xref: /openbmc/linux/arch/x86/kvm/x86.c (revision c3531edc79a7adbe4451d34084e8bf227c022872)
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
19 
20 #include <linux/kvm_host.h>
21 #include "irq.h"
22 #include "ioapic.h"
23 #include "mmu.h"
24 #include "i8254.h"
25 #include "tss.h"
26 #include "kvm_cache_regs.h"
27 #include "kvm_emulate.h"
28 #include "x86.h"
29 #include "cpuid.h"
30 #include "pmu.h"
31 #include "hyperv.h"
32 #include "lapic.h"
33 #include "xen.h"
34 #include "smm.h"
35 
36 #include <linux/clocksource.h>
37 #include <linux/interrupt.h>
38 #include <linux/kvm.h>
39 #include <linux/fs.h>
40 #include <linux/vmalloc.h>
41 #include <linux/export.h>
42 #include <linux/moduleparam.h>
43 #include <linux/mman.h>
44 #include <linux/highmem.h>
45 #include <linux/iommu.h>
46 #include <linux/cpufreq.h>
47 #include <linux/user-return-notifier.h>
48 #include <linux/srcu.h>
49 #include <linux/slab.h>
50 #include <linux/perf_event.h>
51 #include <linux/uaccess.h>
52 #include <linux/hash.h>
53 #include <linux/pci.h>
54 #include <linux/timekeeper_internal.h>
55 #include <linux/pvclock_gtod.h>
56 #include <linux/kvm_irqfd.h>
57 #include <linux/irqbypass.h>
58 #include <linux/sched/stat.h>
59 #include <linux/sched/isolation.h>
60 #include <linux/mem_encrypt.h>
61 #include <linux/entry-kvm.h>
62 #include <linux/suspend.h>
63 
64 #include <trace/events/kvm.h>
65 
66 #include <asm/debugreg.h>
67 #include <asm/msr.h>
68 #include <asm/desc.h>
69 #include <asm/mce.h>
70 #include <asm/pkru.h>
71 #include <linux/kernel_stat.h>
72 #include <asm/fpu/api.h>
73 #include <asm/fpu/xcr.h>
74 #include <asm/fpu/xstate.h>
75 #include <asm/pvclock.h>
76 #include <asm/div64.h>
77 #include <asm/irq_remapping.h>
78 #include <asm/mshyperv.h>
79 #include <asm/hypervisor.h>
80 #include <asm/tlbflush.h>
81 #include <asm/intel_pt.h>
82 #include <asm/emulate_prefix.h>
83 #include <asm/sgx.h>
84 #include <clocksource/hyperv_timer.h>
85 
86 #define CREATE_TRACE_POINTS
87 #include "trace.h"
88 
89 #define MAX_IO_MSRS 256
90 #define KVM_MAX_MCE_BANKS 32
91 
92 struct kvm_caps kvm_caps __read_mostly = {
93 	.supported_mce_cap = MCG_CTL_P | MCG_SER_P,
94 };
95 EXPORT_SYMBOL_GPL(kvm_caps);
96 
97 #define  ERR_PTR_USR(e)  ((void __user *)ERR_PTR(e))
98 
99 #define emul_to_vcpu(ctxt) \
100 	((struct kvm_vcpu *)(ctxt)->vcpu)
101 
102 /* EFER defaults:
103  * - enable syscall per default because its emulated by KVM
104  * - enable LME and LMA per default on 64 bit KVM
105  */
106 #ifdef CONFIG_X86_64
107 static
108 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
109 #else
110 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
111 #endif
112 
113 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
114 
115 #define KVM_EXIT_HYPERCALL_VALID_MASK (1 << KVM_HC_MAP_GPA_RANGE)
116 
117 #define KVM_CAP_PMU_VALID_MASK KVM_PMU_CAP_DISABLE
118 
119 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
120                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
121 
122 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
123 static void process_nmi(struct kvm_vcpu *vcpu);
124 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
125 static void store_regs(struct kvm_vcpu *vcpu);
126 static int sync_regs(struct kvm_vcpu *vcpu);
127 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu);
128 
129 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
130 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
131 
132 static DEFINE_MUTEX(vendor_module_lock);
133 struct kvm_x86_ops kvm_x86_ops __read_mostly;
134 
135 #define KVM_X86_OP(func)					     \
136 	DEFINE_STATIC_CALL_NULL(kvm_x86_##func,			     \
137 				*(((struct kvm_x86_ops *)0)->func));
138 #define KVM_X86_OP_OPTIONAL KVM_X86_OP
139 #define KVM_X86_OP_OPTIONAL_RET0 KVM_X86_OP
140 #include <asm/kvm-x86-ops.h>
141 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
142 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
143 
144 static bool __read_mostly ignore_msrs = 0;
145 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
146 
147 bool __read_mostly report_ignored_msrs = true;
148 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
149 EXPORT_SYMBOL_GPL(report_ignored_msrs);
150 
151 unsigned int min_timer_period_us = 200;
152 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
153 
154 static bool __read_mostly kvmclock_periodic_sync = true;
155 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
156 
157 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
158 static u32 __read_mostly tsc_tolerance_ppm = 250;
159 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
160 
161 /*
162  * lapic timer advance (tscdeadline mode only) in nanoseconds.  '-1' enables
163  * adaptive tuning starting from default advancement of 1000ns.  '0' disables
164  * advancement entirely.  Any other value is used as-is and disables adaptive
165  * tuning, i.e. allows privileged userspace to set an exact advancement time.
166  */
167 static int __read_mostly lapic_timer_advance_ns = -1;
168 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
169 
170 static bool __read_mostly vector_hashing = true;
171 module_param(vector_hashing, bool, S_IRUGO);
172 
173 bool __read_mostly enable_vmware_backdoor = false;
174 module_param(enable_vmware_backdoor, bool, S_IRUGO);
175 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
176 
177 /*
178  * Flags to manipulate forced emulation behavior (any non-zero value will
179  * enable forced emulation).
180  */
181 #define KVM_FEP_CLEAR_RFLAGS_RF	BIT(1)
182 static int __read_mostly force_emulation_prefix;
183 module_param(force_emulation_prefix, int, 0644);
184 
185 int __read_mostly pi_inject_timer = -1;
186 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
187 
188 /* Enable/disable PMU virtualization */
189 bool __read_mostly enable_pmu = true;
190 EXPORT_SYMBOL_GPL(enable_pmu);
191 module_param(enable_pmu, bool, 0444);
192 
193 bool __read_mostly eager_page_split = true;
194 module_param(eager_page_split, bool, 0644);
195 
196 /*
197  * Restoring the host value for MSRs that are only consumed when running in
198  * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
199  * returns to userspace, i.e. the kernel can run with the guest's value.
200  */
201 #define KVM_MAX_NR_USER_RETURN_MSRS 16
202 
203 struct kvm_user_return_msrs {
204 	struct user_return_notifier urn;
205 	bool registered;
206 	struct kvm_user_return_msr_values {
207 		u64 host;
208 		u64 curr;
209 	} values[KVM_MAX_NR_USER_RETURN_MSRS];
210 };
211 
212 u32 __read_mostly kvm_nr_uret_msrs;
213 EXPORT_SYMBOL_GPL(kvm_nr_uret_msrs);
214 static u32 __read_mostly kvm_uret_msrs_list[KVM_MAX_NR_USER_RETURN_MSRS];
215 static struct kvm_user_return_msrs __percpu *user_return_msrs;
216 
217 #define KVM_SUPPORTED_XCR0     (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
218 				| XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
219 				| XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
220 				| XFEATURE_MASK_PKRU | XFEATURE_MASK_XTILE)
221 
222 u64 __read_mostly host_efer;
223 EXPORT_SYMBOL_GPL(host_efer);
224 
225 bool __read_mostly allow_smaller_maxphyaddr = 0;
226 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
227 
228 bool __read_mostly enable_apicv = true;
229 EXPORT_SYMBOL_GPL(enable_apicv);
230 
231 u64 __read_mostly host_xss;
232 EXPORT_SYMBOL_GPL(host_xss);
233 
234 const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
235 	KVM_GENERIC_VM_STATS(),
236 	STATS_DESC_COUNTER(VM, mmu_shadow_zapped),
237 	STATS_DESC_COUNTER(VM, mmu_pte_write),
238 	STATS_DESC_COUNTER(VM, mmu_pde_zapped),
239 	STATS_DESC_COUNTER(VM, mmu_flooded),
240 	STATS_DESC_COUNTER(VM, mmu_recycled),
241 	STATS_DESC_COUNTER(VM, mmu_cache_miss),
242 	STATS_DESC_ICOUNTER(VM, mmu_unsync),
243 	STATS_DESC_ICOUNTER(VM, pages_4k),
244 	STATS_DESC_ICOUNTER(VM, pages_2m),
245 	STATS_DESC_ICOUNTER(VM, pages_1g),
246 	STATS_DESC_ICOUNTER(VM, nx_lpage_splits),
247 	STATS_DESC_PCOUNTER(VM, max_mmu_rmap_size),
248 	STATS_DESC_PCOUNTER(VM, max_mmu_page_hash_collisions)
249 };
250 
251 const struct kvm_stats_header kvm_vm_stats_header = {
252 	.name_size = KVM_STATS_NAME_SIZE,
253 	.num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
254 	.id_offset = sizeof(struct kvm_stats_header),
255 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
256 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
257 		       sizeof(kvm_vm_stats_desc),
258 };
259 
260 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
261 	KVM_GENERIC_VCPU_STATS(),
262 	STATS_DESC_COUNTER(VCPU, pf_taken),
263 	STATS_DESC_COUNTER(VCPU, pf_fixed),
264 	STATS_DESC_COUNTER(VCPU, pf_emulate),
265 	STATS_DESC_COUNTER(VCPU, pf_spurious),
266 	STATS_DESC_COUNTER(VCPU, pf_fast),
267 	STATS_DESC_COUNTER(VCPU, pf_mmio_spte_created),
268 	STATS_DESC_COUNTER(VCPU, pf_guest),
269 	STATS_DESC_COUNTER(VCPU, tlb_flush),
270 	STATS_DESC_COUNTER(VCPU, invlpg),
271 	STATS_DESC_COUNTER(VCPU, exits),
272 	STATS_DESC_COUNTER(VCPU, io_exits),
273 	STATS_DESC_COUNTER(VCPU, mmio_exits),
274 	STATS_DESC_COUNTER(VCPU, signal_exits),
275 	STATS_DESC_COUNTER(VCPU, irq_window_exits),
276 	STATS_DESC_COUNTER(VCPU, nmi_window_exits),
277 	STATS_DESC_COUNTER(VCPU, l1d_flush),
278 	STATS_DESC_COUNTER(VCPU, halt_exits),
279 	STATS_DESC_COUNTER(VCPU, request_irq_exits),
280 	STATS_DESC_COUNTER(VCPU, irq_exits),
281 	STATS_DESC_COUNTER(VCPU, host_state_reload),
282 	STATS_DESC_COUNTER(VCPU, fpu_reload),
283 	STATS_DESC_COUNTER(VCPU, insn_emulation),
284 	STATS_DESC_COUNTER(VCPU, insn_emulation_fail),
285 	STATS_DESC_COUNTER(VCPU, hypercalls),
286 	STATS_DESC_COUNTER(VCPU, irq_injections),
287 	STATS_DESC_COUNTER(VCPU, nmi_injections),
288 	STATS_DESC_COUNTER(VCPU, req_event),
289 	STATS_DESC_COUNTER(VCPU, nested_run),
290 	STATS_DESC_COUNTER(VCPU, directed_yield_attempted),
291 	STATS_DESC_COUNTER(VCPU, directed_yield_successful),
292 	STATS_DESC_COUNTER(VCPU, preemption_reported),
293 	STATS_DESC_COUNTER(VCPU, preemption_other),
294 	STATS_DESC_IBOOLEAN(VCPU, guest_mode),
295 	STATS_DESC_COUNTER(VCPU, notify_window_exits),
296 };
297 
298 const struct kvm_stats_header kvm_vcpu_stats_header = {
299 	.name_size = KVM_STATS_NAME_SIZE,
300 	.num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
301 	.id_offset = sizeof(struct kvm_stats_header),
302 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
303 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
304 		       sizeof(kvm_vcpu_stats_desc),
305 };
306 
307 u64 __read_mostly host_xcr0;
308 
309 static struct kmem_cache *x86_emulator_cache;
310 
311 /*
312  * When called, it means the previous get/set msr reached an invalid msr.
313  * Return true if we want to ignore/silent this failed msr access.
314  */
315 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
316 {
317 	const char *op = write ? "wrmsr" : "rdmsr";
318 
319 	if (ignore_msrs) {
320 		if (report_ignored_msrs)
321 			kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
322 				      op, msr, data);
323 		/* Mask the error */
324 		return true;
325 	} else {
326 		kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
327 				      op, msr, data);
328 		return false;
329 	}
330 }
331 
332 static struct kmem_cache *kvm_alloc_emulator_cache(void)
333 {
334 	unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
335 	unsigned int size = sizeof(struct x86_emulate_ctxt);
336 
337 	return kmem_cache_create_usercopy("x86_emulator", size,
338 					  __alignof__(struct x86_emulate_ctxt),
339 					  SLAB_ACCOUNT, useroffset,
340 					  size - useroffset, NULL);
341 }
342 
343 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
344 
345 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
346 {
347 	int i;
348 	for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
349 		vcpu->arch.apf.gfns[i] = ~0;
350 }
351 
352 static void kvm_on_user_return(struct user_return_notifier *urn)
353 {
354 	unsigned slot;
355 	struct kvm_user_return_msrs *msrs
356 		= container_of(urn, struct kvm_user_return_msrs, urn);
357 	struct kvm_user_return_msr_values *values;
358 	unsigned long flags;
359 
360 	/*
361 	 * Disabling irqs at this point since the following code could be
362 	 * interrupted and executed through kvm_arch_hardware_disable()
363 	 */
364 	local_irq_save(flags);
365 	if (msrs->registered) {
366 		msrs->registered = false;
367 		user_return_notifier_unregister(urn);
368 	}
369 	local_irq_restore(flags);
370 	for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
371 		values = &msrs->values[slot];
372 		if (values->host != values->curr) {
373 			wrmsrl(kvm_uret_msrs_list[slot], values->host);
374 			values->curr = values->host;
375 		}
376 	}
377 }
378 
379 static int kvm_probe_user_return_msr(u32 msr)
380 {
381 	u64 val;
382 	int ret;
383 
384 	preempt_disable();
385 	ret = rdmsrl_safe(msr, &val);
386 	if (ret)
387 		goto out;
388 	ret = wrmsrl_safe(msr, val);
389 out:
390 	preempt_enable();
391 	return ret;
392 }
393 
394 int kvm_add_user_return_msr(u32 msr)
395 {
396 	BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
397 
398 	if (kvm_probe_user_return_msr(msr))
399 		return -1;
400 
401 	kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
402 	return kvm_nr_uret_msrs++;
403 }
404 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
405 
406 int kvm_find_user_return_msr(u32 msr)
407 {
408 	int i;
409 
410 	for (i = 0; i < kvm_nr_uret_msrs; ++i) {
411 		if (kvm_uret_msrs_list[i] == msr)
412 			return i;
413 	}
414 	return -1;
415 }
416 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
417 
418 static void kvm_user_return_msr_cpu_online(void)
419 {
420 	unsigned int cpu = smp_processor_id();
421 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
422 	u64 value;
423 	int i;
424 
425 	for (i = 0; i < kvm_nr_uret_msrs; ++i) {
426 		rdmsrl_safe(kvm_uret_msrs_list[i], &value);
427 		msrs->values[i].host = value;
428 		msrs->values[i].curr = value;
429 	}
430 }
431 
432 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
433 {
434 	unsigned int cpu = smp_processor_id();
435 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
436 	int err;
437 
438 	value = (value & mask) | (msrs->values[slot].host & ~mask);
439 	if (value == msrs->values[slot].curr)
440 		return 0;
441 	err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
442 	if (err)
443 		return 1;
444 
445 	msrs->values[slot].curr = value;
446 	if (!msrs->registered) {
447 		msrs->urn.on_user_return = kvm_on_user_return;
448 		user_return_notifier_register(&msrs->urn);
449 		msrs->registered = true;
450 	}
451 	return 0;
452 }
453 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
454 
455 static void drop_user_return_notifiers(void)
456 {
457 	unsigned int cpu = smp_processor_id();
458 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
459 
460 	if (msrs->registered)
461 		kvm_on_user_return(&msrs->urn);
462 }
463 
464 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
465 {
466 	return vcpu->arch.apic_base;
467 }
468 
469 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
470 {
471 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
472 }
473 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
474 
475 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
476 {
477 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
478 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
479 	u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
480 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
481 
482 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
483 		return 1;
484 	if (!msr_info->host_initiated) {
485 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
486 			return 1;
487 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
488 			return 1;
489 	}
490 
491 	kvm_lapic_set_base(vcpu, msr_info->data);
492 	kvm_recalculate_apic_map(vcpu->kvm);
493 	return 0;
494 }
495 
496 /*
497  * Handle a fault on a hardware virtualization (VMX or SVM) instruction.
498  *
499  * Hardware virtualization extension instructions may fault if a reboot turns
500  * off virtualization while processes are running.  Usually after catching the
501  * fault we just panic; during reboot instead the instruction is ignored.
502  */
503 noinstr void kvm_spurious_fault(void)
504 {
505 	/* Fault while not rebooting.  We want the trace. */
506 	BUG_ON(!kvm_rebooting);
507 }
508 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
509 
510 #define EXCPT_BENIGN		0
511 #define EXCPT_CONTRIBUTORY	1
512 #define EXCPT_PF		2
513 
514 static int exception_class(int vector)
515 {
516 	switch (vector) {
517 	case PF_VECTOR:
518 		return EXCPT_PF;
519 	case DE_VECTOR:
520 	case TS_VECTOR:
521 	case NP_VECTOR:
522 	case SS_VECTOR:
523 	case GP_VECTOR:
524 		return EXCPT_CONTRIBUTORY;
525 	default:
526 		break;
527 	}
528 	return EXCPT_BENIGN;
529 }
530 
531 #define EXCPT_FAULT		0
532 #define EXCPT_TRAP		1
533 #define EXCPT_ABORT		2
534 #define EXCPT_INTERRUPT		3
535 #define EXCPT_DB		4
536 
537 static int exception_type(int vector)
538 {
539 	unsigned int mask;
540 
541 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
542 		return EXCPT_INTERRUPT;
543 
544 	mask = 1 << vector;
545 
546 	/*
547 	 * #DBs can be trap-like or fault-like, the caller must check other CPU
548 	 * state, e.g. DR6, to determine whether a #DB is a trap or fault.
549 	 */
550 	if (mask & (1 << DB_VECTOR))
551 		return EXCPT_DB;
552 
553 	if (mask & ((1 << BP_VECTOR) | (1 << OF_VECTOR)))
554 		return EXCPT_TRAP;
555 
556 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
557 		return EXCPT_ABORT;
558 
559 	/* Reserved exceptions will result in fault */
560 	return EXCPT_FAULT;
561 }
562 
563 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu,
564 				   struct kvm_queued_exception *ex)
565 {
566 	if (!ex->has_payload)
567 		return;
568 
569 	switch (ex->vector) {
570 	case DB_VECTOR:
571 		/*
572 		 * "Certain debug exceptions may clear bit 0-3.  The
573 		 * remaining contents of the DR6 register are never
574 		 * cleared by the processor".
575 		 */
576 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
577 		/*
578 		 * In order to reflect the #DB exception payload in guest
579 		 * dr6, three components need to be considered: active low
580 		 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
581 		 * DR6_BS and DR6_BT)
582 		 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
583 		 * In the target guest dr6:
584 		 * FIXED_1 bits should always be set.
585 		 * Active low bits should be cleared if 1-setting in payload.
586 		 * Active high bits should be set if 1-setting in payload.
587 		 *
588 		 * Note, the payload is compatible with the pending debug
589 		 * exceptions/exit qualification under VMX, that active_low bits
590 		 * are active high in payload.
591 		 * So they need to be flipped for DR6.
592 		 */
593 		vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
594 		vcpu->arch.dr6 |= ex->payload;
595 		vcpu->arch.dr6 ^= ex->payload & DR6_ACTIVE_LOW;
596 
597 		/*
598 		 * The #DB payload is defined as compatible with the 'pending
599 		 * debug exceptions' field under VMX, not DR6. While bit 12 is
600 		 * defined in the 'pending debug exceptions' field (enabled
601 		 * breakpoint), it is reserved and must be zero in DR6.
602 		 */
603 		vcpu->arch.dr6 &= ~BIT(12);
604 		break;
605 	case PF_VECTOR:
606 		vcpu->arch.cr2 = ex->payload;
607 		break;
608 	}
609 
610 	ex->has_payload = false;
611 	ex->payload = 0;
612 }
613 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
614 
615 static void kvm_queue_exception_vmexit(struct kvm_vcpu *vcpu, unsigned int vector,
616 				       bool has_error_code, u32 error_code,
617 				       bool has_payload, unsigned long payload)
618 {
619 	struct kvm_queued_exception *ex = &vcpu->arch.exception_vmexit;
620 
621 	ex->vector = vector;
622 	ex->injected = false;
623 	ex->pending = true;
624 	ex->has_error_code = has_error_code;
625 	ex->error_code = error_code;
626 	ex->has_payload = has_payload;
627 	ex->payload = payload;
628 }
629 
630 /* Forcibly leave the nested mode in cases like a vCPU reset */
631 static void kvm_leave_nested(struct kvm_vcpu *vcpu)
632 {
633 	kvm_x86_ops.nested_ops->leave_nested(vcpu);
634 }
635 
636 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
637 		unsigned nr, bool has_error, u32 error_code,
638 	        bool has_payload, unsigned long payload, bool reinject)
639 {
640 	u32 prev_nr;
641 	int class1, class2;
642 
643 	kvm_make_request(KVM_REQ_EVENT, vcpu);
644 
645 	/*
646 	 * If the exception is destined for L2 and isn't being reinjected,
647 	 * morph it to a VM-Exit if L1 wants to intercept the exception.  A
648 	 * previously injected exception is not checked because it was checked
649 	 * when it was original queued, and re-checking is incorrect if _L1_
650 	 * injected the exception, in which case it's exempt from interception.
651 	 */
652 	if (!reinject && is_guest_mode(vcpu) &&
653 	    kvm_x86_ops.nested_ops->is_exception_vmexit(vcpu, nr, error_code)) {
654 		kvm_queue_exception_vmexit(vcpu, nr, has_error, error_code,
655 					   has_payload, payload);
656 		return;
657 	}
658 
659 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
660 	queue:
661 		if (reinject) {
662 			/*
663 			 * On VM-Entry, an exception can be pending if and only
664 			 * if event injection was blocked by nested_run_pending.
665 			 * In that case, however, vcpu_enter_guest() requests an
666 			 * immediate exit, and the guest shouldn't proceed far
667 			 * enough to need reinjection.
668 			 */
669 			WARN_ON_ONCE(kvm_is_exception_pending(vcpu));
670 			vcpu->arch.exception.injected = true;
671 			if (WARN_ON_ONCE(has_payload)) {
672 				/*
673 				 * A reinjected event has already
674 				 * delivered its payload.
675 				 */
676 				has_payload = false;
677 				payload = 0;
678 			}
679 		} else {
680 			vcpu->arch.exception.pending = true;
681 			vcpu->arch.exception.injected = false;
682 		}
683 		vcpu->arch.exception.has_error_code = has_error;
684 		vcpu->arch.exception.vector = nr;
685 		vcpu->arch.exception.error_code = error_code;
686 		vcpu->arch.exception.has_payload = has_payload;
687 		vcpu->arch.exception.payload = payload;
688 		if (!is_guest_mode(vcpu))
689 			kvm_deliver_exception_payload(vcpu,
690 						      &vcpu->arch.exception);
691 		return;
692 	}
693 
694 	/* to check exception */
695 	prev_nr = vcpu->arch.exception.vector;
696 	if (prev_nr == DF_VECTOR) {
697 		/* triple fault -> shutdown */
698 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
699 		return;
700 	}
701 	class1 = exception_class(prev_nr);
702 	class2 = exception_class(nr);
703 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY) ||
704 	    (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
705 		/*
706 		 * Synthesize #DF.  Clear the previously injected or pending
707 		 * exception so as not to incorrectly trigger shutdown.
708 		 */
709 		vcpu->arch.exception.injected = false;
710 		vcpu->arch.exception.pending = false;
711 
712 		kvm_queue_exception_e(vcpu, DF_VECTOR, 0);
713 	} else {
714 		/* replace previous exception with a new one in a hope
715 		   that instruction re-execution will regenerate lost
716 		   exception */
717 		goto queue;
718 	}
719 }
720 
721 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
722 {
723 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
724 }
725 EXPORT_SYMBOL_GPL(kvm_queue_exception);
726 
727 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
728 {
729 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
730 }
731 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
732 
733 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
734 			   unsigned long payload)
735 {
736 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
737 }
738 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
739 
740 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
741 				    u32 error_code, unsigned long payload)
742 {
743 	kvm_multiple_exception(vcpu, nr, true, error_code,
744 			       true, payload, false);
745 }
746 
747 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
748 {
749 	if (err)
750 		kvm_inject_gp(vcpu, 0);
751 	else
752 		return kvm_skip_emulated_instruction(vcpu);
753 
754 	return 1;
755 }
756 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
757 
758 static int complete_emulated_insn_gp(struct kvm_vcpu *vcpu, int err)
759 {
760 	if (err) {
761 		kvm_inject_gp(vcpu, 0);
762 		return 1;
763 	}
764 
765 	return kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE | EMULTYPE_SKIP |
766 				       EMULTYPE_COMPLETE_USER_EXIT);
767 }
768 
769 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
770 {
771 	++vcpu->stat.pf_guest;
772 
773 	/*
774 	 * Async #PF in L2 is always forwarded to L1 as a VM-Exit regardless of
775 	 * whether or not L1 wants to intercept "regular" #PF.
776 	 */
777 	if (is_guest_mode(vcpu) && fault->async_page_fault)
778 		kvm_queue_exception_vmexit(vcpu, PF_VECTOR,
779 					   true, fault->error_code,
780 					   true, fault->address);
781 	else
782 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
783 					fault->address);
784 }
785 
786 void kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
787 				    struct x86_exception *fault)
788 {
789 	struct kvm_mmu *fault_mmu;
790 	WARN_ON_ONCE(fault->vector != PF_VECTOR);
791 
792 	fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
793 					       vcpu->arch.walk_mmu;
794 
795 	/*
796 	 * Invalidate the TLB entry for the faulting address, if it exists,
797 	 * else the access will fault indefinitely (and to emulate hardware).
798 	 */
799 	if ((fault->error_code & PFERR_PRESENT_MASK) &&
800 	    !(fault->error_code & PFERR_RSVD_MASK))
801 		kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
802 				       fault_mmu->root.hpa);
803 
804 	fault_mmu->inject_page_fault(vcpu, fault);
805 }
806 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
807 
808 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
809 {
810 	atomic_inc(&vcpu->arch.nmi_queued);
811 	kvm_make_request(KVM_REQ_NMI, vcpu);
812 }
813 
814 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
815 {
816 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
817 }
818 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
819 
820 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
821 {
822 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
823 }
824 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
825 
826 /*
827  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
828  * a #GP and return false.
829  */
830 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
831 {
832 	if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
833 		return true;
834 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
835 	return false;
836 }
837 
838 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
839 {
840 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
841 		return true;
842 
843 	kvm_queue_exception(vcpu, UD_VECTOR);
844 	return false;
845 }
846 EXPORT_SYMBOL_GPL(kvm_require_dr);
847 
848 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
849 {
850 	return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
851 }
852 
853 /*
854  * Load the pae pdptrs.  Return 1 if they are all valid, 0 otherwise.
855  */
856 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
857 {
858 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
859 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
860 	gpa_t real_gpa;
861 	int i;
862 	int ret;
863 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
864 
865 	/*
866 	 * If the MMU is nested, CR3 holds an L2 GPA and needs to be translated
867 	 * to an L1 GPA.
868 	 */
869 	real_gpa = kvm_translate_gpa(vcpu, mmu, gfn_to_gpa(pdpt_gfn),
870 				     PFERR_USER_MASK | PFERR_WRITE_MASK, NULL);
871 	if (real_gpa == INVALID_GPA)
872 		return 0;
873 
874 	/* Note the offset, PDPTRs are 32 byte aligned when using PAE paging. */
875 	ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(real_gpa), pdpte,
876 				       cr3 & GENMASK(11, 5), sizeof(pdpte));
877 	if (ret < 0)
878 		return 0;
879 
880 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
881 		if ((pdpte[i] & PT_PRESENT_MASK) &&
882 		    (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
883 			return 0;
884 		}
885 	}
886 
887 	/*
888 	 * Marking VCPU_EXREG_PDPTR dirty doesn't work for !tdp_enabled.
889 	 * Shadow page roots need to be reconstructed instead.
890 	 */
891 	if (!tdp_enabled && memcmp(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs)))
892 		kvm_mmu_free_roots(vcpu->kvm, mmu, KVM_MMU_ROOT_CURRENT);
893 
894 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
895 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
896 	kvm_make_request(KVM_REQ_LOAD_MMU_PGD, vcpu);
897 	vcpu->arch.pdptrs_from_userspace = false;
898 
899 	return 1;
900 }
901 EXPORT_SYMBOL_GPL(load_pdptrs);
902 
903 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
904 {
905 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
906 		kvm_clear_async_pf_completion_queue(vcpu);
907 		kvm_async_pf_hash_reset(vcpu);
908 
909 		/*
910 		 * Clearing CR0.PG is defined to flush the TLB from the guest's
911 		 * perspective.
912 		 */
913 		if (!(cr0 & X86_CR0_PG))
914 			kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
915 	}
916 
917 	if ((cr0 ^ old_cr0) & KVM_MMU_CR0_ROLE_BITS)
918 		kvm_mmu_reset_context(vcpu);
919 
920 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
921 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
922 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
923 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
924 }
925 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
926 
927 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
928 {
929 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
930 
931 	cr0 |= X86_CR0_ET;
932 
933 #ifdef CONFIG_X86_64
934 	if (cr0 & 0xffffffff00000000UL)
935 		return 1;
936 #endif
937 
938 	cr0 &= ~CR0_RESERVED_BITS;
939 
940 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
941 		return 1;
942 
943 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
944 		return 1;
945 
946 #ifdef CONFIG_X86_64
947 	if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
948 	    (cr0 & X86_CR0_PG)) {
949 		int cs_db, cs_l;
950 
951 		if (!is_pae(vcpu))
952 			return 1;
953 		static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
954 		if (cs_l)
955 			return 1;
956 	}
957 #endif
958 	if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
959 	    is_pae(vcpu) && ((cr0 ^ old_cr0) & X86_CR0_PDPTR_BITS) &&
960 	    !load_pdptrs(vcpu, kvm_read_cr3(vcpu)))
961 		return 1;
962 
963 	if (!(cr0 & X86_CR0_PG) &&
964 	    (is_64_bit_mode(vcpu) || kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)))
965 		return 1;
966 
967 	static_call(kvm_x86_set_cr0)(vcpu, cr0);
968 
969 	kvm_post_set_cr0(vcpu, old_cr0, cr0);
970 
971 	return 0;
972 }
973 EXPORT_SYMBOL_GPL(kvm_set_cr0);
974 
975 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
976 {
977 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
978 }
979 EXPORT_SYMBOL_GPL(kvm_lmsw);
980 
981 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
982 {
983 	if (vcpu->arch.guest_state_protected)
984 		return;
985 
986 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
987 
988 		if (vcpu->arch.xcr0 != host_xcr0)
989 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
990 
991 		if (vcpu->arch.xsaves_enabled &&
992 		    vcpu->arch.ia32_xss != host_xss)
993 			wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
994 	}
995 
996 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
997 	if (static_cpu_has(X86_FEATURE_PKU) &&
998 	    vcpu->arch.pkru != vcpu->arch.host_pkru &&
999 	    ((vcpu->arch.xcr0 & XFEATURE_MASK_PKRU) ||
1000 	     kvm_read_cr4_bits(vcpu, X86_CR4_PKE)))
1001 		write_pkru(vcpu->arch.pkru);
1002 #endif /* CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS */
1003 }
1004 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
1005 
1006 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
1007 {
1008 	if (vcpu->arch.guest_state_protected)
1009 		return;
1010 
1011 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
1012 	if (static_cpu_has(X86_FEATURE_PKU) &&
1013 	    ((vcpu->arch.xcr0 & XFEATURE_MASK_PKRU) ||
1014 	     kvm_read_cr4_bits(vcpu, X86_CR4_PKE))) {
1015 		vcpu->arch.pkru = rdpkru();
1016 		if (vcpu->arch.pkru != vcpu->arch.host_pkru)
1017 			write_pkru(vcpu->arch.host_pkru);
1018 	}
1019 #endif /* CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS */
1020 
1021 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
1022 
1023 		if (vcpu->arch.xcr0 != host_xcr0)
1024 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
1025 
1026 		if (vcpu->arch.xsaves_enabled &&
1027 		    vcpu->arch.ia32_xss != host_xss)
1028 			wrmsrl(MSR_IA32_XSS, host_xss);
1029 	}
1030 
1031 }
1032 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
1033 
1034 #ifdef CONFIG_X86_64
1035 static inline u64 kvm_guest_supported_xfd(struct kvm_vcpu *vcpu)
1036 {
1037 	return vcpu->arch.guest_supported_xcr0 & XFEATURE_MASK_USER_DYNAMIC;
1038 }
1039 #endif
1040 
1041 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
1042 {
1043 	u64 xcr0 = xcr;
1044 	u64 old_xcr0 = vcpu->arch.xcr0;
1045 	u64 valid_bits;
1046 
1047 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
1048 	if (index != XCR_XFEATURE_ENABLED_MASK)
1049 		return 1;
1050 	if (!(xcr0 & XFEATURE_MASK_FP))
1051 		return 1;
1052 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
1053 		return 1;
1054 
1055 	/*
1056 	 * Do not allow the guest to set bits that we do not support
1057 	 * saving.  However, xcr0 bit 0 is always set, even if the
1058 	 * emulated CPU does not support XSAVE (see kvm_vcpu_reset()).
1059 	 */
1060 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
1061 	if (xcr0 & ~valid_bits)
1062 		return 1;
1063 
1064 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
1065 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
1066 		return 1;
1067 
1068 	if (xcr0 & XFEATURE_MASK_AVX512) {
1069 		if (!(xcr0 & XFEATURE_MASK_YMM))
1070 			return 1;
1071 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
1072 			return 1;
1073 	}
1074 
1075 	if ((xcr0 & XFEATURE_MASK_XTILE) &&
1076 	    ((xcr0 & XFEATURE_MASK_XTILE) != XFEATURE_MASK_XTILE))
1077 		return 1;
1078 
1079 	vcpu->arch.xcr0 = xcr0;
1080 
1081 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
1082 		kvm_update_cpuid_runtime(vcpu);
1083 	return 0;
1084 }
1085 
1086 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
1087 {
1088 	/* Note, #UD due to CR4.OSXSAVE=0 has priority over the intercept. */
1089 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
1090 	    __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
1091 		kvm_inject_gp(vcpu, 0);
1092 		return 1;
1093 	}
1094 
1095 	return kvm_skip_emulated_instruction(vcpu);
1096 }
1097 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1098 
1099 bool __kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1100 {
1101 	if (cr4 & cr4_reserved_bits)
1102 		return false;
1103 
1104 	if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1105 		return false;
1106 
1107 	return true;
1108 }
1109 EXPORT_SYMBOL_GPL(__kvm_is_valid_cr4);
1110 
1111 static bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1112 {
1113 	return __kvm_is_valid_cr4(vcpu, cr4) &&
1114 	       static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1115 }
1116 
1117 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1118 {
1119 	if ((cr4 ^ old_cr4) & KVM_MMU_CR4_ROLE_BITS)
1120 		kvm_mmu_reset_context(vcpu);
1121 
1122 	/*
1123 	 * If CR4.PCIDE is changed 0 -> 1, there is no need to flush the TLB
1124 	 * according to the SDM; however, stale prev_roots could be reused
1125 	 * incorrectly in the future after a MOV to CR3 with NOFLUSH=1, so we
1126 	 * free them all.  This is *not* a superset of KVM_REQ_TLB_FLUSH_GUEST
1127 	 * or KVM_REQ_TLB_FLUSH_CURRENT, because the hardware TLB is not flushed,
1128 	 * so fall through.
1129 	 */
1130 	if (!tdp_enabled &&
1131 	    (cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE))
1132 		kvm_mmu_unload(vcpu);
1133 
1134 	/*
1135 	 * The TLB has to be flushed for all PCIDs if any of the following
1136 	 * (architecturally required) changes happen:
1137 	 * - CR4.PCIDE is changed from 1 to 0
1138 	 * - CR4.PGE is toggled
1139 	 *
1140 	 * This is a superset of KVM_REQ_TLB_FLUSH_CURRENT.
1141 	 */
1142 	if (((cr4 ^ old_cr4) & X86_CR4_PGE) ||
1143 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1144 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1145 
1146 	/*
1147 	 * The TLB has to be flushed for the current PCID if any of the
1148 	 * following (architecturally required) changes happen:
1149 	 * - CR4.SMEP is changed from 0 to 1
1150 	 * - CR4.PAE is toggled
1151 	 */
1152 	else if (((cr4 ^ old_cr4) & X86_CR4_PAE) ||
1153 		 ((cr4 & X86_CR4_SMEP) && !(old_cr4 & X86_CR4_SMEP)))
1154 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1155 
1156 }
1157 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1158 
1159 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1160 {
1161 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
1162 
1163 	if (!kvm_is_valid_cr4(vcpu, cr4))
1164 		return 1;
1165 
1166 	if (is_long_mode(vcpu)) {
1167 		if (!(cr4 & X86_CR4_PAE))
1168 			return 1;
1169 		if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1170 			return 1;
1171 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1172 		   && ((cr4 ^ old_cr4) & X86_CR4_PDPTR_BITS)
1173 		   && !load_pdptrs(vcpu, kvm_read_cr3(vcpu)))
1174 		return 1;
1175 
1176 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1177 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1178 			return 1;
1179 
1180 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1181 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1182 			return 1;
1183 	}
1184 
1185 	static_call(kvm_x86_set_cr4)(vcpu, cr4);
1186 
1187 	kvm_post_set_cr4(vcpu, old_cr4, cr4);
1188 
1189 	return 0;
1190 }
1191 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1192 
1193 static void kvm_invalidate_pcid(struct kvm_vcpu *vcpu, unsigned long pcid)
1194 {
1195 	struct kvm_mmu *mmu = vcpu->arch.mmu;
1196 	unsigned long roots_to_free = 0;
1197 	int i;
1198 
1199 	/*
1200 	 * MOV CR3 and INVPCID are usually not intercepted when using TDP, but
1201 	 * this is reachable when running EPT=1 and unrestricted_guest=0,  and
1202 	 * also via the emulator.  KVM's TDP page tables are not in the scope of
1203 	 * the invalidation, but the guest's TLB entries need to be flushed as
1204 	 * the CPU may have cached entries in its TLB for the target PCID.
1205 	 */
1206 	if (unlikely(tdp_enabled)) {
1207 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1208 		return;
1209 	}
1210 
1211 	/*
1212 	 * If neither the current CR3 nor any of the prev_roots use the given
1213 	 * PCID, then nothing needs to be done here because a resync will
1214 	 * happen anyway before switching to any other CR3.
1215 	 */
1216 	if (kvm_get_active_pcid(vcpu) == pcid) {
1217 		kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
1218 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1219 	}
1220 
1221 	/*
1222 	 * If PCID is disabled, there is no need to free prev_roots even if the
1223 	 * PCIDs for them are also 0, because MOV to CR3 always flushes the TLB
1224 	 * with PCIDE=0.
1225 	 */
1226 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
1227 		return;
1228 
1229 	for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
1230 		if (kvm_get_pcid(vcpu, mmu->prev_roots[i].pgd) == pcid)
1231 			roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
1232 
1233 	kvm_mmu_free_roots(vcpu->kvm, mmu, roots_to_free);
1234 }
1235 
1236 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1237 {
1238 	bool skip_tlb_flush = false;
1239 	unsigned long pcid = 0;
1240 #ifdef CONFIG_X86_64
1241 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1242 
1243 	if (pcid_enabled) {
1244 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1245 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
1246 		pcid = cr3 & X86_CR3_PCID_MASK;
1247 	}
1248 #endif
1249 
1250 	/* PDPTRs are always reloaded for PAE paging. */
1251 	if (cr3 == kvm_read_cr3(vcpu) && !is_pae_paging(vcpu))
1252 		goto handle_tlb_flush;
1253 
1254 	/*
1255 	 * Do not condition the GPA check on long mode, this helper is used to
1256 	 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1257 	 * the current vCPU mode is accurate.
1258 	 */
1259 	if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1260 		return 1;
1261 
1262 	if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, cr3))
1263 		return 1;
1264 
1265 	if (cr3 != kvm_read_cr3(vcpu))
1266 		kvm_mmu_new_pgd(vcpu, cr3);
1267 
1268 	vcpu->arch.cr3 = cr3;
1269 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
1270 	/* Do not call post_set_cr3, we do not get here for confidential guests.  */
1271 
1272 handle_tlb_flush:
1273 	/*
1274 	 * A load of CR3 that flushes the TLB flushes only the current PCID,
1275 	 * even if PCID is disabled, in which case PCID=0 is flushed.  It's a
1276 	 * moot point in the end because _disabling_ PCID will flush all PCIDs,
1277 	 * and it's impossible to use a non-zero PCID when PCID is disabled,
1278 	 * i.e. only PCID=0 can be relevant.
1279 	 */
1280 	if (!skip_tlb_flush)
1281 		kvm_invalidate_pcid(vcpu, pcid);
1282 
1283 	return 0;
1284 }
1285 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1286 
1287 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1288 {
1289 	if (cr8 & CR8_RESERVED_BITS)
1290 		return 1;
1291 	if (lapic_in_kernel(vcpu))
1292 		kvm_lapic_set_tpr(vcpu, cr8);
1293 	else
1294 		vcpu->arch.cr8 = cr8;
1295 	return 0;
1296 }
1297 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1298 
1299 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1300 {
1301 	if (lapic_in_kernel(vcpu))
1302 		return kvm_lapic_get_cr8(vcpu);
1303 	else
1304 		return vcpu->arch.cr8;
1305 }
1306 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1307 
1308 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1309 {
1310 	int i;
1311 
1312 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1313 		for (i = 0; i < KVM_NR_DB_REGS; i++)
1314 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1315 	}
1316 }
1317 
1318 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1319 {
1320 	unsigned long dr7;
1321 
1322 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1323 		dr7 = vcpu->arch.guest_debug_dr7;
1324 	else
1325 		dr7 = vcpu->arch.dr7;
1326 	static_call(kvm_x86_set_dr7)(vcpu, dr7);
1327 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1328 	if (dr7 & DR7_BP_EN_MASK)
1329 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1330 }
1331 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1332 
1333 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1334 {
1335 	u64 fixed = DR6_FIXED_1;
1336 
1337 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1338 		fixed |= DR6_RTM;
1339 
1340 	if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1341 		fixed |= DR6_BUS_LOCK;
1342 	return fixed;
1343 }
1344 
1345 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1346 {
1347 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1348 
1349 	switch (dr) {
1350 	case 0 ... 3:
1351 		vcpu->arch.db[array_index_nospec(dr, size)] = val;
1352 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1353 			vcpu->arch.eff_db[dr] = val;
1354 		break;
1355 	case 4:
1356 	case 6:
1357 		if (!kvm_dr6_valid(val))
1358 			return 1; /* #GP */
1359 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1360 		break;
1361 	case 5:
1362 	default: /* 7 */
1363 		if (!kvm_dr7_valid(val))
1364 			return 1; /* #GP */
1365 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1366 		kvm_update_dr7(vcpu);
1367 		break;
1368 	}
1369 
1370 	return 0;
1371 }
1372 EXPORT_SYMBOL_GPL(kvm_set_dr);
1373 
1374 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1375 {
1376 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1377 
1378 	switch (dr) {
1379 	case 0 ... 3:
1380 		*val = vcpu->arch.db[array_index_nospec(dr, size)];
1381 		break;
1382 	case 4:
1383 	case 6:
1384 		*val = vcpu->arch.dr6;
1385 		break;
1386 	case 5:
1387 	default: /* 7 */
1388 		*val = vcpu->arch.dr7;
1389 		break;
1390 	}
1391 }
1392 EXPORT_SYMBOL_GPL(kvm_get_dr);
1393 
1394 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1395 {
1396 	u32 ecx = kvm_rcx_read(vcpu);
1397 	u64 data;
1398 
1399 	if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1400 		kvm_inject_gp(vcpu, 0);
1401 		return 1;
1402 	}
1403 
1404 	kvm_rax_write(vcpu, (u32)data);
1405 	kvm_rdx_write(vcpu, data >> 32);
1406 	return kvm_skip_emulated_instruction(vcpu);
1407 }
1408 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1409 
1410 /*
1411  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1412  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1413  *
1414  * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1415  * extract the supported MSRs from the related const lists.
1416  * msrs_to_save is selected from the msrs_to_save_all to reflect the
1417  * capabilities of the host cpu. This capabilities test skips MSRs that are
1418  * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1419  * may depend on host virtualization features rather than host cpu features.
1420  */
1421 
1422 static const u32 msrs_to_save_base[] = {
1423 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1424 	MSR_STAR,
1425 #ifdef CONFIG_X86_64
1426 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1427 #endif
1428 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1429 	MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1430 	MSR_IA32_SPEC_CTRL,
1431 	MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1432 	MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1433 	MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1434 	MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1435 	MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1436 	MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1437 	MSR_IA32_UMWAIT_CONTROL,
1438 
1439 	MSR_IA32_XFD, MSR_IA32_XFD_ERR,
1440 };
1441 
1442 static const u32 msrs_to_save_pmu[] = {
1443 	MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1444 	MSR_ARCH_PERFMON_FIXED_CTR0 + 2,
1445 	MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1446 	MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1447 	MSR_IA32_PEBS_ENABLE, MSR_IA32_DS_AREA, MSR_PEBS_DATA_CFG,
1448 
1449 	/* This part of MSRs should match KVM_INTEL_PMC_MAX_GENERIC. */
1450 	MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1451 	MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1452 	MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1453 	MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1454 	MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1455 	MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1456 	MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1457 	MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1458 
1459 	MSR_K7_EVNTSEL0, MSR_K7_EVNTSEL1, MSR_K7_EVNTSEL2, MSR_K7_EVNTSEL3,
1460 	MSR_K7_PERFCTR0, MSR_K7_PERFCTR1, MSR_K7_PERFCTR2, MSR_K7_PERFCTR3,
1461 
1462 	/* This part of MSRs should match KVM_AMD_PMC_MAX_GENERIC. */
1463 	MSR_F15H_PERF_CTL0, MSR_F15H_PERF_CTL1, MSR_F15H_PERF_CTL2,
1464 	MSR_F15H_PERF_CTL3, MSR_F15H_PERF_CTL4, MSR_F15H_PERF_CTL5,
1465 	MSR_F15H_PERF_CTR0, MSR_F15H_PERF_CTR1, MSR_F15H_PERF_CTR2,
1466 	MSR_F15H_PERF_CTR3, MSR_F15H_PERF_CTR4, MSR_F15H_PERF_CTR5,
1467 };
1468 
1469 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_base) +
1470 			ARRAY_SIZE(msrs_to_save_pmu)];
1471 static unsigned num_msrs_to_save;
1472 
1473 static const u32 emulated_msrs_all[] = {
1474 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1475 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1476 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1477 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1478 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1479 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1480 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1481 	HV_X64_MSR_RESET,
1482 	HV_X64_MSR_VP_INDEX,
1483 	HV_X64_MSR_VP_RUNTIME,
1484 	HV_X64_MSR_SCONTROL,
1485 	HV_X64_MSR_STIMER0_CONFIG,
1486 	HV_X64_MSR_VP_ASSIST_PAGE,
1487 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1488 	HV_X64_MSR_TSC_EMULATION_STATUS, HV_X64_MSR_TSC_INVARIANT_CONTROL,
1489 	HV_X64_MSR_SYNDBG_OPTIONS,
1490 	HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1491 	HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1492 	HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1493 
1494 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1495 	MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1496 
1497 	MSR_IA32_TSC_ADJUST,
1498 	MSR_IA32_TSC_DEADLINE,
1499 	MSR_IA32_ARCH_CAPABILITIES,
1500 	MSR_IA32_PERF_CAPABILITIES,
1501 	MSR_IA32_MISC_ENABLE,
1502 	MSR_IA32_MCG_STATUS,
1503 	MSR_IA32_MCG_CTL,
1504 	MSR_IA32_MCG_EXT_CTL,
1505 	MSR_IA32_SMBASE,
1506 	MSR_SMI_COUNT,
1507 	MSR_PLATFORM_INFO,
1508 	MSR_MISC_FEATURES_ENABLES,
1509 	MSR_AMD64_VIRT_SPEC_CTRL,
1510 	MSR_AMD64_TSC_RATIO,
1511 	MSR_IA32_POWER_CTL,
1512 	MSR_IA32_UCODE_REV,
1513 
1514 	/*
1515 	 * The following list leaves out MSRs whose values are determined
1516 	 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1517 	 * We always support the "true" VMX control MSRs, even if the host
1518 	 * processor does not, so I am putting these registers here rather
1519 	 * than in msrs_to_save_all.
1520 	 */
1521 	MSR_IA32_VMX_BASIC,
1522 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1523 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1524 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1525 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1526 	MSR_IA32_VMX_MISC,
1527 	MSR_IA32_VMX_CR0_FIXED0,
1528 	MSR_IA32_VMX_CR4_FIXED0,
1529 	MSR_IA32_VMX_VMCS_ENUM,
1530 	MSR_IA32_VMX_PROCBASED_CTLS2,
1531 	MSR_IA32_VMX_EPT_VPID_CAP,
1532 	MSR_IA32_VMX_VMFUNC,
1533 
1534 	MSR_K7_HWCR,
1535 	MSR_KVM_POLL_CONTROL,
1536 };
1537 
1538 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1539 static unsigned num_emulated_msrs;
1540 
1541 /*
1542  * List of msr numbers which are used to expose MSR-based features that
1543  * can be used by a hypervisor to validate requested CPU features.
1544  */
1545 static const u32 msr_based_features_all[] = {
1546 	MSR_IA32_VMX_BASIC,
1547 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1548 	MSR_IA32_VMX_PINBASED_CTLS,
1549 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1550 	MSR_IA32_VMX_PROCBASED_CTLS,
1551 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1552 	MSR_IA32_VMX_EXIT_CTLS,
1553 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1554 	MSR_IA32_VMX_ENTRY_CTLS,
1555 	MSR_IA32_VMX_MISC,
1556 	MSR_IA32_VMX_CR0_FIXED0,
1557 	MSR_IA32_VMX_CR0_FIXED1,
1558 	MSR_IA32_VMX_CR4_FIXED0,
1559 	MSR_IA32_VMX_CR4_FIXED1,
1560 	MSR_IA32_VMX_VMCS_ENUM,
1561 	MSR_IA32_VMX_PROCBASED_CTLS2,
1562 	MSR_IA32_VMX_EPT_VPID_CAP,
1563 	MSR_IA32_VMX_VMFUNC,
1564 
1565 	MSR_AMD64_DE_CFG,
1566 	MSR_IA32_UCODE_REV,
1567 	MSR_IA32_ARCH_CAPABILITIES,
1568 	MSR_IA32_PERF_CAPABILITIES,
1569 };
1570 
1571 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1572 static unsigned int num_msr_based_features;
1573 
1574 /*
1575  * Some IA32_ARCH_CAPABILITIES bits have dependencies on MSRs that KVM
1576  * does not yet virtualize. These include:
1577  *   10 - MISC_PACKAGE_CTRLS
1578  *   11 - ENERGY_FILTERING_CTL
1579  *   12 - DOITM
1580  *   18 - FB_CLEAR_CTRL
1581  *   21 - XAPIC_DISABLE_STATUS
1582  *   23 - OVERCLOCKING_STATUS
1583  */
1584 
1585 #define KVM_SUPPORTED_ARCH_CAP \
1586 	(ARCH_CAP_RDCL_NO | ARCH_CAP_IBRS_ALL | ARCH_CAP_RSBA | \
1587 	 ARCH_CAP_SKIP_VMENTRY_L1DFLUSH | ARCH_CAP_SSB_NO | ARCH_CAP_MDS_NO | \
1588 	 ARCH_CAP_PSCHANGE_MC_NO | ARCH_CAP_TSX_CTRL_MSR | ARCH_CAP_TAA_NO | \
1589 	 ARCH_CAP_SBDR_SSDP_NO | ARCH_CAP_FBSDP_NO | ARCH_CAP_PSDP_NO | \
1590 	 ARCH_CAP_FB_CLEAR | ARCH_CAP_RRSBA | ARCH_CAP_PBRSB_NO)
1591 
1592 static u64 kvm_get_arch_capabilities(void)
1593 {
1594 	u64 data = 0;
1595 
1596 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) {
1597 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1598 		data &= KVM_SUPPORTED_ARCH_CAP;
1599 	}
1600 
1601 	/*
1602 	 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1603 	 * the nested hypervisor runs with NX huge pages.  If it is not,
1604 	 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1605 	 * L1 guests, so it need not worry about its own (L2) guests.
1606 	 */
1607 	data |= ARCH_CAP_PSCHANGE_MC_NO;
1608 
1609 	/*
1610 	 * If we're doing cache flushes (either "always" or "cond")
1611 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1612 	 * If an outer hypervisor is doing the cache flush for us
1613 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1614 	 * capability to the guest too, and if EPT is disabled we're not
1615 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1616 	 * require a nested hypervisor to do a flush of its own.
1617 	 */
1618 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1619 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1620 
1621 	if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1622 		data |= ARCH_CAP_RDCL_NO;
1623 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1624 		data |= ARCH_CAP_SSB_NO;
1625 	if (!boot_cpu_has_bug(X86_BUG_MDS))
1626 		data |= ARCH_CAP_MDS_NO;
1627 
1628 	if (!boot_cpu_has(X86_FEATURE_RTM)) {
1629 		/*
1630 		 * If RTM=0 because the kernel has disabled TSX, the host might
1631 		 * have TAA_NO or TSX_CTRL.  Clear TAA_NO (the guest sees RTM=0
1632 		 * and therefore knows that there cannot be TAA) but keep
1633 		 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1634 		 * and we want to allow migrating those guests to tsx=off hosts.
1635 		 */
1636 		data &= ~ARCH_CAP_TAA_NO;
1637 	} else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1638 		data |= ARCH_CAP_TAA_NO;
1639 	} else {
1640 		/*
1641 		 * Nothing to do here; we emulate TSX_CTRL if present on the
1642 		 * host so the guest can choose between disabling TSX or
1643 		 * using VERW to clear CPU buffers.
1644 		 */
1645 	}
1646 
1647 	return data;
1648 }
1649 
1650 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1651 {
1652 	switch (msr->index) {
1653 	case MSR_IA32_ARCH_CAPABILITIES:
1654 		msr->data = kvm_get_arch_capabilities();
1655 		break;
1656 	case MSR_IA32_PERF_CAPABILITIES:
1657 		msr->data = kvm_caps.supported_perf_cap;
1658 		break;
1659 	case MSR_IA32_UCODE_REV:
1660 		rdmsrl_safe(msr->index, &msr->data);
1661 		break;
1662 	default:
1663 		return static_call(kvm_x86_get_msr_feature)(msr);
1664 	}
1665 	return 0;
1666 }
1667 
1668 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1669 {
1670 	struct kvm_msr_entry msr;
1671 	int r;
1672 
1673 	msr.index = index;
1674 	r = kvm_get_msr_feature(&msr);
1675 
1676 	if (r == KVM_MSR_RET_INVALID) {
1677 		/* Unconditionally clear the output for simplicity */
1678 		*data = 0;
1679 		if (kvm_msr_ignored_check(index, 0, false))
1680 			r = 0;
1681 	}
1682 
1683 	if (r)
1684 		return r;
1685 
1686 	*data = msr.data;
1687 
1688 	return 0;
1689 }
1690 
1691 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1692 {
1693 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1694 		return false;
1695 
1696 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1697 		return false;
1698 
1699 	if (efer & (EFER_LME | EFER_LMA) &&
1700 	    !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1701 		return false;
1702 
1703 	if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1704 		return false;
1705 
1706 	return true;
1707 
1708 }
1709 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1710 {
1711 	if (efer & efer_reserved_bits)
1712 		return false;
1713 
1714 	return __kvm_valid_efer(vcpu, efer);
1715 }
1716 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1717 
1718 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1719 {
1720 	u64 old_efer = vcpu->arch.efer;
1721 	u64 efer = msr_info->data;
1722 	int r;
1723 
1724 	if (efer & efer_reserved_bits)
1725 		return 1;
1726 
1727 	if (!msr_info->host_initiated) {
1728 		if (!__kvm_valid_efer(vcpu, efer))
1729 			return 1;
1730 
1731 		if (is_paging(vcpu) &&
1732 		    (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1733 			return 1;
1734 	}
1735 
1736 	efer &= ~EFER_LMA;
1737 	efer |= vcpu->arch.efer & EFER_LMA;
1738 
1739 	r = static_call(kvm_x86_set_efer)(vcpu, efer);
1740 	if (r) {
1741 		WARN_ON(r > 0);
1742 		return r;
1743 	}
1744 
1745 	if ((efer ^ old_efer) & KVM_MMU_EFER_ROLE_BITS)
1746 		kvm_mmu_reset_context(vcpu);
1747 
1748 	return 0;
1749 }
1750 
1751 void kvm_enable_efer_bits(u64 mask)
1752 {
1753        efer_reserved_bits &= ~mask;
1754 }
1755 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1756 
1757 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1758 {
1759 	struct kvm_x86_msr_filter *msr_filter;
1760 	struct msr_bitmap_range *ranges;
1761 	struct kvm *kvm = vcpu->kvm;
1762 	bool allowed;
1763 	int idx;
1764 	u32 i;
1765 
1766 	/* x2APIC MSRs do not support filtering. */
1767 	if (index >= 0x800 && index <= 0x8ff)
1768 		return true;
1769 
1770 	idx = srcu_read_lock(&kvm->srcu);
1771 
1772 	msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1773 	if (!msr_filter) {
1774 		allowed = true;
1775 		goto out;
1776 	}
1777 
1778 	allowed = msr_filter->default_allow;
1779 	ranges = msr_filter->ranges;
1780 
1781 	for (i = 0; i < msr_filter->count; i++) {
1782 		u32 start = ranges[i].base;
1783 		u32 end = start + ranges[i].nmsrs;
1784 		u32 flags = ranges[i].flags;
1785 		unsigned long *bitmap = ranges[i].bitmap;
1786 
1787 		if ((index >= start) && (index < end) && (flags & type)) {
1788 			allowed = !!test_bit(index - start, bitmap);
1789 			break;
1790 		}
1791 	}
1792 
1793 out:
1794 	srcu_read_unlock(&kvm->srcu, idx);
1795 
1796 	return allowed;
1797 }
1798 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1799 
1800 /*
1801  * Write @data into the MSR specified by @index.  Select MSR specific fault
1802  * checks are bypassed if @host_initiated is %true.
1803  * Returns 0 on success, non-0 otherwise.
1804  * Assumes vcpu_load() was already called.
1805  */
1806 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1807 			 bool host_initiated)
1808 {
1809 	struct msr_data msr;
1810 
1811 	switch (index) {
1812 	case MSR_FS_BASE:
1813 	case MSR_GS_BASE:
1814 	case MSR_KERNEL_GS_BASE:
1815 	case MSR_CSTAR:
1816 	case MSR_LSTAR:
1817 		if (is_noncanonical_address(data, vcpu))
1818 			return 1;
1819 		break;
1820 	case MSR_IA32_SYSENTER_EIP:
1821 	case MSR_IA32_SYSENTER_ESP:
1822 		/*
1823 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1824 		 * non-canonical address is written on Intel but not on
1825 		 * AMD (which ignores the top 32-bits, because it does
1826 		 * not implement 64-bit SYSENTER).
1827 		 *
1828 		 * 64-bit code should hence be able to write a non-canonical
1829 		 * value on AMD.  Making the address canonical ensures that
1830 		 * vmentry does not fail on Intel after writing a non-canonical
1831 		 * value, and that something deterministic happens if the guest
1832 		 * invokes 64-bit SYSENTER.
1833 		 */
1834 		data = __canonical_address(data, vcpu_virt_addr_bits(vcpu));
1835 		break;
1836 	case MSR_TSC_AUX:
1837 		if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1838 			return 1;
1839 
1840 		if (!host_initiated &&
1841 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1842 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1843 			return 1;
1844 
1845 		/*
1846 		 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1847 		 * incomplete and conflicting architectural behavior.  Current
1848 		 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1849 		 * reserved and always read as zeros.  Enforce Intel's reserved
1850 		 * bits check if and only if the guest CPU is Intel, and clear
1851 		 * the bits in all other cases.  This ensures cross-vendor
1852 		 * migration will provide consistent behavior for the guest.
1853 		 */
1854 		if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1855 			return 1;
1856 
1857 		data = (u32)data;
1858 		break;
1859 	}
1860 
1861 	msr.data = data;
1862 	msr.index = index;
1863 	msr.host_initiated = host_initiated;
1864 
1865 	return static_call(kvm_x86_set_msr)(vcpu, &msr);
1866 }
1867 
1868 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1869 				     u32 index, u64 data, bool host_initiated)
1870 {
1871 	int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1872 
1873 	if (ret == KVM_MSR_RET_INVALID)
1874 		if (kvm_msr_ignored_check(index, data, true))
1875 			ret = 0;
1876 
1877 	return ret;
1878 }
1879 
1880 /*
1881  * Read the MSR specified by @index into @data.  Select MSR specific fault
1882  * checks are bypassed if @host_initiated is %true.
1883  * Returns 0 on success, non-0 otherwise.
1884  * Assumes vcpu_load() was already called.
1885  */
1886 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1887 		  bool host_initiated)
1888 {
1889 	struct msr_data msr;
1890 	int ret;
1891 
1892 	switch (index) {
1893 	case MSR_TSC_AUX:
1894 		if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1895 			return 1;
1896 
1897 		if (!host_initiated &&
1898 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1899 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1900 			return 1;
1901 		break;
1902 	}
1903 
1904 	msr.index = index;
1905 	msr.host_initiated = host_initiated;
1906 
1907 	ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1908 	if (!ret)
1909 		*data = msr.data;
1910 	return ret;
1911 }
1912 
1913 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1914 				     u32 index, u64 *data, bool host_initiated)
1915 {
1916 	int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1917 
1918 	if (ret == KVM_MSR_RET_INVALID) {
1919 		/* Unconditionally clear *data for simplicity */
1920 		*data = 0;
1921 		if (kvm_msr_ignored_check(index, 0, false))
1922 			ret = 0;
1923 	}
1924 
1925 	return ret;
1926 }
1927 
1928 static int kvm_get_msr_with_filter(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1929 {
1930 	if (!kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1931 		return KVM_MSR_RET_FILTERED;
1932 	return kvm_get_msr_ignored_check(vcpu, index, data, false);
1933 }
1934 
1935 static int kvm_set_msr_with_filter(struct kvm_vcpu *vcpu, u32 index, u64 data)
1936 {
1937 	if (!kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1938 		return KVM_MSR_RET_FILTERED;
1939 	return kvm_set_msr_ignored_check(vcpu, index, data, false);
1940 }
1941 
1942 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1943 {
1944 	return kvm_get_msr_ignored_check(vcpu, index, data, false);
1945 }
1946 EXPORT_SYMBOL_GPL(kvm_get_msr);
1947 
1948 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1949 {
1950 	return kvm_set_msr_ignored_check(vcpu, index, data, false);
1951 }
1952 EXPORT_SYMBOL_GPL(kvm_set_msr);
1953 
1954 static void complete_userspace_rdmsr(struct kvm_vcpu *vcpu)
1955 {
1956 	if (!vcpu->run->msr.error) {
1957 		kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1958 		kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1959 	}
1960 }
1961 
1962 static int complete_emulated_msr_access(struct kvm_vcpu *vcpu)
1963 {
1964 	return complete_emulated_insn_gp(vcpu, vcpu->run->msr.error);
1965 }
1966 
1967 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1968 {
1969 	complete_userspace_rdmsr(vcpu);
1970 	return complete_emulated_msr_access(vcpu);
1971 }
1972 
1973 static int complete_fast_msr_access(struct kvm_vcpu *vcpu)
1974 {
1975 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1976 }
1977 
1978 static int complete_fast_rdmsr(struct kvm_vcpu *vcpu)
1979 {
1980 	complete_userspace_rdmsr(vcpu);
1981 	return complete_fast_msr_access(vcpu);
1982 }
1983 
1984 static u64 kvm_msr_reason(int r)
1985 {
1986 	switch (r) {
1987 	case KVM_MSR_RET_INVALID:
1988 		return KVM_MSR_EXIT_REASON_UNKNOWN;
1989 	case KVM_MSR_RET_FILTERED:
1990 		return KVM_MSR_EXIT_REASON_FILTER;
1991 	default:
1992 		return KVM_MSR_EXIT_REASON_INVAL;
1993 	}
1994 }
1995 
1996 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1997 			      u32 exit_reason, u64 data,
1998 			      int (*completion)(struct kvm_vcpu *vcpu),
1999 			      int r)
2000 {
2001 	u64 msr_reason = kvm_msr_reason(r);
2002 
2003 	/* Check if the user wanted to know about this MSR fault */
2004 	if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
2005 		return 0;
2006 
2007 	vcpu->run->exit_reason = exit_reason;
2008 	vcpu->run->msr.error = 0;
2009 	memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
2010 	vcpu->run->msr.reason = msr_reason;
2011 	vcpu->run->msr.index = index;
2012 	vcpu->run->msr.data = data;
2013 	vcpu->arch.complete_userspace_io = completion;
2014 
2015 	return 1;
2016 }
2017 
2018 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
2019 {
2020 	u32 ecx = kvm_rcx_read(vcpu);
2021 	u64 data;
2022 	int r;
2023 
2024 	r = kvm_get_msr_with_filter(vcpu, ecx, &data);
2025 
2026 	if (!r) {
2027 		trace_kvm_msr_read(ecx, data);
2028 
2029 		kvm_rax_write(vcpu, data & -1u);
2030 		kvm_rdx_write(vcpu, (data >> 32) & -1u);
2031 	} else {
2032 		/* MSR read failed? See if we should ask user space */
2033 		if (kvm_msr_user_space(vcpu, ecx, KVM_EXIT_X86_RDMSR, 0,
2034 				       complete_fast_rdmsr, r))
2035 			return 0;
2036 		trace_kvm_msr_read_ex(ecx);
2037 	}
2038 
2039 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
2040 }
2041 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
2042 
2043 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
2044 {
2045 	u32 ecx = kvm_rcx_read(vcpu);
2046 	u64 data = kvm_read_edx_eax(vcpu);
2047 	int r;
2048 
2049 	r = kvm_set_msr_with_filter(vcpu, ecx, data);
2050 
2051 	if (!r) {
2052 		trace_kvm_msr_write(ecx, data);
2053 	} else {
2054 		/* MSR write failed? See if we should ask user space */
2055 		if (kvm_msr_user_space(vcpu, ecx, KVM_EXIT_X86_WRMSR, data,
2056 				       complete_fast_msr_access, r))
2057 			return 0;
2058 		/* Signal all other negative errors to userspace */
2059 		if (r < 0)
2060 			return r;
2061 		trace_kvm_msr_write_ex(ecx, data);
2062 	}
2063 
2064 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
2065 }
2066 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
2067 
2068 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
2069 {
2070 	return kvm_skip_emulated_instruction(vcpu);
2071 }
2072 
2073 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
2074 {
2075 	/* Treat an INVD instruction as a NOP and just skip it. */
2076 	return kvm_emulate_as_nop(vcpu);
2077 }
2078 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
2079 
2080 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
2081 {
2082 	kvm_queue_exception(vcpu, UD_VECTOR);
2083 	return 1;
2084 }
2085 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
2086 
2087 
2088 static int kvm_emulate_monitor_mwait(struct kvm_vcpu *vcpu, const char *insn)
2089 {
2090 	if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MWAIT_NEVER_UD_FAULTS) &&
2091 	    !guest_cpuid_has(vcpu, X86_FEATURE_MWAIT))
2092 		return kvm_handle_invalid_op(vcpu);
2093 
2094 	pr_warn_once("%s instruction emulated as NOP!\n", insn);
2095 	return kvm_emulate_as_nop(vcpu);
2096 }
2097 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
2098 {
2099 	return kvm_emulate_monitor_mwait(vcpu, "MWAIT");
2100 }
2101 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
2102 
2103 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
2104 {
2105 	return kvm_emulate_monitor_mwait(vcpu, "MONITOR");
2106 }
2107 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
2108 
2109 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
2110 {
2111 	xfer_to_guest_mode_prepare();
2112 	return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
2113 		xfer_to_guest_mode_work_pending();
2114 }
2115 
2116 /*
2117  * The fast path for frequent and performance sensitive wrmsr emulation,
2118  * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
2119  * the latency of virtual IPI by avoiding the expensive bits of transitioning
2120  * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
2121  * other cases which must be called after interrupts are enabled on the host.
2122  */
2123 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
2124 {
2125 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
2126 		return 1;
2127 
2128 	if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
2129 	    ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
2130 	    ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
2131 	    ((u32)(data >> 32) != X2APIC_BROADCAST))
2132 		return kvm_x2apic_icr_write(vcpu->arch.apic, data);
2133 
2134 	return 1;
2135 }
2136 
2137 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
2138 {
2139 	if (!kvm_can_use_hv_timer(vcpu))
2140 		return 1;
2141 
2142 	kvm_set_lapic_tscdeadline_msr(vcpu, data);
2143 	return 0;
2144 }
2145 
2146 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
2147 {
2148 	u32 msr = kvm_rcx_read(vcpu);
2149 	u64 data;
2150 	fastpath_t ret = EXIT_FASTPATH_NONE;
2151 
2152 	switch (msr) {
2153 	case APIC_BASE_MSR + (APIC_ICR >> 4):
2154 		data = kvm_read_edx_eax(vcpu);
2155 		if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
2156 			kvm_skip_emulated_instruction(vcpu);
2157 			ret = EXIT_FASTPATH_EXIT_HANDLED;
2158 		}
2159 		break;
2160 	case MSR_IA32_TSC_DEADLINE:
2161 		data = kvm_read_edx_eax(vcpu);
2162 		if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
2163 			kvm_skip_emulated_instruction(vcpu);
2164 			ret = EXIT_FASTPATH_REENTER_GUEST;
2165 		}
2166 		break;
2167 	default:
2168 		break;
2169 	}
2170 
2171 	if (ret != EXIT_FASTPATH_NONE)
2172 		trace_kvm_msr_write(msr, data);
2173 
2174 	return ret;
2175 }
2176 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
2177 
2178 /*
2179  * Adapt set_msr() to msr_io()'s calling convention
2180  */
2181 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2182 {
2183 	return kvm_get_msr_ignored_check(vcpu, index, data, true);
2184 }
2185 
2186 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2187 {
2188 	return kvm_set_msr_ignored_check(vcpu, index, *data, true);
2189 }
2190 
2191 #ifdef CONFIG_X86_64
2192 struct pvclock_clock {
2193 	int vclock_mode;
2194 	u64 cycle_last;
2195 	u64 mask;
2196 	u32 mult;
2197 	u32 shift;
2198 	u64 base_cycles;
2199 	u64 offset;
2200 };
2201 
2202 struct pvclock_gtod_data {
2203 	seqcount_t	seq;
2204 
2205 	struct pvclock_clock clock; /* extract of a clocksource struct */
2206 	struct pvclock_clock raw_clock; /* extract of a clocksource struct */
2207 
2208 	ktime_t		offs_boot;
2209 	u64		wall_time_sec;
2210 };
2211 
2212 static struct pvclock_gtod_data pvclock_gtod_data;
2213 
2214 static void update_pvclock_gtod(struct timekeeper *tk)
2215 {
2216 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2217 
2218 	write_seqcount_begin(&vdata->seq);
2219 
2220 	/* copy pvclock gtod data */
2221 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->vdso_clock_mode;
2222 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
2223 	vdata->clock.mask		= tk->tkr_mono.mask;
2224 	vdata->clock.mult		= tk->tkr_mono.mult;
2225 	vdata->clock.shift		= tk->tkr_mono.shift;
2226 	vdata->clock.base_cycles	= tk->tkr_mono.xtime_nsec;
2227 	vdata->clock.offset		= tk->tkr_mono.base;
2228 
2229 	vdata->raw_clock.vclock_mode	= tk->tkr_raw.clock->vdso_clock_mode;
2230 	vdata->raw_clock.cycle_last	= tk->tkr_raw.cycle_last;
2231 	vdata->raw_clock.mask		= tk->tkr_raw.mask;
2232 	vdata->raw_clock.mult		= tk->tkr_raw.mult;
2233 	vdata->raw_clock.shift		= tk->tkr_raw.shift;
2234 	vdata->raw_clock.base_cycles	= tk->tkr_raw.xtime_nsec;
2235 	vdata->raw_clock.offset		= tk->tkr_raw.base;
2236 
2237 	vdata->wall_time_sec            = tk->xtime_sec;
2238 
2239 	vdata->offs_boot		= tk->offs_boot;
2240 
2241 	write_seqcount_end(&vdata->seq);
2242 }
2243 
2244 static s64 get_kvmclock_base_ns(void)
2245 {
2246 	/* Count up from boot time, but with the frequency of the raw clock.  */
2247 	return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2248 }
2249 #else
2250 static s64 get_kvmclock_base_ns(void)
2251 {
2252 	/* Master clock not used, so we can just use CLOCK_BOOTTIME.  */
2253 	return ktime_get_boottime_ns();
2254 }
2255 #endif
2256 
2257 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2258 {
2259 	int version;
2260 	int r;
2261 	struct pvclock_wall_clock wc;
2262 	u32 wc_sec_hi;
2263 	u64 wall_nsec;
2264 
2265 	if (!wall_clock)
2266 		return;
2267 
2268 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2269 	if (r)
2270 		return;
2271 
2272 	if (version & 1)
2273 		++version;  /* first time write, random junk */
2274 
2275 	++version;
2276 
2277 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2278 		return;
2279 
2280 	/*
2281 	 * The guest calculates current wall clock time by adding
2282 	 * system time (updated by kvm_guest_time_update below) to the
2283 	 * wall clock specified here.  We do the reverse here.
2284 	 */
2285 	wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2286 
2287 	wc.nsec = do_div(wall_nsec, 1000000000);
2288 	wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2289 	wc.version = version;
2290 
2291 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2292 
2293 	if (sec_hi_ofs) {
2294 		wc_sec_hi = wall_nsec >> 32;
2295 		kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2296 				&wc_sec_hi, sizeof(wc_sec_hi));
2297 	}
2298 
2299 	version++;
2300 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2301 }
2302 
2303 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2304 				  bool old_msr, bool host_initiated)
2305 {
2306 	struct kvm_arch *ka = &vcpu->kvm->arch;
2307 
2308 	if (vcpu->vcpu_id == 0 && !host_initiated) {
2309 		if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2310 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2311 
2312 		ka->boot_vcpu_runs_old_kvmclock = old_msr;
2313 	}
2314 
2315 	vcpu->arch.time = system_time;
2316 	kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2317 
2318 	/* we verify if the enable bit is set... */
2319 	if (system_time & 1)
2320 		kvm_gpc_activate(&vcpu->arch.pv_time, system_time & ~1ULL,
2321 				 sizeof(struct pvclock_vcpu_time_info));
2322 	else
2323 		kvm_gpc_deactivate(&vcpu->arch.pv_time);
2324 
2325 	return;
2326 }
2327 
2328 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2329 {
2330 	do_shl32_div32(dividend, divisor);
2331 	return dividend;
2332 }
2333 
2334 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2335 			       s8 *pshift, u32 *pmultiplier)
2336 {
2337 	uint64_t scaled64;
2338 	int32_t  shift = 0;
2339 	uint64_t tps64;
2340 	uint32_t tps32;
2341 
2342 	tps64 = base_hz;
2343 	scaled64 = scaled_hz;
2344 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2345 		tps64 >>= 1;
2346 		shift--;
2347 	}
2348 
2349 	tps32 = (uint32_t)tps64;
2350 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2351 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2352 			scaled64 >>= 1;
2353 		else
2354 			tps32 <<= 1;
2355 		shift++;
2356 	}
2357 
2358 	*pshift = shift;
2359 	*pmultiplier = div_frac(scaled64, tps32);
2360 }
2361 
2362 #ifdef CONFIG_X86_64
2363 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2364 #endif
2365 
2366 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2367 static unsigned long max_tsc_khz;
2368 
2369 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2370 {
2371 	u64 v = (u64)khz * (1000000 + ppm);
2372 	do_div(v, 1000000);
2373 	return v;
2374 }
2375 
2376 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier);
2377 
2378 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2379 {
2380 	u64 ratio;
2381 
2382 	/* Guest TSC same frequency as host TSC? */
2383 	if (!scale) {
2384 		kvm_vcpu_write_tsc_multiplier(vcpu, kvm_caps.default_tsc_scaling_ratio);
2385 		return 0;
2386 	}
2387 
2388 	/* TSC scaling supported? */
2389 	if (!kvm_caps.has_tsc_control) {
2390 		if (user_tsc_khz > tsc_khz) {
2391 			vcpu->arch.tsc_catchup = 1;
2392 			vcpu->arch.tsc_always_catchup = 1;
2393 			return 0;
2394 		} else {
2395 			pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2396 			return -1;
2397 		}
2398 	}
2399 
2400 	/* TSC scaling required  - calculate ratio */
2401 	ratio = mul_u64_u32_div(1ULL << kvm_caps.tsc_scaling_ratio_frac_bits,
2402 				user_tsc_khz, tsc_khz);
2403 
2404 	if (ratio == 0 || ratio >= kvm_caps.max_tsc_scaling_ratio) {
2405 		pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2406 			            user_tsc_khz);
2407 		return -1;
2408 	}
2409 
2410 	kvm_vcpu_write_tsc_multiplier(vcpu, ratio);
2411 	return 0;
2412 }
2413 
2414 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2415 {
2416 	u32 thresh_lo, thresh_hi;
2417 	int use_scaling = 0;
2418 
2419 	/* tsc_khz can be zero if TSC calibration fails */
2420 	if (user_tsc_khz == 0) {
2421 		/* set tsc_scaling_ratio to a safe value */
2422 		kvm_vcpu_write_tsc_multiplier(vcpu, kvm_caps.default_tsc_scaling_ratio);
2423 		return -1;
2424 	}
2425 
2426 	/* Compute a scale to convert nanoseconds in TSC cycles */
2427 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2428 			   &vcpu->arch.virtual_tsc_shift,
2429 			   &vcpu->arch.virtual_tsc_mult);
2430 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2431 
2432 	/*
2433 	 * Compute the variation in TSC rate which is acceptable
2434 	 * within the range of tolerance and decide if the
2435 	 * rate being applied is within that bounds of the hardware
2436 	 * rate.  If so, no scaling or compensation need be done.
2437 	 */
2438 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2439 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2440 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2441 		pr_debug("requested TSC rate %u falls outside tolerance [%u,%u]\n",
2442 			 user_tsc_khz, thresh_lo, thresh_hi);
2443 		use_scaling = 1;
2444 	}
2445 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2446 }
2447 
2448 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2449 {
2450 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2451 				      vcpu->arch.virtual_tsc_mult,
2452 				      vcpu->arch.virtual_tsc_shift);
2453 	tsc += vcpu->arch.this_tsc_write;
2454 	return tsc;
2455 }
2456 
2457 #ifdef CONFIG_X86_64
2458 static inline int gtod_is_based_on_tsc(int mode)
2459 {
2460 	return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2461 }
2462 #endif
2463 
2464 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2465 {
2466 #ifdef CONFIG_X86_64
2467 	bool vcpus_matched;
2468 	struct kvm_arch *ka = &vcpu->kvm->arch;
2469 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2470 
2471 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2472 			 atomic_read(&vcpu->kvm->online_vcpus));
2473 
2474 	/*
2475 	 * Once the masterclock is enabled, always perform request in
2476 	 * order to update it.
2477 	 *
2478 	 * In order to enable masterclock, the host clocksource must be TSC
2479 	 * and the vcpus need to have matched TSCs.  When that happens,
2480 	 * perform request to enable masterclock.
2481 	 */
2482 	if (ka->use_master_clock ||
2483 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2484 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2485 
2486 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2487 			    atomic_read(&vcpu->kvm->online_vcpus),
2488 		            ka->use_master_clock, gtod->clock.vclock_mode);
2489 #endif
2490 }
2491 
2492 /*
2493  * Multiply tsc by a fixed point number represented by ratio.
2494  *
2495  * The most significant 64-N bits (mult) of ratio represent the
2496  * integral part of the fixed point number; the remaining N bits
2497  * (frac) represent the fractional part, ie. ratio represents a fixed
2498  * point number (mult + frac * 2^(-N)).
2499  *
2500  * N equals to kvm_caps.tsc_scaling_ratio_frac_bits.
2501  */
2502 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2503 {
2504 	return mul_u64_u64_shr(tsc, ratio, kvm_caps.tsc_scaling_ratio_frac_bits);
2505 }
2506 
2507 u64 kvm_scale_tsc(u64 tsc, u64 ratio)
2508 {
2509 	u64 _tsc = tsc;
2510 
2511 	if (ratio != kvm_caps.default_tsc_scaling_ratio)
2512 		_tsc = __scale_tsc(ratio, tsc);
2513 
2514 	return _tsc;
2515 }
2516 
2517 static u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2518 {
2519 	u64 tsc;
2520 
2521 	tsc = kvm_scale_tsc(rdtsc(), vcpu->arch.l1_tsc_scaling_ratio);
2522 
2523 	return target_tsc - tsc;
2524 }
2525 
2526 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2527 {
2528 	return vcpu->arch.l1_tsc_offset +
2529 		kvm_scale_tsc(host_tsc, vcpu->arch.l1_tsc_scaling_ratio);
2530 }
2531 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2532 
2533 u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier)
2534 {
2535 	u64 nested_offset;
2536 
2537 	if (l2_multiplier == kvm_caps.default_tsc_scaling_ratio)
2538 		nested_offset = l1_offset;
2539 	else
2540 		nested_offset = mul_s64_u64_shr((s64) l1_offset, l2_multiplier,
2541 						kvm_caps.tsc_scaling_ratio_frac_bits);
2542 
2543 	nested_offset += l2_offset;
2544 	return nested_offset;
2545 }
2546 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_offset);
2547 
2548 u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier)
2549 {
2550 	if (l2_multiplier != kvm_caps.default_tsc_scaling_ratio)
2551 		return mul_u64_u64_shr(l1_multiplier, l2_multiplier,
2552 				       kvm_caps.tsc_scaling_ratio_frac_bits);
2553 
2554 	return l1_multiplier;
2555 }
2556 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_multiplier);
2557 
2558 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset)
2559 {
2560 	trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2561 				   vcpu->arch.l1_tsc_offset,
2562 				   l1_offset);
2563 
2564 	vcpu->arch.l1_tsc_offset = l1_offset;
2565 
2566 	/*
2567 	 * If we are here because L1 chose not to trap WRMSR to TSC then
2568 	 * according to the spec this should set L1's TSC (as opposed to
2569 	 * setting L1's offset for L2).
2570 	 */
2571 	if (is_guest_mode(vcpu))
2572 		vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2573 			l1_offset,
2574 			static_call(kvm_x86_get_l2_tsc_offset)(vcpu),
2575 			static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2576 	else
2577 		vcpu->arch.tsc_offset = l1_offset;
2578 
2579 	static_call(kvm_x86_write_tsc_offset)(vcpu, vcpu->arch.tsc_offset);
2580 }
2581 
2582 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier)
2583 {
2584 	vcpu->arch.l1_tsc_scaling_ratio = l1_multiplier;
2585 
2586 	/* Userspace is changing the multiplier while L2 is active */
2587 	if (is_guest_mode(vcpu))
2588 		vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2589 			l1_multiplier,
2590 			static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2591 	else
2592 		vcpu->arch.tsc_scaling_ratio = l1_multiplier;
2593 
2594 	if (kvm_caps.has_tsc_control)
2595 		static_call(kvm_x86_write_tsc_multiplier)(
2596 			vcpu, vcpu->arch.tsc_scaling_ratio);
2597 }
2598 
2599 static inline bool kvm_check_tsc_unstable(void)
2600 {
2601 #ifdef CONFIG_X86_64
2602 	/*
2603 	 * TSC is marked unstable when we're running on Hyper-V,
2604 	 * 'TSC page' clocksource is good.
2605 	 */
2606 	if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2607 		return false;
2608 #endif
2609 	return check_tsc_unstable();
2610 }
2611 
2612 /*
2613  * Infers attempts to synchronize the guest's tsc from host writes. Sets the
2614  * offset for the vcpu and tracks the TSC matching generation that the vcpu
2615  * participates in.
2616  */
2617 static void __kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 offset, u64 tsc,
2618 				  u64 ns, bool matched)
2619 {
2620 	struct kvm *kvm = vcpu->kvm;
2621 
2622 	lockdep_assert_held(&kvm->arch.tsc_write_lock);
2623 
2624 	/*
2625 	 * We also track th most recent recorded KHZ, write and time to
2626 	 * allow the matching interval to be extended at each write.
2627 	 */
2628 	kvm->arch.last_tsc_nsec = ns;
2629 	kvm->arch.last_tsc_write = tsc;
2630 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2631 	kvm->arch.last_tsc_offset = offset;
2632 
2633 	vcpu->arch.last_guest_tsc = tsc;
2634 
2635 	kvm_vcpu_write_tsc_offset(vcpu, offset);
2636 
2637 	if (!matched) {
2638 		/*
2639 		 * We split periods of matched TSC writes into generations.
2640 		 * For each generation, we track the original measured
2641 		 * nanosecond time, offset, and write, so if TSCs are in
2642 		 * sync, we can match exact offset, and if not, we can match
2643 		 * exact software computation in compute_guest_tsc()
2644 		 *
2645 		 * These values are tracked in kvm->arch.cur_xxx variables.
2646 		 */
2647 		kvm->arch.cur_tsc_generation++;
2648 		kvm->arch.cur_tsc_nsec = ns;
2649 		kvm->arch.cur_tsc_write = tsc;
2650 		kvm->arch.cur_tsc_offset = offset;
2651 		kvm->arch.nr_vcpus_matched_tsc = 0;
2652 	} else if (vcpu->arch.this_tsc_generation != kvm->arch.cur_tsc_generation) {
2653 		kvm->arch.nr_vcpus_matched_tsc++;
2654 	}
2655 
2656 	/* Keep track of which generation this VCPU has synchronized to */
2657 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2658 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2659 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2660 
2661 	kvm_track_tsc_matching(vcpu);
2662 }
2663 
2664 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2665 {
2666 	struct kvm *kvm = vcpu->kvm;
2667 	u64 offset, ns, elapsed;
2668 	unsigned long flags;
2669 	bool matched = false;
2670 	bool synchronizing = false;
2671 
2672 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2673 	offset = kvm_compute_l1_tsc_offset(vcpu, data);
2674 	ns = get_kvmclock_base_ns();
2675 	elapsed = ns - kvm->arch.last_tsc_nsec;
2676 
2677 	if (vcpu->arch.virtual_tsc_khz) {
2678 		if (data == 0) {
2679 			/*
2680 			 * detection of vcpu initialization -- need to sync
2681 			 * with other vCPUs. This particularly helps to keep
2682 			 * kvm_clock stable after CPU hotplug
2683 			 */
2684 			synchronizing = true;
2685 		} else {
2686 			u64 tsc_exp = kvm->arch.last_tsc_write +
2687 						nsec_to_cycles(vcpu, elapsed);
2688 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2689 			/*
2690 			 * Special case: TSC write with a small delta (1 second)
2691 			 * of virtual cycle time against real time is
2692 			 * interpreted as an attempt to synchronize the CPU.
2693 			 */
2694 			synchronizing = data < tsc_exp + tsc_hz &&
2695 					data + tsc_hz > tsc_exp;
2696 		}
2697 	}
2698 
2699 	/*
2700 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
2701 	 * TSC, we add elapsed time in this computation.  We could let the
2702 	 * compensation code attempt to catch up if we fall behind, but
2703 	 * it's better to try to match offsets from the beginning.
2704          */
2705 	if (synchronizing &&
2706 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2707 		if (!kvm_check_tsc_unstable()) {
2708 			offset = kvm->arch.cur_tsc_offset;
2709 		} else {
2710 			u64 delta = nsec_to_cycles(vcpu, elapsed);
2711 			data += delta;
2712 			offset = kvm_compute_l1_tsc_offset(vcpu, data);
2713 		}
2714 		matched = true;
2715 	}
2716 
2717 	__kvm_synchronize_tsc(vcpu, offset, data, ns, matched);
2718 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2719 }
2720 
2721 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2722 					   s64 adjustment)
2723 {
2724 	u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2725 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2726 }
2727 
2728 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2729 {
2730 	if (vcpu->arch.l1_tsc_scaling_ratio != kvm_caps.default_tsc_scaling_ratio)
2731 		WARN_ON(adjustment < 0);
2732 	adjustment = kvm_scale_tsc((u64) adjustment,
2733 				   vcpu->arch.l1_tsc_scaling_ratio);
2734 	adjust_tsc_offset_guest(vcpu, adjustment);
2735 }
2736 
2737 #ifdef CONFIG_X86_64
2738 
2739 static u64 read_tsc(void)
2740 {
2741 	u64 ret = (u64)rdtsc_ordered();
2742 	u64 last = pvclock_gtod_data.clock.cycle_last;
2743 
2744 	if (likely(ret >= last))
2745 		return ret;
2746 
2747 	/*
2748 	 * GCC likes to generate cmov here, but this branch is extremely
2749 	 * predictable (it's just a function of time and the likely is
2750 	 * very likely) and there's a data dependence, so force GCC
2751 	 * to generate a branch instead.  I don't barrier() because
2752 	 * we don't actually need a barrier, and if this function
2753 	 * ever gets inlined it will generate worse code.
2754 	 */
2755 	asm volatile ("");
2756 	return last;
2757 }
2758 
2759 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2760 			  int *mode)
2761 {
2762 	long v;
2763 	u64 tsc_pg_val;
2764 
2765 	switch (clock->vclock_mode) {
2766 	case VDSO_CLOCKMODE_HVCLOCK:
2767 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2768 						  tsc_timestamp);
2769 		if (tsc_pg_val != U64_MAX) {
2770 			/* TSC page valid */
2771 			*mode = VDSO_CLOCKMODE_HVCLOCK;
2772 			v = (tsc_pg_val - clock->cycle_last) &
2773 				clock->mask;
2774 		} else {
2775 			/* TSC page invalid */
2776 			*mode = VDSO_CLOCKMODE_NONE;
2777 		}
2778 		break;
2779 	case VDSO_CLOCKMODE_TSC:
2780 		*mode = VDSO_CLOCKMODE_TSC;
2781 		*tsc_timestamp = read_tsc();
2782 		v = (*tsc_timestamp - clock->cycle_last) &
2783 			clock->mask;
2784 		break;
2785 	default:
2786 		*mode = VDSO_CLOCKMODE_NONE;
2787 	}
2788 
2789 	if (*mode == VDSO_CLOCKMODE_NONE)
2790 		*tsc_timestamp = v = 0;
2791 
2792 	return v * clock->mult;
2793 }
2794 
2795 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2796 {
2797 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2798 	unsigned long seq;
2799 	int mode;
2800 	u64 ns;
2801 
2802 	do {
2803 		seq = read_seqcount_begin(&gtod->seq);
2804 		ns = gtod->raw_clock.base_cycles;
2805 		ns += vgettsc(&gtod->raw_clock, tsc_timestamp, &mode);
2806 		ns >>= gtod->raw_clock.shift;
2807 		ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2808 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2809 	*t = ns;
2810 
2811 	return mode;
2812 }
2813 
2814 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2815 {
2816 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2817 	unsigned long seq;
2818 	int mode;
2819 	u64 ns;
2820 
2821 	do {
2822 		seq = read_seqcount_begin(&gtod->seq);
2823 		ts->tv_sec = gtod->wall_time_sec;
2824 		ns = gtod->clock.base_cycles;
2825 		ns += vgettsc(&gtod->clock, tsc_timestamp, &mode);
2826 		ns >>= gtod->clock.shift;
2827 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2828 
2829 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2830 	ts->tv_nsec = ns;
2831 
2832 	return mode;
2833 }
2834 
2835 /* returns true if host is using TSC based clocksource */
2836 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2837 {
2838 	/* checked again under seqlock below */
2839 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2840 		return false;
2841 
2842 	return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2843 						      tsc_timestamp));
2844 }
2845 
2846 /* returns true if host is using TSC based clocksource */
2847 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2848 					   u64 *tsc_timestamp)
2849 {
2850 	/* checked again under seqlock below */
2851 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2852 		return false;
2853 
2854 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2855 }
2856 #endif
2857 
2858 /*
2859  *
2860  * Assuming a stable TSC across physical CPUS, and a stable TSC
2861  * across virtual CPUs, the following condition is possible.
2862  * Each numbered line represents an event visible to both
2863  * CPUs at the next numbered event.
2864  *
2865  * "timespecX" represents host monotonic time. "tscX" represents
2866  * RDTSC value.
2867  *
2868  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
2869  *
2870  * 1.  read timespec0,tsc0
2871  * 2.					| timespec1 = timespec0 + N
2872  * 					| tsc1 = tsc0 + M
2873  * 3. transition to guest		| transition to guest
2874  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2875  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
2876  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2877  *
2878  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2879  *
2880  * 	- ret0 < ret1
2881  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2882  *		...
2883  *	- 0 < N - M => M < N
2884  *
2885  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2886  * always the case (the difference between two distinct xtime instances
2887  * might be smaller then the difference between corresponding TSC reads,
2888  * when updating guest vcpus pvclock areas).
2889  *
2890  * To avoid that problem, do not allow visibility of distinct
2891  * system_timestamp/tsc_timestamp values simultaneously: use a master
2892  * copy of host monotonic time values. Update that master copy
2893  * in lockstep.
2894  *
2895  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2896  *
2897  */
2898 
2899 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2900 {
2901 #ifdef CONFIG_X86_64
2902 	struct kvm_arch *ka = &kvm->arch;
2903 	int vclock_mode;
2904 	bool host_tsc_clocksource, vcpus_matched;
2905 
2906 	lockdep_assert_held(&kvm->arch.tsc_write_lock);
2907 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2908 			atomic_read(&kvm->online_vcpus));
2909 
2910 	/*
2911 	 * If the host uses TSC clock, then passthrough TSC as stable
2912 	 * to the guest.
2913 	 */
2914 	host_tsc_clocksource = kvm_get_time_and_clockread(
2915 					&ka->master_kernel_ns,
2916 					&ka->master_cycle_now);
2917 
2918 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2919 				&& !ka->backwards_tsc_observed
2920 				&& !ka->boot_vcpu_runs_old_kvmclock;
2921 
2922 	if (ka->use_master_clock)
2923 		atomic_set(&kvm_guest_has_master_clock, 1);
2924 
2925 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2926 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2927 					vcpus_matched);
2928 #endif
2929 }
2930 
2931 static void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2932 {
2933 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2934 }
2935 
2936 static void __kvm_start_pvclock_update(struct kvm *kvm)
2937 {
2938 	raw_spin_lock_irq(&kvm->arch.tsc_write_lock);
2939 	write_seqcount_begin(&kvm->arch.pvclock_sc);
2940 }
2941 
2942 static void kvm_start_pvclock_update(struct kvm *kvm)
2943 {
2944 	kvm_make_mclock_inprogress_request(kvm);
2945 
2946 	/* no guest entries from this point */
2947 	__kvm_start_pvclock_update(kvm);
2948 }
2949 
2950 static void kvm_end_pvclock_update(struct kvm *kvm)
2951 {
2952 	struct kvm_arch *ka = &kvm->arch;
2953 	struct kvm_vcpu *vcpu;
2954 	unsigned long i;
2955 
2956 	write_seqcount_end(&ka->pvclock_sc);
2957 	raw_spin_unlock_irq(&ka->tsc_write_lock);
2958 	kvm_for_each_vcpu(i, vcpu, kvm)
2959 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2960 
2961 	/* guest entries allowed */
2962 	kvm_for_each_vcpu(i, vcpu, kvm)
2963 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2964 }
2965 
2966 static void kvm_update_masterclock(struct kvm *kvm)
2967 {
2968 	kvm_hv_request_tsc_page_update(kvm);
2969 	kvm_start_pvclock_update(kvm);
2970 	pvclock_update_vm_gtod_copy(kvm);
2971 	kvm_end_pvclock_update(kvm);
2972 }
2973 
2974 /*
2975  * Use the kernel's tsc_khz directly if the TSC is constant, otherwise use KVM's
2976  * per-CPU value (which may be zero if a CPU is going offline).  Note, tsc_khz
2977  * can change during boot even if the TSC is constant, as it's possible for KVM
2978  * to be loaded before TSC calibration completes.  Ideally, KVM would get a
2979  * notification when calibration completes, but practically speaking calibration
2980  * will complete before userspace is alive enough to create VMs.
2981  */
2982 static unsigned long get_cpu_tsc_khz(void)
2983 {
2984 	if (static_cpu_has(X86_FEATURE_CONSTANT_TSC))
2985 		return tsc_khz;
2986 	else
2987 		return __this_cpu_read(cpu_tsc_khz);
2988 }
2989 
2990 /* Called within read_seqcount_begin/retry for kvm->pvclock_sc.  */
2991 static void __get_kvmclock(struct kvm *kvm, struct kvm_clock_data *data)
2992 {
2993 	struct kvm_arch *ka = &kvm->arch;
2994 	struct pvclock_vcpu_time_info hv_clock;
2995 
2996 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2997 	get_cpu();
2998 
2999 	data->flags = 0;
3000 	if (ka->use_master_clock &&
3001 	    (static_cpu_has(X86_FEATURE_CONSTANT_TSC) || __this_cpu_read(cpu_tsc_khz))) {
3002 #ifdef CONFIG_X86_64
3003 		struct timespec64 ts;
3004 
3005 		if (kvm_get_walltime_and_clockread(&ts, &data->host_tsc)) {
3006 			data->realtime = ts.tv_nsec + NSEC_PER_SEC * ts.tv_sec;
3007 			data->flags |= KVM_CLOCK_REALTIME | KVM_CLOCK_HOST_TSC;
3008 		} else
3009 #endif
3010 		data->host_tsc = rdtsc();
3011 
3012 		data->flags |= KVM_CLOCK_TSC_STABLE;
3013 		hv_clock.tsc_timestamp = ka->master_cycle_now;
3014 		hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
3015 		kvm_get_time_scale(NSEC_PER_SEC, get_cpu_tsc_khz() * 1000LL,
3016 				   &hv_clock.tsc_shift,
3017 				   &hv_clock.tsc_to_system_mul);
3018 		data->clock = __pvclock_read_cycles(&hv_clock, data->host_tsc);
3019 	} else {
3020 		data->clock = get_kvmclock_base_ns() + ka->kvmclock_offset;
3021 	}
3022 
3023 	put_cpu();
3024 }
3025 
3026 static void get_kvmclock(struct kvm *kvm, struct kvm_clock_data *data)
3027 {
3028 	struct kvm_arch *ka = &kvm->arch;
3029 	unsigned seq;
3030 
3031 	do {
3032 		seq = read_seqcount_begin(&ka->pvclock_sc);
3033 		__get_kvmclock(kvm, data);
3034 	} while (read_seqcount_retry(&ka->pvclock_sc, seq));
3035 }
3036 
3037 u64 get_kvmclock_ns(struct kvm *kvm)
3038 {
3039 	struct kvm_clock_data data;
3040 
3041 	get_kvmclock(kvm, &data);
3042 	return data.clock;
3043 }
3044 
3045 static void kvm_setup_guest_pvclock(struct kvm_vcpu *v,
3046 				    struct gfn_to_pfn_cache *gpc,
3047 				    unsigned int offset)
3048 {
3049 	struct kvm_vcpu_arch *vcpu = &v->arch;
3050 	struct pvclock_vcpu_time_info *guest_hv_clock;
3051 	unsigned long flags;
3052 
3053 	read_lock_irqsave(&gpc->lock, flags);
3054 	while (!kvm_gpc_check(gpc, offset + sizeof(*guest_hv_clock))) {
3055 		read_unlock_irqrestore(&gpc->lock, flags);
3056 
3057 		if (kvm_gpc_refresh(gpc, offset + sizeof(*guest_hv_clock)))
3058 			return;
3059 
3060 		read_lock_irqsave(&gpc->lock, flags);
3061 	}
3062 
3063 	guest_hv_clock = (void *)(gpc->khva + offset);
3064 
3065 	/*
3066 	 * This VCPU is paused, but it's legal for a guest to read another
3067 	 * VCPU's kvmclock, so we really have to follow the specification where
3068 	 * it says that version is odd if data is being modified, and even after
3069 	 * it is consistent.
3070 	 */
3071 
3072 	guest_hv_clock->version = vcpu->hv_clock.version = (guest_hv_clock->version + 1) | 1;
3073 	smp_wmb();
3074 
3075 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
3076 	vcpu->hv_clock.flags |= (guest_hv_clock->flags & PVCLOCK_GUEST_STOPPED);
3077 
3078 	if (vcpu->pvclock_set_guest_stopped_request) {
3079 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
3080 		vcpu->pvclock_set_guest_stopped_request = false;
3081 	}
3082 
3083 	memcpy(guest_hv_clock, &vcpu->hv_clock, sizeof(*guest_hv_clock));
3084 	smp_wmb();
3085 
3086 	guest_hv_clock->version = ++vcpu->hv_clock.version;
3087 
3088 	mark_page_dirty_in_slot(v->kvm, gpc->memslot, gpc->gpa >> PAGE_SHIFT);
3089 	read_unlock_irqrestore(&gpc->lock, flags);
3090 
3091 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
3092 }
3093 
3094 static int kvm_guest_time_update(struct kvm_vcpu *v)
3095 {
3096 	unsigned long flags, tgt_tsc_khz;
3097 	unsigned seq;
3098 	struct kvm_vcpu_arch *vcpu = &v->arch;
3099 	struct kvm_arch *ka = &v->kvm->arch;
3100 	s64 kernel_ns;
3101 	u64 tsc_timestamp, host_tsc;
3102 	u8 pvclock_flags;
3103 	bool use_master_clock;
3104 
3105 	kernel_ns = 0;
3106 	host_tsc = 0;
3107 
3108 	/*
3109 	 * If the host uses TSC clock, then passthrough TSC as stable
3110 	 * to the guest.
3111 	 */
3112 	do {
3113 		seq = read_seqcount_begin(&ka->pvclock_sc);
3114 		use_master_clock = ka->use_master_clock;
3115 		if (use_master_clock) {
3116 			host_tsc = ka->master_cycle_now;
3117 			kernel_ns = ka->master_kernel_ns;
3118 		}
3119 	} while (read_seqcount_retry(&ka->pvclock_sc, seq));
3120 
3121 	/* Keep irq disabled to prevent changes to the clock */
3122 	local_irq_save(flags);
3123 	tgt_tsc_khz = get_cpu_tsc_khz();
3124 	if (unlikely(tgt_tsc_khz == 0)) {
3125 		local_irq_restore(flags);
3126 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3127 		return 1;
3128 	}
3129 	if (!use_master_clock) {
3130 		host_tsc = rdtsc();
3131 		kernel_ns = get_kvmclock_base_ns();
3132 	}
3133 
3134 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
3135 
3136 	/*
3137 	 * We may have to catch up the TSC to match elapsed wall clock
3138 	 * time for two reasons, even if kvmclock is used.
3139 	 *   1) CPU could have been running below the maximum TSC rate
3140 	 *   2) Broken TSC compensation resets the base at each VCPU
3141 	 *      entry to avoid unknown leaps of TSC even when running
3142 	 *      again on the same CPU.  This may cause apparent elapsed
3143 	 *      time to disappear, and the guest to stand still or run
3144 	 *	very slowly.
3145 	 */
3146 	if (vcpu->tsc_catchup) {
3147 		u64 tsc = compute_guest_tsc(v, kernel_ns);
3148 		if (tsc > tsc_timestamp) {
3149 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
3150 			tsc_timestamp = tsc;
3151 		}
3152 	}
3153 
3154 	local_irq_restore(flags);
3155 
3156 	/* With all the info we got, fill in the values */
3157 
3158 	if (kvm_caps.has_tsc_control)
3159 		tgt_tsc_khz = kvm_scale_tsc(tgt_tsc_khz,
3160 					    v->arch.l1_tsc_scaling_ratio);
3161 
3162 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
3163 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
3164 				   &vcpu->hv_clock.tsc_shift,
3165 				   &vcpu->hv_clock.tsc_to_system_mul);
3166 		vcpu->hw_tsc_khz = tgt_tsc_khz;
3167 	}
3168 
3169 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
3170 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
3171 	vcpu->last_guest_tsc = tsc_timestamp;
3172 
3173 	/* If the host uses TSC clocksource, then it is stable */
3174 	pvclock_flags = 0;
3175 	if (use_master_clock)
3176 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
3177 
3178 	vcpu->hv_clock.flags = pvclock_flags;
3179 
3180 	if (vcpu->pv_time.active)
3181 		kvm_setup_guest_pvclock(v, &vcpu->pv_time, 0);
3182 	if (vcpu->xen.vcpu_info_cache.active)
3183 		kvm_setup_guest_pvclock(v, &vcpu->xen.vcpu_info_cache,
3184 					offsetof(struct compat_vcpu_info, time));
3185 	if (vcpu->xen.vcpu_time_info_cache.active)
3186 		kvm_setup_guest_pvclock(v, &vcpu->xen.vcpu_time_info_cache, 0);
3187 	kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
3188 	return 0;
3189 }
3190 
3191 /*
3192  * kvmclock updates which are isolated to a given vcpu, such as
3193  * vcpu->cpu migration, should not allow system_timestamp from
3194  * the rest of the vcpus to remain static. Otherwise ntp frequency
3195  * correction applies to one vcpu's system_timestamp but not
3196  * the others.
3197  *
3198  * So in those cases, request a kvmclock update for all vcpus.
3199  * We need to rate-limit these requests though, as they can
3200  * considerably slow guests that have a large number of vcpus.
3201  * The time for a remote vcpu to update its kvmclock is bound
3202  * by the delay we use to rate-limit the updates.
3203  */
3204 
3205 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
3206 
3207 static void kvmclock_update_fn(struct work_struct *work)
3208 {
3209 	unsigned long i;
3210 	struct delayed_work *dwork = to_delayed_work(work);
3211 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3212 					   kvmclock_update_work);
3213 	struct kvm *kvm = container_of(ka, struct kvm, arch);
3214 	struct kvm_vcpu *vcpu;
3215 
3216 	kvm_for_each_vcpu(i, vcpu, kvm) {
3217 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3218 		kvm_vcpu_kick(vcpu);
3219 	}
3220 }
3221 
3222 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
3223 {
3224 	struct kvm *kvm = v->kvm;
3225 
3226 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3227 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
3228 					KVMCLOCK_UPDATE_DELAY);
3229 }
3230 
3231 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
3232 
3233 static void kvmclock_sync_fn(struct work_struct *work)
3234 {
3235 	struct delayed_work *dwork = to_delayed_work(work);
3236 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3237 					   kvmclock_sync_work);
3238 	struct kvm *kvm = container_of(ka, struct kvm, arch);
3239 
3240 	if (!kvmclock_periodic_sync)
3241 		return;
3242 
3243 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
3244 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
3245 					KVMCLOCK_SYNC_PERIOD);
3246 }
3247 
3248 /* These helpers are safe iff @msr is known to be an MCx bank MSR. */
3249 static bool is_mci_control_msr(u32 msr)
3250 {
3251 	return (msr & 3) == 0;
3252 }
3253 static bool is_mci_status_msr(u32 msr)
3254 {
3255 	return (msr & 3) == 1;
3256 }
3257 
3258 /*
3259  * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
3260  */
3261 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
3262 {
3263 	/* McStatusWrEn enabled? */
3264 	if (guest_cpuid_is_amd_or_hygon(vcpu))
3265 		return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
3266 
3267 	return false;
3268 }
3269 
3270 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3271 {
3272 	u64 mcg_cap = vcpu->arch.mcg_cap;
3273 	unsigned bank_num = mcg_cap & 0xff;
3274 	u32 msr = msr_info->index;
3275 	u64 data = msr_info->data;
3276 	u32 offset, last_msr;
3277 
3278 	switch (msr) {
3279 	case MSR_IA32_MCG_STATUS:
3280 		vcpu->arch.mcg_status = data;
3281 		break;
3282 	case MSR_IA32_MCG_CTL:
3283 		if (!(mcg_cap & MCG_CTL_P) &&
3284 		    (data || !msr_info->host_initiated))
3285 			return 1;
3286 		if (data != 0 && data != ~(u64)0)
3287 			return 1;
3288 		vcpu->arch.mcg_ctl = data;
3289 		break;
3290 	case MSR_IA32_MC0_CTL2 ... MSR_IA32_MCx_CTL2(KVM_MAX_MCE_BANKS) - 1:
3291 		last_msr = MSR_IA32_MCx_CTL2(bank_num) - 1;
3292 		if (msr > last_msr)
3293 			return 1;
3294 
3295 		if (!(mcg_cap & MCG_CMCI_P) && (data || !msr_info->host_initiated))
3296 			return 1;
3297 		/* An attempt to write a 1 to a reserved bit raises #GP */
3298 		if (data & ~(MCI_CTL2_CMCI_EN | MCI_CTL2_CMCI_THRESHOLD_MASK))
3299 			return 1;
3300 		offset = array_index_nospec(msr - MSR_IA32_MC0_CTL2,
3301 					    last_msr + 1 - MSR_IA32_MC0_CTL2);
3302 		vcpu->arch.mci_ctl2_banks[offset] = data;
3303 		break;
3304 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3305 		last_msr = MSR_IA32_MCx_CTL(bank_num) - 1;
3306 		if (msr > last_msr)
3307 			return 1;
3308 
3309 		/*
3310 		 * Only 0 or all 1s can be written to IA32_MCi_CTL, all other
3311 		 * values are architecturally undefined.  But, some Linux
3312 		 * kernels clear bit 10 in bank 4 to workaround a BIOS/GART TLB
3313 		 * issue on AMD K8s, allow bit 10 to be clear when setting all
3314 		 * other bits in order to avoid an uncaught #GP in the guest.
3315 		 *
3316 		 * UNIXWARE clears bit 0 of MC1_CTL to ignore correctable,
3317 		 * single-bit ECC data errors.
3318 		 */
3319 		if (is_mci_control_msr(msr) &&
3320 		    data != 0 && (data | (1 << 10) | 1) != ~(u64)0)
3321 			return 1;
3322 
3323 		/*
3324 		 * All CPUs allow writing 0 to MCi_STATUS MSRs to clear the MSR.
3325 		 * AMD-based CPUs allow non-zero values, but if and only if
3326 		 * HWCR[McStatusWrEn] is set.
3327 		 */
3328 		if (!msr_info->host_initiated && is_mci_status_msr(msr) &&
3329 		    data != 0 && !can_set_mci_status(vcpu))
3330 			return 1;
3331 
3332 		offset = array_index_nospec(msr - MSR_IA32_MC0_CTL,
3333 					    last_msr + 1 - MSR_IA32_MC0_CTL);
3334 		vcpu->arch.mce_banks[offset] = data;
3335 		break;
3336 	default:
3337 		return 1;
3338 	}
3339 	return 0;
3340 }
3341 
3342 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
3343 {
3344 	u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3345 
3346 	return (vcpu->arch.apf.msr_en_val & mask) == mask;
3347 }
3348 
3349 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3350 {
3351 	gpa_t gpa = data & ~0x3f;
3352 
3353 	/* Bits 4:5 are reserved, Should be zero */
3354 	if (data & 0x30)
3355 		return 1;
3356 
3357 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3358 	    (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3359 		return 1;
3360 
3361 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3362 	    (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3363 		return 1;
3364 
3365 	if (!lapic_in_kernel(vcpu))
3366 		return data ? 1 : 0;
3367 
3368 	vcpu->arch.apf.msr_en_val = data;
3369 
3370 	if (!kvm_pv_async_pf_enabled(vcpu)) {
3371 		kvm_clear_async_pf_completion_queue(vcpu);
3372 		kvm_async_pf_hash_reset(vcpu);
3373 		return 0;
3374 	}
3375 
3376 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3377 					sizeof(u64)))
3378 		return 1;
3379 
3380 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3381 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3382 
3383 	kvm_async_pf_wakeup_all(vcpu);
3384 
3385 	return 0;
3386 }
3387 
3388 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3389 {
3390 	/* Bits 8-63 are reserved */
3391 	if (data >> 8)
3392 		return 1;
3393 
3394 	if (!lapic_in_kernel(vcpu))
3395 		return 1;
3396 
3397 	vcpu->arch.apf.msr_int_val = data;
3398 
3399 	vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3400 
3401 	return 0;
3402 }
3403 
3404 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3405 {
3406 	kvm_gpc_deactivate(&vcpu->arch.pv_time);
3407 	vcpu->arch.time = 0;
3408 }
3409 
3410 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3411 {
3412 	++vcpu->stat.tlb_flush;
3413 	static_call(kvm_x86_flush_tlb_all)(vcpu);
3414 
3415 	/* Flushing all ASIDs flushes the current ASID... */
3416 	kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
3417 }
3418 
3419 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3420 {
3421 	++vcpu->stat.tlb_flush;
3422 
3423 	if (!tdp_enabled) {
3424 		/*
3425 		 * A TLB flush on behalf of the guest is equivalent to
3426 		 * INVPCID(all), toggling CR4.PGE, etc., which requires
3427 		 * a forced sync of the shadow page tables.  Ensure all the
3428 		 * roots are synced and the guest TLB in hardware is clean.
3429 		 */
3430 		kvm_mmu_sync_roots(vcpu);
3431 		kvm_mmu_sync_prev_roots(vcpu);
3432 	}
3433 
3434 	static_call(kvm_x86_flush_tlb_guest)(vcpu);
3435 
3436 	/*
3437 	 * Flushing all "guest" TLB is always a superset of Hyper-V's fine
3438 	 * grained flushing.
3439 	 */
3440 	kvm_hv_vcpu_purge_flush_tlb(vcpu);
3441 }
3442 
3443 
3444 static inline void kvm_vcpu_flush_tlb_current(struct kvm_vcpu *vcpu)
3445 {
3446 	++vcpu->stat.tlb_flush;
3447 	static_call(kvm_x86_flush_tlb_current)(vcpu);
3448 }
3449 
3450 /*
3451  * Service "local" TLB flush requests, which are specific to the current MMU
3452  * context.  In addition to the generic event handling in vcpu_enter_guest(),
3453  * TLB flushes that are targeted at an MMU context also need to be serviced
3454  * prior before nested VM-Enter/VM-Exit.
3455  */
3456 void kvm_service_local_tlb_flush_requests(struct kvm_vcpu *vcpu)
3457 {
3458 	if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
3459 		kvm_vcpu_flush_tlb_current(vcpu);
3460 
3461 	if (kvm_check_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu))
3462 		kvm_vcpu_flush_tlb_guest(vcpu);
3463 }
3464 EXPORT_SYMBOL_GPL(kvm_service_local_tlb_flush_requests);
3465 
3466 static void record_steal_time(struct kvm_vcpu *vcpu)
3467 {
3468 	struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
3469 	struct kvm_steal_time __user *st;
3470 	struct kvm_memslots *slots;
3471 	gpa_t gpa = vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS;
3472 	u64 steal;
3473 	u32 version;
3474 
3475 	if (kvm_xen_msr_enabled(vcpu->kvm)) {
3476 		kvm_xen_runstate_set_running(vcpu);
3477 		return;
3478 	}
3479 
3480 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3481 		return;
3482 
3483 	if (WARN_ON_ONCE(current->mm != vcpu->kvm->mm))
3484 		return;
3485 
3486 	slots = kvm_memslots(vcpu->kvm);
3487 
3488 	if (unlikely(slots->generation != ghc->generation ||
3489 		     gpa != ghc->gpa ||
3490 		     kvm_is_error_hva(ghc->hva) || !ghc->memslot)) {
3491 		/* We rely on the fact that it fits in a single page. */
3492 		BUILD_BUG_ON((sizeof(*st) - 1) & KVM_STEAL_VALID_BITS);
3493 
3494 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, gpa, sizeof(*st)) ||
3495 		    kvm_is_error_hva(ghc->hva) || !ghc->memslot)
3496 			return;
3497 	}
3498 
3499 	st = (struct kvm_steal_time __user *)ghc->hva;
3500 	/*
3501 	 * Doing a TLB flush here, on the guest's behalf, can avoid
3502 	 * expensive IPIs.
3503 	 */
3504 	if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3505 		u8 st_preempted = 0;
3506 		int err = -EFAULT;
3507 
3508 		if (!user_access_begin(st, sizeof(*st)))
3509 			return;
3510 
3511 		asm volatile("1: xchgb %0, %2\n"
3512 			     "xor %1, %1\n"
3513 			     "2:\n"
3514 			     _ASM_EXTABLE_UA(1b, 2b)
3515 			     : "+q" (st_preempted),
3516 			       "+&r" (err),
3517 			       "+m" (st->preempted));
3518 		if (err)
3519 			goto out;
3520 
3521 		user_access_end();
3522 
3523 		vcpu->arch.st.preempted = 0;
3524 
3525 		trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3526 				       st_preempted & KVM_VCPU_FLUSH_TLB);
3527 		if (st_preempted & KVM_VCPU_FLUSH_TLB)
3528 			kvm_vcpu_flush_tlb_guest(vcpu);
3529 
3530 		if (!user_access_begin(st, sizeof(*st)))
3531 			goto dirty;
3532 	} else {
3533 		if (!user_access_begin(st, sizeof(*st)))
3534 			return;
3535 
3536 		unsafe_put_user(0, &st->preempted, out);
3537 		vcpu->arch.st.preempted = 0;
3538 	}
3539 
3540 	unsafe_get_user(version, &st->version, out);
3541 	if (version & 1)
3542 		version += 1;  /* first time write, random junk */
3543 
3544 	version += 1;
3545 	unsafe_put_user(version, &st->version, out);
3546 
3547 	smp_wmb();
3548 
3549 	unsafe_get_user(steal, &st->steal, out);
3550 	steal += current->sched_info.run_delay -
3551 		vcpu->arch.st.last_steal;
3552 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
3553 	unsafe_put_user(steal, &st->steal, out);
3554 
3555 	version += 1;
3556 	unsafe_put_user(version, &st->version, out);
3557 
3558  out:
3559 	user_access_end();
3560  dirty:
3561 	mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
3562 }
3563 
3564 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3565 {
3566 	u32 msr = msr_info->index;
3567 	u64 data = msr_info->data;
3568 
3569 	if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3570 		return kvm_xen_write_hypercall_page(vcpu, data);
3571 
3572 	switch (msr) {
3573 	case MSR_AMD64_NB_CFG:
3574 	case MSR_IA32_UCODE_WRITE:
3575 	case MSR_VM_HSAVE_PA:
3576 	case MSR_AMD64_PATCH_LOADER:
3577 	case MSR_AMD64_BU_CFG2:
3578 	case MSR_AMD64_DC_CFG:
3579 	case MSR_F15H_EX_CFG:
3580 		break;
3581 
3582 	case MSR_IA32_UCODE_REV:
3583 		if (msr_info->host_initiated)
3584 			vcpu->arch.microcode_version = data;
3585 		break;
3586 	case MSR_IA32_ARCH_CAPABILITIES:
3587 		if (!msr_info->host_initiated)
3588 			return 1;
3589 		vcpu->arch.arch_capabilities = data;
3590 		break;
3591 	case MSR_IA32_PERF_CAPABILITIES:
3592 		if (!msr_info->host_initiated)
3593 			return 1;
3594 		if (data & ~kvm_caps.supported_perf_cap)
3595 			return 1;
3596 
3597 		vcpu->arch.perf_capabilities = data;
3598 		kvm_pmu_refresh(vcpu);
3599 		return 0;
3600 	case MSR_EFER:
3601 		return set_efer(vcpu, msr_info);
3602 	case MSR_K7_HWCR:
3603 		data &= ~(u64)0x40;	/* ignore flush filter disable */
3604 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
3605 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
3606 
3607 		/* Handle McStatusWrEn */
3608 		if (data == BIT_ULL(18)) {
3609 			vcpu->arch.msr_hwcr = data;
3610 		} else if (data != 0) {
3611 			kvm_pr_unimpl_wrmsr(vcpu, msr, data);
3612 			return 1;
3613 		}
3614 		break;
3615 	case MSR_FAM10H_MMIO_CONF_BASE:
3616 		if (data != 0) {
3617 			kvm_pr_unimpl_wrmsr(vcpu, msr, data);
3618 			return 1;
3619 		}
3620 		break;
3621 	case 0x200 ... MSR_IA32_MC0_CTL2 - 1:
3622 	case MSR_IA32_MCx_CTL2(KVM_MAX_MCE_BANKS) ... 0x2ff:
3623 		return kvm_mtrr_set_msr(vcpu, msr, data);
3624 	case MSR_IA32_APICBASE:
3625 		return kvm_set_apic_base(vcpu, msr_info);
3626 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3627 		return kvm_x2apic_msr_write(vcpu, msr, data);
3628 	case MSR_IA32_TSC_DEADLINE:
3629 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
3630 		break;
3631 	case MSR_IA32_TSC_ADJUST:
3632 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3633 			if (!msr_info->host_initiated) {
3634 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3635 				adjust_tsc_offset_guest(vcpu, adj);
3636 				/* Before back to guest, tsc_timestamp must be adjusted
3637 				 * as well, otherwise guest's percpu pvclock time could jump.
3638 				 */
3639 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3640 			}
3641 			vcpu->arch.ia32_tsc_adjust_msr = data;
3642 		}
3643 		break;
3644 	case MSR_IA32_MISC_ENABLE: {
3645 		u64 old_val = vcpu->arch.ia32_misc_enable_msr;
3646 
3647 		if (!msr_info->host_initiated) {
3648 			/* RO bits */
3649 			if ((old_val ^ data) & MSR_IA32_MISC_ENABLE_PMU_RO_MASK)
3650 				return 1;
3651 
3652 			/* R bits, i.e. writes are ignored, but don't fault. */
3653 			data = data & ~MSR_IA32_MISC_ENABLE_EMON;
3654 			data |= old_val & MSR_IA32_MISC_ENABLE_EMON;
3655 		}
3656 
3657 		if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3658 		    ((old_val ^ data)  & MSR_IA32_MISC_ENABLE_MWAIT)) {
3659 			if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3660 				return 1;
3661 			vcpu->arch.ia32_misc_enable_msr = data;
3662 			kvm_update_cpuid_runtime(vcpu);
3663 		} else {
3664 			vcpu->arch.ia32_misc_enable_msr = data;
3665 		}
3666 		break;
3667 	}
3668 	case MSR_IA32_SMBASE:
3669 		if (!IS_ENABLED(CONFIG_KVM_SMM) || !msr_info->host_initiated)
3670 			return 1;
3671 		vcpu->arch.smbase = data;
3672 		break;
3673 	case MSR_IA32_POWER_CTL:
3674 		vcpu->arch.msr_ia32_power_ctl = data;
3675 		break;
3676 	case MSR_IA32_TSC:
3677 		if (msr_info->host_initiated) {
3678 			kvm_synchronize_tsc(vcpu, data);
3679 		} else {
3680 			u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3681 			adjust_tsc_offset_guest(vcpu, adj);
3682 			vcpu->arch.ia32_tsc_adjust_msr += adj;
3683 		}
3684 		break;
3685 	case MSR_IA32_XSS:
3686 		if (!msr_info->host_initiated &&
3687 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3688 			return 1;
3689 		/*
3690 		 * KVM supports exposing PT to the guest, but does not support
3691 		 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3692 		 * XSAVES/XRSTORS to save/restore PT MSRs.
3693 		 */
3694 		if (data & ~kvm_caps.supported_xss)
3695 			return 1;
3696 		vcpu->arch.ia32_xss = data;
3697 		kvm_update_cpuid_runtime(vcpu);
3698 		break;
3699 	case MSR_SMI_COUNT:
3700 		if (!msr_info->host_initiated)
3701 			return 1;
3702 		vcpu->arch.smi_count = data;
3703 		break;
3704 	case MSR_KVM_WALL_CLOCK_NEW:
3705 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3706 			return 1;
3707 
3708 		vcpu->kvm->arch.wall_clock = data;
3709 		kvm_write_wall_clock(vcpu->kvm, data, 0);
3710 		break;
3711 	case MSR_KVM_WALL_CLOCK:
3712 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3713 			return 1;
3714 
3715 		vcpu->kvm->arch.wall_clock = data;
3716 		kvm_write_wall_clock(vcpu->kvm, data, 0);
3717 		break;
3718 	case MSR_KVM_SYSTEM_TIME_NEW:
3719 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3720 			return 1;
3721 
3722 		kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3723 		break;
3724 	case MSR_KVM_SYSTEM_TIME:
3725 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3726 			return 1;
3727 
3728 		kvm_write_system_time(vcpu, data, true,  msr_info->host_initiated);
3729 		break;
3730 	case MSR_KVM_ASYNC_PF_EN:
3731 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3732 			return 1;
3733 
3734 		if (kvm_pv_enable_async_pf(vcpu, data))
3735 			return 1;
3736 		break;
3737 	case MSR_KVM_ASYNC_PF_INT:
3738 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3739 			return 1;
3740 
3741 		if (kvm_pv_enable_async_pf_int(vcpu, data))
3742 			return 1;
3743 		break;
3744 	case MSR_KVM_ASYNC_PF_ACK:
3745 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3746 			return 1;
3747 		if (data & 0x1) {
3748 			vcpu->arch.apf.pageready_pending = false;
3749 			kvm_check_async_pf_completion(vcpu);
3750 		}
3751 		break;
3752 	case MSR_KVM_STEAL_TIME:
3753 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3754 			return 1;
3755 
3756 		if (unlikely(!sched_info_on()))
3757 			return 1;
3758 
3759 		if (data & KVM_STEAL_RESERVED_MASK)
3760 			return 1;
3761 
3762 		vcpu->arch.st.msr_val = data;
3763 
3764 		if (!(data & KVM_MSR_ENABLED))
3765 			break;
3766 
3767 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3768 
3769 		break;
3770 	case MSR_KVM_PV_EOI_EN:
3771 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3772 			return 1;
3773 
3774 		if (kvm_lapic_set_pv_eoi(vcpu, data, sizeof(u8)))
3775 			return 1;
3776 		break;
3777 
3778 	case MSR_KVM_POLL_CONTROL:
3779 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3780 			return 1;
3781 
3782 		/* only enable bit supported */
3783 		if (data & (-1ULL << 1))
3784 			return 1;
3785 
3786 		vcpu->arch.msr_kvm_poll_control = data;
3787 		break;
3788 
3789 	case MSR_IA32_MCG_CTL:
3790 	case MSR_IA32_MCG_STATUS:
3791 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3792 	case MSR_IA32_MC0_CTL2 ... MSR_IA32_MCx_CTL2(KVM_MAX_MCE_BANKS) - 1:
3793 		return set_msr_mce(vcpu, msr_info);
3794 
3795 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3796 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3797 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3798 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3799 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3800 			return kvm_pmu_set_msr(vcpu, msr_info);
3801 
3802 		if (data)
3803 			kvm_pr_unimpl_wrmsr(vcpu, msr, data);
3804 		break;
3805 	case MSR_K7_CLK_CTL:
3806 		/*
3807 		 * Ignore all writes to this no longer documented MSR.
3808 		 * Writes are only relevant for old K7 processors,
3809 		 * all pre-dating SVM, but a recommended workaround from
3810 		 * AMD for these chips. It is possible to specify the
3811 		 * affected processor models on the command line, hence
3812 		 * the need to ignore the workaround.
3813 		 */
3814 		break;
3815 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3816 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3817 	case HV_X64_MSR_SYNDBG_OPTIONS:
3818 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3819 	case HV_X64_MSR_CRASH_CTL:
3820 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3821 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3822 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3823 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3824 	case HV_X64_MSR_TSC_INVARIANT_CONTROL:
3825 		return kvm_hv_set_msr_common(vcpu, msr, data,
3826 					     msr_info->host_initiated);
3827 	case MSR_IA32_BBL_CR_CTL3:
3828 		/* Drop writes to this legacy MSR -- see rdmsr
3829 		 * counterpart for further detail.
3830 		 */
3831 		kvm_pr_unimpl_wrmsr(vcpu, msr, data);
3832 		break;
3833 	case MSR_AMD64_OSVW_ID_LENGTH:
3834 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3835 			return 1;
3836 		vcpu->arch.osvw.length = data;
3837 		break;
3838 	case MSR_AMD64_OSVW_STATUS:
3839 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3840 			return 1;
3841 		vcpu->arch.osvw.status = data;
3842 		break;
3843 	case MSR_PLATFORM_INFO:
3844 		if (!msr_info->host_initiated ||
3845 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3846 		     cpuid_fault_enabled(vcpu)))
3847 			return 1;
3848 		vcpu->arch.msr_platform_info = data;
3849 		break;
3850 	case MSR_MISC_FEATURES_ENABLES:
3851 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3852 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3853 		     !supports_cpuid_fault(vcpu)))
3854 			return 1;
3855 		vcpu->arch.msr_misc_features_enables = data;
3856 		break;
3857 #ifdef CONFIG_X86_64
3858 	case MSR_IA32_XFD:
3859 		if (!msr_info->host_initiated &&
3860 		    !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
3861 			return 1;
3862 
3863 		if (data & ~kvm_guest_supported_xfd(vcpu))
3864 			return 1;
3865 
3866 		fpu_update_guest_xfd(&vcpu->arch.guest_fpu, data);
3867 		break;
3868 	case MSR_IA32_XFD_ERR:
3869 		if (!msr_info->host_initiated &&
3870 		    !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
3871 			return 1;
3872 
3873 		if (data & ~kvm_guest_supported_xfd(vcpu))
3874 			return 1;
3875 
3876 		vcpu->arch.guest_fpu.xfd_err = data;
3877 		break;
3878 #endif
3879 	case MSR_IA32_PEBS_ENABLE:
3880 	case MSR_IA32_DS_AREA:
3881 	case MSR_PEBS_DATA_CFG:
3882 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3883 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3884 			return kvm_pmu_set_msr(vcpu, msr_info);
3885 		/*
3886 		 * Userspace is allowed to write '0' to MSRs that KVM reports
3887 		 * as to-be-saved, even if an MSRs isn't fully supported.
3888 		 */
3889 		return !msr_info->host_initiated || data;
3890 	default:
3891 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3892 			return kvm_pmu_set_msr(vcpu, msr_info);
3893 		return KVM_MSR_RET_INVALID;
3894 	}
3895 	return 0;
3896 }
3897 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3898 
3899 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3900 {
3901 	u64 data;
3902 	u64 mcg_cap = vcpu->arch.mcg_cap;
3903 	unsigned bank_num = mcg_cap & 0xff;
3904 	u32 offset, last_msr;
3905 
3906 	switch (msr) {
3907 	case MSR_IA32_P5_MC_ADDR:
3908 	case MSR_IA32_P5_MC_TYPE:
3909 		data = 0;
3910 		break;
3911 	case MSR_IA32_MCG_CAP:
3912 		data = vcpu->arch.mcg_cap;
3913 		break;
3914 	case MSR_IA32_MCG_CTL:
3915 		if (!(mcg_cap & MCG_CTL_P) && !host)
3916 			return 1;
3917 		data = vcpu->arch.mcg_ctl;
3918 		break;
3919 	case MSR_IA32_MCG_STATUS:
3920 		data = vcpu->arch.mcg_status;
3921 		break;
3922 	case MSR_IA32_MC0_CTL2 ... MSR_IA32_MCx_CTL2(KVM_MAX_MCE_BANKS) - 1:
3923 		last_msr = MSR_IA32_MCx_CTL2(bank_num) - 1;
3924 		if (msr > last_msr)
3925 			return 1;
3926 
3927 		if (!(mcg_cap & MCG_CMCI_P) && !host)
3928 			return 1;
3929 		offset = array_index_nospec(msr - MSR_IA32_MC0_CTL2,
3930 					    last_msr + 1 - MSR_IA32_MC0_CTL2);
3931 		data = vcpu->arch.mci_ctl2_banks[offset];
3932 		break;
3933 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3934 		last_msr = MSR_IA32_MCx_CTL(bank_num) - 1;
3935 		if (msr > last_msr)
3936 			return 1;
3937 
3938 		offset = array_index_nospec(msr - MSR_IA32_MC0_CTL,
3939 					    last_msr + 1 - MSR_IA32_MC0_CTL);
3940 		data = vcpu->arch.mce_banks[offset];
3941 		break;
3942 	default:
3943 		return 1;
3944 	}
3945 	*pdata = data;
3946 	return 0;
3947 }
3948 
3949 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3950 {
3951 	switch (msr_info->index) {
3952 	case MSR_IA32_PLATFORM_ID:
3953 	case MSR_IA32_EBL_CR_POWERON:
3954 	case MSR_IA32_LASTBRANCHFROMIP:
3955 	case MSR_IA32_LASTBRANCHTOIP:
3956 	case MSR_IA32_LASTINTFROMIP:
3957 	case MSR_IA32_LASTINTTOIP:
3958 	case MSR_AMD64_SYSCFG:
3959 	case MSR_K8_TSEG_ADDR:
3960 	case MSR_K8_TSEG_MASK:
3961 	case MSR_VM_HSAVE_PA:
3962 	case MSR_K8_INT_PENDING_MSG:
3963 	case MSR_AMD64_NB_CFG:
3964 	case MSR_FAM10H_MMIO_CONF_BASE:
3965 	case MSR_AMD64_BU_CFG2:
3966 	case MSR_IA32_PERF_CTL:
3967 	case MSR_AMD64_DC_CFG:
3968 	case MSR_F15H_EX_CFG:
3969 	/*
3970 	 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3971 	 * limit) MSRs. Just return 0, as we do not want to expose the host
3972 	 * data here. Do not conditionalize this on CPUID, as KVM does not do
3973 	 * so for existing CPU-specific MSRs.
3974 	 */
3975 	case MSR_RAPL_POWER_UNIT:
3976 	case MSR_PP0_ENERGY_STATUS:	/* Power plane 0 (core) */
3977 	case MSR_PP1_ENERGY_STATUS:	/* Power plane 1 (graphics uncore) */
3978 	case MSR_PKG_ENERGY_STATUS:	/* Total package */
3979 	case MSR_DRAM_ENERGY_STATUS:	/* DRAM controller */
3980 		msr_info->data = 0;
3981 		break;
3982 	case MSR_IA32_PEBS_ENABLE:
3983 	case MSR_IA32_DS_AREA:
3984 	case MSR_PEBS_DATA_CFG:
3985 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3986 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3987 			return kvm_pmu_get_msr(vcpu, msr_info);
3988 		/*
3989 		 * Userspace is allowed to read MSRs that KVM reports as
3990 		 * to-be-saved, even if an MSR isn't fully supported.
3991 		 */
3992 		if (!msr_info->host_initiated)
3993 			return 1;
3994 		msr_info->data = 0;
3995 		break;
3996 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3997 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3998 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3999 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
4000 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
4001 			return kvm_pmu_get_msr(vcpu, msr_info);
4002 		msr_info->data = 0;
4003 		break;
4004 	case MSR_IA32_UCODE_REV:
4005 		msr_info->data = vcpu->arch.microcode_version;
4006 		break;
4007 	case MSR_IA32_ARCH_CAPABILITIES:
4008 		if (!msr_info->host_initiated &&
4009 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
4010 			return 1;
4011 		msr_info->data = vcpu->arch.arch_capabilities;
4012 		break;
4013 	case MSR_IA32_PERF_CAPABILITIES:
4014 		if (!msr_info->host_initiated &&
4015 		    !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
4016 			return 1;
4017 		msr_info->data = vcpu->arch.perf_capabilities;
4018 		break;
4019 	case MSR_IA32_POWER_CTL:
4020 		msr_info->data = vcpu->arch.msr_ia32_power_ctl;
4021 		break;
4022 	case MSR_IA32_TSC: {
4023 		/*
4024 		 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
4025 		 * even when not intercepted. AMD manual doesn't explicitly
4026 		 * state this but appears to behave the same.
4027 		 *
4028 		 * On userspace reads and writes, however, we unconditionally
4029 		 * return L1's TSC value to ensure backwards-compatible
4030 		 * behavior for migration.
4031 		 */
4032 		u64 offset, ratio;
4033 
4034 		if (msr_info->host_initiated) {
4035 			offset = vcpu->arch.l1_tsc_offset;
4036 			ratio = vcpu->arch.l1_tsc_scaling_ratio;
4037 		} else {
4038 			offset = vcpu->arch.tsc_offset;
4039 			ratio = vcpu->arch.tsc_scaling_ratio;
4040 		}
4041 
4042 		msr_info->data = kvm_scale_tsc(rdtsc(), ratio) + offset;
4043 		break;
4044 	}
4045 	case MSR_MTRRcap:
4046 	case 0x200 ... MSR_IA32_MC0_CTL2 - 1:
4047 	case MSR_IA32_MCx_CTL2(KVM_MAX_MCE_BANKS) ... 0x2ff:
4048 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
4049 	case 0xcd: /* fsb frequency */
4050 		msr_info->data = 3;
4051 		break;
4052 		/*
4053 		 * MSR_EBC_FREQUENCY_ID
4054 		 * Conservative value valid for even the basic CPU models.
4055 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
4056 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
4057 		 * and 266MHz for model 3, or 4. Set Core Clock
4058 		 * Frequency to System Bus Frequency Ratio to 1 (bits
4059 		 * 31:24) even though these are only valid for CPU
4060 		 * models > 2, however guests may end up dividing or
4061 		 * multiplying by zero otherwise.
4062 		 */
4063 	case MSR_EBC_FREQUENCY_ID:
4064 		msr_info->data = 1 << 24;
4065 		break;
4066 	case MSR_IA32_APICBASE:
4067 		msr_info->data = kvm_get_apic_base(vcpu);
4068 		break;
4069 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
4070 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
4071 	case MSR_IA32_TSC_DEADLINE:
4072 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
4073 		break;
4074 	case MSR_IA32_TSC_ADJUST:
4075 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
4076 		break;
4077 	case MSR_IA32_MISC_ENABLE:
4078 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
4079 		break;
4080 	case MSR_IA32_SMBASE:
4081 		if (!IS_ENABLED(CONFIG_KVM_SMM) || !msr_info->host_initiated)
4082 			return 1;
4083 		msr_info->data = vcpu->arch.smbase;
4084 		break;
4085 	case MSR_SMI_COUNT:
4086 		msr_info->data = vcpu->arch.smi_count;
4087 		break;
4088 	case MSR_IA32_PERF_STATUS:
4089 		/* TSC increment by tick */
4090 		msr_info->data = 1000ULL;
4091 		/* CPU multiplier */
4092 		msr_info->data |= (((uint64_t)4ULL) << 40);
4093 		break;
4094 	case MSR_EFER:
4095 		msr_info->data = vcpu->arch.efer;
4096 		break;
4097 	case MSR_KVM_WALL_CLOCK:
4098 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
4099 			return 1;
4100 
4101 		msr_info->data = vcpu->kvm->arch.wall_clock;
4102 		break;
4103 	case MSR_KVM_WALL_CLOCK_NEW:
4104 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
4105 			return 1;
4106 
4107 		msr_info->data = vcpu->kvm->arch.wall_clock;
4108 		break;
4109 	case MSR_KVM_SYSTEM_TIME:
4110 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
4111 			return 1;
4112 
4113 		msr_info->data = vcpu->arch.time;
4114 		break;
4115 	case MSR_KVM_SYSTEM_TIME_NEW:
4116 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
4117 			return 1;
4118 
4119 		msr_info->data = vcpu->arch.time;
4120 		break;
4121 	case MSR_KVM_ASYNC_PF_EN:
4122 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
4123 			return 1;
4124 
4125 		msr_info->data = vcpu->arch.apf.msr_en_val;
4126 		break;
4127 	case MSR_KVM_ASYNC_PF_INT:
4128 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
4129 			return 1;
4130 
4131 		msr_info->data = vcpu->arch.apf.msr_int_val;
4132 		break;
4133 	case MSR_KVM_ASYNC_PF_ACK:
4134 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
4135 			return 1;
4136 
4137 		msr_info->data = 0;
4138 		break;
4139 	case MSR_KVM_STEAL_TIME:
4140 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
4141 			return 1;
4142 
4143 		msr_info->data = vcpu->arch.st.msr_val;
4144 		break;
4145 	case MSR_KVM_PV_EOI_EN:
4146 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
4147 			return 1;
4148 
4149 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
4150 		break;
4151 	case MSR_KVM_POLL_CONTROL:
4152 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
4153 			return 1;
4154 
4155 		msr_info->data = vcpu->arch.msr_kvm_poll_control;
4156 		break;
4157 	case MSR_IA32_P5_MC_ADDR:
4158 	case MSR_IA32_P5_MC_TYPE:
4159 	case MSR_IA32_MCG_CAP:
4160 	case MSR_IA32_MCG_CTL:
4161 	case MSR_IA32_MCG_STATUS:
4162 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
4163 	case MSR_IA32_MC0_CTL2 ... MSR_IA32_MCx_CTL2(KVM_MAX_MCE_BANKS) - 1:
4164 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
4165 				   msr_info->host_initiated);
4166 	case MSR_IA32_XSS:
4167 		if (!msr_info->host_initiated &&
4168 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
4169 			return 1;
4170 		msr_info->data = vcpu->arch.ia32_xss;
4171 		break;
4172 	case MSR_K7_CLK_CTL:
4173 		/*
4174 		 * Provide expected ramp-up count for K7. All other
4175 		 * are set to zero, indicating minimum divisors for
4176 		 * every field.
4177 		 *
4178 		 * This prevents guest kernels on AMD host with CPU
4179 		 * type 6, model 8 and higher from exploding due to
4180 		 * the rdmsr failing.
4181 		 */
4182 		msr_info->data = 0x20000000;
4183 		break;
4184 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
4185 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
4186 	case HV_X64_MSR_SYNDBG_OPTIONS:
4187 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
4188 	case HV_X64_MSR_CRASH_CTL:
4189 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
4190 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
4191 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
4192 	case HV_X64_MSR_TSC_EMULATION_STATUS:
4193 	case HV_X64_MSR_TSC_INVARIANT_CONTROL:
4194 		return kvm_hv_get_msr_common(vcpu,
4195 					     msr_info->index, &msr_info->data,
4196 					     msr_info->host_initiated);
4197 	case MSR_IA32_BBL_CR_CTL3:
4198 		/* This legacy MSR exists but isn't fully documented in current
4199 		 * silicon.  It is however accessed by winxp in very narrow
4200 		 * scenarios where it sets bit #19, itself documented as
4201 		 * a "reserved" bit.  Best effort attempt to source coherent
4202 		 * read data here should the balance of the register be
4203 		 * interpreted by the guest:
4204 		 *
4205 		 * L2 cache control register 3: 64GB range, 256KB size,
4206 		 * enabled, latency 0x1, configured
4207 		 */
4208 		msr_info->data = 0xbe702111;
4209 		break;
4210 	case MSR_AMD64_OSVW_ID_LENGTH:
4211 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
4212 			return 1;
4213 		msr_info->data = vcpu->arch.osvw.length;
4214 		break;
4215 	case MSR_AMD64_OSVW_STATUS:
4216 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
4217 			return 1;
4218 		msr_info->data = vcpu->arch.osvw.status;
4219 		break;
4220 	case MSR_PLATFORM_INFO:
4221 		if (!msr_info->host_initiated &&
4222 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
4223 			return 1;
4224 		msr_info->data = vcpu->arch.msr_platform_info;
4225 		break;
4226 	case MSR_MISC_FEATURES_ENABLES:
4227 		msr_info->data = vcpu->arch.msr_misc_features_enables;
4228 		break;
4229 	case MSR_K7_HWCR:
4230 		msr_info->data = vcpu->arch.msr_hwcr;
4231 		break;
4232 #ifdef CONFIG_X86_64
4233 	case MSR_IA32_XFD:
4234 		if (!msr_info->host_initiated &&
4235 		    !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
4236 			return 1;
4237 
4238 		msr_info->data = vcpu->arch.guest_fpu.fpstate->xfd;
4239 		break;
4240 	case MSR_IA32_XFD_ERR:
4241 		if (!msr_info->host_initiated &&
4242 		    !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
4243 			return 1;
4244 
4245 		msr_info->data = vcpu->arch.guest_fpu.xfd_err;
4246 		break;
4247 #endif
4248 	default:
4249 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
4250 			return kvm_pmu_get_msr(vcpu, msr_info);
4251 		return KVM_MSR_RET_INVALID;
4252 	}
4253 	return 0;
4254 }
4255 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
4256 
4257 /*
4258  * Read or write a bunch of msrs. All parameters are kernel addresses.
4259  *
4260  * @return number of msrs set successfully.
4261  */
4262 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
4263 		    struct kvm_msr_entry *entries,
4264 		    int (*do_msr)(struct kvm_vcpu *vcpu,
4265 				  unsigned index, u64 *data))
4266 {
4267 	int i;
4268 
4269 	for (i = 0; i < msrs->nmsrs; ++i)
4270 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
4271 			break;
4272 
4273 	return i;
4274 }
4275 
4276 /*
4277  * Read or write a bunch of msrs. Parameters are user addresses.
4278  *
4279  * @return number of msrs set successfully.
4280  */
4281 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
4282 		  int (*do_msr)(struct kvm_vcpu *vcpu,
4283 				unsigned index, u64 *data),
4284 		  int writeback)
4285 {
4286 	struct kvm_msrs msrs;
4287 	struct kvm_msr_entry *entries;
4288 	int r, n;
4289 	unsigned size;
4290 
4291 	r = -EFAULT;
4292 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
4293 		goto out;
4294 
4295 	r = -E2BIG;
4296 	if (msrs.nmsrs >= MAX_IO_MSRS)
4297 		goto out;
4298 
4299 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
4300 	entries = memdup_user(user_msrs->entries, size);
4301 	if (IS_ERR(entries)) {
4302 		r = PTR_ERR(entries);
4303 		goto out;
4304 	}
4305 
4306 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
4307 	if (r < 0)
4308 		goto out_free;
4309 
4310 	r = -EFAULT;
4311 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
4312 		goto out_free;
4313 
4314 	r = n;
4315 
4316 out_free:
4317 	kfree(entries);
4318 out:
4319 	return r;
4320 }
4321 
4322 static inline bool kvm_can_mwait_in_guest(void)
4323 {
4324 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
4325 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
4326 		boot_cpu_has(X86_FEATURE_ARAT);
4327 }
4328 
4329 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
4330 					    struct kvm_cpuid2 __user *cpuid_arg)
4331 {
4332 	struct kvm_cpuid2 cpuid;
4333 	int r;
4334 
4335 	r = -EFAULT;
4336 	if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4337 		return r;
4338 
4339 	r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4340 	if (r)
4341 		return r;
4342 
4343 	r = -EFAULT;
4344 	if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4345 		return r;
4346 
4347 	return 0;
4348 }
4349 
4350 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
4351 {
4352 	int r = 0;
4353 
4354 	switch (ext) {
4355 	case KVM_CAP_IRQCHIP:
4356 	case KVM_CAP_HLT:
4357 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
4358 	case KVM_CAP_SET_TSS_ADDR:
4359 	case KVM_CAP_EXT_CPUID:
4360 	case KVM_CAP_EXT_EMUL_CPUID:
4361 	case KVM_CAP_CLOCKSOURCE:
4362 	case KVM_CAP_PIT:
4363 	case KVM_CAP_NOP_IO_DELAY:
4364 	case KVM_CAP_MP_STATE:
4365 	case KVM_CAP_SYNC_MMU:
4366 	case KVM_CAP_USER_NMI:
4367 	case KVM_CAP_REINJECT_CONTROL:
4368 	case KVM_CAP_IRQ_INJECT_STATUS:
4369 	case KVM_CAP_IOEVENTFD:
4370 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
4371 	case KVM_CAP_PIT2:
4372 	case KVM_CAP_PIT_STATE2:
4373 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
4374 	case KVM_CAP_VCPU_EVENTS:
4375 	case KVM_CAP_HYPERV:
4376 	case KVM_CAP_HYPERV_VAPIC:
4377 	case KVM_CAP_HYPERV_SPIN:
4378 	case KVM_CAP_HYPERV_SYNIC:
4379 	case KVM_CAP_HYPERV_SYNIC2:
4380 	case KVM_CAP_HYPERV_VP_INDEX:
4381 	case KVM_CAP_HYPERV_EVENTFD:
4382 	case KVM_CAP_HYPERV_TLBFLUSH:
4383 	case KVM_CAP_HYPERV_SEND_IPI:
4384 	case KVM_CAP_HYPERV_CPUID:
4385 	case KVM_CAP_HYPERV_ENFORCE_CPUID:
4386 	case KVM_CAP_SYS_HYPERV_CPUID:
4387 	case KVM_CAP_PCI_SEGMENT:
4388 	case KVM_CAP_DEBUGREGS:
4389 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
4390 	case KVM_CAP_XSAVE:
4391 	case KVM_CAP_ASYNC_PF:
4392 	case KVM_CAP_ASYNC_PF_INT:
4393 	case KVM_CAP_GET_TSC_KHZ:
4394 	case KVM_CAP_KVMCLOCK_CTRL:
4395 	case KVM_CAP_READONLY_MEM:
4396 	case KVM_CAP_HYPERV_TIME:
4397 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
4398 	case KVM_CAP_TSC_DEADLINE_TIMER:
4399 	case KVM_CAP_DISABLE_QUIRKS:
4400 	case KVM_CAP_SET_BOOT_CPU_ID:
4401  	case KVM_CAP_SPLIT_IRQCHIP:
4402 	case KVM_CAP_IMMEDIATE_EXIT:
4403 	case KVM_CAP_PMU_EVENT_FILTER:
4404 	case KVM_CAP_PMU_EVENT_MASKED_EVENTS:
4405 	case KVM_CAP_GET_MSR_FEATURES:
4406 	case KVM_CAP_MSR_PLATFORM_INFO:
4407 	case KVM_CAP_EXCEPTION_PAYLOAD:
4408 	case KVM_CAP_X86_TRIPLE_FAULT_EVENT:
4409 	case KVM_CAP_SET_GUEST_DEBUG:
4410 	case KVM_CAP_LAST_CPU:
4411 	case KVM_CAP_X86_USER_SPACE_MSR:
4412 	case KVM_CAP_X86_MSR_FILTER:
4413 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4414 #ifdef CONFIG_X86_SGX_KVM
4415 	case KVM_CAP_SGX_ATTRIBUTE:
4416 #endif
4417 	case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
4418 	case KVM_CAP_VM_MOVE_ENC_CONTEXT_FROM:
4419 	case KVM_CAP_SREGS2:
4420 	case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
4421 	case KVM_CAP_VCPU_ATTRIBUTES:
4422 	case KVM_CAP_SYS_ATTRIBUTES:
4423 	case KVM_CAP_VAPIC:
4424 	case KVM_CAP_ENABLE_CAP:
4425 	case KVM_CAP_VM_DISABLE_NX_HUGE_PAGES:
4426 		r = 1;
4427 		break;
4428 	case KVM_CAP_EXIT_HYPERCALL:
4429 		r = KVM_EXIT_HYPERCALL_VALID_MASK;
4430 		break;
4431 	case KVM_CAP_SET_GUEST_DEBUG2:
4432 		return KVM_GUESTDBG_VALID_MASK;
4433 #ifdef CONFIG_KVM_XEN
4434 	case KVM_CAP_XEN_HVM:
4435 		r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
4436 		    KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
4437 		    KVM_XEN_HVM_CONFIG_SHARED_INFO |
4438 		    KVM_XEN_HVM_CONFIG_EVTCHN_2LEVEL |
4439 		    KVM_XEN_HVM_CONFIG_EVTCHN_SEND;
4440 		if (sched_info_on())
4441 			r |= KVM_XEN_HVM_CONFIG_RUNSTATE |
4442 			     KVM_XEN_HVM_CONFIG_RUNSTATE_UPDATE_FLAG;
4443 		break;
4444 #endif
4445 	case KVM_CAP_SYNC_REGS:
4446 		r = KVM_SYNC_X86_VALID_FIELDS;
4447 		break;
4448 	case KVM_CAP_ADJUST_CLOCK:
4449 		r = KVM_CLOCK_VALID_FLAGS;
4450 		break;
4451 	case KVM_CAP_X86_DISABLE_EXITS:
4452 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
4453 		      KVM_X86_DISABLE_EXITS_CSTATE;
4454 		if(kvm_can_mwait_in_guest())
4455 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
4456 		break;
4457 	case KVM_CAP_X86_SMM:
4458 		if (!IS_ENABLED(CONFIG_KVM_SMM))
4459 			break;
4460 
4461 		/* SMBASE is usually relocated above 1M on modern chipsets,
4462 		 * and SMM handlers might indeed rely on 4G segment limits,
4463 		 * so do not report SMM to be available if real mode is
4464 		 * emulated via vm86 mode.  Still, do not go to great lengths
4465 		 * to avoid userspace's usage of the feature, because it is a
4466 		 * fringe case that is not enabled except via specific settings
4467 		 * of the module parameters.
4468 		 */
4469 		r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
4470 		break;
4471 	case KVM_CAP_NR_VCPUS:
4472 		r = min_t(unsigned int, num_online_cpus(), KVM_MAX_VCPUS);
4473 		break;
4474 	case KVM_CAP_MAX_VCPUS:
4475 		r = KVM_MAX_VCPUS;
4476 		break;
4477 	case KVM_CAP_MAX_VCPU_ID:
4478 		r = KVM_MAX_VCPU_IDS;
4479 		break;
4480 	case KVM_CAP_PV_MMU:	/* obsolete */
4481 		r = 0;
4482 		break;
4483 	case KVM_CAP_MCE:
4484 		r = KVM_MAX_MCE_BANKS;
4485 		break;
4486 	case KVM_CAP_XCRS:
4487 		r = boot_cpu_has(X86_FEATURE_XSAVE);
4488 		break;
4489 	case KVM_CAP_TSC_CONTROL:
4490 	case KVM_CAP_VM_TSC_CONTROL:
4491 		r = kvm_caps.has_tsc_control;
4492 		break;
4493 	case KVM_CAP_X2APIC_API:
4494 		r = KVM_X2APIC_API_VALID_FLAGS;
4495 		break;
4496 	case KVM_CAP_NESTED_STATE:
4497 		r = kvm_x86_ops.nested_ops->get_state ?
4498 			kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
4499 		break;
4500 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4501 		r = kvm_x86_ops.enable_l2_tlb_flush != NULL;
4502 		break;
4503 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4504 		r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
4505 		break;
4506 	case KVM_CAP_SMALLER_MAXPHYADDR:
4507 		r = (int) allow_smaller_maxphyaddr;
4508 		break;
4509 	case KVM_CAP_STEAL_TIME:
4510 		r = sched_info_on();
4511 		break;
4512 	case KVM_CAP_X86_BUS_LOCK_EXIT:
4513 		if (kvm_caps.has_bus_lock_exit)
4514 			r = KVM_BUS_LOCK_DETECTION_OFF |
4515 			    KVM_BUS_LOCK_DETECTION_EXIT;
4516 		else
4517 			r = 0;
4518 		break;
4519 	case KVM_CAP_XSAVE2: {
4520 		u64 guest_perm = xstate_get_guest_group_perm();
4521 
4522 		r = xstate_required_size(kvm_caps.supported_xcr0 & guest_perm, false);
4523 		if (r < sizeof(struct kvm_xsave))
4524 			r = sizeof(struct kvm_xsave);
4525 		break;
4526 	}
4527 	case KVM_CAP_PMU_CAPABILITY:
4528 		r = enable_pmu ? KVM_CAP_PMU_VALID_MASK : 0;
4529 		break;
4530 	case KVM_CAP_DISABLE_QUIRKS2:
4531 		r = KVM_X86_VALID_QUIRKS;
4532 		break;
4533 	case KVM_CAP_X86_NOTIFY_VMEXIT:
4534 		r = kvm_caps.has_notify_vmexit;
4535 		break;
4536 	default:
4537 		break;
4538 	}
4539 	return r;
4540 }
4541 
4542 static inline void __user *kvm_get_attr_addr(struct kvm_device_attr *attr)
4543 {
4544 	void __user *uaddr = (void __user*)(unsigned long)attr->addr;
4545 
4546 	if ((u64)(unsigned long)uaddr != attr->addr)
4547 		return ERR_PTR_USR(-EFAULT);
4548 	return uaddr;
4549 }
4550 
4551 static int kvm_x86_dev_get_attr(struct kvm_device_attr *attr)
4552 {
4553 	u64 __user *uaddr = kvm_get_attr_addr(attr);
4554 
4555 	if (attr->group)
4556 		return -ENXIO;
4557 
4558 	if (IS_ERR(uaddr))
4559 		return PTR_ERR(uaddr);
4560 
4561 	switch (attr->attr) {
4562 	case KVM_X86_XCOMP_GUEST_SUPP:
4563 		if (put_user(kvm_caps.supported_xcr0, uaddr))
4564 			return -EFAULT;
4565 		return 0;
4566 	default:
4567 		return -ENXIO;
4568 		break;
4569 	}
4570 }
4571 
4572 static int kvm_x86_dev_has_attr(struct kvm_device_attr *attr)
4573 {
4574 	if (attr->group)
4575 		return -ENXIO;
4576 
4577 	switch (attr->attr) {
4578 	case KVM_X86_XCOMP_GUEST_SUPP:
4579 		return 0;
4580 	default:
4581 		return -ENXIO;
4582 	}
4583 }
4584 
4585 long kvm_arch_dev_ioctl(struct file *filp,
4586 			unsigned int ioctl, unsigned long arg)
4587 {
4588 	void __user *argp = (void __user *)arg;
4589 	long r;
4590 
4591 	switch (ioctl) {
4592 	case KVM_GET_MSR_INDEX_LIST: {
4593 		struct kvm_msr_list __user *user_msr_list = argp;
4594 		struct kvm_msr_list msr_list;
4595 		unsigned n;
4596 
4597 		r = -EFAULT;
4598 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4599 			goto out;
4600 		n = msr_list.nmsrs;
4601 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4602 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4603 			goto out;
4604 		r = -E2BIG;
4605 		if (n < msr_list.nmsrs)
4606 			goto out;
4607 		r = -EFAULT;
4608 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4609 				 num_msrs_to_save * sizeof(u32)))
4610 			goto out;
4611 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4612 				 &emulated_msrs,
4613 				 num_emulated_msrs * sizeof(u32)))
4614 			goto out;
4615 		r = 0;
4616 		break;
4617 	}
4618 	case KVM_GET_SUPPORTED_CPUID:
4619 	case KVM_GET_EMULATED_CPUID: {
4620 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4621 		struct kvm_cpuid2 cpuid;
4622 
4623 		r = -EFAULT;
4624 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4625 			goto out;
4626 
4627 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4628 					    ioctl);
4629 		if (r)
4630 			goto out;
4631 
4632 		r = -EFAULT;
4633 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4634 			goto out;
4635 		r = 0;
4636 		break;
4637 	}
4638 	case KVM_X86_GET_MCE_CAP_SUPPORTED:
4639 		r = -EFAULT;
4640 		if (copy_to_user(argp, &kvm_caps.supported_mce_cap,
4641 				 sizeof(kvm_caps.supported_mce_cap)))
4642 			goto out;
4643 		r = 0;
4644 		break;
4645 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4646 		struct kvm_msr_list __user *user_msr_list = argp;
4647 		struct kvm_msr_list msr_list;
4648 		unsigned int n;
4649 
4650 		r = -EFAULT;
4651 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4652 			goto out;
4653 		n = msr_list.nmsrs;
4654 		msr_list.nmsrs = num_msr_based_features;
4655 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4656 			goto out;
4657 		r = -E2BIG;
4658 		if (n < msr_list.nmsrs)
4659 			goto out;
4660 		r = -EFAULT;
4661 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
4662 				 num_msr_based_features * sizeof(u32)))
4663 			goto out;
4664 		r = 0;
4665 		break;
4666 	}
4667 	case KVM_GET_MSRS:
4668 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
4669 		break;
4670 	case KVM_GET_SUPPORTED_HV_CPUID:
4671 		r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4672 		break;
4673 	case KVM_GET_DEVICE_ATTR: {
4674 		struct kvm_device_attr attr;
4675 		r = -EFAULT;
4676 		if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4677 			break;
4678 		r = kvm_x86_dev_get_attr(&attr);
4679 		break;
4680 	}
4681 	case KVM_HAS_DEVICE_ATTR: {
4682 		struct kvm_device_attr attr;
4683 		r = -EFAULT;
4684 		if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4685 			break;
4686 		r = kvm_x86_dev_has_attr(&attr);
4687 		break;
4688 	}
4689 	default:
4690 		r = -EINVAL;
4691 		break;
4692 	}
4693 out:
4694 	return r;
4695 }
4696 
4697 static void wbinvd_ipi(void *garbage)
4698 {
4699 	wbinvd();
4700 }
4701 
4702 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4703 {
4704 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4705 }
4706 
4707 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4708 {
4709 	/* Address WBINVD may be executed by guest */
4710 	if (need_emulate_wbinvd(vcpu)) {
4711 		if (static_call(kvm_x86_has_wbinvd_exit)())
4712 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4713 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4714 			smp_call_function_single(vcpu->cpu,
4715 					wbinvd_ipi, NULL, 1);
4716 	}
4717 
4718 	static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4719 
4720 	/* Save host pkru register if supported */
4721 	vcpu->arch.host_pkru = read_pkru();
4722 
4723 	/* Apply any externally detected TSC adjustments (due to suspend) */
4724 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4725 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4726 		vcpu->arch.tsc_offset_adjustment = 0;
4727 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4728 	}
4729 
4730 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4731 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4732 				rdtsc() - vcpu->arch.last_host_tsc;
4733 		if (tsc_delta < 0)
4734 			mark_tsc_unstable("KVM discovered backwards TSC");
4735 
4736 		if (kvm_check_tsc_unstable()) {
4737 			u64 offset = kvm_compute_l1_tsc_offset(vcpu,
4738 						vcpu->arch.last_guest_tsc);
4739 			kvm_vcpu_write_tsc_offset(vcpu, offset);
4740 			vcpu->arch.tsc_catchup = 1;
4741 		}
4742 
4743 		if (kvm_lapic_hv_timer_in_use(vcpu))
4744 			kvm_lapic_restart_hv_timer(vcpu);
4745 
4746 		/*
4747 		 * On a host with synchronized TSC, there is no need to update
4748 		 * kvmclock on vcpu->cpu migration
4749 		 */
4750 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4751 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4752 		if (vcpu->cpu != cpu)
4753 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4754 		vcpu->cpu = cpu;
4755 	}
4756 
4757 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4758 }
4759 
4760 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4761 {
4762 	struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
4763 	struct kvm_steal_time __user *st;
4764 	struct kvm_memslots *slots;
4765 	static const u8 preempted = KVM_VCPU_PREEMPTED;
4766 	gpa_t gpa = vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS;
4767 
4768 	/*
4769 	 * The vCPU can be marked preempted if and only if the VM-Exit was on
4770 	 * an instruction boundary and will not trigger guest emulation of any
4771 	 * kind (see vcpu_run).  Vendor specific code controls (conservatively)
4772 	 * when this is true, for example allowing the vCPU to be marked
4773 	 * preempted if and only if the VM-Exit was due to a host interrupt.
4774 	 */
4775 	if (!vcpu->arch.at_instruction_boundary) {
4776 		vcpu->stat.preemption_other++;
4777 		return;
4778 	}
4779 
4780 	vcpu->stat.preemption_reported++;
4781 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4782 		return;
4783 
4784 	if (vcpu->arch.st.preempted)
4785 		return;
4786 
4787 	/* This happens on process exit */
4788 	if (unlikely(current->mm != vcpu->kvm->mm))
4789 		return;
4790 
4791 	slots = kvm_memslots(vcpu->kvm);
4792 
4793 	if (unlikely(slots->generation != ghc->generation ||
4794 		     gpa != ghc->gpa ||
4795 		     kvm_is_error_hva(ghc->hva) || !ghc->memslot))
4796 		return;
4797 
4798 	st = (struct kvm_steal_time __user *)ghc->hva;
4799 	BUILD_BUG_ON(sizeof(st->preempted) != sizeof(preempted));
4800 
4801 	if (!copy_to_user_nofault(&st->preempted, &preempted, sizeof(preempted)))
4802 		vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4803 
4804 	mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
4805 }
4806 
4807 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4808 {
4809 	int idx;
4810 
4811 	if (vcpu->preempted) {
4812 		if (!vcpu->arch.guest_state_protected)
4813 			vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4814 
4815 		/*
4816 		 * Take the srcu lock as memslots will be accessed to check the gfn
4817 		 * cache generation against the memslots generation.
4818 		 */
4819 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4820 		if (kvm_xen_msr_enabled(vcpu->kvm))
4821 			kvm_xen_runstate_set_preempted(vcpu);
4822 		else
4823 			kvm_steal_time_set_preempted(vcpu);
4824 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4825 	}
4826 
4827 	static_call(kvm_x86_vcpu_put)(vcpu);
4828 	vcpu->arch.last_host_tsc = rdtsc();
4829 }
4830 
4831 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4832 				    struct kvm_lapic_state *s)
4833 {
4834 	static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
4835 
4836 	return kvm_apic_get_state(vcpu, s);
4837 }
4838 
4839 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4840 				    struct kvm_lapic_state *s)
4841 {
4842 	int r;
4843 
4844 	r = kvm_apic_set_state(vcpu, s);
4845 	if (r)
4846 		return r;
4847 	update_cr8_intercept(vcpu);
4848 
4849 	return 0;
4850 }
4851 
4852 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4853 {
4854 	/*
4855 	 * We can accept userspace's request for interrupt injection
4856 	 * as long as we have a place to store the interrupt number.
4857 	 * The actual injection will happen when the CPU is able to
4858 	 * deliver the interrupt.
4859 	 */
4860 	if (kvm_cpu_has_extint(vcpu))
4861 		return false;
4862 
4863 	/* Acknowledging ExtINT does not happen if LINT0 is masked.  */
4864 	return (!lapic_in_kernel(vcpu) ||
4865 		kvm_apic_accept_pic_intr(vcpu));
4866 }
4867 
4868 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4869 {
4870 	/*
4871 	 * Do not cause an interrupt window exit if an exception
4872 	 * is pending or an event needs reinjection; userspace
4873 	 * might want to inject the interrupt manually using KVM_SET_REGS
4874 	 * or KVM_SET_SREGS.  For that to work, we must be at an
4875 	 * instruction boundary and with no events half-injected.
4876 	 */
4877 	return (kvm_arch_interrupt_allowed(vcpu) &&
4878 		kvm_cpu_accept_dm_intr(vcpu) &&
4879 		!kvm_event_needs_reinjection(vcpu) &&
4880 		!kvm_is_exception_pending(vcpu));
4881 }
4882 
4883 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4884 				    struct kvm_interrupt *irq)
4885 {
4886 	if (irq->irq >= KVM_NR_INTERRUPTS)
4887 		return -EINVAL;
4888 
4889 	if (!irqchip_in_kernel(vcpu->kvm)) {
4890 		kvm_queue_interrupt(vcpu, irq->irq, false);
4891 		kvm_make_request(KVM_REQ_EVENT, vcpu);
4892 		return 0;
4893 	}
4894 
4895 	/*
4896 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4897 	 * fail for in-kernel 8259.
4898 	 */
4899 	if (pic_in_kernel(vcpu->kvm))
4900 		return -ENXIO;
4901 
4902 	if (vcpu->arch.pending_external_vector != -1)
4903 		return -EEXIST;
4904 
4905 	vcpu->arch.pending_external_vector = irq->irq;
4906 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4907 	return 0;
4908 }
4909 
4910 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4911 {
4912 	kvm_inject_nmi(vcpu);
4913 
4914 	return 0;
4915 }
4916 
4917 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4918 					   struct kvm_tpr_access_ctl *tac)
4919 {
4920 	if (tac->flags)
4921 		return -EINVAL;
4922 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
4923 	return 0;
4924 }
4925 
4926 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4927 					u64 mcg_cap)
4928 {
4929 	int r;
4930 	unsigned bank_num = mcg_cap & 0xff, bank;
4931 
4932 	r = -EINVAL;
4933 	if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4934 		goto out;
4935 	if (mcg_cap & ~(kvm_caps.supported_mce_cap | 0xff | 0xff0000))
4936 		goto out;
4937 	r = 0;
4938 	vcpu->arch.mcg_cap = mcg_cap;
4939 	/* Init IA32_MCG_CTL to all 1s */
4940 	if (mcg_cap & MCG_CTL_P)
4941 		vcpu->arch.mcg_ctl = ~(u64)0;
4942 	/* Init IA32_MCi_CTL to all 1s, IA32_MCi_CTL2 to all 0s */
4943 	for (bank = 0; bank < bank_num; bank++) {
4944 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4945 		if (mcg_cap & MCG_CMCI_P)
4946 			vcpu->arch.mci_ctl2_banks[bank] = 0;
4947 	}
4948 
4949 	kvm_apic_after_set_mcg_cap(vcpu);
4950 
4951 	static_call(kvm_x86_setup_mce)(vcpu);
4952 out:
4953 	return r;
4954 }
4955 
4956 /*
4957  * Validate this is an UCNA (uncorrectable no action) error by checking the
4958  * MCG_STATUS and MCi_STATUS registers:
4959  * - none of the bits for Machine Check Exceptions are set
4960  * - both the VAL (valid) and UC (uncorrectable) bits are set
4961  * MCI_STATUS_PCC - Processor Context Corrupted
4962  * MCI_STATUS_S - Signaled as a Machine Check Exception
4963  * MCI_STATUS_AR - Software recoverable Action Required
4964  */
4965 static bool is_ucna(struct kvm_x86_mce *mce)
4966 {
4967 	return	!mce->mcg_status &&
4968 		!(mce->status & (MCI_STATUS_PCC | MCI_STATUS_S | MCI_STATUS_AR)) &&
4969 		(mce->status & MCI_STATUS_VAL) &&
4970 		(mce->status & MCI_STATUS_UC);
4971 }
4972 
4973 static int kvm_vcpu_x86_set_ucna(struct kvm_vcpu *vcpu, struct kvm_x86_mce *mce, u64* banks)
4974 {
4975 	u64 mcg_cap = vcpu->arch.mcg_cap;
4976 
4977 	banks[1] = mce->status;
4978 	banks[2] = mce->addr;
4979 	banks[3] = mce->misc;
4980 	vcpu->arch.mcg_status = mce->mcg_status;
4981 
4982 	if (!(mcg_cap & MCG_CMCI_P) ||
4983 	    !(vcpu->arch.mci_ctl2_banks[mce->bank] & MCI_CTL2_CMCI_EN))
4984 		return 0;
4985 
4986 	if (lapic_in_kernel(vcpu))
4987 		kvm_apic_local_deliver(vcpu->arch.apic, APIC_LVTCMCI);
4988 
4989 	return 0;
4990 }
4991 
4992 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4993 				      struct kvm_x86_mce *mce)
4994 {
4995 	u64 mcg_cap = vcpu->arch.mcg_cap;
4996 	unsigned bank_num = mcg_cap & 0xff;
4997 	u64 *banks = vcpu->arch.mce_banks;
4998 
4999 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
5000 		return -EINVAL;
5001 
5002 	banks += array_index_nospec(4 * mce->bank, 4 * bank_num);
5003 
5004 	if (is_ucna(mce))
5005 		return kvm_vcpu_x86_set_ucna(vcpu, mce, banks);
5006 
5007 	/*
5008 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
5009 	 * reporting is disabled
5010 	 */
5011 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
5012 	    vcpu->arch.mcg_ctl != ~(u64)0)
5013 		return 0;
5014 	/*
5015 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
5016 	 * reporting is disabled for the bank
5017 	 */
5018 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
5019 		return 0;
5020 	if (mce->status & MCI_STATUS_UC) {
5021 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
5022 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
5023 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5024 			return 0;
5025 		}
5026 		if (banks[1] & MCI_STATUS_VAL)
5027 			mce->status |= MCI_STATUS_OVER;
5028 		banks[2] = mce->addr;
5029 		banks[3] = mce->misc;
5030 		vcpu->arch.mcg_status = mce->mcg_status;
5031 		banks[1] = mce->status;
5032 		kvm_queue_exception(vcpu, MC_VECTOR);
5033 	} else if (!(banks[1] & MCI_STATUS_VAL)
5034 		   || !(banks[1] & MCI_STATUS_UC)) {
5035 		if (banks[1] & MCI_STATUS_VAL)
5036 			mce->status |= MCI_STATUS_OVER;
5037 		banks[2] = mce->addr;
5038 		banks[3] = mce->misc;
5039 		banks[1] = mce->status;
5040 	} else
5041 		banks[1] |= MCI_STATUS_OVER;
5042 	return 0;
5043 }
5044 
5045 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
5046 					       struct kvm_vcpu_events *events)
5047 {
5048 	struct kvm_queued_exception *ex;
5049 
5050 	process_nmi(vcpu);
5051 
5052 #ifdef CONFIG_KVM_SMM
5053 	if (kvm_check_request(KVM_REQ_SMI, vcpu))
5054 		process_smi(vcpu);
5055 #endif
5056 
5057 	/*
5058 	 * KVM's ABI only allows for one exception to be migrated.  Luckily,
5059 	 * the only time there can be two queued exceptions is if there's a
5060 	 * non-exiting _injected_ exception, and a pending exiting exception.
5061 	 * In that case, ignore the VM-Exiting exception as it's an extension
5062 	 * of the injected exception.
5063 	 */
5064 	if (vcpu->arch.exception_vmexit.pending &&
5065 	    !vcpu->arch.exception.pending &&
5066 	    !vcpu->arch.exception.injected)
5067 		ex = &vcpu->arch.exception_vmexit;
5068 	else
5069 		ex = &vcpu->arch.exception;
5070 
5071 	/*
5072 	 * In guest mode, payload delivery should be deferred if the exception
5073 	 * will be intercepted by L1, e.g. KVM should not modifying CR2 if L1
5074 	 * intercepts #PF, ditto for DR6 and #DBs.  If the per-VM capability,
5075 	 * KVM_CAP_EXCEPTION_PAYLOAD, is not set, userspace may or may not
5076 	 * propagate the payload and so it cannot be safely deferred.  Deliver
5077 	 * the payload if the capability hasn't been requested.
5078 	 */
5079 	if (!vcpu->kvm->arch.exception_payload_enabled &&
5080 	    ex->pending && ex->has_payload)
5081 		kvm_deliver_exception_payload(vcpu, ex);
5082 
5083 	memset(events, 0, sizeof(*events));
5084 
5085 	/*
5086 	 * The API doesn't provide the instruction length for software
5087 	 * exceptions, so don't report them. As long as the guest RIP
5088 	 * isn't advanced, we should expect to encounter the exception
5089 	 * again.
5090 	 */
5091 	if (!kvm_exception_is_soft(ex->vector)) {
5092 		events->exception.injected = ex->injected;
5093 		events->exception.pending = ex->pending;
5094 		/*
5095 		 * For ABI compatibility, deliberately conflate
5096 		 * pending and injected exceptions when
5097 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
5098 		 */
5099 		if (!vcpu->kvm->arch.exception_payload_enabled)
5100 			events->exception.injected |= ex->pending;
5101 	}
5102 	events->exception.nr = ex->vector;
5103 	events->exception.has_error_code = ex->has_error_code;
5104 	events->exception.error_code = ex->error_code;
5105 	events->exception_has_payload = ex->has_payload;
5106 	events->exception_payload = ex->payload;
5107 
5108 	events->interrupt.injected =
5109 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
5110 	events->interrupt.nr = vcpu->arch.interrupt.nr;
5111 	events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
5112 
5113 	events->nmi.injected = vcpu->arch.nmi_injected;
5114 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
5115 	events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
5116 
5117 	/* events->sipi_vector is never valid when reporting to user space */
5118 
5119 #ifdef CONFIG_KVM_SMM
5120 	events->smi.smm = is_smm(vcpu);
5121 	events->smi.pending = vcpu->arch.smi_pending;
5122 	events->smi.smm_inside_nmi =
5123 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
5124 #endif
5125 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
5126 
5127 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
5128 			 | KVM_VCPUEVENT_VALID_SHADOW
5129 			 | KVM_VCPUEVENT_VALID_SMM);
5130 	if (vcpu->kvm->arch.exception_payload_enabled)
5131 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
5132 	if (vcpu->kvm->arch.triple_fault_event) {
5133 		events->triple_fault.pending = kvm_test_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5134 		events->flags |= KVM_VCPUEVENT_VALID_TRIPLE_FAULT;
5135 	}
5136 }
5137 
5138 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
5139 					      struct kvm_vcpu_events *events)
5140 {
5141 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
5142 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
5143 			      | KVM_VCPUEVENT_VALID_SHADOW
5144 			      | KVM_VCPUEVENT_VALID_SMM
5145 			      | KVM_VCPUEVENT_VALID_PAYLOAD
5146 			      | KVM_VCPUEVENT_VALID_TRIPLE_FAULT))
5147 		return -EINVAL;
5148 
5149 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
5150 		if (!vcpu->kvm->arch.exception_payload_enabled)
5151 			return -EINVAL;
5152 		if (events->exception.pending)
5153 			events->exception.injected = 0;
5154 		else
5155 			events->exception_has_payload = 0;
5156 	} else {
5157 		events->exception.pending = 0;
5158 		events->exception_has_payload = 0;
5159 	}
5160 
5161 	if ((events->exception.injected || events->exception.pending) &&
5162 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
5163 		return -EINVAL;
5164 
5165 	/* INITs are latched while in SMM */
5166 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
5167 	    (events->smi.smm || events->smi.pending) &&
5168 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
5169 		return -EINVAL;
5170 
5171 	process_nmi(vcpu);
5172 
5173 	/*
5174 	 * Flag that userspace is stuffing an exception, the next KVM_RUN will
5175 	 * morph the exception to a VM-Exit if appropriate.  Do this only for
5176 	 * pending exceptions, already-injected exceptions are not subject to
5177 	 * intercpetion.  Note, userspace that conflates pending and injected
5178 	 * is hosed, and will incorrectly convert an injected exception into a
5179 	 * pending exception, which in turn may cause a spurious VM-Exit.
5180 	 */
5181 	vcpu->arch.exception_from_userspace = events->exception.pending;
5182 
5183 	vcpu->arch.exception_vmexit.pending = false;
5184 
5185 	vcpu->arch.exception.injected = events->exception.injected;
5186 	vcpu->arch.exception.pending = events->exception.pending;
5187 	vcpu->arch.exception.vector = events->exception.nr;
5188 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
5189 	vcpu->arch.exception.error_code = events->exception.error_code;
5190 	vcpu->arch.exception.has_payload = events->exception_has_payload;
5191 	vcpu->arch.exception.payload = events->exception_payload;
5192 
5193 	vcpu->arch.interrupt.injected = events->interrupt.injected;
5194 	vcpu->arch.interrupt.nr = events->interrupt.nr;
5195 	vcpu->arch.interrupt.soft = events->interrupt.soft;
5196 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
5197 		static_call(kvm_x86_set_interrupt_shadow)(vcpu,
5198 						events->interrupt.shadow);
5199 
5200 	vcpu->arch.nmi_injected = events->nmi.injected;
5201 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
5202 		vcpu->arch.nmi_pending = events->nmi.pending;
5203 	static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
5204 
5205 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
5206 	    lapic_in_kernel(vcpu))
5207 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
5208 
5209 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
5210 #ifdef CONFIG_KVM_SMM
5211 		if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
5212 			kvm_leave_nested(vcpu);
5213 			kvm_smm_changed(vcpu, events->smi.smm);
5214 		}
5215 
5216 		vcpu->arch.smi_pending = events->smi.pending;
5217 
5218 		if (events->smi.smm) {
5219 			if (events->smi.smm_inside_nmi)
5220 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
5221 			else
5222 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
5223 		}
5224 
5225 #else
5226 		if (events->smi.smm || events->smi.pending ||
5227 		    events->smi.smm_inside_nmi)
5228 			return -EINVAL;
5229 #endif
5230 
5231 		if (lapic_in_kernel(vcpu)) {
5232 			if (events->smi.latched_init)
5233 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
5234 			else
5235 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
5236 		}
5237 	}
5238 
5239 	if (events->flags & KVM_VCPUEVENT_VALID_TRIPLE_FAULT) {
5240 		if (!vcpu->kvm->arch.triple_fault_event)
5241 			return -EINVAL;
5242 		if (events->triple_fault.pending)
5243 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5244 		else
5245 			kvm_clear_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5246 	}
5247 
5248 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5249 
5250 	return 0;
5251 }
5252 
5253 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
5254 					     struct kvm_debugregs *dbgregs)
5255 {
5256 	unsigned long val;
5257 
5258 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
5259 	kvm_get_dr(vcpu, 6, &val);
5260 	dbgregs->dr6 = val;
5261 	dbgregs->dr7 = vcpu->arch.dr7;
5262 	dbgregs->flags = 0;
5263 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
5264 }
5265 
5266 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
5267 					    struct kvm_debugregs *dbgregs)
5268 {
5269 	if (dbgregs->flags)
5270 		return -EINVAL;
5271 
5272 	if (!kvm_dr6_valid(dbgregs->dr6))
5273 		return -EINVAL;
5274 	if (!kvm_dr7_valid(dbgregs->dr7))
5275 		return -EINVAL;
5276 
5277 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
5278 	kvm_update_dr0123(vcpu);
5279 	vcpu->arch.dr6 = dbgregs->dr6;
5280 	vcpu->arch.dr7 = dbgregs->dr7;
5281 	kvm_update_dr7(vcpu);
5282 
5283 	return 0;
5284 }
5285 
5286 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
5287 					 struct kvm_xsave *guest_xsave)
5288 {
5289 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
5290 		return;
5291 
5292 	fpu_copy_guest_fpstate_to_uabi(&vcpu->arch.guest_fpu,
5293 				       guest_xsave->region,
5294 				       sizeof(guest_xsave->region),
5295 				       vcpu->arch.pkru);
5296 }
5297 
5298 static void kvm_vcpu_ioctl_x86_get_xsave2(struct kvm_vcpu *vcpu,
5299 					  u8 *state, unsigned int size)
5300 {
5301 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
5302 		return;
5303 
5304 	fpu_copy_guest_fpstate_to_uabi(&vcpu->arch.guest_fpu,
5305 				       state, size, vcpu->arch.pkru);
5306 }
5307 
5308 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
5309 					struct kvm_xsave *guest_xsave)
5310 {
5311 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
5312 		return 0;
5313 
5314 	return fpu_copy_uabi_to_guest_fpstate(&vcpu->arch.guest_fpu,
5315 					      guest_xsave->region,
5316 					      kvm_caps.supported_xcr0,
5317 					      &vcpu->arch.pkru);
5318 }
5319 
5320 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
5321 					struct kvm_xcrs *guest_xcrs)
5322 {
5323 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
5324 		guest_xcrs->nr_xcrs = 0;
5325 		return;
5326 	}
5327 
5328 	guest_xcrs->nr_xcrs = 1;
5329 	guest_xcrs->flags = 0;
5330 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
5331 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
5332 }
5333 
5334 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
5335 				       struct kvm_xcrs *guest_xcrs)
5336 {
5337 	int i, r = 0;
5338 
5339 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
5340 		return -EINVAL;
5341 
5342 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
5343 		return -EINVAL;
5344 
5345 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
5346 		/* Only support XCR0 currently */
5347 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
5348 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
5349 				guest_xcrs->xcrs[i].value);
5350 			break;
5351 		}
5352 	if (r)
5353 		r = -EINVAL;
5354 	return r;
5355 }
5356 
5357 /*
5358  * kvm_set_guest_paused() indicates to the guest kernel that it has been
5359  * stopped by the hypervisor.  This function will be called from the host only.
5360  * EINVAL is returned when the host attempts to set the flag for a guest that
5361  * does not support pv clocks.
5362  */
5363 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
5364 {
5365 	if (!vcpu->arch.pv_time.active)
5366 		return -EINVAL;
5367 	vcpu->arch.pvclock_set_guest_stopped_request = true;
5368 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5369 	return 0;
5370 }
5371 
5372 static int kvm_arch_tsc_has_attr(struct kvm_vcpu *vcpu,
5373 				 struct kvm_device_attr *attr)
5374 {
5375 	int r;
5376 
5377 	switch (attr->attr) {
5378 	case KVM_VCPU_TSC_OFFSET:
5379 		r = 0;
5380 		break;
5381 	default:
5382 		r = -ENXIO;
5383 	}
5384 
5385 	return r;
5386 }
5387 
5388 static int kvm_arch_tsc_get_attr(struct kvm_vcpu *vcpu,
5389 				 struct kvm_device_attr *attr)
5390 {
5391 	u64 __user *uaddr = kvm_get_attr_addr(attr);
5392 	int r;
5393 
5394 	if (IS_ERR(uaddr))
5395 		return PTR_ERR(uaddr);
5396 
5397 	switch (attr->attr) {
5398 	case KVM_VCPU_TSC_OFFSET:
5399 		r = -EFAULT;
5400 		if (put_user(vcpu->arch.l1_tsc_offset, uaddr))
5401 			break;
5402 		r = 0;
5403 		break;
5404 	default:
5405 		r = -ENXIO;
5406 	}
5407 
5408 	return r;
5409 }
5410 
5411 static int kvm_arch_tsc_set_attr(struct kvm_vcpu *vcpu,
5412 				 struct kvm_device_attr *attr)
5413 {
5414 	u64 __user *uaddr = kvm_get_attr_addr(attr);
5415 	struct kvm *kvm = vcpu->kvm;
5416 	int r;
5417 
5418 	if (IS_ERR(uaddr))
5419 		return PTR_ERR(uaddr);
5420 
5421 	switch (attr->attr) {
5422 	case KVM_VCPU_TSC_OFFSET: {
5423 		u64 offset, tsc, ns;
5424 		unsigned long flags;
5425 		bool matched;
5426 
5427 		r = -EFAULT;
5428 		if (get_user(offset, uaddr))
5429 			break;
5430 
5431 		raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
5432 
5433 		matched = (vcpu->arch.virtual_tsc_khz &&
5434 			   kvm->arch.last_tsc_khz == vcpu->arch.virtual_tsc_khz &&
5435 			   kvm->arch.last_tsc_offset == offset);
5436 
5437 		tsc = kvm_scale_tsc(rdtsc(), vcpu->arch.l1_tsc_scaling_ratio) + offset;
5438 		ns = get_kvmclock_base_ns();
5439 
5440 		__kvm_synchronize_tsc(vcpu, offset, tsc, ns, matched);
5441 		raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
5442 
5443 		r = 0;
5444 		break;
5445 	}
5446 	default:
5447 		r = -ENXIO;
5448 	}
5449 
5450 	return r;
5451 }
5452 
5453 static int kvm_vcpu_ioctl_device_attr(struct kvm_vcpu *vcpu,
5454 				      unsigned int ioctl,
5455 				      void __user *argp)
5456 {
5457 	struct kvm_device_attr attr;
5458 	int r;
5459 
5460 	if (copy_from_user(&attr, argp, sizeof(attr)))
5461 		return -EFAULT;
5462 
5463 	if (attr.group != KVM_VCPU_TSC_CTRL)
5464 		return -ENXIO;
5465 
5466 	switch (ioctl) {
5467 	case KVM_HAS_DEVICE_ATTR:
5468 		r = kvm_arch_tsc_has_attr(vcpu, &attr);
5469 		break;
5470 	case KVM_GET_DEVICE_ATTR:
5471 		r = kvm_arch_tsc_get_attr(vcpu, &attr);
5472 		break;
5473 	case KVM_SET_DEVICE_ATTR:
5474 		r = kvm_arch_tsc_set_attr(vcpu, &attr);
5475 		break;
5476 	}
5477 
5478 	return r;
5479 }
5480 
5481 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
5482 				     struct kvm_enable_cap *cap)
5483 {
5484 	int r;
5485 	uint16_t vmcs_version;
5486 	void __user *user_ptr;
5487 
5488 	if (cap->flags)
5489 		return -EINVAL;
5490 
5491 	switch (cap->cap) {
5492 	case KVM_CAP_HYPERV_SYNIC2:
5493 		if (cap->args[0])
5494 			return -EINVAL;
5495 		fallthrough;
5496 
5497 	case KVM_CAP_HYPERV_SYNIC:
5498 		if (!irqchip_in_kernel(vcpu->kvm))
5499 			return -EINVAL;
5500 		return kvm_hv_activate_synic(vcpu, cap->cap ==
5501 					     KVM_CAP_HYPERV_SYNIC2);
5502 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
5503 		if (!kvm_x86_ops.nested_ops->enable_evmcs)
5504 			return -ENOTTY;
5505 		r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
5506 		if (!r) {
5507 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
5508 			if (copy_to_user(user_ptr, &vmcs_version,
5509 					 sizeof(vmcs_version)))
5510 				r = -EFAULT;
5511 		}
5512 		return r;
5513 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
5514 		if (!kvm_x86_ops.enable_l2_tlb_flush)
5515 			return -ENOTTY;
5516 
5517 		return static_call(kvm_x86_enable_l2_tlb_flush)(vcpu);
5518 
5519 	case KVM_CAP_HYPERV_ENFORCE_CPUID:
5520 		return kvm_hv_set_enforce_cpuid(vcpu, cap->args[0]);
5521 
5522 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
5523 		vcpu->arch.pv_cpuid.enforce = cap->args[0];
5524 		if (vcpu->arch.pv_cpuid.enforce)
5525 			kvm_update_pv_runtime(vcpu);
5526 
5527 		return 0;
5528 	default:
5529 		return -EINVAL;
5530 	}
5531 }
5532 
5533 long kvm_arch_vcpu_ioctl(struct file *filp,
5534 			 unsigned int ioctl, unsigned long arg)
5535 {
5536 	struct kvm_vcpu *vcpu = filp->private_data;
5537 	void __user *argp = (void __user *)arg;
5538 	int r;
5539 	union {
5540 		struct kvm_sregs2 *sregs2;
5541 		struct kvm_lapic_state *lapic;
5542 		struct kvm_xsave *xsave;
5543 		struct kvm_xcrs *xcrs;
5544 		void *buffer;
5545 	} u;
5546 
5547 	vcpu_load(vcpu);
5548 
5549 	u.buffer = NULL;
5550 	switch (ioctl) {
5551 	case KVM_GET_LAPIC: {
5552 		r = -EINVAL;
5553 		if (!lapic_in_kernel(vcpu))
5554 			goto out;
5555 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
5556 				GFP_KERNEL_ACCOUNT);
5557 
5558 		r = -ENOMEM;
5559 		if (!u.lapic)
5560 			goto out;
5561 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
5562 		if (r)
5563 			goto out;
5564 		r = -EFAULT;
5565 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
5566 			goto out;
5567 		r = 0;
5568 		break;
5569 	}
5570 	case KVM_SET_LAPIC: {
5571 		r = -EINVAL;
5572 		if (!lapic_in_kernel(vcpu))
5573 			goto out;
5574 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
5575 		if (IS_ERR(u.lapic)) {
5576 			r = PTR_ERR(u.lapic);
5577 			goto out_nofree;
5578 		}
5579 
5580 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
5581 		break;
5582 	}
5583 	case KVM_INTERRUPT: {
5584 		struct kvm_interrupt irq;
5585 
5586 		r = -EFAULT;
5587 		if (copy_from_user(&irq, argp, sizeof(irq)))
5588 			goto out;
5589 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
5590 		break;
5591 	}
5592 	case KVM_NMI: {
5593 		r = kvm_vcpu_ioctl_nmi(vcpu);
5594 		break;
5595 	}
5596 	case KVM_SMI: {
5597 		r = kvm_inject_smi(vcpu);
5598 		break;
5599 	}
5600 	case KVM_SET_CPUID: {
5601 		struct kvm_cpuid __user *cpuid_arg = argp;
5602 		struct kvm_cpuid cpuid;
5603 
5604 		r = -EFAULT;
5605 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5606 			goto out;
5607 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
5608 		break;
5609 	}
5610 	case KVM_SET_CPUID2: {
5611 		struct kvm_cpuid2 __user *cpuid_arg = argp;
5612 		struct kvm_cpuid2 cpuid;
5613 
5614 		r = -EFAULT;
5615 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5616 			goto out;
5617 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
5618 					      cpuid_arg->entries);
5619 		break;
5620 	}
5621 	case KVM_GET_CPUID2: {
5622 		struct kvm_cpuid2 __user *cpuid_arg = argp;
5623 		struct kvm_cpuid2 cpuid;
5624 
5625 		r = -EFAULT;
5626 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5627 			goto out;
5628 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
5629 					      cpuid_arg->entries);
5630 		if (r)
5631 			goto out;
5632 		r = -EFAULT;
5633 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
5634 			goto out;
5635 		r = 0;
5636 		break;
5637 	}
5638 	case KVM_GET_MSRS: {
5639 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
5640 		r = msr_io(vcpu, argp, do_get_msr, 1);
5641 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5642 		break;
5643 	}
5644 	case KVM_SET_MSRS: {
5645 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
5646 		r = msr_io(vcpu, argp, do_set_msr, 0);
5647 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5648 		break;
5649 	}
5650 	case KVM_TPR_ACCESS_REPORTING: {
5651 		struct kvm_tpr_access_ctl tac;
5652 
5653 		r = -EFAULT;
5654 		if (copy_from_user(&tac, argp, sizeof(tac)))
5655 			goto out;
5656 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
5657 		if (r)
5658 			goto out;
5659 		r = -EFAULT;
5660 		if (copy_to_user(argp, &tac, sizeof(tac)))
5661 			goto out;
5662 		r = 0;
5663 		break;
5664 	};
5665 	case KVM_SET_VAPIC_ADDR: {
5666 		struct kvm_vapic_addr va;
5667 		int idx;
5668 
5669 		r = -EINVAL;
5670 		if (!lapic_in_kernel(vcpu))
5671 			goto out;
5672 		r = -EFAULT;
5673 		if (copy_from_user(&va, argp, sizeof(va)))
5674 			goto out;
5675 		idx = srcu_read_lock(&vcpu->kvm->srcu);
5676 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
5677 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5678 		break;
5679 	}
5680 	case KVM_X86_SETUP_MCE: {
5681 		u64 mcg_cap;
5682 
5683 		r = -EFAULT;
5684 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
5685 			goto out;
5686 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
5687 		break;
5688 	}
5689 	case KVM_X86_SET_MCE: {
5690 		struct kvm_x86_mce mce;
5691 
5692 		r = -EFAULT;
5693 		if (copy_from_user(&mce, argp, sizeof(mce)))
5694 			goto out;
5695 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
5696 		break;
5697 	}
5698 	case KVM_GET_VCPU_EVENTS: {
5699 		struct kvm_vcpu_events events;
5700 
5701 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
5702 
5703 		r = -EFAULT;
5704 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
5705 			break;
5706 		r = 0;
5707 		break;
5708 	}
5709 	case KVM_SET_VCPU_EVENTS: {
5710 		struct kvm_vcpu_events events;
5711 
5712 		r = -EFAULT;
5713 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5714 			break;
5715 
5716 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5717 		break;
5718 	}
5719 	case KVM_GET_DEBUGREGS: {
5720 		struct kvm_debugregs dbgregs;
5721 
5722 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5723 
5724 		r = -EFAULT;
5725 		if (copy_to_user(argp, &dbgregs,
5726 				 sizeof(struct kvm_debugregs)))
5727 			break;
5728 		r = 0;
5729 		break;
5730 	}
5731 	case KVM_SET_DEBUGREGS: {
5732 		struct kvm_debugregs dbgregs;
5733 
5734 		r = -EFAULT;
5735 		if (copy_from_user(&dbgregs, argp,
5736 				   sizeof(struct kvm_debugregs)))
5737 			break;
5738 
5739 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5740 		break;
5741 	}
5742 	case KVM_GET_XSAVE: {
5743 		r = -EINVAL;
5744 		if (vcpu->arch.guest_fpu.uabi_size > sizeof(struct kvm_xsave))
5745 			break;
5746 
5747 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5748 		r = -ENOMEM;
5749 		if (!u.xsave)
5750 			break;
5751 
5752 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5753 
5754 		r = -EFAULT;
5755 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5756 			break;
5757 		r = 0;
5758 		break;
5759 	}
5760 	case KVM_SET_XSAVE: {
5761 		int size = vcpu->arch.guest_fpu.uabi_size;
5762 
5763 		u.xsave = memdup_user(argp, size);
5764 		if (IS_ERR(u.xsave)) {
5765 			r = PTR_ERR(u.xsave);
5766 			goto out_nofree;
5767 		}
5768 
5769 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5770 		break;
5771 	}
5772 
5773 	case KVM_GET_XSAVE2: {
5774 		int size = vcpu->arch.guest_fpu.uabi_size;
5775 
5776 		u.xsave = kzalloc(size, GFP_KERNEL_ACCOUNT);
5777 		r = -ENOMEM;
5778 		if (!u.xsave)
5779 			break;
5780 
5781 		kvm_vcpu_ioctl_x86_get_xsave2(vcpu, u.buffer, size);
5782 
5783 		r = -EFAULT;
5784 		if (copy_to_user(argp, u.xsave, size))
5785 			break;
5786 
5787 		r = 0;
5788 		break;
5789 	}
5790 
5791 	case KVM_GET_XCRS: {
5792 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5793 		r = -ENOMEM;
5794 		if (!u.xcrs)
5795 			break;
5796 
5797 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5798 
5799 		r = -EFAULT;
5800 		if (copy_to_user(argp, u.xcrs,
5801 				 sizeof(struct kvm_xcrs)))
5802 			break;
5803 		r = 0;
5804 		break;
5805 	}
5806 	case KVM_SET_XCRS: {
5807 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5808 		if (IS_ERR(u.xcrs)) {
5809 			r = PTR_ERR(u.xcrs);
5810 			goto out_nofree;
5811 		}
5812 
5813 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5814 		break;
5815 	}
5816 	case KVM_SET_TSC_KHZ: {
5817 		u32 user_tsc_khz;
5818 
5819 		r = -EINVAL;
5820 		user_tsc_khz = (u32)arg;
5821 
5822 		if (kvm_caps.has_tsc_control &&
5823 		    user_tsc_khz >= kvm_caps.max_guest_tsc_khz)
5824 			goto out;
5825 
5826 		if (user_tsc_khz == 0)
5827 			user_tsc_khz = tsc_khz;
5828 
5829 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5830 			r = 0;
5831 
5832 		goto out;
5833 	}
5834 	case KVM_GET_TSC_KHZ: {
5835 		r = vcpu->arch.virtual_tsc_khz;
5836 		goto out;
5837 	}
5838 	case KVM_KVMCLOCK_CTRL: {
5839 		r = kvm_set_guest_paused(vcpu);
5840 		goto out;
5841 	}
5842 	case KVM_ENABLE_CAP: {
5843 		struct kvm_enable_cap cap;
5844 
5845 		r = -EFAULT;
5846 		if (copy_from_user(&cap, argp, sizeof(cap)))
5847 			goto out;
5848 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5849 		break;
5850 	}
5851 	case KVM_GET_NESTED_STATE: {
5852 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5853 		u32 user_data_size;
5854 
5855 		r = -EINVAL;
5856 		if (!kvm_x86_ops.nested_ops->get_state)
5857 			break;
5858 
5859 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5860 		r = -EFAULT;
5861 		if (get_user(user_data_size, &user_kvm_nested_state->size))
5862 			break;
5863 
5864 		r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5865 						     user_data_size);
5866 		if (r < 0)
5867 			break;
5868 
5869 		if (r > user_data_size) {
5870 			if (put_user(r, &user_kvm_nested_state->size))
5871 				r = -EFAULT;
5872 			else
5873 				r = -E2BIG;
5874 			break;
5875 		}
5876 
5877 		r = 0;
5878 		break;
5879 	}
5880 	case KVM_SET_NESTED_STATE: {
5881 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5882 		struct kvm_nested_state kvm_state;
5883 		int idx;
5884 
5885 		r = -EINVAL;
5886 		if (!kvm_x86_ops.nested_ops->set_state)
5887 			break;
5888 
5889 		r = -EFAULT;
5890 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5891 			break;
5892 
5893 		r = -EINVAL;
5894 		if (kvm_state.size < sizeof(kvm_state))
5895 			break;
5896 
5897 		if (kvm_state.flags &
5898 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5899 		      | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5900 		      | KVM_STATE_NESTED_GIF_SET))
5901 			break;
5902 
5903 		/* nested_run_pending implies guest_mode.  */
5904 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5905 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5906 			break;
5907 
5908 		idx = srcu_read_lock(&vcpu->kvm->srcu);
5909 		r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5910 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5911 		break;
5912 	}
5913 	case KVM_GET_SUPPORTED_HV_CPUID:
5914 		r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5915 		break;
5916 #ifdef CONFIG_KVM_XEN
5917 	case KVM_XEN_VCPU_GET_ATTR: {
5918 		struct kvm_xen_vcpu_attr xva;
5919 
5920 		r = -EFAULT;
5921 		if (copy_from_user(&xva, argp, sizeof(xva)))
5922 			goto out;
5923 		r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5924 		if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5925 			r = -EFAULT;
5926 		break;
5927 	}
5928 	case KVM_XEN_VCPU_SET_ATTR: {
5929 		struct kvm_xen_vcpu_attr xva;
5930 
5931 		r = -EFAULT;
5932 		if (copy_from_user(&xva, argp, sizeof(xva)))
5933 			goto out;
5934 		r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5935 		break;
5936 	}
5937 #endif
5938 	case KVM_GET_SREGS2: {
5939 		u.sregs2 = kzalloc(sizeof(struct kvm_sregs2), GFP_KERNEL);
5940 		r = -ENOMEM;
5941 		if (!u.sregs2)
5942 			goto out;
5943 		__get_sregs2(vcpu, u.sregs2);
5944 		r = -EFAULT;
5945 		if (copy_to_user(argp, u.sregs2, sizeof(struct kvm_sregs2)))
5946 			goto out;
5947 		r = 0;
5948 		break;
5949 	}
5950 	case KVM_SET_SREGS2: {
5951 		u.sregs2 = memdup_user(argp, sizeof(struct kvm_sregs2));
5952 		if (IS_ERR(u.sregs2)) {
5953 			r = PTR_ERR(u.sregs2);
5954 			u.sregs2 = NULL;
5955 			goto out;
5956 		}
5957 		r = __set_sregs2(vcpu, u.sregs2);
5958 		break;
5959 	}
5960 	case KVM_HAS_DEVICE_ATTR:
5961 	case KVM_GET_DEVICE_ATTR:
5962 	case KVM_SET_DEVICE_ATTR:
5963 		r = kvm_vcpu_ioctl_device_attr(vcpu, ioctl, argp);
5964 		break;
5965 	default:
5966 		r = -EINVAL;
5967 	}
5968 out:
5969 	kfree(u.buffer);
5970 out_nofree:
5971 	vcpu_put(vcpu);
5972 	return r;
5973 }
5974 
5975 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5976 {
5977 	return VM_FAULT_SIGBUS;
5978 }
5979 
5980 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5981 {
5982 	int ret;
5983 
5984 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
5985 		return -EINVAL;
5986 	ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5987 	return ret;
5988 }
5989 
5990 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5991 					      u64 ident_addr)
5992 {
5993 	return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5994 }
5995 
5996 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5997 					 unsigned long kvm_nr_mmu_pages)
5998 {
5999 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
6000 		return -EINVAL;
6001 
6002 	mutex_lock(&kvm->slots_lock);
6003 
6004 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
6005 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
6006 
6007 	mutex_unlock(&kvm->slots_lock);
6008 	return 0;
6009 }
6010 
6011 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
6012 {
6013 	return kvm->arch.n_max_mmu_pages;
6014 }
6015 
6016 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
6017 {
6018 	struct kvm_pic *pic = kvm->arch.vpic;
6019 	int r;
6020 
6021 	r = 0;
6022 	switch (chip->chip_id) {
6023 	case KVM_IRQCHIP_PIC_MASTER:
6024 		memcpy(&chip->chip.pic, &pic->pics[0],
6025 			sizeof(struct kvm_pic_state));
6026 		break;
6027 	case KVM_IRQCHIP_PIC_SLAVE:
6028 		memcpy(&chip->chip.pic, &pic->pics[1],
6029 			sizeof(struct kvm_pic_state));
6030 		break;
6031 	case KVM_IRQCHIP_IOAPIC:
6032 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
6033 		break;
6034 	default:
6035 		r = -EINVAL;
6036 		break;
6037 	}
6038 	return r;
6039 }
6040 
6041 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
6042 {
6043 	struct kvm_pic *pic = kvm->arch.vpic;
6044 	int r;
6045 
6046 	r = 0;
6047 	switch (chip->chip_id) {
6048 	case KVM_IRQCHIP_PIC_MASTER:
6049 		spin_lock(&pic->lock);
6050 		memcpy(&pic->pics[0], &chip->chip.pic,
6051 			sizeof(struct kvm_pic_state));
6052 		spin_unlock(&pic->lock);
6053 		break;
6054 	case KVM_IRQCHIP_PIC_SLAVE:
6055 		spin_lock(&pic->lock);
6056 		memcpy(&pic->pics[1], &chip->chip.pic,
6057 			sizeof(struct kvm_pic_state));
6058 		spin_unlock(&pic->lock);
6059 		break;
6060 	case KVM_IRQCHIP_IOAPIC:
6061 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
6062 		break;
6063 	default:
6064 		r = -EINVAL;
6065 		break;
6066 	}
6067 	kvm_pic_update_irq(pic);
6068 	return r;
6069 }
6070 
6071 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
6072 {
6073 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
6074 
6075 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
6076 
6077 	mutex_lock(&kps->lock);
6078 	memcpy(ps, &kps->channels, sizeof(*ps));
6079 	mutex_unlock(&kps->lock);
6080 	return 0;
6081 }
6082 
6083 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
6084 {
6085 	int i;
6086 	struct kvm_pit *pit = kvm->arch.vpit;
6087 
6088 	mutex_lock(&pit->pit_state.lock);
6089 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
6090 	for (i = 0; i < 3; i++)
6091 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
6092 	mutex_unlock(&pit->pit_state.lock);
6093 	return 0;
6094 }
6095 
6096 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
6097 {
6098 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
6099 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
6100 		sizeof(ps->channels));
6101 	ps->flags = kvm->arch.vpit->pit_state.flags;
6102 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
6103 	memset(&ps->reserved, 0, sizeof(ps->reserved));
6104 	return 0;
6105 }
6106 
6107 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
6108 {
6109 	int start = 0;
6110 	int i;
6111 	u32 prev_legacy, cur_legacy;
6112 	struct kvm_pit *pit = kvm->arch.vpit;
6113 
6114 	mutex_lock(&pit->pit_state.lock);
6115 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
6116 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
6117 	if (!prev_legacy && cur_legacy)
6118 		start = 1;
6119 	memcpy(&pit->pit_state.channels, &ps->channels,
6120 	       sizeof(pit->pit_state.channels));
6121 	pit->pit_state.flags = ps->flags;
6122 	for (i = 0; i < 3; i++)
6123 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
6124 				   start && i == 0);
6125 	mutex_unlock(&pit->pit_state.lock);
6126 	return 0;
6127 }
6128 
6129 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
6130 				 struct kvm_reinject_control *control)
6131 {
6132 	struct kvm_pit *pit = kvm->arch.vpit;
6133 
6134 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
6135 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
6136 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
6137 	 */
6138 	mutex_lock(&pit->pit_state.lock);
6139 	kvm_pit_set_reinject(pit, control->pit_reinject);
6140 	mutex_unlock(&pit->pit_state.lock);
6141 
6142 	return 0;
6143 }
6144 
6145 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
6146 {
6147 
6148 	/*
6149 	 * Flush all CPUs' dirty log buffers to the  dirty_bitmap.  Called
6150 	 * before reporting dirty_bitmap to userspace.  KVM flushes the buffers
6151 	 * on all VM-Exits, thus we only need to kick running vCPUs to force a
6152 	 * VM-Exit.
6153 	 */
6154 	struct kvm_vcpu *vcpu;
6155 	unsigned long i;
6156 
6157 	kvm_for_each_vcpu(i, vcpu, kvm)
6158 		kvm_vcpu_kick(vcpu);
6159 }
6160 
6161 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
6162 			bool line_status)
6163 {
6164 	if (!irqchip_in_kernel(kvm))
6165 		return -ENXIO;
6166 
6167 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
6168 					irq_event->irq, irq_event->level,
6169 					line_status);
6170 	return 0;
6171 }
6172 
6173 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
6174 			    struct kvm_enable_cap *cap)
6175 {
6176 	int r;
6177 
6178 	if (cap->flags)
6179 		return -EINVAL;
6180 
6181 	switch (cap->cap) {
6182 	case KVM_CAP_DISABLE_QUIRKS2:
6183 		r = -EINVAL;
6184 		if (cap->args[0] & ~KVM_X86_VALID_QUIRKS)
6185 			break;
6186 		fallthrough;
6187 	case KVM_CAP_DISABLE_QUIRKS:
6188 		kvm->arch.disabled_quirks = cap->args[0];
6189 		r = 0;
6190 		break;
6191 	case KVM_CAP_SPLIT_IRQCHIP: {
6192 		mutex_lock(&kvm->lock);
6193 		r = -EINVAL;
6194 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
6195 			goto split_irqchip_unlock;
6196 		r = -EEXIST;
6197 		if (irqchip_in_kernel(kvm))
6198 			goto split_irqchip_unlock;
6199 		if (kvm->created_vcpus)
6200 			goto split_irqchip_unlock;
6201 		r = kvm_setup_empty_irq_routing(kvm);
6202 		if (r)
6203 			goto split_irqchip_unlock;
6204 		/* Pairs with irqchip_in_kernel. */
6205 		smp_wmb();
6206 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
6207 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
6208 		kvm_clear_apicv_inhibit(kvm, APICV_INHIBIT_REASON_ABSENT);
6209 		r = 0;
6210 split_irqchip_unlock:
6211 		mutex_unlock(&kvm->lock);
6212 		break;
6213 	}
6214 	case KVM_CAP_X2APIC_API:
6215 		r = -EINVAL;
6216 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
6217 			break;
6218 
6219 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
6220 			kvm->arch.x2apic_format = true;
6221 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
6222 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
6223 
6224 		r = 0;
6225 		break;
6226 	case KVM_CAP_X86_DISABLE_EXITS:
6227 		r = -EINVAL;
6228 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
6229 			break;
6230 
6231 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
6232 			kvm_can_mwait_in_guest())
6233 			kvm->arch.mwait_in_guest = true;
6234 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
6235 			kvm->arch.hlt_in_guest = true;
6236 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
6237 			kvm->arch.pause_in_guest = true;
6238 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
6239 			kvm->arch.cstate_in_guest = true;
6240 		r = 0;
6241 		break;
6242 	case KVM_CAP_MSR_PLATFORM_INFO:
6243 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
6244 		r = 0;
6245 		break;
6246 	case KVM_CAP_EXCEPTION_PAYLOAD:
6247 		kvm->arch.exception_payload_enabled = cap->args[0];
6248 		r = 0;
6249 		break;
6250 	case KVM_CAP_X86_TRIPLE_FAULT_EVENT:
6251 		kvm->arch.triple_fault_event = cap->args[0];
6252 		r = 0;
6253 		break;
6254 	case KVM_CAP_X86_USER_SPACE_MSR:
6255 		r = -EINVAL;
6256 		if (cap->args[0] & ~KVM_MSR_EXIT_REASON_VALID_MASK)
6257 			break;
6258 		kvm->arch.user_space_msr_mask = cap->args[0];
6259 		r = 0;
6260 		break;
6261 	case KVM_CAP_X86_BUS_LOCK_EXIT:
6262 		r = -EINVAL;
6263 		if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
6264 			break;
6265 
6266 		if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
6267 		    (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
6268 			break;
6269 
6270 		if (kvm_caps.has_bus_lock_exit &&
6271 		    cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
6272 			kvm->arch.bus_lock_detection_enabled = true;
6273 		r = 0;
6274 		break;
6275 #ifdef CONFIG_X86_SGX_KVM
6276 	case KVM_CAP_SGX_ATTRIBUTE: {
6277 		unsigned long allowed_attributes = 0;
6278 
6279 		r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
6280 		if (r)
6281 			break;
6282 
6283 		/* KVM only supports the PROVISIONKEY privileged attribute. */
6284 		if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
6285 		    !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
6286 			kvm->arch.sgx_provisioning_allowed = true;
6287 		else
6288 			r = -EINVAL;
6289 		break;
6290 	}
6291 #endif
6292 	case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
6293 		r = -EINVAL;
6294 		if (!kvm_x86_ops.vm_copy_enc_context_from)
6295 			break;
6296 
6297 		r = static_call(kvm_x86_vm_copy_enc_context_from)(kvm, cap->args[0]);
6298 		break;
6299 	case KVM_CAP_VM_MOVE_ENC_CONTEXT_FROM:
6300 		r = -EINVAL;
6301 		if (!kvm_x86_ops.vm_move_enc_context_from)
6302 			break;
6303 
6304 		r = static_call(kvm_x86_vm_move_enc_context_from)(kvm, cap->args[0]);
6305 		break;
6306 	case KVM_CAP_EXIT_HYPERCALL:
6307 		if (cap->args[0] & ~KVM_EXIT_HYPERCALL_VALID_MASK) {
6308 			r = -EINVAL;
6309 			break;
6310 		}
6311 		kvm->arch.hypercall_exit_enabled = cap->args[0];
6312 		r = 0;
6313 		break;
6314 	case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
6315 		r = -EINVAL;
6316 		if (cap->args[0] & ~1)
6317 			break;
6318 		kvm->arch.exit_on_emulation_error = cap->args[0];
6319 		r = 0;
6320 		break;
6321 	case KVM_CAP_PMU_CAPABILITY:
6322 		r = -EINVAL;
6323 		if (!enable_pmu || (cap->args[0] & ~KVM_CAP_PMU_VALID_MASK))
6324 			break;
6325 
6326 		mutex_lock(&kvm->lock);
6327 		if (!kvm->created_vcpus) {
6328 			kvm->arch.enable_pmu = !(cap->args[0] & KVM_PMU_CAP_DISABLE);
6329 			r = 0;
6330 		}
6331 		mutex_unlock(&kvm->lock);
6332 		break;
6333 	case KVM_CAP_MAX_VCPU_ID:
6334 		r = -EINVAL;
6335 		if (cap->args[0] > KVM_MAX_VCPU_IDS)
6336 			break;
6337 
6338 		mutex_lock(&kvm->lock);
6339 		if (kvm->arch.max_vcpu_ids == cap->args[0]) {
6340 			r = 0;
6341 		} else if (!kvm->arch.max_vcpu_ids) {
6342 			kvm->arch.max_vcpu_ids = cap->args[0];
6343 			r = 0;
6344 		}
6345 		mutex_unlock(&kvm->lock);
6346 		break;
6347 	case KVM_CAP_X86_NOTIFY_VMEXIT:
6348 		r = -EINVAL;
6349 		if ((u32)cap->args[0] & ~KVM_X86_NOTIFY_VMEXIT_VALID_BITS)
6350 			break;
6351 		if (!kvm_caps.has_notify_vmexit)
6352 			break;
6353 		if (!((u32)cap->args[0] & KVM_X86_NOTIFY_VMEXIT_ENABLED))
6354 			break;
6355 		mutex_lock(&kvm->lock);
6356 		if (!kvm->created_vcpus) {
6357 			kvm->arch.notify_window = cap->args[0] >> 32;
6358 			kvm->arch.notify_vmexit_flags = (u32)cap->args[0];
6359 			r = 0;
6360 		}
6361 		mutex_unlock(&kvm->lock);
6362 		break;
6363 	case KVM_CAP_VM_DISABLE_NX_HUGE_PAGES:
6364 		r = -EINVAL;
6365 
6366 		/*
6367 		 * Since the risk of disabling NX hugepages is a guest crashing
6368 		 * the system, ensure the userspace process has permission to
6369 		 * reboot the system.
6370 		 *
6371 		 * Note that unlike the reboot() syscall, the process must have
6372 		 * this capability in the root namespace because exposing
6373 		 * /dev/kvm into a container does not limit the scope of the
6374 		 * iTLB multihit bug to that container. In other words,
6375 		 * this must use capable(), not ns_capable().
6376 		 */
6377 		if (!capable(CAP_SYS_BOOT)) {
6378 			r = -EPERM;
6379 			break;
6380 		}
6381 
6382 		if (cap->args[0])
6383 			break;
6384 
6385 		mutex_lock(&kvm->lock);
6386 		if (!kvm->created_vcpus) {
6387 			kvm->arch.disable_nx_huge_pages = true;
6388 			r = 0;
6389 		}
6390 		mutex_unlock(&kvm->lock);
6391 		break;
6392 	default:
6393 		r = -EINVAL;
6394 		break;
6395 	}
6396 	return r;
6397 }
6398 
6399 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
6400 {
6401 	struct kvm_x86_msr_filter *msr_filter;
6402 
6403 	msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
6404 	if (!msr_filter)
6405 		return NULL;
6406 
6407 	msr_filter->default_allow = default_allow;
6408 	return msr_filter;
6409 }
6410 
6411 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
6412 {
6413 	u32 i;
6414 
6415 	if (!msr_filter)
6416 		return;
6417 
6418 	for (i = 0; i < msr_filter->count; i++)
6419 		kfree(msr_filter->ranges[i].bitmap);
6420 
6421 	kfree(msr_filter);
6422 }
6423 
6424 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
6425 			      struct kvm_msr_filter_range *user_range)
6426 {
6427 	unsigned long *bitmap = NULL;
6428 	size_t bitmap_size;
6429 
6430 	if (!user_range->nmsrs)
6431 		return 0;
6432 
6433 	if (user_range->flags & ~KVM_MSR_FILTER_RANGE_VALID_MASK)
6434 		return -EINVAL;
6435 
6436 	if (!user_range->flags)
6437 		return -EINVAL;
6438 
6439 	bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
6440 	if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
6441 		return -EINVAL;
6442 
6443 	bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
6444 	if (IS_ERR(bitmap))
6445 		return PTR_ERR(bitmap);
6446 
6447 	msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
6448 		.flags = user_range->flags,
6449 		.base = user_range->base,
6450 		.nmsrs = user_range->nmsrs,
6451 		.bitmap = bitmap,
6452 	};
6453 
6454 	msr_filter->count++;
6455 	return 0;
6456 }
6457 
6458 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm,
6459 				       struct kvm_msr_filter *filter)
6460 {
6461 	struct kvm_x86_msr_filter *new_filter, *old_filter;
6462 	bool default_allow;
6463 	bool empty = true;
6464 	int r = 0;
6465 	u32 i;
6466 
6467 	if (filter->flags & ~KVM_MSR_FILTER_VALID_MASK)
6468 		return -EINVAL;
6469 
6470 	for (i = 0; i < ARRAY_SIZE(filter->ranges); i++)
6471 		empty &= !filter->ranges[i].nmsrs;
6472 
6473 	default_allow = !(filter->flags & KVM_MSR_FILTER_DEFAULT_DENY);
6474 	if (empty && !default_allow)
6475 		return -EINVAL;
6476 
6477 	new_filter = kvm_alloc_msr_filter(default_allow);
6478 	if (!new_filter)
6479 		return -ENOMEM;
6480 
6481 	for (i = 0; i < ARRAY_SIZE(filter->ranges); i++) {
6482 		r = kvm_add_msr_filter(new_filter, &filter->ranges[i]);
6483 		if (r) {
6484 			kvm_free_msr_filter(new_filter);
6485 			return r;
6486 		}
6487 	}
6488 
6489 	mutex_lock(&kvm->lock);
6490 
6491 	/* The per-VM filter is protected by kvm->lock... */
6492 	old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
6493 
6494 	rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
6495 	synchronize_srcu(&kvm->srcu);
6496 
6497 	kvm_free_msr_filter(old_filter);
6498 
6499 	kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
6500 	mutex_unlock(&kvm->lock);
6501 
6502 	return 0;
6503 }
6504 
6505 #ifdef CONFIG_KVM_COMPAT
6506 /* for KVM_X86_SET_MSR_FILTER */
6507 struct kvm_msr_filter_range_compat {
6508 	__u32 flags;
6509 	__u32 nmsrs;
6510 	__u32 base;
6511 	__u32 bitmap;
6512 };
6513 
6514 struct kvm_msr_filter_compat {
6515 	__u32 flags;
6516 	struct kvm_msr_filter_range_compat ranges[KVM_MSR_FILTER_MAX_RANGES];
6517 };
6518 
6519 #define KVM_X86_SET_MSR_FILTER_COMPAT _IOW(KVMIO, 0xc6, struct kvm_msr_filter_compat)
6520 
6521 long kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl,
6522 			      unsigned long arg)
6523 {
6524 	void __user *argp = (void __user *)arg;
6525 	struct kvm *kvm = filp->private_data;
6526 	long r = -ENOTTY;
6527 
6528 	switch (ioctl) {
6529 	case KVM_X86_SET_MSR_FILTER_COMPAT: {
6530 		struct kvm_msr_filter __user *user_msr_filter = argp;
6531 		struct kvm_msr_filter_compat filter_compat;
6532 		struct kvm_msr_filter filter;
6533 		int i;
6534 
6535 		if (copy_from_user(&filter_compat, user_msr_filter,
6536 				   sizeof(filter_compat)))
6537 			return -EFAULT;
6538 
6539 		filter.flags = filter_compat.flags;
6540 		for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
6541 			struct kvm_msr_filter_range_compat *cr;
6542 
6543 			cr = &filter_compat.ranges[i];
6544 			filter.ranges[i] = (struct kvm_msr_filter_range) {
6545 				.flags = cr->flags,
6546 				.nmsrs = cr->nmsrs,
6547 				.base = cr->base,
6548 				.bitmap = (__u8 *)(ulong)cr->bitmap,
6549 			};
6550 		}
6551 
6552 		r = kvm_vm_ioctl_set_msr_filter(kvm, &filter);
6553 		break;
6554 	}
6555 	}
6556 
6557 	return r;
6558 }
6559 #endif
6560 
6561 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
6562 static int kvm_arch_suspend_notifier(struct kvm *kvm)
6563 {
6564 	struct kvm_vcpu *vcpu;
6565 	unsigned long i;
6566 	int ret = 0;
6567 
6568 	mutex_lock(&kvm->lock);
6569 	kvm_for_each_vcpu(i, vcpu, kvm) {
6570 		if (!vcpu->arch.pv_time.active)
6571 			continue;
6572 
6573 		ret = kvm_set_guest_paused(vcpu);
6574 		if (ret) {
6575 			kvm_err("Failed to pause guest VCPU%d: %d\n",
6576 				vcpu->vcpu_id, ret);
6577 			break;
6578 		}
6579 	}
6580 	mutex_unlock(&kvm->lock);
6581 
6582 	return ret ? NOTIFY_BAD : NOTIFY_DONE;
6583 }
6584 
6585 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state)
6586 {
6587 	switch (state) {
6588 	case PM_HIBERNATION_PREPARE:
6589 	case PM_SUSPEND_PREPARE:
6590 		return kvm_arch_suspend_notifier(kvm);
6591 	}
6592 
6593 	return NOTIFY_DONE;
6594 }
6595 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
6596 
6597 static int kvm_vm_ioctl_get_clock(struct kvm *kvm, void __user *argp)
6598 {
6599 	struct kvm_clock_data data = { 0 };
6600 
6601 	get_kvmclock(kvm, &data);
6602 	if (copy_to_user(argp, &data, sizeof(data)))
6603 		return -EFAULT;
6604 
6605 	return 0;
6606 }
6607 
6608 static int kvm_vm_ioctl_set_clock(struct kvm *kvm, void __user *argp)
6609 {
6610 	struct kvm_arch *ka = &kvm->arch;
6611 	struct kvm_clock_data data;
6612 	u64 now_raw_ns;
6613 
6614 	if (copy_from_user(&data, argp, sizeof(data)))
6615 		return -EFAULT;
6616 
6617 	/*
6618 	 * Only KVM_CLOCK_REALTIME is used, but allow passing the
6619 	 * result of KVM_GET_CLOCK back to KVM_SET_CLOCK.
6620 	 */
6621 	if (data.flags & ~KVM_CLOCK_VALID_FLAGS)
6622 		return -EINVAL;
6623 
6624 	kvm_hv_request_tsc_page_update(kvm);
6625 	kvm_start_pvclock_update(kvm);
6626 	pvclock_update_vm_gtod_copy(kvm);
6627 
6628 	/*
6629 	 * This pairs with kvm_guest_time_update(): when masterclock is
6630 	 * in use, we use master_kernel_ns + kvmclock_offset to set
6631 	 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
6632 	 * is slightly ahead) here we risk going negative on unsigned
6633 	 * 'system_time' when 'data.clock' is very small.
6634 	 */
6635 	if (data.flags & KVM_CLOCK_REALTIME) {
6636 		u64 now_real_ns = ktime_get_real_ns();
6637 
6638 		/*
6639 		 * Avoid stepping the kvmclock backwards.
6640 		 */
6641 		if (now_real_ns > data.realtime)
6642 			data.clock += now_real_ns - data.realtime;
6643 	}
6644 
6645 	if (ka->use_master_clock)
6646 		now_raw_ns = ka->master_kernel_ns;
6647 	else
6648 		now_raw_ns = get_kvmclock_base_ns();
6649 	ka->kvmclock_offset = data.clock - now_raw_ns;
6650 	kvm_end_pvclock_update(kvm);
6651 	return 0;
6652 }
6653 
6654 long kvm_arch_vm_ioctl(struct file *filp,
6655 		       unsigned int ioctl, unsigned long arg)
6656 {
6657 	struct kvm *kvm = filp->private_data;
6658 	void __user *argp = (void __user *)arg;
6659 	int r = -ENOTTY;
6660 	/*
6661 	 * This union makes it completely explicit to gcc-3.x
6662 	 * that these two variables' stack usage should be
6663 	 * combined, not added together.
6664 	 */
6665 	union {
6666 		struct kvm_pit_state ps;
6667 		struct kvm_pit_state2 ps2;
6668 		struct kvm_pit_config pit_config;
6669 	} u;
6670 
6671 	switch (ioctl) {
6672 	case KVM_SET_TSS_ADDR:
6673 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
6674 		break;
6675 	case KVM_SET_IDENTITY_MAP_ADDR: {
6676 		u64 ident_addr;
6677 
6678 		mutex_lock(&kvm->lock);
6679 		r = -EINVAL;
6680 		if (kvm->created_vcpus)
6681 			goto set_identity_unlock;
6682 		r = -EFAULT;
6683 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
6684 			goto set_identity_unlock;
6685 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
6686 set_identity_unlock:
6687 		mutex_unlock(&kvm->lock);
6688 		break;
6689 	}
6690 	case KVM_SET_NR_MMU_PAGES:
6691 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
6692 		break;
6693 	case KVM_GET_NR_MMU_PAGES:
6694 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
6695 		break;
6696 	case KVM_CREATE_IRQCHIP: {
6697 		mutex_lock(&kvm->lock);
6698 
6699 		r = -EEXIST;
6700 		if (irqchip_in_kernel(kvm))
6701 			goto create_irqchip_unlock;
6702 
6703 		r = -EINVAL;
6704 		if (kvm->created_vcpus)
6705 			goto create_irqchip_unlock;
6706 
6707 		r = kvm_pic_init(kvm);
6708 		if (r)
6709 			goto create_irqchip_unlock;
6710 
6711 		r = kvm_ioapic_init(kvm);
6712 		if (r) {
6713 			kvm_pic_destroy(kvm);
6714 			goto create_irqchip_unlock;
6715 		}
6716 
6717 		r = kvm_setup_default_irq_routing(kvm);
6718 		if (r) {
6719 			kvm_ioapic_destroy(kvm);
6720 			kvm_pic_destroy(kvm);
6721 			goto create_irqchip_unlock;
6722 		}
6723 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
6724 		smp_wmb();
6725 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
6726 		kvm_clear_apicv_inhibit(kvm, APICV_INHIBIT_REASON_ABSENT);
6727 	create_irqchip_unlock:
6728 		mutex_unlock(&kvm->lock);
6729 		break;
6730 	}
6731 	case KVM_CREATE_PIT:
6732 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
6733 		goto create_pit;
6734 	case KVM_CREATE_PIT2:
6735 		r = -EFAULT;
6736 		if (copy_from_user(&u.pit_config, argp,
6737 				   sizeof(struct kvm_pit_config)))
6738 			goto out;
6739 	create_pit:
6740 		mutex_lock(&kvm->lock);
6741 		r = -EEXIST;
6742 		if (kvm->arch.vpit)
6743 			goto create_pit_unlock;
6744 		r = -ENOMEM;
6745 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
6746 		if (kvm->arch.vpit)
6747 			r = 0;
6748 	create_pit_unlock:
6749 		mutex_unlock(&kvm->lock);
6750 		break;
6751 	case KVM_GET_IRQCHIP: {
6752 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6753 		struct kvm_irqchip *chip;
6754 
6755 		chip = memdup_user(argp, sizeof(*chip));
6756 		if (IS_ERR(chip)) {
6757 			r = PTR_ERR(chip);
6758 			goto out;
6759 		}
6760 
6761 		r = -ENXIO;
6762 		if (!irqchip_kernel(kvm))
6763 			goto get_irqchip_out;
6764 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
6765 		if (r)
6766 			goto get_irqchip_out;
6767 		r = -EFAULT;
6768 		if (copy_to_user(argp, chip, sizeof(*chip)))
6769 			goto get_irqchip_out;
6770 		r = 0;
6771 	get_irqchip_out:
6772 		kfree(chip);
6773 		break;
6774 	}
6775 	case KVM_SET_IRQCHIP: {
6776 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6777 		struct kvm_irqchip *chip;
6778 
6779 		chip = memdup_user(argp, sizeof(*chip));
6780 		if (IS_ERR(chip)) {
6781 			r = PTR_ERR(chip);
6782 			goto out;
6783 		}
6784 
6785 		r = -ENXIO;
6786 		if (!irqchip_kernel(kvm))
6787 			goto set_irqchip_out;
6788 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
6789 	set_irqchip_out:
6790 		kfree(chip);
6791 		break;
6792 	}
6793 	case KVM_GET_PIT: {
6794 		r = -EFAULT;
6795 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
6796 			goto out;
6797 		r = -ENXIO;
6798 		if (!kvm->arch.vpit)
6799 			goto out;
6800 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
6801 		if (r)
6802 			goto out;
6803 		r = -EFAULT;
6804 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
6805 			goto out;
6806 		r = 0;
6807 		break;
6808 	}
6809 	case KVM_SET_PIT: {
6810 		r = -EFAULT;
6811 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
6812 			goto out;
6813 		mutex_lock(&kvm->lock);
6814 		r = -ENXIO;
6815 		if (!kvm->arch.vpit)
6816 			goto set_pit_out;
6817 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
6818 set_pit_out:
6819 		mutex_unlock(&kvm->lock);
6820 		break;
6821 	}
6822 	case KVM_GET_PIT2: {
6823 		r = -ENXIO;
6824 		if (!kvm->arch.vpit)
6825 			goto out;
6826 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
6827 		if (r)
6828 			goto out;
6829 		r = -EFAULT;
6830 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
6831 			goto out;
6832 		r = 0;
6833 		break;
6834 	}
6835 	case KVM_SET_PIT2: {
6836 		r = -EFAULT;
6837 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
6838 			goto out;
6839 		mutex_lock(&kvm->lock);
6840 		r = -ENXIO;
6841 		if (!kvm->arch.vpit)
6842 			goto set_pit2_out;
6843 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
6844 set_pit2_out:
6845 		mutex_unlock(&kvm->lock);
6846 		break;
6847 	}
6848 	case KVM_REINJECT_CONTROL: {
6849 		struct kvm_reinject_control control;
6850 		r =  -EFAULT;
6851 		if (copy_from_user(&control, argp, sizeof(control)))
6852 			goto out;
6853 		r = -ENXIO;
6854 		if (!kvm->arch.vpit)
6855 			goto out;
6856 		r = kvm_vm_ioctl_reinject(kvm, &control);
6857 		break;
6858 	}
6859 	case KVM_SET_BOOT_CPU_ID:
6860 		r = 0;
6861 		mutex_lock(&kvm->lock);
6862 		if (kvm->created_vcpus)
6863 			r = -EBUSY;
6864 		else
6865 			kvm->arch.bsp_vcpu_id = arg;
6866 		mutex_unlock(&kvm->lock);
6867 		break;
6868 #ifdef CONFIG_KVM_XEN
6869 	case KVM_XEN_HVM_CONFIG: {
6870 		struct kvm_xen_hvm_config xhc;
6871 		r = -EFAULT;
6872 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
6873 			goto out;
6874 		r = kvm_xen_hvm_config(kvm, &xhc);
6875 		break;
6876 	}
6877 	case KVM_XEN_HVM_GET_ATTR: {
6878 		struct kvm_xen_hvm_attr xha;
6879 
6880 		r = -EFAULT;
6881 		if (copy_from_user(&xha, argp, sizeof(xha)))
6882 			goto out;
6883 		r = kvm_xen_hvm_get_attr(kvm, &xha);
6884 		if (!r && copy_to_user(argp, &xha, sizeof(xha)))
6885 			r = -EFAULT;
6886 		break;
6887 	}
6888 	case KVM_XEN_HVM_SET_ATTR: {
6889 		struct kvm_xen_hvm_attr xha;
6890 
6891 		r = -EFAULT;
6892 		if (copy_from_user(&xha, argp, sizeof(xha)))
6893 			goto out;
6894 		r = kvm_xen_hvm_set_attr(kvm, &xha);
6895 		break;
6896 	}
6897 	case KVM_XEN_HVM_EVTCHN_SEND: {
6898 		struct kvm_irq_routing_xen_evtchn uxe;
6899 
6900 		r = -EFAULT;
6901 		if (copy_from_user(&uxe, argp, sizeof(uxe)))
6902 			goto out;
6903 		r = kvm_xen_hvm_evtchn_send(kvm, &uxe);
6904 		break;
6905 	}
6906 #endif
6907 	case KVM_SET_CLOCK:
6908 		r = kvm_vm_ioctl_set_clock(kvm, argp);
6909 		break;
6910 	case KVM_GET_CLOCK:
6911 		r = kvm_vm_ioctl_get_clock(kvm, argp);
6912 		break;
6913 	case KVM_SET_TSC_KHZ: {
6914 		u32 user_tsc_khz;
6915 
6916 		r = -EINVAL;
6917 		user_tsc_khz = (u32)arg;
6918 
6919 		if (kvm_caps.has_tsc_control &&
6920 		    user_tsc_khz >= kvm_caps.max_guest_tsc_khz)
6921 			goto out;
6922 
6923 		if (user_tsc_khz == 0)
6924 			user_tsc_khz = tsc_khz;
6925 
6926 		WRITE_ONCE(kvm->arch.default_tsc_khz, user_tsc_khz);
6927 		r = 0;
6928 
6929 		goto out;
6930 	}
6931 	case KVM_GET_TSC_KHZ: {
6932 		r = READ_ONCE(kvm->arch.default_tsc_khz);
6933 		goto out;
6934 	}
6935 	case KVM_MEMORY_ENCRYPT_OP: {
6936 		r = -ENOTTY;
6937 		if (!kvm_x86_ops.mem_enc_ioctl)
6938 			goto out;
6939 
6940 		r = static_call(kvm_x86_mem_enc_ioctl)(kvm, argp);
6941 		break;
6942 	}
6943 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
6944 		struct kvm_enc_region region;
6945 
6946 		r = -EFAULT;
6947 		if (copy_from_user(&region, argp, sizeof(region)))
6948 			goto out;
6949 
6950 		r = -ENOTTY;
6951 		if (!kvm_x86_ops.mem_enc_register_region)
6952 			goto out;
6953 
6954 		r = static_call(kvm_x86_mem_enc_register_region)(kvm, &region);
6955 		break;
6956 	}
6957 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
6958 		struct kvm_enc_region region;
6959 
6960 		r = -EFAULT;
6961 		if (copy_from_user(&region, argp, sizeof(region)))
6962 			goto out;
6963 
6964 		r = -ENOTTY;
6965 		if (!kvm_x86_ops.mem_enc_unregister_region)
6966 			goto out;
6967 
6968 		r = static_call(kvm_x86_mem_enc_unregister_region)(kvm, &region);
6969 		break;
6970 	}
6971 	case KVM_HYPERV_EVENTFD: {
6972 		struct kvm_hyperv_eventfd hvevfd;
6973 
6974 		r = -EFAULT;
6975 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
6976 			goto out;
6977 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
6978 		break;
6979 	}
6980 	case KVM_SET_PMU_EVENT_FILTER:
6981 		r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
6982 		break;
6983 	case KVM_X86_SET_MSR_FILTER: {
6984 		struct kvm_msr_filter __user *user_msr_filter = argp;
6985 		struct kvm_msr_filter filter;
6986 
6987 		if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
6988 			return -EFAULT;
6989 
6990 		r = kvm_vm_ioctl_set_msr_filter(kvm, &filter);
6991 		break;
6992 	}
6993 	default:
6994 		r = -ENOTTY;
6995 	}
6996 out:
6997 	return r;
6998 }
6999 
7000 static void kvm_probe_msr_to_save(u32 msr_index)
7001 {
7002 	u32 dummy[2];
7003 
7004 	if (rdmsr_safe(msr_index, &dummy[0], &dummy[1]))
7005 		return;
7006 
7007 	/*
7008 	 * Even MSRs that are valid in the host may not be exposed to guests in
7009 	 * some cases.
7010 	 */
7011 	switch (msr_index) {
7012 	case MSR_IA32_BNDCFGS:
7013 		if (!kvm_mpx_supported())
7014 			return;
7015 		break;
7016 	case MSR_TSC_AUX:
7017 		if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
7018 		    !kvm_cpu_cap_has(X86_FEATURE_RDPID))
7019 			return;
7020 		break;
7021 	case MSR_IA32_UMWAIT_CONTROL:
7022 		if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
7023 			return;
7024 		break;
7025 	case MSR_IA32_RTIT_CTL:
7026 	case MSR_IA32_RTIT_STATUS:
7027 		if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
7028 			return;
7029 		break;
7030 	case MSR_IA32_RTIT_CR3_MATCH:
7031 		if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
7032 		    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
7033 			return;
7034 		break;
7035 	case MSR_IA32_RTIT_OUTPUT_BASE:
7036 	case MSR_IA32_RTIT_OUTPUT_MASK:
7037 		if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
7038 		    (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
7039 		     !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
7040 			return;
7041 		break;
7042 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
7043 		if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
7044 		    (msr_index - MSR_IA32_RTIT_ADDR0_A >=
7045 		     intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2))
7046 			return;
7047 		break;
7048 	case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR_MAX:
7049 		if (msr_index - MSR_ARCH_PERFMON_PERFCTR0 >=
7050 		    kvm_pmu_cap.num_counters_gp)
7051 			return;
7052 		break;
7053 	case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL_MAX:
7054 		if (msr_index - MSR_ARCH_PERFMON_EVENTSEL0 >=
7055 		    kvm_pmu_cap.num_counters_gp)
7056 			return;
7057 		break;
7058 	case MSR_IA32_XFD:
7059 	case MSR_IA32_XFD_ERR:
7060 		if (!kvm_cpu_cap_has(X86_FEATURE_XFD))
7061 			return;
7062 		break;
7063 	default:
7064 		break;
7065 	}
7066 
7067 	msrs_to_save[num_msrs_to_save++] = msr_index;
7068 }
7069 
7070 static void kvm_init_msr_list(void)
7071 {
7072 	unsigned i;
7073 
7074 	BUILD_BUG_ON_MSG(KVM_PMC_MAX_FIXED != 3,
7075 			 "Please update the fixed PMCs in msrs_to_save_pmu[]");
7076 
7077 	num_msrs_to_save = 0;
7078 	num_emulated_msrs = 0;
7079 	num_msr_based_features = 0;
7080 
7081 	for (i = 0; i < ARRAY_SIZE(msrs_to_save_base); i++)
7082 		kvm_probe_msr_to_save(msrs_to_save_base[i]);
7083 
7084 	if (enable_pmu) {
7085 		for (i = 0; i < ARRAY_SIZE(msrs_to_save_pmu); i++)
7086 			kvm_probe_msr_to_save(msrs_to_save_pmu[i]);
7087 	}
7088 
7089 	for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
7090 		if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
7091 			continue;
7092 
7093 		emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
7094 	}
7095 
7096 	for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
7097 		struct kvm_msr_entry msr;
7098 
7099 		msr.index = msr_based_features_all[i];
7100 		if (kvm_get_msr_feature(&msr))
7101 			continue;
7102 
7103 		msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
7104 	}
7105 }
7106 
7107 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
7108 			   const void *v)
7109 {
7110 	int handled = 0;
7111 	int n;
7112 
7113 	do {
7114 		n = min(len, 8);
7115 		if (!(lapic_in_kernel(vcpu) &&
7116 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
7117 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
7118 			break;
7119 		handled += n;
7120 		addr += n;
7121 		len -= n;
7122 		v += n;
7123 	} while (len);
7124 
7125 	return handled;
7126 }
7127 
7128 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
7129 {
7130 	int handled = 0;
7131 	int n;
7132 
7133 	do {
7134 		n = min(len, 8);
7135 		if (!(lapic_in_kernel(vcpu) &&
7136 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
7137 					 addr, n, v))
7138 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
7139 			break;
7140 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
7141 		handled += n;
7142 		addr += n;
7143 		len -= n;
7144 		v += n;
7145 	} while (len);
7146 
7147 	return handled;
7148 }
7149 
7150 void kvm_set_segment(struct kvm_vcpu *vcpu,
7151 		     struct kvm_segment *var, int seg)
7152 {
7153 	static_call(kvm_x86_set_segment)(vcpu, var, seg);
7154 }
7155 
7156 void kvm_get_segment(struct kvm_vcpu *vcpu,
7157 		     struct kvm_segment *var, int seg)
7158 {
7159 	static_call(kvm_x86_get_segment)(vcpu, var, seg);
7160 }
7161 
7162 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u64 access,
7163 			   struct x86_exception *exception)
7164 {
7165 	struct kvm_mmu *mmu = vcpu->arch.mmu;
7166 	gpa_t t_gpa;
7167 
7168 	BUG_ON(!mmu_is_nested(vcpu));
7169 
7170 	/* NPT walks are always user-walks */
7171 	access |= PFERR_USER_MASK;
7172 	t_gpa  = mmu->gva_to_gpa(vcpu, mmu, gpa, access, exception);
7173 
7174 	return t_gpa;
7175 }
7176 
7177 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
7178 			      struct x86_exception *exception)
7179 {
7180 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7181 
7182 	u64 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
7183 	return mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
7184 }
7185 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
7186 
7187 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
7188 			       struct x86_exception *exception)
7189 {
7190 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7191 
7192 	u64 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
7193 	access |= PFERR_WRITE_MASK;
7194 	return mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
7195 }
7196 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
7197 
7198 /* uses this to access any guest's mapped memory without checking CPL */
7199 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
7200 				struct x86_exception *exception)
7201 {
7202 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7203 
7204 	return mmu->gva_to_gpa(vcpu, mmu, gva, 0, exception);
7205 }
7206 
7207 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
7208 				      struct kvm_vcpu *vcpu, u64 access,
7209 				      struct x86_exception *exception)
7210 {
7211 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7212 	void *data = val;
7213 	int r = X86EMUL_CONTINUE;
7214 
7215 	while (bytes) {
7216 		gpa_t gpa = mmu->gva_to_gpa(vcpu, mmu, addr, access, exception);
7217 		unsigned offset = addr & (PAGE_SIZE-1);
7218 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
7219 		int ret;
7220 
7221 		if (gpa == INVALID_GPA)
7222 			return X86EMUL_PROPAGATE_FAULT;
7223 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
7224 					       offset, toread);
7225 		if (ret < 0) {
7226 			r = X86EMUL_IO_NEEDED;
7227 			goto out;
7228 		}
7229 
7230 		bytes -= toread;
7231 		data += toread;
7232 		addr += toread;
7233 	}
7234 out:
7235 	return r;
7236 }
7237 
7238 /* used for instruction fetching */
7239 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
7240 				gva_t addr, void *val, unsigned int bytes,
7241 				struct x86_exception *exception)
7242 {
7243 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7244 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7245 	u64 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
7246 	unsigned offset;
7247 	int ret;
7248 
7249 	/* Inline kvm_read_guest_virt_helper for speed.  */
7250 	gpa_t gpa = mmu->gva_to_gpa(vcpu, mmu, addr, access|PFERR_FETCH_MASK,
7251 				    exception);
7252 	if (unlikely(gpa == INVALID_GPA))
7253 		return X86EMUL_PROPAGATE_FAULT;
7254 
7255 	offset = addr & (PAGE_SIZE-1);
7256 	if (WARN_ON(offset + bytes > PAGE_SIZE))
7257 		bytes = (unsigned)PAGE_SIZE - offset;
7258 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
7259 				       offset, bytes);
7260 	if (unlikely(ret < 0))
7261 		return X86EMUL_IO_NEEDED;
7262 
7263 	return X86EMUL_CONTINUE;
7264 }
7265 
7266 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
7267 			       gva_t addr, void *val, unsigned int bytes,
7268 			       struct x86_exception *exception)
7269 {
7270 	u64 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
7271 
7272 	/*
7273 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
7274 	 * is returned, but our callers are not ready for that and they blindly
7275 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
7276 	 * uninitialized kernel stack memory into cr2 and error code.
7277 	 */
7278 	memset(exception, 0, sizeof(*exception));
7279 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
7280 					  exception);
7281 }
7282 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
7283 
7284 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
7285 			     gva_t addr, void *val, unsigned int bytes,
7286 			     struct x86_exception *exception, bool system)
7287 {
7288 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7289 	u64 access = 0;
7290 
7291 	if (system)
7292 		access |= PFERR_IMPLICIT_ACCESS;
7293 	else if (static_call(kvm_x86_get_cpl)(vcpu) == 3)
7294 		access |= PFERR_USER_MASK;
7295 
7296 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
7297 }
7298 
7299 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
7300 				      struct kvm_vcpu *vcpu, u64 access,
7301 				      struct x86_exception *exception)
7302 {
7303 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7304 	void *data = val;
7305 	int r = X86EMUL_CONTINUE;
7306 
7307 	while (bytes) {
7308 		gpa_t gpa = mmu->gva_to_gpa(vcpu, mmu, addr, access, exception);
7309 		unsigned offset = addr & (PAGE_SIZE-1);
7310 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
7311 		int ret;
7312 
7313 		if (gpa == INVALID_GPA)
7314 			return X86EMUL_PROPAGATE_FAULT;
7315 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
7316 		if (ret < 0) {
7317 			r = X86EMUL_IO_NEEDED;
7318 			goto out;
7319 		}
7320 
7321 		bytes -= towrite;
7322 		data += towrite;
7323 		addr += towrite;
7324 	}
7325 out:
7326 	return r;
7327 }
7328 
7329 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
7330 			      unsigned int bytes, struct x86_exception *exception,
7331 			      bool system)
7332 {
7333 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7334 	u64 access = PFERR_WRITE_MASK;
7335 
7336 	if (system)
7337 		access |= PFERR_IMPLICIT_ACCESS;
7338 	else if (static_call(kvm_x86_get_cpl)(vcpu) == 3)
7339 		access |= PFERR_USER_MASK;
7340 
7341 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
7342 					   access, exception);
7343 }
7344 
7345 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
7346 				unsigned int bytes, struct x86_exception *exception)
7347 {
7348 	/* kvm_write_guest_virt_system can pull in tons of pages. */
7349 	vcpu->arch.l1tf_flush_l1d = true;
7350 
7351 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
7352 					   PFERR_WRITE_MASK, exception);
7353 }
7354 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
7355 
7356 static int kvm_can_emulate_insn(struct kvm_vcpu *vcpu, int emul_type,
7357 				void *insn, int insn_len)
7358 {
7359 	return static_call(kvm_x86_can_emulate_instruction)(vcpu, emul_type,
7360 							    insn, insn_len);
7361 }
7362 
7363 int handle_ud(struct kvm_vcpu *vcpu)
7364 {
7365 	static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
7366 	int fep_flags = READ_ONCE(force_emulation_prefix);
7367 	int emul_type = EMULTYPE_TRAP_UD;
7368 	char sig[5]; /* ud2; .ascii "kvm" */
7369 	struct x86_exception e;
7370 
7371 	if (unlikely(!kvm_can_emulate_insn(vcpu, emul_type, NULL, 0)))
7372 		return 1;
7373 
7374 	if (fep_flags &&
7375 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
7376 				sig, sizeof(sig), &e) == 0 &&
7377 	    memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
7378 		if (fep_flags & KVM_FEP_CLEAR_RFLAGS_RF)
7379 			kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) & ~X86_EFLAGS_RF);
7380 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
7381 		emul_type = EMULTYPE_TRAP_UD_FORCED;
7382 	}
7383 
7384 	return kvm_emulate_instruction(vcpu, emul_type);
7385 }
7386 EXPORT_SYMBOL_GPL(handle_ud);
7387 
7388 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
7389 			    gpa_t gpa, bool write)
7390 {
7391 	/* For APIC access vmexit */
7392 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
7393 		return 1;
7394 
7395 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
7396 		trace_vcpu_match_mmio(gva, gpa, write, true);
7397 		return 1;
7398 	}
7399 
7400 	return 0;
7401 }
7402 
7403 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
7404 				gpa_t *gpa, struct x86_exception *exception,
7405 				bool write)
7406 {
7407 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7408 	u64 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
7409 		| (write ? PFERR_WRITE_MASK : 0);
7410 
7411 	/*
7412 	 * currently PKRU is only applied to ept enabled guest so
7413 	 * there is no pkey in EPT page table for L1 guest or EPT
7414 	 * shadow page table for L2 guest.
7415 	 */
7416 	if (vcpu_match_mmio_gva(vcpu, gva) && (!is_paging(vcpu) ||
7417 	    !permission_fault(vcpu, vcpu->arch.walk_mmu,
7418 			      vcpu->arch.mmio_access, 0, access))) {
7419 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
7420 					(gva & (PAGE_SIZE - 1));
7421 		trace_vcpu_match_mmio(gva, *gpa, write, false);
7422 		return 1;
7423 	}
7424 
7425 	*gpa = mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
7426 
7427 	if (*gpa == INVALID_GPA)
7428 		return -1;
7429 
7430 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
7431 }
7432 
7433 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
7434 			const void *val, int bytes)
7435 {
7436 	int ret;
7437 
7438 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
7439 	if (ret < 0)
7440 		return 0;
7441 	kvm_page_track_write(vcpu, gpa, val, bytes);
7442 	return 1;
7443 }
7444 
7445 struct read_write_emulator_ops {
7446 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
7447 				  int bytes);
7448 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
7449 				  void *val, int bytes);
7450 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
7451 			       int bytes, void *val);
7452 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
7453 				    void *val, int bytes);
7454 	bool write;
7455 };
7456 
7457 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
7458 {
7459 	if (vcpu->mmio_read_completed) {
7460 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
7461 			       vcpu->mmio_fragments[0].gpa, val);
7462 		vcpu->mmio_read_completed = 0;
7463 		return 1;
7464 	}
7465 
7466 	return 0;
7467 }
7468 
7469 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
7470 			void *val, int bytes)
7471 {
7472 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
7473 }
7474 
7475 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
7476 			 void *val, int bytes)
7477 {
7478 	return emulator_write_phys(vcpu, gpa, val, bytes);
7479 }
7480 
7481 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
7482 {
7483 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
7484 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
7485 }
7486 
7487 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
7488 			  void *val, int bytes)
7489 {
7490 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
7491 	return X86EMUL_IO_NEEDED;
7492 }
7493 
7494 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
7495 			   void *val, int bytes)
7496 {
7497 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
7498 
7499 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
7500 	return X86EMUL_CONTINUE;
7501 }
7502 
7503 static const struct read_write_emulator_ops read_emultor = {
7504 	.read_write_prepare = read_prepare,
7505 	.read_write_emulate = read_emulate,
7506 	.read_write_mmio = vcpu_mmio_read,
7507 	.read_write_exit_mmio = read_exit_mmio,
7508 };
7509 
7510 static const struct read_write_emulator_ops write_emultor = {
7511 	.read_write_emulate = write_emulate,
7512 	.read_write_mmio = write_mmio,
7513 	.read_write_exit_mmio = write_exit_mmio,
7514 	.write = true,
7515 };
7516 
7517 static int emulator_read_write_onepage(unsigned long addr, void *val,
7518 				       unsigned int bytes,
7519 				       struct x86_exception *exception,
7520 				       struct kvm_vcpu *vcpu,
7521 				       const struct read_write_emulator_ops *ops)
7522 {
7523 	gpa_t gpa;
7524 	int handled, ret;
7525 	bool write = ops->write;
7526 	struct kvm_mmio_fragment *frag;
7527 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7528 
7529 	/*
7530 	 * If the exit was due to a NPF we may already have a GPA.
7531 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
7532 	 * Note, this cannot be used on string operations since string
7533 	 * operation using rep will only have the initial GPA from the NPF
7534 	 * occurred.
7535 	 */
7536 	if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
7537 	    (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
7538 		gpa = ctxt->gpa_val;
7539 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
7540 	} else {
7541 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
7542 		if (ret < 0)
7543 			return X86EMUL_PROPAGATE_FAULT;
7544 	}
7545 
7546 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
7547 		return X86EMUL_CONTINUE;
7548 
7549 	/*
7550 	 * Is this MMIO handled locally?
7551 	 */
7552 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
7553 	if (handled == bytes)
7554 		return X86EMUL_CONTINUE;
7555 
7556 	gpa += handled;
7557 	bytes -= handled;
7558 	val += handled;
7559 
7560 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
7561 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
7562 	frag->gpa = gpa;
7563 	frag->data = val;
7564 	frag->len = bytes;
7565 	return X86EMUL_CONTINUE;
7566 }
7567 
7568 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
7569 			unsigned long addr,
7570 			void *val, unsigned int bytes,
7571 			struct x86_exception *exception,
7572 			const struct read_write_emulator_ops *ops)
7573 {
7574 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7575 	gpa_t gpa;
7576 	int rc;
7577 
7578 	if (ops->read_write_prepare &&
7579 		  ops->read_write_prepare(vcpu, val, bytes))
7580 		return X86EMUL_CONTINUE;
7581 
7582 	vcpu->mmio_nr_fragments = 0;
7583 
7584 	/* Crossing a page boundary? */
7585 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
7586 		int now;
7587 
7588 		now = -addr & ~PAGE_MASK;
7589 		rc = emulator_read_write_onepage(addr, val, now, exception,
7590 						 vcpu, ops);
7591 
7592 		if (rc != X86EMUL_CONTINUE)
7593 			return rc;
7594 		addr += now;
7595 		if (ctxt->mode != X86EMUL_MODE_PROT64)
7596 			addr = (u32)addr;
7597 		val += now;
7598 		bytes -= now;
7599 	}
7600 
7601 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
7602 					 vcpu, ops);
7603 	if (rc != X86EMUL_CONTINUE)
7604 		return rc;
7605 
7606 	if (!vcpu->mmio_nr_fragments)
7607 		return rc;
7608 
7609 	gpa = vcpu->mmio_fragments[0].gpa;
7610 
7611 	vcpu->mmio_needed = 1;
7612 	vcpu->mmio_cur_fragment = 0;
7613 
7614 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
7615 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
7616 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
7617 	vcpu->run->mmio.phys_addr = gpa;
7618 
7619 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
7620 }
7621 
7622 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
7623 				  unsigned long addr,
7624 				  void *val,
7625 				  unsigned int bytes,
7626 				  struct x86_exception *exception)
7627 {
7628 	return emulator_read_write(ctxt, addr, val, bytes,
7629 				   exception, &read_emultor);
7630 }
7631 
7632 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
7633 			    unsigned long addr,
7634 			    const void *val,
7635 			    unsigned int bytes,
7636 			    struct x86_exception *exception)
7637 {
7638 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
7639 				   exception, &write_emultor);
7640 }
7641 
7642 #define emulator_try_cmpxchg_user(t, ptr, old, new) \
7643 	(__try_cmpxchg_user((t __user *)(ptr), (t *)(old), *(t *)(new), efault ## t))
7644 
7645 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
7646 				     unsigned long addr,
7647 				     const void *old,
7648 				     const void *new,
7649 				     unsigned int bytes,
7650 				     struct x86_exception *exception)
7651 {
7652 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7653 	u64 page_line_mask;
7654 	unsigned long hva;
7655 	gpa_t gpa;
7656 	int r;
7657 
7658 	/* guests cmpxchg8b have to be emulated atomically */
7659 	if (bytes > 8 || (bytes & (bytes - 1)))
7660 		goto emul_write;
7661 
7662 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
7663 
7664 	if (gpa == INVALID_GPA ||
7665 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
7666 		goto emul_write;
7667 
7668 	/*
7669 	 * Emulate the atomic as a straight write to avoid #AC if SLD is
7670 	 * enabled in the host and the access splits a cache line.
7671 	 */
7672 	if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
7673 		page_line_mask = ~(cache_line_size() - 1);
7674 	else
7675 		page_line_mask = PAGE_MASK;
7676 
7677 	if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
7678 		goto emul_write;
7679 
7680 	hva = kvm_vcpu_gfn_to_hva(vcpu, gpa_to_gfn(gpa));
7681 	if (kvm_is_error_hva(hva))
7682 		goto emul_write;
7683 
7684 	hva += offset_in_page(gpa);
7685 
7686 	switch (bytes) {
7687 	case 1:
7688 		r = emulator_try_cmpxchg_user(u8, hva, old, new);
7689 		break;
7690 	case 2:
7691 		r = emulator_try_cmpxchg_user(u16, hva, old, new);
7692 		break;
7693 	case 4:
7694 		r = emulator_try_cmpxchg_user(u32, hva, old, new);
7695 		break;
7696 	case 8:
7697 		r = emulator_try_cmpxchg_user(u64, hva, old, new);
7698 		break;
7699 	default:
7700 		BUG();
7701 	}
7702 
7703 	if (r < 0)
7704 		return X86EMUL_UNHANDLEABLE;
7705 	if (r)
7706 		return X86EMUL_CMPXCHG_FAILED;
7707 
7708 	kvm_page_track_write(vcpu, gpa, new, bytes);
7709 
7710 	return X86EMUL_CONTINUE;
7711 
7712 emul_write:
7713 	pr_warn_once("emulating exchange as write\n");
7714 
7715 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
7716 }
7717 
7718 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
7719 			       unsigned short port, void *data,
7720 			       unsigned int count, bool in)
7721 {
7722 	unsigned i;
7723 	int r;
7724 
7725 	WARN_ON_ONCE(vcpu->arch.pio.count);
7726 	for (i = 0; i < count; i++) {
7727 		if (in)
7728 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, port, size, data);
7729 		else
7730 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS, port, size, data);
7731 
7732 		if (r) {
7733 			if (i == 0)
7734 				goto userspace_io;
7735 
7736 			/*
7737 			 * Userspace must have unregistered the device while PIO
7738 			 * was running.  Drop writes / read as 0.
7739 			 */
7740 			if (in)
7741 				memset(data, 0, size * (count - i));
7742 			break;
7743 		}
7744 
7745 		data += size;
7746 	}
7747 	return 1;
7748 
7749 userspace_io:
7750 	vcpu->arch.pio.port = port;
7751 	vcpu->arch.pio.in = in;
7752 	vcpu->arch.pio.count = count;
7753 	vcpu->arch.pio.size = size;
7754 
7755 	if (in)
7756 		memset(vcpu->arch.pio_data, 0, size * count);
7757 	else
7758 		memcpy(vcpu->arch.pio_data, data, size * count);
7759 
7760 	vcpu->run->exit_reason = KVM_EXIT_IO;
7761 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
7762 	vcpu->run->io.size = size;
7763 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
7764 	vcpu->run->io.count = count;
7765 	vcpu->run->io.port = port;
7766 	return 0;
7767 }
7768 
7769 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
7770       			   unsigned short port, void *val, unsigned int count)
7771 {
7772 	int r = emulator_pio_in_out(vcpu, size, port, val, count, true);
7773 	if (r)
7774 		trace_kvm_pio(KVM_PIO_IN, port, size, count, val);
7775 
7776 	return r;
7777 }
7778 
7779 static void complete_emulator_pio_in(struct kvm_vcpu *vcpu, void *val)
7780 {
7781 	int size = vcpu->arch.pio.size;
7782 	unsigned int count = vcpu->arch.pio.count;
7783 	memcpy(val, vcpu->arch.pio_data, size * count);
7784 	trace_kvm_pio(KVM_PIO_IN, vcpu->arch.pio.port, size, count, vcpu->arch.pio_data);
7785 	vcpu->arch.pio.count = 0;
7786 }
7787 
7788 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
7789 				    int size, unsigned short port, void *val,
7790 				    unsigned int count)
7791 {
7792 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7793 	if (vcpu->arch.pio.count) {
7794 		/*
7795 		 * Complete a previous iteration that required userspace I/O.
7796 		 * Note, @count isn't guaranteed to match pio.count as userspace
7797 		 * can modify ECX before rerunning the vCPU.  Ignore any such
7798 		 * shenanigans as KVM doesn't support modifying the rep count,
7799 		 * and the emulator ensures @count doesn't overflow the buffer.
7800 		 */
7801 		complete_emulator_pio_in(vcpu, val);
7802 		return 1;
7803 	}
7804 
7805 	return emulator_pio_in(vcpu, size, port, val, count);
7806 }
7807 
7808 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
7809 			    unsigned short port, const void *val,
7810 			    unsigned int count)
7811 {
7812 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, val);
7813 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
7814 }
7815 
7816 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
7817 				     int size, unsigned short port,
7818 				     const void *val, unsigned int count)
7819 {
7820 	return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
7821 }
7822 
7823 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
7824 {
7825 	return static_call(kvm_x86_get_segment_base)(vcpu, seg);
7826 }
7827 
7828 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
7829 {
7830 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
7831 }
7832 
7833 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
7834 {
7835 	if (!need_emulate_wbinvd(vcpu))
7836 		return X86EMUL_CONTINUE;
7837 
7838 	if (static_call(kvm_x86_has_wbinvd_exit)()) {
7839 		int cpu = get_cpu();
7840 
7841 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
7842 		on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
7843 				wbinvd_ipi, NULL, 1);
7844 		put_cpu();
7845 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
7846 	} else
7847 		wbinvd();
7848 	return X86EMUL_CONTINUE;
7849 }
7850 
7851 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
7852 {
7853 	kvm_emulate_wbinvd_noskip(vcpu);
7854 	return kvm_skip_emulated_instruction(vcpu);
7855 }
7856 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
7857 
7858 
7859 
7860 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
7861 {
7862 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
7863 }
7864 
7865 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
7866 			    unsigned long *dest)
7867 {
7868 	kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
7869 }
7870 
7871 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
7872 			   unsigned long value)
7873 {
7874 
7875 	return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
7876 }
7877 
7878 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
7879 {
7880 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
7881 }
7882 
7883 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
7884 {
7885 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7886 	unsigned long value;
7887 
7888 	switch (cr) {
7889 	case 0:
7890 		value = kvm_read_cr0(vcpu);
7891 		break;
7892 	case 2:
7893 		value = vcpu->arch.cr2;
7894 		break;
7895 	case 3:
7896 		value = kvm_read_cr3(vcpu);
7897 		break;
7898 	case 4:
7899 		value = kvm_read_cr4(vcpu);
7900 		break;
7901 	case 8:
7902 		value = kvm_get_cr8(vcpu);
7903 		break;
7904 	default:
7905 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
7906 		return 0;
7907 	}
7908 
7909 	return value;
7910 }
7911 
7912 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
7913 {
7914 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7915 	int res = 0;
7916 
7917 	switch (cr) {
7918 	case 0:
7919 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
7920 		break;
7921 	case 2:
7922 		vcpu->arch.cr2 = val;
7923 		break;
7924 	case 3:
7925 		res = kvm_set_cr3(vcpu, val);
7926 		break;
7927 	case 4:
7928 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
7929 		break;
7930 	case 8:
7931 		res = kvm_set_cr8(vcpu, val);
7932 		break;
7933 	default:
7934 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
7935 		res = -1;
7936 	}
7937 
7938 	return res;
7939 }
7940 
7941 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
7942 {
7943 	return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
7944 }
7945 
7946 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7947 {
7948 	static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
7949 }
7950 
7951 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7952 {
7953 	static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
7954 }
7955 
7956 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7957 {
7958 	static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
7959 }
7960 
7961 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7962 {
7963 	static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
7964 }
7965 
7966 static unsigned long emulator_get_cached_segment_base(
7967 	struct x86_emulate_ctxt *ctxt, int seg)
7968 {
7969 	return get_segment_base(emul_to_vcpu(ctxt), seg);
7970 }
7971 
7972 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
7973 				 struct desc_struct *desc, u32 *base3,
7974 				 int seg)
7975 {
7976 	struct kvm_segment var;
7977 
7978 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
7979 	*selector = var.selector;
7980 
7981 	if (var.unusable) {
7982 		memset(desc, 0, sizeof(*desc));
7983 		if (base3)
7984 			*base3 = 0;
7985 		return false;
7986 	}
7987 
7988 	if (var.g)
7989 		var.limit >>= 12;
7990 	set_desc_limit(desc, var.limit);
7991 	set_desc_base(desc, (unsigned long)var.base);
7992 #ifdef CONFIG_X86_64
7993 	if (base3)
7994 		*base3 = var.base >> 32;
7995 #endif
7996 	desc->type = var.type;
7997 	desc->s = var.s;
7998 	desc->dpl = var.dpl;
7999 	desc->p = var.present;
8000 	desc->avl = var.avl;
8001 	desc->l = var.l;
8002 	desc->d = var.db;
8003 	desc->g = var.g;
8004 
8005 	return true;
8006 }
8007 
8008 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
8009 				 struct desc_struct *desc, u32 base3,
8010 				 int seg)
8011 {
8012 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8013 	struct kvm_segment var;
8014 
8015 	var.selector = selector;
8016 	var.base = get_desc_base(desc);
8017 #ifdef CONFIG_X86_64
8018 	var.base |= ((u64)base3) << 32;
8019 #endif
8020 	var.limit = get_desc_limit(desc);
8021 	if (desc->g)
8022 		var.limit = (var.limit << 12) | 0xfff;
8023 	var.type = desc->type;
8024 	var.dpl = desc->dpl;
8025 	var.db = desc->d;
8026 	var.s = desc->s;
8027 	var.l = desc->l;
8028 	var.g = desc->g;
8029 	var.avl = desc->avl;
8030 	var.present = desc->p;
8031 	var.unusable = !var.present;
8032 	var.padding = 0;
8033 
8034 	kvm_set_segment(vcpu, &var, seg);
8035 	return;
8036 }
8037 
8038 static int emulator_get_msr_with_filter(struct x86_emulate_ctxt *ctxt,
8039 					u32 msr_index, u64 *pdata)
8040 {
8041 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8042 	int r;
8043 
8044 	r = kvm_get_msr_with_filter(vcpu, msr_index, pdata);
8045 	if (r < 0)
8046 		return X86EMUL_UNHANDLEABLE;
8047 
8048 	if (r) {
8049 		if (kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_RDMSR, 0,
8050 				       complete_emulated_rdmsr, r))
8051 			return X86EMUL_IO_NEEDED;
8052 
8053 		trace_kvm_msr_read_ex(msr_index);
8054 		return X86EMUL_PROPAGATE_FAULT;
8055 	}
8056 
8057 	trace_kvm_msr_read(msr_index, *pdata);
8058 	return X86EMUL_CONTINUE;
8059 }
8060 
8061 static int emulator_set_msr_with_filter(struct x86_emulate_ctxt *ctxt,
8062 					u32 msr_index, u64 data)
8063 {
8064 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8065 	int r;
8066 
8067 	r = kvm_set_msr_with_filter(vcpu, msr_index, data);
8068 	if (r < 0)
8069 		return X86EMUL_UNHANDLEABLE;
8070 
8071 	if (r) {
8072 		if (kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_WRMSR, data,
8073 				       complete_emulated_msr_access, r))
8074 			return X86EMUL_IO_NEEDED;
8075 
8076 		trace_kvm_msr_write_ex(msr_index, data);
8077 		return X86EMUL_PROPAGATE_FAULT;
8078 	}
8079 
8080 	trace_kvm_msr_write(msr_index, data);
8081 	return X86EMUL_CONTINUE;
8082 }
8083 
8084 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
8085 			    u32 msr_index, u64 *pdata)
8086 {
8087 	return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
8088 }
8089 
8090 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
8091 			      u32 pmc)
8092 {
8093 	if (kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc))
8094 		return 0;
8095 	return -EINVAL;
8096 }
8097 
8098 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
8099 			     u32 pmc, u64 *pdata)
8100 {
8101 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
8102 }
8103 
8104 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
8105 {
8106 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
8107 }
8108 
8109 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
8110 			      struct x86_instruction_info *info,
8111 			      enum x86_intercept_stage stage)
8112 {
8113 	return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
8114 					    &ctxt->exception);
8115 }
8116 
8117 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
8118 			      u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
8119 			      bool exact_only)
8120 {
8121 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
8122 }
8123 
8124 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
8125 {
8126 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
8127 }
8128 
8129 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
8130 {
8131 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
8132 }
8133 
8134 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
8135 {
8136 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
8137 }
8138 
8139 static bool emulator_guest_has_rdpid(struct x86_emulate_ctxt *ctxt)
8140 {
8141 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_RDPID);
8142 }
8143 
8144 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
8145 {
8146 	return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
8147 }
8148 
8149 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
8150 {
8151 	kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
8152 }
8153 
8154 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
8155 {
8156 	static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
8157 }
8158 
8159 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
8160 {
8161 	return emul_to_vcpu(ctxt)->arch.hflags;
8162 }
8163 
8164 #ifndef CONFIG_KVM_SMM
8165 static int emulator_leave_smm(struct x86_emulate_ctxt *ctxt)
8166 {
8167 	WARN_ON_ONCE(1);
8168 	return X86EMUL_UNHANDLEABLE;
8169 }
8170 #endif
8171 
8172 static void emulator_triple_fault(struct x86_emulate_ctxt *ctxt)
8173 {
8174 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, emul_to_vcpu(ctxt));
8175 }
8176 
8177 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
8178 {
8179 	return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
8180 }
8181 
8182 static void emulator_vm_bugged(struct x86_emulate_ctxt *ctxt)
8183 {
8184 	struct kvm *kvm = emul_to_vcpu(ctxt)->kvm;
8185 
8186 	if (!kvm->vm_bugged)
8187 		kvm_vm_bugged(kvm);
8188 }
8189 
8190 static const struct x86_emulate_ops emulate_ops = {
8191 	.vm_bugged           = emulator_vm_bugged,
8192 	.read_gpr            = emulator_read_gpr,
8193 	.write_gpr           = emulator_write_gpr,
8194 	.read_std            = emulator_read_std,
8195 	.write_std           = emulator_write_std,
8196 	.fetch               = kvm_fetch_guest_virt,
8197 	.read_emulated       = emulator_read_emulated,
8198 	.write_emulated      = emulator_write_emulated,
8199 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
8200 	.invlpg              = emulator_invlpg,
8201 	.pio_in_emulated     = emulator_pio_in_emulated,
8202 	.pio_out_emulated    = emulator_pio_out_emulated,
8203 	.get_segment         = emulator_get_segment,
8204 	.set_segment         = emulator_set_segment,
8205 	.get_cached_segment_base = emulator_get_cached_segment_base,
8206 	.get_gdt             = emulator_get_gdt,
8207 	.get_idt	     = emulator_get_idt,
8208 	.set_gdt             = emulator_set_gdt,
8209 	.set_idt	     = emulator_set_idt,
8210 	.get_cr              = emulator_get_cr,
8211 	.set_cr              = emulator_set_cr,
8212 	.cpl                 = emulator_get_cpl,
8213 	.get_dr              = emulator_get_dr,
8214 	.set_dr              = emulator_set_dr,
8215 	.set_msr_with_filter = emulator_set_msr_with_filter,
8216 	.get_msr_with_filter = emulator_get_msr_with_filter,
8217 	.get_msr             = emulator_get_msr,
8218 	.check_pmc	     = emulator_check_pmc,
8219 	.read_pmc            = emulator_read_pmc,
8220 	.halt                = emulator_halt,
8221 	.wbinvd              = emulator_wbinvd,
8222 	.fix_hypercall       = emulator_fix_hypercall,
8223 	.intercept           = emulator_intercept,
8224 	.get_cpuid           = emulator_get_cpuid,
8225 	.guest_has_long_mode = emulator_guest_has_long_mode,
8226 	.guest_has_movbe     = emulator_guest_has_movbe,
8227 	.guest_has_fxsr      = emulator_guest_has_fxsr,
8228 	.guest_has_rdpid     = emulator_guest_has_rdpid,
8229 	.set_nmi_mask        = emulator_set_nmi_mask,
8230 	.get_hflags          = emulator_get_hflags,
8231 	.leave_smm           = emulator_leave_smm,
8232 	.triple_fault        = emulator_triple_fault,
8233 	.set_xcr             = emulator_set_xcr,
8234 };
8235 
8236 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
8237 {
8238 	u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
8239 	/*
8240 	 * an sti; sti; sequence only disable interrupts for the first
8241 	 * instruction. So, if the last instruction, be it emulated or
8242 	 * not, left the system with the INT_STI flag enabled, it
8243 	 * means that the last instruction is an sti. We should not
8244 	 * leave the flag on in this case. The same goes for mov ss
8245 	 */
8246 	if (int_shadow & mask)
8247 		mask = 0;
8248 	if (unlikely(int_shadow || mask)) {
8249 		static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
8250 		if (!mask)
8251 			kvm_make_request(KVM_REQ_EVENT, vcpu);
8252 	}
8253 }
8254 
8255 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
8256 {
8257 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8258 
8259 	if (ctxt->exception.vector == PF_VECTOR)
8260 		kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
8261 	else if (ctxt->exception.error_code_valid)
8262 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
8263 				      ctxt->exception.error_code);
8264 	else
8265 		kvm_queue_exception(vcpu, ctxt->exception.vector);
8266 }
8267 
8268 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
8269 {
8270 	struct x86_emulate_ctxt *ctxt;
8271 
8272 	ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
8273 	if (!ctxt) {
8274 		pr_err("failed to allocate vcpu's emulator\n");
8275 		return NULL;
8276 	}
8277 
8278 	ctxt->vcpu = vcpu;
8279 	ctxt->ops = &emulate_ops;
8280 	vcpu->arch.emulate_ctxt = ctxt;
8281 
8282 	return ctxt;
8283 }
8284 
8285 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
8286 {
8287 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8288 	int cs_db, cs_l;
8289 
8290 	static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
8291 
8292 	ctxt->gpa_available = false;
8293 	ctxt->eflags = kvm_get_rflags(vcpu);
8294 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
8295 
8296 	ctxt->eip = kvm_rip_read(vcpu);
8297 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
8298 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
8299 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
8300 		     cs_db				? X86EMUL_MODE_PROT32 :
8301 							  X86EMUL_MODE_PROT16;
8302 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
8303 
8304 	ctxt->interruptibility = 0;
8305 	ctxt->have_exception = false;
8306 	ctxt->exception.vector = -1;
8307 	ctxt->perm_ok = false;
8308 
8309 	init_decode_cache(ctxt);
8310 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8311 }
8312 
8313 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
8314 {
8315 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8316 	int ret;
8317 
8318 	init_emulate_ctxt(vcpu);
8319 
8320 	ctxt->op_bytes = 2;
8321 	ctxt->ad_bytes = 2;
8322 	ctxt->_eip = ctxt->eip + inc_eip;
8323 	ret = emulate_int_real(ctxt, irq);
8324 
8325 	if (ret != X86EMUL_CONTINUE) {
8326 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
8327 	} else {
8328 		ctxt->eip = ctxt->_eip;
8329 		kvm_rip_write(vcpu, ctxt->eip);
8330 		kvm_set_rflags(vcpu, ctxt->eflags);
8331 	}
8332 }
8333 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
8334 
8335 static void prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, u64 *data,
8336 					   u8 ndata, u8 *insn_bytes, u8 insn_size)
8337 {
8338 	struct kvm_run *run = vcpu->run;
8339 	u64 info[5];
8340 	u8 info_start;
8341 
8342 	/*
8343 	 * Zero the whole array used to retrieve the exit info, as casting to
8344 	 * u32 for select entries will leave some chunks uninitialized.
8345 	 */
8346 	memset(&info, 0, sizeof(info));
8347 
8348 	static_call(kvm_x86_get_exit_info)(vcpu, (u32 *)&info[0], &info[1],
8349 					   &info[2], (u32 *)&info[3],
8350 					   (u32 *)&info[4]);
8351 
8352 	run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8353 	run->emulation_failure.suberror = KVM_INTERNAL_ERROR_EMULATION;
8354 
8355 	/*
8356 	 * There's currently space for 13 entries, but 5 are used for the exit
8357 	 * reason and info.  Restrict to 4 to reduce the maintenance burden
8358 	 * when expanding kvm_run.emulation_failure in the future.
8359 	 */
8360 	if (WARN_ON_ONCE(ndata > 4))
8361 		ndata = 4;
8362 
8363 	/* Always include the flags as a 'data' entry. */
8364 	info_start = 1;
8365 	run->emulation_failure.flags = 0;
8366 
8367 	if (insn_size) {
8368 		BUILD_BUG_ON((sizeof(run->emulation_failure.insn_size) +
8369 			      sizeof(run->emulation_failure.insn_bytes) != 16));
8370 		info_start += 2;
8371 		run->emulation_failure.flags |=
8372 			KVM_INTERNAL_ERROR_EMULATION_FLAG_INSTRUCTION_BYTES;
8373 		run->emulation_failure.insn_size = insn_size;
8374 		memset(run->emulation_failure.insn_bytes, 0x90,
8375 		       sizeof(run->emulation_failure.insn_bytes));
8376 		memcpy(run->emulation_failure.insn_bytes, insn_bytes, insn_size);
8377 	}
8378 
8379 	memcpy(&run->internal.data[info_start], info, sizeof(info));
8380 	memcpy(&run->internal.data[info_start + ARRAY_SIZE(info)], data,
8381 	       ndata * sizeof(data[0]));
8382 
8383 	run->emulation_failure.ndata = info_start + ARRAY_SIZE(info) + ndata;
8384 }
8385 
8386 static void prepare_emulation_ctxt_failure_exit(struct kvm_vcpu *vcpu)
8387 {
8388 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8389 
8390 	prepare_emulation_failure_exit(vcpu, NULL, 0, ctxt->fetch.data,
8391 				       ctxt->fetch.end - ctxt->fetch.data);
8392 }
8393 
8394 void __kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, u64 *data,
8395 					  u8 ndata)
8396 {
8397 	prepare_emulation_failure_exit(vcpu, data, ndata, NULL, 0);
8398 }
8399 EXPORT_SYMBOL_GPL(__kvm_prepare_emulation_failure_exit);
8400 
8401 void kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu)
8402 {
8403 	__kvm_prepare_emulation_failure_exit(vcpu, NULL, 0);
8404 }
8405 EXPORT_SYMBOL_GPL(kvm_prepare_emulation_failure_exit);
8406 
8407 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
8408 {
8409 	struct kvm *kvm = vcpu->kvm;
8410 
8411 	++vcpu->stat.insn_emulation_fail;
8412 	trace_kvm_emulate_insn_failed(vcpu);
8413 
8414 	if (emulation_type & EMULTYPE_VMWARE_GP) {
8415 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
8416 		return 1;
8417 	}
8418 
8419 	if (kvm->arch.exit_on_emulation_error ||
8420 	    (emulation_type & EMULTYPE_SKIP)) {
8421 		prepare_emulation_ctxt_failure_exit(vcpu);
8422 		return 0;
8423 	}
8424 
8425 	kvm_queue_exception(vcpu, UD_VECTOR);
8426 
8427 	if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
8428 		prepare_emulation_ctxt_failure_exit(vcpu);
8429 		return 0;
8430 	}
8431 
8432 	return 1;
8433 }
8434 
8435 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
8436 				  bool write_fault_to_shadow_pgtable,
8437 				  int emulation_type)
8438 {
8439 	gpa_t gpa = cr2_or_gpa;
8440 	kvm_pfn_t pfn;
8441 
8442 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
8443 		return false;
8444 
8445 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
8446 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
8447 		return false;
8448 
8449 	if (!vcpu->arch.mmu->root_role.direct) {
8450 		/*
8451 		 * Write permission should be allowed since only
8452 		 * write access need to be emulated.
8453 		 */
8454 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
8455 
8456 		/*
8457 		 * If the mapping is invalid in guest, let cpu retry
8458 		 * it to generate fault.
8459 		 */
8460 		if (gpa == INVALID_GPA)
8461 			return true;
8462 	}
8463 
8464 	/*
8465 	 * Do not retry the unhandleable instruction if it faults on the
8466 	 * readonly host memory, otherwise it will goto a infinite loop:
8467 	 * retry instruction -> write #PF -> emulation fail -> retry
8468 	 * instruction -> ...
8469 	 */
8470 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
8471 
8472 	/*
8473 	 * If the instruction failed on the error pfn, it can not be fixed,
8474 	 * report the error to userspace.
8475 	 */
8476 	if (is_error_noslot_pfn(pfn))
8477 		return false;
8478 
8479 	kvm_release_pfn_clean(pfn);
8480 
8481 	/* The instructions are well-emulated on direct mmu. */
8482 	if (vcpu->arch.mmu->root_role.direct) {
8483 		unsigned int indirect_shadow_pages;
8484 
8485 		write_lock(&vcpu->kvm->mmu_lock);
8486 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
8487 		write_unlock(&vcpu->kvm->mmu_lock);
8488 
8489 		if (indirect_shadow_pages)
8490 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
8491 
8492 		return true;
8493 	}
8494 
8495 	/*
8496 	 * if emulation was due to access to shadowed page table
8497 	 * and it failed try to unshadow page and re-enter the
8498 	 * guest to let CPU execute the instruction.
8499 	 */
8500 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
8501 
8502 	/*
8503 	 * If the access faults on its page table, it can not
8504 	 * be fixed by unprotecting shadow page and it should
8505 	 * be reported to userspace.
8506 	 */
8507 	return !write_fault_to_shadow_pgtable;
8508 }
8509 
8510 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
8511 			      gpa_t cr2_or_gpa,  int emulation_type)
8512 {
8513 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8514 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
8515 
8516 	last_retry_eip = vcpu->arch.last_retry_eip;
8517 	last_retry_addr = vcpu->arch.last_retry_addr;
8518 
8519 	/*
8520 	 * If the emulation is caused by #PF and it is non-page_table
8521 	 * writing instruction, it means the VM-EXIT is caused by shadow
8522 	 * page protected, we can zap the shadow page and retry this
8523 	 * instruction directly.
8524 	 *
8525 	 * Note: if the guest uses a non-page-table modifying instruction
8526 	 * on the PDE that points to the instruction, then we will unmap
8527 	 * the instruction and go to an infinite loop. So, we cache the
8528 	 * last retried eip and the last fault address, if we meet the eip
8529 	 * and the address again, we can break out of the potential infinite
8530 	 * loop.
8531 	 */
8532 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
8533 
8534 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
8535 		return false;
8536 
8537 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
8538 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
8539 		return false;
8540 
8541 	if (x86_page_table_writing_insn(ctxt))
8542 		return false;
8543 
8544 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
8545 		return false;
8546 
8547 	vcpu->arch.last_retry_eip = ctxt->eip;
8548 	vcpu->arch.last_retry_addr = cr2_or_gpa;
8549 
8550 	if (!vcpu->arch.mmu->root_role.direct)
8551 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
8552 
8553 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
8554 
8555 	return true;
8556 }
8557 
8558 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
8559 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
8560 
8561 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
8562 				unsigned long *db)
8563 {
8564 	u32 dr6 = 0;
8565 	int i;
8566 	u32 enable, rwlen;
8567 
8568 	enable = dr7;
8569 	rwlen = dr7 >> 16;
8570 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
8571 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
8572 			dr6 |= (1 << i);
8573 	return dr6;
8574 }
8575 
8576 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
8577 {
8578 	struct kvm_run *kvm_run = vcpu->run;
8579 
8580 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
8581 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
8582 		kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
8583 		kvm_run->debug.arch.exception = DB_VECTOR;
8584 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
8585 		return 0;
8586 	}
8587 	kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
8588 	return 1;
8589 }
8590 
8591 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
8592 {
8593 	unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
8594 	int r;
8595 
8596 	r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
8597 	if (unlikely(!r))
8598 		return 0;
8599 
8600 	kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_INSTRUCTIONS);
8601 
8602 	/*
8603 	 * rflags is the old, "raw" value of the flags.  The new value has
8604 	 * not been saved yet.
8605 	 *
8606 	 * This is correct even for TF set by the guest, because "the
8607 	 * processor will not generate this exception after the instruction
8608 	 * that sets the TF flag".
8609 	 */
8610 	if (unlikely(rflags & X86_EFLAGS_TF))
8611 		r = kvm_vcpu_do_singlestep(vcpu);
8612 	return r;
8613 }
8614 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
8615 
8616 static bool kvm_is_code_breakpoint_inhibited(struct kvm_vcpu *vcpu)
8617 {
8618 	u32 shadow;
8619 
8620 	if (kvm_get_rflags(vcpu) & X86_EFLAGS_RF)
8621 		return true;
8622 
8623 	/*
8624 	 * Intel CPUs inhibit code #DBs when MOV/POP SS blocking is active,
8625 	 * but AMD CPUs do not.  MOV/POP SS blocking is rare, check that first
8626 	 * to avoid the relatively expensive CPUID lookup.
8627 	 */
8628 	shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
8629 	return (shadow & KVM_X86_SHADOW_INT_MOV_SS) &&
8630 	       guest_cpuid_is_intel(vcpu);
8631 }
8632 
8633 static bool kvm_vcpu_check_code_breakpoint(struct kvm_vcpu *vcpu,
8634 					   int emulation_type, int *r)
8635 {
8636 	WARN_ON_ONCE(emulation_type & EMULTYPE_NO_DECODE);
8637 
8638 	/*
8639 	 * Do not check for code breakpoints if hardware has already done the
8640 	 * checks, as inferred from the emulation type.  On NO_DECODE and SKIP,
8641 	 * the instruction has passed all exception checks, and all intercepted
8642 	 * exceptions that trigger emulation have lower priority than code
8643 	 * breakpoints, i.e. the fact that the intercepted exception occurred
8644 	 * means any code breakpoints have already been serviced.
8645 	 *
8646 	 * Note, KVM needs to check for code #DBs on EMULTYPE_TRAP_UD_FORCED as
8647 	 * hardware has checked the RIP of the magic prefix, but not the RIP of
8648 	 * the instruction being emulated.  The intent of forced emulation is
8649 	 * to behave as if KVM intercepted the instruction without an exception
8650 	 * and without a prefix.
8651 	 */
8652 	if (emulation_type & (EMULTYPE_NO_DECODE | EMULTYPE_SKIP |
8653 			      EMULTYPE_TRAP_UD | EMULTYPE_VMWARE_GP | EMULTYPE_PF))
8654 		return false;
8655 
8656 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
8657 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
8658 		struct kvm_run *kvm_run = vcpu->run;
8659 		unsigned long eip = kvm_get_linear_rip(vcpu);
8660 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
8661 					   vcpu->arch.guest_debug_dr7,
8662 					   vcpu->arch.eff_db);
8663 
8664 		if (dr6 != 0) {
8665 			kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
8666 			kvm_run->debug.arch.pc = eip;
8667 			kvm_run->debug.arch.exception = DB_VECTOR;
8668 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
8669 			*r = 0;
8670 			return true;
8671 		}
8672 	}
8673 
8674 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
8675 	    !kvm_is_code_breakpoint_inhibited(vcpu)) {
8676 		unsigned long eip = kvm_get_linear_rip(vcpu);
8677 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
8678 					   vcpu->arch.dr7,
8679 					   vcpu->arch.db);
8680 
8681 		if (dr6 != 0) {
8682 			kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
8683 			*r = 1;
8684 			return true;
8685 		}
8686 	}
8687 
8688 	return false;
8689 }
8690 
8691 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
8692 {
8693 	switch (ctxt->opcode_len) {
8694 	case 1:
8695 		switch (ctxt->b) {
8696 		case 0xe4:	/* IN */
8697 		case 0xe5:
8698 		case 0xec:
8699 		case 0xed:
8700 		case 0xe6:	/* OUT */
8701 		case 0xe7:
8702 		case 0xee:
8703 		case 0xef:
8704 		case 0x6c:	/* INS */
8705 		case 0x6d:
8706 		case 0x6e:	/* OUTS */
8707 		case 0x6f:
8708 			return true;
8709 		}
8710 		break;
8711 	case 2:
8712 		switch (ctxt->b) {
8713 		case 0x33:	/* RDPMC */
8714 			return true;
8715 		}
8716 		break;
8717 	}
8718 
8719 	return false;
8720 }
8721 
8722 /*
8723  * Decode an instruction for emulation.  The caller is responsible for handling
8724  * code breakpoints.  Note, manually detecting code breakpoints is unnecessary
8725  * (and wrong) when emulating on an intercepted fault-like exception[*], as
8726  * code breakpoints have higher priority and thus have already been done by
8727  * hardware.
8728  *
8729  * [*] Except #MC, which is higher priority, but KVM should never emulate in
8730  *     response to a machine check.
8731  */
8732 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
8733 				    void *insn, int insn_len)
8734 {
8735 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8736 	int r;
8737 
8738 	init_emulate_ctxt(vcpu);
8739 
8740 	r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
8741 
8742 	trace_kvm_emulate_insn_start(vcpu);
8743 	++vcpu->stat.insn_emulation;
8744 
8745 	return r;
8746 }
8747 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
8748 
8749 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
8750 			    int emulation_type, void *insn, int insn_len)
8751 {
8752 	int r;
8753 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8754 	bool writeback = true;
8755 	bool write_fault_to_spt;
8756 
8757 	if (unlikely(!kvm_can_emulate_insn(vcpu, emulation_type, insn, insn_len)))
8758 		return 1;
8759 
8760 	vcpu->arch.l1tf_flush_l1d = true;
8761 
8762 	/*
8763 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
8764 	 * never reused.
8765 	 */
8766 	write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
8767 	vcpu->arch.write_fault_to_shadow_pgtable = false;
8768 
8769 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
8770 		kvm_clear_exception_queue(vcpu);
8771 
8772 		/*
8773 		 * Return immediately if RIP hits a code breakpoint, such #DBs
8774 		 * are fault-like and are higher priority than any faults on
8775 		 * the code fetch itself.
8776 		 */
8777 		if (kvm_vcpu_check_code_breakpoint(vcpu, emulation_type, &r))
8778 			return r;
8779 
8780 		r = x86_decode_emulated_instruction(vcpu, emulation_type,
8781 						    insn, insn_len);
8782 		if (r != EMULATION_OK)  {
8783 			if ((emulation_type & EMULTYPE_TRAP_UD) ||
8784 			    (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
8785 				kvm_queue_exception(vcpu, UD_VECTOR);
8786 				return 1;
8787 			}
8788 			if (reexecute_instruction(vcpu, cr2_or_gpa,
8789 						  write_fault_to_spt,
8790 						  emulation_type))
8791 				return 1;
8792 
8793 			if (ctxt->have_exception &&
8794 			    !(emulation_type & EMULTYPE_SKIP)) {
8795 				/*
8796 				 * #UD should result in just EMULATION_FAILED, and trap-like
8797 				 * exception should not be encountered during decode.
8798 				 */
8799 				WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
8800 					     exception_type(ctxt->exception.vector) == EXCPT_TRAP);
8801 				inject_emulated_exception(vcpu);
8802 				return 1;
8803 			}
8804 			return handle_emulation_failure(vcpu, emulation_type);
8805 		}
8806 	}
8807 
8808 	if ((emulation_type & EMULTYPE_VMWARE_GP) &&
8809 	    !is_vmware_backdoor_opcode(ctxt)) {
8810 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
8811 		return 1;
8812 	}
8813 
8814 	/*
8815 	 * EMULTYPE_SKIP without EMULTYPE_COMPLETE_USER_EXIT is intended for
8816 	 * use *only* by vendor callbacks for kvm_skip_emulated_instruction().
8817 	 * The caller is responsible for updating interruptibility state and
8818 	 * injecting single-step #DBs.
8819 	 */
8820 	if (emulation_type & EMULTYPE_SKIP) {
8821 		if (ctxt->mode != X86EMUL_MODE_PROT64)
8822 			ctxt->eip = (u32)ctxt->_eip;
8823 		else
8824 			ctxt->eip = ctxt->_eip;
8825 
8826 		if (emulation_type & EMULTYPE_COMPLETE_USER_EXIT) {
8827 			r = 1;
8828 			goto writeback;
8829 		}
8830 
8831 		kvm_rip_write(vcpu, ctxt->eip);
8832 		if (ctxt->eflags & X86_EFLAGS_RF)
8833 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
8834 		return 1;
8835 	}
8836 
8837 	if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
8838 		return 1;
8839 
8840 	/* this is needed for vmware backdoor interface to work since it
8841 	   changes registers values  during IO operation */
8842 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
8843 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8844 		emulator_invalidate_register_cache(ctxt);
8845 	}
8846 
8847 restart:
8848 	if (emulation_type & EMULTYPE_PF) {
8849 		/* Save the faulting GPA (cr2) in the address field */
8850 		ctxt->exception.address = cr2_or_gpa;
8851 
8852 		/* With shadow page tables, cr2 contains a GVA or nGPA. */
8853 		if (vcpu->arch.mmu->root_role.direct) {
8854 			ctxt->gpa_available = true;
8855 			ctxt->gpa_val = cr2_or_gpa;
8856 		}
8857 	} else {
8858 		/* Sanitize the address out of an abundance of paranoia. */
8859 		ctxt->exception.address = 0;
8860 	}
8861 
8862 	r = x86_emulate_insn(ctxt);
8863 
8864 	if (r == EMULATION_INTERCEPTED)
8865 		return 1;
8866 
8867 	if (r == EMULATION_FAILED) {
8868 		if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
8869 					emulation_type))
8870 			return 1;
8871 
8872 		return handle_emulation_failure(vcpu, emulation_type);
8873 	}
8874 
8875 	if (ctxt->have_exception) {
8876 		r = 1;
8877 		inject_emulated_exception(vcpu);
8878 	} else if (vcpu->arch.pio.count) {
8879 		if (!vcpu->arch.pio.in) {
8880 			/* FIXME: return into emulator if single-stepping.  */
8881 			vcpu->arch.pio.count = 0;
8882 		} else {
8883 			writeback = false;
8884 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
8885 		}
8886 		r = 0;
8887 	} else if (vcpu->mmio_needed) {
8888 		++vcpu->stat.mmio_exits;
8889 
8890 		if (!vcpu->mmio_is_write)
8891 			writeback = false;
8892 		r = 0;
8893 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8894 	} else if (vcpu->arch.complete_userspace_io) {
8895 		writeback = false;
8896 		r = 0;
8897 	} else if (r == EMULATION_RESTART)
8898 		goto restart;
8899 	else
8900 		r = 1;
8901 
8902 writeback:
8903 	if (writeback) {
8904 		unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
8905 		toggle_interruptibility(vcpu, ctxt->interruptibility);
8906 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8907 
8908 		/*
8909 		 * Note, EXCPT_DB is assumed to be fault-like as the emulator
8910 		 * only supports code breakpoints and general detect #DB, both
8911 		 * of which are fault-like.
8912 		 */
8913 		if (!ctxt->have_exception ||
8914 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
8915 			kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_INSTRUCTIONS);
8916 			if (ctxt->is_branch)
8917 				kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_BRANCH_INSTRUCTIONS);
8918 			kvm_rip_write(vcpu, ctxt->eip);
8919 			if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
8920 				r = kvm_vcpu_do_singlestep(vcpu);
8921 			static_call_cond(kvm_x86_update_emulated_instruction)(vcpu);
8922 			__kvm_set_rflags(vcpu, ctxt->eflags);
8923 		}
8924 
8925 		/*
8926 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
8927 		 * do nothing, and it will be requested again as soon as
8928 		 * the shadow expires.  But we still need to check here,
8929 		 * because POPF has no interrupt shadow.
8930 		 */
8931 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
8932 			kvm_make_request(KVM_REQ_EVENT, vcpu);
8933 	} else
8934 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
8935 
8936 	return r;
8937 }
8938 
8939 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
8940 {
8941 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
8942 }
8943 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
8944 
8945 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
8946 					void *insn, int insn_len)
8947 {
8948 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
8949 }
8950 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
8951 
8952 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
8953 {
8954 	vcpu->arch.pio.count = 0;
8955 	return 1;
8956 }
8957 
8958 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
8959 {
8960 	vcpu->arch.pio.count = 0;
8961 
8962 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
8963 		return 1;
8964 
8965 	return kvm_skip_emulated_instruction(vcpu);
8966 }
8967 
8968 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
8969 			    unsigned short port)
8970 {
8971 	unsigned long val = kvm_rax_read(vcpu);
8972 	int ret = emulator_pio_out(vcpu, size, port, &val, 1);
8973 
8974 	if (ret)
8975 		return ret;
8976 
8977 	/*
8978 	 * Workaround userspace that relies on old KVM behavior of %rip being
8979 	 * incremented prior to exiting to userspace to handle "OUT 0x7e".
8980 	 */
8981 	if (port == 0x7e &&
8982 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
8983 		vcpu->arch.complete_userspace_io =
8984 			complete_fast_pio_out_port_0x7e;
8985 		kvm_skip_emulated_instruction(vcpu);
8986 	} else {
8987 		vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8988 		vcpu->arch.complete_userspace_io = complete_fast_pio_out;
8989 	}
8990 	return 0;
8991 }
8992 
8993 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
8994 {
8995 	unsigned long val;
8996 
8997 	/* We should only ever be called with arch.pio.count equal to 1 */
8998 	BUG_ON(vcpu->arch.pio.count != 1);
8999 
9000 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
9001 		vcpu->arch.pio.count = 0;
9002 		return 1;
9003 	}
9004 
9005 	/* For size less than 4 we merge, else we zero extend */
9006 	val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
9007 
9008 	complete_emulator_pio_in(vcpu, &val);
9009 	kvm_rax_write(vcpu, val);
9010 
9011 	return kvm_skip_emulated_instruction(vcpu);
9012 }
9013 
9014 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
9015 			   unsigned short port)
9016 {
9017 	unsigned long val;
9018 	int ret;
9019 
9020 	/* For size less than 4 we merge, else we zero extend */
9021 	val = (size < 4) ? kvm_rax_read(vcpu) : 0;
9022 
9023 	ret = emulator_pio_in(vcpu, size, port, &val, 1);
9024 	if (ret) {
9025 		kvm_rax_write(vcpu, val);
9026 		return ret;
9027 	}
9028 
9029 	vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
9030 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
9031 
9032 	return 0;
9033 }
9034 
9035 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
9036 {
9037 	int ret;
9038 
9039 	if (in)
9040 		ret = kvm_fast_pio_in(vcpu, size, port);
9041 	else
9042 		ret = kvm_fast_pio_out(vcpu, size, port);
9043 	return ret && kvm_skip_emulated_instruction(vcpu);
9044 }
9045 EXPORT_SYMBOL_GPL(kvm_fast_pio);
9046 
9047 static int kvmclock_cpu_down_prep(unsigned int cpu)
9048 {
9049 	__this_cpu_write(cpu_tsc_khz, 0);
9050 	return 0;
9051 }
9052 
9053 static void tsc_khz_changed(void *data)
9054 {
9055 	struct cpufreq_freqs *freq = data;
9056 	unsigned long khz = 0;
9057 
9058 	WARN_ON_ONCE(boot_cpu_has(X86_FEATURE_CONSTANT_TSC));
9059 
9060 	if (data)
9061 		khz = freq->new;
9062 	else
9063 		khz = cpufreq_quick_get(raw_smp_processor_id());
9064 	if (!khz)
9065 		khz = tsc_khz;
9066 	__this_cpu_write(cpu_tsc_khz, khz);
9067 }
9068 
9069 #ifdef CONFIG_X86_64
9070 static void kvm_hyperv_tsc_notifier(void)
9071 {
9072 	struct kvm *kvm;
9073 	int cpu;
9074 
9075 	mutex_lock(&kvm_lock);
9076 	list_for_each_entry(kvm, &vm_list, vm_list)
9077 		kvm_make_mclock_inprogress_request(kvm);
9078 
9079 	/* no guest entries from this point */
9080 	hyperv_stop_tsc_emulation();
9081 
9082 	/* TSC frequency always matches when on Hyper-V */
9083 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
9084 		for_each_present_cpu(cpu)
9085 			per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
9086 	}
9087 	kvm_caps.max_guest_tsc_khz = tsc_khz;
9088 
9089 	list_for_each_entry(kvm, &vm_list, vm_list) {
9090 		__kvm_start_pvclock_update(kvm);
9091 		pvclock_update_vm_gtod_copy(kvm);
9092 		kvm_end_pvclock_update(kvm);
9093 	}
9094 
9095 	mutex_unlock(&kvm_lock);
9096 }
9097 #endif
9098 
9099 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
9100 {
9101 	struct kvm *kvm;
9102 	struct kvm_vcpu *vcpu;
9103 	int send_ipi = 0;
9104 	unsigned long i;
9105 
9106 	/*
9107 	 * We allow guests to temporarily run on slowing clocks,
9108 	 * provided we notify them after, or to run on accelerating
9109 	 * clocks, provided we notify them before.  Thus time never
9110 	 * goes backwards.
9111 	 *
9112 	 * However, we have a problem.  We can't atomically update
9113 	 * the frequency of a given CPU from this function; it is
9114 	 * merely a notifier, which can be called from any CPU.
9115 	 * Changing the TSC frequency at arbitrary points in time
9116 	 * requires a recomputation of local variables related to
9117 	 * the TSC for each VCPU.  We must flag these local variables
9118 	 * to be updated and be sure the update takes place with the
9119 	 * new frequency before any guests proceed.
9120 	 *
9121 	 * Unfortunately, the combination of hotplug CPU and frequency
9122 	 * change creates an intractable locking scenario; the order
9123 	 * of when these callouts happen is undefined with respect to
9124 	 * CPU hotplug, and they can race with each other.  As such,
9125 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
9126 	 * undefined; you can actually have a CPU frequency change take
9127 	 * place in between the computation of X and the setting of the
9128 	 * variable.  To protect against this problem, all updates of
9129 	 * the per_cpu tsc_khz variable are done in an interrupt
9130 	 * protected IPI, and all callers wishing to update the value
9131 	 * must wait for a synchronous IPI to complete (which is trivial
9132 	 * if the caller is on the CPU already).  This establishes the
9133 	 * necessary total order on variable updates.
9134 	 *
9135 	 * Note that because a guest time update may take place
9136 	 * anytime after the setting of the VCPU's request bit, the
9137 	 * correct TSC value must be set before the request.  However,
9138 	 * to ensure the update actually makes it to any guest which
9139 	 * starts running in hardware virtualization between the set
9140 	 * and the acquisition of the spinlock, we must also ping the
9141 	 * CPU after setting the request bit.
9142 	 *
9143 	 */
9144 
9145 	smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
9146 
9147 	mutex_lock(&kvm_lock);
9148 	list_for_each_entry(kvm, &vm_list, vm_list) {
9149 		kvm_for_each_vcpu(i, vcpu, kvm) {
9150 			if (vcpu->cpu != cpu)
9151 				continue;
9152 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9153 			if (vcpu->cpu != raw_smp_processor_id())
9154 				send_ipi = 1;
9155 		}
9156 	}
9157 	mutex_unlock(&kvm_lock);
9158 
9159 	if (freq->old < freq->new && send_ipi) {
9160 		/*
9161 		 * We upscale the frequency.  Must make the guest
9162 		 * doesn't see old kvmclock values while running with
9163 		 * the new frequency, otherwise we risk the guest sees
9164 		 * time go backwards.
9165 		 *
9166 		 * In case we update the frequency for another cpu
9167 		 * (which might be in guest context) send an interrupt
9168 		 * to kick the cpu out of guest context.  Next time
9169 		 * guest context is entered kvmclock will be updated,
9170 		 * so the guest will not see stale values.
9171 		 */
9172 		smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
9173 	}
9174 }
9175 
9176 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
9177 				     void *data)
9178 {
9179 	struct cpufreq_freqs *freq = data;
9180 	int cpu;
9181 
9182 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
9183 		return 0;
9184 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
9185 		return 0;
9186 
9187 	for_each_cpu(cpu, freq->policy->cpus)
9188 		__kvmclock_cpufreq_notifier(freq, cpu);
9189 
9190 	return 0;
9191 }
9192 
9193 static struct notifier_block kvmclock_cpufreq_notifier_block = {
9194 	.notifier_call  = kvmclock_cpufreq_notifier
9195 };
9196 
9197 static int kvmclock_cpu_online(unsigned int cpu)
9198 {
9199 	tsc_khz_changed(NULL);
9200 	return 0;
9201 }
9202 
9203 static void kvm_timer_init(void)
9204 {
9205 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
9206 		max_tsc_khz = tsc_khz;
9207 
9208 		if (IS_ENABLED(CONFIG_CPU_FREQ)) {
9209 			struct cpufreq_policy *policy;
9210 			int cpu;
9211 
9212 			cpu = get_cpu();
9213 			policy = cpufreq_cpu_get(cpu);
9214 			if (policy) {
9215 				if (policy->cpuinfo.max_freq)
9216 					max_tsc_khz = policy->cpuinfo.max_freq;
9217 				cpufreq_cpu_put(policy);
9218 			}
9219 			put_cpu();
9220 		}
9221 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
9222 					  CPUFREQ_TRANSITION_NOTIFIER);
9223 
9224 		cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
9225 				  kvmclock_cpu_online, kvmclock_cpu_down_prep);
9226 	}
9227 }
9228 
9229 #ifdef CONFIG_X86_64
9230 static void pvclock_gtod_update_fn(struct work_struct *work)
9231 {
9232 	struct kvm *kvm;
9233 	struct kvm_vcpu *vcpu;
9234 	unsigned long i;
9235 
9236 	mutex_lock(&kvm_lock);
9237 	list_for_each_entry(kvm, &vm_list, vm_list)
9238 		kvm_for_each_vcpu(i, vcpu, kvm)
9239 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9240 	atomic_set(&kvm_guest_has_master_clock, 0);
9241 	mutex_unlock(&kvm_lock);
9242 }
9243 
9244 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
9245 
9246 /*
9247  * Indirection to move queue_work() out of the tk_core.seq write held
9248  * region to prevent possible deadlocks against time accessors which
9249  * are invoked with work related locks held.
9250  */
9251 static void pvclock_irq_work_fn(struct irq_work *w)
9252 {
9253 	queue_work(system_long_wq, &pvclock_gtod_work);
9254 }
9255 
9256 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
9257 
9258 /*
9259  * Notification about pvclock gtod data update.
9260  */
9261 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
9262 			       void *priv)
9263 {
9264 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
9265 	struct timekeeper *tk = priv;
9266 
9267 	update_pvclock_gtod(tk);
9268 
9269 	/*
9270 	 * Disable master clock if host does not trust, or does not use,
9271 	 * TSC based clocksource. Delegate queue_work() to irq_work as
9272 	 * this is invoked with tk_core.seq write held.
9273 	 */
9274 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
9275 	    atomic_read(&kvm_guest_has_master_clock) != 0)
9276 		irq_work_queue(&pvclock_irq_work);
9277 	return 0;
9278 }
9279 
9280 static struct notifier_block pvclock_gtod_notifier = {
9281 	.notifier_call = pvclock_gtod_notify,
9282 };
9283 #endif
9284 
9285 static inline void kvm_ops_update(struct kvm_x86_init_ops *ops)
9286 {
9287 	memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
9288 
9289 #define __KVM_X86_OP(func) \
9290 	static_call_update(kvm_x86_##func, kvm_x86_ops.func);
9291 #define KVM_X86_OP(func) \
9292 	WARN_ON(!kvm_x86_ops.func); __KVM_X86_OP(func)
9293 #define KVM_X86_OP_OPTIONAL __KVM_X86_OP
9294 #define KVM_X86_OP_OPTIONAL_RET0(func) \
9295 	static_call_update(kvm_x86_##func, (void *)kvm_x86_ops.func ? : \
9296 					   (void *)__static_call_return0);
9297 #include <asm/kvm-x86-ops.h>
9298 #undef __KVM_X86_OP
9299 
9300 	kvm_pmu_ops_update(ops->pmu_ops);
9301 }
9302 
9303 static int kvm_x86_check_processor_compatibility(void)
9304 {
9305 	int cpu = smp_processor_id();
9306 	struct cpuinfo_x86 *c = &cpu_data(cpu);
9307 
9308 	/*
9309 	 * Compatibility checks are done when loading KVM and when enabling
9310 	 * hardware, e.g. during CPU hotplug, to ensure all online CPUs are
9311 	 * compatible, i.e. KVM should never perform a compatibility check on
9312 	 * an offline CPU.
9313 	 */
9314 	WARN_ON(!cpu_online(cpu));
9315 
9316 	if (__cr4_reserved_bits(cpu_has, c) !=
9317 	    __cr4_reserved_bits(cpu_has, &boot_cpu_data))
9318 		return -EIO;
9319 
9320 	return static_call(kvm_x86_check_processor_compatibility)();
9321 }
9322 
9323 static void kvm_x86_check_cpu_compat(void *ret)
9324 {
9325 	*(int *)ret = kvm_x86_check_processor_compatibility();
9326 }
9327 
9328 static int __kvm_x86_vendor_init(struct kvm_x86_init_ops *ops)
9329 {
9330 	u64 host_pat;
9331 	int r, cpu;
9332 
9333 	if (kvm_x86_ops.hardware_enable) {
9334 		pr_err("already loaded vendor module '%s'\n", kvm_x86_ops.name);
9335 		return -EEXIST;
9336 	}
9337 
9338 	/*
9339 	 * KVM explicitly assumes that the guest has an FPU and
9340 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
9341 	 * vCPU's FPU state as a fxregs_state struct.
9342 	 */
9343 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
9344 		pr_err("inadequate fpu\n");
9345 		return -EOPNOTSUPP;
9346 	}
9347 
9348 	if (IS_ENABLED(CONFIG_PREEMPT_RT) && !boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
9349 		pr_err("RT requires X86_FEATURE_CONSTANT_TSC\n");
9350 		return -EOPNOTSUPP;
9351 	}
9352 
9353 	/*
9354 	 * KVM assumes that PAT entry '0' encodes WB memtype and simply zeroes
9355 	 * the PAT bits in SPTEs.  Bail if PAT[0] is programmed to something
9356 	 * other than WB.  Note, EPT doesn't utilize the PAT, but don't bother
9357 	 * with an exception.  PAT[0] is set to WB on RESET and also by the
9358 	 * kernel, i.e. failure indicates a kernel bug or broken firmware.
9359 	 */
9360 	if (rdmsrl_safe(MSR_IA32_CR_PAT, &host_pat) ||
9361 	    (host_pat & GENMASK(2, 0)) != 6) {
9362 		pr_err("host PAT[0] is not WB\n");
9363 		return -EIO;
9364 	}
9365 
9366 	x86_emulator_cache = kvm_alloc_emulator_cache();
9367 	if (!x86_emulator_cache) {
9368 		pr_err("failed to allocate cache for x86 emulator\n");
9369 		return -ENOMEM;
9370 	}
9371 
9372 	user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
9373 	if (!user_return_msrs) {
9374 		pr_err("failed to allocate percpu kvm_user_return_msrs\n");
9375 		r = -ENOMEM;
9376 		goto out_free_x86_emulator_cache;
9377 	}
9378 	kvm_nr_uret_msrs = 0;
9379 
9380 	r = kvm_mmu_vendor_module_init();
9381 	if (r)
9382 		goto out_free_percpu;
9383 
9384 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
9385 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
9386 		kvm_caps.supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
9387 	}
9388 
9389 	rdmsrl_safe(MSR_EFER, &host_efer);
9390 
9391 	if (boot_cpu_has(X86_FEATURE_XSAVES))
9392 		rdmsrl(MSR_IA32_XSS, host_xss);
9393 
9394 	kvm_init_pmu_capability(ops->pmu_ops);
9395 
9396 	r = ops->hardware_setup();
9397 	if (r != 0)
9398 		goto out_mmu_exit;
9399 
9400 	kvm_ops_update(ops);
9401 
9402 	for_each_online_cpu(cpu) {
9403 		smp_call_function_single(cpu, kvm_x86_check_cpu_compat, &r, 1);
9404 		if (r < 0)
9405 			goto out_unwind_ops;
9406 	}
9407 
9408 	/*
9409 	 * Point of no return!  DO NOT add error paths below this point unless
9410 	 * absolutely necessary, as most operations from this point forward
9411 	 * require unwinding.
9412 	 */
9413 	kvm_timer_init();
9414 
9415 	if (pi_inject_timer == -1)
9416 		pi_inject_timer = housekeeping_enabled(HK_TYPE_TIMER);
9417 #ifdef CONFIG_X86_64
9418 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
9419 
9420 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
9421 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
9422 #endif
9423 
9424 	kvm_register_perf_callbacks(ops->handle_intel_pt_intr);
9425 
9426 	if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
9427 		kvm_caps.supported_xss = 0;
9428 
9429 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
9430 	cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
9431 #undef __kvm_cpu_cap_has
9432 
9433 	if (kvm_caps.has_tsc_control) {
9434 		/*
9435 		 * Make sure the user can only configure tsc_khz values that
9436 		 * fit into a signed integer.
9437 		 * A min value is not calculated because it will always
9438 		 * be 1 on all machines.
9439 		 */
9440 		u64 max = min(0x7fffffffULL,
9441 			      __scale_tsc(kvm_caps.max_tsc_scaling_ratio, tsc_khz));
9442 		kvm_caps.max_guest_tsc_khz = max;
9443 	}
9444 	kvm_caps.default_tsc_scaling_ratio = 1ULL << kvm_caps.tsc_scaling_ratio_frac_bits;
9445 	kvm_init_msr_list();
9446 	return 0;
9447 
9448 out_unwind_ops:
9449 	kvm_x86_ops.hardware_enable = NULL;
9450 	static_call(kvm_x86_hardware_unsetup)();
9451 out_mmu_exit:
9452 	kvm_mmu_vendor_module_exit();
9453 out_free_percpu:
9454 	free_percpu(user_return_msrs);
9455 out_free_x86_emulator_cache:
9456 	kmem_cache_destroy(x86_emulator_cache);
9457 	return r;
9458 }
9459 
9460 int kvm_x86_vendor_init(struct kvm_x86_init_ops *ops)
9461 {
9462 	int r;
9463 
9464 	mutex_lock(&vendor_module_lock);
9465 	r = __kvm_x86_vendor_init(ops);
9466 	mutex_unlock(&vendor_module_lock);
9467 
9468 	return r;
9469 }
9470 EXPORT_SYMBOL_GPL(kvm_x86_vendor_init);
9471 
9472 void kvm_x86_vendor_exit(void)
9473 {
9474 	kvm_unregister_perf_callbacks();
9475 
9476 #ifdef CONFIG_X86_64
9477 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
9478 		clear_hv_tscchange_cb();
9479 #endif
9480 	kvm_lapic_exit();
9481 
9482 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
9483 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
9484 					    CPUFREQ_TRANSITION_NOTIFIER);
9485 		cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
9486 	}
9487 #ifdef CONFIG_X86_64
9488 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
9489 	irq_work_sync(&pvclock_irq_work);
9490 	cancel_work_sync(&pvclock_gtod_work);
9491 #endif
9492 	static_call(kvm_x86_hardware_unsetup)();
9493 	kvm_mmu_vendor_module_exit();
9494 	free_percpu(user_return_msrs);
9495 	kmem_cache_destroy(x86_emulator_cache);
9496 #ifdef CONFIG_KVM_XEN
9497 	static_key_deferred_flush(&kvm_xen_enabled);
9498 	WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
9499 #endif
9500 	mutex_lock(&vendor_module_lock);
9501 	kvm_x86_ops.hardware_enable = NULL;
9502 	mutex_unlock(&vendor_module_lock);
9503 }
9504 EXPORT_SYMBOL_GPL(kvm_x86_vendor_exit);
9505 
9506 static int __kvm_emulate_halt(struct kvm_vcpu *vcpu, int state, int reason)
9507 {
9508 	/*
9509 	 * The vCPU has halted, e.g. executed HLT.  Update the run state if the
9510 	 * local APIC is in-kernel, the run loop will detect the non-runnable
9511 	 * state and halt the vCPU.  Exit to userspace if the local APIC is
9512 	 * managed by userspace, in which case userspace is responsible for
9513 	 * handling wake events.
9514 	 */
9515 	++vcpu->stat.halt_exits;
9516 	if (lapic_in_kernel(vcpu)) {
9517 		vcpu->arch.mp_state = state;
9518 		return 1;
9519 	} else {
9520 		vcpu->run->exit_reason = reason;
9521 		return 0;
9522 	}
9523 }
9524 
9525 int kvm_emulate_halt_noskip(struct kvm_vcpu *vcpu)
9526 {
9527 	return __kvm_emulate_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
9528 }
9529 EXPORT_SYMBOL_GPL(kvm_emulate_halt_noskip);
9530 
9531 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
9532 {
9533 	int ret = kvm_skip_emulated_instruction(vcpu);
9534 	/*
9535 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
9536 	 * KVM_EXIT_DEBUG here.
9537 	 */
9538 	return kvm_emulate_halt_noskip(vcpu) && ret;
9539 }
9540 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
9541 
9542 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
9543 {
9544 	int ret = kvm_skip_emulated_instruction(vcpu);
9545 
9546 	return __kvm_emulate_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD,
9547 					KVM_EXIT_AP_RESET_HOLD) && ret;
9548 }
9549 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
9550 
9551 #ifdef CONFIG_X86_64
9552 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
9553 			        unsigned long clock_type)
9554 {
9555 	struct kvm_clock_pairing clock_pairing;
9556 	struct timespec64 ts;
9557 	u64 cycle;
9558 	int ret;
9559 
9560 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
9561 		return -KVM_EOPNOTSUPP;
9562 
9563 	/*
9564 	 * When tsc is in permanent catchup mode guests won't be able to use
9565 	 * pvclock_read_retry loop to get consistent view of pvclock
9566 	 */
9567 	if (vcpu->arch.tsc_always_catchup)
9568 		return -KVM_EOPNOTSUPP;
9569 
9570 	if (!kvm_get_walltime_and_clockread(&ts, &cycle))
9571 		return -KVM_EOPNOTSUPP;
9572 
9573 	clock_pairing.sec = ts.tv_sec;
9574 	clock_pairing.nsec = ts.tv_nsec;
9575 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
9576 	clock_pairing.flags = 0;
9577 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
9578 
9579 	ret = 0;
9580 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
9581 			    sizeof(struct kvm_clock_pairing)))
9582 		ret = -KVM_EFAULT;
9583 
9584 	return ret;
9585 }
9586 #endif
9587 
9588 /*
9589  * kvm_pv_kick_cpu_op:  Kick a vcpu.
9590  *
9591  * @apicid - apicid of vcpu to be kicked.
9592  */
9593 static void kvm_pv_kick_cpu_op(struct kvm *kvm, int apicid)
9594 {
9595 	/*
9596 	 * All other fields are unused for APIC_DM_REMRD, but may be consumed by
9597 	 * common code, e.g. for tracing. Defer initialization to the compiler.
9598 	 */
9599 	struct kvm_lapic_irq lapic_irq = {
9600 		.delivery_mode = APIC_DM_REMRD,
9601 		.dest_mode = APIC_DEST_PHYSICAL,
9602 		.shorthand = APIC_DEST_NOSHORT,
9603 		.dest_id = apicid,
9604 	};
9605 
9606 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
9607 }
9608 
9609 bool kvm_apicv_activated(struct kvm *kvm)
9610 {
9611 	return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
9612 }
9613 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
9614 
9615 bool kvm_vcpu_apicv_activated(struct kvm_vcpu *vcpu)
9616 {
9617 	ulong vm_reasons = READ_ONCE(vcpu->kvm->arch.apicv_inhibit_reasons);
9618 	ulong vcpu_reasons = static_call(kvm_x86_vcpu_get_apicv_inhibit_reasons)(vcpu);
9619 
9620 	return (vm_reasons | vcpu_reasons) == 0;
9621 }
9622 EXPORT_SYMBOL_GPL(kvm_vcpu_apicv_activated);
9623 
9624 static void set_or_clear_apicv_inhibit(unsigned long *inhibits,
9625 				       enum kvm_apicv_inhibit reason, bool set)
9626 {
9627 	if (set)
9628 		__set_bit(reason, inhibits);
9629 	else
9630 		__clear_bit(reason, inhibits);
9631 
9632 	trace_kvm_apicv_inhibit_changed(reason, set, *inhibits);
9633 }
9634 
9635 static void kvm_apicv_init(struct kvm *kvm)
9636 {
9637 	unsigned long *inhibits = &kvm->arch.apicv_inhibit_reasons;
9638 
9639 	init_rwsem(&kvm->arch.apicv_update_lock);
9640 
9641 	set_or_clear_apicv_inhibit(inhibits, APICV_INHIBIT_REASON_ABSENT, true);
9642 
9643 	if (!enable_apicv)
9644 		set_or_clear_apicv_inhibit(inhibits,
9645 					   APICV_INHIBIT_REASON_DISABLE, true);
9646 }
9647 
9648 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
9649 {
9650 	struct kvm_vcpu *target = NULL;
9651 	struct kvm_apic_map *map;
9652 
9653 	vcpu->stat.directed_yield_attempted++;
9654 
9655 	if (single_task_running())
9656 		goto no_yield;
9657 
9658 	rcu_read_lock();
9659 	map = rcu_dereference(vcpu->kvm->arch.apic_map);
9660 
9661 	if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
9662 		target = map->phys_map[dest_id]->vcpu;
9663 
9664 	rcu_read_unlock();
9665 
9666 	if (!target || !READ_ONCE(target->ready))
9667 		goto no_yield;
9668 
9669 	/* Ignore requests to yield to self */
9670 	if (vcpu == target)
9671 		goto no_yield;
9672 
9673 	if (kvm_vcpu_yield_to(target) <= 0)
9674 		goto no_yield;
9675 
9676 	vcpu->stat.directed_yield_successful++;
9677 
9678 no_yield:
9679 	return;
9680 }
9681 
9682 static int complete_hypercall_exit(struct kvm_vcpu *vcpu)
9683 {
9684 	u64 ret = vcpu->run->hypercall.ret;
9685 
9686 	if (!is_64_bit_mode(vcpu))
9687 		ret = (u32)ret;
9688 	kvm_rax_write(vcpu, ret);
9689 	++vcpu->stat.hypercalls;
9690 	return kvm_skip_emulated_instruction(vcpu);
9691 }
9692 
9693 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
9694 {
9695 	unsigned long nr, a0, a1, a2, a3, ret;
9696 	int op_64_bit;
9697 
9698 	if (kvm_xen_hypercall_enabled(vcpu->kvm))
9699 		return kvm_xen_hypercall(vcpu);
9700 
9701 	if (kvm_hv_hypercall_enabled(vcpu))
9702 		return kvm_hv_hypercall(vcpu);
9703 
9704 	nr = kvm_rax_read(vcpu);
9705 	a0 = kvm_rbx_read(vcpu);
9706 	a1 = kvm_rcx_read(vcpu);
9707 	a2 = kvm_rdx_read(vcpu);
9708 	a3 = kvm_rsi_read(vcpu);
9709 
9710 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
9711 
9712 	op_64_bit = is_64_bit_hypercall(vcpu);
9713 	if (!op_64_bit) {
9714 		nr &= 0xFFFFFFFF;
9715 		a0 &= 0xFFFFFFFF;
9716 		a1 &= 0xFFFFFFFF;
9717 		a2 &= 0xFFFFFFFF;
9718 		a3 &= 0xFFFFFFFF;
9719 	}
9720 
9721 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
9722 		ret = -KVM_EPERM;
9723 		goto out;
9724 	}
9725 
9726 	ret = -KVM_ENOSYS;
9727 
9728 	switch (nr) {
9729 	case KVM_HC_VAPIC_POLL_IRQ:
9730 		ret = 0;
9731 		break;
9732 	case KVM_HC_KICK_CPU:
9733 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
9734 			break;
9735 
9736 		kvm_pv_kick_cpu_op(vcpu->kvm, a1);
9737 		kvm_sched_yield(vcpu, a1);
9738 		ret = 0;
9739 		break;
9740 #ifdef CONFIG_X86_64
9741 	case KVM_HC_CLOCK_PAIRING:
9742 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
9743 		break;
9744 #endif
9745 	case KVM_HC_SEND_IPI:
9746 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
9747 			break;
9748 
9749 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
9750 		break;
9751 	case KVM_HC_SCHED_YIELD:
9752 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
9753 			break;
9754 
9755 		kvm_sched_yield(vcpu, a0);
9756 		ret = 0;
9757 		break;
9758 	case KVM_HC_MAP_GPA_RANGE: {
9759 		u64 gpa = a0, npages = a1, attrs = a2;
9760 
9761 		ret = -KVM_ENOSYS;
9762 		if (!(vcpu->kvm->arch.hypercall_exit_enabled & (1 << KVM_HC_MAP_GPA_RANGE)))
9763 			break;
9764 
9765 		if (!PAGE_ALIGNED(gpa) || !npages ||
9766 		    gpa_to_gfn(gpa) + npages <= gpa_to_gfn(gpa)) {
9767 			ret = -KVM_EINVAL;
9768 			break;
9769 		}
9770 
9771 		vcpu->run->exit_reason        = KVM_EXIT_HYPERCALL;
9772 		vcpu->run->hypercall.nr       = KVM_HC_MAP_GPA_RANGE;
9773 		vcpu->run->hypercall.args[0]  = gpa;
9774 		vcpu->run->hypercall.args[1]  = npages;
9775 		vcpu->run->hypercall.args[2]  = attrs;
9776 		vcpu->run->hypercall.longmode = op_64_bit;
9777 		vcpu->arch.complete_userspace_io = complete_hypercall_exit;
9778 		return 0;
9779 	}
9780 	default:
9781 		ret = -KVM_ENOSYS;
9782 		break;
9783 	}
9784 out:
9785 	if (!op_64_bit)
9786 		ret = (u32)ret;
9787 	kvm_rax_write(vcpu, ret);
9788 
9789 	++vcpu->stat.hypercalls;
9790 	return kvm_skip_emulated_instruction(vcpu);
9791 }
9792 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
9793 
9794 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
9795 {
9796 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
9797 	char instruction[3];
9798 	unsigned long rip = kvm_rip_read(vcpu);
9799 
9800 	/*
9801 	 * If the quirk is disabled, synthesize a #UD and let the guest pick up
9802 	 * the pieces.
9803 	 */
9804 	if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_FIX_HYPERCALL_INSN)) {
9805 		ctxt->exception.error_code_valid = false;
9806 		ctxt->exception.vector = UD_VECTOR;
9807 		ctxt->have_exception = true;
9808 		return X86EMUL_PROPAGATE_FAULT;
9809 	}
9810 
9811 	static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
9812 
9813 	return emulator_write_emulated(ctxt, rip, instruction, 3,
9814 		&ctxt->exception);
9815 }
9816 
9817 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
9818 {
9819 	return vcpu->run->request_interrupt_window &&
9820 		likely(!pic_in_kernel(vcpu->kvm));
9821 }
9822 
9823 /* Called within kvm->srcu read side.  */
9824 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
9825 {
9826 	struct kvm_run *kvm_run = vcpu->run;
9827 
9828 	kvm_run->if_flag = static_call(kvm_x86_get_if_flag)(vcpu);
9829 	kvm_run->cr8 = kvm_get_cr8(vcpu);
9830 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
9831 
9832 	kvm_run->ready_for_interrupt_injection =
9833 		pic_in_kernel(vcpu->kvm) ||
9834 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
9835 
9836 	if (is_smm(vcpu))
9837 		kvm_run->flags |= KVM_RUN_X86_SMM;
9838 }
9839 
9840 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
9841 {
9842 	int max_irr, tpr;
9843 
9844 	if (!kvm_x86_ops.update_cr8_intercept)
9845 		return;
9846 
9847 	if (!lapic_in_kernel(vcpu))
9848 		return;
9849 
9850 	if (vcpu->arch.apic->apicv_active)
9851 		return;
9852 
9853 	if (!vcpu->arch.apic->vapic_addr)
9854 		max_irr = kvm_lapic_find_highest_irr(vcpu);
9855 	else
9856 		max_irr = -1;
9857 
9858 	if (max_irr != -1)
9859 		max_irr >>= 4;
9860 
9861 	tpr = kvm_lapic_get_cr8(vcpu);
9862 
9863 	static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
9864 }
9865 
9866 
9867 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
9868 {
9869 	if (kvm_test_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9870 		kvm_x86_ops.nested_ops->triple_fault(vcpu);
9871 		return 1;
9872 	}
9873 
9874 	return kvm_x86_ops.nested_ops->check_events(vcpu);
9875 }
9876 
9877 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
9878 {
9879 	trace_kvm_inj_exception(vcpu->arch.exception.vector,
9880 				vcpu->arch.exception.has_error_code,
9881 				vcpu->arch.exception.error_code,
9882 				vcpu->arch.exception.injected);
9883 
9884 	if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
9885 		vcpu->arch.exception.error_code = false;
9886 	static_call(kvm_x86_inject_exception)(vcpu);
9887 }
9888 
9889 /*
9890  * Check for any event (interrupt or exception) that is ready to be injected,
9891  * and if there is at least one event, inject the event with the highest
9892  * priority.  This handles both "pending" events, i.e. events that have never
9893  * been injected into the guest, and "injected" events, i.e. events that were
9894  * injected as part of a previous VM-Enter, but weren't successfully delivered
9895  * and need to be re-injected.
9896  *
9897  * Note, this is not guaranteed to be invoked on a guest instruction boundary,
9898  * i.e. doesn't guarantee that there's an event window in the guest.  KVM must
9899  * be able to inject exceptions in the "middle" of an instruction, and so must
9900  * also be able to re-inject NMIs and IRQs in the middle of an instruction.
9901  * I.e. for exceptions and re-injected events, NOT invoking this on instruction
9902  * boundaries is necessary and correct.
9903  *
9904  * For simplicity, KVM uses a single path to inject all events (except events
9905  * that are injected directly from L1 to L2) and doesn't explicitly track
9906  * instruction boundaries for asynchronous events.  However, because VM-Exits
9907  * that can occur during instruction execution typically result in KVM skipping
9908  * the instruction or injecting an exception, e.g. instruction and exception
9909  * intercepts, and because pending exceptions have higher priority than pending
9910  * interrupts, KVM still honors instruction boundaries in most scenarios.
9911  *
9912  * But, if a VM-Exit occurs during instruction execution, and KVM does NOT skip
9913  * the instruction or inject an exception, then KVM can incorrecty inject a new
9914  * asynchrounous event if the event became pending after the CPU fetched the
9915  * instruction (in the guest).  E.g. if a page fault (#PF, #NPF, EPT violation)
9916  * occurs and is resolved by KVM, a coincident NMI, SMI, IRQ, etc... can be
9917  * injected on the restarted instruction instead of being deferred until the
9918  * instruction completes.
9919  *
9920  * In practice, this virtualization hole is unlikely to be observed by the
9921  * guest, and even less likely to cause functional problems.  To detect the
9922  * hole, the guest would have to trigger an event on a side effect of an early
9923  * phase of instruction execution, e.g. on the instruction fetch from memory.
9924  * And for it to be a functional problem, the guest would need to depend on the
9925  * ordering between that side effect, the instruction completing, _and_ the
9926  * delivery of the asynchronous event.
9927  */
9928 static int kvm_check_and_inject_events(struct kvm_vcpu *vcpu,
9929 				       bool *req_immediate_exit)
9930 {
9931 	bool can_inject;
9932 	int r;
9933 
9934 	/*
9935 	 * Process nested events first, as nested VM-Exit supercedes event
9936 	 * re-injection.  If there's an event queued for re-injection, it will
9937 	 * be saved into the appropriate vmc{b,s}12 fields on nested VM-Exit.
9938 	 */
9939 	if (is_guest_mode(vcpu))
9940 		r = kvm_check_nested_events(vcpu);
9941 	else
9942 		r = 0;
9943 
9944 	/*
9945 	 * Re-inject exceptions and events *especially* if immediate entry+exit
9946 	 * to/from L2 is needed, as any event that has already been injected
9947 	 * into L2 needs to complete its lifecycle before injecting a new event.
9948 	 *
9949 	 * Don't re-inject an NMI or interrupt if there is a pending exception.
9950 	 * This collision arises if an exception occurred while vectoring the
9951 	 * injected event, KVM intercepted said exception, and KVM ultimately
9952 	 * determined the fault belongs to the guest and queues the exception
9953 	 * for injection back into the guest.
9954 	 *
9955 	 * "Injected" interrupts can also collide with pending exceptions if
9956 	 * userspace ignores the "ready for injection" flag and blindly queues
9957 	 * an interrupt.  In that case, prioritizing the exception is correct,
9958 	 * as the exception "occurred" before the exit to userspace.  Trap-like
9959 	 * exceptions, e.g. most #DBs, have higher priority than interrupts.
9960 	 * And while fault-like exceptions, e.g. #GP and #PF, are the lowest
9961 	 * priority, they're only generated (pended) during instruction
9962 	 * execution, and interrupts are recognized at instruction boundaries.
9963 	 * Thus a pending fault-like exception means the fault occurred on the
9964 	 * *previous* instruction and must be serviced prior to recognizing any
9965 	 * new events in order to fully complete the previous instruction.
9966 	 */
9967 	if (vcpu->arch.exception.injected)
9968 		kvm_inject_exception(vcpu);
9969 	else if (kvm_is_exception_pending(vcpu))
9970 		; /* see above */
9971 	else if (vcpu->arch.nmi_injected)
9972 		static_call(kvm_x86_inject_nmi)(vcpu);
9973 	else if (vcpu->arch.interrupt.injected)
9974 		static_call(kvm_x86_inject_irq)(vcpu, true);
9975 
9976 	/*
9977 	 * Exceptions that morph to VM-Exits are handled above, and pending
9978 	 * exceptions on top of injected exceptions that do not VM-Exit should
9979 	 * either morph to #DF or, sadly, override the injected exception.
9980 	 */
9981 	WARN_ON_ONCE(vcpu->arch.exception.injected &&
9982 		     vcpu->arch.exception.pending);
9983 
9984 	/*
9985 	 * Bail if immediate entry+exit to/from the guest is needed to complete
9986 	 * nested VM-Enter or event re-injection so that a different pending
9987 	 * event can be serviced (or if KVM needs to exit to userspace).
9988 	 *
9989 	 * Otherwise, continue processing events even if VM-Exit occurred.  The
9990 	 * VM-Exit will have cleared exceptions that were meant for L2, but
9991 	 * there may now be events that can be injected into L1.
9992 	 */
9993 	if (r < 0)
9994 		goto out;
9995 
9996 	/*
9997 	 * A pending exception VM-Exit should either result in nested VM-Exit
9998 	 * or force an immediate re-entry and exit to/from L2, and exception
9999 	 * VM-Exits cannot be injected (flag should _never_ be set).
10000 	 */
10001 	WARN_ON_ONCE(vcpu->arch.exception_vmexit.injected ||
10002 		     vcpu->arch.exception_vmexit.pending);
10003 
10004 	/*
10005 	 * New events, other than exceptions, cannot be injected if KVM needs
10006 	 * to re-inject a previous event.  See above comments on re-injecting
10007 	 * for why pending exceptions get priority.
10008 	 */
10009 	can_inject = !kvm_event_needs_reinjection(vcpu);
10010 
10011 	if (vcpu->arch.exception.pending) {
10012 		/*
10013 		 * Fault-class exceptions, except #DBs, set RF=1 in the RFLAGS
10014 		 * value pushed on the stack.  Trap-like exception and all #DBs
10015 		 * leave RF as-is (KVM follows Intel's behavior in this regard;
10016 		 * AMD states that code breakpoint #DBs excplitly clear RF=0).
10017 		 *
10018 		 * Note, most versions of Intel's SDM and AMD's APM incorrectly
10019 		 * describe the behavior of General Detect #DBs, which are
10020 		 * fault-like.  They do _not_ set RF, a la code breakpoints.
10021 		 */
10022 		if (exception_type(vcpu->arch.exception.vector) == EXCPT_FAULT)
10023 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
10024 					     X86_EFLAGS_RF);
10025 
10026 		if (vcpu->arch.exception.vector == DB_VECTOR) {
10027 			kvm_deliver_exception_payload(vcpu, &vcpu->arch.exception);
10028 			if (vcpu->arch.dr7 & DR7_GD) {
10029 				vcpu->arch.dr7 &= ~DR7_GD;
10030 				kvm_update_dr7(vcpu);
10031 			}
10032 		}
10033 
10034 		kvm_inject_exception(vcpu);
10035 
10036 		vcpu->arch.exception.pending = false;
10037 		vcpu->arch.exception.injected = true;
10038 
10039 		can_inject = false;
10040 	}
10041 
10042 	/* Don't inject interrupts if the user asked to avoid doing so */
10043 	if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ)
10044 		return 0;
10045 
10046 	/*
10047 	 * Finally, inject interrupt events.  If an event cannot be injected
10048 	 * due to architectural conditions (e.g. IF=0) a window-open exit
10049 	 * will re-request KVM_REQ_EVENT.  Sometimes however an event is pending
10050 	 * and can architecturally be injected, but we cannot do it right now:
10051 	 * an interrupt could have arrived just now and we have to inject it
10052 	 * as a vmexit, or there could already an event in the queue, which is
10053 	 * indicated by can_inject.  In that case we request an immediate exit
10054 	 * in order to make progress and get back here for another iteration.
10055 	 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
10056 	 */
10057 #ifdef CONFIG_KVM_SMM
10058 	if (vcpu->arch.smi_pending) {
10059 		r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
10060 		if (r < 0)
10061 			goto out;
10062 		if (r) {
10063 			vcpu->arch.smi_pending = false;
10064 			++vcpu->arch.smi_count;
10065 			enter_smm(vcpu);
10066 			can_inject = false;
10067 		} else
10068 			static_call(kvm_x86_enable_smi_window)(vcpu);
10069 	}
10070 #endif
10071 
10072 	if (vcpu->arch.nmi_pending) {
10073 		r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
10074 		if (r < 0)
10075 			goto out;
10076 		if (r) {
10077 			--vcpu->arch.nmi_pending;
10078 			vcpu->arch.nmi_injected = true;
10079 			static_call(kvm_x86_inject_nmi)(vcpu);
10080 			can_inject = false;
10081 			WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
10082 		}
10083 		if (vcpu->arch.nmi_pending)
10084 			static_call(kvm_x86_enable_nmi_window)(vcpu);
10085 	}
10086 
10087 	if (kvm_cpu_has_injectable_intr(vcpu)) {
10088 		r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
10089 		if (r < 0)
10090 			goto out;
10091 		if (r) {
10092 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
10093 			static_call(kvm_x86_inject_irq)(vcpu, false);
10094 			WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
10095 		}
10096 		if (kvm_cpu_has_injectable_intr(vcpu))
10097 			static_call(kvm_x86_enable_irq_window)(vcpu);
10098 	}
10099 
10100 	if (is_guest_mode(vcpu) &&
10101 	    kvm_x86_ops.nested_ops->has_events &&
10102 	    kvm_x86_ops.nested_ops->has_events(vcpu))
10103 		*req_immediate_exit = true;
10104 
10105 	/*
10106 	 * KVM must never queue a new exception while injecting an event; KVM
10107 	 * is done emulating and should only propagate the to-be-injected event
10108 	 * to the VMCS/VMCB.  Queueing a new exception can put the vCPU into an
10109 	 * infinite loop as KVM will bail from VM-Enter to inject the pending
10110 	 * exception and start the cycle all over.
10111 	 *
10112 	 * Exempt triple faults as they have special handling and won't put the
10113 	 * vCPU into an infinite loop.  Triple fault can be queued when running
10114 	 * VMX without unrestricted guest, as that requires KVM to emulate Real
10115 	 * Mode events (see kvm_inject_realmode_interrupt()).
10116 	 */
10117 	WARN_ON_ONCE(vcpu->arch.exception.pending ||
10118 		     vcpu->arch.exception_vmexit.pending);
10119 	return 0;
10120 
10121 out:
10122 	if (r == -EBUSY) {
10123 		*req_immediate_exit = true;
10124 		r = 0;
10125 	}
10126 	return r;
10127 }
10128 
10129 static void process_nmi(struct kvm_vcpu *vcpu)
10130 {
10131 	unsigned limit = 2;
10132 
10133 	/*
10134 	 * x86 is limited to one NMI running, and one NMI pending after it.
10135 	 * If an NMI is already in progress, limit further NMIs to just one.
10136 	 * Otherwise, allow two (and we'll inject the first one immediately).
10137 	 */
10138 	if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
10139 		limit = 1;
10140 
10141 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
10142 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
10143 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10144 }
10145 
10146 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
10147 				       unsigned long *vcpu_bitmap)
10148 {
10149 	kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC, vcpu_bitmap);
10150 }
10151 
10152 void kvm_make_scan_ioapic_request(struct kvm *kvm)
10153 {
10154 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
10155 }
10156 
10157 void __kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
10158 {
10159 	struct kvm_lapic *apic = vcpu->arch.apic;
10160 	bool activate;
10161 
10162 	if (!lapic_in_kernel(vcpu))
10163 		return;
10164 
10165 	down_read(&vcpu->kvm->arch.apicv_update_lock);
10166 	preempt_disable();
10167 
10168 	/* Do not activate APICV when APIC is disabled */
10169 	activate = kvm_vcpu_apicv_activated(vcpu) &&
10170 		   (kvm_get_apic_mode(vcpu) != LAPIC_MODE_DISABLED);
10171 
10172 	if (apic->apicv_active == activate)
10173 		goto out;
10174 
10175 	apic->apicv_active = activate;
10176 	kvm_apic_update_apicv(vcpu);
10177 	static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
10178 
10179 	/*
10180 	 * When APICv gets disabled, we may still have injected interrupts
10181 	 * pending. At the same time, KVM_REQ_EVENT may not be set as APICv was
10182 	 * still active when the interrupt got accepted. Make sure
10183 	 * kvm_check_and_inject_events() is called to check for that.
10184 	 */
10185 	if (!apic->apicv_active)
10186 		kvm_make_request(KVM_REQ_EVENT, vcpu);
10187 
10188 out:
10189 	preempt_enable();
10190 	up_read(&vcpu->kvm->arch.apicv_update_lock);
10191 }
10192 EXPORT_SYMBOL_GPL(__kvm_vcpu_update_apicv);
10193 
10194 static void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
10195 {
10196 	if (!lapic_in_kernel(vcpu))
10197 		return;
10198 
10199 	/*
10200 	 * Due to sharing page tables across vCPUs, the xAPIC memslot must be
10201 	 * deleted if any vCPU has xAPIC virtualization and x2APIC enabled, but
10202 	 * and hardware doesn't support x2APIC virtualization.  E.g. some AMD
10203 	 * CPUs support AVIC but not x2APIC.  KVM still allows enabling AVIC in
10204 	 * this case so that KVM can the AVIC doorbell to inject interrupts to
10205 	 * running vCPUs, but KVM must not create SPTEs for the APIC base as
10206 	 * the vCPU would incorrectly be able to access the vAPIC page via MMIO
10207 	 * despite being in x2APIC mode.  For simplicity, inhibiting the APIC
10208 	 * access page is sticky.
10209 	 */
10210 	if (apic_x2apic_mode(vcpu->arch.apic) &&
10211 	    kvm_x86_ops.allow_apicv_in_x2apic_without_x2apic_virtualization)
10212 		kvm_inhibit_apic_access_page(vcpu);
10213 
10214 	__kvm_vcpu_update_apicv(vcpu);
10215 }
10216 
10217 void __kvm_set_or_clear_apicv_inhibit(struct kvm *kvm,
10218 				      enum kvm_apicv_inhibit reason, bool set)
10219 {
10220 	unsigned long old, new;
10221 
10222 	lockdep_assert_held_write(&kvm->arch.apicv_update_lock);
10223 
10224 	if (!(kvm_x86_ops.required_apicv_inhibits & BIT(reason)))
10225 		return;
10226 
10227 	old = new = kvm->arch.apicv_inhibit_reasons;
10228 
10229 	set_or_clear_apicv_inhibit(&new, reason, set);
10230 
10231 	if (!!old != !!new) {
10232 		/*
10233 		 * Kick all vCPUs before setting apicv_inhibit_reasons to avoid
10234 		 * false positives in the sanity check WARN in svm_vcpu_run().
10235 		 * This task will wait for all vCPUs to ack the kick IRQ before
10236 		 * updating apicv_inhibit_reasons, and all other vCPUs will
10237 		 * block on acquiring apicv_update_lock so that vCPUs can't
10238 		 * redo svm_vcpu_run() without seeing the new inhibit state.
10239 		 *
10240 		 * Note, holding apicv_update_lock and taking it in the read
10241 		 * side (handling the request) also prevents other vCPUs from
10242 		 * servicing the request with a stale apicv_inhibit_reasons.
10243 		 */
10244 		kvm_make_all_cpus_request(kvm, KVM_REQ_APICV_UPDATE);
10245 		kvm->arch.apicv_inhibit_reasons = new;
10246 		if (new) {
10247 			unsigned long gfn = gpa_to_gfn(APIC_DEFAULT_PHYS_BASE);
10248 			int idx = srcu_read_lock(&kvm->srcu);
10249 
10250 			kvm_zap_gfn_range(kvm, gfn, gfn+1);
10251 			srcu_read_unlock(&kvm->srcu, idx);
10252 		}
10253 	} else {
10254 		kvm->arch.apicv_inhibit_reasons = new;
10255 	}
10256 }
10257 
10258 void kvm_set_or_clear_apicv_inhibit(struct kvm *kvm,
10259 				    enum kvm_apicv_inhibit reason, bool set)
10260 {
10261 	if (!enable_apicv)
10262 		return;
10263 
10264 	down_write(&kvm->arch.apicv_update_lock);
10265 	__kvm_set_or_clear_apicv_inhibit(kvm, reason, set);
10266 	up_write(&kvm->arch.apicv_update_lock);
10267 }
10268 EXPORT_SYMBOL_GPL(kvm_set_or_clear_apicv_inhibit);
10269 
10270 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
10271 {
10272 	if (!kvm_apic_present(vcpu))
10273 		return;
10274 
10275 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
10276 
10277 	if (irqchip_split(vcpu->kvm))
10278 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
10279 	else {
10280 		static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
10281 		if (ioapic_in_kernel(vcpu->kvm))
10282 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
10283 	}
10284 
10285 	if (is_guest_mode(vcpu))
10286 		vcpu->arch.load_eoi_exitmap_pending = true;
10287 	else
10288 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
10289 }
10290 
10291 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
10292 {
10293 	u64 eoi_exit_bitmap[4];
10294 
10295 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
10296 		return;
10297 
10298 	if (to_hv_vcpu(vcpu)) {
10299 		bitmap_or((ulong *)eoi_exit_bitmap,
10300 			  vcpu->arch.ioapic_handled_vectors,
10301 			  to_hv_synic(vcpu)->vec_bitmap, 256);
10302 		static_call_cond(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
10303 		return;
10304 	}
10305 
10306 	static_call_cond(kvm_x86_load_eoi_exitmap)(
10307 		vcpu, (u64 *)vcpu->arch.ioapic_handled_vectors);
10308 }
10309 
10310 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
10311 					    unsigned long start, unsigned long end)
10312 {
10313 	unsigned long apic_address;
10314 
10315 	/*
10316 	 * The physical address of apic access page is stored in the VMCS.
10317 	 * Update it when it becomes invalid.
10318 	 */
10319 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
10320 	if (start <= apic_address && apic_address < end)
10321 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
10322 }
10323 
10324 void kvm_arch_guest_memory_reclaimed(struct kvm *kvm)
10325 {
10326 	static_call_cond(kvm_x86_guest_memory_reclaimed)(kvm);
10327 }
10328 
10329 static void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
10330 {
10331 	if (!lapic_in_kernel(vcpu))
10332 		return;
10333 
10334 	static_call_cond(kvm_x86_set_apic_access_page_addr)(vcpu);
10335 }
10336 
10337 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
10338 {
10339 	smp_send_reschedule(vcpu->cpu);
10340 }
10341 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
10342 
10343 /*
10344  * Called within kvm->srcu read side.
10345  * Returns 1 to let vcpu_run() continue the guest execution loop without
10346  * exiting to the userspace.  Otherwise, the value will be returned to the
10347  * userspace.
10348  */
10349 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
10350 {
10351 	int r;
10352 	bool req_int_win =
10353 		dm_request_for_irq_injection(vcpu) &&
10354 		kvm_cpu_accept_dm_intr(vcpu);
10355 	fastpath_t exit_fastpath;
10356 
10357 	bool req_immediate_exit = false;
10358 
10359 	if (kvm_request_pending(vcpu)) {
10360 		if (kvm_check_request(KVM_REQ_VM_DEAD, vcpu)) {
10361 			r = -EIO;
10362 			goto out;
10363 		}
10364 
10365 		if (kvm_dirty_ring_check_request(vcpu)) {
10366 			r = 0;
10367 			goto out;
10368 		}
10369 
10370 		if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
10371 			if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
10372 				r = 0;
10373 				goto out;
10374 			}
10375 		}
10376 		if (kvm_check_request(KVM_REQ_MMU_FREE_OBSOLETE_ROOTS, vcpu))
10377 			kvm_mmu_free_obsolete_roots(vcpu);
10378 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
10379 			__kvm_migrate_timers(vcpu);
10380 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
10381 			kvm_update_masterclock(vcpu->kvm);
10382 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
10383 			kvm_gen_kvmclock_update(vcpu);
10384 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
10385 			r = kvm_guest_time_update(vcpu);
10386 			if (unlikely(r))
10387 				goto out;
10388 		}
10389 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
10390 			kvm_mmu_sync_roots(vcpu);
10391 		if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
10392 			kvm_mmu_load_pgd(vcpu);
10393 
10394 		/*
10395 		 * Note, the order matters here, as flushing "all" TLB entries
10396 		 * also flushes the "current" TLB entries, i.e. servicing the
10397 		 * flush "all" will clear any request to flush "current".
10398 		 */
10399 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
10400 			kvm_vcpu_flush_tlb_all(vcpu);
10401 
10402 		kvm_service_local_tlb_flush_requests(vcpu);
10403 
10404 		/*
10405 		 * Fall back to a "full" guest flush if Hyper-V's precise
10406 		 * flushing fails.  Note, Hyper-V's flushing is per-vCPU, but
10407 		 * the flushes are considered "remote" and not "local" because
10408 		 * the requests can be initiated from other vCPUs.
10409 		 */
10410 		if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu) &&
10411 		    kvm_hv_vcpu_flush_tlb(vcpu))
10412 			kvm_vcpu_flush_tlb_guest(vcpu);
10413 
10414 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
10415 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
10416 			r = 0;
10417 			goto out;
10418 		}
10419 		if (kvm_test_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
10420 			if (is_guest_mode(vcpu))
10421 				kvm_x86_ops.nested_ops->triple_fault(vcpu);
10422 
10423 			if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
10424 				vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
10425 				vcpu->mmio_needed = 0;
10426 				r = 0;
10427 				goto out;
10428 			}
10429 		}
10430 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
10431 			/* Page is swapped out. Do synthetic halt */
10432 			vcpu->arch.apf.halted = true;
10433 			r = 1;
10434 			goto out;
10435 		}
10436 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
10437 			record_steal_time(vcpu);
10438 #ifdef CONFIG_KVM_SMM
10439 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
10440 			process_smi(vcpu);
10441 #endif
10442 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
10443 			process_nmi(vcpu);
10444 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
10445 			kvm_pmu_handle_event(vcpu);
10446 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
10447 			kvm_pmu_deliver_pmi(vcpu);
10448 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
10449 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
10450 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
10451 				     vcpu->arch.ioapic_handled_vectors)) {
10452 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
10453 				vcpu->run->eoi.vector =
10454 						vcpu->arch.pending_ioapic_eoi;
10455 				r = 0;
10456 				goto out;
10457 			}
10458 		}
10459 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
10460 			vcpu_scan_ioapic(vcpu);
10461 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
10462 			vcpu_load_eoi_exitmap(vcpu);
10463 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
10464 			kvm_vcpu_reload_apic_access_page(vcpu);
10465 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
10466 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
10467 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
10468 			vcpu->run->system_event.ndata = 0;
10469 			r = 0;
10470 			goto out;
10471 		}
10472 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
10473 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
10474 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
10475 			vcpu->run->system_event.ndata = 0;
10476 			r = 0;
10477 			goto out;
10478 		}
10479 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
10480 			struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
10481 
10482 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
10483 			vcpu->run->hyperv = hv_vcpu->exit;
10484 			r = 0;
10485 			goto out;
10486 		}
10487 
10488 		/*
10489 		 * KVM_REQ_HV_STIMER has to be processed after
10490 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
10491 		 * depend on the guest clock being up-to-date
10492 		 */
10493 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
10494 			kvm_hv_process_stimers(vcpu);
10495 		if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
10496 			kvm_vcpu_update_apicv(vcpu);
10497 		if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
10498 			kvm_check_async_pf_completion(vcpu);
10499 		if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
10500 			static_call(kvm_x86_msr_filter_changed)(vcpu);
10501 
10502 		if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
10503 			static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
10504 	}
10505 
10506 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
10507 	    kvm_xen_has_interrupt(vcpu)) {
10508 		++vcpu->stat.req_event;
10509 		r = kvm_apic_accept_events(vcpu);
10510 		if (r < 0) {
10511 			r = 0;
10512 			goto out;
10513 		}
10514 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
10515 			r = 1;
10516 			goto out;
10517 		}
10518 
10519 		r = kvm_check_and_inject_events(vcpu, &req_immediate_exit);
10520 		if (r < 0) {
10521 			r = 0;
10522 			goto out;
10523 		}
10524 		if (req_int_win)
10525 			static_call(kvm_x86_enable_irq_window)(vcpu);
10526 
10527 		if (kvm_lapic_enabled(vcpu)) {
10528 			update_cr8_intercept(vcpu);
10529 			kvm_lapic_sync_to_vapic(vcpu);
10530 		}
10531 	}
10532 
10533 	r = kvm_mmu_reload(vcpu);
10534 	if (unlikely(r)) {
10535 		goto cancel_injection;
10536 	}
10537 
10538 	preempt_disable();
10539 
10540 	static_call(kvm_x86_prepare_switch_to_guest)(vcpu);
10541 
10542 	/*
10543 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
10544 	 * IPI are then delayed after guest entry, which ensures that they
10545 	 * result in virtual interrupt delivery.
10546 	 */
10547 	local_irq_disable();
10548 
10549 	/* Store vcpu->apicv_active before vcpu->mode.  */
10550 	smp_store_release(&vcpu->mode, IN_GUEST_MODE);
10551 
10552 	kvm_vcpu_srcu_read_unlock(vcpu);
10553 
10554 	/*
10555 	 * 1) We should set ->mode before checking ->requests.  Please see
10556 	 * the comment in kvm_vcpu_exiting_guest_mode().
10557 	 *
10558 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
10559 	 * pairs with the memory barrier implicit in pi_test_and_set_on
10560 	 * (see vmx_deliver_posted_interrupt).
10561 	 *
10562 	 * 3) This also orders the write to mode from any reads to the page
10563 	 * tables done while the VCPU is running.  Please see the comment
10564 	 * in kvm_flush_remote_tlbs.
10565 	 */
10566 	smp_mb__after_srcu_read_unlock();
10567 
10568 	/*
10569 	 * Process pending posted interrupts to handle the case where the
10570 	 * notification IRQ arrived in the host, or was never sent (because the
10571 	 * target vCPU wasn't running).  Do this regardless of the vCPU's APICv
10572 	 * status, KVM doesn't update assigned devices when APICv is inhibited,
10573 	 * i.e. they can post interrupts even if APICv is temporarily disabled.
10574 	 */
10575 	if (kvm_lapic_enabled(vcpu))
10576 		static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
10577 
10578 	if (kvm_vcpu_exit_request(vcpu)) {
10579 		vcpu->mode = OUTSIDE_GUEST_MODE;
10580 		smp_wmb();
10581 		local_irq_enable();
10582 		preempt_enable();
10583 		kvm_vcpu_srcu_read_lock(vcpu);
10584 		r = 1;
10585 		goto cancel_injection;
10586 	}
10587 
10588 	if (req_immediate_exit) {
10589 		kvm_make_request(KVM_REQ_EVENT, vcpu);
10590 		static_call(kvm_x86_request_immediate_exit)(vcpu);
10591 	}
10592 
10593 	fpregs_assert_state_consistent();
10594 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
10595 		switch_fpu_return();
10596 
10597 	if (vcpu->arch.guest_fpu.xfd_err)
10598 		wrmsrl(MSR_IA32_XFD_ERR, vcpu->arch.guest_fpu.xfd_err);
10599 
10600 	if (unlikely(vcpu->arch.switch_db_regs)) {
10601 		set_debugreg(0, 7);
10602 		set_debugreg(vcpu->arch.eff_db[0], 0);
10603 		set_debugreg(vcpu->arch.eff_db[1], 1);
10604 		set_debugreg(vcpu->arch.eff_db[2], 2);
10605 		set_debugreg(vcpu->arch.eff_db[3], 3);
10606 	} else if (unlikely(hw_breakpoint_active())) {
10607 		set_debugreg(0, 7);
10608 	}
10609 
10610 	guest_timing_enter_irqoff();
10611 
10612 	for (;;) {
10613 		/*
10614 		 * Assert that vCPU vs. VM APICv state is consistent.  An APICv
10615 		 * update must kick and wait for all vCPUs before toggling the
10616 		 * per-VM state, and responsing vCPUs must wait for the update
10617 		 * to complete before servicing KVM_REQ_APICV_UPDATE.
10618 		 */
10619 		WARN_ON_ONCE((kvm_vcpu_apicv_activated(vcpu) != kvm_vcpu_apicv_active(vcpu)) &&
10620 			     (kvm_get_apic_mode(vcpu) != LAPIC_MODE_DISABLED));
10621 
10622 		exit_fastpath = static_call(kvm_x86_vcpu_run)(vcpu);
10623 		if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
10624 			break;
10625 
10626 		if (kvm_lapic_enabled(vcpu))
10627 			static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
10628 
10629 		if (unlikely(kvm_vcpu_exit_request(vcpu))) {
10630 			exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
10631 			break;
10632 		}
10633 	}
10634 
10635 	/*
10636 	 * Do this here before restoring debug registers on the host.  And
10637 	 * since we do this before handling the vmexit, a DR access vmexit
10638 	 * can (a) read the correct value of the debug registers, (b) set
10639 	 * KVM_DEBUGREG_WONT_EXIT again.
10640 	 */
10641 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
10642 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
10643 		static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
10644 		kvm_update_dr0123(vcpu);
10645 		kvm_update_dr7(vcpu);
10646 	}
10647 
10648 	/*
10649 	 * If the guest has used debug registers, at least dr7
10650 	 * will be disabled while returning to the host.
10651 	 * If we don't have active breakpoints in the host, we don't
10652 	 * care about the messed up debug address registers. But if
10653 	 * we have some of them active, restore the old state.
10654 	 */
10655 	if (hw_breakpoint_active())
10656 		hw_breakpoint_restore();
10657 
10658 	vcpu->arch.last_vmentry_cpu = vcpu->cpu;
10659 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
10660 
10661 	vcpu->mode = OUTSIDE_GUEST_MODE;
10662 	smp_wmb();
10663 
10664 	/*
10665 	 * Sync xfd before calling handle_exit_irqoff() which may
10666 	 * rely on the fact that guest_fpu::xfd is up-to-date (e.g.
10667 	 * in #NM irqoff handler).
10668 	 */
10669 	if (vcpu->arch.xfd_no_write_intercept)
10670 		fpu_sync_guest_vmexit_xfd_state();
10671 
10672 	static_call(kvm_x86_handle_exit_irqoff)(vcpu);
10673 
10674 	if (vcpu->arch.guest_fpu.xfd_err)
10675 		wrmsrl(MSR_IA32_XFD_ERR, 0);
10676 
10677 	/*
10678 	 * Consume any pending interrupts, including the possible source of
10679 	 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
10680 	 * An instruction is required after local_irq_enable() to fully unblock
10681 	 * interrupts on processors that implement an interrupt shadow, the
10682 	 * stat.exits increment will do nicely.
10683 	 */
10684 	kvm_before_interrupt(vcpu, KVM_HANDLING_IRQ);
10685 	local_irq_enable();
10686 	++vcpu->stat.exits;
10687 	local_irq_disable();
10688 	kvm_after_interrupt(vcpu);
10689 
10690 	/*
10691 	 * Wait until after servicing IRQs to account guest time so that any
10692 	 * ticks that occurred while running the guest are properly accounted
10693 	 * to the guest.  Waiting until IRQs are enabled degrades the accuracy
10694 	 * of accounting via context tracking, but the loss of accuracy is
10695 	 * acceptable for all known use cases.
10696 	 */
10697 	guest_timing_exit_irqoff();
10698 
10699 	local_irq_enable();
10700 	preempt_enable();
10701 
10702 	kvm_vcpu_srcu_read_lock(vcpu);
10703 
10704 	/*
10705 	 * Profile KVM exit RIPs:
10706 	 */
10707 	if (unlikely(prof_on == KVM_PROFILING)) {
10708 		unsigned long rip = kvm_rip_read(vcpu);
10709 		profile_hit(KVM_PROFILING, (void *)rip);
10710 	}
10711 
10712 	if (unlikely(vcpu->arch.tsc_always_catchup))
10713 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10714 
10715 	if (vcpu->arch.apic_attention)
10716 		kvm_lapic_sync_from_vapic(vcpu);
10717 
10718 	r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
10719 	return r;
10720 
10721 cancel_injection:
10722 	if (req_immediate_exit)
10723 		kvm_make_request(KVM_REQ_EVENT, vcpu);
10724 	static_call(kvm_x86_cancel_injection)(vcpu);
10725 	if (unlikely(vcpu->arch.apic_attention))
10726 		kvm_lapic_sync_from_vapic(vcpu);
10727 out:
10728 	return r;
10729 }
10730 
10731 /* Called within kvm->srcu read side.  */
10732 static inline int vcpu_block(struct kvm_vcpu *vcpu)
10733 {
10734 	bool hv_timer;
10735 
10736 	if (!kvm_arch_vcpu_runnable(vcpu)) {
10737 		/*
10738 		 * Switch to the software timer before halt-polling/blocking as
10739 		 * the guest's timer may be a break event for the vCPU, and the
10740 		 * hypervisor timer runs only when the CPU is in guest mode.
10741 		 * Switch before halt-polling so that KVM recognizes an expired
10742 		 * timer before blocking.
10743 		 */
10744 		hv_timer = kvm_lapic_hv_timer_in_use(vcpu);
10745 		if (hv_timer)
10746 			kvm_lapic_switch_to_sw_timer(vcpu);
10747 
10748 		kvm_vcpu_srcu_read_unlock(vcpu);
10749 		if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
10750 			kvm_vcpu_halt(vcpu);
10751 		else
10752 			kvm_vcpu_block(vcpu);
10753 		kvm_vcpu_srcu_read_lock(vcpu);
10754 
10755 		if (hv_timer)
10756 			kvm_lapic_switch_to_hv_timer(vcpu);
10757 
10758 		/*
10759 		 * If the vCPU is not runnable, a signal or another host event
10760 		 * of some kind is pending; service it without changing the
10761 		 * vCPU's activity state.
10762 		 */
10763 		if (!kvm_arch_vcpu_runnable(vcpu))
10764 			return 1;
10765 	}
10766 
10767 	/*
10768 	 * Evaluate nested events before exiting the halted state.  This allows
10769 	 * the halt state to be recorded properly in the VMCS12's activity
10770 	 * state field (AMD does not have a similar field and a VM-Exit always
10771 	 * causes a spurious wakeup from HLT).
10772 	 */
10773 	if (is_guest_mode(vcpu)) {
10774 		if (kvm_check_nested_events(vcpu) < 0)
10775 			return 0;
10776 	}
10777 
10778 	if (kvm_apic_accept_events(vcpu) < 0)
10779 		return 0;
10780 	switch(vcpu->arch.mp_state) {
10781 	case KVM_MP_STATE_HALTED:
10782 	case KVM_MP_STATE_AP_RESET_HOLD:
10783 		vcpu->arch.pv.pv_unhalted = false;
10784 		vcpu->arch.mp_state =
10785 			KVM_MP_STATE_RUNNABLE;
10786 		fallthrough;
10787 	case KVM_MP_STATE_RUNNABLE:
10788 		vcpu->arch.apf.halted = false;
10789 		break;
10790 	case KVM_MP_STATE_INIT_RECEIVED:
10791 		break;
10792 	default:
10793 		WARN_ON_ONCE(1);
10794 		break;
10795 	}
10796 	return 1;
10797 }
10798 
10799 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
10800 {
10801 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
10802 		!vcpu->arch.apf.halted);
10803 }
10804 
10805 /* Called within kvm->srcu read side.  */
10806 static int vcpu_run(struct kvm_vcpu *vcpu)
10807 {
10808 	int r;
10809 
10810 	vcpu->arch.l1tf_flush_l1d = true;
10811 
10812 	for (;;) {
10813 		/*
10814 		 * If another guest vCPU requests a PV TLB flush in the middle
10815 		 * of instruction emulation, the rest of the emulation could
10816 		 * use a stale page translation. Assume that any code after
10817 		 * this point can start executing an instruction.
10818 		 */
10819 		vcpu->arch.at_instruction_boundary = false;
10820 		if (kvm_vcpu_running(vcpu)) {
10821 			r = vcpu_enter_guest(vcpu);
10822 		} else {
10823 			r = vcpu_block(vcpu);
10824 		}
10825 
10826 		if (r <= 0)
10827 			break;
10828 
10829 		kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
10830 		if (kvm_xen_has_pending_events(vcpu))
10831 			kvm_xen_inject_pending_events(vcpu);
10832 
10833 		if (kvm_cpu_has_pending_timer(vcpu))
10834 			kvm_inject_pending_timer_irqs(vcpu);
10835 
10836 		if (dm_request_for_irq_injection(vcpu) &&
10837 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
10838 			r = 0;
10839 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
10840 			++vcpu->stat.request_irq_exits;
10841 			break;
10842 		}
10843 
10844 		if (__xfer_to_guest_mode_work_pending()) {
10845 			kvm_vcpu_srcu_read_unlock(vcpu);
10846 			r = xfer_to_guest_mode_handle_work(vcpu);
10847 			kvm_vcpu_srcu_read_lock(vcpu);
10848 			if (r)
10849 				return r;
10850 		}
10851 	}
10852 
10853 	return r;
10854 }
10855 
10856 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
10857 {
10858 	return kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
10859 }
10860 
10861 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
10862 {
10863 	BUG_ON(!vcpu->arch.pio.count);
10864 
10865 	return complete_emulated_io(vcpu);
10866 }
10867 
10868 /*
10869  * Implements the following, as a state machine:
10870  *
10871  * read:
10872  *   for each fragment
10873  *     for each mmio piece in the fragment
10874  *       write gpa, len
10875  *       exit
10876  *       copy data
10877  *   execute insn
10878  *
10879  * write:
10880  *   for each fragment
10881  *     for each mmio piece in the fragment
10882  *       write gpa, len
10883  *       copy data
10884  *       exit
10885  */
10886 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
10887 {
10888 	struct kvm_run *run = vcpu->run;
10889 	struct kvm_mmio_fragment *frag;
10890 	unsigned len;
10891 
10892 	BUG_ON(!vcpu->mmio_needed);
10893 
10894 	/* Complete previous fragment */
10895 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
10896 	len = min(8u, frag->len);
10897 	if (!vcpu->mmio_is_write)
10898 		memcpy(frag->data, run->mmio.data, len);
10899 
10900 	if (frag->len <= 8) {
10901 		/* Switch to the next fragment. */
10902 		frag++;
10903 		vcpu->mmio_cur_fragment++;
10904 	} else {
10905 		/* Go forward to the next mmio piece. */
10906 		frag->data += len;
10907 		frag->gpa += len;
10908 		frag->len -= len;
10909 	}
10910 
10911 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
10912 		vcpu->mmio_needed = 0;
10913 
10914 		/* FIXME: return into emulator if single-stepping.  */
10915 		if (vcpu->mmio_is_write)
10916 			return 1;
10917 		vcpu->mmio_read_completed = 1;
10918 		return complete_emulated_io(vcpu);
10919 	}
10920 
10921 	run->exit_reason = KVM_EXIT_MMIO;
10922 	run->mmio.phys_addr = frag->gpa;
10923 	if (vcpu->mmio_is_write)
10924 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
10925 	run->mmio.len = min(8u, frag->len);
10926 	run->mmio.is_write = vcpu->mmio_is_write;
10927 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
10928 	return 0;
10929 }
10930 
10931 /* Swap (qemu) user FPU context for the guest FPU context. */
10932 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
10933 {
10934 	/* Exclude PKRU, it's restored separately immediately after VM-Exit. */
10935 	fpu_swap_kvm_fpstate(&vcpu->arch.guest_fpu, true);
10936 	trace_kvm_fpu(1);
10937 }
10938 
10939 /* When vcpu_run ends, restore user space FPU context. */
10940 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
10941 {
10942 	fpu_swap_kvm_fpstate(&vcpu->arch.guest_fpu, false);
10943 	++vcpu->stat.fpu_reload;
10944 	trace_kvm_fpu(0);
10945 }
10946 
10947 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
10948 {
10949 	struct kvm_queued_exception *ex = &vcpu->arch.exception;
10950 	struct kvm_run *kvm_run = vcpu->run;
10951 	int r;
10952 
10953 	vcpu_load(vcpu);
10954 	kvm_sigset_activate(vcpu);
10955 	kvm_run->flags = 0;
10956 	kvm_load_guest_fpu(vcpu);
10957 
10958 	kvm_vcpu_srcu_read_lock(vcpu);
10959 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
10960 		if (kvm_run->immediate_exit) {
10961 			r = -EINTR;
10962 			goto out;
10963 		}
10964 		/*
10965 		 * It should be impossible for the hypervisor timer to be in
10966 		 * use before KVM has ever run the vCPU.
10967 		 */
10968 		WARN_ON_ONCE(kvm_lapic_hv_timer_in_use(vcpu));
10969 
10970 		kvm_vcpu_srcu_read_unlock(vcpu);
10971 		kvm_vcpu_block(vcpu);
10972 		kvm_vcpu_srcu_read_lock(vcpu);
10973 
10974 		if (kvm_apic_accept_events(vcpu) < 0) {
10975 			r = 0;
10976 			goto out;
10977 		}
10978 		r = -EAGAIN;
10979 		if (signal_pending(current)) {
10980 			r = -EINTR;
10981 			kvm_run->exit_reason = KVM_EXIT_INTR;
10982 			++vcpu->stat.signal_exits;
10983 		}
10984 		goto out;
10985 	}
10986 
10987 	if ((kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) ||
10988 	    (kvm_run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)) {
10989 		r = -EINVAL;
10990 		goto out;
10991 	}
10992 
10993 	if (kvm_run->kvm_dirty_regs) {
10994 		r = sync_regs(vcpu);
10995 		if (r != 0)
10996 			goto out;
10997 	}
10998 
10999 	/* re-sync apic's tpr */
11000 	if (!lapic_in_kernel(vcpu)) {
11001 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
11002 			r = -EINVAL;
11003 			goto out;
11004 		}
11005 	}
11006 
11007 	/*
11008 	 * If userspace set a pending exception and L2 is active, convert it to
11009 	 * a pending VM-Exit if L1 wants to intercept the exception.
11010 	 */
11011 	if (vcpu->arch.exception_from_userspace && is_guest_mode(vcpu) &&
11012 	    kvm_x86_ops.nested_ops->is_exception_vmexit(vcpu, ex->vector,
11013 							ex->error_code)) {
11014 		kvm_queue_exception_vmexit(vcpu, ex->vector,
11015 					   ex->has_error_code, ex->error_code,
11016 					   ex->has_payload, ex->payload);
11017 		ex->injected = false;
11018 		ex->pending = false;
11019 	}
11020 	vcpu->arch.exception_from_userspace = false;
11021 
11022 	if (unlikely(vcpu->arch.complete_userspace_io)) {
11023 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
11024 		vcpu->arch.complete_userspace_io = NULL;
11025 		r = cui(vcpu);
11026 		if (r <= 0)
11027 			goto out;
11028 	} else {
11029 		WARN_ON_ONCE(vcpu->arch.pio.count);
11030 		WARN_ON_ONCE(vcpu->mmio_needed);
11031 	}
11032 
11033 	if (kvm_run->immediate_exit) {
11034 		r = -EINTR;
11035 		goto out;
11036 	}
11037 
11038 	r = static_call(kvm_x86_vcpu_pre_run)(vcpu);
11039 	if (r <= 0)
11040 		goto out;
11041 
11042 	r = vcpu_run(vcpu);
11043 
11044 out:
11045 	kvm_put_guest_fpu(vcpu);
11046 	if (kvm_run->kvm_valid_regs)
11047 		store_regs(vcpu);
11048 	post_kvm_run_save(vcpu);
11049 	kvm_vcpu_srcu_read_unlock(vcpu);
11050 
11051 	kvm_sigset_deactivate(vcpu);
11052 	vcpu_put(vcpu);
11053 	return r;
11054 }
11055 
11056 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
11057 {
11058 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
11059 		/*
11060 		 * We are here if userspace calls get_regs() in the middle of
11061 		 * instruction emulation. Registers state needs to be copied
11062 		 * back from emulation context to vcpu. Userspace shouldn't do
11063 		 * that usually, but some bad designed PV devices (vmware
11064 		 * backdoor interface) need this to work
11065 		 */
11066 		emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
11067 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
11068 	}
11069 	regs->rax = kvm_rax_read(vcpu);
11070 	regs->rbx = kvm_rbx_read(vcpu);
11071 	regs->rcx = kvm_rcx_read(vcpu);
11072 	regs->rdx = kvm_rdx_read(vcpu);
11073 	regs->rsi = kvm_rsi_read(vcpu);
11074 	regs->rdi = kvm_rdi_read(vcpu);
11075 	regs->rsp = kvm_rsp_read(vcpu);
11076 	regs->rbp = kvm_rbp_read(vcpu);
11077 #ifdef CONFIG_X86_64
11078 	regs->r8 = kvm_r8_read(vcpu);
11079 	regs->r9 = kvm_r9_read(vcpu);
11080 	regs->r10 = kvm_r10_read(vcpu);
11081 	regs->r11 = kvm_r11_read(vcpu);
11082 	regs->r12 = kvm_r12_read(vcpu);
11083 	regs->r13 = kvm_r13_read(vcpu);
11084 	regs->r14 = kvm_r14_read(vcpu);
11085 	regs->r15 = kvm_r15_read(vcpu);
11086 #endif
11087 
11088 	regs->rip = kvm_rip_read(vcpu);
11089 	regs->rflags = kvm_get_rflags(vcpu);
11090 }
11091 
11092 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
11093 {
11094 	vcpu_load(vcpu);
11095 	__get_regs(vcpu, regs);
11096 	vcpu_put(vcpu);
11097 	return 0;
11098 }
11099 
11100 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
11101 {
11102 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
11103 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
11104 
11105 	kvm_rax_write(vcpu, regs->rax);
11106 	kvm_rbx_write(vcpu, regs->rbx);
11107 	kvm_rcx_write(vcpu, regs->rcx);
11108 	kvm_rdx_write(vcpu, regs->rdx);
11109 	kvm_rsi_write(vcpu, regs->rsi);
11110 	kvm_rdi_write(vcpu, regs->rdi);
11111 	kvm_rsp_write(vcpu, regs->rsp);
11112 	kvm_rbp_write(vcpu, regs->rbp);
11113 #ifdef CONFIG_X86_64
11114 	kvm_r8_write(vcpu, regs->r8);
11115 	kvm_r9_write(vcpu, regs->r9);
11116 	kvm_r10_write(vcpu, regs->r10);
11117 	kvm_r11_write(vcpu, regs->r11);
11118 	kvm_r12_write(vcpu, regs->r12);
11119 	kvm_r13_write(vcpu, regs->r13);
11120 	kvm_r14_write(vcpu, regs->r14);
11121 	kvm_r15_write(vcpu, regs->r15);
11122 #endif
11123 
11124 	kvm_rip_write(vcpu, regs->rip);
11125 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
11126 
11127 	vcpu->arch.exception.pending = false;
11128 	vcpu->arch.exception_vmexit.pending = false;
11129 
11130 	kvm_make_request(KVM_REQ_EVENT, vcpu);
11131 }
11132 
11133 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
11134 {
11135 	vcpu_load(vcpu);
11136 	__set_regs(vcpu, regs);
11137 	vcpu_put(vcpu);
11138 	return 0;
11139 }
11140 
11141 static void __get_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
11142 {
11143 	struct desc_ptr dt;
11144 
11145 	if (vcpu->arch.guest_state_protected)
11146 		goto skip_protected_regs;
11147 
11148 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
11149 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
11150 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
11151 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
11152 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
11153 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
11154 
11155 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
11156 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
11157 
11158 	static_call(kvm_x86_get_idt)(vcpu, &dt);
11159 	sregs->idt.limit = dt.size;
11160 	sregs->idt.base = dt.address;
11161 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
11162 	sregs->gdt.limit = dt.size;
11163 	sregs->gdt.base = dt.address;
11164 
11165 	sregs->cr2 = vcpu->arch.cr2;
11166 	sregs->cr3 = kvm_read_cr3(vcpu);
11167 
11168 skip_protected_regs:
11169 	sregs->cr0 = kvm_read_cr0(vcpu);
11170 	sregs->cr4 = kvm_read_cr4(vcpu);
11171 	sregs->cr8 = kvm_get_cr8(vcpu);
11172 	sregs->efer = vcpu->arch.efer;
11173 	sregs->apic_base = kvm_get_apic_base(vcpu);
11174 }
11175 
11176 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
11177 {
11178 	__get_sregs_common(vcpu, sregs);
11179 
11180 	if (vcpu->arch.guest_state_protected)
11181 		return;
11182 
11183 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
11184 		set_bit(vcpu->arch.interrupt.nr,
11185 			(unsigned long *)sregs->interrupt_bitmap);
11186 }
11187 
11188 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
11189 {
11190 	int i;
11191 
11192 	__get_sregs_common(vcpu, (struct kvm_sregs *)sregs2);
11193 
11194 	if (vcpu->arch.guest_state_protected)
11195 		return;
11196 
11197 	if (is_pae_paging(vcpu)) {
11198 		for (i = 0 ; i < 4 ; i++)
11199 			sregs2->pdptrs[i] = kvm_pdptr_read(vcpu, i);
11200 		sregs2->flags |= KVM_SREGS2_FLAGS_PDPTRS_VALID;
11201 	}
11202 }
11203 
11204 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
11205 				  struct kvm_sregs *sregs)
11206 {
11207 	vcpu_load(vcpu);
11208 	__get_sregs(vcpu, sregs);
11209 	vcpu_put(vcpu);
11210 	return 0;
11211 }
11212 
11213 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
11214 				    struct kvm_mp_state *mp_state)
11215 {
11216 	int r;
11217 
11218 	vcpu_load(vcpu);
11219 	if (kvm_mpx_supported())
11220 		kvm_load_guest_fpu(vcpu);
11221 
11222 	r = kvm_apic_accept_events(vcpu);
11223 	if (r < 0)
11224 		goto out;
11225 	r = 0;
11226 
11227 	if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
11228 	     vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
11229 	    vcpu->arch.pv.pv_unhalted)
11230 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
11231 	else
11232 		mp_state->mp_state = vcpu->arch.mp_state;
11233 
11234 out:
11235 	if (kvm_mpx_supported())
11236 		kvm_put_guest_fpu(vcpu);
11237 	vcpu_put(vcpu);
11238 	return r;
11239 }
11240 
11241 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
11242 				    struct kvm_mp_state *mp_state)
11243 {
11244 	int ret = -EINVAL;
11245 
11246 	vcpu_load(vcpu);
11247 
11248 	switch (mp_state->mp_state) {
11249 	case KVM_MP_STATE_UNINITIALIZED:
11250 	case KVM_MP_STATE_HALTED:
11251 	case KVM_MP_STATE_AP_RESET_HOLD:
11252 	case KVM_MP_STATE_INIT_RECEIVED:
11253 	case KVM_MP_STATE_SIPI_RECEIVED:
11254 		if (!lapic_in_kernel(vcpu))
11255 			goto out;
11256 		break;
11257 
11258 	case KVM_MP_STATE_RUNNABLE:
11259 		break;
11260 
11261 	default:
11262 		goto out;
11263 	}
11264 
11265 	/*
11266 	 * Pending INITs are reported using KVM_SET_VCPU_EVENTS, disallow
11267 	 * forcing the guest into INIT/SIPI if those events are supposed to be
11268 	 * blocked.  KVM prioritizes SMI over INIT, so reject INIT/SIPI state
11269 	 * if an SMI is pending as well.
11270 	 */
11271 	if ((!kvm_apic_init_sipi_allowed(vcpu) || vcpu->arch.smi_pending) &&
11272 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
11273 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
11274 		goto out;
11275 
11276 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
11277 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
11278 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
11279 	} else
11280 		vcpu->arch.mp_state = mp_state->mp_state;
11281 	kvm_make_request(KVM_REQ_EVENT, vcpu);
11282 
11283 	ret = 0;
11284 out:
11285 	vcpu_put(vcpu);
11286 	return ret;
11287 }
11288 
11289 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
11290 		    int reason, bool has_error_code, u32 error_code)
11291 {
11292 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
11293 	int ret;
11294 
11295 	init_emulate_ctxt(vcpu);
11296 
11297 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
11298 				   has_error_code, error_code);
11299 	if (ret) {
11300 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
11301 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
11302 		vcpu->run->internal.ndata = 0;
11303 		return 0;
11304 	}
11305 
11306 	kvm_rip_write(vcpu, ctxt->eip);
11307 	kvm_set_rflags(vcpu, ctxt->eflags);
11308 	return 1;
11309 }
11310 EXPORT_SYMBOL_GPL(kvm_task_switch);
11311 
11312 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
11313 {
11314 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
11315 		/*
11316 		 * When EFER.LME and CR0.PG are set, the processor is in
11317 		 * 64-bit mode (though maybe in a 32-bit code segment).
11318 		 * CR4.PAE and EFER.LMA must be set.
11319 		 */
11320 		if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
11321 			return false;
11322 		if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
11323 			return false;
11324 	} else {
11325 		/*
11326 		 * Not in 64-bit mode: EFER.LMA is clear and the code
11327 		 * segment cannot be 64-bit.
11328 		 */
11329 		if (sregs->efer & EFER_LMA || sregs->cs.l)
11330 			return false;
11331 	}
11332 
11333 	return kvm_is_valid_cr4(vcpu, sregs->cr4);
11334 }
11335 
11336 static int __set_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs,
11337 		int *mmu_reset_needed, bool update_pdptrs)
11338 {
11339 	struct msr_data apic_base_msr;
11340 	int idx;
11341 	struct desc_ptr dt;
11342 
11343 	if (!kvm_is_valid_sregs(vcpu, sregs))
11344 		return -EINVAL;
11345 
11346 	apic_base_msr.data = sregs->apic_base;
11347 	apic_base_msr.host_initiated = true;
11348 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
11349 		return -EINVAL;
11350 
11351 	if (vcpu->arch.guest_state_protected)
11352 		return 0;
11353 
11354 	dt.size = sregs->idt.limit;
11355 	dt.address = sregs->idt.base;
11356 	static_call(kvm_x86_set_idt)(vcpu, &dt);
11357 	dt.size = sregs->gdt.limit;
11358 	dt.address = sregs->gdt.base;
11359 	static_call(kvm_x86_set_gdt)(vcpu, &dt);
11360 
11361 	vcpu->arch.cr2 = sregs->cr2;
11362 	*mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
11363 	vcpu->arch.cr3 = sregs->cr3;
11364 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
11365 	static_call_cond(kvm_x86_post_set_cr3)(vcpu, sregs->cr3);
11366 
11367 	kvm_set_cr8(vcpu, sregs->cr8);
11368 
11369 	*mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
11370 	static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
11371 
11372 	*mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
11373 	static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
11374 	vcpu->arch.cr0 = sregs->cr0;
11375 
11376 	*mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
11377 	static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
11378 
11379 	if (update_pdptrs) {
11380 		idx = srcu_read_lock(&vcpu->kvm->srcu);
11381 		if (is_pae_paging(vcpu)) {
11382 			load_pdptrs(vcpu, kvm_read_cr3(vcpu));
11383 			*mmu_reset_needed = 1;
11384 		}
11385 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
11386 	}
11387 
11388 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
11389 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
11390 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
11391 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
11392 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
11393 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
11394 
11395 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
11396 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
11397 
11398 	update_cr8_intercept(vcpu);
11399 
11400 	/* Older userspace won't unhalt the vcpu on reset. */
11401 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
11402 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
11403 	    !is_protmode(vcpu))
11404 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11405 
11406 	return 0;
11407 }
11408 
11409 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
11410 {
11411 	int pending_vec, max_bits;
11412 	int mmu_reset_needed = 0;
11413 	int ret = __set_sregs_common(vcpu, sregs, &mmu_reset_needed, true);
11414 
11415 	if (ret)
11416 		return ret;
11417 
11418 	if (mmu_reset_needed)
11419 		kvm_mmu_reset_context(vcpu);
11420 
11421 	max_bits = KVM_NR_INTERRUPTS;
11422 	pending_vec = find_first_bit(
11423 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
11424 
11425 	if (pending_vec < max_bits) {
11426 		kvm_queue_interrupt(vcpu, pending_vec, false);
11427 		pr_debug("Set back pending irq %d\n", pending_vec);
11428 		kvm_make_request(KVM_REQ_EVENT, vcpu);
11429 	}
11430 	return 0;
11431 }
11432 
11433 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
11434 {
11435 	int mmu_reset_needed = 0;
11436 	bool valid_pdptrs = sregs2->flags & KVM_SREGS2_FLAGS_PDPTRS_VALID;
11437 	bool pae = (sregs2->cr0 & X86_CR0_PG) && (sregs2->cr4 & X86_CR4_PAE) &&
11438 		!(sregs2->efer & EFER_LMA);
11439 	int i, ret;
11440 
11441 	if (sregs2->flags & ~KVM_SREGS2_FLAGS_PDPTRS_VALID)
11442 		return -EINVAL;
11443 
11444 	if (valid_pdptrs && (!pae || vcpu->arch.guest_state_protected))
11445 		return -EINVAL;
11446 
11447 	ret = __set_sregs_common(vcpu, (struct kvm_sregs *)sregs2,
11448 				 &mmu_reset_needed, !valid_pdptrs);
11449 	if (ret)
11450 		return ret;
11451 
11452 	if (valid_pdptrs) {
11453 		for (i = 0; i < 4 ; i++)
11454 			kvm_pdptr_write(vcpu, i, sregs2->pdptrs[i]);
11455 
11456 		kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
11457 		mmu_reset_needed = 1;
11458 		vcpu->arch.pdptrs_from_userspace = true;
11459 	}
11460 	if (mmu_reset_needed)
11461 		kvm_mmu_reset_context(vcpu);
11462 	return 0;
11463 }
11464 
11465 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
11466 				  struct kvm_sregs *sregs)
11467 {
11468 	int ret;
11469 
11470 	vcpu_load(vcpu);
11471 	ret = __set_sregs(vcpu, sregs);
11472 	vcpu_put(vcpu);
11473 	return ret;
11474 }
11475 
11476 static void kvm_arch_vcpu_guestdbg_update_apicv_inhibit(struct kvm *kvm)
11477 {
11478 	bool set = false;
11479 	struct kvm_vcpu *vcpu;
11480 	unsigned long i;
11481 
11482 	if (!enable_apicv)
11483 		return;
11484 
11485 	down_write(&kvm->arch.apicv_update_lock);
11486 
11487 	kvm_for_each_vcpu(i, vcpu, kvm) {
11488 		if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ) {
11489 			set = true;
11490 			break;
11491 		}
11492 	}
11493 	__kvm_set_or_clear_apicv_inhibit(kvm, APICV_INHIBIT_REASON_BLOCKIRQ, set);
11494 	up_write(&kvm->arch.apicv_update_lock);
11495 }
11496 
11497 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
11498 					struct kvm_guest_debug *dbg)
11499 {
11500 	unsigned long rflags;
11501 	int i, r;
11502 
11503 	if (vcpu->arch.guest_state_protected)
11504 		return -EINVAL;
11505 
11506 	vcpu_load(vcpu);
11507 
11508 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
11509 		r = -EBUSY;
11510 		if (kvm_is_exception_pending(vcpu))
11511 			goto out;
11512 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
11513 			kvm_queue_exception(vcpu, DB_VECTOR);
11514 		else
11515 			kvm_queue_exception(vcpu, BP_VECTOR);
11516 	}
11517 
11518 	/*
11519 	 * Read rflags as long as potentially injected trace flags are still
11520 	 * filtered out.
11521 	 */
11522 	rflags = kvm_get_rflags(vcpu);
11523 
11524 	vcpu->guest_debug = dbg->control;
11525 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
11526 		vcpu->guest_debug = 0;
11527 
11528 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
11529 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
11530 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
11531 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
11532 	} else {
11533 		for (i = 0; i < KVM_NR_DB_REGS; i++)
11534 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
11535 	}
11536 	kvm_update_dr7(vcpu);
11537 
11538 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11539 		vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
11540 
11541 	/*
11542 	 * Trigger an rflags update that will inject or remove the trace
11543 	 * flags.
11544 	 */
11545 	kvm_set_rflags(vcpu, rflags);
11546 
11547 	static_call(kvm_x86_update_exception_bitmap)(vcpu);
11548 
11549 	kvm_arch_vcpu_guestdbg_update_apicv_inhibit(vcpu->kvm);
11550 
11551 	r = 0;
11552 
11553 out:
11554 	vcpu_put(vcpu);
11555 	return r;
11556 }
11557 
11558 /*
11559  * Translate a guest virtual address to a guest physical address.
11560  */
11561 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
11562 				    struct kvm_translation *tr)
11563 {
11564 	unsigned long vaddr = tr->linear_address;
11565 	gpa_t gpa;
11566 	int idx;
11567 
11568 	vcpu_load(vcpu);
11569 
11570 	idx = srcu_read_lock(&vcpu->kvm->srcu);
11571 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
11572 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
11573 	tr->physical_address = gpa;
11574 	tr->valid = gpa != INVALID_GPA;
11575 	tr->writeable = 1;
11576 	tr->usermode = 0;
11577 
11578 	vcpu_put(vcpu);
11579 	return 0;
11580 }
11581 
11582 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
11583 {
11584 	struct fxregs_state *fxsave;
11585 
11586 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
11587 		return 0;
11588 
11589 	vcpu_load(vcpu);
11590 
11591 	fxsave = &vcpu->arch.guest_fpu.fpstate->regs.fxsave;
11592 	memcpy(fpu->fpr, fxsave->st_space, 128);
11593 	fpu->fcw = fxsave->cwd;
11594 	fpu->fsw = fxsave->swd;
11595 	fpu->ftwx = fxsave->twd;
11596 	fpu->last_opcode = fxsave->fop;
11597 	fpu->last_ip = fxsave->rip;
11598 	fpu->last_dp = fxsave->rdp;
11599 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
11600 
11601 	vcpu_put(vcpu);
11602 	return 0;
11603 }
11604 
11605 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
11606 {
11607 	struct fxregs_state *fxsave;
11608 
11609 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
11610 		return 0;
11611 
11612 	vcpu_load(vcpu);
11613 
11614 	fxsave = &vcpu->arch.guest_fpu.fpstate->regs.fxsave;
11615 
11616 	memcpy(fxsave->st_space, fpu->fpr, 128);
11617 	fxsave->cwd = fpu->fcw;
11618 	fxsave->swd = fpu->fsw;
11619 	fxsave->twd = fpu->ftwx;
11620 	fxsave->fop = fpu->last_opcode;
11621 	fxsave->rip = fpu->last_ip;
11622 	fxsave->rdp = fpu->last_dp;
11623 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
11624 
11625 	vcpu_put(vcpu);
11626 	return 0;
11627 }
11628 
11629 static void store_regs(struct kvm_vcpu *vcpu)
11630 {
11631 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
11632 
11633 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
11634 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
11635 
11636 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
11637 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
11638 
11639 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
11640 		kvm_vcpu_ioctl_x86_get_vcpu_events(
11641 				vcpu, &vcpu->run->s.regs.events);
11642 }
11643 
11644 static int sync_regs(struct kvm_vcpu *vcpu)
11645 {
11646 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
11647 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
11648 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
11649 	}
11650 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
11651 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
11652 			return -EINVAL;
11653 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
11654 	}
11655 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
11656 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
11657 				vcpu, &vcpu->run->s.regs.events))
11658 			return -EINVAL;
11659 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
11660 	}
11661 
11662 	return 0;
11663 }
11664 
11665 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
11666 {
11667 	if (kvm_check_tsc_unstable() && kvm->created_vcpus)
11668 		pr_warn_once("SMP vm created on host with unstable TSC; "
11669 			     "guest TSC will not be reliable\n");
11670 
11671 	if (!kvm->arch.max_vcpu_ids)
11672 		kvm->arch.max_vcpu_ids = KVM_MAX_VCPU_IDS;
11673 
11674 	if (id >= kvm->arch.max_vcpu_ids)
11675 		return -EINVAL;
11676 
11677 	return static_call(kvm_x86_vcpu_precreate)(kvm);
11678 }
11679 
11680 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
11681 {
11682 	struct page *page;
11683 	int r;
11684 
11685 	vcpu->arch.last_vmentry_cpu = -1;
11686 	vcpu->arch.regs_avail = ~0;
11687 	vcpu->arch.regs_dirty = ~0;
11688 
11689 	kvm_gpc_init(&vcpu->arch.pv_time, vcpu->kvm, vcpu, KVM_HOST_USES_PFN);
11690 
11691 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
11692 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11693 	else
11694 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
11695 
11696 	r = kvm_mmu_create(vcpu);
11697 	if (r < 0)
11698 		return r;
11699 
11700 	if (irqchip_in_kernel(vcpu->kvm)) {
11701 		r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
11702 		if (r < 0)
11703 			goto fail_mmu_destroy;
11704 
11705 		/*
11706 		 * Defer evaluating inhibits until the vCPU is first run, as
11707 		 * this vCPU will not get notified of any changes until this
11708 		 * vCPU is visible to other vCPUs (marked online and added to
11709 		 * the set of vCPUs).  Opportunistically mark APICv active as
11710 		 * VMX in particularly is highly unlikely to have inhibits.
11711 		 * Ignore the current per-VM APICv state so that vCPU creation
11712 		 * is guaranteed to run with a deterministic value, the request
11713 		 * will ensure the vCPU gets the correct state before VM-Entry.
11714 		 */
11715 		if (enable_apicv) {
11716 			vcpu->arch.apic->apicv_active = true;
11717 			kvm_make_request(KVM_REQ_APICV_UPDATE, vcpu);
11718 		}
11719 	} else
11720 		static_branch_inc(&kvm_has_noapic_vcpu);
11721 
11722 	r = -ENOMEM;
11723 
11724 	page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
11725 	if (!page)
11726 		goto fail_free_lapic;
11727 	vcpu->arch.pio_data = page_address(page);
11728 
11729 	vcpu->arch.mce_banks = kcalloc(KVM_MAX_MCE_BANKS * 4, sizeof(u64),
11730 				       GFP_KERNEL_ACCOUNT);
11731 	vcpu->arch.mci_ctl2_banks = kcalloc(KVM_MAX_MCE_BANKS, sizeof(u64),
11732 					    GFP_KERNEL_ACCOUNT);
11733 	if (!vcpu->arch.mce_banks || !vcpu->arch.mci_ctl2_banks)
11734 		goto fail_free_mce_banks;
11735 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
11736 
11737 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
11738 				GFP_KERNEL_ACCOUNT))
11739 		goto fail_free_mce_banks;
11740 
11741 	if (!alloc_emulate_ctxt(vcpu))
11742 		goto free_wbinvd_dirty_mask;
11743 
11744 	if (!fpu_alloc_guest_fpstate(&vcpu->arch.guest_fpu)) {
11745 		pr_err("failed to allocate vcpu's fpu\n");
11746 		goto free_emulate_ctxt;
11747 	}
11748 
11749 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
11750 	vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
11751 
11752 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
11753 
11754 	kvm_async_pf_hash_reset(vcpu);
11755 
11756 	vcpu->arch.perf_capabilities = kvm_caps.supported_perf_cap;
11757 	kvm_pmu_init(vcpu);
11758 
11759 	vcpu->arch.pending_external_vector = -1;
11760 	vcpu->arch.preempted_in_kernel = false;
11761 
11762 #if IS_ENABLED(CONFIG_HYPERV)
11763 	vcpu->arch.hv_root_tdp = INVALID_PAGE;
11764 #endif
11765 
11766 	r = static_call(kvm_x86_vcpu_create)(vcpu);
11767 	if (r)
11768 		goto free_guest_fpu;
11769 
11770 	vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
11771 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
11772 	kvm_xen_init_vcpu(vcpu);
11773 	kvm_vcpu_mtrr_init(vcpu);
11774 	vcpu_load(vcpu);
11775 	kvm_set_tsc_khz(vcpu, vcpu->kvm->arch.default_tsc_khz);
11776 	kvm_vcpu_reset(vcpu, false);
11777 	kvm_init_mmu(vcpu);
11778 	vcpu_put(vcpu);
11779 	return 0;
11780 
11781 free_guest_fpu:
11782 	fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
11783 free_emulate_ctxt:
11784 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
11785 free_wbinvd_dirty_mask:
11786 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
11787 fail_free_mce_banks:
11788 	kfree(vcpu->arch.mce_banks);
11789 	kfree(vcpu->arch.mci_ctl2_banks);
11790 	free_page((unsigned long)vcpu->arch.pio_data);
11791 fail_free_lapic:
11792 	kvm_free_lapic(vcpu);
11793 fail_mmu_destroy:
11794 	kvm_mmu_destroy(vcpu);
11795 	return r;
11796 }
11797 
11798 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
11799 {
11800 	struct kvm *kvm = vcpu->kvm;
11801 
11802 	if (mutex_lock_killable(&vcpu->mutex))
11803 		return;
11804 	vcpu_load(vcpu);
11805 	kvm_synchronize_tsc(vcpu, 0);
11806 	vcpu_put(vcpu);
11807 
11808 	/* poll control enabled by default */
11809 	vcpu->arch.msr_kvm_poll_control = 1;
11810 
11811 	mutex_unlock(&vcpu->mutex);
11812 
11813 	if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
11814 		schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
11815 						KVMCLOCK_SYNC_PERIOD);
11816 }
11817 
11818 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
11819 {
11820 	int idx;
11821 
11822 	kvmclock_reset(vcpu);
11823 
11824 	static_call(kvm_x86_vcpu_free)(vcpu);
11825 
11826 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
11827 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
11828 	fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
11829 
11830 	kvm_xen_destroy_vcpu(vcpu);
11831 	kvm_hv_vcpu_uninit(vcpu);
11832 	kvm_pmu_destroy(vcpu);
11833 	kfree(vcpu->arch.mce_banks);
11834 	kfree(vcpu->arch.mci_ctl2_banks);
11835 	kvm_free_lapic(vcpu);
11836 	idx = srcu_read_lock(&vcpu->kvm->srcu);
11837 	kvm_mmu_destroy(vcpu);
11838 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
11839 	free_page((unsigned long)vcpu->arch.pio_data);
11840 	kvfree(vcpu->arch.cpuid_entries);
11841 	if (!lapic_in_kernel(vcpu))
11842 		static_branch_dec(&kvm_has_noapic_vcpu);
11843 }
11844 
11845 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
11846 {
11847 	struct kvm_cpuid_entry2 *cpuid_0x1;
11848 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
11849 	unsigned long new_cr0;
11850 
11851 	/*
11852 	 * Several of the "set" flows, e.g. ->set_cr0(), read other registers
11853 	 * to handle side effects.  RESET emulation hits those flows and relies
11854 	 * on emulated/virtualized registers, including those that are loaded
11855 	 * into hardware, to be zeroed at vCPU creation.  Use CRs as a sentinel
11856 	 * to detect improper or missing initialization.
11857 	 */
11858 	WARN_ON_ONCE(!init_event &&
11859 		     (old_cr0 || kvm_read_cr3(vcpu) || kvm_read_cr4(vcpu)));
11860 
11861 	/*
11862 	 * SVM doesn't unconditionally VM-Exit on INIT and SHUTDOWN, thus it's
11863 	 * possible to INIT the vCPU while L2 is active.  Force the vCPU back
11864 	 * into L1 as EFER.SVME is cleared on INIT (along with all other EFER
11865 	 * bits), i.e. virtualization is disabled.
11866 	 */
11867 	if (is_guest_mode(vcpu))
11868 		kvm_leave_nested(vcpu);
11869 
11870 	kvm_lapic_reset(vcpu, init_event);
11871 
11872 	WARN_ON_ONCE(is_guest_mode(vcpu) || is_smm(vcpu));
11873 	vcpu->arch.hflags = 0;
11874 
11875 	vcpu->arch.smi_pending = 0;
11876 	vcpu->arch.smi_count = 0;
11877 	atomic_set(&vcpu->arch.nmi_queued, 0);
11878 	vcpu->arch.nmi_pending = 0;
11879 	vcpu->arch.nmi_injected = false;
11880 	kvm_clear_interrupt_queue(vcpu);
11881 	kvm_clear_exception_queue(vcpu);
11882 
11883 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
11884 	kvm_update_dr0123(vcpu);
11885 	vcpu->arch.dr6 = DR6_ACTIVE_LOW;
11886 	vcpu->arch.dr7 = DR7_FIXED_1;
11887 	kvm_update_dr7(vcpu);
11888 
11889 	vcpu->arch.cr2 = 0;
11890 
11891 	kvm_make_request(KVM_REQ_EVENT, vcpu);
11892 	vcpu->arch.apf.msr_en_val = 0;
11893 	vcpu->arch.apf.msr_int_val = 0;
11894 	vcpu->arch.st.msr_val = 0;
11895 
11896 	kvmclock_reset(vcpu);
11897 
11898 	kvm_clear_async_pf_completion_queue(vcpu);
11899 	kvm_async_pf_hash_reset(vcpu);
11900 	vcpu->arch.apf.halted = false;
11901 
11902 	if (vcpu->arch.guest_fpu.fpstate && kvm_mpx_supported()) {
11903 		struct fpstate *fpstate = vcpu->arch.guest_fpu.fpstate;
11904 
11905 		/*
11906 		 * All paths that lead to INIT are required to load the guest's
11907 		 * FPU state (because most paths are buried in KVM_RUN).
11908 		 */
11909 		if (init_event)
11910 			kvm_put_guest_fpu(vcpu);
11911 
11912 		fpstate_clear_xstate_component(fpstate, XFEATURE_BNDREGS);
11913 		fpstate_clear_xstate_component(fpstate, XFEATURE_BNDCSR);
11914 
11915 		if (init_event)
11916 			kvm_load_guest_fpu(vcpu);
11917 	}
11918 
11919 	if (!init_event) {
11920 		kvm_pmu_reset(vcpu);
11921 		vcpu->arch.smbase = 0x30000;
11922 
11923 		vcpu->arch.msr_misc_features_enables = 0;
11924 		vcpu->arch.ia32_misc_enable_msr = MSR_IA32_MISC_ENABLE_PEBS_UNAVAIL |
11925 						  MSR_IA32_MISC_ENABLE_BTS_UNAVAIL;
11926 
11927 		__kvm_set_xcr(vcpu, 0, XFEATURE_MASK_FP);
11928 		__kvm_set_msr(vcpu, MSR_IA32_XSS, 0, true);
11929 	}
11930 
11931 	/* All GPRs except RDX (handled below) are zeroed on RESET/INIT. */
11932 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
11933 	kvm_register_mark_dirty(vcpu, VCPU_REGS_RSP);
11934 
11935 	/*
11936 	 * Fall back to KVM's default Family/Model/Stepping of 0x600 (P6/Athlon)
11937 	 * if no CPUID match is found.  Note, it's impossible to get a match at
11938 	 * RESET since KVM emulates RESET before exposing the vCPU to userspace,
11939 	 * i.e. it's impossible for kvm_find_cpuid_entry() to find a valid entry
11940 	 * on RESET.  But, go through the motions in case that's ever remedied.
11941 	 */
11942 	cpuid_0x1 = kvm_find_cpuid_entry(vcpu, 1);
11943 	kvm_rdx_write(vcpu, cpuid_0x1 ? cpuid_0x1->eax : 0x600);
11944 
11945 	static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
11946 
11947 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
11948 	kvm_rip_write(vcpu, 0xfff0);
11949 
11950 	vcpu->arch.cr3 = 0;
11951 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
11952 
11953 	/*
11954 	 * CR0.CD/NW are set on RESET, preserved on INIT.  Note, some versions
11955 	 * of Intel's SDM list CD/NW as being set on INIT, but they contradict
11956 	 * (or qualify) that with a footnote stating that CD/NW are preserved.
11957 	 */
11958 	new_cr0 = X86_CR0_ET;
11959 	if (init_event)
11960 		new_cr0 |= (old_cr0 & (X86_CR0_NW | X86_CR0_CD));
11961 	else
11962 		new_cr0 |= X86_CR0_NW | X86_CR0_CD;
11963 
11964 	static_call(kvm_x86_set_cr0)(vcpu, new_cr0);
11965 	static_call(kvm_x86_set_cr4)(vcpu, 0);
11966 	static_call(kvm_x86_set_efer)(vcpu, 0);
11967 	static_call(kvm_x86_update_exception_bitmap)(vcpu);
11968 
11969 	/*
11970 	 * On the standard CR0/CR4/EFER modification paths, there are several
11971 	 * complex conditions determining whether the MMU has to be reset and/or
11972 	 * which PCIDs have to be flushed.  However, CR0.WP and the paging-related
11973 	 * bits in CR4 and EFER are irrelevant if CR0.PG was '0'; and a reset+flush
11974 	 * is needed anyway if CR0.PG was '1' (which can only happen for INIT, as
11975 	 * CR0 will be '0' prior to RESET).  So we only need to check CR0.PG here.
11976 	 */
11977 	if (old_cr0 & X86_CR0_PG) {
11978 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
11979 		kvm_mmu_reset_context(vcpu);
11980 	}
11981 
11982 	/*
11983 	 * Intel's SDM states that all TLB entries are flushed on INIT.  AMD's
11984 	 * APM states the TLBs are untouched by INIT, but it also states that
11985 	 * the TLBs are flushed on "External initialization of the processor."
11986 	 * Flush the guest TLB regardless of vendor, there is no meaningful
11987 	 * benefit in relying on the guest to flush the TLB immediately after
11988 	 * INIT.  A spurious TLB flush is benign and likely negligible from a
11989 	 * performance perspective.
11990 	 */
11991 	if (init_event)
11992 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
11993 }
11994 EXPORT_SYMBOL_GPL(kvm_vcpu_reset);
11995 
11996 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
11997 {
11998 	struct kvm_segment cs;
11999 
12000 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
12001 	cs.selector = vector << 8;
12002 	cs.base = vector << 12;
12003 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
12004 	kvm_rip_write(vcpu, 0);
12005 }
12006 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
12007 
12008 int kvm_arch_hardware_enable(void)
12009 {
12010 	struct kvm *kvm;
12011 	struct kvm_vcpu *vcpu;
12012 	unsigned long i;
12013 	int ret;
12014 	u64 local_tsc;
12015 	u64 max_tsc = 0;
12016 	bool stable, backwards_tsc = false;
12017 
12018 	kvm_user_return_msr_cpu_online();
12019 
12020 	ret = kvm_x86_check_processor_compatibility();
12021 	if (ret)
12022 		return ret;
12023 
12024 	ret = static_call(kvm_x86_hardware_enable)();
12025 	if (ret != 0)
12026 		return ret;
12027 
12028 	local_tsc = rdtsc();
12029 	stable = !kvm_check_tsc_unstable();
12030 	list_for_each_entry(kvm, &vm_list, vm_list) {
12031 		kvm_for_each_vcpu(i, vcpu, kvm) {
12032 			if (!stable && vcpu->cpu == smp_processor_id())
12033 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
12034 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
12035 				backwards_tsc = true;
12036 				if (vcpu->arch.last_host_tsc > max_tsc)
12037 					max_tsc = vcpu->arch.last_host_tsc;
12038 			}
12039 		}
12040 	}
12041 
12042 	/*
12043 	 * Sometimes, even reliable TSCs go backwards.  This happens on
12044 	 * platforms that reset TSC during suspend or hibernate actions, but
12045 	 * maintain synchronization.  We must compensate.  Fortunately, we can
12046 	 * detect that condition here, which happens early in CPU bringup,
12047 	 * before any KVM threads can be running.  Unfortunately, we can't
12048 	 * bring the TSCs fully up to date with real time, as we aren't yet far
12049 	 * enough into CPU bringup that we know how much real time has actually
12050 	 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
12051 	 * variables that haven't been updated yet.
12052 	 *
12053 	 * So we simply find the maximum observed TSC above, then record the
12054 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
12055 	 * the adjustment will be applied.  Note that we accumulate
12056 	 * adjustments, in case multiple suspend cycles happen before some VCPU
12057 	 * gets a chance to run again.  In the event that no KVM threads get a
12058 	 * chance to run, we will miss the entire elapsed period, as we'll have
12059 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
12060 	 * loose cycle time.  This isn't too big a deal, since the loss will be
12061 	 * uniform across all VCPUs (not to mention the scenario is extremely
12062 	 * unlikely). It is possible that a second hibernate recovery happens
12063 	 * much faster than a first, causing the observed TSC here to be
12064 	 * smaller; this would require additional padding adjustment, which is
12065 	 * why we set last_host_tsc to the local tsc observed here.
12066 	 *
12067 	 * N.B. - this code below runs only on platforms with reliable TSC,
12068 	 * as that is the only way backwards_tsc is set above.  Also note
12069 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
12070 	 * have the same delta_cyc adjustment applied if backwards_tsc
12071 	 * is detected.  Note further, this adjustment is only done once,
12072 	 * as we reset last_host_tsc on all VCPUs to stop this from being
12073 	 * called multiple times (one for each physical CPU bringup).
12074 	 *
12075 	 * Platforms with unreliable TSCs don't have to deal with this, they
12076 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
12077 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
12078 	 * guarantee that they stay in perfect synchronization.
12079 	 */
12080 	if (backwards_tsc) {
12081 		u64 delta_cyc = max_tsc - local_tsc;
12082 		list_for_each_entry(kvm, &vm_list, vm_list) {
12083 			kvm->arch.backwards_tsc_observed = true;
12084 			kvm_for_each_vcpu(i, vcpu, kvm) {
12085 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
12086 				vcpu->arch.last_host_tsc = local_tsc;
12087 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
12088 			}
12089 
12090 			/*
12091 			 * We have to disable TSC offset matching.. if you were
12092 			 * booting a VM while issuing an S4 host suspend....
12093 			 * you may have some problem.  Solving this issue is
12094 			 * left as an exercise to the reader.
12095 			 */
12096 			kvm->arch.last_tsc_nsec = 0;
12097 			kvm->arch.last_tsc_write = 0;
12098 		}
12099 
12100 	}
12101 	return 0;
12102 }
12103 
12104 void kvm_arch_hardware_disable(void)
12105 {
12106 	static_call(kvm_x86_hardware_disable)();
12107 	drop_user_return_notifiers();
12108 }
12109 
12110 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
12111 {
12112 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
12113 }
12114 
12115 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
12116 {
12117 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
12118 }
12119 
12120 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
12121 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
12122 
12123 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
12124 {
12125 	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
12126 
12127 	vcpu->arch.l1tf_flush_l1d = true;
12128 	if (pmu->version && unlikely(pmu->event_count)) {
12129 		pmu->need_cleanup = true;
12130 		kvm_make_request(KVM_REQ_PMU, vcpu);
12131 	}
12132 	static_call(kvm_x86_sched_in)(vcpu, cpu);
12133 }
12134 
12135 void kvm_arch_free_vm(struct kvm *kvm)
12136 {
12137 	kfree(to_kvm_hv(kvm)->hv_pa_pg);
12138 	__kvm_arch_free_vm(kvm);
12139 }
12140 
12141 
12142 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
12143 {
12144 	int ret;
12145 	unsigned long flags;
12146 
12147 	if (type)
12148 		return -EINVAL;
12149 
12150 	ret = kvm_page_track_init(kvm);
12151 	if (ret)
12152 		goto out;
12153 
12154 	ret = kvm_mmu_init_vm(kvm);
12155 	if (ret)
12156 		goto out_page_track;
12157 
12158 	ret = static_call(kvm_x86_vm_init)(kvm);
12159 	if (ret)
12160 		goto out_uninit_mmu;
12161 
12162 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
12163 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
12164 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
12165 
12166 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
12167 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
12168 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
12169 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
12170 		&kvm->arch.irq_sources_bitmap);
12171 
12172 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
12173 	mutex_init(&kvm->arch.apic_map_lock);
12174 	seqcount_raw_spinlock_init(&kvm->arch.pvclock_sc, &kvm->arch.tsc_write_lock);
12175 	kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
12176 
12177 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
12178 	pvclock_update_vm_gtod_copy(kvm);
12179 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
12180 
12181 	kvm->arch.default_tsc_khz = max_tsc_khz ? : tsc_khz;
12182 	kvm->arch.guest_can_read_msr_platform_info = true;
12183 	kvm->arch.enable_pmu = enable_pmu;
12184 
12185 #if IS_ENABLED(CONFIG_HYPERV)
12186 	spin_lock_init(&kvm->arch.hv_root_tdp_lock);
12187 	kvm->arch.hv_root_tdp = INVALID_PAGE;
12188 #endif
12189 
12190 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
12191 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
12192 
12193 	kvm_apicv_init(kvm);
12194 	kvm_hv_init_vm(kvm);
12195 	kvm_xen_init_vm(kvm);
12196 
12197 	return 0;
12198 
12199 out_uninit_mmu:
12200 	kvm_mmu_uninit_vm(kvm);
12201 out_page_track:
12202 	kvm_page_track_cleanup(kvm);
12203 out:
12204 	return ret;
12205 }
12206 
12207 int kvm_arch_post_init_vm(struct kvm *kvm)
12208 {
12209 	return kvm_mmu_post_init_vm(kvm);
12210 }
12211 
12212 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
12213 {
12214 	vcpu_load(vcpu);
12215 	kvm_mmu_unload(vcpu);
12216 	vcpu_put(vcpu);
12217 }
12218 
12219 static void kvm_unload_vcpu_mmus(struct kvm *kvm)
12220 {
12221 	unsigned long i;
12222 	struct kvm_vcpu *vcpu;
12223 
12224 	kvm_for_each_vcpu(i, vcpu, kvm) {
12225 		kvm_clear_async_pf_completion_queue(vcpu);
12226 		kvm_unload_vcpu_mmu(vcpu);
12227 	}
12228 }
12229 
12230 void kvm_arch_sync_events(struct kvm *kvm)
12231 {
12232 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
12233 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
12234 	kvm_free_pit(kvm);
12235 }
12236 
12237 /**
12238  * __x86_set_memory_region: Setup KVM internal memory slot
12239  *
12240  * @kvm: the kvm pointer to the VM.
12241  * @id: the slot ID to setup.
12242  * @gpa: the GPA to install the slot (unused when @size == 0).
12243  * @size: the size of the slot. Set to zero to uninstall a slot.
12244  *
12245  * This function helps to setup a KVM internal memory slot.  Specify
12246  * @size > 0 to install a new slot, while @size == 0 to uninstall a
12247  * slot.  The return code can be one of the following:
12248  *
12249  *   HVA:           on success (uninstall will return a bogus HVA)
12250  *   -errno:        on error
12251  *
12252  * The caller should always use IS_ERR() to check the return value
12253  * before use.  Note, the KVM internal memory slots are guaranteed to
12254  * remain valid and unchanged until the VM is destroyed, i.e., the
12255  * GPA->HVA translation will not change.  However, the HVA is a user
12256  * address, i.e. its accessibility is not guaranteed, and must be
12257  * accessed via __copy_{to,from}_user().
12258  */
12259 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
12260 				      u32 size)
12261 {
12262 	int i, r;
12263 	unsigned long hva, old_npages;
12264 	struct kvm_memslots *slots = kvm_memslots(kvm);
12265 	struct kvm_memory_slot *slot;
12266 
12267 	/* Called with kvm->slots_lock held.  */
12268 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
12269 		return ERR_PTR_USR(-EINVAL);
12270 
12271 	slot = id_to_memslot(slots, id);
12272 	if (size) {
12273 		if (slot && slot->npages)
12274 			return ERR_PTR_USR(-EEXIST);
12275 
12276 		/*
12277 		 * MAP_SHARED to prevent internal slot pages from being moved
12278 		 * by fork()/COW.
12279 		 */
12280 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
12281 			      MAP_SHARED | MAP_ANONYMOUS, 0);
12282 		if (IS_ERR((void *)hva))
12283 			return (void __user *)hva;
12284 	} else {
12285 		if (!slot || !slot->npages)
12286 			return NULL;
12287 
12288 		old_npages = slot->npages;
12289 		hva = slot->userspace_addr;
12290 	}
12291 
12292 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
12293 		struct kvm_userspace_memory_region m;
12294 
12295 		m.slot = id | (i << 16);
12296 		m.flags = 0;
12297 		m.guest_phys_addr = gpa;
12298 		m.userspace_addr = hva;
12299 		m.memory_size = size;
12300 		r = __kvm_set_memory_region(kvm, &m);
12301 		if (r < 0)
12302 			return ERR_PTR_USR(r);
12303 	}
12304 
12305 	if (!size)
12306 		vm_munmap(hva, old_npages * PAGE_SIZE);
12307 
12308 	return (void __user *)hva;
12309 }
12310 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
12311 
12312 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
12313 {
12314 	kvm_mmu_pre_destroy_vm(kvm);
12315 }
12316 
12317 void kvm_arch_destroy_vm(struct kvm *kvm)
12318 {
12319 	if (current->mm == kvm->mm) {
12320 		/*
12321 		 * Free memory regions allocated on behalf of userspace,
12322 		 * unless the memory map has changed due to process exit
12323 		 * or fd copying.
12324 		 */
12325 		mutex_lock(&kvm->slots_lock);
12326 		__x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
12327 					0, 0);
12328 		__x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
12329 					0, 0);
12330 		__x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
12331 		mutex_unlock(&kvm->slots_lock);
12332 	}
12333 	kvm_unload_vcpu_mmus(kvm);
12334 	static_call_cond(kvm_x86_vm_destroy)(kvm);
12335 	kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
12336 	kvm_pic_destroy(kvm);
12337 	kvm_ioapic_destroy(kvm);
12338 	kvm_destroy_vcpus(kvm);
12339 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
12340 	kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
12341 	kvm_mmu_uninit_vm(kvm);
12342 	kvm_page_track_cleanup(kvm);
12343 	kvm_xen_destroy_vm(kvm);
12344 	kvm_hv_destroy_vm(kvm);
12345 }
12346 
12347 static void memslot_rmap_free(struct kvm_memory_slot *slot)
12348 {
12349 	int i;
12350 
12351 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
12352 		kvfree(slot->arch.rmap[i]);
12353 		slot->arch.rmap[i] = NULL;
12354 	}
12355 }
12356 
12357 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
12358 {
12359 	int i;
12360 
12361 	memslot_rmap_free(slot);
12362 
12363 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
12364 		kvfree(slot->arch.lpage_info[i - 1]);
12365 		slot->arch.lpage_info[i - 1] = NULL;
12366 	}
12367 
12368 	kvm_page_track_free_memslot(slot);
12369 }
12370 
12371 int memslot_rmap_alloc(struct kvm_memory_slot *slot, unsigned long npages)
12372 {
12373 	const int sz = sizeof(*slot->arch.rmap[0]);
12374 	int i;
12375 
12376 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
12377 		int level = i + 1;
12378 		int lpages = __kvm_mmu_slot_lpages(slot, npages, level);
12379 
12380 		if (slot->arch.rmap[i])
12381 			continue;
12382 
12383 		slot->arch.rmap[i] = __vcalloc(lpages, sz, GFP_KERNEL_ACCOUNT);
12384 		if (!slot->arch.rmap[i]) {
12385 			memslot_rmap_free(slot);
12386 			return -ENOMEM;
12387 		}
12388 	}
12389 
12390 	return 0;
12391 }
12392 
12393 static int kvm_alloc_memslot_metadata(struct kvm *kvm,
12394 				      struct kvm_memory_slot *slot)
12395 {
12396 	unsigned long npages = slot->npages;
12397 	int i, r;
12398 
12399 	/*
12400 	 * Clear out the previous array pointers for the KVM_MR_MOVE case.  The
12401 	 * old arrays will be freed by __kvm_set_memory_region() if installing
12402 	 * the new memslot is successful.
12403 	 */
12404 	memset(&slot->arch, 0, sizeof(slot->arch));
12405 
12406 	if (kvm_memslots_have_rmaps(kvm)) {
12407 		r = memslot_rmap_alloc(slot, npages);
12408 		if (r)
12409 			return r;
12410 	}
12411 
12412 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
12413 		struct kvm_lpage_info *linfo;
12414 		unsigned long ugfn;
12415 		int lpages;
12416 		int level = i + 1;
12417 
12418 		lpages = __kvm_mmu_slot_lpages(slot, npages, level);
12419 
12420 		linfo = __vcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
12421 		if (!linfo)
12422 			goto out_free;
12423 
12424 		slot->arch.lpage_info[i - 1] = linfo;
12425 
12426 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
12427 			linfo[0].disallow_lpage = 1;
12428 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
12429 			linfo[lpages - 1].disallow_lpage = 1;
12430 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
12431 		/*
12432 		 * If the gfn and userspace address are not aligned wrt each
12433 		 * other, disable large page support for this slot.
12434 		 */
12435 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
12436 			unsigned long j;
12437 
12438 			for (j = 0; j < lpages; ++j)
12439 				linfo[j].disallow_lpage = 1;
12440 		}
12441 	}
12442 
12443 	if (kvm_page_track_create_memslot(kvm, slot, npages))
12444 		goto out_free;
12445 
12446 	return 0;
12447 
12448 out_free:
12449 	memslot_rmap_free(slot);
12450 
12451 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
12452 		kvfree(slot->arch.lpage_info[i - 1]);
12453 		slot->arch.lpage_info[i - 1] = NULL;
12454 	}
12455 	return -ENOMEM;
12456 }
12457 
12458 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
12459 {
12460 	struct kvm_vcpu *vcpu;
12461 	unsigned long i;
12462 
12463 	/*
12464 	 * memslots->generation has been incremented.
12465 	 * mmio generation may have reached its maximum value.
12466 	 */
12467 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
12468 
12469 	/* Force re-initialization of steal_time cache */
12470 	kvm_for_each_vcpu(i, vcpu, kvm)
12471 		kvm_vcpu_kick(vcpu);
12472 }
12473 
12474 int kvm_arch_prepare_memory_region(struct kvm *kvm,
12475 				   const struct kvm_memory_slot *old,
12476 				   struct kvm_memory_slot *new,
12477 				   enum kvm_mr_change change)
12478 {
12479 	if (change == KVM_MR_CREATE || change == KVM_MR_MOVE) {
12480 		if ((new->base_gfn + new->npages - 1) > kvm_mmu_max_gfn())
12481 			return -EINVAL;
12482 
12483 		return kvm_alloc_memslot_metadata(kvm, new);
12484 	}
12485 
12486 	if (change == KVM_MR_FLAGS_ONLY)
12487 		memcpy(&new->arch, &old->arch, sizeof(old->arch));
12488 	else if (WARN_ON_ONCE(change != KVM_MR_DELETE))
12489 		return -EIO;
12490 
12491 	return 0;
12492 }
12493 
12494 
12495 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
12496 {
12497 	struct kvm_arch *ka = &kvm->arch;
12498 
12499 	if (!kvm_x86_ops.cpu_dirty_log_size)
12500 		return;
12501 
12502 	if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
12503 	    (!enable && --ka->cpu_dirty_logging_count == 0))
12504 		kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
12505 
12506 	WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
12507 }
12508 
12509 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
12510 				     struct kvm_memory_slot *old,
12511 				     const struct kvm_memory_slot *new,
12512 				     enum kvm_mr_change change)
12513 {
12514 	u32 old_flags = old ? old->flags : 0;
12515 	u32 new_flags = new ? new->flags : 0;
12516 	bool log_dirty_pages = new_flags & KVM_MEM_LOG_DIRTY_PAGES;
12517 
12518 	/*
12519 	 * Update CPU dirty logging if dirty logging is being toggled.  This
12520 	 * applies to all operations.
12521 	 */
12522 	if ((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES)
12523 		kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
12524 
12525 	/*
12526 	 * Nothing more to do for RO slots (which can't be dirtied and can't be
12527 	 * made writable) or CREATE/MOVE/DELETE of a slot.
12528 	 *
12529 	 * For a memslot with dirty logging disabled:
12530 	 * CREATE:      No dirty mappings will already exist.
12531 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
12532 	 *		kvm_arch_flush_shadow_memslot()
12533 	 *
12534 	 * For a memslot with dirty logging enabled:
12535 	 * CREATE:      No shadow pages exist, thus nothing to write-protect
12536 	 *		and no dirty bits to clear.
12537 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
12538 	 *		kvm_arch_flush_shadow_memslot().
12539 	 */
12540 	if ((change != KVM_MR_FLAGS_ONLY) || (new_flags & KVM_MEM_READONLY))
12541 		return;
12542 
12543 	/*
12544 	 * READONLY and non-flags changes were filtered out above, and the only
12545 	 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
12546 	 * logging isn't being toggled on or off.
12547 	 */
12548 	if (WARN_ON_ONCE(!((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES)))
12549 		return;
12550 
12551 	if (!log_dirty_pages) {
12552 		/*
12553 		 * Dirty logging tracks sptes in 4k granularity, meaning that
12554 		 * large sptes have to be split.  If live migration succeeds,
12555 		 * the guest in the source machine will be destroyed and large
12556 		 * sptes will be created in the destination.  However, if the
12557 		 * guest continues to run in the source machine (for example if
12558 		 * live migration fails), small sptes will remain around and
12559 		 * cause bad performance.
12560 		 *
12561 		 * Scan sptes if dirty logging has been stopped, dropping those
12562 		 * which can be collapsed into a single large-page spte.  Later
12563 		 * page faults will create the large-page sptes.
12564 		 */
12565 		kvm_mmu_zap_collapsible_sptes(kvm, new);
12566 	} else {
12567 		/*
12568 		 * Initially-all-set does not require write protecting any page,
12569 		 * because they're all assumed to be dirty.
12570 		 */
12571 		if (kvm_dirty_log_manual_protect_and_init_set(kvm))
12572 			return;
12573 
12574 		if (READ_ONCE(eager_page_split))
12575 			kvm_mmu_slot_try_split_huge_pages(kvm, new, PG_LEVEL_4K);
12576 
12577 		if (kvm_x86_ops.cpu_dirty_log_size) {
12578 			kvm_mmu_slot_leaf_clear_dirty(kvm, new);
12579 			kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M);
12580 		} else {
12581 			kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
12582 		}
12583 
12584 		/*
12585 		 * Unconditionally flush the TLBs after enabling dirty logging.
12586 		 * A flush is almost always going to be necessary (see below),
12587 		 * and unconditionally flushing allows the helpers to omit
12588 		 * the subtly complex checks when removing write access.
12589 		 *
12590 		 * Do the flush outside of mmu_lock to reduce the amount of
12591 		 * time mmu_lock is held.  Flushing after dropping mmu_lock is
12592 		 * safe as KVM only needs to guarantee the slot is fully
12593 		 * write-protected before returning to userspace, i.e. before
12594 		 * userspace can consume the dirty status.
12595 		 *
12596 		 * Flushing outside of mmu_lock requires KVM to be careful when
12597 		 * making decisions based on writable status of an SPTE, e.g. a
12598 		 * !writable SPTE doesn't guarantee a CPU can't perform writes.
12599 		 *
12600 		 * Specifically, KVM also write-protects guest page tables to
12601 		 * monitor changes when using shadow paging, and must guarantee
12602 		 * no CPUs can write to those page before mmu_lock is dropped.
12603 		 * Because CPUs may have stale TLB entries at this point, a
12604 		 * !writable SPTE doesn't guarantee CPUs can't perform writes.
12605 		 *
12606 		 * KVM also allows making SPTES writable outside of mmu_lock,
12607 		 * e.g. to allow dirty logging without taking mmu_lock.
12608 		 *
12609 		 * To handle these scenarios, KVM uses a separate software-only
12610 		 * bit (MMU-writable) to track if a SPTE is !writable due to
12611 		 * a guest page table being write-protected (KVM clears the
12612 		 * MMU-writable flag when write-protecting for shadow paging).
12613 		 *
12614 		 * The use of MMU-writable is also the primary motivation for
12615 		 * the unconditional flush.  Because KVM must guarantee that a
12616 		 * CPU doesn't contain stale, writable TLB entries for a
12617 		 * !MMU-writable SPTE, KVM must flush if it encounters any
12618 		 * MMU-writable SPTE regardless of whether the actual hardware
12619 		 * writable bit was set.  I.e. KVM is almost guaranteed to need
12620 		 * to flush, while unconditionally flushing allows the "remove
12621 		 * write access" helpers to ignore MMU-writable entirely.
12622 		 *
12623 		 * See is_writable_pte() for more details (the case involving
12624 		 * access-tracked SPTEs is particularly relevant).
12625 		 */
12626 		kvm_arch_flush_remote_tlbs_memslot(kvm, new);
12627 	}
12628 }
12629 
12630 void kvm_arch_commit_memory_region(struct kvm *kvm,
12631 				struct kvm_memory_slot *old,
12632 				const struct kvm_memory_slot *new,
12633 				enum kvm_mr_change change)
12634 {
12635 	if (!kvm->arch.n_requested_mmu_pages &&
12636 	    (change == KVM_MR_CREATE || change == KVM_MR_DELETE)) {
12637 		unsigned long nr_mmu_pages;
12638 
12639 		nr_mmu_pages = kvm->nr_memslot_pages / KVM_MEMSLOT_PAGES_TO_MMU_PAGES_RATIO;
12640 		nr_mmu_pages = max(nr_mmu_pages, KVM_MIN_ALLOC_MMU_PAGES);
12641 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
12642 	}
12643 
12644 	kvm_mmu_slot_apply_flags(kvm, old, new, change);
12645 
12646 	/* Free the arrays associated with the old memslot. */
12647 	if (change == KVM_MR_MOVE)
12648 		kvm_arch_free_memslot(kvm, old);
12649 }
12650 
12651 void kvm_arch_flush_shadow_all(struct kvm *kvm)
12652 {
12653 	kvm_mmu_zap_all(kvm);
12654 }
12655 
12656 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
12657 				   struct kvm_memory_slot *slot)
12658 {
12659 	kvm_page_track_flush_slot(kvm, slot);
12660 }
12661 
12662 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
12663 {
12664 	return (is_guest_mode(vcpu) &&
12665 		static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
12666 }
12667 
12668 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
12669 {
12670 	if (!list_empty_careful(&vcpu->async_pf.done))
12671 		return true;
12672 
12673 	if (kvm_apic_has_pending_init_or_sipi(vcpu) &&
12674 	    kvm_apic_init_sipi_allowed(vcpu))
12675 		return true;
12676 
12677 	if (vcpu->arch.pv.pv_unhalted)
12678 		return true;
12679 
12680 	if (kvm_is_exception_pending(vcpu))
12681 		return true;
12682 
12683 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
12684 	    (vcpu->arch.nmi_pending &&
12685 	     static_call(kvm_x86_nmi_allowed)(vcpu, false)))
12686 		return true;
12687 
12688 #ifdef CONFIG_KVM_SMM
12689 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
12690 	    (vcpu->arch.smi_pending &&
12691 	     static_call(kvm_x86_smi_allowed)(vcpu, false)))
12692 		return true;
12693 #endif
12694 
12695 	if (kvm_arch_interrupt_allowed(vcpu) &&
12696 	    (kvm_cpu_has_interrupt(vcpu) ||
12697 	    kvm_guest_apic_has_interrupt(vcpu)))
12698 		return true;
12699 
12700 	if (kvm_hv_has_stimer_pending(vcpu))
12701 		return true;
12702 
12703 	if (is_guest_mode(vcpu) &&
12704 	    kvm_x86_ops.nested_ops->has_events &&
12705 	    kvm_x86_ops.nested_ops->has_events(vcpu))
12706 		return true;
12707 
12708 	if (kvm_xen_has_pending_events(vcpu))
12709 		return true;
12710 
12711 	return false;
12712 }
12713 
12714 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
12715 {
12716 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
12717 }
12718 
12719 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
12720 {
12721 	if (kvm_vcpu_apicv_active(vcpu) &&
12722 	    static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
12723 		return true;
12724 
12725 	return false;
12726 }
12727 
12728 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
12729 {
12730 	if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
12731 		return true;
12732 
12733 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
12734 #ifdef CONFIG_KVM_SMM
12735 		kvm_test_request(KVM_REQ_SMI, vcpu) ||
12736 #endif
12737 		 kvm_test_request(KVM_REQ_EVENT, vcpu))
12738 		return true;
12739 
12740 	return kvm_arch_dy_has_pending_interrupt(vcpu);
12741 }
12742 
12743 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
12744 {
12745 	if (vcpu->arch.guest_state_protected)
12746 		return true;
12747 
12748 	return vcpu->arch.preempted_in_kernel;
12749 }
12750 
12751 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
12752 {
12753 	return kvm_rip_read(vcpu);
12754 }
12755 
12756 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
12757 {
12758 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
12759 }
12760 
12761 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
12762 {
12763 	return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
12764 }
12765 
12766 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
12767 {
12768 	/* Can't read the RIP when guest state is protected, just return 0 */
12769 	if (vcpu->arch.guest_state_protected)
12770 		return 0;
12771 
12772 	if (is_64_bit_mode(vcpu))
12773 		return kvm_rip_read(vcpu);
12774 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
12775 		     kvm_rip_read(vcpu));
12776 }
12777 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
12778 
12779 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
12780 {
12781 	return kvm_get_linear_rip(vcpu) == linear_rip;
12782 }
12783 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
12784 
12785 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
12786 {
12787 	unsigned long rflags;
12788 
12789 	rflags = static_call(kvm_x86_get_rflags)(vcpu);
12790 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
12791 		rflags &= ~X86_EFLAGS_TF;
12792 	return rflags;
12793 }
12794 EXPORT_SYMBOL_GPL(kvm_get_rflags);
12795 
12796 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
12797 {
12798 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
12799 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
12800 		rflags |= X86_EFLAGS_TF;
12801 	static_call(kvm_x86_set_rflags)(vcpu, rflags);
12802 }
12803 
12804 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
12805 {
12806 	__kvm_set_rflags(vcpu, rflags);
12807 	kvm_make_request(KVM_REQ_EVENT, vcpu);
12808 }
12809 EXPORT_SYMBOL_GPL(kvm_set_rflags);
12810 
12811 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
12812 {
12813 	BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
12814 
12815 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
12816 }
12817 
12818 static inline u32 kvm_async_pf_next_probe(u32 key)
12819 {
12820 	return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
12821 }
12822 
12823 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12824 {
12825 	u32 key = kvm_async_pf_hash_fn(gfn);
12826 
12827 	while (vcpu->arch.apf.gfns[key] != ~0)
12828 		key = kvm_async_pf_next_probe(key);
12829 
12830 	vcpu->arch.apf.gfns[key] = gfn;
12831 }
12832 
12833 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
12834 {
12835 	int i;
12836 	u32 key = kvm_async_pf_hash_fn(gfn);
12837 
12838 	for (i = 0; i < ASYNC_PF_PER_VCPU &&
12839 		     (vcpu->arch.apf.gfns[key] != gfn &&
12840 		      vcpu->arch.apf.gfns[key] != ~0); i++)
12841 		key = kvm_async_pf_next_probe(key);
12842 
12843 	return key;
12844 }
12845 
12846 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12847 {
12848 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
12849 }
12850 
12851 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12852 {
12853 	u32 i, j, k;
12854 
12855 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
12856 
12857 	if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
12858 		return;
12859 
12860 	while (true) {
12861 		vcpu->arch.apf.gfns[i] = ~0;
12862 		do {
12863 			j = kvm_async_pf_next_probe(j);
12864 			if (vcpu->arch.apf.gfns[j] == ~0)
12865 				return;
12866 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
12867 			/*
12868 			 * k lies cyclically in ]i,j]
12869 			 * |    i.k.j |
12870 			 * |....j i.k.| or  |.k..j i...|
12871 			 */
12872 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
12873 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
12874 		i = j;
12875 	}
12876 }
12877 
12878 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
12879 {
12880 	u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
12881 
12882 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
12883 				      sizeof(reason));
12884 }
12885 
12886 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
12887 {
12888 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
12889 
12890 	return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
12891 					     &token, offset, sizeof(token));
12892 }
12893 
12894 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
12895 {
12896 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
12897 	u32 val;
12898 
12899 	if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
12900 					 &val, offset, sizeof(val)))
12901 		return false;
12902 
12903 	return !val;
12904 }
12905 
12906 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
12907 {
12908 
12909 	if (!kvm_pv_async_pf_enabled(vcpu))
12910 		return false;
12911 
12912 	if (vcpu->arch.apf.send_user_only &&
12913 	    static_call(kvm_x86_get_cpl)(vcpu) == 0)
12914 		return false;
12915 
12916 	if (is_guest_mode(vcpu)) {
12917 		/*
12918 		 * L1 needs to opt into the special #PF vmexits that are
12919 		 * used to deliver async page faults.
12920 		 */
12921 		return vcpu->arch.apf.delivery_as_pf_vmexit;
12922 	} else {
12923 		/*
12924 		 * Play it safe in case the guest temporarily disables paging.
12925 		 * The real mode IDT in particular is unlikely to have a #PF
12926 		 * exception setup.
12927 		 */
12928 		return is_paging(vcpu);
12929 	}
12930 }
12931 
12932 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
12933 {
12934 	if (unlikely(!lapic_in_kernel(vcpu) ||
12935 		     kvm_event_needs_reinjection(vcpu) ||
12936 		     kvm_is_exception_pending(vcpu)))
12937 		return false;
12938 
12939 	if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
12940 		return false;
12941 
12942 	/*
12943 	 * If interrupts are off we cannot even use an artificial
12944 	 * halt state.
12945 	 */
12946 	return kvm_arch_interrupt_allowed(vcpu);
12947 }
12948 
12949 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
12950 				     struct kvm_async_pf *work)
12951 {
12952 	struct x86_exception fault;
12953 
12954 	trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
12955 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
12956 
12957 	if (kvm_can_deliver_async_pf(vcpu) &&
12958 	    !apf_put_user_notpresent(vcpu)) {
12959 		fault.vector = PF_VECTOR;
12960 		fault.error_code_valid = true;
12961 		fault.error_code = 0;
12962 		fault.nested_page_fault = false;
12963 		fault.address = work->arch.token;
12964 		fault.async_page_fault = true;
12965 		kvm_inject_page_fault(vcpu, &fault);
12966 		return true;
12967 	} else {
12968 		/*
12969 		 * It is not possible to deliver a paravirtualized asynchronous
12970 		 * page fault, but putting the guest in an artificial halt state
12971 		 * can be beneficial nevertheless: if an interrupt arrives, we
12972 		 * can deliver it timely and perhaps the guest will schedule
12973 		 * another process.  When the instruction that triggered a page
12974 		 * fault is retried, hopefully the page will be ready in the host.
12975 		 */
12976 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
12977 		return false;
12978 	}
12979 }
12980 
12981 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
12982 				 struct kvm_async_pf *work)
12983 {
12984 	struct kvm_lapic_irq irq = {
12985 		.delivery_mode = APIC_DM_FIXED,
12986 		.vector = vcpu->arch.apf.vec
12987 	};
12988 
12989 	if (work->wakeup_all)
12990 		work->arch.token = ~0; /* broadcast wakeup */
12991 	else
12992 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
12993 	trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
12994 
12995 	if ((work->wakeup_all || work->notpresent_injected) &&
12996 	    kvm_pv_async_pf_enabled(vcpu) &&
12997 	    !apf_put_user_ready(vcpu, work->arch.token)) {
12998 		vcpu->arch.apf.pageready_pending = true;
12999 		kvm_apic_set_irq(vcpu, &irq, NULL);
13000 	}
13001 
13002 	vcpu->arch.apf.halted = false;
13003 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
13004 }
13005 
13006 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
13007 {
13008 	kvm_make_request(KVM_REQ_APF_READY, vcpu);
13009 	if (!vcpu->arch.apf.pageready_pending)
13010 		kvm_vcpu_kick(vcpu);
13011 }
13012 
13013 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
13014 {
13015 	if (!kvm_pv_async_pf_enabled(vcpu))
13016 		return true;
13017 	else
13018 		return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
13019 }
13020 
13021 void kvm_arch_start_assignment(struct kvm *kvm)
13022 {
13023 	if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
13024 		static_call_cond(kvm_x86_pi_start_assignment)(kvm);
13025 }
13026 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
13027 
13028 void kvm_arch_end_assignment(struct kvm *kvm)
13029 {
13030 	atomic_dec(&kvm->arch.assigned_device_count);
13031 }
13032 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
13033 
13034 bool noinstr kvm_arch_has_assigned_device(struct kvm *kvm)
13035 {
13036 	return arch_atomic_read(&kvm->arch.assigned_device_count);
13037 }
13038 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
13039 
13040 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
13041 {
13042 	atomic_inc(&kvm->arch.noncoherent_dma_count);
13043 }
13044 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
13045 
13046 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
13047 {
13048 	atomic_dec(&kvm->arch.noncoherent_dma_count);
13049 }
13050 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
13051 
13052 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
13053 {
13054 	return atomic_read(&kvm->arch.noncoherent_dma_count);
13055 }
13056 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
13057 
13058 bool kvm_arch_has_irq_bypass(void)
13059 {
13060 	return true;
13061 }
13062 
13063 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
13064 				      struct irq_bypass_producer *prod)
13065 {
13066 	struct kvm_kernel_irqfd *irqfd =
13067 		container_of(cons, struct kvm_kernel_irqfd, consumer);
13068 	int ret;
13069 
13070 	irqfd->producer = prod;
13071 	kvm_arch_start_assignment(irqfd->kvm);
13072 	ret = static_call(kvm_x86_pi_update_irte)(irqfd->kvm,
13073 					 prod->irq, irqfd->gsi, 1);
13074 
13075 	if (ret)
13076 		kvm_arch_end_assignment(irqfd->kvm);
13077 
13078 	return ret;
13079 }
13080 
13081 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
13082 				      struct irq_bypass_producer *prod)
13083 {
13084 	int ret;
13085 	struct kvm_kernel_irqfd *irqfd =
13086 		container_of(cons, struct kvm_kernel_irqfd, consumer);
13087 
13088 	WARN_ON(irqfd->producer != prod);
13089 	irqfd->producer = NULL;
13090 
13091 	/*
13092 	 * When producer of consumer is unregistered, we change back to
13093 	 * remapped mode, so we can re-use the current implementation
13094 	 * when the irq is masked/disabled or the consumer side (KVM
13095 	 * int this case doesn't want to receive the interrupts.
13096 	*/
13097 	ret = static_call(kvm_x86_pi_update_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
13098 	if (ret)
13099 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
13100 		       " fails: %d\n", irqfd->consumer.token, ret);
13101 
13102 	kvm_arch_end_assignment(irqfd->kvm);
13103 }
13104 
13105 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
13106 				   uint32_t guest_irq, bool set)
13107 {
13108 	return static_call(kvm_x86_pi_update_irte)(kvm, host_irq, guest_irq, set);
13109 }
13110 
13111 bool kvm_arch_irqfd_route_changed(struct kvm_kernel_irq_routing_entry *old,
13112 				  struct kvm_kernel_irq_routing_entry *new)
13113 {
13114 	if (new->type != KVM_IRQ_ROUTING_MSI)
13115 		return true;
13116 
13117 	return !!memcmp(&old->msi, &new->msi, sizeof(new->msi));
13118 }
13119 
13120 bool kvm_vector_hashing_enabled(void)
13121 {
13122 	return vector_hashing;
13123 }
13124 
13125 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
13126 {
13127 	return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
13128 }
13129 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
13130 
13131 
13132 int kvm_spec_ctrl_test_value(u64 value)
13133 {
13134 	/*
13135 	 * test that setting IA32_SPEC_CTRL to given value
13136 	 * is allowed by the host processor
13137 	 */
13138 
13139 	u64 saved_value;
13140 	unsigned long flags;
13141 	int ret = 0;
13142 
13143 	local_irq_save(flags);
13144 
13145 	if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
13146 		ret = 1;
13147 	else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
13148 		ret = 1;
13149 	else
13150 		wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
13151 
13152 	local_irq_restore(flags);
13153 
13154 	return ret;
13155 }
13156 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
13157 
13158 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
13159 {
13160 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
13161 	struct x86_exception fault;
13162 	u64 access = error_code &
13163 		(PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
13164 
13165 	if (!(error_code & PFERR_PRESENT_MASK) ||
13166 	    mmu->gva_to_gpa(vcpu, mmu, gva, access, &fault) != INVALID_GPA) {
13167 		/*
13168 		 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
13169 		 * tables probably do not match the TLB.  Just proceed
13170 		 * with the error code that the processor gave.
13171 		 */
13172 		fault.vector = PF_VECTOR;
13173 		fault.error_code_valid = true;
13174 		fault.error_code = error_code;
13175 		fault.nested_page_fault = false;
13176 		fault.address = gva;
13177 		fault.async_page_fault = false;
13178 	}
13179 	vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
13180 }
13181 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
13182 
13183 /*
13184  * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
13185  * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
13186  * indicates whether exit to userspace is needed.
13187  */
13188 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
13189 			      struct x86_exception *e)
13190 {
13191 	if (r == X86EMUL_PROPAGATE_FAULT) {
13192 		if (KVM_BUG_ON(!e, vcpu->kvm))
13193 			return -EIO;
13194 
13195 		kvm_inject_emulated_page_fault(vcpu, e);
13196 		return 1;
13197 	}
13198 
13199 	/*
13200 	 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
13201 	 * while handling a VMX instruction KVM could've handled the request
13202 	 * correctly by exiting to userspace and performing I/O but there
13203 	 * doesn't seem to be a real use-case behind such requests, just return
13204 	 * KVM_EXIT_INTERNAL_ERROR for now.
13205 	 */
13206 	kvm_prepare_emulation_failure_exit(vcpu);
13207 
13208 	return 0;
13209 }
13210 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
13211 
13212 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
13213 {
13214 	bool pcid_enabled;
13215 	struct x86_exception e;
13216 	struct {
13217 		u64 pcid;
13218 		u64 gla;
13219 	} operand;
13220 	int r;
13221 
13222 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
13223 	if (r != X86EMUL_CONTINUE)
13224 		return kvm_handle_memory_failure(vcpu, r, &e);
13225 
13226 	if (operand.pcid >> 12 != 0) {
13227 		kvm_inject_gp(vcpu, 0);
13228 		return 1;
13229 	}
13230 
13231 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
13232 
13233 	switch (type) {
13234 	case INVPCID_TYPE_INDIV_ADDR:
13235 		if ((!pcid_enabled && (operand.pcid != 0)) ||
13236 		    is_noncanonical_address(operand.gla, vcpu)) {
13237 			kvm_inject_gp(vcpu, 0);
13238 			return 1;
13239 		}
13240 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
13241 		return kvm_skip_emulated_instruction(vcpu);
13242 
13243 	case INVPCID_TYPE_SINGLE_CTXT:
13244 		if (!pcid_enabled && (operand.pcid != 0)) {
13245 			kvm_inject_gp(vcpu, 0);
13246 			return 1;
13247 		}
13248 
13249 		kvm_invalidate_pcid(vcpu, operand.pcid);
13250 		return kvm_skip_emulated_instruction(vcpu);
13251 
13252 	case INVPCID_TYPE_ALL_NON_GLOBAL:
13253 		/*
13254 		 * Currently, KVM doesn't mark global entries in the shadow
13255 		 * page tables, so a non-global flush just degenerates to a
13256 		 * global flush. If needed, we could optimize this later by
13257 		 * keeping track of global entries in shadow page tables.
13258 		 */
13259 
13260 		fallthrough;
13261 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
13262 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
13263 		return kvm_skip_emulated_instruction(vcpu);
13264 
13265 	default:
13266 		kvm_inject_gp(vcpu, 0);
13267 		return 1;
13268 	}
13269 }
13270 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
13271 
13272 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
13273 {
13274 	struct kvm_run *run = vcpu->run;
13275 	struct kvm_mmio_fragment *frag;
13276 	unsigned int len;
13277 
13278 	BUG_ON(!vcpu->mmio_needed);
13279 
13280 	/* Complete previous fragment */
13281 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
13282 	len = min(8u, frag->len);
13283 	if (!vcpu->mmio_is_write)
13284 		memcpy(frag->data, run->mmio.data, len);
13285 
13286 	if (frag->len <= 8) {
13287 		/* Switch to the next fragment. */
13288 		frag++;
13289 		vcpu->mmio_cur_fragment++;
13290 	} else {
13291 		/* Go forward to the next mmio piece. */
13292 		frag->data += len;
13293 		frag->gpa += len;
13294 		frag->len -= len;
13295 	}
13296 
13297 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
13298 		vcpu->mmio_needed = 0;
13299 
13300 		// VMG change, at this point, we're always done
13301 		// RIP has already been advanced
13302 		return 1;
13303 	}
13304 
13305 	// More MMIO is needed
13306 	run->mmio.phys_addr = frag->gpa;
13307 	run->mmio.len = min(8u, frag->len);
13308 	run->mmio.is_write = vcpu->mmio_is_write;
13309 	if (run->mmio.is_write)
13310 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
13311 	run->exit_reason = KVM_EXIT_MMIO;
13312 
13313 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
13314 
13315 	return 0;
13316 }
13317 
13318 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
13319 			  void *data)
13320 {
13321 	int handled;
13322 	struct kvm_mmio_fragment *frag;
13323 
13324 	if (!data)
13325 		return -EINVAL;
13326 
13327 	handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
13328 	if (handled == bytes)
13329 		return 1;
13330 
13331 	bytes -= handled;
13332 	gpa += handled;
13333 	data += handled;
13334 
13335 	/*TODO: Check if need to increment number of frags */
13336 	frag = vcpu->mmio_fragments;
13337 	vcpu->mmio_nr_fragments = 1;
13338 	frag->len = bytes;
13339 	frag->gpa = gpa;
13340 	frag->data = data;
13341 
13342 	vcpu->mmio_needed = 1;
13343 	vcpu->mmio_cur_fragment = 0;
13344 
13345 	vcpu->run->mmio.phys_addr = gpa;
13346 	vcpu->run->mmio.len = min(8u, frag->len);
13347 	vcpu->run->mmio.is_write = 1;
13348 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
13349 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
13350 
13351 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
13352 
13353 	return 0;
13354 }
13355 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
13356 
13357 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
13358 			 void *data)
13359 {
13360 	int handled;
13361 	struct kvm_mmio_fragment *frag;
13362 
13363 	if (!data)
13364 		return -EINVAL;
13365 
13366 	handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
13367 	if (handled == bytes)
13368 		return 1;
13369 
13370 	bytes -= handled;
13371 	gpa += handled;
13372 	data += handled;
13373 
13374 	/*TODO: Check if need to increment number of frags */
13375 	frag = vcpu->mmio_fragments;
13376 	vcpu->mmio_nr_fragments = 1;
13377 	frag->len = bytes;
13378 	frag->gpa = gpa;
13379 	frag->data = data;
13380 
13381 	vcpu->mmio_needed = 1;
13382 	vcpu->mmio_cur_fragment = 0;
13383 
13384 	vcpu->run->mmio.phys_addr = gpa;
13385 	vcpu->run->mmio.len = min(8u, frag->len);
13386 	vcpu->run->mmio.is_write = 0;
13387 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
13388 
13389 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
13390 
13391 	return 0;
13392 }
13393 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
13394 
13395 static void advance_sev_es_emulated_pio(struct kvm_vcpu *vcpu, unsigned count, int size)
13396 {
13397 	vcpu->arch.sev_pio_count -= count;
13398 	vcpu->arch.sev_pio_data += count * size;
13399 }
13400 
13401 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
13402 			   unsigned int port);
13403 
13404 static int complete_sev_es_emulated_outs(struct kvm_vcpu *vcpu)
13405 {
13406 	int size = vcpu->arch.pio.size;
13407 	int port = vcpu->arch.pio.port;
13408 
13409 	vcpu->arch.pio.count = 0;
13410 	if (vcpu->arch.sev_pio_count)
13411 		return kvm_sev_es_outs(vcpu, size, port);
13412 	return 1;
13413 }
13414 
13415 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
13416 			   unsigned int port)
13417 {
13418 	for (;;) {
13419 		unsigned int count =
13420 			min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
13421 		int ret = emulator_pio_out(vcpu, size, port, vcpu->arch.sev_pio_data, count);
13422 
13423 		/* memcpy done already by emulator_pio_out.  */
13424 		advance_sev_es_emulated_pio(vcpu, count, size);
13425 		if (!ret)
13426 			break;
13427 
13428 		/* Emulation done by the kernel.  */
13429 		if (!vcpu->arch.sev_pio_count)
13430 			return 1;
13431 	}
13432 
13433 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_outs;
13434 	return 0;
13435 }
13436 
13437 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
13438 			  unsigned int port);
13439 
13440 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
13441 {
13442 	unsigned count = vcpu->arch.pio.count;
13443 	int size = vcpu->arch.pio.size;
13444 	int port = vcpu->arch.pio.port;
13445 
13446 	complete_emulator_pio_in(vcpu, vcpu->arch.sev_pio_data);
13447 	advance_sev_es_emulated_pio(vcpu, count, size);
13448 	if (vcpu->arch.sev_pio_count)
13449 		return kvm_sev_es_ins(vcpu, size, port);
13450 	return 1;
13451 }
13452 
13453 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
13454 			  unsigned int port)
13455 {
13456 	for (;;) {
13457 		unsigned int count =
13458 			min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
13459 		if (!emulator_pio_in(vcpu, size, port, vcpu->arch.sev_pio_data, count))
13460 			break;
13461 
13462 		/* Emulation done by the kernel.  */
13463 		advance_sev_es_emulated_pio(vcpu, count, size);
13464 		if (!vcpu->arch.sev_pio_count)
13465 			return 1;
13466 	}
13467 
13468 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
13469 	return 0;
13470 }
13471 
13472 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
13473 			 unsigned int port, void *data,  unsigned int count,
13474 			 int in)
13475 {
13476 	vcpu->arch.sev_pio_data = data;
13477 	vcpu->arch.sev_pio_count = count;
13478 	return in ? kvm_sev_es_ins(vcpu, size, port)
13479 		  : kvm_sev_es_outs(vcpu, size, port);
13480 }
13481 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
13482 
13483 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
13484 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
13485 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
13486 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
13487 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
13488 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
13489 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
13490 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter);
13491 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
13492 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
13493 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
13494 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
13495 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
13496 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
13497 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
13498 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
13499 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
13500 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
13501 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
13502 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
13503 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
13504 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
13505 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_kick_vcpu_slowpath);
13506 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_doorbell);
13507 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_accept_irq);
13508 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
13509 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
13510 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
13511 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);
13512 
13513 static int __init kvm_x86_init(void)
13514 {
13515 	kvm_mmu_x86_module_init();
13516 	return 0;
13517 }
13518 module_init(kvm_x86_init);
13519 
13520 static void __exit kvm_x86_exit(void)
13521 {
13522 	/*
13523 	 * If module_init() is implemented, module_exit() must also be
13524 	 * implemented to allow module unload.
13525 	 */
13526 }
13527 module_exit(kvm_x86_exit);
13528