r) {
9148 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
9149 			static_call(kvm_x86_set_irq)(vcpu);
9150 			WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
9151 		}
9152 		if (kvm_cpu_has_injectable_intr(vcpu))
9153 			static_call(kvm_x86_enable_irq_window)(vcpu);
9154 	}
9155 
9156 	if (is_guest_mode(vcpu) &&
9157 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
9158 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
9159 		*req_immediate_exit = true;
9160 
9161 	WARN_ON(vcpu->arch.exception.pending);
9162 	return 0;
9163 
9164 out:
9165 	if (r == -EBUSY) {
9166 		*req_immediate_exit = true;
9167 		r = 0;
9168 	}
9169 	return r;
9170 }
9171 
9172 static void process_nmi(struct kvm_vcpu *vcpu)
9173 {
9174 	unsigned limit = 2;
9175 
9176 	/*
9177 	 * x86 is limited to one NMI running, and one NMI pending after it.
9178 	 * If an NMI is already in progress, limit further NMIs to just one.
9179 	 * Otherwise, allow two (and we'll inject the first one immediately).
9180 	 */
9181 	if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
9182 		limit = 1;
9183 
9184 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
9185 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
9186 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9187 }
9188 
9189 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
9190 {
9191 	u32 flags = 0;
9192 	flags |= seg->g       << 23;
9193 	flags |= seg->db      << 22;
9194 	flags |= seg->l       << 21;
9195 	flags |= seg->avl     << 20;
9196 	flags |= seg->present << 15;
9197 	flags |= seg->dpl     << 13;
9198 	flags |= seg->s       << 12;
9199 	flags |= seg->type    << 8;
9200 	return flags;
9201 }
9202 
9203 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
9204 {
9205 	struct kvm_segment seg;
9206 	int offset;
9207 
9208 	kvm_get_segment(vcpu, &seg, n);
9209 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
9210 
9211 	if (n < 3)
9212 		offset = 0x7f84 + n * 12;
9213 	else
9214 		offset = 0x7f2c + (n - 3) * 12;
9215 
9216 	put_smstate(u32, buf, offset + 8, seg.base);
9217 	put_smstate(u32, buf, offset + 4, seg.limit);
9218 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
9219 }
9220 
9221 #ifdef CONFIG_X86_64
9222 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
9223 {
9224 	struct kvm_segment seg;
9225 	int offset;
9226 	u16 flags;
9227 
9228 	kvm_get_segment(vcpu, &seg, n);
9229 	offset = 0x7e00 + n * 16;
9230 
9231 	flags = enter_smm_get_segment_flags(&seg) >> 8;
9232 	put_smstate(u16, buf, offset, seg.selector);
9233 	put_smstate(u16, buf, offset + 2, flags);
9234 	put_smstate(u32, buf, offset + 4, seg.limit);
9235 	put_smstate(u64, buf, offset + 8, seg.base);
9236 }
9237 #endif
9238 
9239 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
9240 {
9241 	struct desc_ptr dt;
9242 	struct kvm_segment seg;
9243 	unsigned long val;
9244 	int i;
9245 
9246 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
9247 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
9248 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
9249 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
9250 
9251 	for (i = 0; i < 8; i++)
9252 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
9253 
9254 	kvm_get_dr(vcpu, 6, &val);
9255 	put_smstate(u32, buf, 0x7fcc, (u32)val);
9256 	kvm_get_dr(vcpu, 7, &val);
9257 	put_smstate(u32, buf, 0x7fc8, (u32)val);
9258 
9259 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9260 	put_smstate(u32, buf, 0x7fc4, seg.selector);
9261 	put_smstate(u32, buf, 0x7f64, seg.base);
9262 	put_smstate(u32, buf, 0x7f60, seg.limit);
9263 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
9264 
9265 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9266 	put_smstate(u32, buf, 0x7fc0, seg.selector);
9267 	put_smstate(u32, buf, 0x7f80, seg.base);
9268 	put_smstate(u32, buf, 0x7f7c, seg.limit);
9269 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
9270 
9271 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
9272 	put_smstate(u32, buf, 0x7f74, dt.address);
9273 	put_smstate(u32, buf, 0x7f70, dt.size);
9274 
9275 	static_call(kvm_x86_get_idt)(vcpu, &dt);
9276 	put_smstate(u32, buf, 0x7f58, dt.address);
9277 	put_smstate(u32, buf, 0x7f54, dt.size);
9278 
9279 	for (i = 0; i < 6; i++)
9280 		enter_smm_save_seg_32(vcpu, buf, i);
9281 
9282 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
9283 
9284 	/* revision id */
9285 	put_smstate(u32, buf, 0x7efc, 0x00020000);
9286 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
9287 }
9288 
9289 #ifdef CONFIG_X86_64
9290 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
9291 {
9292 	struct desc_ptr dt;
9293 	struct kvm_segment seg;
9294 	unsigned long val;
9295 	int i;
9296 
9297 	for (i = 0; i < 16; i++)
9298 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
9299 
9300 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
9301 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
9302 
9303 	kvm_get_dr(vcpu, 6, &val);
9304 	put_smstate(u64, buf, 0x7f68, val);
9305 	kvm_get_dr(vcpu, 7, &val);
9306 	put_smstate(u64, buf, 0x7f60, val);
9307 
9308 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
9309 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
9310 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
9311 
9312 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
9313 
9314 	/* revision id */
9315 	put_smstate(u32, buf, 0x7efc, 0x00020064);
9316 
9317 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
9318 
9319 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9320 	put_smstate(u16, buf, 0x7e90, seg.selector);
9321 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
9322 	put_smstate(u32, buf, 0x7e94, seg.limit);
9323 	put_smstate(u64, buf, 0x7e98, seg.base);
9324 
9325 	static_call(kvm_x86_get_idt)(vcpu, &dt);
9326 	put_smstate(u32, buf, 0x7e84, dt.size);
9327 	put_smstate(u64, buf, 0x7e88, dt.address);
9328 
9329 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9330 	put_smstate(u16, buf, 0x7e70, seg.selector);
9331 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
9332 	put_smstate(u32, buf, 0x7e74, seg.limit);
9333 	put_smstate(u64, buf, 0x7e78, seg.base);
9334 
9335 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
9336 	put_smstate(u32, buf, 0x7e64, dt.size);
9337 	put_smstate(u64, buf, 0x7e68, dt.address);
9338 
9339 	for (i = 0; i < 6; i++)
9340 		enter_smm_save_seg_64(vcpu, buf, i);
9341 }
9342 #endif
9343 
9344 static void enter_smm(struct kvm_vcpu *vcpu)
9345 {
9346 	struct kvm_segment cs, ds;
9347 	struct desc_ptr dt;
9348 	unsigned long cr0;
9349 	char buf[512];
9350 
9351 	memset(buf, 0, 512);
9352 #ifdef CONFIG_X86_64
9353 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9354 		enter_smm_save_state_64(vcpu, buf);
9355 	else
9356 #endif
9357 		enter_smm_save_state_32(vcpu, buf);
9358 
9359 	/*
9360 	 * Give enter_smm() a chance to make ISA-specific changes to the vCPU
9361 	 * state (e.g. leave guest mode) after we've saved the state into the
9362 	 * SMM state-save area.
9363 	 */
9364 	static_call(kvm_x86_enter_smm)(vcpu, buf);
9365 
9366 	kvm_smm_changed(vcpu, true);
9367 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
9368 
9369 	if (static_call(kvm_x86_get_nmi_mask)(vcpu))
9370 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
9371 	else
9372 		static_call(kvm_x86_set_nmi_mask)(vcpu, true);
9373 
9374 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
9375 	kvm_rip_write(vcpu, 0x8000);
9376 
9377 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
9378 	static_call(kvm_x86_set_cr0)(vcpu, cr0);
9379 	vcpu->arch.cr0 = cr0;
9380 
9381 	static_call(kvm_x86_set_cr4)(vcpu, 0);
9382 
9383 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
9384 	dt.address = dt.size = 0;
9385 	static_call(kvm_x86_set_idt)(vcpu, &dt);
9386 
9387 	kvm_set_dr(vcpu, 7, DR7_FIXED_1);
9388 
9389 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
9390 	cs.base = vcpu->arch.smbase;
9391 
9392 	ds.selector = 0;
9393 	ds.base = 0;
9394 
9395 	cs.limit    = ds.limit = 0xffffffff;
9396 	cs.type     = ds.type = 0x3;
9397 	cs.dpl      = ds.dpl = 0;
9398 	cs.db       = ds.db = 0;
9399 	cs.s        = ds.s = 1;
9400 	cs.l        = ds.l = 0;
9401 	cs.g        = ds.g = 1;
9402 	cs.avl      = ds.avl = 0;
9403 	cs.present  = ds.present = 1;
9404 	cs.unusable = ds.unusable = 0;
9405 	cs.padding  = ds.padding = 0;
9406 
9407 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9408 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
9409 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
9410 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
9411 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
9412 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
9413 
9414 #ifdef CONFIG_X86_64
9415 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9416 		static_call(kvm_x86_set_efer)(vcpu, 0);
9417 #endif
9418 
9419 	kvm_update_cpuid_runtime(vcpu);
9420 	kvm_mmu_reset_context(vcpu);
9421 }
9422 
9423 static void process_smi(struct kvm_vcpu *vcpu)
9424 {
9425 	vcpu->arch.smi_pending = true;
9426 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9427 }
9428 
9429 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
9430 				       unsigned long *vcpu_bitmap)
9431 {
9432 	kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC, vcpu_bitmap);
9433 }
9434 
9435 void kvm_make_scan_ioapic_request(struct kvm *kvm)
9436 {
9437 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
9438 }
9439 
9440 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
9441 {
9442 	bool activate;
9443 
9444 	if (!lapic_in_kernel(vcpu))
9445 		return;
9446 
9447 	down_read(&vcpu->kvm->arch.apicv_update_lock);
9448 
9449 	activate = kvm_apicv_activated(vcpu->kvm);
9450 	if (vcpu->arch.apicv_active == activate)
9451 		goto out;
9452 
9453 	vcpu->arch.apicv_active = activate;
9454 	kvm_apic_update_apicv(vcpu);
9455 	static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
9456 
9457 	/*
9458 	 * When APICv gets disabled, we may still have injected interrupts
9459 	 * pending. At the same time, KVM_REQ_EVENT may not be set as APICv was
9460 	 * still active when the interrupt got accepted. Make sure
9461 	 * inject_pending_event() is called to check for that.
9462 	 */
9463 	if (!vcpu->arch.apicv_active)
9464 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9465 
9466 out:
9467 	up_read(&vcpu->kvm->arch.apicv_update_lock);
9468 }
9469 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
9470 
9471 void __kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9472 {
9473 	unsigned long old, new;
9474 
9475 	lockdep_assert_held_write(&kvm->arch.apicv_update_lock);
9476 
9477 	if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
9478 	    !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
9479 		return;
9480 
9481 	old = new = kvm->arch.apicv_inhibit_reasons;
9482 
9483 	if (activate)
9484 		__clear_bit(bit, &new);
9485 	else
9486 		__set_bit(bit, &new);
9487 
9488 	if (!!old != !!new) {
9489 		trace_kvm_apicv_update_request(activate, bit);
9490 		/*
9491 		 * Kick all vCPUs before setting apicv_inhibit_reasons to avoid
9492 		 * false positives in the sanity check WARN in svm_vcpu_run().
9493 		 * This task will wait for all vCPUs to ack the kick IRQ before
9494 		 * updating apicv_inhibit_reasons, and all other vCPUs will
9495 		 * block on acquiring apicv_update_lock so that vCPUs can't
9496 		 * redo svm_vcpu_run() without seeing the new inhibit state.
9497 		 *
9498 		 * Note, holding apicv_update_lock and taking it in the read
9499 		 * side (handling the request) also prevents other vCPUs from
9500 		 * servicing the request with a stale apicv_inhibit_reasons.
9501 		 */
9502 		kvm_make_all_cpus_request(kvm, KVM_REQ_APICV_UPDATE);
9503 		kvm->arch.apicv_inhibit_reasons = new;
9504 		if (new) {
9505 			unsigned long gfn = gpa_to_gfn(APIC_DEFAULT_PHYS_BASE);
9506 			kvm_zap_gfn_range(kvm, gfn, gfn+1);
9507 		}
9508 	} else
9509 		kvm->arch.apicv_inhibit_reasons = new;
9510 }
9511 EXPORT_SYMBOL_GPL(__kvm_request_apicv_update);
9512 
9513 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9514 {
9515 	down_write(&kvm->arch.apicv_update_lock);
9516 	__kvm_request_apicv_update(kvm, activate, bit);
9517 	up_write(&kvm->arch.apicv_update_lock);
9518 }
9519 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
9520 
9521 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
9522 {
9523 	if (!kvm_apic_present(vcpu))
9524 		return;
9525 
9526 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
9527 
9528 	if (irqchip_split(vcpu->kvm))
9529 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
9530 	else {
9531 		if (vcpu->arch.apicv_active)
9532 			static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9533 		if (ioapic_in_kernel(vcpu->kvm))
9534 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
9535 	}
9536 
9537 	if (is_guest_mode(vcpu))
9538 		vcpu->arch.load_eoi_exitmap_pending = true;
9539 	else
9540 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
9541 }
9542 
9543 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
9544 {
9545 	u64 eoi_exit_bitmap[4];
9546 
9547 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
9548 		return;
9549 
9550 	if (to_hv_vcpu(vcpu))
9551 		bitmap_or((ulong *)eoi_exit_bitmap,
9552 			  vcpu->arch.ioapic_handled_vectors,
9553 			  to_hv_synic(vcpu)->vec_bitmap, 256);
9554 
9555 	static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
9556 }
9557 
9558 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
9559 					    unsigned long start, unsigned long end)
9560 {
9561 	unsigned long apic_address;
9562 
9563 	/*
9564 	 * The physical address of apic access page is stored in the VMCS.
9565 	 * Update it when it becomes invalid.
9566 	 */
9567 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9568 	if (start <= apic_address && apic_address < end)
9569 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9570 }
9571 
9572 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9573 {
9574 	if (!lapic_in_kernel(vcpu))
9575 		return;
9576 
9577 	if (!kvm_x86_ops.set_apic_access_page_addr)
9578 		return;
9579 
9580 	static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9581 }
9582 
9583 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9584 {
9585 	smp_send_reschedule(vcpu->cpu);
9586 }
9587 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9588 
9589 /*
9590  * Returns 1 to let vcpu_run() continue the guest execution loop without
9591  * exiting to the userspace.  Otherwise, the value will be returned to the
9592  * userspace.
9593  */
9594 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9595 {
9596 	int r;
9597 	bool req_int_win =
9598 		dm_request_for_irq_injection(vcpu) &&
9599 		kvm_cpu_accept_dm_intr(vcpu);
9600 	fastpath_t exit_fastpath;
9601 
9602 	bool req_immediate_exit = false;
9603 
9604 	/* Forbid vmenter if vcpu dirty ring is soft-full */
9605 	if (unlikely(vcpu->kvm->dirty_ring_size &&
9606 		     kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9607 		vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9608 		trace_kvm_dirty_ring_exit(vcpu);
9609 		r = 0;
9610 		goto out;
9611 	}
9612 
9613 	if (kvm_request_pending(vcpu)) {
9614 		if (kvm_check_request(KVM_REQ_VM_DEAD, vcpu)) {
9615 			r = -EIO;
9616 			goto out;
9617 		}
9618 		if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9619 			if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9620 				r = 0;
9621 				goto out;
9622 			}
9623 		}
9624 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9625 			kvm_mmu_unload(vcpu);
9626 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9627 			__kvm_migrate_timers(vcpu);
9628 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9629 			kvm_update_masterclock(vcpu->kvm);
9630 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9631 			kvm_gen_kvmclock_update(vcpu);
9632 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9633 			r = kvm_guest_time_update(vcpu);
9634 			if (unlikely(r))
9635 				goto out;
9636 		}
9637 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9638 			kvm_mmu_sync_roots(vcpu);
9639 		if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9640 			kvm_mmu_load_pgd(vcpu);
9641 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9642 			kvm_vcpu_flush_tlb_all(vcpu);
9643 
9644 			/* Flushing all ASIDs flushes the current ASID... */
9645 			kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9646 		}
9647 		if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
9648 			kvm_vcpu_flush_tlb_current(vcpu);
9649 		if (kvm_check_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu))
9650 			kvm_vcpu_flush_tlb_guest(vcpu);
9651 
9652 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9653 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9654 			r = 0;
9655 			goto out;
9656 		}
9657 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9658 			if (is_guest_mode(vcpu)) {
9659 				kvm_x86_ops.nested_ops->triple_fault(vcpu);
9660 			} else {
9661 				vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9662 				vcpu->mmio_needed = 0;
9663 				r = 0;
9664 				goto out;
9665 			}
9666 		}
9667 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9668 			/* Page is swapped out. Do synthetic halt */
9669 			vcpu->arch.apf.halted = true;
9670 			r = 1;
9671 			goto out;
9672 		}
9673 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9674 			record_steal_time(vcpu);
9675 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
9676 			process_smi(vcpu);
9677 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
9678 			process_nmi(vcpu);
9679 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
9680 			kvm_pmu_handle_event(vcpu);
9681 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
9682 			kvm_pmu_deliver_pmi(vcpu);
9683 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9684 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9685 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
9686 				     vcpu->arch.ioapic_handled_vectors)) {
9687 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9688 				vcpu->run->eoi.vector =
9689 						vcpu->arch.pending_ioapic_eoi;
9690 				r = 0;
9691 				goto out;
9692 			}
9693 		}
9694 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9695 			vcpu_scan_ioapic(vcpu);
9696 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9697 			vcpu_load_eoi_exitmap(vcpu);
9698 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9699 			kvm_vcpu_reload_apic_access_page(vcpu);
9700 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9701 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9702 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9703 			r = 0;
9704 			goto out;
9705 		}
9706 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9707 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9708 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9709 			r = 0;
9710 			goto out;
9711 		}
9712 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9713 			struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9714 
9715 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9716 			vcpu->run->hyperv = hv_vcpu->exit;
9717 			r = 0;
9718 			goto out;
9719 		}
9720 
9721 		/*
9722 		 * KVM_REQ_HV_STIMER has to be processed after
9723 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9724 		 * depend on the guest clock being up-to-date
9725 		 */
9726 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9727 			kvm_hv_process_stimers(vcpu);
9728 		if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9729 			kvm_vcpu_update_apicv(vcpu);
9730 		if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9731 			kvm_check_async_pf_completion(vcpu);
9732 		if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9733 			static_call(kvm_x86_msr_filter_changed)(vcpu);
9734 
9735 		if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9736 			static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9737 	}
9738 
9739 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9740 	    kvm_xen_has_interrupt(vcpu)) {
9741 		++vcpu->stat.req_event;
9742 		r = kvm_apic_accept_events(vcpu);
9743 		if (r < 0) {
9744 			r = 0;
9745 			goto out;
9746 		}
9747 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9748 			r = 1;
9749 			goto out;
9750 		}
9751 
9752 		r = inject_pending_event(vcpu, &req_immediate_exit);
9753 		if (r < 0) {
9754 			r = 0;
9755 			goto out;
9756 		}
9757 		if (req_int_win)
9758 			static_call(kvm_x86_enable_irq_window)(vcpu);
9759 
9760 		if (kvm_lapic_enabled(vcpu)) {
9761 			update_cr8_intercept(vcpu);
9762 			kvm_lapic_sync_to_vapic(vcpu);
9763 		}
9764 	}
9765 
9766 	r = kvm_mmu_reload(vcpu);
9767 	if (unlikely(r)) {
9768 		goto cancel_injection;
9769 	}
9770 
9771 	preempt_disable();
9772 
9773 	static_call(kvm_x86_prepare_guest_switch)(vcpu);
9774 
9775 	/*
9776 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
9777 	 * IPI are then delayed after guest entry, which ensures that they
9778 	 * result in virtual interrupt delivery.
9779 	 */
9780 	local_irq_disable();
9781 	vcpu->mode = IN_GUEST_MODE;
9782 
9783 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9784 
9785 	/*
9786 	 * 1) We should set ->mode before checking ->requests.  Please see
9787 	 * the comment in kvm_vcpu_exiting_guest_mode().
9788 	 *
9789 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
9790 	 * pairs with the memory barrier implicit in pi_test_and_set_on
9791 	 * (see vmx_deliver_posted_interrupt).
9792 	 *
9793 	 * 3) This also orders the write to mode from any reads to the page
9794 	 * tables done while the VCPU is running.  Please see the comment
9795 	 * in kvm_flush_remote_tlbs.
9796 	 */
9797 	smp_mb__after_srcu_read_unlock();
9798 
9799 	/*
9800 	 * This handles the case where a posted interrupt was
9801 	 * notified with kvm_vcpu_kick.
9802 	 */
9803 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
9804 		static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9805 
9806 	if (kvm_vcpu_exit_request(vcpu)) {
9807 		vcpu->mode = OUTSIDE_GUEST_MODE;
9808 		smp_wmb();
9809 		local_irq_enable();
9810 		preempt_enable();
9811 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9812 		r = 1;
9813 		goto cancel_injection;
9814 	}
9815 
9816 	if (req_immediate_exit) {
9817 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9818 		static_call(kvm_x86_request_immediate_exit)(vcpu);
9819 	}
9820 
9821 	fpregs_assert_state_consistent();
9822 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
9823 		switch_fpu_return();
9824 
9825 	if (unlikely(vcpu->arch.switch_db_regs)) {
9826 		set_debugreg(0, 7);
9827 		set_debugreg(vcpu->arch.eff_db[0], 0);
9828 		set_debugreg(vcpu->arch.eff_db[1], 1);
9829 		set_debugreg(vcpu->arch.eff_db[2], 2);
9830 		set_debugreg(vcpu->arch.eff_db[3], 3);
9831 	} else if (unlikely(hw_breakpoint_active())) {
9832 		set_debugreg(0, 7);
9833 	}
9834 
9835 	for (;;) {
9836 		/*
9837 		 * Assert that vCPU vs. VM APICv state is consistent.  An APICv
9838 		 * update must kick and wait for all vCPUs before toggling the
9839 		 * per-VM state, and responsing vCPUs must wait for the update
9840 		 * to complete before servicing KVM_REQ_APICV_UPDATE.
9841 		 */
9842 		WARN_ON_ONCE(kvm_apicv_activated(vcpu->kvm) != kvm_vcpu_apicv_active(vcpu));
9843 
9844 		exit_fastpath = static_call(kvm_x86_run)(vcpu);
9845 		if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9846 			break;
9847 
9848 		if (vcpu->arch.apicv_active)
9849 			static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9850 
9851 		if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9852 			exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9853 			break;
9854 		}
9855 	}
9856 
9857 	/*
9858 	 * Do this here before restoring debug registers on the host.  And
9859 	 * since we do this before handling the vmexit, a DR access vmexit
9860 	 * can (a) read the correct value of the debug registers, (b) set
9861 	 * KVM_DEBUGREG_WONT_EXIT again.
9862 	 */
9863 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9864 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9865 		static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9866 		kvm_update_dr0123(vcpu);
9867 		kvm_update_dr7(vcpu);
9868 	}
9869 
9870 	/*
9871 	 * If the guest has used debug registers, at least dr7
9872 	 * will be disabled while returning to the host.
9873 	 * If we don't have active breakpoints in the host, we don't
9874 	 * care about the messed up debug address registers. But if
9875 	 * we have some of them active, restore the old state.
9876 	 */
9877 	if (hw_breakpoint_active())
9878 		hw_breakpoint_restore();
9879 
9880 	vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9881 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9882 
9883 	vcpu->mode = OUTSIDE_GUEST_MODE;
9884 	smp_wmb();
9885 
9886 	static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9887 
9888 	/*
9889 	 * Consume any pending interrupts, including the possible source of
9890 	 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9891 	 * An instruction is required after local_irq_enable() to fully unblock
9892 	 * interrupts on processors that implement an interrupt shadow, the
9893 	 * stat.exits increment will do nicely.
9894 	 */
9895 	kvm_before_interrupt(vcpu);
9896 	local_irq_enable();
9897 	++vcpu->stat.exits;
9898 	local_irq_disable();
9899 	kvm_after_interrupt(vcpu);
9900 
9901 	/*
9902 	 * Wait until after servicing IRQs to account guest time so that any
9903 	 * ticks that occurred while running the guest are properly accounted
9904 	 * to the guest.  Waiting until IRQs are enabled degrades the accuracy
9905 	 * of accounting via context tracking, but the loss of accuracy is
9906 	 * acceptable for all known use cases.
9907 	 */
9908 	vtime_account_guest_exit();
9909 
9910 	if (lapic_in_kernel(vcpu)) {
9911 		s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9912 		if (delta != S64_MIN) {
9913 			trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9914 			vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9915 		}
9916 	}
9917 
9918 	local_irq_enable();
9919 	preempt_enable();
9920 
9921 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9922 
9923 	/*
9924 	 * Profile KVM exit RIPs:
9925 	 */
9926 	if (unlikely(prof_on == KVM_PROFILING)) {
9927 		unsigned long rip = kvm_rip_read(vcpu);
9928 		profile_hit(KVM_PROFILING, (void *)rip);
9929 	}
9930 
9931 	if (unlikely(vcpu->arch.tsc_always_catchup))
9932 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9933 
9934 	if (vcpu->arch.apic_attention)
9935 		kvm_lapic_sync_from_vapic(vcpu);
9936 
9937 	r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
9938 	return r;
9939 
9940 cancel_injection:
9941 	if (req_immediate_exit)
9942 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9943 	static_call(kvm_x86_cancel_injection)(vcpu);
9944 	if (unlikely(vcpu->arch.apic_attention))
9945 		kvm_lapic_sync_from_vapic(vcpu);
9946 out:
9947 	return r;
9948 }
9949 
9950 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9951 {
9952 	if (!kvm_arch_vcpu_runnable(vcpu) &&
9953 	    (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
9954 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9955 		kvm_vcpu_block(vcpu);
9956 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9957 
9958 		if (kvm_x86_ops.post_block)
9959 			static_call(kvm_x86_post_block)(vcpu);
9960 
9961 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9962 			return 1;
9963 	}
9964 
9965 	if (kvm_apic_accept_events(vcpu) < 0)
9966 		return 0;
9967 	switch(vcpu->arch.mp_state) {
9968 	case KVM_MP_STATE_HALTED:
9969 	case KVM_MP_STATE_AP_RESET_HOLD:
9970 		vcpu->arch.pv.pv_unhalted = false;
9971 		vcpu->arch.mp_state =
9972 			KVM_MP_STATE_RUNNABLE;
9973 		fallthrough;
9974 	case KVM_MP_STATE_RUNNABLE:
9975 		vcpu->arch.apf.halted = false;
9976 		break;
9977 	case KVM_MP_STATE_INIT_RECEIVED:
9978 		break;
9979 	default:
9980 		return -EINTR;
9981 	}
9982 	return 1;
9983 }
9984 
9985 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9986 {
9987 	if (is_guest_mode(vcpu))
9988 		kvm_check_nested_events(vcpu);
9989 
9990 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9991 		!vcpu->arch.apf.halted);
9992 }
9993 
9994 static int vcpu_run(struct kvm_vcpu *vcpu)
9995 {
9996 	int r;
9997 	struct kvm *kvm = vcpu->kvm;
9998 
9999 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
10000 	vcpu->arch.l1tf_flush_l1d = true;
10001 
10002 	for (;;) {
10003 		if (kvm_vcpu_running(vcpu)) {
10004 			r = vcpu_enter_guest(vcpu);
10005 		} else {
10006 			r = vcpu_block(kvm, vcpu);
10007 		}
10008 
10009 		if (r <= 0)
10010 			break;
10011 
10012 		kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
10013 		if (kvm_cpu_has_pending_timer(vcpu))
10014 			kvm_inject_pending_timer_irqs(vcpu);
10015 
10016 		if (dm_request_for_irq_injection(vcpu) &&
10017 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
10018 			r = 0;
10019 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
10020 			++vcpu->stat.request_irq_exits;
10021 			break;
10022 		}
10023 
10024 		if (__xfer_to_guest_mode_work_pending()) {
10025 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
10026 			r = xfer_to_guest_mode_handle_work(vcpu);
10027 			if (r)
10028 				return r;
10029 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
10030 		}
10031 	}
10032 
10033 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
10034 
10035 	return r;
10036 }
10037 
10038 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
10039 {
10040 	int r;
10041 
10042 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
10043 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
10044 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
10045 	return r;
10046 }
10047 
10048 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
10049 {
10050 	BUG_ON(!vcpu->arch.pio.count);
10051 
10052 	return complete_emulated_io(vcpu);
10053 }
10054 
10055 /*
10056  * Implements the following, as a state machine:
10057  *
10058  * read:
10059  *   for each fragment
10060  *     for each mmio piece in the fragment
10061  *       write gpa, len
10062  *       exit
10063  *       copy data
10064  *   execute insn
10065  *
10066  * write:
10067  *   for each fragment
10068  *     for each mmio piece in the fragment
10069  *       write gpa, len
10070  *       copy data
10071  *       exit
10072  */
10073 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
10074 {
10075 	struct kvm_run *run = vcpu->run;
10076 	struct kvm_mmio_fragment *frag;
10077 	unsigned len;
10078 
10079 	BUG_ON(!vcpu->mmio_needed);
10080 
10081 	/* Complete previous fragment */
10082 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
10083 	len = min(8u, frag->len);
10084 	if (!vcpu->mmio_is_write)
10085 		memcpy(frag->data, run->mmio.data, len);
10086 
10087 	if (frag->len <= 8) {
10088 		/* Switch to the next fragment. */
10089 		frag++;
10090 		vcpu->mmio_cur_fragment++;
10091 	} else {
10092 		/* Go forward to the next mmio piece. */
10093 		frag->data += len;
10094 		frag->gpa += len;
10095 		frag->len -= len;
10096 	}
10097 
10098 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
10099 		vcpu->mmio_needed = 0;
10100 
10101 		/* FIXME: return into emulator if single-stepping.  */
10102 		if (vcpu->mmio_is_write)
10103 			return 1;
10104 		vcpu->mmio_read_completed = 1;
10105 		return complete_emulated_io(vcpu);
10106 	}
10107 
10108 	run->exit_reason = KVM_EXIT_MMIO;
10109 	run->mmio.phys_addr = frag->gpa;
10110 	if (vcpu->mmio_is_write)
10111 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
10112 	run->mmio.len = min(8u, frag->len);
10113 	run->mmio.is_write = vcpu->mmio_is_write;
10114 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
10115 	return 0;
10116 }
10117 
10118 /* Swap (qemu) user FPU context for the guest FPU context. */
10119 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
10120 {
10121 	/*
10122 	 * Exclude PKRU from restore as restored separately in
10123 	 * kvm_x86_ops.run().
10124 	 */
10125 	fpu_swap_kvm_fpstate(&vcpu->arch.guest_fpu, true);
10126 	trace_kvm_fpu(1);
10127 }
10128 
10129 /* When vcpu_run ends, restore user space FPU context. */
10130 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
10131 {
10132 	fpu_swap_kvm_fpstate(&vcpu->arch.guest_fpu, false);
10133 	++vcpu->stat.fpu_reload;
10134 	trace_kvm_fpu(0);
10135 }
10136 
10137 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
10138 {
10139 	struct kvm_run *kvm_run = vcpu->run;
10140 	int r;
10141 
10142 	vcpu_load(vcpu);
10143 	kvm_sigset_activate(vcpu);
10144 	kvm_run->flags = 0;
10145 	kvm_load_guest_fpu(vcpu);
10146 
10147 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
10148 		if (kvm_run->immediate_exit) {
10149 			r = -EINTR;
10150 			goto out;
10151 		}
10152 		kvm_vcpu_block(vcpu);
10153 		if (kvm_apic_accept_events(vcpu) < 0) {
10154 			r = 0;
10155 			goto out;
10156 		}
10157 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
10158 		r = -EAGAIN;
10159 		if (signal_pending(current)) {
10160 			r = -EINTR;
10161 			kvm_run->exit_reason = KVM_EXIT_INTR;
10162 			++vcpu->stat.signal_exits;
10163 		}
10164 		goto out;
10165 	}
10166 
10167 	if ((kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) ||
10168 	    (kvm_run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)) {
10169 		r = -EINVAL;
10170 		goto out;
10171 	}
10172 
10173 	if (kvm_run->kvm_dirty_regs) {
10174 		r = sync_regs(vcpu);
10175 		if (r != 0)
10176 			goto out;
10177 	}
10178 
10179 	/* re-sync apic's tpr */
10180 	if (!lapic_in_kernel(vcpu)) {
10181 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
10182 			r = -EINVAL;
10183 			goto out;
10184 		}
10185 	}
10186 
10187 	if (unlikely(vcpu->arch.complete_userspace_io)) {
10188 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
10189 		vcpu->arch.complete_userspace_io = NULL;
10190 		r = cui(vcpu);
10191 		if (r <= 0)
10192 			goto out;
10193 	} else
10194 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
10195 
10196 	if (kvm_run->immediate_exit)
10197 		r = -EINTR;
10198 	else
10199 		r = vcpu_run(vcpu);
10200 
10201 out:
10202 	kvm_put_guest_fpu(vcpu);
10203 	if (kvm_run->kvm_valid_regs)
10204 		store_regs(vcpu);
10205 	post_kvm_run_save(vcpu);
10206 	kvm_sigset_deactivate(vcpu);
10207 
10208 	vcpu_put(vcpu);
10209 	return r;
10210 }
10211 
10212 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10213 {
10214 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
10215 		/*
10216 		 * We are here if userspace calls get_regs() in the middle of
10217 		 * instruction emulation. Registers state needs to be copied
10218 		 * back from emulation context to vcpu. Userspace shouldn't do
10219 		 * that usually, but some bad designed PV devices (vmware
10220 		 * backdoor interface) need this to work
10221 		 */
10222 		emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
10223 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10224 	}
10225 	regs->rax = kvm_rax_read(vcpu);
10226 	regs->rbx = kvm_rbx_read(vcpu);
10227 	regs->rcx = kvm_rcx_read(vcpu);
10228 	regs->rdx = kvm_rdx_read(vcpu);
10229 	regs->rsi = kvm_rsi_read(vcpu);
10230 	regs->rdi = kvm_rdi_read(vcpu);
10231 	regs->rsp = kvm_rsp_read(vcpu);
10232 	regs->rbp = kvm_rbp_read(vcpu);
10233 #ifdef CONFIG_X86_64
10234 	regs->r8 = kvm_r8_read(vcpu);
10235 	regs->r9 = kvm_r9_read(vcpu);
10236 	regs->r10 = kvm_r10_read(vcpu);
10237 	regs->r11 = kvm_r11_read(vcpu);
10238 	regs->r12 = kvm_r12_read(vcpu);
10239 	regs->r13 = kvm_r13_read(vcpu);
10240 	regs->r14 = kvm_r14_read(vcpu);
10241 	regs->r15 = kvm_r15_read(vcpu);
10242 #endif
10243 
10244 	regs->rip = kvm_rip_read(vcpu);
10245 	regs->rflags = kvm_get_rflags(vcpu);
10246 }
10247 
10248 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10249 {
10250 	vcpu_load(vcpu);
10251 	__get_regs(vcpu, regs);
10252 	vcpu_put(vcpu);
10253 	return 0;
10254 }
10255 
10256 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10257 {
10258 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
10259 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10260 
10261 	kvm_rax_write(vcpu, regs->rax);
10262 	kvm_rbx_write(vcpu, regs->rbx);
10263 	kvm_rcx_write(vcpu, regs->rcx);
10264 	kvm_rdx_write(vcpu, regs->rdx);
10265 	kvm_rsi_write(vcpu, regs->rsi);
10266 	kvm_rdi_write(vcpu, regs->rdi);
10267 	kvm_rsp_write(vcpu, regs->rsp);
10268 	kvm_rbp_write(vcpu, regs->rbp);
10269 #ifdef CONFIG_X86_64
10270 	kvm_r8_write(vcpu, regs->r8);
10271 	kvm_r9_write(vcpu, regs->r9);
10272 	kvm_r10_write(vcpu, regs->r10);
10273 	kvm_r11_write(vcpu, regs->r11);
10274 	kvm_r12_write(vcpu, regs->r12);
10275 	kvm_r13_write(vcpu, regs->r13);
10276 	kvm_r14_write(vcpu, regs->r14);
10277 	kvm_r15_write(vcpu, regs->r15);
10278 #endif
10279 
10280 	kvm_rip_write(vcpu, regs->rip);
10281 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
10282 
10283 	vcpu->arch.exception.pending = false;
10284 
10285 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10286 }
10287 
10288 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10289 {
10290 	vcpu_load(vcpu);
10291 	__set_regs(vcpu, regs);
10292 	vcpu_put(vcpu);
10293 	return 0;
10294 }
10295 
10296 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
10297 {
10298 	struct kvm_segment cs;
10299 
10300 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10301 	*db = cs.db;
10302 	*l = cs.l;
10303 }
10304 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
10305 
10306 static void __get_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10307 {
10308 	struct desc_ptr dt;
10309 
10310 	if (vcpu->arch.guest_state_protected)
10311 		goto skip_protected_regs;
10312 
10313 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10314 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10315 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10316 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10317 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10318 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10319 
10320 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10321 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10322 
10323 	static_call(kvm_x86_get_idt)(vcpu, &dt);
10324 	sregs->idt.limit = dt.size;
10325 	sregs->idt.base = dt.address;
10326 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
10327 	sregs->gdt.limit = dt.size;
10328 	sregs->gdt.base = dt.address;
10329 
10330 	sregs->cr2 = vcpu->arch.cr2;
10331 	sregs->cr3 = kvm_read_cr3(vcpu);
10332 
10333 skip_protected_regs:
10334 	sregs->cr0 = kvm_read_cr0(vcpu);
10335 	sregs->cr4 = kvm_read_cr4(vcpu);
10336 	sregs->cr8 = kvm_get_cr8(vcpu);
10337 	sregs->efer = vcpu->arch.efer;
10338 	sregs->apic_base = kvm_get_apic_base(vcpu);
10339 }
10340 
10341 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10342 {
10343 	__get_sregs_common(vcpu, sregs);
10344 
10345 	if (vcpu->arch.guest_state_protected)
10346 		return;
10347 
10348 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
10349 		set_bit(vcpu->arch.interrupt.nr,
10350 			(unsigned long *)sregs->interrupt_bitmap);
10351 }
10352 
10353 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10354 {
10355 	int i;
10356 
10357 	__get_sregs_common(vcpu, (struct kvm_sregs *)sregs2);
10358 
10359 	if (vcpu->arch.guest_state_protected)
10360 		return;
10361 
10362 	if (is_pae_paging(vcpu)) {
10363 		for (i = 0 ; i < 4 ; i++)
10364 			sregs2->pdptrs[i] = kvm_pdptr_read(vcpu, i);
10365 		sregs2->flags |= KVM_SREGS2_FLAGS_PDPTRS_VALID;
10366 	}
10367 }
10368 
10369 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
10370 				  struct kvm_sregs *sregs)
10371 {
10372 	vcpu_load(vcpu);
10373 	__get_sregs(vcpu, sregs);
10374 	vcpu_put(vcpu);
10375 	return 0;
10376 }
10377 
10378 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
10379 				    struct kvm_mp_state *mp_state)
10380 {
10381 	int r;
10382 
10383 	vcpu_load(vcpu);
10384 	if (kvm_mpx_supported())
10385 		kvm_load_guest_fpu(vcpu);
10386 
10387 	r = kvm_apic_accept_events(vcpu);
10388 	if (r < 0)
10389 		goto out;
10390 	r = 0;
10391 
10392 	if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
10393 	     vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
10394 	    vcpu->arch.pv.pv_unhalted)
10395 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
10396 	else
10397 		mp_state->mp_state = vcpu->arch.mp_state;
10398 
10399 out:
10400 	if (kvm_mpx_supported())
10401 		kvm_put_guest_fpu(vcpu);
10402 	vcpu_put(vcpu);
10403 	return r;
10404 }
10405 
10406 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
10407 				    struct kvm_mp_state *mp_state)
10408 {
10409 	int ret = -EINVAL;
10410 
10411 	vcpu_load(vcpu);
10412 
10413 	if (!lapic_in_kernel(vcpu) &&
10414 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
10415 		goto out;
10416 
10417 	/*
10418 	 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
10419 	 * INIT state; latched init should be reported using
10420 	 * KVM_SET_VCPU_EVENTS, so reject it here.
10421 	 */
10422 	if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
10423 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
10424 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
10425 		goto out;
10426 
10427 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
10428 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
10429 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
10430 	} else
10431 		vcpu->arch.mp_state = mp_state->mp_state;
10432 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10433 
10434 	ret = 0;
10435 out:
10436 	vcpu_put(vcpu);
10437 	return ret;
10438 }
10439 
10440 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
10441 		    int reason, bool has_error_code, u32 error_code)
10442 {
10443 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
10444 	int ret;
10445 
10446 	init_emulate_ctxt(vcpu);
10447 
10448 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
10449 				   has_error_code, error_code);
10450 	if (ret) {
10451 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
10452 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
10453 		vcpu->run->internal.ndata = 0;
10454 		return 0;
10455 	}
10456 
10457 	kvm_rip_write(vcpu, ctxt->eip);
10458 	kvm_set_rflags(vcpu, ctxt->eflags);
10459 	return 1;
10460 }
10461 EXPORT_SYMBOL_GPL(kvm_task_switch);
10462 
10463 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10464 {
10465 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
10466 		/*
10467 		 * When EFER.LME and CR0.PG are set, the processor is in
10468 		 * 64-bit mode (though maybe in a 32-bit code segment).
10469 		 * CR4.PAE and EFER.LMA must be set.
10470 		 */
10471 		if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
10472 			return false;
10473 		if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
10474 			return false;
10475 	} else {
10476 		/*
10477 		 * Not in 64-bit mode: EFER.LMA is clear and the code
10478 		 * segment cannot be 64-bit.
10479 		 */
10480 		if (sregs->efer & EFER_LMA || sregs->cs.l)
10481 			return false;
10482 	}
10483 
10484 	return kvm_is_valid_cr4(vcpu, sregs->cr4);
10485 }
10486 
10487 static int __set_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs,
10488 		int *mmu_reset_needed, bool update_pdptrs)
10489 {
10490 	struct msr_data apic_base_msr;
10491 	int idx;
10492 	struct desc_ptr dt;
10493 
10494 	if (!kvm_is_valid_sregs(vcpu, sregs))
10495 		return -EINVAL;
10496 
10497 	apic_base_msr.data = sregs->apic_base;
10498 	apic_base_msr.host_initiated = true;
10499 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
10500 		return -EINVAL;
10501 
10502 	if (vcpu->arch.guest_state_protected)
10503 		return 0;
10504 
10505 	dt.size = sregs->idt.limit;
10506 	dt.address = sregs->idt.base;
10507 	static_call(kvm_x86_set_idt)(vcpu, &dt);
10508 	dt.size = sregs->gdt.limit;
10509 	dt.address = sregs->gdt.base;
10510 	static_call(kvm_x86_set_gdt)(vcpu, &dt);
10511 
10512 	vcpu->arch.cr2 = sregs->cr2;
10513 	*mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
10514 	vcpu->arch.cr3 = sregs->cr3;
10515 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
10516 
10517 	kvm_set_cr8(vcpu, sregs->cr8);
10518 
10519 	*mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
10520 	static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
10521 
10522 	*mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
10523 	static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
10524 	vcpu->arch.cr0 = sregs->cr0;
10525 
10526 	*mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
10527 	static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
10528 
10529 	if (update_pdptrs) {
10530 		idx = srcu_read_lock(&vcpu->kvm->srcu);
10531 		if (is_pae_paging(vcpu)) {
10532 			load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
10533 			*mmu_reset_needed = 1;
10534 		}
10535 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
10536 	}
10537 
10538 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10539 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10540 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10541 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10542 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10543 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10544 
10545 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10546 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10547 
10548 	update_cr8_intercept(vcpu);
10549 
10550 	/* Older userspace won't unhalt the vcpu on reset. */
10551 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
10552 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
10553 	    !is_protmode(vcpu))
10554 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10555 
10556 	return 0;
10557 }
10558 
10559 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10560 {
10561 	int pending_vec, max_bits;
10562 	int mmu_reset_needed = 0;
10563 	int ret = __set_sregs_common(vcpu, sregs, &mmu_reset_needed, true);
10564 
10565 	if (ret)
10566 		return ret;
10567 
10568 	if (mmu_reset_needed)
10569 		kvm_mmu_reset_context(vcpu);
10570 
10571 	max_bits = KVM_NR_INTERRUPTS;
10572 	pending_vec = find_first_bit(
10573 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
10574 
10575 	if (pending_vec < max_bits) {
10576 		kvm_queue_interrupt(vcpu, pending_vec, false);
10577 		pr_debug("Set back pending irq %d\n", pending_vec);
10578 		kvm_make_request(KVM_REQ_EVENT, vcpu);
10579 	}
10580 	return 0;
10581 }
10582 
10583 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10584 {
10585 	int mmu_reset_needed = 0;
10586 	bool valid_pdptrs = sregs2->flags & KVM_SREGS2_FLAGS_PDPTRS_VALID;
10587 	bool pae = (sregs2->cr0 & X86_CR0_PG) && (sregs2->cr4 & X86_CR4_PAE) &&
10588 		!(sregs2->efer & EFER_LMA);
10589 	int i, ret;
10590 
10591 	if (sregs2->flags & ~KVM_SREGS2_FLAGS_PDPTRS_VALID)
10592 		return -EINVAL;
10593 
10594 	if (valid_pdptrs && (!pae || vcpu->arch.guest_state_protected))
10595 		return -EINVAL;
10596 
10597 	ret = __set_sregs_common(vcpu, (struct kvm_sregs *)sregs2,
10598 				 &mmu_reset_needed, !valid_pdptrs);
10599 	if (ret)
10600 		return ret;
10601 
10602 	if (valid_pdptrs) {
10603 		for (i = 0; i < 4 ; i++)
10604 			kvm_pdptr_write(vcpu, i, sregs2->pdptrs[i]);
10605 
10606 		kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
10607 		mmu_reset_needed = 1;
10608 		vcpu->arch.pdptrs_from_userspace = true;
10609 	}
10610 	if (mmu_reset_needed)
10611 		kvm_mmu_reset_context(vcpu);
10612 	return 0;
10613 }
10614 
10615 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
10616 				  struct kvm_sregs *sregs)
10617 {
10618 	int ret;
10619 
10620 	vcpu_load(vcpu);
10621 	ret = __set_sregs(vcpu, sregs);
10622 	vcpu_put(vcpu);
10623 	return ret;
10624 }
10625 
10626 static void kvm_arch_vcpu_guestdbg_update_apicv_inhibit(struct kvm *kvm)
10627 {
10628 	bool inhibit = false;
10629 	struct kvm_vcpu *vcpu;
10630 	int i;
10631 
10632 	down_write(&kvm->arch.apicv_update_lock);
10633 
10634 	kvm_for_each_vcpu(i, vcpu, kvm) {
10635 		if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ) {
10636 			inhibit = true;
10637 			break;
10638 		}
10639 	}
10640 	__kvm_request_apicv_update(kvm, !inhibit, APICV_INHIBIT_REASON_BLOCKIRQ);
10641 	up_write(&kvm->arch.apicv_update_lock);
10642 }
10643 
10644 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
10645 					struct kvm_guest_debug *dbg)
10646 {
10647 	unsigned long rflags;
10648 	int i, r;
10649 
10650 	if (vcpu->arch.guest_state_protected)
10651 		return -EINVAL;
10652 
10653 	vcpu_load(vcpu);
10654 
10655 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10656 		r = -EBUSY;
10657 		if (vcpu->arch.exception.pending)
10658 			goto out;
10659 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10660 			kvm_queue_exception(vcpu, DB_VECTOR);
10661 		else
10662 			kvm_queue_exception(vcpu, BP_VECTOR);
10663 	}
10664 
10665 	/*
10666 	 * Read rflags as long as potentially injected trace flags are still
10667 	 * filtered out.
10668 	 */
10669 	rflags = kvm_get_rflags(vcpu);
10670 
10671 	vcpu->guest_debug = dbg->control;
10672 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10673 		vcpu->guest_debug = 0;
10674 
10675 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10676 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
10677 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10678 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10679 	} else {
10680 		for (i = 0; i < KVM_NR_DB_REGS; i++)
10681 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10682 	}
10683 	kvm_update_dr7(vcpu);
10684 
10685 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10686 		vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
10687 
10688 	/*
10689 	 * Trigger an rflags update that will inject or remove the trace
10690 	 * flags.
10691 	 */
10692 	kvm_set_rflags(vcpu, rflags);
10693 
10694 	static_call(kvm_x86_update_exception_bitmap)(vcpu);
10695 
10696 	kvm_arch_vcpu_guestdbg_update_apicv_inhibit(vcpu->kvm);
10697 
10698 	r = 0;
10699 
10700 out:
10701 	vcpu_put(vcpu);
10702 	return r;
10703 }
10704 
10705 /*
10706  * Translate a guest virtual address to a guest physical address.
10707  */
10708 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10709 				    struct kvm_translation *tr)
10710 {
10711 	unsigned long vaddr = tr->linear_address;
10712 	gpa_t gpa;
10713 	int idx;
10714 
10715 	vcpu_load(vcpu);
10716 
10717 	idx = srcu_read_lock(&vcpu->kvm->srcu);
10718 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10719 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
10720 	tr->physical_address = gpa;
10721 	tr->valid = gpa != UNMAPPED_GVA;
10722 	tr->writeable = 1;
10723 	tr->usermode = 0;
10724 
10725 	vcpu_put(vcpu);
10726 	return 0;
10727 }
10728 
10729 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10730 {
10731 	struct fxregs_state *fxsave;
10732 
10733 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
10734 		return 0;
10735 
10736 	vcpu_load(vcpu);
10737 
10738 	fxsave = &vcpu->arch.guest_fpu.fpstate->regs.fxsave;
10739 	memcpy(fpu->fpr, fxsave->st_space, 128);
10740 	fpu->fcw = fxsave->cwd;
10741 	fpu->fsw = fxsave->swd;
10742 	fpu->ftwx = fxsave->twd;
10743 	fpu->last_opcode = fxsave->fop;
10744 	fpu->last_ip = fxsave->rip;
10745 	fpu->last_dp = fxsave->rdp;
10746 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10747 
10748 	vcpu_put(vcpu);
10749 	return 0;
10750 }
10751 
10752 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10753 {
10754 	struct fxregs_state *fxsave;
10755 
10756 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
10757 		return 0;
10758 
10759 	vcpu_load(vcpu);
10760 
10761 	fxsave = &vcpu->arch.guest_fpu.fpstate->regs.fxsave;
10762 
10763 	memcpy(fxsave->st_space, fpu->fpr, 128);
10764 	fxsave->cwd = fpu->fcw;
10765 	fxsave->swd = fpu->fsw;
10766 	fxsave->twd = fpu->ftwx;
10767 	fxsave->fop = fpu->last_opcode;
10768 	fxsave->rip = fpu->last_ip;
10769 	fxsave->rdp = fpu->last_dp;
10770 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10771 
10772 	vcpu_put(vcpu);
10773 	return 0;
10774 }
10775 
10776 static void store_regs(struct kvm_vcpu *vcpu)
10777 {
10778 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10779 
10780 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10781 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
10782 
10783 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10784 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10785 
10786 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10787 		kvm_vcpu_ioctl_x86_get_vcpu_events(
10788 				vcpu, &vcpu->run->s.regs.events);
10789 }
10790 
10791 static int sync_regs(struct kvm_vcpu *vcpu)
10792 {
10793 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10794 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
10795 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10796 	}
10797 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10798 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10799 			return -EINVAL;
10800 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10801 	}
10802 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10803 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10804 				vcpu, &vcpu->run->s.regs.events))
10805 			return -EINVAL;
10806 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10807 	}
10808 
10809 	return 0;
10810 }
10811 
10812 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10813 {
10814 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10815 		pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10816 			     "guest TSC will not be reliable\n");
10817 
10818 	return 0;
10819 }
10820 
10821 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10822 {
10823 	struct page *page;
10824 	int r;
10825 
10826 	vcpu->arch.last_vmentry_cpu = -1;
10827 	vcpu->arch.regs_avail = ~0;
10828 	vcpu->arch.regs_dirty = ~0;
10829 
10830 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10831 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10832 	else
10833 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10834 
10835 	r = kvm_mmu_create(vcpu);
10836 	if (r < 0)
10837 		return r;
10838 
10839 	if (irqchip_in_kernel(vcpu->kvm)) {
10840 		r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10841 		if (r < 0)
10842 			goto fail_mmu_destroy;
10843 		if (kvm_apicv_activated(vcpu->kvm))
10844 			vcpu->arch.apicv_active = true;
10845 	} else
10846 		static_branch_inc(&kvm_has_noapic_vcpu);
10847 
10848 	r = -ENOMEM;
10849 
10850 	page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10851 	if (!page)
10852 		goto fail_free_lapic;
10853 	vcpu->arch.pio_data = page_address(page);
10854 
10855 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10856 				       GFP_KERNEL_ACCOUNT);
10857 	if (!vcpu->arch.mce_banks)
10858 		goto fail_free_pio_data;
10859 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
10860 
10861 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
10862 				GFP_KERNEL_ACCOUNT))
10863 		goto fail_free_mce_banks;
10864 
10865 	if (!alloc_emulate_ctxt(vcpu))
10866 		goto free_wbinvd_dirty_mask;
10867 
10868 	if (!fpu_alloc_guest_fpstate(&vcpu->arch.guest_fpu)) {
10869 		pr_err("kvm: failed to allocate vcpu's fpu\n");
10870 		goto free_emulate_ctxt;
10871 	}
10872 
10873 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10874 	vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
10875 
10876 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10877 
10878 	kvm_async_pf_hash_reset(vcpu);
10879 	kvm_pmu_init(vcpu);
10880 
10881 	vcpu->arch.pending_external_vector = -1;
10882 	vcpu->arch.preempted_in_kernel = false;
10883 
10884 #if IS_ENABLED(CONFIG_HYPERV)
10885 	vcpu->arch.hv_root_tdp = INVALID_PAGE;
10886 #endif
10887 
10888 	r = static_call(kvm_x86_vcpu_create)(vcpu);
10889 	if (r)
10890 		goto free_guest_fpu;
10891 
10892 	vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10893 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10894 	kvm_vcpu_mtrr_init(vcpu);
10895 	vcpu_load(vcpu);
10896 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
10897 	kvm_vcpu_reset(vcpu, false);
10898 	kvm_init_mmu(vcpu);
10899 	vcpu_put(vcpu);
10900 	return 0;
10901 
10902 free_guest_fpu:
10903 	fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
10904 free_emulate_ctxt:
10905 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10906 free_wbinvd_dirty_mask:
10907 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10908 fail_free_mce_banks:
10909 	kfree(vcpu->arch.mce_banks);
10910 fail_free_pio_data:
10911 	free_page((unsigned long)vcpu->arch.pio_data);
10912 fail_free_lapic:
10913 	kvm_free_lapic(vcpu);
10914 fail_mmu_destroy:
10915 	kvm_mmu_destroy(vcpu);
10916 	return r;
10917 }
10918 
10919 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10920 {
10921 	struct kvm *kvm = vcpu->kvm;
10922 
10923 	if (mutex_lock_killable(&vcpu->mutex))
10924 		return;
10925 	vcpu_load(vcpu);
10926 	kvm_synchronize_tsc(vcpu, 0);
10927 	vcpu_put(vcpu);
10928 
10929 	/* poll control enabled by default */
10930 	vcpu->arch.msr_kvm_poll_control = 1;
10931 
10932 	mutex_unlock(&vcpu->mutex);
10933 
10934 	if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10935 		schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10936 						KVMCLOCK_SYNC_PERIOD);
10937 }
10938 
10939 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10940 {
10941 	int idx;
10942 
10943 	kvmclock_reset(vcpu);
10944 
10945 	static_call(kvm_x86_vcpu_free)(vcpu);
10946 
10947 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10948 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10949 	fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
10950 
10951 	kvm_hv_vcpu_uninit(vcpu);
10952 	kvm_pmu_destroy(vcpu);
10953 	kfree(vcpu->arch.mce_banks);
10954 	kvm_free_lapic(vcpu);
10955 	idx = srcu_read_lock(&vcpu->kvm->srcu);
10956 	kvm_mmu_destroy(vcpu);
10957 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
10958 	free_page((unsigned long)vcpu->arch.pio_data);
10959 	kvfree(vcpu->arch.cpuid_entries);
10960 	if (!lapic_in_kernel(vcpu))
10961 		static_branch_dec(&kvm_has_noapic_vcpu);
10962 }
10963 
10964 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10965 {
10966 	struct kvm_cpuid_entry2 *cpuid_0x1;
10967 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
10968 	unsigned long new_cr0;
10969 
10970 	/*
10971 	 * Several of the "set" flows, e.g. ->set_cr0(), read other registers
10972 	 * to handle side effects.  RESET emulation hits those flows and relies
10973 	 * on emulated/virtualized registers, including those that are loaded
10974 	 * into hardware, to be zeroed at vCPU creation.  Use CRs as a sentinel
10975 	 * to detect improper or missing initialization.
10976 	 */
10977 	WARN_ON_ONCE(!init_event &&
10978 		     (old_cr0 || kvm_read_cr3(vcpu) || kvm_read_cr4(vcpu)));
10979 
10980 	kvm_lapic_reset(vcpu, init_event);
10981 
10982 	vcpu->arch.hflags = 0;
10983 
10984 	vcpu->arch.smi_pending = 0;
10985 	vcpu->arch.smi_count = 0;
10986 	atomic_set(&vcpu->arch.nmi_queued, 0);
10987 	vcpu->arch.nmi_pending = 0;
10988 	vcpu->arch.nmi_injected = false;
10989 	kvm_clear_interrupt_queue(vcpu);
10990 	kvm_clear_exception_queue(vcpu);
10991 
10992 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10993 	kvm_update_dr0123(vcpu);
10994 	vcpu->arch.dr6 = DR6_ACTIVE_LOW;
10995 	vcpu->arch.dr7 = DR7_FIXED_1;
10996 	kvm_update_dr7(vcpu);
10997 
10998 	vcpu->arch.cr2 = 0;
10999 
11000 	kvm_make_request(KVM_REQ_EVENT, vcpu);
11001 	vcpu->arch.apf.msr_en_val = 0;
11002 	vcpu->arch.apf.msr_int_val = 0;
11003 	vcpu->arch.st.msr_val = 0;
11004 
11005 	kvmclock_reset(vcpu);
11006 
11007 	kvm_clear_async_pf_completion_queue(vcpu);
11008 	kvm_async_pf_hash_reset(vcpu);
11009 	vcpu->arch.apf.halted = false;
11010 
11011 	if (vcpu->arch.guest_fpu.fpstate && kvm_mpx_supported()) {
11012 		struct fpstate *fpstate = vcpu->arch.guest_fpu.fpstate;
11013 
11014 		/*
11015 		 * To avoid have the INIT path from kvm_apic_has_events() that be
11016 		 * called with loaded FPU and does not let userspace fix the state.
11017 		 */
11018 		if (init_event)
11019 			kvm_put_guest_fpu(vcpu);
11020 
11021 		fpstate_clear_xstate_component(fpstate, XFEATURE_BNDREGS);
11022 		fpstate_clear_xstate_component(fpstate, XFEATURE_BNDCSR);
11023 
11024 		if (init_event)
11025 			kvm_load_guest_fpu(vcpu);
11026 	}
11027 
11028 	if (!init_event) {
11029 		kvm_pmu_reset(vcpu);
11030 		vcpu->arch.smbase = 0x30000;
11031 
11032 		vcpu->arch.msr_misc_features_enables = 0;
11033 
11034 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
11035 	}
11036 
11037 	/* All GPRs except RDX (handled below) are zeroed on RESET/INIT. */
11038 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
11039 	kvm_register_mark_dirty(vcpu, VCPU_REGS_RSP);
11040 
11041 	/*
11042 	 * Fall back to KVM's default Family/Model/Stepping of 0x600 (P6/Athlon)
11043 	 * if no CPUID match is found.  Note, it's impossible to get a match at
11044 	 * RESET since KVM emulates RESET before exposing the vCPU to userspace,
11045 	 * i.e. it's impossible for kvm_find_cpuid_entry() to find a valid entry
11046 	 * on RESET.  But, go through the motions in case that's ever remedied.
11047 	 */
11048 	cpuid_0x1 = kvm_find_cpuid_entry(vcpu, 1, 0);
11049 	kvm_rdx_write(vcpu, cpuid_0x1 ? cpuid_0x1->eax : 0x600);
11050 
11051 	vcpu->arch.ia32_xss = 0;
11052 
11053 	static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
11054 
11055 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
11056 	kvm_rip_write(vcpu, 0xfff0);
11057 
11058 	vcpu->arch.cr3 = 0;
11059 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
11060 
11061 	/*
11062 	 * CR0.CD/NW are set on RESET, preserved on INIT.  Note, some versions
11063 	 * of Intel's SDM list CD/NW as being set on INIT, but they contradict
11064 	 * (or qualify) that with a footnote stating that CD/NW are preserved.
11065 	 */
11066 	new_cr0 = X86_CR0_ET;
11067 	if (init_event)
11068 		new_cr0 |= (old_cr0 & (X86_CR0_NW | X86_CR0_CD));
11069 	else
11070 		new_cr0 |= X86_CR0_NW | X86_CR0_CD;
11071 
11072 	static_call(kvm_x86_set_cr0)(vcpu, new_cr0);
11073 	static_call(kvm_x86_set_cr4)(vcpu, 0);
11074 	static_call(kvm_x86_set_efer)(vcpu, 0);
11075 	static_call(kvm_x86_update_exception_bitmap)(vcpu);
11076 
11077 	/*
11078 	 * Reset the MMU context if paging was enabled prior to INIT (which is
11079 	 * implied if CR0.PG=1 as CR0 will be '0' prior to RESET).  Unlike the
11080 	 * standard CR0/CR4/EFER modification paths, only CR0.PG needs to be
11081 	 * checked because it is unconditionally cleared on INIT and all other
11082 	 * paging related bits are ignored if paging is disabled, i.e. CR0.WP,
11083 	 * CR4, and EFER changes are all irrelevant if CR0.PG was '0'.
11084 	 */
11085 	if (old_cr0 & X86_CR0_PG)
11086 		kvm_mmu_reset_context(vcpu);
11087 
11088 	/*
11089 	 * Intel's SDM states that all TLB entries are flushed on INIT.  AMD's
11090 	 * APM states the TLBs are untouched by INIT, but it also states that
11091 	 * the TLBs are flushed on "External initialization of the processor."
11092 	 * Flush the guest TLB regardless of vendor, there is no meaningful
11093 	 * benefit in relying on the guest to flush the TLB immediately after
11094 	 * INIT.  A spurious TLB flush is benign and likely negligible from a
11095 	 * performance perspective.
11096 	 */
11097 	if (init_event)
11098 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
11099 }
11100 EXPORT_SYMBOL_GPL(kvm_vcpu_reset);
11101 
11102 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
11103 {
11104 	struct kvm_segment cs;
11105 
11106 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
11107 	cs.selector = vector << 8;
11108 	cs.base = vector << 12;
11109 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
11110 	kvm_rip_write(vcpu, 0);
11111 }
11112 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
11113 
11114 int kvm_arch_hardware_enable(void)
11115 {
11116 	struct kvm *kvm;
11117 	struct kvm_vcpu *vcpu;
11118 	int i;
11119 	int ret;
11120 	u64 local_tsc;
11121 	u64 max_tsc = 0;
11122 	bool stable, backwards_tsc = false;
11123 
11124 	kvm_user_return_msr_cpu_online();
11125 	ret = static_call(kvm_x86_hardware_enable)();
11126 	if (ret != 0)
11127 		return ret;
11128 
11129 	local_tsc = rdtsc();
11130 	stable = !kvm_check_tsc_unstable();
11131 	list_for_each_entry(kvm, &vm_list, vm_list) {
11132 		kvm_for_each_vcpu(i, vcpu, kvm) {
11133 			if (!stable && vcpu->cpu == smp_processor_id())
11134 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
11135 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
11136 				backwards_tsc = true;
11137 				if (vcpu->arch.last_host_tsc > max_tsc)
11138 					max_tsc = vcpu->arch.last_host_tsc;
11139 			}
11140 		}
11141 	}
11142 
11143 	/*
11144 	 * Sometimes, even reliable TSCs go backwards.  This happens on
11145 	 * platforms that reset TSC during suspend or hibernate actions, but
11146 	 * maintain synchronization.  We must compensate.  Fortunately, we can
11147 	 * detect that condition here, which happens early in CPU bringup,
11148 	 * before any KVM threads can be running.  Unfortunately, we can't
11149 	 * bring the TSCs fully up to date with real time, as we aren't yet far
11150 	 * enough into CPU bringup that we know how much real time has actually
11151 	 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
11152 	 * variables that haven't been updated yet.
11153 	 *
11154 	 * So we simply find the maximum observed TSC above, then record the
11155 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
11156 	 * the adjustment will be applied.  Note that we accumulate
11157 	 * adjustments, in case multiple suspend cycles happen before some VCPU
11158 	 * gets a chance to run again.  In the event that no KVM threads get a
11159 	 * chance to run, we will miss the entire elapsed period, as we'll have
11160 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
11161 	 * loose cycle time.  This isn't too big a deal, since the loss will be
11162 	 * uniform across all VCPUs (not to mention the scenario is extremely
11163 	 * unlikely). It is possible that a second hibernate recovery happens
11164 	 * much faster than a first, causing the observed TSC here to be
11165 	 * smaller; this would require additional padding adjustment, which is
11166 	 * why we set last_host_tsc to the local tsc observed here.
11167 	 *
11168 	 * N.B. - this code below runs only on platforms with reliable TSC,
11169 	 * as that is the only way backwards_tsc is set above.  Also note
11170 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
11171 	 * have the same delta_cyc adjustment applied if backwards_tsc
11172 	 * is detected.  Note further, this adjustment is only done once,
11173 	 * as we reset last_host_tsc on all VCPUs to stop this from being
11174 	 * called multiple times (one for each physical CPU bringup).
11175 	 *
11176 	 * Platforms with unreliable TSCs don't have to deal with this, they
11177 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
11178 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
11179 	 * guarantee that they stay in perfect synchronization.
11180 	 */
11181 	if (backwards_tsc) {
11182 		u64 delta_cyc = max_tsc - local_tsc;
11183 		list_for_each_entry(kvm, &vm_list, vm_list) {
11184 			kvm->arch.backwards_tsc_observed = true;
11185 			kvm_for_each_vcpu(i, vcpu, kvm) {
11186 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
11187 				vcpu->arch.last_host_tsc = local_tsc;
11188 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
11189 			}
11190 
11191 			/*
11192 			 * We have to disable TSC offset matching.. if you were
11193 			 * booting a VM while issuing an S4 host suspend....
11194 			 * you may have some problem.  Solving this issue is
11195 			 * left as an exercise to the reader.
11196 			 */
11197 			kvm->arch.last_tsc_nsec = 0;
11198 			kvm->arch.last_tsc_write = 0;
11199 		}
11200 
11201 	}
11202 	return 0;
11203 }
11204 
11205 void kvm_arch_hardware_disable(void)
11206 {
11207 	static_call(kvm_x86_hardware_disable)();
11208 	drop_user_return_notifiers();
11209 }
11210 
11211 int kvm_arch_hardware_setup(void *opaque)
11212 {
11213 	struct kvm_x86_init_ops *ops = opaque;
11214 	int r;
11215 
11216 	rdmsrl_safe(MSR_EFER, &host_efer);
11217 
11218 	if (boot_cpu_has(X86_FEATURE_XSAVES))
11219 		rdmsrl(MSR_IA32_XSS, host_xss);
11220 
11221 	r = ops->hardware_setup();
11222 	if (r != 0)
11223 		return r;
11224 
11225 	memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
11226 	kvm_ops_static_call_update();
11227 
11228 	if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
11229 		supported_xss = 0;
11230 
11231 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
11232 	cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
11233 #undef __kvm_cpu_cap_has
11234 
11235 	if (kvm_has_tsc_control) {
11236 		/*
11237 		 * Make sure the user can only configure tsc_khz values that
11238 		 * fit into a signed integer.
11239 		 * A min value is not calculated because it will always
11240 		 * be 1 on all machines.
11241 		 */
11242 		u64 max = min(0x7fffffffULL,
11243 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
11244 		kvm_max_guest_tsc_khz = max;
11245 
11246 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
11247 	}
11248 
11249 	kvm_init_msr_list();
11250 	return 0;
11251 }
11252 
11253 void kvm_arch_hardware_unsetup(void)
11254 {
11255 	static_call(kvm_x86_hardware_unsetup)();
11256 }
11257 
11258 int kvm_arch_check_processor_compat(void *opaque)
11259 {
11260 	struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
11261 	struct kvm_x86_init_ops *ops = opaque;
11262 
11263 	WARN_ON(!irqs_disabled());
11264 
11265 	if (__cr4_reserved_bits(cpu_has, c) !=
11266 	    __cr4_reserved_bits(cpu_has, &boot_cpu_data))
11267 		return -EIO;
11268 
11269 	return ops->check_processor_compatibility();
11270 }
11271 
11272 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
11273 {
11274 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
11275 }
11276 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
11277 
11278 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
11279 {
11280 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
11281 }
11282 
11283 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
11284 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
11285 
11286 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
11287 {
11288 	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
11289 
11290 	vcpu->arch.l1tf_flush_l1d = true;
11291 	if (pmu->version && unlikely(pmu->event_count)) {
11292 		pmu->need_cleanup = true;
11293 		kvm_make_request(KVM_REQ_PMU, vcpu);
11294 	}
11295 	static_call(kvm_x86_sched_in)(vcpu, cpu);
11296 }
11297 
11298 void kvm_arch_free_vm(struct kvm *kvm)
11299 {
11300 	kfree(to_kvm_hv(kvm)->hv_pa_pg);
11301 	__kvm_arch_free_vm(kvm);
11302 }
11303 
11304 
11305 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
11306 {
11307 	int ret;
11308 	unsigned long flags;
11309 
11310 	if (type)
11311 		return -EINVAL;
11312 
11313 	ret = kvm_page_track_init(kvm);
11314 	if (ret)
11315 		return ret;
11316 
11317 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
11318 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
11319 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
11320 	INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
11321 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
11322 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
11323 
11324 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
11325 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
11326 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
11327 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
11328 		&kvm->arch.irq_sources_bitmap);
11329 
11330 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
11331 	mutex_init(&kvm->arch.apic_map_lock);
11332 	seqcount_raw_spinlock_init(&kvm->arch.pvclock_sc, &kvm->arch.tsc_write_lock);
11333 	kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
11334 
11335 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
11336 	pvclock_update_vm_gtod_copy(kvm);
11337 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
11338 
11339 	kvm->arch.guest_can_read_msr_platform_info = true;
11340 
11341 #if IS_ENABLED(CONFIG_HYPERV)
11342 	spin_lock_init(&kvm->arch.hv_root_tdp_lock);
11343 	kvm->arch.hv_root_tdp = INVALID_PAGE;
11344 #endif
11345 
11346 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
11347 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
11348 
11349 	kvm_apicv_init(kvm);
11350 	kvm_hv_init_vm(kvm);
11351 	kvm_mmu_init_vm(kvm);
11352 	kvm_xen_init_vm(kvm);
11353 
11354 	return static_call(kvm_x86_vm_init)(kvm);
11355 }
11356 
11357 int kvm_arch_post_init_vm(struct kvm *kvm)
11358 {
11359 	return kvm_mmu_post_init_vm(kvm);
11360 }
11361 
11362 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
11363 {
11364 	vcpu_load(vcpu);
11365 	kvm_mmu_unload(vcpu);
11366 	vcpu_put(vcpu);
11367 }
11368 
11369 static void kvm_free_vcpus(struct kvm *kvm)
11370 {
11371 	unsigned int i;
11372 	struct kvm_vcpu *vcpu;
11373 
11374 	/*
11375 	 * Unpin any mmu pages first.
11376 	 */
11377 	kvm_for_each_vcpu(i, vcpu, kvm) {
11378 		kvm_clear_async_pf_completion_queue(vcpu);
11379 		kvm_unload_vcpu_mmu(vcpu);
11380 	}
11381 	kvm_for_each_vcpu(i, vcpu, kvm)
11382 		kvm_vcpu_destroy(vcpu);
11383 
11384 	mutex_lock(&kvm->lock);
11385 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
11386 		kvm->vcpus[i] = NULL;
11387 
11388 	atomic_set(&kvm->online_vcpus, 0);
11389 	mutex_unlock(&kvm->lock);
11390 }
11391 
11392 void kvm_arch_sync_events(struct kvm *kvm)
11393 {
11394 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
11395 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
11396 	kvm_free_pit(kvm);
11397 }
11398 
11399 #define  ERR_PTR_USR(e)  ((void __user *)ERR_PTR(e))
11400 
11401 /**
11402  * __x86_set_memory_region: Setup KVM internal memory slot
11403  *
11404  * @kvm: the kvm pointer to the VM.
11405  * @id: the slot ID to setup.
11406  * @gpa: the GPA to install the slot (unused when @size == 0).
11407  * @size: the size of the slot. Set to zero to uninstall a slot.
11408  *
11409  * This function helps to setup a KVM internal memory slot.  Specify
11410  * @size > 0 to install a new slot, while @size == 0 to uninstall a
11411  * slot.  The return code can be one of the following:
11412  *
11413  *   HVA:           on success (uninstall will return a bogus HVA)
11414  *   -errno:        on error
11415  *
11416  * The caller should always use IS_ERR() to check the return value
11417  * before use.  Note, the KVM internal memory slots are guaranteed to
11418  * remain valid and unchanged until the VM is destroyed, i.e., the
11419  * GPA->HVA translation will not change.  However, the HVA is a user
11420  * address, i.e. its accessibility is not guaranteed, and must be
11421  * accessed via __copy_{to,from}_user().
11422  */
11423 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
11424 				      u32 size)
11425 {
11426 	int i, r;
11427 	unsigned long hva, old_npages;
11428 	struct kvm_memslots *slots = kvm_memslots(kvm);
11429 	struct kvm_memory_slot *slot;
11430 
11431 	/* Called with kvm->slots_lock held.  */
11432 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
11433 		return ERR_PTR_USR(-EINVAL);
11434 
11435 	slot = id_to_memslot(slots, id);
11436 	if (size) {
11437 		if (slot && slot->npages)
11438 			return ERR_PTR_USR(-EEXIST);
11439 
11440 		/*
11441 		 * MAP_SHARED to prevent internal slot pages from being moved
11442 		 * by fork()/COW.
11443 		 */
11444 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
11445 			      MAP_SHARED | MAP_ANONYMOUS, 0);
11446 		if (IS_ERR((void *)hva))
11447 			return (void __user *)hva;
11448 	} else {
11449 		if (!slot || !slot->npages)
11450 			return NULL;
11451 
11452 		old_npages = slot->npages;
11453 		hva = slot->userspace_addr;
11454 	}
11455 
11456 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11457 		struct kvm_userspace_memory_region m;
11458 
11459 		m.slot = id | (i << 16);
11460 		m.flags = 0;
11461 		m.guest_phys_addr = gpa;
11462 		m.userspace_addr = hva;
11463 		m.memory_size = size;
11464 		r = __kvm_set_memory_region(kvm, &m);
11465 		if (r < 0)
11466 			return ERR_PTR_USR(r);
11467 	}
11468 
11469 	if (!size)
11470 		vm_munmap(hva, old_npages * PAGE_SIZE);
11471 
11472 	return (void __user *)hva;
11473 }
11474 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
11475 
11476 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
11477 {
11478 	kvm_mmu_pre_destroy_vm(kvm);
11479 }
11480 
11481 void kvm_arch_destroy_vm(struct kvm *kvm)
11482 {
11483 	if (current->mm == kvm->mm) {
11484 		/*
11485 		 * Free memory regions allocated on behalf of userspace,
11486 		 * unless the the memory map has changed due to process exit
11487 		 * or fd copying.
11488 		 */
11489 		mutex_lock(&kvm->slots_lock);
11490 		__x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
11491 					0, 0);
11492 		__x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
11493 					0, 0);
11494 		__x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
11495 		mutex_unlock(&kvm->slots_lock);
11496 	}
11497 	static_call_cond(kvm_x86_vm_destroy)(kvm);
11498 	kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
11499 	kvm_pic_destroy(kvm);
11500 	kvm_ioapic_destroy(kvm);
11501 	kvm_free_vcpus(kvm);
11502 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
11503 	kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
11504 	kvm_mmu_uninit_vm(kvm);
11505 	kvm_page_track_cleanup(kvm);
11506 	kvm_xen_destroy_vm(kvm);
11507 	kvm_hv_destroy_vm(kvm);
11508 }
11509 
11510 static void memslot_rmap_free(struct kvm_memory_slot *slot)
11511 {
11512 	int i;
11513 
11514 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11515 		kvfree(slot->arch.rmap[i]);
11516 		slot->arch.rmap[i] = NULL;
11517 	}
11518 }
11519 
11520 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
11521 {
11522 	int i;
11523 
11524 	memslot_rmap_free(slot);
11525 
11526 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11527 		kvfree(slot->arch.lpage_info[i - 1]);
11528 		slot->arch.lpage_info[i - 1] = NULL;
11529 	}
11530 
11531 	kvm_page_track_free_memslot(slot);
11532 }
11533 
11534 int memslot_rmap_alloc(struct kvm_memory_slot *slot, unsigned long npages)
11535 {
11536 	const int sz = sizeof(*slot->arch.rmap[0]);
11537 	int i;
11538 
11539 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11540 		int level = i + 1;
11541 		int lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11542 
11543 		if (slot->arch.rmap[i])
11544 			continue;
11545 
11546 		slot->arch.rmap[i] = kvcalloc(lpages, sz, GFP_KERNEL_ACCOUNT);
11547 		if (!slot->arch.rmap[i]) {
11548 			memslot_rmap_free(slot);
11549 			return -ENOMEM;
11550 		}
11551 	}
11552 
11553 	return 0;
11554 }
11555 
11556 static int kvm_alloc_memslot_metadata(struct kvm *kvm,
11557 				      struct kvm_memory_slot *slot,
11558 				      unsigned long npages)
11559 {
11560 	int i, r;
11561 
11562 	/*
11563 	 * Clear out the previous array pointers for the KVM_MR_MOVE case.  The
11564 	 * old arrays will be freed by __kvm_set_memory_region() if installing
11565 	 * the new memslot is successful.
11566 	 */
11567 	memset(&slot->arch, 0, sizeof(slot->arch));
11568 
11569 	if (kvm_memslots_have_rmaps(kvm)) {
11570 		r = memslot_rmap_alloc(slot, npages);
11571 		if (r)
11572 			return r;
11573 	}
11574 
11575 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11576 		struct kvm_lpage_info *linfo;
11577 		unsigned long ugfn;
11578 		int lpages;
11579 		int level = i + 1;
11580 
11581 		lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11582 
11583 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
11584 		if (!linfo)
11585 			goto out_free;
11586 
11587 		slot->arch.lpage_info[i - 1] = linfo;
11588 
11589 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
11590 			linfo[0].disallow_lpage = 1;
11591 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
11592 			linfo[lpages - 1].disallow_lpage = 1;
11593 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
11594 		/*
11595 		 * If the gfn and userspace address are not aligned wrt each
11596 		 * other, disable large page support for this slot.
11597 		 */
11598 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
11599 			unsigned long j;
11600 
11601 			for (j = 0; j < lpages; ++j)
11602 				linfo[j].disallow_lpage = 1;
11603 		}
11604 	}
11605 
11606 	if (kvm_page_track_create_memslot(kvm, slot, npages))
11607 		goto out_free;
11608 
11609 	return 0;
11610 
11611 out_free:
11612 	memslot_rmap_free(slot);
11613 
11614 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11615 		kvfree(slot->arch.lpage_info[i - 1]);
11616 		slot->arch.lpage_info[i - 1] = NULL;
11617 	}
11618 	return -ENOMEM;
11619 }
11620 
11621 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
11622 {
11623 	struct kvm_vcpu *vcpu;
11624 	int i;
11625 
11626 	/*
11627 	 * memslots->generation has been incremented.
11628 	 * mmio generation may have reached its maximum value.
11629 	 */
11630 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
11631 
11632 	/* Force re-initialization of steal_time cache */
11633 	kvm_for_each_vcpu(i, vcpu, kvm)
11634 		kvm_vcpu_kick(vcpu);
11635 }
11636 
11637 int kvm_arch_prepare_memory_region(struct kvm *kvm,
11638 				struct kvm_memory_slot *memslot,
11639 				const struct kvm_userspace_memory_region *mem,
11640 				enum kvm_mr_change change)
11641 {
11642 	if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
11643 		return kvm_alloc_memslot_metadata(kvm, memslot,
11644 						  mem->memory_size >> PAGE_SHIFT);
11645 	return 0;
11646 }
11647 
11648 
11649 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
11650 {
11651 	struct kvm_arch *ka = &kvm->arch;
11652 
11653 	if (!kvm_x86_ops.cpu_dirty_log_size)
11654 		return;
11655 
11656 	if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
11657 	    (!enable && --ka->cpu_dirty_logging_count == 0))
11658 		kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
11659 
11660 	WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
11661 }
11662 
11663 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
11664 				     struct kvm_memory_slot *old,
11665 				     const struct kvm_memory_slot *new,
11666 				     enum kvm_mr_change change)
11667 {
11668 	bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES;
11669 
11670 	/*
11671 	 * Update CPU dirty logging if dirty logging is being toggled.  This
11672 	 * applies to all operations.
11673 	 */
11674 	if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)
11675 		kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
11676 
11677 	/*
11678 	 * Nothing more to do for RO slots (which can't be dirtied and can't be
11679 	 * made writable) or CREATE/MOVE/DELETE of a slot.
11680 	 *
11681 	 * For a memslot with dirty logging disabled:
11682 	 * CREATE:      No dirty mappings will already exist.
11683 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
11684 	 *		kvm_arch_flush_shadow_memslot()
11685 	 *
11686 	 * For a memslot with dirty logging enabled:
11687 	 * CREATE:      No shadow pages exist, thus nothing to write-protect
11688 	 *		and no dirty bits to clear.
11689 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
11690 	 *		kvm_arch_flush_shadow_memslot().
11691 	 */
11692 	if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
11693 		return;
11694 
11695 	/*
11696 	 * READONLY and non-flags changes were filtered out above, and the only
11697 	 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
11698 	 * logging isn't being toggled on or off.
11699 	 */
11700 	if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)))
11701 		return;
11702 
11703 	if (!log_dirty_pages) {
11704 		/*
11705 		 * Dirty logging tracks sptes in 4k granularity, meaning that
11706 		 * large sptes have to be split.  If live migration succeeds,
11707 		 * the guest in the source machine will be destroyed and large
11708 		 * sptes will be created in the destination.  However, if the
11709 		 * guest continues to run in the source machine (for example if
11710 		 * live migration fails), small sptes will remain around and
11711 		 * cause bad performance.
11712 		 *
11713 		 * Scan sptes if dirty logging has been stopped, dropping those
11714 		 * which can be collapsed into a single large-page spte.  Later
11715 		 * page faults will create the large-page sptes.
11716 		 */
11717 		kvm_mmu_zap_collapsible_sptes(kvm, new);
11718 	} else {
11719 		/*
11720 		 * Initially-all-set does not require write protecting any page,
11721 		 * because they're all assumed to be dirty.
11722 		 */
11723 		if (kvm_dirty_log_manual_protect_and_init_set(kvm))
11724 			return;
11725 
11726 		if (kvm_x86_ops.cpu_dirty_log_size) {
11727 			kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11728 			kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M);
11729 		} else {
11730 			kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
11731 		}
11732 	}
11733 }
11734 
11735 void kvm_arch_commit_memory_region(struct kvm *kvm,
11736 				const struct kvm_userspace_memory_region *mem,
11737 				struct kvm_memory_slot *old,
11738 				const struct kvm_memory_slot *new,
11739 				enum kvm_mr_change change)
11740 {
11741 	if (!kvm->arch.n_requested_mmu_pages)
11742 		kvm_mmu_change_mmu_pages(kvm,
11743 				kvm_mmu_calculate_default_mmu_pages(kvm));
11744 
11745 	kvm_mmu_slot_apply_flags(kvm, old, new, change);
11746 
11747 	/* Free the arrays associated with the old memslot. */
11748 	if (change == KVM_MR_MOVE)
11749 		kvm_arch_free_memslot(kvm, old);
11750 }
11751 
11752 void kvm_arch_flush_shadow_all(struct kvm *kvm)
11753 {
11754 	kvm_mmu_zap_all(kvm);
11755 }
11756 
11757 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
11758 				   struct kvm_memory_slot *slot)
11759 {
11760 	kvm_page_track_flush_slot(kvm, slot);
11761 }
11762 
11763 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
11764 {
11765 	return (is_guest_mode(vcpu) &&
11766 			kvm_x86_ops.guest_apic_has_interrupt &&
11767 			static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
11768 }
11769 
11770 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
11771 {
11772 	if (!list_empty_careful(&vcpu->async_pf.done))
11773 		return true;
11774 
11775 	if (kvm_apic_has_events(vcpu))
11776 		return true;
11777 
11778 	if (vcpu->arch.pv.pv_unhalted)
11779 		return true;
11780 
11781 	if (vcpu->arch.exception.pending)
11782 		return true;
11783 
11784 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11785 	    (vcpu->arch.nmi_pending &&
11786 	     static_call(kvm_x86_nmi_allowed)(vcpu, false)))
11787 		return true;
11788 
11789 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
11790 	    (vcpu->arch.smi_pending &&
11791 	     static_call(kvm_x86_smi_allowed)(vcpu, false)))
11792 		return true;
11793 
11794 	if (kvm_arch_interrupt_allowed(vcpu) &&
11795 	    (kvm_cpu_has_interrupt(vcpu) ||
11796 	    kvm_guest_apic_has_interrupt(vcpu)))
11797 		return true;
11798 
11799 	if (kvm_hv_has_stimer_pending(vcpu))
11800 		return true;
11801 
11802 	if (is_guest_mode(vcpu) &&
11803 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
11804 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
11805 		return true;
11806 
11807 	return false;
11808 }
11809 
11810 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
11811 {
11812 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
11813 }
11814 
11815 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
11816 {
11817 	if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11818 		return true;
11819 
11820 	return false;
11821 }
11822 
11823 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
11824 {
11825 	if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
11826 		return true;
11827 
11828 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11829 		kvm_test_request(KVM_REQ_SMI, vcpu) ||
11830 		 kvm_test_request(KVM_REQ_EVENT, vcpu))
11831 		return true;
11832 
11833 	return kvm_arch_dy_has_pending_interrupt(vcpu);
11834 }
11835 
11836 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
11837 {
11838 	if (vcpu->arch.guest_state_protected)
11839 		return true;
11840 
11841 	return vcpu->arch.preempted_in_kernel;
11842 }
11843 
11844 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
11845 {
11846 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
11847 }
11848 
11849 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
11850 {
11851 	return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
11852 }
11853 
11854 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
11855 {
11856 	/* Can't read the RIP when guest state is protected, just return 0 */
11857 	if (vcpu->arch.guest_state_protected)
11858 		return 0;
11859 
11860 	if (is_64_bit_mode(vcpu))
11861 		return kvm_rip_read(vcpu);
11862 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
11863 		     kvm_rip_read(vcpu));
11864 }
11865 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
11866 
11867 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
11868 {
11869 	return kvm_get_linear_rip(vcpu) == linear_rip;
11870 }
11871 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
11872 
11873 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
11874 {
11875 	unsigned long rflags;
11876 
11877 	rflags = static_call(kvm_x86_get_rflags)(vcpu);
11878 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11879 		rflags &= ~X86_EFLAGS_TF;
11880 	return rflags;
11881 }
11882 EXPORT_SYMBOL_GPL(kvm_get_rflags);
11883 
11884 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11885 {
11886 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
11887 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
11888 		rflags |= X86_EFLAGS_TF;
11889 	static_call(kvm_x86_set_rflags)(vcpu, rflags);
11890 }
11891 
11892 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11893 {
11894 	__kvm_set_rflags(vcpu, rflags);
11895 	kvm_make_request(KVM_REQ_EVENT, vcpu);
11896 }
11897 EXPORT_SYMBOL_GPL(kvm_set_rflags);
11898 
11899 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
11900 {
11901 	int r;
11902 
11903 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
11904 	      work->wakeup_all)
11905 		return;
11906 
11907 	r = kvm_mmu_reload(vcpu);
11908 	if (unlikely(r))
11909 		return;
11910 
11911 	if (!vcpu->arch.mmu->direct_map &&
11912 	      work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
11913 		return;
11914 
11915 	kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
11916 }
11917 
11918 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
11919 {
11920 	BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
11921 
11922 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
11923 }
11924 
11925 static inline u32 kvm_async_pf_next_probe(u32 key)
11926 {
11927 	return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
11928 }
11929 
11930 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11931 {
11932 	u32 key = kvm_async_pf_hash_fn(gfn);
11933 
11934 	while (vcpu->arch.apf.gfns[key] != ~0)
11935 		key = kvm_async_pf_next_probe(key);
11936 
11937 	vcpu->arch.apf.gfns[key] = gfn;
11938 }
11939 
11940 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
11941 {
11942 	int i;
11943 	u32 key = kvm_async_pf_hash_fn(gfn);
11944 
11945 	for (i = 0; i < ASYNC_PF_PER_VCPU &&
11946 		     (vcpu->arch.apf.gfns[key] != gfn &&
11947 		      vcpu->arch.apf.gfns[key] != ~0); i++)
11948 		key = kvm_async_pf_next_probe(key);
11949 
11950 	return key;
11951 }
11952 
11953 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11954 {
11955 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
11956 }
11957 
11958 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11959 {
11960 	u32 i, j, k;
11961 
11962 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
11963 
11964 	if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
11965 		return;
11966 
11967 	while (true) {
11968 		vcpu->arch.apf.gfns[i] = ~0;
11969 		do {
11970 			j = kvm_async_pf_next_probe(j);
11971 			if (vcpu->arch.apf.gfns[j] == ~0)
11972 				return;
11973 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
11974 			/*
11975 			 * k lies cyclically in ]i,j]
11976 			 * |    i.k.j |
11977 			 * |....j i.k.| or  |.k..j i...|
11978 			 */
11979 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
11980 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11981 		i = j;
11982 	}
11983 }
11984 
11985 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11986 {
11987 	u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11988 
11989 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11990 				      sizeof(reason));
11991 }
11992 
11993 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11994 {
11995 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11996 
11997 	return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11998 					     &token, offset, sizeof(token));
11999 }
12000 
12001 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
12002 {
12003 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
12004 	u32 val;
12005 
12006 	if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
12007 					 &val, offset, sizeof(val)))
12008 		return false;
12009 
12010 	return !val;
12011 }
12012 
12013 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
12014 {
12015 	if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
12016 		return false;
12017 
12018 	if (!kvm_pv_async_pf_enabled(vcpu) ||
12019 	    (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
12020 		return false;
12021 
12022 	return true;
12023 }
12024 
12025 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
12026 {
12027 	if (unlikely(!lapic_in_kernel(vcpu) ||
12028 		     kvm_event_needs_reinjection(vcpu) ||
12029 		     vcpu->arch.exception.pending))
12030 		return false;
12031 
12032 	if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
12033 		return false;
12034 
12035 	/*
12036 	 * If interrupts are off we cannot even use an artificial
12037 	 * halt state.
12038 	 */
12039 	return kvm_arch_interrupt_allowed(vcpu);
12040 }
12041 
12042 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
12043 				     struct kvm_async_pf *work)
12044 {
12045 	struct x86_exception fault;
12046 
12047 	trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
12048 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
12049 
12050 	if (kvm_can_deliver_async_pf(vcpu) &&
12051 	    !apf_put_user_notpresent(vcpu)) {
12052 		fault.vector = PF_VECTOR;
12053 		fault.error_code_valid = true;
12054 		fault.error_code = 0;
12055 		fault.nested_page_fault = false;
12056 		fault.address = work->arch.token;
12057 		fault.async_page_fault = true;
12058 		kvm_inject_page_fault(vcpu, &fault);
12059 		return true;
12060 	} else {
12061 		/*
12062 		 * It is not possible to deliver a paravirtualized asynchronous
12063 		 * page fault, but putting the guest in an artificial halt state
12064 		 * can be beneficial nevertheless: if an interrupt arrives, we
12065 		 * can deliver it timely and perhaps the guest will schedule
12066 		 * another process.  When the instruction that triggered a page
12067 		 * fault is retried, hopefully the page will be ready in the host.
12068 		 */
12069 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
12070 		return false;
12071 	}
12072 }
12073 
12074 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
12075 				 struct kvm_async_pf *work)
12076 {
12077 	struct kvm_lapic_irq irq = {
12078 		.delivery_mode = APIC_DM_FIXED,
12079 		.vector = vcpu->arch.apf.vec
12080 	};
12081 
12082 	if (work->wakeup_all)
12083 		work->arch.token = ~0; /* broadcast wakeup */
12084 	else
12085 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
12086 	trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
12087 
12088 	if ((work->wakeup_all || work->notpresent_injected) &&
12089 	    kvm_pv_async_pf_enabled(vcpu) &&
12090 	    !apf_put_user_ready(vcpu, work->arch.token)) {
12091 		vcpu->arch.apf.pageready_pending = true;
12092 		kvm_apic_set_irq(vcpu, &irq, NULL);
12093 	}
12094 
12095 	vcpu->arch.apf.halted = false;
12096 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
12097 }
12098 
12099 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
12100 {
12101 	kvm_make_request(KVM_REQ_APF_READY, vcpu);
12102 	if (!vcpu->arch.apf.pageready_pending)
12103 		kvm_vcpu_kick(vcpu);
12104 }
12105 
12106 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
12107 {
12108 	if (!kvm_pv_async_pf_enabled(vcpu))
12109 		return true;
12110 	else
12111 		return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
12112 }
12113 
12114 void kvm_arch_start_assignment(struct kvm *kvm)
12115 {
12116 	if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
12117 		static_call_cond(kvm_x86_start_assignment)(kvm);
12118 }
12119 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
12120 
12121 void kvm_arch_end_assignment(struct kvm *kvm)
12122 {
12123 	atomic_dec(&kvm->arch.assigned_device_count);
12124 }
12125 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
12126 
12127 bool kvm_arch_has_assigned_device(struct kvm *kvm)
12128 {
12129 	return atomic_read(&kvm->arch.assigned_device_count);
12130 }
12131 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
12132 
12133 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
12134 {
12135 	atomic_inc(&kvm->arch.noncoherent_dma_count);
12136 }
12137 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
12138 
12139 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
12140 {
12141 	atomic_dec(&kvm->arch.noncoherent_dma_count);
12142 }
12143 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
12144 
12145 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
12146 {
12147 	return atomic_read(&kvm->arch.noncoherent_dma_count);
12148 }
12149 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
12150 
12151 bool kvm_arch_has_irq_bypass(void)
12152 {
12153 	return true;
12154 }
12155 
12156 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
12157 				      struct irq_bypass_producer *prod)
12158 {
12159 	struct kvm_kernel_irqfd *irqfd =
12160 		container_of(cons, struct kvm_kernel_irqfd, consumer);
12161 	int ret;
12162 
12163 	irqfd->producer = prod;
12164 	kvm_arch_start_assignment(irqfd->kvm);
12165 	ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
12166 					 prod->irq, irqfd->gsi, 1);
12167 
12168 	if (ret)
12169 		kvm_arch_end_assignment(irqfd->kvm);
12170 
12171 	return ret;
12172 }
12173 
12174 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
12175 				      struct irq_bypass_producer *prod)
12176 {
12177 	int ret;
12178 	struct kvm_kernel_irqfd *irqfd =
12179 		container_of(cons, struct kvm_kernel_irqfd, consumer);
12180 
12181 	WARN_ON(irqfd->producer != prod);
12182 	irqfd->producer = NULL;
12183 
12184 	/*
12185 	 * When producer of consumer is unregistered, we change back to
12186 	 * remapped mode, so we can re-use the current implementation
12187 	 * when the irq is masked/disabled or the consumer side (KVM
12188 	 * int this case doesn't want to receive the interrupts.
12189 	*/
12190 	ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
12191 	if (ret)
12192 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
12193 		       " fails: %d\n", irqfd->consumer.token, ret);
12194 
12195 	kvm_arch_end_assignment(irqfd->kvm);
12196 }
12197 
12198 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
12199 				   uint32_t guest_irq, bool set)
12200 {
12201 	return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
12202 }
12203 
12204 bool kvm_arch_irqfd_route_changed(struct kvm_kernel_irq_routing_entry *old,
12205 				  struct kvm_kernel_irq_routing_entry *new)
12206 {
12207 	if (new->type != KVM_IRQ_ROUTING_MSI)
12208 		return true;
12209 
12210 	return !!memcmp(&old->msi, &new->msi, sizeof(new->msi));
12211 }
12212 
12213 bool kvm_vector_hashing_enabled(void)
12214 {
12215 	return vector_hashing;
12216 }
12217 
12218 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
12219 {
12220 	return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
12221 }
12222 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
12223 
12224 
12225 int kvm_spec_ctrl_test_value(u64 value)
12226 {
12227 	/*
12228 	 * test that setting IA32_SPEC_CTRL to given value
12229 	 * is allowed by the host processor
12230 	 */
12231 
12232 	u64 saved_value;
12233 	unsigned long flags;
12234 	int ret = 0;
12235 
12236 	local_irq_save(flags);
12237 
12238 	if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
12239 		ret = 1;
12240 	else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
12241 		ret = 1;
12242 	else
12243 		wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
12244 
12245 	local_irq_restore(flags);
12246 
12247 	return ret;
12248 }
12249 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
12250 
12251 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
12252 {
12253 	struct x86_exception fault;
12254 	u32 access = error_code &
12255 		(PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
12256 
12257 	if (!(error_code & PFERR_PRESENT_MASK) ||
12258 	    vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
12259 		/*
12260 		 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
12261 		 * tables probably do not match the TLB.  Just proceed
12262 		 * with the error code that the processor gave.
12263 		 */
12264 		fault.vector = PF_VECTOR;
12265 		fault.error_code_valid = true;
12266 		fault.error_code = error_code;
12267 		fault.nested_page_fault = false;
12268 		fault.address = gva;
12269 	}
12270 	vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
12271 }
12272 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
12273 
12274 /*
12275  * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
12276  * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
12277  * indicates whether exit to userspace is needed.
12278  */
12279 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
12280 			      struct x86_exception *e)
12281 {
12282 	if (r == X86EMUL_PROPAGATE_FAULT) {
12283 		kvm_inject_emulated_page_fault(vcpu, e);
12284 		return 1;
12285 	}
12286 
12287 	/*
12288 	 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
12289 	 * while handling a VMX instruction KVM could've handled the request
12290 	 * correctly by exiting to userspace and performing I/O but there
12291 	 * doesn't seem to be a real use-case behind such requests, just return
12292 	 * KVM_EXIT_INTERNAL_ERROR for now.
12293 	 */
12294 	kvm_prepare_emulation_failure_exit(vcpu);
12295 
12296 	return 0;
12297 }
12298 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
12299 
12300 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
12301 {
12302 	bool pcid_enabled;
12303 	struct x86_exception e;
12304 	struct {
12305 		u64 pcid;
12306 		u64 gla;
12307 	} operand;
12308 	int r;
12309 
12310 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
12311 	if (r != X86EMUL_CONTINUE)
12312 		return kvm_handle_memory_failure(vcpu, r, &e);
12313 
12314 	if (operand.pcid >> 12 != 0) {
12315 		kvm_inject_gp(vcpu, 0);
12316 		return 1;
12317 	}
12318 
12319 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
12320 
12321 	switch (type) {
12322 	case INVPCID_TYPE_INDIV_ADDR:
12323 		if ((!pcid_enabled && (operand.pcid != 0)) ||
12324 		    is_noncanonical_address(operand.gla, vcpu)) {
12325 			kvm_inject_gp(vcpu, 0);
12326 			return 1;
12327 		}
12328 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
12329 		return kvm_skip_emulated_instruction(vcpu);
12330 
12331 	case INVPCID_TYPE_SINGLE_CTXT:
12332 		if (!pcid_enabled && (operand.pcid != 0)) {
12333 			kvm_inject_gp(vcpu, 0);
12334 			return 1;
12335 		}
12336 
12337 		kvm_invalidate_pcid(vcpu, operand.pcid);
12338 		return kvm_skip_emulated_instruction(vcpu);
12339 
12340 	case INVPCID_TYPE_ALL_NON_GLOBAL:
12341 		/*
12342 		 * Currently, KVM doesn't mark global entries in the shadow
12343 		 * page tables, so a non-global flush just degenerates to a
12344 		 * global flush. If needed, we could optimize this later by
12345 		 * keeping track of global entries in shadow page tables.
12346 		 */
12347 
12348 		fallthrough;
12349 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
12350 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
12351 		return kvm_skip_emulated_instruction(vcpu);
12352 
12353 	default:
12354 		kvm_inject_gp(vcpu, 0);
12355 		return 1;
12356 	}
12357 }
12358 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
12359 
12360 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
12361 {
12362 	struct kvm_run *run = vcpu->run;
12363 	struct kvm_mmio_fragment *frag;
12364 	unsigned int len;
12365 
12366 	BUG_ON(!vcpu->mmio_needed);
12367 
12368 	/* Complete previous fragment */
12369 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
12370 	len = min(8u, frag->len);
12371 	if (!vcpu->mmio_is_write)
12372 		memcpy(frag->data, run->mmio.data, len);
12373 
12374 	if (frag->len <= 8) {
12375 		/* Switch to the next fragment. */
12376 		frag++;
12377 		vcpu->mmio_cur_fragment++;
12378 	} else {
12379 		/* Go forward to the next mmio piece. */
12380 		frag->data += len;
12381 		frag->gpa += len;
12382 		frag->len -= len;
12383 	}
12384 
12385 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
12386 		vcpu->mmio_needed = 0;
12387 
12388 		// VMG change, at this point, we're always done
12389 		// RIP has already been advanced
12390 		return 1;
12391 	}
12392 
12393 	// More MMIO is needed
12394 	run->mmio.phys_addr = frag->gpa;
12395 	run->mmio.len = min(8u, frag->len);
12396 	run->mmio.is_write = vcpu->mmio_is_write;
12397 	if (run->mmio.is_write)
12398 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
12399 	run->exit_reason = KVM_EXIT_MMIO;
12400 
12401 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12402 
12403 	return 0;
12404 }
12405 
12406 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12407 			  void *data)
12408 {
12409 	int handled;
12410 	struct kvm_mmio_fragment *frag;
12411 
12412 	if (!data)
12413 		return -EINVAL;
12414 
12415 	handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12416 	if (handled == bytes)
12417 		return 1;
12418 
12419 	bytes -= handled;
12420 	gpa += handled;
12421 	data += handled;
12422 
12423 	/*TODO: Check if need to increment number of frags */
12424 	frag = vcpu->mmio_fragments;
12425 	vcpu->mmio_nr_fragments = 1;
12426 	frag->len = bytes;
12427 	frag->gpa = gpa;
12428 	frag->data = data;
12429 
12430 	vcpu->mmio_needed = 1;
12431 	vcpu->mmio_cur_fragment = 0;
12432 
12433 	vcpu->run->mmio.phys_addr = gpa;
12434 	vcpu->run->mmio.len = min(8u, frag->len);
12435 	vcpu->run->mmio.is_write = 1;
12436 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
12437 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
12438 
12439 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12440 
12441 	return 0;
12442 }
12443 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
12444 
12445 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12446 			 void *data)
12447 {
12448 	int handled;
12449 	struct kvm_mmio_fragment *frag;
12450 
12451 	if (!data)
12452 		return -EINVAL;
12453 
12454 	handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12455 	if (handled == bytes)
12456 		return 1;
12457 
12458 	bytes -= handled;
12459 	gpa += handled;
12460 	data += handled;
12461 
12462 	/*TODO: Check if need to increment number of frags */
12463 	frag = vcpu->mmio_fragments;
12464 	vcpu->mmio_nr_fragments = 1;
12465 	frag->len = bytes;
12466 	frag->gpa = gpa;
12467 	frag->data = data;
12468 
12469 	vcpu->mmio_needed = 1;
12470 	vcpu->mmio_cur_fragment = 0;
12471 
12472 	vcpu->run->mmio.phys_addr = gpa;
12473 	vcpu->run->mmio.len = min(8u, frag->len);
12474 	vcpu->run->mmio.is_write = 0;
12475 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
12476 
12477 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12478 
12479 	return 0;
12480 }
12481 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
12482 
12483 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12484 			   unsigned int port);
12485 
12486 static int complete_sev_es_emulated_outs(struct kvm_vcpu *vcpu)
12487 {
12488 	int size = vcpu->arch.pio.size;
12489 	int port = vcpu->arch.pio.port;
12490 
12491 	vcpu->arch.pio.count = 0;
12492 	if (vcpu->arch.sev_pio_count)
12493 		return kvm_sev_es_outs(vcpu, size, port);
12494 	return 1;
12495 }
12496 
12497 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12498 			   unsigned int port)
12499 {
12500 	for (;;) {
12501 		unsigned int count =
12502 			min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
12503 		int ret = emulator_pio_out(vcpu, size, port, vcpu->arch.sev_pio_data, count);
12504 
12505 		/* memcpy done already by emulator_pio_out.  */
12506 		vcpu->arch.sev_pio_count -= count;
12507 		vcpu->arch.sev_pio_data += count * vcpu->arch.pio.size;
12508 		if (!ret)
12509 			break;
12510 
12511 		/* Emulation done by the kernel.  */
12512 		if (!vcpu->arch.sev_pio_count)
12513 			return 1;
12514 	}
12515 
12516 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_outs;
12517 	return 0;
12518 }
12519 
12520 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12521 			  unsigned int port);
12522 
12523 static void advance_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12524 {
12525 	unsigned count = vcpu->arch.pio.count;
12526 	complete_emulator_pio_in(vcpu, vcpu->arch.sev_pio_data);
12527 	vcpu->arch.sev_pio_count -= count;
12528 	vcpu->arch.sev_pio_data += count * vcpu->arch.pio.size;
12529 }
12530 
12531 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12532 {
12533 	int size = vcpu->arch.pio.size;
12534 	int port = vcpu->arch.pio.port;
12535 
12536 	advance_sev_es_emulated_ins(vcpu);
12537 	if (vcpu->arch.sev_pio_count)
12538 		return kvm_sev_es_ins(vcpu, size, port);
12539 	return 1;
12540 }
12541 
12542 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12543 			  unsigned int port)
12544 {
12545 	for (;;) {
12546 		unsigned int count =
12547 			min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
12548 		if (!__emulator_pio_in(vcpu, size, port, count))
12549 			break;
12550 
12551 		/* Emulation done by the kernel.  */
12552 		advance_sev_es_emulated_ins(vcpu);
12553 		if (!vcpu->arch.sev_pio_count)
12554 			return 1;
12555 	}
12556 
12557 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
12558 	return 0;
12559 }
12560 
12561 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
12562 			 unsigned int port, void *data,  unsigned int count,
12563 			 int in)
12564 {
12565 	vcpu->arch.sev_pio_data = data;
12566 	vcpu->arch.sev_pio_count = count;
12567 	return in ? kvm_sev_es_ins(vcpu, size, port)
12568 		  : kvm_sev_es_outs(vcpu, size, port);
12569 }
12570 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
12571 
12572 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
12573 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
12574 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
12575 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
12576 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
12577 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
12578 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
12579 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
12580 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
12581 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
12582 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
12583 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
12584 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
12585 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
12586 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
12587 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
12588 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
12589 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
12590 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
12591 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
12592 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
12593 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
12594 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
12595 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
12596 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
12597 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
12598 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);
12599