xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 359745d7)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * derived from drivers/kvm/kvm_main.c
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright (C) 2008 Qumranet, Inc.
9  * Copyright IBM Corporation, 2008
10  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11  *
12  * Authors:
13  *   Avi Kivity   <avi@qumranet.com>
14  *   Yaniv Kamay  <yaniv@qumranet.com>
15  *   Amit Shah    <amit.shah@qumranet.com>
16  *   Ben-Ami Yassour <benami@il.ibm.com>
17  */
18 
19 #include <linux/kvm_host.h>
20 #include "irq.h"
21 #include "ioapic.h"
22 #include "mmu.h"
23 #include "i8254.h"
24 #include "tss.h"
25 #include "kvm_cache_regs.h"
26 #include "kvm_emulate.h"
27 #include "x86.h"
28 #include "cpuid.h"
29 #include "pmu.h"
30 #include "hyperv.h"
31 #include "lapic.h"
32 #include "xen.h"
33 
34 #include <linux/clocksource.h>
35 #include <linux/interrupt.h>
36 #include <linux/kvm.h>
37 #include <linux/fs.h>
38 #include <linux/vmalloc.h>
39 #include <linux/export.h>
40 #include <linux/moduleparam.h>
41 #include <linux/mman.h>
42 #include <linux/highmem.h>
43 #include <linux/iommu.h>
44 #include <linux/intel-iommu.h>
45 #include <linux/cpufreq.h>
46 #include <linux/user-return-notifier.h>
47 #include <linux/srcu.h>
48 #include <linux/slab.h>
49 #include <linux/perf_event.h>
50 #include <linux/uaccess.h>
51 #include <linux/hash.h>
52 #include <linux/pci.h>
53 #include <linux/timekeeper_internal.h>
54 #include <linux/pvclock_gtod.h>
55 #include <linux/kvm_irqfd.h>
56 #include <linux/irqbypass.h>
57 #include <linux/sched/stat.h>
58 #include <linux/sched/isolation.h>
59 #include <linux/mem_encrypt.h>
60 #include <linux/entry-kvm.h>
61 #include <linux/suspend.h>
62 
63 #include <trace/events/kvm.h>
64 
65 #include <asm/debugreg.h>
66 #include <asm/msr.h>
67 #include <asm/desc.h>
68 #include <asm/mce.h>
69 #include <asm/pkru.h>
70 #include <linux/kernel_stat.h>
71 #include <asm/fpu/api.h>
72 #include <asm/fpu/xcr.h>
73 #include <asm/fpu/xstate.h>
74 #include <asm/pvclock.h>
75 #include <asm/div64.h>
76 #include <asm/irq_remapping.h>
77 #include <asm/mshyperv.h>
78 #include <asm/hypervisor.h>
79 #include <asm/tlbflush.h>
80 #include <asm/intel_pt.h>
81 #include <asm/emulate_prefix.h>
82 #include <asm/sgx.h>
83 #include <clocksource/hyperv_timer.h>
84 
85 #define CREATE_TRACE_POINTS
86 #include "trace.h"
87 
88 #define MAX_IO_MSRS 256
89 #define KVM_MAX_MCE_BANKS 32
90 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
91 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
92 
93 #define emul_to_vcpu(ctxt) \
94 	((struct kvm_vcpu *)(ctxt)->vcpu)
95 
96 /* EFER defaults:
97  * - enable syscall per default because its emulated by KVM
98  * - enable LME and LMA per default on 64 bit KVM
99  */
100 #ifdef CONFIG_X86_64
101 static
102 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
103 #else
104 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
105 #endif
106 
107 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
108 
109 #define KVM_EXIT_HYPERCALL_VALID_MASK (1 << KVM_HC_MAP_GPA_RANGE)
110 
111 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
112                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
113 
114 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
115 static void process_nmi(struct kvm_vcpu *vcpu);
116 static void process_smi(struct kvm_vcpu *vcpu);
117 static void enter_smm(struct kvm_vcpu *vcpu);
118 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
119 static void store_regs(struct kvm_vcpu *vcpu);
120 static int sync_regs(struct kvm_vcpu *vcpu);
121 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu);
122 
123 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
124 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
125 
126 struct kvm_x86_ops kvm_x86_ops __read_mostly;
127 EXPORT_SYMBOL_GPL(kvm_x86_ops);
128 
129 #define KVM_X86_OP(func)					     \
130 	DEFINE_STATIC_CALL_NULL(kvm_x86_##func,			     \
131 				*(((struct kvm_x86_ops *)0)->func));
132 #define KVM_X86_OP_NULL KVM_X86_OP
133 #include <asm/kvm-x86-ops.h>
134 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
135 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
136 EXPORT_STATIC_CALL_GPL(kvm_x86_tlb_flush_current);
137 
138 static bool __read_mostly ignore_msrs = 0;
139 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
140 
141 bool __read_mostly report_ignored_msrs = true;
142 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
143 EXPORT_SYMBOL_GPL(report_ignored_msrs);
144 
145 unsigned int min_timer_period_us = 200;
146 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
147 
148 static bool __read_mostly kvmclock_periodic_sync = true;
149 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
150 
151 bool __read_mostly kvm_has_tsc_control;
152 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
153 u32  __read_mostly kvm_max_guest_tsc_khz;
154 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
155 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
156 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
157 u64  __read_mostly kvm_max_tsc_scaling_ratio;
158 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
159 u64 __read_mostly kvm_default_tsc_scaling_ratio;
160 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
161 bool __read_mostly kvm_has_bus_lock_exit;
162 EXPORT_SYMBOL_GPL(kvm_has_bus_lock_exit);
163 
164 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
165 static u32 __read_mostly tsc_tolerance_ppm = 250;
166 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
167 
168 /*
169  * lapic timer advance (tscdeadline mode only) in nanoseconds.  '-1' enables
170  * adaptive tuning starting from default advancement of 1000ns.  '0' disables
171  * advancement entirely.  Any other value is used as-is and disables adaptive
172  * tuning, i.e. allows privileged userspace to set an exact advancement time.
173  */
174 static int __read_mostly lapic_timer_advance_ns = -1;
175 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
176 
177 static bool __read_mostly vector_hashing = true;
178 module_param(vector_hashing, bool, S_IRUGO);
179 
180 bool __read_mostly enable_vmware_backdoor = false;
181 module_param(enable_vmware_backdoor, bool, S_IRUGO);
182 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
183 
184 static bool __read_mostly force_emulation_prefix = false;
185 module_param(force_emulation_prefix, bool, S_IRUGO);
186 
187 int __read_mostly pi_inject_timer = -1;
188 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
189 
190 /*
191  * Restoring the host value for MSRs that are only consumed when running in
192  * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
193  * returns to userspace, i.e. the kernel can run with the guest's value.
194  */
195 #define KVM_MAX_NR_USER_RETURN_MSRS 16
196 
197 struct kvm_user_return_msrs {
198 	struct user_return_notifier urn;
199 	bool registered;
200 	struct kvm_user_return_msr_values {
201 		u64 host;
202 		u64 curr;
203 	} values[KVM_MAX_NR_USER_RETURN_MSRS];
204 };
205 
206 u32 __read_mostly kvm_nr_uret_msrs;
207 EXPORT_SYMBOL_GPL(kvm_nr_uret_msrs);
208 static u32 __read_mostly kvm_uret_msrs_list[KVM_MAX_NR_USER_RETURN_MSRS];
209 static struct kvm_user_return_msrs __percpu *user_return_msrs;
210 
211 #define KVM_SUPPORTED_XCR0     (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
212 				| XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
213 				| XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
214 				| XFEATURE_MASK_PKRU | XFEATURE_MASK_XTILE)
215 
216 u64 __read_mostly host_efer;
217 EXPORT_SYMBOL_GPL(host_efer);
218 
219 bool __read_mostly allow_smaller_maxphyaddr = 0;
220 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
221 
222 bool __read_mostly enable_apicv = true;
223 EXPORT_SYMBOL_GPL(enable_apicv);
224 
225 u64 __read_mostly host_xss;
226 EXPORT_SYMBOL_GPL(host_xss);
227 u64 __read_mostly supported_xss;
228 EXPORT_SYMBOL_GPL(supported_xss);
229 
230 const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
231 	KVM_GENERIC_VM_STATS(),
232 	STATS_DESC_COUNTER(VM, mmu_shadow_zapped),
233 	STATS_DESC_COUNTER(VM, mmu_pte_write),
234 	STATS_DESC_COUNTER(VM, mmu_pde_zapped),
235 	STATS_DESC_COUNTER(VM, mmu_flooded),
236 	STATS_DESC_COUNTER(VM, mmu_recycled),
237 	STATS_DESC_COUNTER(VM, mmu_cache_miss),
238 	STATS_DESC_ICOUNTER(VM, mmu_unsync),
239 	STATS_DESC_ICOUNTER(VM, pages_4k),
240 	STATS_DESC_ICOUNTER(VM, pages_2m),
241 	STATS_DESC_ICOUNTER(VM, pages_1g),
242 	STATS_DESC_ICOUNTER(VM, nx_lpage_splits),
243 	STATS_DESC_PCOUNTER(VM, max_mmu_rmap_size),
244 	STATS_DESC_PCOUNTER(VM, max_mmu_page_hash_collisions)
245 };
246 
247 const struct kvm_stats_header kvm_vm_stats_header = {
248 	.name_size = KVM_STATS_NAME_SIZE,
249 	.num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
250 	.id_offset = sizeof(struct kvm_stats_header),
251 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
252 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
253 		       sizeof(kvm_vm_stats_desc),
254 };
255 
256 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
257 	KVM_GENERIC_VCPU_STATS(),
258 	STATS_DESC_COUNTER(VCPU, pf_fixed),
259 	STATS_DESC_COUNTER(VCPU, pf_guest),
260 	STATS_DESC_COUNTER(VCPU, tlb_flush),
261 	STATS_DESC_COUNTER(VCPU, invlpg),
262 	STATS_DESC_COUNTER(VCPU, exits),
263 	STATS_DESC_COUNTER(VCPU, io_exits),
264 	STATS_DESC_COUNTER(VCPU, mmio_exits),
265 	STATS_DESC_COUNTER(VCPU, signal_exits),
266 	STATS_DESC_COUNTER(VCPU, irq_window_exits),
267 	STATS_DESC_COUNTER(VCPU, nmi_window_exits),
268 	STATS_DESC_COUNTER(VCPU, l1d_flush),
269 	STATS_DESC_COUNTER(VCPU, halt_exits),
270 	STATS_DESC_COUNTER(VCPU, request_irq_exits),
271 	STATS_DESC_COUNTER(VCPU, irq_exits),
272 	STATS_DESC_COUNTER(VCPU, host_state_reload),
273 	STATS_DESC_COUNTER(VCPU, fpu_reload),
274 	STATS_DESC_COUNTER(VCPU, insn_emulation),
275 	STATS_DESC_COUNTER(VCPU, insn_emulation_fail),
276 	STATS_DESC_COUNTER(VCPU, hypercalls),
277 	STATS_DESC_COUNTER(VCPU, irq_injections),
278 	STATS_DESC_COUNTER(VCPU, nmi_injections),
279 	STATS_DESC_COUNTER(VCPU, req_event),
280 	STATS_DESC_COUNTER(VCPU, nested_run),
281 	STATS_DESC_COUNTER(VCPU, directed_yield_attempted),
282 	STATS_DESC_COUNTER(VCPU, directed_yield_successful),
283 	STATS_DESC_ICOUNTER(VCPU, guest_mode)
284 };
285 
286 const struct kvm_stats_header kvm_vcpu_stats_header = {
287 	.name_size = KVM_STATS_NAME_SIZE,
288 	.num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
289 	.id_offset = sizeof(struct kvm_stats_header),
290 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
291 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
292 		       sizeof(kvm_vcpu_stats_desc),
293 };
294 
295 u64 __read_mostly host_xcr0;
296 u64 __read_mostly supported_xcr0;
297 EXPORT_SYMBOL_GPL(supported_xcr0);
298 
299 static struct kmem_cache *x86_emulator_cache;
300 
301 /*
302  * When called, it means the previous get/set msr reached an invalid msr.
303  * Return true if we want to ignore/silent this failed msr access.
304  */
305 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
306 {
307 	const char *op = write ? "wrmsr" : "rdmsr";
308 
309 	if (ignore_msrs) {
310 		if (report_ignored_msrs)
311 			kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
312 				      op, msr, data);
313 		/* Mask the error */
314 		return true;
315 	} else {
316 		kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
317 				      op, msr, data);
318 		return false;
319 	}
320 }
321 
322 static struct kmem_cache *kvm_alloc_emulator_cache(void)
323 {
324 	unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
325 	unsigned int size = sizeof(struct x86_emulate_ctxt);
326 
327 	return kmem_cache_create_usercopy("x86_emulator", size,
328 					  __alignof__(struct x86_emulate_ctxt),
329 					  SLAB_ACCOUNT, useroffset,
330 					  size - useroffset, NULL);
331 }
332 
333 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
334 
335 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
336 {
337 	int i;
338 	for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
339 		vcpu->arch.apf.gfns[i] = ~0;
340 }
341 
342 static void kvm_on_user_return(struct user_return_notifier *urn)
343 {
344 	unsigned slot;
345 	struct kvm_user_return_msrs *msrs
346 		= container_of(urn, struct kvm_user_return_msrs, urn);
347 	struct kvm_user_return_msr_values *values;
348 	unsigned long flags;
349 
350 	/*
351 	 * Disabling irqs at this point since the following code could be
352 	 * interrupted and executed through kvm_arch_hardware_disable()
353 	 */
354 	local_irq_save(flags);
355 	if (msrs->registered) {
356 		msrs->registered = false;
357 		user_return_notifier_unregister(urn);
358 	}
359 	local_irq_restore(flags);
360 	for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
361 		values = &msrs->values[slot];
362 		if (values->host != values->curr) {
363 			wrmsrl(kvm_uret_msrs_list[slot], values->host);
364 			values->curr = values->host;
365 		}
366 	}
367 }
368 
369 static int kvm_probe_user_return_msr(u32 msr)
370 {
371 	u64 val;
372 	int ret;
373 
374 	preempt_disable();
375 	ret = rdmsrl_safe(msr, &val);
376 	if (ret)
377 		goto out;
378 	ret = wrmsrl_safe(msr, val);
379 out:
380 	preempt_enable();
381 	return ret;
382 }
383 
384 int kvm_add_user_return_msr(u32 msr)
385 {
386 	BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
387 
388 	if (kvm_probe_user_return_msr(msr))
389 		return -1;
390 
391 	kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
392 	return kvm_nr_uret_msrs++;
393 }
394 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
395 
396 int kvm_find_user_return_msr(u32 msr)
397 {
398 	int i;
399 
400 	for (i = 0; i < kvm_nr_uret_msrs; ++i) {
401 		if (kvm_uret_msrs_list[i] == msr)
402 			return i;
403 	}
404 	return -1;
405 }
406 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
407 
408 static void kvm_user_return_msr_cpu_online(void)
409 {
410 	unsigned int cpu = smp_processor_id();
411 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
412 	u64 value;
413 	int i;
414 
415 	for (i = 0; i < kvm_nr_uret_msrs; ++i) {
416 		rdmsrl_safe(kvm_uret_msrs_list[i], &value);
417 		msrs->values[i].host = value;
418 		msrs->values[i].curr = value;
419 	}
420 }
421 
422 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
423 {
424 	unsigned int cpu = smp_processor_id();
425 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
426 	int err;
427 
428 	value = (value & mask) | (msrs->values[slot].host & ~mask);
429 	if (value == msrs->values[slot].curr)
430 		return 0;
431 	err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
432 	if (err)
433 		return 1;
434 
435 	msrs->values[slot].curr = value;
436 	if (!msrs->registered) {
437 		msrs->urn.on_user_return = kvm_on_user_return;
438 		user_return_notifier_register(&msrs->urn);
439 		msrs->registered = true;
440 	}
441 	return 0;
442 }
443 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
444 
445 static void drop_user_return_notifiers(void)
446 {
447 	unsigned int cpu = smp_processor_id();
448 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
449 
450 	if (msrs->registered)
451 		kvm_on_user_return(&msrs->urn);
452 }
453 
454 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
455 {
456 	return vcpu->arch.apic_base;
457 }
458 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
459 
460 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
461 {
462 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
463 }
464 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
465 
466 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
467 {
468 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
469 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
470 	u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
471 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
472 
473 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
474 		return 1;
475 	if (!msr_info->host_initiated) {
476 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
477 			return 1;
478 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
479 			return 1;
480 	}
481 
482 	kvm_lapic_set_base(vcpu, msr_info->data);
483 	kvm_recalculate_apic_map(vcpu->kvm);
484 	return 0;
485 }
486 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
487 
488 /*
489  * Handle a fault on a hardware virtualization (VMX or SVM) instruction.
490  *
491  * Hardware virtualization extension instructions may fault if a reboot turns
492  * off virtualization while processes are running.  Usually after catching the
493  * fault we just panic; during reboot instead the instruction is ignored.
494  */
495 noinstr void kvm_spurious_fault(void)
496 {
497 	/* Fault while not rebooting.  We want the trace. */
498 	BUG_ON(!kvm_rebooting);
499 }
500 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
501 
502 #define EXCPT_BENIGN		0
503 #define EXCPT_CONTRIBUTORY	1
504 #define EXCPT_PF		2
505 
506 static int exception_class(int vector)
507 {
508 	switch (vector) {
509 	case PF_VECTOR:
510 		return EXCPT_PF;
511 	case DE_VECTOR:
512 	case TS_VECTOR:
513 	case NP_VECTOR:
514 	case SS_VECTOR:
515 	case GP_VECTOR:
516 		return EXCPT_CONTRIBUTORY;
517 	default:
518 		break;
519 	}
520 	return EXCPT_BENIGN;
521 }
522 
523 #define EXCPT_FAULT		0
524 #define EXCPT_TRAP		1
525 #define EXCPT_ABORT		2
526 #define EXCPT_INTERRUPT		3
527 
528 static int exception_type(int vector)
529 {
530 	unsigned int mask;
531 
532 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
533 		return EXCPT_INTERRUPT;
534 
535 	mask = 1 << vector;
536 
537 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
538 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
539 		return EXCPT_TRAP;
540 
541 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
542 		return EXCPT_ABORT;
543 
544 	/* Reserved exceptions will result in fault */
545 	return EXCPT_FAULT;
546 }
547 
548 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
549 {
550 	unsigned nr = vcpu->arch.exception.nr;
551 	bool has_payload = vcpu->arch.exception.has_payload;
552 	unsigned long payload = vcpu->arch.exception.payload;
553 
554 	if (!has_payload)
555 		return;
556 
557 	switch (nr) {
558 	case DB_VECTOR:
559 		/*
560 		 * "Certain debug exceptions may clear bit 0-3.  The
561 		 * remaining contents of the DR6 register are never
562 		 * cleared by the processor".
563 		 */
564 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
565 		/*
566 		 * In order to reflect the #DB exception payload in guest
567 		 * dr6, three components need to be considered: active low
568 		 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
569 		 * DR6_BS and DR6_BT)
570 		 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
571 		 * In the target guest dr6:
572 		 * FIXED_1 bits should always be set.
573 		 * Active low bits should be cleared if 1-setting in payload.
574 		 * Active high bits should be set if 1-setting in payload.
575 		 *
576 		 * Note, the payload is compatible with the pending debug
577 		 * exceptions/exit qualification under VMX, that active_low bits
578 		 * are active high in payload.
579 		 * So they need to be flipped for DR6.
580 		 */
581 		vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
582 		vcpu->arch.dr6 |= payload;
583 		vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
584 
585 		/*
586 		 * The #DB payload is defined as compatible with the 'pending
587 		 * debug exceptions' field under VMX, not DR6. While bit 12 is
588 		 * defined in the 'pending debug exceptions' field (enabled
589 		 * breakpoint), it is reserved and must be zero in DR6.
590 		 */
591 		vcpu->arch.dr6 &= ~BIT(12);
592 		break;
593 	case PF_VECTOR:
594 		vcpu->arch.cr2 = payload;
595 		break;
596 	}
597 
598 	vcpu->arch.exception.has_payload = false;
599 	vcpu->arch.exception.payload = 0;
600 }
601 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
602 
603 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
604 		unsigned nr, bool has_error, u32 error_code,
605 	        bool has_payload, unsigned long payload, bool reinject)
606 {
607 	u32 prev_nr;
608 	int class1, class2;
609 
610 	kvm_make_request(KVM_REQ_EVENT, vcpu);
611 
612 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
613 	queue:
614 		if (reinject) {
615 			/*
616 			 * On vmentry, vcpu->arch.exception.pending is only
617 			 * true if an event injection was blocked by
618 			 * nested_run_pending.  In that case, however,
619 			 * vcpu_enter_guest requests an immediate exit,
620 			 * and the guest shouldn't proceed far enough to
621 			 * need reinjection.
622 			 */
623 			WARN_ON_ONCE(vcpu->arch.exception.pending);
624 			vcpu->arch.exception.injected = true;
625 			if (WARN_ON_ONCE(has_payload)) {
626 				/*
627 				 * A reinjected event has already
628 				 * delivered its payload.
629 				 */
630 				has_payload = false;
631 				payload = 0;
632 			}
633 		} else {
634 			vcpu->arch.exception.pending = true;
635 			vcpu->arch.exception.injected = false;
636 		}
637 		vcpu->arch.exception.has_error_code = has_error;
638 		vcpu->arch.exception.nr = nr;
639 		vcpu->arch.exception.error_code = error_code;
640 		vcpu->arch.exception.has_payload = has_payload;
641 		vcpu->arch.exception.payload = payload;
642 		if (!is_guest_mode(vcpu))
643 			kvm_deliver_exception_payload(vcpu);
644 		return;
645 	}
646 
647 	/* to check exception */
648 	prev_nr = vcpu->arch.exception.nr;
649 	if (prev_nr == DF_VECTOR) {
650 		/* triple fault -> shutdown */
651 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
652 		return;
653 	}
654 	class1 = exception_class(prev_nr);
655 	class2 = exception_class(nr);
656 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
657 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
658 		/*
659 		 * Generate double fault per SDM Table 5-5.  Set
660 		 * exception.pending = true so that the double fault
661 		 * can trigger a nested vmexit.
662 		 */
663 		vcpu->arch.exception.pending = true;
664 		vcpu->arch.exception.injected = false;
665 		vcpu->arch.exception.has_error_code = true;
666 		vcpu->arch.exception.nr = DF_VECTOR;
667 		vcpu->arch.exception.error_code = 0;
668 		vcpu->arch.exception.has_payload = false;
669 		vcpu->arch.exception.payload = 0;
670 	} else
671 		/* replace previous exception with a new one in a hope
672 		   that instruction re-execution will regenerate lost
673 		   exception */
674 		goto queue;
675 }
676 
677 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
678 {
679 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
680 }
681 EXPORT_SYMBOL_GPL(kvm_queue_exception);
682 
683 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
684 {
685 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
686 }
687 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
688 
689 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
690 			   unsigned long payload)
691 {
692 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
693 }
694 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
695 
696 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
697 				    u32 error_code, unsigned long payload)
698 {
699 	kvm_multiple_exception(vcpu, nr, true, error_code,
700 			       true, payload, false);
701 }
702 
703 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
704 {
705 	if (err)
706 		kvm_inject_gp(vcpu, 0);
707 	else
708 		return kvm_skip_emulated_instruction(vcpu);
709 
710 	return 1;
711 }
712 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
713 
714 static int complete_emulated_insn_gp(struct kvm_vcpu *vcpu, int err)
715 {
716 	if (err) {
717 		kvm_inject_gp(vcpu, 0);
718 		return 1;
719 	}
720 
721 	return kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE | EMULTYPE_SKIP |
722 				       EMULTYPE_COMPLETE_USER_EXIT);
723 }
724 
725 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
726 {
727 	++vcpu->stat.pf_guest;
728 	vcpu->arch.exception.nested_apf =
729 		is_guest_mode(vcpu) && fault->async_page_fault;
730 	if (vcpu->arch.exception.nested_apf) {
731 		vcpu->arch.apf.nested_apf_token = fault->address;
732 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
733 	} else {
734 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
735 					fault->address);
736 	}
737 }
738 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
739 
740 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
741 				    struct x86_exception *fault)
742 {
743 	struct kvm_mmu *fault_mmu;
744 	WARN_ON_ONCE(fault->vector != PF_VECTOR);
745 
746 	fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
747 					       vcpu->arch.walk_mmu;
748 
749 	/*
750 	 * Invalidate the TLB entry for the faulting address, if it exists,
751 	 * else the access will fault indefinitely (and to emulate hardware).
752 	 */
753 	if ((fault->error_code & PFERR_PRESENT_MASK) &&
754 	    !(fault->error_code & PFERR_RSVD_MASK))
755 		kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
756 				       fault_mmu->root_hpa);
757 
758 	fault_mmu->inject_page_fault(vcpu, fault);
759 	return fault->nested_page_fault;
760 }
761 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
762 
763 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
764 {
765 	atomic_inc(&vcpu->arch.nmi_queued);
766 	kvm_make_request(KVM_REQ_NMI, vcpu);
767 }
768 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
769 
770 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
771 {
772 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
773 }
774 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
775 
776 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
777 {
778 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
779 }
780 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
781 
782 /*
783  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
784  * a #GP and return false.
785  */
786 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
787 {
788 	if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
789 		return true;
790 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
791 	return false;
792 }
793 EXPORT_SYMBOL_GPL(kvm_require_cpl);
794 
795 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
796 {
797 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
798 		return true;
799 
800 	kvm_queue_exception(vcpu, UD_VECTOR);
801 	return false;
802 }
803 EXPORT_SYMBOL_GPL(kvm_require_dr);
804 
805 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
806 {
807 	return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
808 }
809 
810 /*
811  * Load the pae pdptrs.  Return 1 if they are all valid, 0 otherwise.
812  */
813 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
814 {
815 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
816 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
817 	gpa_t real_gpa;
818 	int i;
819 	int ret;
820 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
821 
822 	/*
823 	 * If the MMU is nested, CR3 holds an L2 GPA and needs to be translated
824 	 * to an L1 GPA.
825 	 */
826 	real_gpa = kvm_translate_gpa(vcpu, mmu, gfn_to_gpa(pdpt_gfn),
827 				     PFERR_USER_MASK | PFERR_WRITE_MASK, NULL);
828 	if (real_gpa == UNMAPPED_GVA)
829 		return 0;
830 
831 	/* Note the offset, PDPTRs are 32 byte aligned when using PAE paging. */
832 	ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(real_gpa), pdpte,
833 				       cr3 & GENMASK(11, 5), sizeof(pdpte));
834 	if (ret < 0)
835 		return 0;
836 
837 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
838 		if ((pdpte[i] & PT_PRESENT_MASK) &&
839 		    (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
840 			return 0;
841 		}
842 	}
843 
844 	/*
845 	 * Marking VCPU_EXREG_PDPTR dirty doesn't work for !tdp_enabled.
846 	 * Shadow page roots need to be reconstructed instead.
847 	 */
848 	if (!tdp_enabled && memcmp(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs)))
849 		kvm_mmu_free_roots(vcpu, mmu, KVM_MMU_ROOT_CURRENT);
850 
851 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
852 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
853 	kvm_make_request(KVM_REQ_LOAD_MMU_PGD, vcpu);
854 	vcpu->arch.pdptrs_from_userspace = false;
855 
856 	return 1;
857 }
858 EXPORT_SYMBOL_GPL(load_pdptrs);
859 
860 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
861 {
862 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
863 		kvm_clear_async_pf_completion_queue(vcpu);
864 		kvm_async_pf_hash_reset(vcpu);
865 	}
866 
867 	if ((cr0 ^ old_cr0) & KVM_MMU_CR0_ROLE_BITS)
868 		kvm_mmu_reset_context(vcpu);
869 
870 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
871 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
872 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
873 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
874 }
875 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
876 
877 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
878 {
879 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
880 
881 	cr0 |= X86_CR0_ET;
882 
883 #ifdef CONFIG_X86_64
884 	if (cr0 & 0xffffffff00000000UL)
885 		return 1;
886 #endif
887 
888 	cr0 &= ~CR0_RESERVED_BITS;
889 
890 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
891 		return 1;
892 
893 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
894 		return 1;
895 
896 #ifdef CONFIG_X86_64
897 	if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
898 	    (cr0 & X86_CR0_PG)) {
899 		int cs_db, cs_l;
900 
901 		if (!is_pae(vcpu))
902 			return 1;
903 		static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
904 		if (cs_l)
905 			return 1;
906 	}
907 #endif
908 	if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
909 	    is_pae(vcpu) && ((cr0 ^ old_cr0) & X86_CR0_PDPTR_BITS) &&
910 	    !load_pdptrs(vcpu, kvm_read_cr3(vcpu)))
911 		return 1;
912 
913 	if (!(cr0 & X86_CR0_PG) &&
914 	    (is_64_bit_mode(vcpu) || kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)))
915 		return 1;
916 
917 	static_call(kvm_x86_set_cr0)(vcpu, cr0);
918 
919 	kvm_post_set_cr0(vcpu, old_cr0, cr0);
920 
921 	return 0;
922 }
923 EXPORT_SYMBOL_GPL(kvm_set_cr0);
924 
925 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
926 {
927 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
928 }
929 EXPORT_SYMBOL_GPL(kvm_lmsw);
930 
931 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
932 {
933 	if (vcpu->arch.guest_state_protected)
934 		return;
935 
936 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
937 
938 		if (vcpu->arch.xcr0 != host_xcr0)
939 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
940 
941 		if (vcpu->arch.xsaves_enabled &&
942 		    vcpu->arch.ia32_xss != host_xss)
943 			wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
944 	}
945 
946 	if (static_cpu_has(X86_FEATURE_PKU) &&
947 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
948 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
949 	    vcpu->arch.pkru != vcpu->arch.host_pkru)
950 		write_pkru(vcpu->arch.pkru);
951 }
952 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
953 
954 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
955 {
956 	if (vcpu->arch.guest_state_protected)
957 		return;
958 
959 	if (static_cpu_has(X86_FEATURE_PKU) &&
960 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
961 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
962 		vcpu->arch.pkru = rdpkru();
963 		if (vcpu->arch.pkru != vcpu->arch.host_pkru)
964 			write_pkru(vcpu->arch.host_pkru);
965 	}
966 
967 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
968 
969 		if (vcpu->arch.xcr0 != host_xcr0)
970 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
971 
972 		if (vcpu->arch.xsaves_enabled &&
973 		    vcpu->arch.ia32_xss != host_xss)
974 			wrmsrl(MSR_IA32_XSS, host_xss);
975 	}
976 
977 }
978 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
979 
980 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
981 {
982 	u64 xcr0 = xcr;
983 	u64 old_xcr0 = vcpu->arch.xcr0;
984 	u64 valid_bits;
985 
986 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
987 	if (index != XCR_XFEATURE_ENABLED_MASK)
988 		return 1;
989 	if (!(xcr0 & XFEATURE_MASK_FP))
990 		return 1;
991 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
992 		return 1;
993 
994 	/*
995 	 * Do not allow the guest to set bits that we do not support
996 	 * saving.  However, xcr0 bit 0 is always set, even if the
997 	 * emulated CPU does not support XSAVE (see kvm_vcpu_reset()).
998 	 */
999 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
1000 	if (xcr0 & ~valid_bits)
1001 		return 1;
1002 
1003 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
1004 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
1005 		return 1;
1006 
1007 	if (xcr0 & XFEATURE_MASK_AVX512) {
1008 		if (!(xcr0 & XFEATURE_MASK_YMM))
1009 			return 1;
1010 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
1011 			return 1;
1012 	}
1013 
1014 	if ((xcr0 & XFEATURE_MASK_XTILE) &&
1015 	    ((xcr0 & XFEATURE_MASK_XTILE) != XFEATURE_MASK_XTILE))
1016 		return 1;
1017 
1018 	vcpu->arch.xcr0 = xcr0;
1019 
1020 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
1021 		kvm_update_cpuid_runtime(vcpu);
1022 	return 0;
1023 }
1024 
1025 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
1026 {
1027 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
1028 	    __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
1029 		kvm_inject_gp(vcpu, 0);
1030 		return 1;
1031 	}
1032 
1033 	return kvm_skip_emulated_instruction(vcpu);
1034 }
1035 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1036 
1037 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1038 {
1039 	if (cr4 & cr4_reserved_bits)
1040 		return false;
1041 
1042 	if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1043 		return false;
1044 
1045 	return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1046 }
1047 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
1048 
1049 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1050 {
1051 	/*
1052 	 * If any role bit is changed, the MMU needs to be reset.
1053 	 *
1054 	 * If CR4.PCIDE is changed 1 -> 0, the guest TLB must be flushed.
1055 	 * If CR4.PCIDE is changed 0 -> 1, there is no need to flush the TLB
1056 	 * according to the SDM; however, stale prev_roots could be reused
1057 	 * incorrectly in the future after a MOV to CR3 with NOFLUSH=1, so we
1058 	 * free them all.  KVM_REQ_MMU_RELOAD is fit for the both cases; it
1059 	 * is slow, but changing CR4.PCIDE is a rare case.
1060 	 *
1061 	 * If CR4.PGE is changed, the guest TLB must be flushed.
1062 	 *
1063 	 * Note: resetting MMU is a superset of KVM_REQ_MMU_RELOAD and
1064 	 * KVM_REQ_MMU_RELOAD is a superset of KVM_REQ_TLB_FLUSH_GUEST, hence
1065 	 * the usage of "else if".
1066 	 */
1067 	if ((cr4 ^ old_cr4) & KVM_MMU_CR4_ROLE_BITS)
1068 		kvm_mmu_reset_context(vcpu);
1069 	else if ((cr4 ^ old_cr4) & X86_CR4_PCIDE)
1070 		kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
1071 	else if ((cr4 ^ old_cr4) & X86_CR4_PGE)
1072 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1073 }
1074 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1075 
1076 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1077 {
1078 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
1079 
1080 	if (!kvm_is_valid_cr4(vcpu, cr4))
1081 		return 1;
1082 
1083 	if (is_long_mode(vcpu)) {
1084 		if (!(cr4 & X86_CR4_PAE))
1085 			return 1;
1086 		if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1087 			return 1;
1088 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1089 		   && ((cr4 ^ old_cr4) & X86_CR4_PDPTR_BITS)
1090 		   && !load_pdptrs(vcpu, kvm_read_cr3(vcpu)))
1091 		return 1;
1092 
1093 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1094 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1095 			return 1;
1096 
1097 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1098 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1099 			return 1;
1100 	}
1101 
1102 	static_call(kvm_x86_set_cr4)(vcpu, cr4);
1103 
1104 	kvm_post_set_cr4(vcpu, old_cr4, cr4);
1105 
1106 	return 0;
1107 }
1108 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1109 
1110 static void kvm_invalidate_pcid(struct kvm_vcpu *vcpu, unsigned long pcid)
1111 {
1112 	struct kvm_mmu *mmu = vcpu->arch.mmu;
1113 	unsigned long roots_to_free = 0;
1114 	int i;
1115 
1116 	/*
1117 	 * MOV CR3 and INVPCID are usually not intercepted when using TDP, but
1118 	 * this is reachable when running EPT=1 and unrestricted_guest=0,  and
1119 	 * also via the emulator.  KVM's TDP page tables are not in the scope of
1120 	 * the invalidation, but the guest's TLB entries need to be flushed as
1121 	 * the CPU may have cached entries in its TLB for the target PCID.
1122 	 */
1123 	if (unlikely(tdp_enabled)) {
1124 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1125 		return;
1126 	}
1127 
1128 	/*
1129 	 * If neither the current CR3 nor any of the prev_roots use the given
1130 	 * PCID, then nothing needs to be done here because a resync will
1131 	 * happen anyway before switching to any other CR3.
1132 	 */
1133 	if (kvm_get_active_pcid(vcpu) == pcid) {
1134 		kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
1135 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1136 	}
1137 
1138 	/*
1139 	 * If PCID is disabled, there is no need to free prev_roots even if the
1140 	 * PCIDs for them are also 0, because MOV to CR3 always flushes the TLB
1141 	 * with PCIDE=0.
1142 	 */
1143 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
1144 		return;
1145 
1146 	for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
1147 		if (kvm_get_pcid(vcpu, mmu->prev_roots[i].pgd) == pcid)
1148 			roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
1149 
1150 	kvm_mmu_free_roots(vcpu, mmu, roots_to_free);
1151 }
1152 
1153 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1154 {
1155 	bool skip_tlb_flush = false;
1156 	unsigned long pcid = 0;
1157 #ifdef CONFIG_X86_64
1158 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1159 
1160 	if (pcid_enabled) {
1161 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1162 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
1163 		pcid = cr3 & X86_CR3_PCID_MASK;
1164 	}
1165 #endif
1166 
1167 	/* PDPTRs are always reloaded for PAE paging. */
1168 	if (cr3 == kvm_read_cr3(vcpu) && !is_pae_paging(vcpu))
1169 		goto handle_tlb_flush;
1170 
1171 	/*
1172 	 * Do not condition the GPA check on long mode, this helper is used to
1173 	 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1174 	 * the current vCPU mode is accurate.
1175 	 */
1176 	if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1177 		return 1;
1178 
1179 	if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, cr3))
1180 		return 1;
1181 
1182 	if (cr3 != kvm_read_cr3(vcpu))
1183 		kvm_mmu_new_pgd(vcpu, cr3);
1184 
1185 	vcpu->arch.cr3 = cr3;
1186 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
1187 	/* Do not call post_set_cr3, we do not get here for confidential guests.  */
1188 
1189 handle_tlb_flush:
1190 	/*
1191 	 * A load of CR3 that flushes the TLB flushes only the current PCID,
1192 	 * even if PCID is disabled, in which case PCID=0 is flushed.  It's a
1193 	 * moot point in the end because _disabling_ PCID will flush all PCIDs,
1194 	 * and it's impossible to use a non-zero PCID when PCID is disabled,
1195 	 * i.e. only PCID=0 can be relevant.
1196 	 */
1197 	if (!skip_tlb_flush)
1198 		kvm_invalidate_pcid(vcpu, pcid);
1199 
1200 	return 0;
1201 }
1202 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1203 
1204 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1205 {
1206 	if (cr8 & CR8_RESERVED_BITS)
1207 		return 1;
1208 	if (lapic_in_kernel(vcpu))
1209 		kvm_lapic_set_tpr(vcpu, cr8);
1210 	else
1211 		vcpu->arch.cr8 = cr8;
1212 	return 0;
1213 }
1214 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1215 
1216 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1217 {
1218 	if (lapic_in_kernel(vcpu))
1219 		return kvm_lapic_get_cr8(vcpu);
1220 	else
1221 		return vcpu->arch.cr8;
1222 }
1223 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1224 
1225 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1226 {
1227 	int i;
1228 
1229 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1230 		for (i = 0; i < KVM_NR_DB_REGS; i++)
1231 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1232 	}
1233 }
1234 
1235 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1236 {
1237 	unsigned long dr7;
1238 
1239 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1240 		dr7 = vcpu->arch.guest_debug_dr7;
1241 	else
1242 		dr7 = vcpu->arch.dr7;
1243 	static_call(kvm_x86_set_dr7)(vcpu, dr7);
1244 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1245 	if (dr7 & DR7_BP_EN_MASK)
1246 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1247 }
1248 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1249 
1250 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1251 {
1252 	u64 fixed = DR6_FIXED_1;
1253 
1254 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1255 		fixed |= DR6_RTM;
1256 
1257 	if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1258 		fixed |= DR6_BUS_LOCK;
1259 	return fixed;
1260 }
1261 
1262 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1263 {
1264 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1265 
1266 	switch (dr) {
1267 	case 0 ... 3:
1268 		vcpu->arch.db[array_index_nospec(dr, size)] = val;
1269 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1270 			vcpu->arch.eff_db[dr] = val;
1271 		break;
1272 	case 4:
1273 	case 6:
1274 		if (!kvm_dr6_valid(val))
1275 			return 1; /* #GP */
1276 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1277 		break;
1278 	case 5:
1279 	default: /* 7 */
1280 		if (!kvm_dr7_valid(val))
1281 			return 1; /* #GP */
1282 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1283 		kvm_update_dr7(vcpu);
1284 		break;
1285 	}
1286 
1287 	return 0;
1288 }
1289 EXPORT_SYMBOL_GPL(kvm_set_dr);
1290 
1291 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1292 {
1293 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1294 
1295 	switch (dr) {
1296 	case 0 ... 3:
1297 		*val = vcpu->arch.db[array_index_nospec(dr, size)];
1298 		break;
1299 	case 4:
1300 	case 6:
1301 		*val = vcpu->arch.dr6;
1302 		break;
1303 	case 5:
1304 	default: /* 7 */
1305 		*val = vcpu->arch.dr7;
1306 		break;
1307 	}
1308 }
1309 EXPORT_SYMBOL_GPL(kvm_get_dr);
1310 
1311 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1312 {
1313 	u32 ecx = kvm_rcx_read(vcpu);
1314 	u64 data;
1315 
1316 	if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1317 		kvm_inject_gp(vcpu, 0);
1318 		return 1;
1319 	}
1320 
1321 	kvm_rax_write(vcpu, (u32)data);
1322 	kvm_rdx_write(vcpu, data >> 32);
1323 	return kvm_skip_emulated_instruction(vcpu);
1324 }
1325 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1326 
1327 /*
1328  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1329  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1330  *
1331  * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1332  * extract the supported MSRs from the related const lists.
1333  * msrs_to_save is selected from the msrs_to_save_all to reflect the
1334  * capabilities of the host cpu. This capabilities test skips MSRs that are
1335  * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1336  * may depend on host virtualization features rather than host cpu features.
1337  */
1338 
1339 static const u32 msrs_to_save_all[] = {
1340 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1341 	MSR_STAR,
1342 #ifdef CONFIG_X86_64
1343 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1344 #endif
1345 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1346 	MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1347 	MSR_IA32_SPEC_CTRL,
1348 	MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1349 	MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1350 	MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1351 	MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1352 	MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1353 	MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1354 	MSR_IA32_UMWAIT_CONTROL,
1355 
1356 	MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1357 	MSR_ARCH_PERFMON_FIXED_CTR0 + 2,
1358 	MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1359 	MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1360 	MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1361 	MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1362 	MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1363 	MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1364 	MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1365 	MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1366 	MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1367 	MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1368 	MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1369 	MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1370 	MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1371 	MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1372 	MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1373 	MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1374 	MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1375 	MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1376 	MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1377 	MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1378 
1379 	MSR_K7_EVNTSEL0, MSR_K7_EVNTSEL1, MSR_K7_EVNTSEL2, MSR_K7_EVNTSEL3,
1380 	MSR_K7_PERFCTR0, MSR_K7_PERFCTR1, MSR_K7_PERFCTR2, MSR_K7_PERFCTR3,
1381 	MSR_F15H_PERF_CTL0, MSR_F15H_PERF_CTL1, MSR_F15H_PERF_CTL2,
1382 	MSR_F15H_PERF_CTL3, MSR_F15H_PERF_CTL4, MSR_F15H_PERF_CTL5,
1383 	MSR_F15H_PERF_CTR0, MSR_F15H_PERF_CTR1, MSR_F15H_PERF_CTR2,
1384 	MSR_F15H_PERF_CTR3, MSR_F15H_PERF_CTR4, MSR_F15H_PERF_CTR5,
1385 	MSR_IA32_XFD, MSR_IA32_XFD_ERR,
1386 };
1387 
1388 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1389 static unsigned num_msrs_to_save;
1390 
1391 static const u32 emulated_msrs_all[] = {
1392 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1393 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1394 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1395 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1396 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1397 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1398 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1399 	HV_X64_MSR_RESET,
1400 	HV_X64_MSR_VP_INDEX,
1401 	HV_X64_MSR_VP_RUNTIME,
1402 	HV_X64_MSR_SCONTROL,
1403 	HV_X64_MSR_STIMER0_CONFIG,
1404 	HV_X64_MSR_VP_ASSIST_PAGE,
1405 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1406 	HV_X64_MSR_TSC_EMULATION_STATUS,
1407 	HV_X64_MSR_SYNDBG_OPTIONS,
1408 	HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1409 	HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1410 	HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1411 
1412 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1413 	MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1414 
1415 	MSR_IA32_TSC_ADJUST,
1416 	MSR_IA32_TSC_DEADLINE,
1417 	MSR_IA32_ARCH_CAPABILITIES,
1418 	MSR_IA32_PERF_CAPABILITIES,
1419 	MSR_IA32_MISC_ENABLE,
1420 	MSR_IA32_MCG_STATUS,
1421 	MSR_IA32_MCG_CTL,
1422 	MSR_IA32_MCG_EXT_CTL,
1423 	MSR_IA32_SMBASE,
1424 	MSR_SMI_COUNT,
1425 	MSR_PLATFORM_INFO,
1426 	MSR_MISC_FEATURES_ENABLES,
1427 	MSR_AMD64_VIRT_SPEC_CTRL,
1428 	MSR_AMD64_TSC_RATIO,
1429 	MSR_IA32_POWER_CTL,
1430 	MSR_IA32_UCODE_REV,
1431 
1432 	/*
1433 	 * The following list leaves out MSRs whose values are determined
1434 	 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1435 	 * We always support the "true" VMX control MSRs, even if the host
1436 	 * processor does not, so I am putting these registers here rather
1437 	 * than in msrs_to_save_all.
1438 	 */
1439 	MSR_IA32_VMX_BASIC,
1440 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1441 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1442 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1443 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1444 	MSR_IA32_VMX_MISC,
1445 	MSR_IA32_VMX_CR0_FIXED0,
1446 	MSR_IA32_VMX_CR4_FIXED0,
1447 	MSR_IA32_VMX_VMCS_ENUM,
1448 	MSR_IA32_VMX_PROCBASED_CTLS2,
1449 	MSR_IA32_VMX_EPT_VPID_CAP,
1450 	MSR_IA32_VMX_VMFUNC,
1451 
1452 	MSR_K7_HWCR,
1453 	MSR_KVM_POLL_CONTROL,
1454 };
1455 
1456 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1457 static unsigned num_emulated_msrs;
1458 
1459 /*
1460  * List of msr numbers which are used to expose MSR-based features that
1461  * can be used by a hypervisor to validate requested CPU features.
1462  */
1463 static const u32 msr_based_features_all[] = {
1464 	MSR_IA32_VMX_BASIC,
1465 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1466 	MSR_IA32_VMX_PINBASED_CTLS,
1467 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1468 	MSR_IA32_VMX_PROCBASED_CTLS,
1469 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1470 	MSR_IA32_VMX_EXIT_CTLS,
1471 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1472 	MSR_IA32_VMX_ENTRY_CTLS,
1473 	MSR_IA32_VMX_MISC,
1474 	MSR_IA32_VMX_CR0_FIXED0,
1475 	MSR_IA32_VMX_CR0_FIXED1,
1476 	MSR_IA32_VMX_CR4_FIXED0,
1477 	MSR_IA32_VMX_CR4_FIXED1,
1478 	MSR_IA32_VMX_VMCS_ENUM,
1479 	MSR_IA32_VMX_PROCBASED_CTLS2,
1480 	MSR_IA32_VMX_EPT_VPID_CAP,
1481 	MSR_IA32_VMX_VMFUNC,
1482 
1483 	MSR_F10H_DECFG,
1484 	MSR_IA32_UCODE_REV,
1485 	MSR_IA32_ARCH_CAPABILITIES,
1486 	MSR_IA32_PERF_CAPABILITIES,
1487 };
1488 
1489 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1490 static unsigned int num_msr_based_features;
1491 
1492 static u64 kvm_get_arch_capabilities(void)
1493 {
1494 	u64 data = 0;
1495 
1496 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1497 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1498 
1499 	/*
1500 	 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1501 	 * the nested hypervisor runs with NX huge pages.  If it is not,
1502 	 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1503 	 * L1 guests, so it need not worry about its own (L2) guests.
1504 	 */
1505 	data |= ARCH_CAP_PSCHANGE_MC_NO;
1506 
1507 	/*
1508 	 * If we're doing cache flushes (either "always" or "cond")
1509 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1510 	 * If an outer hypervisor is doing the cache flush for us
1511 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1512 	 * capability to the guest too, and if EPT is disabled we're not
1513 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1514 	 * require a nested hypervisor to do a flush of its own.
1515 	 */
1516 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1517 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1518 
1519 	if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1520 		data |= ARCH_CAP_RDCL_NO;
1521 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1522 		data |= ARCH_CAP_SSB_NO;
1523 	if (!boot_cpu_has_bug(X86_BUG_MDS))
1524 		data |= ARCH_CAP_MDS_NO;
1525 
1526 	if (!boot_cpu_has(X86_FEATURE_RTM)) {
1527 		/*
1528 		 * If RTM=0 because the kernel has disabled TSX, the host might
1529 		 * have TAA_NO or TSX_CTRL.  Clear TAA_NO (the guest sees RTM=0
1530 		 * and therefore knows that there cannot be TAA) but keep
1531 		 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1532 		 * and we want to allow migrating those guests to tsx=off hosts.
1533 		 */
1534 		data &= ~ARCH_CAP_TAA_NO;
1535 	} else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1536 		data |= ARCH_CAP_TAA_NO;
1537 	} else {
1538 		/*
1539 		 * Nothing to do here; we emulate TSX_CTRL if present on the
1540 		 * host so the guest can choose between disabling TSX or
1541 		 * using VERW to clear CPU buffers.
1542 		 */
1543 	}
1544 
1545 	return data;
1546 }
1547 
1548 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1549 {
1550 	switch (msr->index) {
1551 	case MSR_IA32_ARCH_CAPABILITIES:
1552 		msr->data = kvm_get_arch_capabilities();
1553 		break;
1554 	case MSR_IA32_UCODE_REV:
1555 		rdmsrl_safe(msr->index, &msr->data);
1556 		break;
1557 	default:
1558 		return static_call(kvm_x86_get_msr_feature)(msr);
1559 	}
1560 	return 0;
1561 }
1562 
1563 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1564 {
1565 	struct kvm_msr_entry msr;
1566 	int r;
1567 
1568 	msr.index = index;
1569 	r = kvm_get_msr_feature(&msr);
1570 
1571 	if (r == KVM_MSR_RET_INVALID) {
1572 		/* Unconditionally clear the output for simplicity */
1573 		*data = 0;
1574 		if (kvm_msr_ignored_check(index, 0, false))
1575 			r = 0;
1576 	}
1577 
1578 	if (r)
1579 		return r;
1580 
1581 	*data = msr.data;
1582 
1583 	return 0;
1584 }
1585 
1586 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1587 {
1588 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1589 		return false;
1590 
1591 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1592 		return false;
1593 
1594 	if (efer & (EFER_LME | EFER_LMA) &&
1595 	    !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1596 		return false;
1597 
1598 	if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1599 		return false;
1600 
1601 	return true;
1602 
1603 }
1604 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1605 {
1606 	if (efer & efer_reserved_bits)
1607 		return false;
1608 
1609 	return __kvm_valid_efer(vcpu, efer);
1610 }
1611 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1612 
1613 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1614 {
1615 	u64 old_efer = vcpu->arch.efer;
1616 	u64 efer = msr_info->data;
1617 	int r;
1618 
1619 	if (efer & efer_reserved_bits)
1620 		return 1;
1621 
1622 	if (!msr_info->host_initiated) {
1623 		if (!__kvm_valid_efer(vcpu, efer))
1624 			return 1;
1625 
1626 		if (is_paging(vcpu) &&
1627 		    (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1628 			return 1;
1629 	}
1630 
1631 	efer &= ~EFER_LMA;
1632 	efer |= vcpu->arch.efer & EFER_LMA;
1633 
1634 	r = static_call(kvm_x86_set_efer)(vcpu, efer);
1635 	if (r) {
1636 		WARN_ON(r > 0);
1637 		return r;
1638 	}
1639 
1640 	/* Update reserved bits */
1641 	if ((efer ^ old_efer) & EFER_NX)
1642 		kvm_mmu_reset_context(vcpu);
1643 
1644 	return 0;
1645 }
1646 
1647 void kvm_enable_efer_bits(u64 mask)
1648 {
1649        efer_reserved_bits &= ~mask;
1650 }
1651 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1652 
1653 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1654 {
1655 	struct kvm_x86_msr_filter *msr_filter;
1656 	struct msr_bitmap_range *ranges;
1657 	struct kvm *kvm = vcpu->kvm;
1658 	bool allowed;
1659 	int idx;
1660 	u32 i;
1661 
1662 	/* x2APIC MSRs do not support filtering. */
1663 	if (index >= 0x800 && index <= 0x8ff)
1664 		return true;
1665 
1666 	idx = srcu_read_lock(&kvm->srcu);
1667 
1668 	msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1669 	if (!msr_filter) {
1670 		allowed = true;
1671 		goto out;
1672 	}
1673 
1674 	allowed = msr_filter->default_allow;
1675 	ranges = msr_filter->ranges;
1676 
1677 	for (i = 0; i < msr_filter->count; i++) {
1678 		u32 start = ranges[i].base;
1679 		u32 end = start + ranges[i].nmsrs;
1680 		u32 flags = ranges[i].flags;
1681 		unsigned long *bitmap = ranges[i].bitmap;
1682 
1683 		if ((index >= start) && (index < end) && (flags & type)) {
1684 			allowed = !!test_bit(index - start, bitmap);
1685 			break;
1686 		}
1687 	}
1688 
1689 out:
1690 	srcu_read_unlock(&kvm->srcu, idx);
1691 
1692 	return allowed;
1693 }
1694 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1695 
1696 /*
1697  * Write @data into the MSR specified by @index.  Select MSR specific fault
1698  * checks are bypassed if @host_initiated is %true.
1699  * Returns 0 on success, non-0 otherwise.
1700  * Assumes vcpu_load() was already called.
1701  */
1702 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1703 			 bool host_initiated)
1704 {
1705 	struct msr_data msr;
1706 
1707 	if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1708 		return KVM_MSR_RET_FILTERED;
1709 
1710 	switch (index) {
1711 	case MSR_FS_BASE:
1712 	case MSR_GS_BASE:
1713 	case MSR_KERNEL_GS_BASE:
1714 	case MSR_CSTAR:
1715 	case MSR_LSTAR:
1716 		if (is_noncanonical_address(data, vcpu))
1717 			return 1;
1718 		break;
1719 	case MSR_IA32_SYSENTER_EIP:
1720 	case MSR_IA32_SYSENTER_ESP:
1721 		/*
1722 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1723 		 * non-canonical address is written on Intel but not on
1724 		 * AMD (which ignores the top 32-bits, because it does
1725 		 * not implement 64-bit SYSENTER).
1726 		 *
1727 		 * 64-bit code should hence be able to write a non-canonical
1728 		 * value on AMD.  Making the address canonical ensures that
1729 		 * vmentry does not fail on Intel after writing a non-canonical
1730 		 * value, and that something deterministic happens if the guest
1731 		 * invokes 64-bit SYSENTER.
1732 		 */
1733 		data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1734 		break;
1735 	case MSR_TSC_AUX:
1736 		if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1737 			return 1;
1738 
1739 		if (!host_initiated &&
1740 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1741 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1742 			return 1;
1743 
1744 		/*
1745 		 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1746 		 * incomplete and conflicting architectural behavior.  Current
1747 		 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1748 		 * reserved and always read as zeros.  Enforce Intel's reserved
1749 		 * bits check if and only if the guest CPU is Intel, and clear
1750 		 * the bits in all other cases.  This ensures cross-vendor
1751 		 * migration will provide consistent behavior for the guest.
1752 		 */
1753 		if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1754 			return 1;
1755 
1756 		data = (u32)data;
1757 		break;
1758 	}
1759 
1760 	msr.data = data;
1761 	msr.index = index;
1762 	msr.host_initiated = host_initiated;
1763 
1764 	return static_call(kvm_x86_set_msr)(vcpu, &msr);
1765 }
1766 
1767 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1768 				     u32 index, u64 data, bool host_initiated)
1769 {
1770 	int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1771 
1772 	if (ret == KVM_MSR_RET_INVALID)
1773 		if (kvm_msr_ignored_check(index, data, true))
1774 			ret = 0;
1775 
1776 	return ret;
1777 }
1778 
1779 /*
1780  * Read the MSR specified by @index into @data.  Select MSR specific fault
1781  * checks are bypassed if @host_initiated is %true.
1782  * Returns 0 on success, non-0 otherwise.
1783  * Assumes vcpu_load() was already called.
1784  */
1785 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1786 		  bool host_initiated)
1787 {
1788 	struct msr_data msr;
1789 	int ret;
1790 
1791 	if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1792 		return KVM_MSR_RET_FILTERED;
1793 
1794 	switch (index) {
1795 	case MSR_TSC_AUX:
1796 		if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1797 			return 1;
1798 
1799 		if (!host_initiated &&
1800 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1801 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1802 			return 1;
1803 		break;
1804 	}
1805 
1806 	msr.index = index;
1807 	msr.host_initiated = host_initiated;
1808 
1809 	ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1810 	if (!ret)
1811 		*data = msr.data;
1812 	return ret;
1813 }
1814 
1815 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1816 				     u32 index, u64 *data, bool host_initiated)
1817 {
1818 	int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1819 
1820 	if (ret == KVM_MSR_RET_INVALID) {
1821 		/* Unconditionally clear *data for simplicity */
1822 		*data = 0;
1823 		if (kvm_msr_ignored_check(index, 0, false))
1824 			ret = 0;
1825 	}
1826 
1827 	return ret;
1828 }
1829 
1830 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1831 {
1832 	return kvm_get_msr_ignored_check(vcpu, index, data, false);
1833 }
1834 EXPORT_SYMBOL_GPL(kvm_get_msr);
1835 
1836 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1837 {
1838 	return kvm_set_msr_ignored_check(vcpu, index, data, false);
1839 }
1840 EXPORT_SYMBOL_GPL(kvm_set_msr);
1841 
1842 static void complete_userspace_rdmsr(struct kvm_vcpu *vcpu)
1843 {
1844 	if (!vcpu->run->msr.error) {
1845 		kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1846 		kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1847 	}
1848 }
1849 
1850 static int complete_emulated_msr_access(struct kvm_vcpu *vcpu)
1851 {
1852 	return complete_emulated_insn_gp(vcpu, vcpu->run->msr.error);
1853 }
1854 
1855 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1856 {
1857 	complete_userspace_rdmsr(vcpu);
1858 	return complete_emulated_msr_access(vcpu);
1859 }
1860 
1861 static int complete_fast_msr_access(struct kvm_vcpu *vcpu)
1862 {
1863 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1864 }
1865 
1866 static int complete_fast_rdmsr(struct kvm_vcpu *vcpu)
1867 {
1868 	complete_userspace_rdmsr(vcpu);
1869 	return complete_fast_msr_access(vcpu);
1870 }
1871 
1872 static u64 kvm_msr_reason(int r)
1873 {
1874 	switch (r) {
1875 	case KVM_MSR_RET_INVALID:
1876 		return KVM_MSR_EXIT_REASON_UNKNOWN;
1877 	case KVM_MSR_RET_FILTERED:
1878 		return KVM_MSR_EXIT_REASON_FILTER;
1879 	default:
1880 		return KVM_MSR_EXIT_REASON_INVAL;
1881 	}
1882 }
1883 
1884 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1885 			      u32 exit_reason, u64 data,
1886 			      int (*completion)(struct kvm_vcpu *vcpu),
1887 			      int r)
1888 {
1889 	u64 msr_reason = kvm_msr_reason(r);
1890 
1891 	/* Check if the user wanted to know about this MSR fault */
1892 	if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1893 		return 0;
1894 
1895 	vcpu->run->exit_reason = exit_reason;
1896 	vcpu->run->msr.error = 0;
1897 	memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1898 	vcpu->run->msr.reason = msr_reason;
1899 	vcpu->run->msr.index = index;
1900 	vcpu->run->msr.data = data;
1901 	vcpu->arch.complete_userspace_io = completion;
1902 
1903 	return 1;
1904 }
1905 
1906 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1907 {
1908 	u32 ecx = kvm_rcx_read(vcpu);
1909 	u64 data;
1910 	int r;
1911 
1912 	r = kvm_get_msr(vcpu, ecx, &data);
1913 
1914 	if (!r) {
1915 		trace_kvm_msr_read(ecx, data);
1916 
1917 		kvm_rax_write(vcpu, data & -1u);
1918 		kvm_rdx_write(vcpu, (data >> 32) & -1u);
1919 	} else {
1920 		/* MSR read failed? See if we should ask user space */
1921 		if (kvm_msr_user_space(vcpu, ecx, KVM_EXIT_X86_RDMSR, 0,
1922 				       complete_fast_rdmsr, r))
1923 			return 0;
1924 		trace_kvm_msr_read_ex(ecx);
1925 	}
1926 
1927 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1928 }
1929 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1930 
1931 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1932 {
1933 	u32 ecx = kvm_rcx_read(vcpu);
1934 	u64 data = kvm_read_edx_eax(vcpu);
1935 	int r;
1936 
1937 	r = kvm_set_msr(vcpu, ecx, data);
1938 
1939 	if (!r) {
1940 		trace_kvm_msr_write(ecx, data);
1941 	} else {
1942 		/* MSR write failed? See if we should ask user space */
1943 		if (kvm_msr_user_space(vcpu, ecx, KVM_EXIT_X86_WRMSR, data,
1944 				       complete_fast_msr_access, r))
1945 			return 0;
1946 		/* Signal all other negative errors to userspace */
1947 		if (r < 0)
1948 			return r;
1949 		trace_kvm_msr_write_ex(ecx, data);
1950 	}
1951 
1952 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1953 }
1954 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1955 
1956 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
1957 {
1958 	return kvm_skip_emulated_instruction(vcpu);
1959 }
1960 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
1961 
1962 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
1963 {
1964 	/* Treat an INVD instruction as a NOP and just skip it. */
1965 	return kvm_emulate_as_nop(vcpu);
1966 }
1967 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
1968 
1969 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
1970 {
1971 	pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
1972 	return kvm_emulate_as_nop(vcpu);
1973 }
1974 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
1975 
1976 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
1977 {
1978 	kvm_queue_exception(vcpu, UD_VECTOR);
1979 	return 1;
1980 }
1981 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
1982 
1983 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
1984 {
1985 	pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
1986 	return kvm_emulate_as_nop(vcpu);
1987 }
1988 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
1989 
1990 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1991 {
1992 	xfer_to_guest_mode_prepare();
1993 	return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1994 		xfer_to_guest_mode_work_pending();
1995 }
1996 
1997 /*
1998  * The fast path for frequent and performance sensitive wrmsr emulation,
1999  * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
2000  * the latency of virtual IPI by avoiding the expensive bits of transitioning
2001  * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
2002  * other cases which must be called after interrupts are enabled on the host.
2003  */
2004 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
2005 {
2006 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
2007 		return 1;
2008 
2009 	if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
2010 		((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
2011 		((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
2012 		((u32)(data >> 32) != X2APIC_BROADCAST)) {
2013 
2014 		data &= ~(1 << 12);
2015 		kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
2016 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
2017 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
2018 		trace_kvm_apic_write(APIC_ICR, (u32)data);
2019 		return 0;
2020 	}
2021 
2022 	return 1;
2023 }
2024 
2025 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
2026 {
2027 	if (!kvm_can_use_hv_timer(vcpu))
2028 		return 1;
2029 
2030 	kvm_set_lapic_tscdeadline_msr(vcpu, data);
2031 	return 0;
2032 }
2033 
2034 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
2035 {
2036 	u32 msr = kvm_rcx_read(vcpu);
2037 	u64 data;
2038 	fastpath_t ret = EXIT_FASTPATH_NONE;
2039 
2040 	switch (msr) {
2041 	case APIC_BASE_MSR + (APIC_ICR >> 4):
2042 		data = kvm_read_edx_eax(vcpu);
2043 		if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
2044 			kvm_skip_emulated_instruction(vcpu);
2045 			ret = EXIT_FASTPATH_EXIT_HANDLED;
2046 		}
2047 		break;
2048 	case MSR_IA32_TSC_DEADLINE:
2049 		data = kvm_read_edx_eax(vcpu);
2050 		if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
2051 			kvm_skip_emulated_instruction(vcpu);
2052 			ret = EXIT_FASTPATH_REENTER_GUEST;
2053 		}
2054 		break;
2055 	default:
2056 		break;
2057 	}
2058 
2059 	if (ret != EXIT_FASTPATH_NONE)
2060 		trace_kvm_msr_write(msr, data);
2061 
2062 	return ret;
2063 }
2064 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
2065 
2066 /*
2067  * Adapt set_msr() to msr_io()'s calling convention
2068  */
2069 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2070 {
2071 	return kvm_get_msr_ignored_check(vcpu, index, data, true);
2072 }
2073 
2074 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2075 {
2076 	return kvm_set_msr_ignored_check(vcpu, index, *data, true);
2077 }
2078 
2079 #ifdef CONFIG_X86_64
2080 struct pvclock_clock {
2081 	int vclock_mode;
2082 	u64 cycle_last;
2083 	u64 mask;
2084 	u32 mult;
2085 	u32 shift;
2086 	u64 base_cycles;
2087 	u64 offset;
2088 };
2089 
2090 struct pvclock_gtod_data {
2091 	seqcount_t	seq;
2092 
2093 	struct pvclock_clock clock; /* extract of a clocksource struct */
2094 	struct pvclock_clock raw_clock; /* extract of a clocksource struct */
2095 
2096 	ktime_t		offs_boot;
2097 	u64		wall_time_sec;
2098 };
2099 
2100 static struct pvclock_gtod_data pvclock_gtod_data;
2101 
2102 static void update_pvclock_gtod(struct timekeeper *tk)
2103 {
2104 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2105 
2106 	write_seqcount_begin(&vdata->seq);
2107 
2108 	/* copy pvclock gtod data */
2109 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->vdso_clock_mode;
2110 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
2111 	vdata->clock.mask		= tk->tkr_mono.mask;
2112 	vdata->clock.mult		= tk->tkr_mono.mult;
2113 	vdata->clock.shift		= tk->tkr_mono.shift;
2114 	vdata->clock.base_cycles	= tk->tkr_mono.xtime_nsec;
2115 	vdata->clock.offset		= tk->tkr_mono.base;
2116 
2117 	vdata->raw_clock.vclock_mode	= tk->tkr_raw.clock->vdso_clock_mode;
2118 	vdata->raw_clock.cycle_last	= tk->tkr_raw.cycle_last;
2119 	vdata->raw_clock.mask		= tk->tkr_raw.mask;
2120 	vdata->raw_clock.mult		= tk->tkr_raw.mult;
2121 	vdata->raw_clock.shift		= tk->tkr_raw.shift;
2122 	vdata->raw_clock.base_cycles	= tk->tkr_raw.xtime_nsec;
2123 	vdata->raw_clock.offset		= tk->tkr_raw.base;
2124 
2125 	vdata->wall_time_sec            = tk->xtime_sec;
2126 
2127 	vdata->offs_boot		= tk->offs_boot;
2128 
2129 	write_seqcount_end(&vdata->seq);
2130 }
2131 
2132 static s64 get_kvmclock_base_ns(void)
2133 {
2134 	/* Count up from boot time, but with the frequency of the raw clock.  */
2135 	return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2136 }
2137 #else
2138 static s64 get_kvmclock_base_ns(void)
2139 {
2140 	/* Master clock not used, so we can just use CLOCK_BOOTTIME.  */
2141 	return ktime_get_boottime_ns();
2142 }
2143 #endif
2144 
2145 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2146 {
2147 	int version;
2148 	int r;
2149 	struct pvclock_wall_clock wc;
2150 	u32 wc_sec_hi;
2151 	u64 wall_nsec;
2152 
2153 	if (!wall_clock)
2154 		return;
2155 
2156 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2157 	if (r)
2158 		return;
2159 
2160 	if (version & 1)
2161 		++version;  /* first time write, random junk */
2162 
2163 	++version;
2164 
2165 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2166 		return;
2167 
2168 	/*
2169 	 * The guest calculates current wall clock time by adding
2170 	 * system time (updated by kvm_guest_time_update below) to the
2171 	 * wall clock specified here.  We do the reverse here.
2172 	 */
2173 	wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2174 
2175 	wc.nsec = do_div(wall_nsec, 1000000000);
2176 	wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2177 	wc.version = version;
2178 
2179 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2180 
2181 	if (sec_hi_ofs) {
2182 		wc_sec_hi = wall_nsec >> 32;
2183 		kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2184 				&wc_sec_hi, sizeof(wc_sec_hi));
2185 	}
2186 
2187 	version++;
2188 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2189 }
2190 
2191 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2192 				  bool old_msr, bool host_initiated)
2193 {
2194 	struct kvm_arch *ka = &vcpu->kvm->arch;
2195 
2196 	if (vcpu->vcpu_id == 0 && !host_initiated) {
2197 		if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2198 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2199 
2200 		ka->boot_vcpu_runs_old_kvmclock = old_msr;
2201 	}
2202 
2203 	vcpu->arch.time = system_time;
2204 	kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2205 
2206 	/* we verify if the enable bit is set... */
2207 	vcpu->arch.pv_time_enabled = false;
2208 	if (!(system_time & 1))
2209 		return;
2210 
2211 	if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2212 				       &vcpu->arch.pv_time, system_time & ~1ULL,
2213 				       sizeof(struct pvclock_vcpu_time_info)))
2214 		vcpu->arch.pv_time_enabled = true;
2215 
2216 	return;
2217 }
2218 
2219 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2220 {
2221 	do_shl32_div32(dividend, divisor);
2222 	return dividend;
2223 }
2224 
2225 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2226 			       s8 *pshift, u32 *pmultiplier)
2227 {
2228 	uint64_t scaled64;
2229 	int32_t  shift = 0;
2230 	uint64_t tps64;
2231 	uint32_t tps32;
2232 
2233 	tps64 = base_hz;
2234 	scaled64 = scaled_hz;
2235 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2236 		tps64 >>= 1;
2237 		shift--;
2238 	}
2239 
2240 	tps32 = (uint32_t)tps64;
2241 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2242 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2243 			scaled64 >>= 1;
2244 		else
2245 			tps32 <<= 1;
2246 		shift++;
2247 	}
2248 
2249 	*pshift = shift;
2250 	*pmultiplier = div_frac(scaled64, tps32);
2251 }
2252 
2253 #ifdef CONFIG_X86_64
2254 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2255 #endif
2256 
2257 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2258 static unsigned long max_tsc_khz;
2259 
2260 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2261 {
2262 	u64 v = (u64)khz * (1000000 + ppm);
2263 	do_div(v, 1000000);
2264 	return v;
2265 }
2266 
2267 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier);
2268 
2269 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2270 {
2271 	u64 ratio;
2272 
2273 	/* Guest TSC same frequency as host TSC? */
2274 	if (!scale) {
2275 		kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2276 		return 0;
2277 	}
2278 
2279 	/* TSC scaling supported? */
2280 	if (!kvm_has_tsc_control) {
2281 		if (user_tsc_khz > tsc_khz) {
2282 			vcpu->arch.tsc_catchup = 1;
2283 			vcpu->arch.tsc_always_catchup = 1;
2284 			return 0;
2285 		} else {
2286 			pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2287 			return -1;
2288 		}
2289 	}
2290 
2291 	/* TSC scaling required  - calculate ratio */
2292 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2293 				user_tsc_khz, tsc_khz);
2294 
2295 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2296 		pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2297 			            user_tsc_khz);
2298 		return -1;
2299 	}
2300 
2301 	kvm_vcpu_write_tsc_multiplier(vcpu, ratio);
2302 	return 0;
2303 }
2304 
2305 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2306 {
2307 	u32 thresh_lo, thresh_hi;
2308 	int use_scaling = 0;
2309 
2310 	/* tsc_khz can be zero if TSC calibration fails */
2311 	if (user_tsc_khz == 0) {
2312 		/* set tsc_scaling_ratio to a safe value */
2313 		kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2314 		return -1;
2315 	}
2316 
2317 	/* Compute a scale to convert nanoseconds in TSC cycles */
2318 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2319 			   &vcpu->arch.virtual_tsc_shift,
2320 			   &vcpu->arch.virtual_tsc_mult);
2321 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2322 
2323 	/*
2324 	 * Compute the variation in TSC rate which is acceptable
2325 	 * within the range of tolerance and decide if the
2326 	 * rate being applied is within that bounds of the hardware
2327 	 * rate.  If so, no scaling or compensation need be done.
2328 	 */
2329 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2330 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2331 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2332 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2333 		use_scaling = 1;
2334 	}
2335 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2336 }
2337 
2338 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2339 {
2340 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2341 				      vcpu->arch.virtual_tsc_mult,
2342 				      vcpu->arch.virtual_tsc_shift);
2343 	tsc += vcpu->arch.this_tsc_write;
2344 	return tsc;
2345 }
2346 
2347 static inline int gtod_is_based_on_tsc(int mode)
2348 {
2349 	return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2350 }
2351 
2352 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2353 {
2354 #ifdef CONFIG_X86_64
2355 	bool vcpus_matched;
2356 	struct kvm_arch *ka = &vcpu->kvm->arch;
2357 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2358 
2359 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2360 			 atomic_read(&vcpu->kvm->online_vcpus));
2361 
2362 	/*
2363 	 * Once the masterclock is enabled, always perform request in
2364 	 * order to update it.
2365 	 *
2366 	 * In order to enable masterclock, the host clocksource must be TSC
2367 	 * and the vcpus need to have matched TSCs.  When that happens,
2368 	 * perform request to enable masterclock.
2369 	 */
2370 	if (ka->use_master_clock ||
2371 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2372 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2373 
2374 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2375 			    atomic_read(&vcpu->kvm->online_vcpus),
2376 		            ka->use_master_clock, gtod->clock.vclock_mode);
2377 #endif
2378 }
2379 
2380 /*
2381  * Multiply tsc by a fixed point number represented by ratio.
2382  *
2383  * The most significant 64-N bits (mult) of ratio represent the
2384  * integral part of the fixed point number; the remaining N bits
2385  * (frac) represent the fractional part, ie. ratio represents a fixed
2386  * point number (mult + frac * 2^(-N)).
2387  *
2388  * N equals to kvm_tsc_scaling_ratio_frac_bits.
2389  */
2390 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2391 {
2392 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2393 }
2394 
2395 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc, u64 ratio)
2396 {
2397 	u64 _tsc = tsc;
2398 
2399 	if (ratio != kvm_default_tsc_scaling_ratio)
2400 		_tsc = __scale_tsc(ratio, tsc);
2401 
2402 	return _tsc;
2403 }
2404 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2405 
2406 static u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2407 {
2408 	u64 tsc;
2409 
2410 	tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio);
2411 
2412 	return target_tsc - tsc;
2413 }
2414 
2415 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2416 {
2417 	return vcpu->arch.l1_tsc_offset +
2418 		kvm_scale_tsc(vcpu, host_tsc, vcpu->arch.l1_tsc_scaling_ratio);
2419 }
2420 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2421 
2422 u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier)
2423 {
2424 	u64 nested_offset;
2425 
2426 	if (l2_multiplier == kvm_default_tsc_scaling_ratio)
2427 		nested_offset = l1_offset;
2428 	else
2429 		nested_offset = mul_s64_u64_shr((s64) l1_offset, l2_multiplier,
2430 						kvm_tsc_scaling_ratio_frac_bits);
2431 
2432 	nested_offset += l2_offset;
2433 	return nested_offset;
2434 }
2435 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_offset);
2436 
2437 u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier)
2438 {
2439 	if (l2_multiplier != kvm_default_tsc_scaling_ratio)
2440 		return mul_u64_u64_shr(l1_multiplier, l2_multiplier,
2441 				       kvm_tsc_scaling_ratio_frac_bits);
2442 
2443 	return l1_multiplier;
2444 }
2445 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_multiplier);
2446 
2447 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset)
2448 {
2449 	trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2450 				   vcpu->arch.l1_tsc_offset,
2451 				   l1_offset);
2452 
2453 	vcpu->arch.l1_tsc_offset = l1_offset;
2454 
2455 	/*
2456 	 * If we are here because L1 chose not to trap WRMSR to TSC then
2457 	 * according to the spec this should set L1's TSC (as opposed to
2458 	 * setting L1's offset for L2).
2459 	 */
2460 	if (is_guest_mode(vcpu))
2461 		vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2462 			l1_offset,
2463 			static_call(kvm_x86_get_l2_tsc_offset)(vcpu),
2464 			static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2465 	else
2466 		vcpu->arch.tsc_offset = l1_offset;
2467 
2468 	static_call(kvm_x86_write_tsc_offset)(vcpu, vcpu->arch.tsc_offset);
2469 }
2470 
2471 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier)
2472 {
2473 	vcpu->arch.l1_tsc_scaling_ratio = l1_multiplier;
2474 
2475 	/* Userspace is changing the multiplier while L2 is active */
2476 	if (is_guest_mode(vcpu))
2477 		vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2478 			l1_multiplier,
2479 			static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2480 	else
2481 		vcpu->arch.tsc_scaling_ratio = l1_multiplier;
2482 
2483 	if (kvm_has_tsc_control)
2484 		static_call(kvm_x86_write_tsc_multiplier)(
2485 			vcpu, vcpu->arch.tsc_scaling_ratio);
2486 }
2487 
2488 static inline bool kvm_check_tsc_unstable(void)
2489 {
2490 #ifdef CONFIG_X86_64
2491 	/*
2492 	 * TSC is marked unstable when we're running on Hyper-V,
2493 	 * 'TSC page' clocksource is good.
2494 	 */
2495 	if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2496 		return false;
2497 #endif
2498 	return check_tsc_unstable();
2499 }
2500 
2501 /*
2502  * Infers attempts to synchronize the guest's tsc from host writes. Sets the
2503  * offset for the vcpu and tracks the TSC matching generation that the vcpu
2504  * participates in.
2505  */
2506 static void __kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 offset, u64 tsc,
2507 				  u64 ns, bool matched)
2508 {
2509 	struct kvm *kvm = vcpu->kvm;
2510 
2511 	lockdep_assert_held(&kvm->arch.tsc_write_lock);
2512 
2513 	/*
2514 	 * We also track th most recent recorded KHZ, write and time to
2515 	 * allow the matching interval to be extended at each write.
2516 	 */
2517 	kvm->arch.last_tsc_nsec = ns;
2518 	kvm->arch.last_tsc_write = tsc;
2519 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2520 	kvm->arch.last_tsc_offset = offset;
2521 
2522 	vcpu->arch.last_guest_tsc = tsc;
2523 
2524 	kvm_vcpu_write_tsc_offset(vcpu, offset);
2525 
2526 	if (!matched) {
2527 		/*
2528 		 * We split periods of matched TSC writes into generations.
2529 		 * For each generation, we track the original measured
2530 		 * nanosecond time, offset, and write, so if TSCs are in
2531 		 * sync, we can match exact offset, and if not, we can match
2532 		 * exact software computation in compute_guest_tsc()
2533 		 *
2534 		 * These values are tracked in kvm->arch.cur_xxx variables.
2535 		 */
2536 		kvm->arch.cur_tsc_generation++;
2537 		kvm->arch.cur_tsc_nsec = ns;
2538 		kvm->arch.cur_tsc_write = tsc;
2539 		kvm->arch.cur_tsc_offset = offset;
2540 		kvm->arch.nr_vcpus_matched_tsc = 0;
2541 	} else if (vcpu->arch.this_tsc_generation != kvm->arch.cur_tsc_generation) {
2542 		kvm->arch.nr_vcpus_matched_tsc++;
2543 	}
2544 
2545 	/* Keep track of which generation this VCPU has synchronized to */
2546 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2547 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2548 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2549 
2550 	kvm_track_tsc_matching(vcpu);
2551 }
2552 
2553 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2554 {
2555 	struct kvm *kvm = vcpu->kvm;
2556 	u64 offset, ns, elapsed;
2557 	unsigned long flags;
2558 	bool matched = false;
2559 	bool synchronizing = false;
2560 
2561 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2562 	offset = kvm_compute_l1_tsc_offset(vcpu, data);
2563 	ns = get_kvmclock_base_ns();
2564 	elapsed = ns - kvm->arch.last_tsc_nsec;
2565 
2566 	if (vcpu->arch.virtual_tsc_khz) {
2567 		if (data == 0) {
2568 			/*
2569 			 * detection of vcpu initialization -- need to sync
2570 			 * with other vCPUs. This particularly helps to keep
2571 			 * kvm_clock stable after CPU hotplug
2572 			 */
2573 			synchronizing = true;
2574 		} else {
2575 			u64 tsc_exp = kvm->arch.last_tsc_write +
2576 						nsec_to_cycles(vcpu, elapsed);
2577 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2578 			/*
2579 			 * Special case: TSC write with a small delta (1 second)
2580 			 * of virtual cycle time against real time is
2581 			 * interpreted as an attempt to synchronize the CPU.
2582 			 */
2583 			synchronizing = data < tsc_exp + tsc_hz &&
2584 					data + tsc_hz > tsc_exp;
2585 		}
2586 	}
2587 
2588 	/*
2589 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
2590 	 * TSC, we add elapsed time in this computation.  We could let the
2591 	 * compensation code attempt to catch up if we fall behind, but
2592 	 * it's better to try to match offsets from the beginning.
2593          */
2594 	if (synchronizing &&
2595 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2596 		if (!kvm_check_tsc_unstable()) {
2597 			offset = kvm->arch.cur_tsc_offset;
2598 		} else {
2599 			u64 delta = nsec_to_cycles(vcpu, elapsed);
2600 			data += delta;
2601 			offset = kvm_compute_l1_tsc_offset(vcpu, data);
2602 		}
2603 		matched = true;
2604 	}
2605 
2606 	__kvm_synchronize_tsc(vcpu, offset, data, ns, matched);
2607 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2608 }
2609 
2610 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2611 					   s64 adjustment)
2612 {
2613 	u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2614 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2615 }
2616 
2617 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2618 {
2619 	if (vcpu->arch.l1_tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2620 		WARN_ON(adjustment < 0);
2621 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment,
2622 				   vcpu->arch.l1_tsc_scaling_ratio);
2623 	adjust_tsc_offset_guest(vcpu, adjustment);
2624 }
2625 
2626 #ifdef CONFIG_X86_64
2627 
2628 static u64 read_tsc(void)
2629 {
2630 	u64 ret = (u64)rdtsc_ordered();
2631 	u64 last = pvclock_gtod_data.clock.cycle_last;
2632 
2633 	if (likely(ret >= last))
2634 		return ret;
2635 
2636 	/*
2637 	 * GCC likes to generate cmov here, but this branch is extremely
2638 	 * predictable (it's just a function of time and the likely is
2639 	 * very likely) and there's a data dependence, so force GCC
2640 	 * to generate a branch instead.  I don't barrier() because
2641 	 * we don't actually need a barrier, and if this function
2642 	 * ever gets inlined it will generate worse code.
2643 	 */
2644 	asm volatile ("");
2645 	return last;
2646 }
2647 
2648 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2649 			  int *mode)
2650 {
2651 	long v;
2652 	u64 tsc_pg_val;
2653 
2654 	switch (clock->vclock_mode) {
2655 	case VDSO_CLOCKMODE_HVCLOCK:
2656 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2657 						  tsc_timestamp);
2658 		if (tsc_pg_val != U64_MAX) {
2659 			/* TSC page valid */
2660 			*mode = VDSO_CLOCKMODE_HVCLOCK;
2661 			v = (tsc_pg_val - clock->cycle_last) &
2662 				clock->mask;
2663 		} else {
2664 			/* TSC page invalid */
2665 			*mode = VDSO_CLOCKMODE_NONE;
2666 		}
2667 		break;
2668 	case VDSO_CLOCKMODE_TSC:
2669 		*mode = VDSO_CLOCKMODE_TSC;
2670 		*tsc_timestamp = read_tsc();
2671 		v = (*tsc_timestamp - clock->cycle_last) &
2672 			clock->mask;
2673 		break;
2674 	default:
2675 		*mode = VDSO_CLOCKMODE_NONE;
2676 	}
2677 
2678 	if (*mode == VDSO_CLOCKMODE_NONE)
2679 		*tsc_timestamp = v = 0;
2680 
2681 	return v * clock->mult;
2682 }
2683 
2684 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2685 {
2686 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2687 	unsigned long seq;
2688 	int mode;
2689 	u64 ns;
2690 
2691 	do {
2692 		seq = read_seqcount_begin(&gtod->seq);
2693 		ns = gtod->raw_clock.base_cycles;
2694 		ns += vgettsc(&gtod->raw_clock, tsc_timestamp, &mode);
2695 		ns >>= gtod->raw_clock.shift;
2696 		ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2697 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2698 	*t = ns;
2699 
2700 	return mode;
2701 }
2702 
2703 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2704 {
2705 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2706 	unsigned long seq;
2707 	int mode;
2708 	u64 ns;
2709 
2710 	do {
2711 		seq = read_seqcount_begin(&gtod->seq);
2712 		ts->tv_sec = gtod->wall_time_sec;
2713 		ns = gtod->clock.base_cycles;
2714 		ns += vgettsc(&gtod->clock, tsc_timestamp, &mode);
2715 		ns >>= gtod->clock.shift;
2716 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2717 
2718 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2719 	ts->tv_nsec = ns;
2720 
2721 	return mode;
2722 }
2723 
2724 /* returns true if host is using TSC based clocksource */
2725 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2726 {
2727 	/* checked again under seqlock below */
2728 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2729 		return false;
2730 
2731 	return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2732 						      tsc_timestamp));
2733 }
2734 
2735 /* returns true if host is using TSC based clocksource */
2736 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2737 					   u64 *tsc_timestamp)
2738 {
2739 	/* checked again under seqlock below */
2740 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2741 		return false;
2742 
2743 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2744 }
2745 #endif
2746 
2747 /*
2748  *
2749  * Assuming a stable TSC across physical CPUS, and a stable TSC
2750  * across virtual CPUs, the following condition is possible.
2751  * Each numbered line represents an event visible to both
2752  * CPUs at the next numbered event.
2753  *
2754  * "timespecX" represents host monotonic time. "tscX" represents
2755  * RDTSC value.
2756  *
2757  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
2758  *
2759  * 1.  read timespec0,tsc0
2760  * 2.					| timespec1 = timespec0 + N
2761  * 					| tsc1 = tsc0 + M
2762  * 3. transition to guest		| transition to guest
2763  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2764  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
2765  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2766  *
2767  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2768  *
2769  * 	- ret0 < ret1
2770  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2771  *		...
2772  *	- 0 < N - M => M < N
2773  *
2774  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2775  * always the case (the difference between two distinct xtime instances
2776  * might be smaller then the difference between corresponding TSC reads,
2777  * when updating guest vcpus pvclock areas).
2778  *
2779  * To avoid that problem, do not allow visibility of distinct
2780  * system_timestamp/tsc_timestamp values simultaneously: use a master
2781  * copy of host monotonic time values. Update that master copy
2782  * in lockstep.
2783  *
2784  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2785  *
2786  */
2787 
2788 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2789 {
2790 #ifdef CONFIG_X86_64
2791 	struct kvm_arch *ka = &kvm->arch;
2792 	int vclock_mode;
2793 	bool host_tsc_clocksource, vcpus_matched;
2794 
2795 	lockdep_assert_held(&kvm->arch.tsc_write_lock);
2796 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2797 			atomic_read(&kvm->online_vcpus));
2798 
2799 	/*
2800 	 * If the host uses TSC clock, then passthrough TSC as stable
2801 	 * to the guest.
2802 	 */
2803 	host_tsc_clocksource = kvm_get_time_and_clockread(
2804 					&ka->master_kernel_ns,
2805 					&ka->master_cycle_now);
2806 
2807 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2808 				&& !ka->backwards_tsc_observed
2809 				&& !ka->boot_vcpu_runs_old_kvmclock;
2810 
2811 	if (ka->use_master_clock)
2812 		atomic_set(&kvm_guest_has_master_clock, 1);
2813 
2814 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2815 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2816 					vcpus_matched);
2817 #endif
2818 }
2819 
2820 static void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2821 {
2822 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2823 }
2824 
2825 static void __kvm_start_pvclock_update(struct kvm *kvm)
2826 {
2827 	raw_spin_lock_irq(&kvm->arch.tsc_write_lock);
2828 	write_seqcount_begin(&kvm->arch.pvclock_sc);
2829 }
2830 
2831 static void kvm_start_pvclock_update(struct kvm *kvm)
2832 {
2833 	kvm_make_mclock_inprogress_request(kvm);
2834 
2835 	/* no guest entries from this point */
2836 	__kvm_start_pvclock_update(kvm);
2837 }
2838 
2839 static void kvm_end_pvclock_update(struct kvm *kvm)
2840 {
2841 	struct kvm_arch *ka = &kvm->arch;
2842 	struct kvm_vcpu *vcpu;
2843 	unsigned long i;
2844 
2845 	write_seqcount_end(&ka->pvclock_sc);
2846 	raw_spin_unlock_irq(&ka->tsc_write_lock);
2847 	kvm_for_each_vcpu(i, vcpu, kvm)
2848 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2849 
2850 	/* guest entries allowed */
2851 	kvm_for_each_vcpu(i, vcpu, kvm)
2852 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2853 }
2854 
2855 static void kvm_update_masterclock(struct kvm *kvm)
2856 {
2857 	kvm_hv_invalidate_tsc_page(kvm);
2858 	kvm_start_pvclock_update(kvm);
2859 	pvclock_update_vm_gtod_copy(kvm);
2860 	kvm_end_pvclock_update(kvm);
2861 }
2862 
2863 /* Called within read_seqcount_begin/retry for kvm->pvclock_sc.  */
2864 static void __get_kvmclock(struct kvm *kvm, struct kvm_clock_data *data)
2865 {
2866 	struct kvm_arch *ka = &kvm->arch;
2867 	struct pvclock_vcpu_time_info hv_clock;
2868 
2869 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2870 	get_cpu();
2871 
2872 	data->flags = 0;
2873 	if (ka->use_master_clock && __this_cpu_read(cpu_tsc_khz)) {
2874 #ifdef CONFIG_X86_64
2875 		struct timespec64 ts;
2876 
2877 		if (kvm_get_walltime_and_clockread(&ts, &data->host_tsc)) {
2878 			data->realtime = ts.tv_nsec + NSEC_PER_SEC * ts.tv_sec;
2879 			data->flags |= KVM_CLOCK_REALTIME | KVM_CLOCK_HOST_TSC;
2880 		} else
2881 #endif
2882 		data->host_tsc = rdtsc();
2883 
2884 		data->flags |= KVM_CLOCK_TSC_STABLE;
2885 		hv_clock.tsc_timestamp = ka->master_cycle_now;
2886 		hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2887 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2888 				   &hv_clock.tsc_shift,
2889 				   &hv_clock.tsc_to_system_mul);
2890 		data->clock = __pvclock_read_cycles(&hv_clock, data->host_tsc);
2891 	} else {
2892 		data->clock = get_kvmclock_base_ns() + ka->kvmclock_offset;
2893 	}
2894 
2895 	put_cpu();
2896 }
2897 
2898 static void get_kvmclock(struct kvm *kvm, struct kvm_clock_data *data)
2899 {
2900 	struct kvm_arch *ka = &kvm->arch;
2901 	unsigned seq;
2902 
2903 	do {
2904 		seq = read_seqcount_begin(&ka->pvclock_sc);
2905 		__get_kvmclock(kvm, data);
2906 	} while (read_seqcount_retry(&ka->pvclock_sc, seq));
2907 }
2908 
2909 u64 get_kvmclock_ns(struct kvm *kvm)
2910 {
2911 	struct kvm_clock_data data;
2912 
2913 	get_kvmclock(kvm, &data);
2914 	return data.clock;
2915 }
2916 
2917 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2918 				   struct gfn_to_hva_cache *cache,
2919 				   unsigned int offset)
2920 {
2921 	struct kvm_vcpu_arch *vcpu = &v->arch;
2922 	struct pvclock_vcpu_time_info guest_hv_clock;
2923 
2924 	if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2925 		&guest_hv_clock, offset, sizeof(guest_hv_clock))))
2926 		return;
2927 
2928 	/* This VCPU is paused, but it's legal for a guest to read another
2929 	 * VCPU's kvmclock, so we really have to follow the specification where
2930 	 * it says that version is odd if data is being modified, and even after
2931 	 * it is consistent.
2932 	 *
2933 	 * Version field updates must be kept separate.  This is because
2934 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2935 	 * writes within a string instruction are weakly ordered.  So there
2936 	 * are three writes overall.
2937 	 *
2938 	 * As a small optimization, only write the version field in the first
2939 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2940 	 * version field is the first in the struct.
2941 	 */
2942 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2943 
2944 	if (guest_hv_clock.version & 1)
2945 		++guest_hv_clock.version;  /* first time write, random junk */
2946 
2947 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2948 	kvm_write_guest_offset_cached(v->kvm, cache,
2949 				      &vcpu->hv_clock, offset,
2950 				      sizeof(vcpu->hv_clock.version));
2951 
2952 	smp_wmb();
2953 
2954 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2955 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2956 
2957 	if (vcpu->pvclock_set_guest_stopped_request) {
2958 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2959 		vcpu->pvclock_set_guest_stopped_request = false;
2960 	}
2961 
2962 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2963 
2964 	kvm_write_guest_offset_cached(v->kvm, cache,
2965 				      &vcpu->hv_clock, offset,
2966 				      sizeof(vcpu->hv_clock));
2967 
2968 	smp_wmb();
2969 
2970 	vcpu->hv_clock.version++;
2971 	kvm_write_guest_offset_cached(v->kvm, cache,
2972 				     &vcpu->hv_clock, offset,
2973 				     sizeof(vcpu->hv_clock.version));
2974 }
2975 
2976 static int kvm_guest_time_update(struct kvm_vcpu *v)
2977 {
2978 	unsigned long flags, tgt_tsc_khz;
2979 	unsigned seq;
2980 	struct kvm_vcpu_arch *vcpu = &v->arch;
2981 	struct kvm_arch *ka = &v->kvm->arch;
2982 	s64 kernel_ns;
2983 	u64 tsc_timestamp, host_tsc;
2984 	u8 pvclock_flags;
2985 	bool use_master_clock;
2986 
2987 	kernel_ns = 0;
2988 	host_tsc = 0;
2989 
2990 	/*
2991 	 * If the host uses TSC clock, then passthrough TSC as stable
2992 	 * to the guest.
2993 	 */
2994 	do {
2995 		seq = read_seqcount_begin(&ka->pvclock_sc);
2996 		use_master_clock = ka->use_master_clock;
2997 		if (use_master_clock) {
2998 			host_tsc = ka->master_cycle_now;
2999 			kernel_ns = ka->master_kernel_ns;
3000 		}
3001 	} while (read_seqcount_retry(&ka->pvclock_sc, seq));
3002 
3003 	/* Keep irq disabled to prevent changes to the clock */
3004 	local_irq_save(flags);
3005 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
3006 	if (unlikely(tgt_tsc_khz == 0)) {
3007 		local_irq_restore(flags);
3008 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3009 		return 1;
3010 	}
3011 	if (!use_master_clock) {
3012 		host_tsc = rdtsc();
3013 		kernel_ns = get_kvmclock_base_ns();
3014 	}
3015 
3016 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
3017 
3018 	/*
3019 	 * We may have to catch up the TSC to match elapsed wall clock
3020 	 * time for two reasons, even if kvmclock is used.
3021 	 *   1) CPU could have been running below the maximum TSC rate
3022 	 *   2) Broken TSC compensation resets the base at each VCPU
3023 	 *      entry to avoid unknown leaps of TSC even when running
3024 	 *      again on the same CPU.  This may cause apparent elapsed
3025 	 *      time to disappear, and the guest to stand still or run
3026 	 *	very slowly.
3027 	 */
3028 	if (vcpu->tsc_catchup) {
3029 		u64 tsc = compute_guest_tsc(v, kernel_ns);
3030 		if (tsc > tsc_timestamp) {
3031 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
3032 			tsc_timestamp = tsc;
3033 		}
3034 	}
3035 
3036 	local_irq_restore(flags);
3037 
3038 	/* With all the info we got, fill in the values */
3039 
3040 	if (kvm_has_tsc_control)
3041 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz,
3042 					    v->arch.l1_tsc_scaling_ratio);
3043 
3044 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
3045 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
3046 				   &vcpu->hv_clock.tsc_shift,
3047 				   &vcpu->hv_clock.tsc_to_system_mul);
3048 		vcpu->hw_tsc_khz = tgt_tsc_khz;
3049 	}
3050 
3051 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
3052 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
3053 	vcpu->last_guest_tsc = tsc_timestamp;
3054 
3055 	/* If the host uses TSC clocksource, then it is stable */
3056 	pvclock_flags = 0;
3057 	if (use_master_clock)
3058 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
3059 
3060 	vcpu->hv_clock.flags = pvclock_flags;
3061 
3062 	if (vcpu->pv_time_enabled)
3063 		kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
3064 	if (vcpu->xen.vcpu_info_set)
3065 		kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
3066 				       offsetof(struct compat_vcpu_info, time));
3067 	if (vcpu->xen.vcpu_time_info_set)
3068 		kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
3069 	if (!v->vcpu_idx)
3070 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
3071 	return 0;
3072 }
3073 
3074 /*
3075  * kvmclock updates which are isolated to a given vcpu, such as
3076  * vcpu->cpu migration, should not allow system_timestamp from
3077  * the rest of the vcpus to remain static. Otherwise ntp frequency
3078  * correction applies to one vcpu's system_timestamp but not
3079  * the others.
3080  *
3081  * So in those cases, request a kvmclock update for all vcpus.
3082  * We need to rate-limit these requests though, as they can
3083  * considerably slow guests that have a large number of vcpus.
3084  * The time for a remote vcpu to update its kvmclock is bound
3085  * by the delay we use to rate-limit the updates.
3086  */
3087 
3088 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
3089 
3090 static void kvmclock_update_fn(struct work_struct *work)
3091 {
3092 	unsigned long i;
3093 	struct delayed_work *dwork = to_delayed_work(work);
3094 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3095 					   kvmclock_update_work);
3096 	struct kvm *kvm = container_of(ka, struct kvm, arch);
3097 	struct kvm_vcpu *vcpu;
3098 
3099 	kvm_for_each_vcpu(i, vcpu, kvm) {
3100 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3101 		kvm_vcpu_kick(vcpu);
3102 	}
3103 }
3104 
3105 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
3106 {
3107 	struct kvm *kvm = v->kvm;
3108 
3109 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3110 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
3111 					KVMCLOCK_UPDATE_DELAY);
3112 }
3113 
3114 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
3115 
3116 static void kvmclock_sync_fn(struct work_struct *work)
3117 {
3118 	struct delayed_work *dwork = to_delayed_work(work);
3119 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3120 					   kvmclock_sync_work);
3121 	struct kvm *kvm = container_of(ka, struct kvm, arch);
3122 
3123 	if (!kvmclock_periodic_sync)
3124 		return;
3125 
3126 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
3127 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
3128 					KVMCLOCK_SYNC_PERIOD);
3129 }
3130 
3131 /*
3132  * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
3133  */
3134 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
3135 {
3136 	/* McStatusWrEn enabled? */
3137 	if (guest_cpuid_is_amd_or_hygon(vcpu))
3138 		return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
3139 
3140 	return false;
3141 }
3142 
3143 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3144 {
3145 	u64 mcg_cap = vcpu->arch.mcg_cap;
3146 	unsigned bank_num = mcg_cap & 0xff;
3147 	u32 msr = msr_info->index;
3148 	u64 data = msr_info->data;
3149 
3150 	switch (msr) {
3151 	case MSR_IA32_MCG_STATUS:
3152 		vcpu->arch.mcg_status = data;
3153 		break;
3154 	case MSR_IA32_MCG_CTL:
3155 		if (!(mcg_cap & MCG_CTL_P) &&
3156 		    (data || !msr_info->host_initiated))
3157 			return 1;
3158 		if (data != 0 && data != ~(u64)0)
3159 			return 1;
3160 		vcpu->arch.mcg_ctl = data;
3161 		break;
3162 	default:
3163 		if (msr >= MSR_IA32_MC0_CTL &&
3164 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
3165 			u32 offset = array_index_nospec(
3166 				msr - MSR_IA32_MC0_CTL,
3167 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3168 
3169 			/* only 0 or all 1s can be written to IA32_MCi_CTL
3170 			 * some Linux kernels though clear bit 10 in bank 4 to
3171 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
3172 			 * this to avoid an uncatched #GP in the guest
3173 			 */
3174 			if ((offset & 0x3) == 0 &&
3175 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
3176 				return -1;
3177 
3178 			/* MCi_STATUS */
3179 			if (!msr_info->host_initiated &&
3180 			    (offset & 0x3) == 1 && data != 0) {
3181 				if (!can_set_mci_status(vcpu))
3182 					return -1;
3183 			}
3184 
3185 			vcpu->arch.mce_banks[offset] = data;
3186 			break;
3187 		}
3188 		return 1;
3189 	}
3190 	return 0;
3191 }
3192 
3193 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
3194 {
3195 	u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3196 
3197 	return (vcpu->arch.apf.msr_en_val & mask) == mask;
3198 }
3199 
3200 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3201 {
3202 	gpa_t gpa = data & ~0x3f;
3203 
3204 	/* Bits 4:5 are reserved, Should be zero */
3205 	if (data & 0x30)
3206 		return 1;
3207 
3208 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3209 	    (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3210 		return 1;
3211 
3212 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3213 	    (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3214 		return 1;
3215 
3216 	if (!lapic_in_kernel(vcpu))
3217 		return data ? 1 : 0;
3218 
3219 	vcpu->arch.apf.msr_en_val = data;
3220 
3221 	if (!kvm_pv_async_pf_enabled(vcpu)) {
3222 		kvm_clear_async_pf_completion_queue(vcpu);
3223 		kvm_async_pf_hash_reset(vcpu);
3224 		return 0;
3225 	}
3226 
3227 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3228 					sizeof(u64)))
3229 		return 1;
3230 
3231 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3232 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3233 
3234 	kvm_async_pf_wakeup_all(vcpu);
3235 
3236 	return 0;
3237 }
3238 
3239 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3240 {
3241 	/* Bits 8-63 are reserved */
3242 	if (data >> 8)
3243 		return 1;
3244 
3245 	if (!lapic_in_kernel(vcpu))
3246 		return 1;
3247 
3248 	vcpu->arch.apf.msr_int_val = data;
3249 
3250 	vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3251 
3252 	return 0;
3253 }
3254 
3255 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3256 {
3257 	vcpu->arch.pv_time_enabled = false;
3258 	vcpu->arch.time = 0;
3259 }
3260 
3261 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3262 {
3263 	++vcpu->stat.tlb_flush;
3264 	static_call(kvm_x86_tlb_flush_all)(vcpu);
3265 }
3266 
3267 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3268 {
3269 	++vcpu->stat.tlb_flush;
3270 
3271 	if (!tdp_enabled) {
3272 		/*
3273 		 * A TLB flush on behalf of the guest is equivalent to
3274 		 * INVPCID(all), toggling CR4.PGE, etc., which requires
3275 		 * a forced sync of the shadow page tables.  Ensure all the
3276 		 * roots are synced and the guest TLB in hardware is clean.
3277 		 */
3278 		kvm_mmu_sync_roots(vcpu);
3279 		kvm_mmu_sync_prev_roots(vcpu);
3280 	}
3281 
3282 	static_call(kvm_x86_tlb_flush_guest)(vcpu);
3283 }
3284 
3285 
3286 static inline void kvm_vcpu_flush_tlb_current(struct kvm_vcpu *vcpu)
3287 {
3288 	++vcpu->stat.tlb_flush;
3289 	static_call(kvm_x86_tlb_flush_current)(vcpu);
3290 }
3291 
3292 /*
3293  * Service "local" TLB flush requests, which are specific to the current MMU
3294  * context.  In addition to the generic event handling in vcpu_enter_guest(),
3295  * TLB flushes that are targeted at an MMU context also need to be serviced
3296  * prior before nested VM-Enter/VM-Exit.
3297  */
3298 void kvm_service_local_tlb_flush_requests(struct kvm_vcpu *vcpu)
3299 {
3300 	if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
3301 		kvm_vcpu_flush_tlb_current(vcpu);
3302 
3303 	if (kvm_check_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu))
3304 		kvm_vcpu_flush_tlb_guest(vcpu);
3305 }
3306 EXPORT_SYMBOL_GPL(kvm_service_local_tlb_flush_requests);
3307 
3308 static void record_steal_time(struct kvm_vcpu *vcpu)
3309 {
3310 	struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
3311 	struct kvm_steal_time __user *st;
3312 	struct kvm_memslots *slots;
3313 	u64 steal;
3314 	u32 version;
3315 
3316 	if (kvm_xen_msr_enabled(vcpu->kvm)) {
3317 		kvm_xen_runstate_set_running(vcpu);
3318 		return;
3319 	}
3320 
3321 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3322 		return;
3323 
3324 	if (WARN_ON_ONCE(current->mm != vcpu->kvm->mm))
3325 		return;
3326 
3327 	slots = kvm_memslots(vcpu->kvm);
3328 
3329 	if (unlikely(slots->generation != ghc->generation ||
3330 		     kvm_is_error_hva(ghc->hva) || !ghc->memslot)) {
3331 		gfn_t gfn = vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS;
3332 
3333 		/* We rely on the fact that it fits in a single page. */
3334 		BUILD_BUG_ON((sizeof(*st) - 1) & KVM_STEAL_VALID_BITS);
3335 
3336 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, gfn, sizeof(*st)) ||
3337 		    kvm_is_error_hva(ghc->hva) || !ghc->memslot)
3338 			return;
3339 	}
3340 
3341 	st = (struct kvm_steal_time __user *)ghc->hva;
3342 	/*
3343 	 * Doing a TLB flush here, on the guest's behalf, can avoid
3344 	 * expensive IPIs.
3345 	 */
3346 	if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3347 		u8 st_preempted = 0;
3348 		int err = -EFAULT;
3349 
3350 		if (!user_access_begin(st, sizeof(*st)))
3351 			return;
3352 
3353 		asm volatile("1: xchgb %0, %2\n"
3354 			     "xor %1, %1\n"
3355 			     "2:\n"
3356 			     _ASM_EXTABLE_UA(1b, 2b)
3357 			     : "+q" (st_preempted),
3358 			       "+&r" (err),
3359 			       "+m" (st->preempted));
3360 		if (err)
3361 			goto out;
3362 
3363 		user_access_end();
3364 
3365 		vcpu->arch.st.preempted = 0;
3366 
3367 		trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3368 				       st_preempted & KVM_VCPU_FLUSH_TLB);
3369 		if (st_preempted & KVM_VCPU_FLUSH_TLB)
3370 			kvm_vcpu_flush_tlb_guest(vcpu);
3371 
3372 		if (!user_access_begin(st, sizeof(*st)))
3373 			goto dirty;
3374 	} else {
3375 		if (!user_access_begin(st, sizeof(*st)))
3376 			return;
3377 
3378 		unsafe_put_user(0, &st->preempted, out);
3379 		vcpu->arch.st.preempted = 0;
3380 	}
3381 
3382 	unsafe_get_user(version, &st->version, out);
3383 	if (version & 1)
3384 		version += 1;  /* first time write, random junk */
3385 
3386 	version += 1;
3387 	unsafe_put_user(version, &st->version, out);
3388 
3389 	smp_wmb();
3390 
3391 	unsafe_get_user(steal, &st->steal, out);
3392 	steal += current->sched_info.run_delay -
3393 		vcpu->arch.st.last_steal;
3394 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
3395 	unsafe_put_user(steal, &st->steal, out);
3396 
3397 	version += 1;
3398 	unsafe_put_user(version, &st->version, out);
3399 
3400  out:
3401 	user_access_end();
3402  dirty:
3403 	mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
3404 }
3405 
3406 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3407 {
3408 	bool pr = false;
3409 	u32 msr = msr_info->index;
3410 	u64 data = msr_info->data;
3411 
3412 	if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3413 		return kvm_xen_write_hypercall_page(vcpu, data);
3414 
3415 	switch (msr) {
3416 	case MSR_AMD64_NB_CFG:
3417 	case MSR_IA32_UCODE_WRITE:
3418 	case MSR_VM_HSAVE_PA:
3419 	case MSR_AMD64_PATCH_LOADER:
3420 	case MSR_AMD64_BU_CFG2:
3421 	case MSR_AMD64_DC_CFG:
3422 	case MSR_F15H_EX_CFG:
3423 		break;
3424 
3425 	case MSR_IA32_UCODE_REV:
3426 		if (msr_info->host_initiated)
3427 			vcpu->arch.microcode_version = data;
3428 		break;
3429 	case MSR_IA32_ARCH_CAPABILITIES:
3430 		if (!msr_info->host_initiated)
3431 			return 1;
3432 		vcpu->arch.arch_capabilities = data;
3433 		break;
3434 	case MSR_IA32_PERF_CAPABILITIES: {
3435 		struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3436 
3437 		if (!msr_info->host_initiated)
3438 			return 1;
3439 		if (kvm_get_msr_feature(&msr_ent))
3440 			return 1;
3441 		if (data & ~msr_ent.data)
3442 			return 1;
3443 
3444 		vcpu->arch.perf_capabilities = data;
3445 
3446 		return 0;
3447 		}
3448 	case MSR_EFER:
3449 		return set_efer(vcpu, msr_info);
3450 	case MSR_K7_HWCR:
3451 		data &= ~(u64)0x40;	/* ignore flush filter disable */
3452 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
3453 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
3454 
3455 		/* Handle McStatusWrEn */
3456 		if (data == BIT_ULL(18)) {
3457 			vcpu->arch.msr_hwcr = data;
3458 		} else if (data != 0) {
3459 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3460 				    data);
3461 			return 1;
3462 		}
3463 		break;
3464 	case MSR_FAM10H_MMIO_CONF_BASE:
3465 		if (data != 0) {
3466 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3467 				    "0x%llx\n", data);
3468 			return 1;
3469 		}
3470 		break;
3471 	case 0x200 ... 0x2ff:
3472 		return kvm_mtrr_set_msr(vcpu, msr, data);
3473 	case MSR_IA32_APICBASE:
3474 		return kvm_set_apic_base(vcpu, msr_info);
3475 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3476 		return kvm_x2apic_msr_write(vcpu, msr, data);
3477 	case MSR_IA32_TSC_DEADLINE:
3478 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
3479 		break;
3480 	case MSR_IA32_TSC_ADJUST:
3481 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3482 			if (!msr_info->host_initiated) {
3483 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3484 				adjust_tsc_offset_guest(vcpu, adj);
3485 				/* Before back to guest, tsc_timestamp must be adjusted
3486 				 * as well, otherwise guest's percpu pvclock time could jump.
3487 				 */
3488 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3489 			}
3490 			vcpu->arch.ia32_tsc_adjust_msr = data;
3491 		}
3492 		break;
3493 	case MSR_IA32_MISC_ENABLE:
3494 		if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3495 		    ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3496 			if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3497 				return 1;
3498 			vcpu->arch.ia32_misc_enable_msr = data;
3499 			kvm_update_cpuid_runtime(vcpu);
3500 		} else {
3501 			vcpu->arch.ia32_misc_enable_msr = data;
3502 		}
3503 		break;
3504 	case MSR_IA32_SMBASE:
3505 		if (!msr_info->host_initiated)
3506 			return 1;
3507 		vcpu->arch.smbase = data;
3508 		break;
3509 	case MSR_IA32_POWER_CTL:
3510 		vcpu->arch.msr_ia32_power_ctl = data;
3511 		break;
3512 	case MSR_IA32_TSC:
3513 		if (msr_info->host_initiated) {
3514 			kvm_synchronize_tsc(vcpu, data);
3515 		} else {
3516 			u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3517 			adjust_tsc_offset_guest(vcpu, adj);
3518 			vcpu->arch.ia32_tsc_adjust_msr += adj;
3519 		}
3520 		break;
3521 	case MSR_IA32_XSS:
3522 		if (!msr_info->host_initiated &&
3523 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3524 			return 1;
3525 		/*
3526 		 * KVM supports exposing PT to the guest, but does not support
3527 		 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3528 		 * XSAVES/XRSTORS to save/restore PT MSRs.
3529 		 */
3530 		if (data & ~supported_xss)
3531 			return 1;
3532 		vcpu->arch.ia32_xss = data;
3533 		break;
3534 	case MSR_SMI_COUNT:
3535 		if (!msr_info->host_initiated)
3536 			return 1;
3537 		vcpu->arch.smi_count = data;
3538 		break;
3539 	case MSR_KVM_WALL_CLOCK_NEW:
3540 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3541 			return 1;
3542 
3543 		vcpu->kvm->arch.wall_clock = data;
3544 		kvm_write_wall_clock(vcpu->kvm, data, 0);
3545 		break;
3546 	case MSR_KVM_WALL_CLOCK:
3547 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3548 			return 1;
3549 
3550 		vcpu->kvm->arch.wall_clock = data;
3551 		kvm_write_wall_clock(vcpu->kvm, data, 0);
3552 		break;
3553 	case MSR_KVM_SYSTEM_TIME_NEW:
3554 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3555 			return 1;
3556 
3557 		kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3558 		break;
3559 	case MSR_KVM_SYSTEM_TIME:
3560 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3561 			return 1;
3562 
3563 		kvm_write_system_time(vcpu, data, true,  msr_info->host_initiated);
3564 		break;
3565 	case MSR_KVM_ASYNC_PF_EN:
3566 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3567 			return 1;
3568 
3569 		if (kvm_pv_enable_async_pf(vcpu, data))
3570 			return 1;
3571 		break;
3572 	case MSR_KVM_ASYNC_PF_INT:
3573 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3574 			return 1;
3575 
3576 		if (kvm_pv_enable_async_pf_int(vcpu, data))
3577 			return 1;
3578 		break;
3579 	case MSR_KVM_ASYNC_PF_ACK:
3580 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3581 			return 1;
3582 		if (data & 0x1) {
3583 			vcpu->arch.apf.pageready_pending = false;
3584 			kvm_check_async_pf_completion(vcpu);
3585 		}
3586 		break;
3587 	case MSR_KVM_STEAL_TIME:
3588 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3589 			return 1;
3590 
3591 		if (unlikely(!sched_info_on()))
3592 			return 1;
3593 
3594 		if (data & KVM_STEAL_RESERVED_MASK)
3595 			return 1;
3596 
3597 		vcpu->arch.st.msr_val = data;
3598 
3599 		if (!(data & KVM_MSR_ENABLED))
3600 			break;
3601 
3602 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3603 
3604 		break;
3605 	case MSR_KVM_PV_EOI_EN:
3606 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3607 			return 1;
3608 
3609 		if (kvm_lapic_set_pv_eoi(vcpu, data, sizeof(u8)))
3610 			return 1;
3611 		break;
3612 
3613 	case MSR_KVM_POLL_CONTROL:
3614 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3615 			return 1;
3616 
3617 		/* only enable bit supported */
3618 		if (data & (-1ULL << 1))
3619 			return 1;
3620 
3621 		vcpu->arch.msr_kvm_poll_control = data;
3622 		break;
3623 
3624 	case MSR_IA32_MCG_CTL:
3625 	case MSR_IA32_MCG_STATUS:
3626 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3627 		return set_msr_mce(vcpu, msr_info);
3628 
3629 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3630 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3631 		pr = true;
3632 		fallthrough;
3633 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3634 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3635 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3636 			return kvm_pmu_set_msr(vcpu, msr_info);
3637 
3638 		if (pr || data != 0)
3639 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3640 				    "0x%x data 0x%llx\n", msr, data);
3641 		break;
3642 	case MSR_K7_CLK_CTL:
3643 		/*
3644 		 * Ignore all writes to this no longer documented MSR.
3645 		 * Writes are only relevant for old K7 processors,
3646 		 * all pre-dating SVM, but a recommended workaround from
3647 		 * AMD for these chips. It is possible to specify the
3648 		 * affected processor models on the command line, hence
3649 		 * the need to ignore the workaround.
3650 		 */
3651 		break;
3652 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3653 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3654 	case HV_X64_MSR_SYNDBG_OPTIONS:
3655 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3656 	case HV_X64_MSR_CRASH_CTL:
3657 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3658 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3659 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3660 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3661 		return kvm_hv_set_msr_common(vcpu, msr, data,
3662 					     msr_info->host_initiated);
3663 	case MSR_IA32_BBL_CR_CTL3:
3664 		/* Drop writes to this legacy MSR -- see rdmsr
3665 		 * counterpart for further detail.
3666 		 */
3667 		if (report_ignored_msrs)
3668 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3669 				msr, data);
3670 		break;
3671 	case MSR_AMD64_OSVW_ID_LENGTH:
3672 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3673 			return 1;
3674 		vcpu->arch.osvw.length = data;
3675 		break;
3676 	case MSR_AMD64_OSVW_STATUS:
3677 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3678 			return 1;
3679 		vcpu->arch.osvw.status = data;
3680 		break;
3681 	case MSR_PLATFORM_INFO:
3682 		if (!msr_info->host_initiated ||
3683 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3684 		     cpuid_fault_enabled(vcpu)))
3685 			return 1;
3686 		vcpu->arch.msr_platform_info = data;
3687 		break;
3688 	case MSR_MISC_FEATURES_ENABLES:
3689 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3690 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3691 		     !supports_cpuid_fault(vcpu)))
3692 			return 1;
3693 		vcpu->arch.msr_misc_features_enables = data;
3694 		break;
3695 #ifdef CONFIG_X86_64
3696 	case MSR_IA32_XFD:
3697 		if (!msr_info->host_initiated &&
3698 		    !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
3699 			return 1;
3700 
3701 		if (data & ~(XFEATURE_MASK_USER_DYNAMIC &
3702 			     vcpu->arch.guest_supported_xcr0))
3703 			return 1;
3704 
3705 		fpu_update_guest_xfd(&vcpu->arch.guest_fpu, data);
3706 		break;
3707 	case MSR_IA32_XFD_ERR:
3708 		if (!msr_info->host_initiated &&
3709 		    !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
3710 			return 1;
3711 
3712 		if (data & ~(XFEATURE_MASK_USER_DYNAMIC &
3713 			     vcpu->arch.guest_supported_xcr0))
3714 			return 1;
3715 
3716 		vcpu->arch.guest_fpu.xfd_err = data;
3717 		break;
3718 #endif
3719 	default:
3720 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3721 			return kvm_pmu_set_msr(vcpu, msr_info);
3722 		return KVM_MSR_RET_INVALID;
3723 	}
3724 	return 0;
3725 }
3726 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3727 
3728 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3729 {
3730 	u64 data;
3731 	u64 mcg_cap = vcpu->arch.mcg_cap;
3732 	unsigned bank_num = mcg_cap & 0xff;
3733 
3734 	switch (msr) {
3735 	case MSR_IA32_P5_MC_ADDR:
3736 	case MSR_IA32_P5_MC_TYPE:
3737 		data = 0;
3738 		break;
3739 	case MSR_IA32_MCG_CAP:
3740 		data = vcpu->arch.mcg_cap;
3741 		break;
3742 	case MSR_IA32_MCG_CTL:
3743 		if (!(mcg_cap & MCG_CTL_P) && !host)
3744 			return 1;
3745 		data = vcpu->arch.mcg_ctl;
3746 		break;
3747 	case MSR_IA32_MCG_STATUS:
3748 		data = vcpu->arch.mcg_status;
3749 		break;
3750 	default:
3751 		if (msr >= MSR_IA32_MC0_CTL &&
3752 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
3753 			u32 offset = array_index_nospec(
3754 				msr - MSR_IA32_MC0_CTL,
3755 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3756 
3757 			data = vcpu->arch.mce_banks[offset];
3758 			break;
3759 		}
3760 		return 1;
3761 	}
3762 	*pdata = data;
3763 	return 0;
3764 }
3765 
3766 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3767 {
3768 	switch (msr_info->index) {
3769 	case MSR_IA32_PLATFORM_ID:
3770 	case MSR_IA32_EBL_CR_POWERON:
3771 	case MSR_IA32_LASTBRANCHFROMIP:
3772 	case MSR_IA32_LASTBRANCHTOIP:
3773 	case MSR_IA32_LASTINTFROMIP:
3774 	case MSR_IA32_LASTINTTOIP:
3775 	case MSR_AMD64_SYSCFG:
3776 	case MSR_K8_TSEG_ADDR:
3777 	case MSR_K8_TSEG_MASK:
3778 	case MSR_VM_HSAVE_PA:
3779 	case MSR_K8_INT_PENDING_MSG:
3780 	case MSR_AMD64_NB_CFG:
3781 	case MSR_FAM10H_MMIO_CONF_BASE:
3782 	case MSR_AMD64_BU_CFG2:
3783 	case MSR_IA32_PERF_CTL:
3784 	case MSR_AMD64_DC_CFG:
3785 	case MSR_F15H_EX_CFG:
3786 	/*
3787 	 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3788 	 * limit) MSRs. Just return 0, as we do not want to expose the host
3789 	 * data here. Do not conditionalize this on CPUID, as KVM does not do
3790 	 * so for existing CPU-specific MSRs.
3791 	 */
3792 	case MSR_RAPL_POWER_UNIT:
3793 	case MSR_PP0_ENERGY_STATUS:	/* Power plane 0 (core) */
3794 	case MSR_PP1_ENERGY_STATUS:	/* Power plane 1 (graphics uncore) */
3795 	case MSR_PKG_ENERGY_STATUS:	/* Total package */
3796 	case MSR_DRAM_ENERGY_STATUS:	/* DRAM controller */
3797 		msr_info->data = 0;
3798 		break;
3799 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3800 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3801 			return kvm_pmu_get_msr(vcpu, msr_info);
3802 		if (!msr_info->host_initiated)
3803 			return 1;
3804 		msr_info->data = 0;
3805 		break;
3806 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3807 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3808 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3809 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3810 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3811 			return kvm_pmu_get_msr(vcpu, msr_info);
3812 		msr_info->data = 0;
3813 		break;
3814 	case MSR_IA32_UCODE_REV:
3815 		msr_info->data = vcpu->arch.microcode_version;
3816 		break;
3817 	case MSR_IA32_ARCH_CAPABILITIES:
3818 		if (!msr_info->host_initiated &&
3819 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3820 			return 1;
3821 		msr_info->data = vcpu->arch.arch_capabilities;
3822 		break;
3823 	case MSR_IA32_PERF_CAPABILITIES:
3824 		if (!msr_info->host_initiated &&
3825 		    !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3826 			return 1;
3827 		msr_info->data = vcpu->arch.perf_capabilities;
3828 		break;
3829 	case MSR_IA32_POWER_CTL:
3830 		msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3831 		break;
3832 	case MSR_IA32_TSC: {
3833 		/*
3834 		 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3835 		 * even when not intercepted. AMD manual doesn't explicitly
3836 		 * state this but appears to behave the same.
3837 		 *
3838 		 * On userspace reads and writes, however, we unconditionally
3839 		 * return L1's TSC value to ensure backwards-compatible
3840 		 * behavior for migration.
3841 		 */
3842 		u64 offset, ratio;
3843 
3844 		if (msr_info->host_initiated) {
3845 			offset = vcpu->arch.l1_tsc_offset;
3846 			ratio = vcpu->arch.l1_tsc_scaling_ratio;
3847 		} else {
3848 			offset = vcpu->arch.tsc_offset;
3849 			ratio = vcpu->arch.tsc_scaling_ratio;
3850 		}
3851 
3852 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc(), ratio) + offset;
3853 		break;
3854 	}
3855 	case MSR_MTRRcap:
3856 	case 0x200 ... 0x2ff:
3857 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3858 	case 0xcd: /* fsb frequency */
3859 		msr_info->data = 3;
3860 		break;
3861 		/*
3862 		 * MSR_EBC_FREQUENCY_ID
3863 		 * Conservative value valid for even the basic CPU models.
3864 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3865 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3866 		 * and 266MHz for model 3, or 4. Set Core Clock
3867 		 * Frequency to System Bus Frequency Ratio to 1 (bits
3868 		 * 31:24) even though these are only valid for CPU
3869 		 * models > 2, however guests may end up dividing or
3870 		 * multiplying by zero otherwise.
3871 		 */
3872 	case MSR_EBC_FREQUENCY_ID:
3873 		msr_info->data = 1 << 24;
3874 		break;
3875 	case MSR_IA32_APICBASE:
3876 		msr_info->data = kvm_get_apic_base(vcpu);
3877 		break;
3878 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3879 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3880 	case MSR_IA32_TSC_DEADLINE:
3881 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3882 		break;
3883 	case MSR_IA32_TSC_ADJUST:
3884 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3885 		break;
3886 	case MSR_IA32_MISC_ENABLE:
3887 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3888 		break;
3889 	case MSR_IA32_SMBASE:
3890 		if (!msr_info->host_initiated)
3891 			return 1;
3892 		msr_info->data = vcpu->arch.smbase;
3893 		break;
3894 	case MSR_SMI_COUNT:
3895 		msr_info->data = vcpu->arch.smi_count;
3896 		break;
3897 	case MSR_IA32_PERF_STATUS:
3898 		/* TSC increment by tick */
3899 		msr_info->data = 1000ULL;
3900 		/* CPU multiplier */
3901 		msr_info->data |= (((uint64_t)4ULL) << 40);
3902 		break;
3903 	case MSR_EFER:
3904 		msr_info->data = vcpu->arch.efer;
3905 		break;
3906 	case MSR_KVM_WALL_CLOCK:
3907 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3908 			return 1;
3909 
3910 		msr_info->data = vcpu->kvm->arch.wall_clock;
3911 		break;
3912 	case MSR_KVM_WALL_CLOCK_NEW:
3913 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3914 			return 1;
3915 
3916 		msr_info->data = vcpu->kvm->arch.wall_clock;
3917 		break;
3918 	case MSR_KVM_SYSTEM_TIME:
3919 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3920 			return 1;
3921 
3922 		msr_info->data = vcpu->arch.time;
3923 		break;
3924 	case MSR_KVM_SYSTEM_TIME_NEW:
3925 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3926 			return 1;
3927 
3928 		msr_info->data = vcpu->arch.time;
3929 		break;
3930 	case MSR_KVM_ASYNC_PF_EN:
3931 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3932 			return 1;
3933 
3934 		msr_info->data = vcpu->arch.apf.msr_en_val;
3935 		break;
3936 	case MSR_KVM_ASYNC_PF_INT:
3937 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3938 			return 1;
3939 
3940 		msr_info->data = vcpu->arch.apf.msr_int_val;
3941 		break;
3942 	case MSR_KVM_ASYNC_PF_ACK:
3943 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3944 			return 1;
3945 
3946 		msr_info->data = 0;
3947 		break;
3948 	case MSR_KVM_STEAL_TIME:
3949 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3950 			return 1;
3951 
3952 		msr_info->data = vcpu->arch.st.msr_val;
3953 		break;
3954 	case MSR_KVM_PV_EOI_EN:
3955 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3956 			return 1;
3957 
3958 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
3959 		break;
3960 	case MSR_KVM_POLL_CONTROL:
3961 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3962 			return 1;
3963 
3964 		msr_info->data = vcpu->arch.msr_kvm_poll_control;
3965 		break;
3966 	case MSR_IA32_P5_MC_ADDR:
3967 	case MSR_IA32_P5_MC_TYPE:
3968 	case MSR_IA32_MCG_CAP:
3969 	case MSR_IA32_MCG_CTL:
3970 	case MSR_IA32_MCG_STATUS:
3971 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3972 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3973 				   msr_info->host_initiated);
3974 	case MSR_IA32_XSS:
3975 		if (!msr_info->host_initiated &&
3976 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3977 			return 1;
3978 		msr_info->data = vcpu->arch.ia32_xss;
3979 		break;
3980 	case MSR_K7_CLK_CTL:
3981 		/*
3982 		 * Provide expected ramp-up count for K7. All other
3983 		 * are set to zero, indicating minimum divisors for
3984 		 * every field.
3985 		 *
3986 		 * This prevents guest kernels on AMD host with CPU
3987 		 * type 6, model 8 and higher from exploding due to
3988 		 * the rdmsr failing.
3989 		 */
3990 		msr_info->data = 0x20000000;
3991 		break;
3992 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3993 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3994 	case HV_X64_MSR_SYNDBG_OPTIONS:
3995 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3996 	case HV_X64_MSR_CRASH_CTL:
3997 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3998 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3999 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
4000 	case HV_X64_MSR_TSC_EMULATION_STATUS:
4001 		return kvm_hv_get_msr_common(vcpu,
4002 					     msr_info->index, &msr_info->data,
4003 					     msr_info->host_initiated);
4004 	case MSR_IA32_BBL_CR_CTL3:
4005 		/* This legacy MSR exists but isn't fully documented in current
4006 		 * silicon.  It is however accessed by winxp in very narrow
4007 		 * scenarios where it sets bit #19, itself documented as
4008 		 * a "reserved" bit.  Best effort attempt to source coherent
4009 		 * read data here should the balance of the register be
4010 		 * interpreted by the guest:
4011 		 *
4012 		 * L2 cache control register 3: 64GB range, 256KB size,
4013 		 * enabled, latency 0x1, configured
4014 		 */
4015 		msr_info->data = 0xbe702111;
4016 		break;
4017 	case MSR_AMD64_OSVW_ID_LENGTH:
4018 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
4019 			return 1;
4020 		msr_info->data = vcpu->arch.osvw.length;
4021 		break;
4022 	case MSR_AMD64_OSVW_STATUS:
4023 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
4024 			return 1;
4025 		msr_info->data = vcpu->arch.osvw.status;
4026 		break;
4027 	case MSR_PLATFORM_INFO:
4028 		if (!msr_info->host_initiated &&
4029 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
4030 			return 1;
4031 		msr_info->data = vcpu->arch.msr_platform_info;
4032 		break;
4033 	case MSR_MISC_FEATURES_ENABLES:
4034 		msr_info->data = vcpu->arch.msr_misc_features_enables;
4035 		break;
4036 	case MSR_K7_HWCR:
4037 		msr_info->data = vcpu->arch.msr_hwcr;
4038 		break;
4039 #ifdef CONFIG_X86_64
4040 	case MSR_IA32_XFD:
4041 		if (!msr_info->host_initiated &&
4042 		    !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
4043 			return 1;
4044 
4045 		msr_info->data = vcpu->arch.guest_fpu.fpstate->xfd;
4046 		break;
4047 	case MSR_IA32_XFD_ERR:
4048 		if (!msr_info->host_initiated &&
4049 		    !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
4050 			return 1;
4051 
4052 		msr_info->data = vcpu->arch.guest_fpu.xfd_err;
4053 		break;
4054 #endif
4055 	default:
4056 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
4057 			return kvm_pmu_get_msr(vcpu, msr_info);
4058 		return KVM_MSR_RET_INVALID;
4059 	}
4060 	return 0;
4061 }
4062 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
4063 
4064 /*
4065  * Read or write a bunch of msrs. All parameters are kernel addresses.
4066  *
4067  * @return number of msrs set successfully.
4068  */
4069 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
4070 		    struct kvm_msr_entry *entries,
4071 		    int (*do_msr)(struct kvm_vcpu *vcpu,
4072 				  unsigned index, u64 *data))
4073 {
4074 	int i;
4075 
4076 	for (i = 0; i < msrs->nmsrs; ++i)
4077 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
4078 			break;
4079 
4080 	return i;
4081 }
4082 
4083 /*
4084  * Read or write a bunch of msrs. Parameters are user addresses.
4085  *
4086  * @return number of msrs set successfully.
4087  */
4088 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
4089 		  int (*do_msr)(struct kvm_vcpu *vcpu,
4090 				unsigned index, u64 *data),
4091 		  int writeback)
4092 {
4093 	struct kvm_msrs msrs;
4094 	struct kvm_msr_entry *entries;
4095 	int r, n;
4096 	unsigned size;
4097 
4098 	r = -EFAULT;
4099 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
4100 		goto out;
4101 
4102 	r = -E2BIG;
4103 	if (msrs.nmsrs >= MAX_IO_MSRS)
4104 		goto out;
4105 
4106 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
4107 	entries = memdup_user(user_msrs->entries, size);
4108 	if (IS_ERR(entries)) {
4109 		r = PTR_ERR(entries);
4110 		goto out;
4111 	}
4112 
4113 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
4114 	if (r < 0)
4115 		goto out_free;
4116 
4117 	r = -EFAULT;
4118 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
4119 		goto out_free;
4120 
4121 	r = n;
4122 
4123 out_free:
4124 	kfree(entries);
4125 out:
4126 	return r;
4127 }
4128 
4129 static inline bool kvm_can_mwait_in_guest(void)
4130 {
4131 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
4132 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
4133 		boot_cpu_has(X86_FEATURE_ARAT);
4134 }
4135 
4136 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
4137 					    struct kvm_cpuid2 __user *cpuid_arg)
4138 {
4139 	struct kvm_cpuid2 cpuid;
4140 	int r;
4141 
4142 	r = -EFAULT;
4143 	if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4144 		return r;
4145 
4146 	r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4147 	if (r)
4148 		return r;
4149 
4150 	r = -EFAULT;
4151 	if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4152 		return r;
4153 
4154 	return 0;
4155 }
4156 
4157 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
4158 {
4159 	int r = 0;
4160 
4161 	switch (ext) {
4162 	case KVM_CAP_IRQCHIP:
4163 	case KVM_CAP_HLT:
4164 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
4165 	case KVM_CAP_SET_TSS_ADDR:
4166 	case KVM_CAP_EXT_CPUID:
4167 	case KVM_CAP_EXT_EMUL_CPUID:
4168 	case KVM_CAP_CLOCKSOURCE:
4169 	case KVM_CAP_PIT:
4170 	case KVM_CAP_NOP_IO_DELAY:
4171 	case KVM_CAP_MP_STATE:
4172 	case KVM_CAP_SYNC_MMU:
4173 	case KVM_CAP_USER_NMI:
4174 	case KVM_CAP_REINJECT_CONTROL:
4175 	case KVM_CAP_IRQ_INJECT_STATUS:
4176 	case KVM_CAP_IOEVENTFD:
4177 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
4178 	case KVM_CAP_PIT2:
4179 	case KVM_CAP_PIT_STATE2:
4180 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
4181 	case KVM_CAP_VCPU_EVENTS:
4182 	case KVM_CAP_HYPERV:
4183 	case KVM_CAP_HYPERV_VAPIC:
4184 	case KVM_CAP_HYPERV_SPIN:
4185 	case KVM_CAP_HYPERV_SYNIC:
4186 	case KVM_CAP_HYPERV_SYNIC2:
4187 	case KVM_CAP_HYPERV_VP_INDEX:
4188 	case KVM_CAP_HYPERV_EVENTFD:
4189 	case KVM_CAP_HYPERV_TLBFLUSH:
4190 	case KVM_CAP_HYPERV_SEND_IPI:
4191 	case KVM_CAP_HYPERV_CPUID:
4192 	case KVM_CAP_HYPERV_ENFORCE_CPUID:
4193 	case KVM_CAP_SYS_HYPERV_CPUID:
4194 	case KVM_CAP_PCI_SEGMENT:
4195 	case KVM_CAP_DEBUGREGS:
4196 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
4197 	case KVM_CAP_XSAVE:
4198 	case KVM_CAP_ASYNC_PF:
4199 	case KVM_CAP_ASYNC_PF_INT:
4200 	case KVM_CAP_GET_TSC_KHZ:
4201 	case KVM_CAP_KVMCLOCK_CTRL:
4202 	case KVM_CAP_READONLY_MEM:
4203 	case KVM_CAP_HYPERV_TIME:
4204 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
4205 	case KVM_CAP_TSC_DEADLINE_TIMER:
4206 	case KVM_CAP_DISABLE_QUIRKS:
4207 	case KVM_CAP_SET_BOOT_CPU_ID:
4208  	case KVM_CAP_SPLIT_IRQCHIP:
4209 	case KVM_CAP_IMMEDIATE_EXIT:
4210 	case KVM_CAP_PMU_EVENT_FILTER:
4211 	case KVM_CAP_GET_MSR_FEATURES:
4212 	case KVM_CAP_MSR_PLATFORM_INFO:
4213 	case KVM_CAP_EXCEPTION_PAYLOAD:
4214 	case KVM_CAP_SET_GUEST_DEBUG:
4215 	case KVM_CAP_LAST_CPU:
4216 	case KVM_CAP_X86_USER_SPACE_MSR:
4217 	case KVM_CAP_X86_MSR_FILTER:
4218 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4219 #ifdef CONFIG_X86_SGX_KVM
4220 	case KVM_CAP_SGX_ATTRIBUTE:
4221 #endif
4222 	case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
4223 	case KVM_CAP_VM_MOVE_ENC_CONTEXT_FROM:
4224 	case KVM_CAP_SREGS2:
4225 	case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
4226 	case KVM_CAP_VCPU_ATTRIBUTES:
4227 		r = 1;
4228 		break;
4229 	case KVM_CAP_EXIT_HYPERCALL:
4230 		r = KVM_EXIT_HYPERCALL_VALID_MASK;
4231 		break;
4232 	case KVM_CAP_SET_GUEST_DEBUG2:
4233 		return KVM_GUESTDBG_VALID_MASK;
4234 #ifdef CONFIG_KVM_XEN
4235 	case KVM_CAP_XEN_HVM:
4236 		r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
4237 		    KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
4238 		    KVM_XEN_HVM_CONFIG_SHARED_INFO |
4239 		    KVM_XEN_HVM_CONFIG_EVTCHN_2LEVEL;
4240 		if (sched_info_on())
4241 			r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
4242 		break;
4243 #endif
4244 	case KVM_CAP_SYNC_REGS:
4245 		r = KVM_SYNC_X86_VALID_FIELDS;
4246 		break;
4247 	case KVM_CAP_ADJUST_CLOCK:
4248 		r = KVM_CLOCK_VALID_FLAGS;
4249 		break;
4250 	case KVM_CAP_X86_DISABLE_EXITS:
4251 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
4252 		      KVM_X86_DISABLE_EXITS_CSTATE;
4253 		if(kvm_can_mwait_in_guest())
4254 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
4255 		break;
4256 	case KVM_CAP_X86_SMM:
4257 		/* SMBASE is usually relocated above 1M on modern chipsets,
4258 		 * and SMM handlers might indeed rely on 4G segment limits,
4259 		 * so do not report SMM to be available if real mode is
4260 		 * emulated via vm86 mode.  Still, do not go to great lengths
4261 		 * to avoid userspace's usage of the feature, because it is a
4262 		 * fringe case that is not enabled except via specific settings
4263 		 * of the module parameters.
4264 		 */
4265 		r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
4266 		break;
4267 	case KVM_CAP_VAPIC:
4268 		r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
4269 		break;
4270 	case KVM_CAP_NR_VCPUS:
4271 		r = min_t(unsigned int, num_online_cpus(), KVM_MAX_VCPUS);
4272 		break;
4273 	case KVM_CAP_MAX_VCPUS:
4274 		r = KVM_MAX_VCPUS;
4275 		break;
4276 	case KVM_CAP_MAX_VCPU_ID:
4277 		r = KVM_MAX_VCPU_IDS;
4278 		break;
4279 	case KVM_CAP_PV_MMU:	/* obsolete */
4280 		r = 0;
4281 		break;
4282 	case KVM_CAP_MCE:
4283 		r = KVM_MAX_MCE_BANKS;
4284 		break;
4285 	case KVM_CAP_XCRS:
4286 		r = boot_cpu_has(X86_FEATURE_XSAVE);
4287 		break;
4288 	case KVM_CAP_TSC_CONTROL:
4289 		r = kvm_has_tsc_control;
4290 		break;
4291 	case KVM_CAP_X2APIC_API:
4292 		r = KVM_X2APIC_API_VALID_FLAGS;
4293 		break;
4294 	case KVM_CAP_NESTED_STATE:
4295 		r = kvm_x86_ops.nested_ops->get_state ?
4296 			kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
4297 		break;
4298 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4299 		r = kvm_x86_ops.enable_direct_tlbflush != NULL;
4300 		break;
4301 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4302 		r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
4303 		break;
4304 	case KVM_CAP_SMALLER_MAXPHYADDR:
4305 		r = (int) allow_smaller_maxphyaddr;
4306 		break;
4307 	case KVM_CAP_STEAL_TIME:
4308 		r = sched_info_on();
4309 		break;
4310 	case KVM_CAP_X86_BUS_LOCK_EXIT:
4311 		if (kvm_has_bus_lock_exit)
4312 			r = KVM_BUS_LOCK_DETECTION_OFF |
4313 			    KVM_BUS_LOCK_DETECTION_EXIT;
4314 		else
4315 			r = 0;
4316 		break;
4317 	case KVM_CAP_XSAVE2: {
4318 		u64 guest_perm = xstate_get_guest_group_perm();
4319 
4320 		r = xstate_required_size(supported_xcr0 & guest_perm, false);
4321 		if (r < sizeof(struct kvm_xsave))
4322 			r = sizeof(struct kvm_xsave);
4323 		break;
4324 	}
4325 	default:
4326 		break;
4327 	}
4328 	return r;
4329 
4330 }
4331 
4332 long kvm_arch_dev_ioctl(struct file *filp,
4333 			unsigned int ioctl, unsigned long arg)
4334 {
4335 	void __user *argp = (void __user *)arg;
4336 	long r;
4337 
4338 	switch (ioctl) {
4339 	case KVM_GET_MSR_INDEX_LIST: {
4340 		struct kvm_msr_list __user *user_msr_list = argp;
4341 		struct kvm_msr_list msr_list;
4342 		unsigned n;
4343 
4344 		r = -EFAULT;
4345 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4346 			goto out;
4347 		n = msr_list.nmsrs;
4348 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4349 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4350 			goto out;
4351 		r = -E2BIG;
4352 		if (n < msr_list.nmsrs)
4353 			goto out;
4354 		r = -EFAULT;
4355 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4356 				 num_msrs_to_save * sizeof(u32)))
4357 			goto out;
4358 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4359 				 &emulated_msrs,
4360 				 num_emulated_msrs * sizeof(u32)))
4361 			goto out;
4362 		r = 0;
4363 		break;
4364 	}
4365 	case KVM_GET_SUPPORTED_CPUID:
4366 	case KVM_GET_EMULATED_CPUID: {
4367 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4368 		struct kvm_cpuid2 cpuid;
4369 
4370 		r = -EFAULT;
4371 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4372 			goto out;
4373 
4374 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4375 					    ioctl);
4376 		if (r)
4377 			goto out;
4378 
4379 		r = -EFAULT;
4380 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4381 			goto out;
4382 		r = 0;
4383 		break;
4384 	}
4385 	case KVM_X86_GET_MCE_CAP_SUPPORTED:
4386 		r = -EFAULT;
4387 		if (copy_to_user(argp, &kvm_mce_cap_supported,
4388 				 sizeof(kvm_mce_cap_supported)))
4389 			goto out;
4390 		r = 0;
4391 		break;
4392 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4393 		struct kvm_msr_list __user *user_msr_list = argp;
4394 		struct kvm_msr_list msr_list;
4395 		unsigned int n;
4396 
4397 		r = -EFAULT;
4398 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4399 			goto out;
4400 		n = msr_list.nmsrs;
4401 		msr_list.nmsrs = num_msr_based_features;
4402 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4403 			goto out;
4404 		r = -E2BIG;
4405 		if (n < msr_list.nmsrs)
4406 			goto out;
4407 		r = -EFAULT;
4408 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
4409 				 num_msr_based_features * sizeof(u32)))
4410 			goto out;
4411 		r = 0;
4412 		break;
4413 	}
4414 	case KVM_GET_MSRS:
4415 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
4416 		break;
4417 	case KVM_GET_SUPPORTED_HV_CPUID:
4418 		r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4419 		break;
4420 	default:
4421 		r = -EINVAL;
4422 		break;
4423 	}
4424 out:
4425 	return r;
4426 }
4427 
4428 static void wbinvd_ipi(void *garbage)
4429 {
4430 	wbinvd();
4431 }
4432 
4433 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4434 {
4435 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4436 }
4437 
4438 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4439 {
4440 	/* Address WBINVD may be executed by guest */
4441 	if (need_emulate_wbinvd(vcpu)) {
4442 		if (static_call(kvm_x86_has_wbinvd_exit)())
4443 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4444 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4445 			smp_call_function_single(vcpu->cpu,
4446 					wbinvd_ipi, NULL, 1);
4447 	}
4448 
4449 	static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4450 
4451 	/* Save host pkru register if supported */
4452 	vcpu->arch.host_pkru = read_pkru();
4453 
4454 	/* Apply any externally detected TSC adjustments (due to suspend) */
4455 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4456 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4457 		vcpu->arch.tsc_offset_adjustment = 0;
4458 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4459 	}
4460 
4461 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4462 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4463 				rdtsc() - vcpu->arch.last_host_tsc;
4464 		if (tsc_delta < 0)
4465 			mark_tsc_unstable("KVM discovered backwards TSC");
4466 
4467 		if (kvm_check_tsc_unstable()) {
4468 			u64 offset = kvm_compute_l1_tsc_offset(vcpu,
4469 						vcpu->arch.last_guest_tsc);
4470 			kvm_vcpu_write_tsc_offset(vcpu, offset);
4471 			vcpu->arch.tsc_catchup = 1;
4472 		}
4473 
4474 		if (kvm_lapic_hv_timer_in_use(vcpu))
4475 			kvm_lapic_restart_hv_timer(vcpu);
4476 
4477 		/*
4478 		 * On a host with synchronized TSC, there is no need to update
4479 		 * kvmclock on vcpu->cpu migration
4480 		 */
4481 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4482 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4483 		if (vcpu->cpu != cpu)
4484 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4485 		vcpu->cpu = cpu;
4486 	}
4487 
4488 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4489 }
4490 
4491 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4492 {
4493 	struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
4494 	struct kvm_steal_time __user *st;
4495 	struct kvm_memslots *slots;
4496 	static const u8 preempted = KVM_VCPU_PREEMPTED;
4497 
4498 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4499 		return;
4500 
4501 	if (vcpu->arch.st.preempted)
4502 		return;
4503 
4504 	/* This happens on process exit */
4505 	if (unlikely(current->mm != vcpu->kvm->mm))
4506 		return;
4507 
4508 	slots = kvm_memslots(vcpu->kvm);
4509 
4510 	if (unlikely(slots->generation != ghc->generation ||
4511 		     kvm_is_error_hva(ghc->hva) || !ghc->memslot))
4512 		return;
4513 
4514 	st = (struct kvm_steal_time __user *)ghc->hva;
4515 	BUILD_BUG_ON(sizeof(st->preempted) != sizeof(preempted));
4516 
4517 	if (!copy_to_user_nofault(&st->preempted, &preempted, sizeof(preempted)))
4518 		vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4519 
4520 	mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
4521 }
4522 
4523 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4524 {
4525 	int idx;
4526 
4527 	if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4528 		vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4529 
4530 	/*
4531 	 * Take the srcu lock as memslots will be accessed to check the gfn
4532 	 * cache generation against the memslots generation.
4533 	 */
4534 	idx = srcu_read_lock(&vcpu->kvm->srcu);
4535 	if (kvm_xen_msr_enabled(vcpu->kvm))
4536 		kvm_xen_runstate_set_preempted(vcpu);
4537 	else
4538 		kvm_steal_time_set_preempted(vcpu);
4539 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
4540 
4541 	static_call(kvm_x86_vcpu_put)(vcpu);
4542 	vcpu->arch.last_host_tsc = rdtsc();
4543 }
4544 
4545 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4546 				    struct kvm_lapic_state *s)
4547 {
4548 	static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
4549 
4550 	return kvm_apic_get_state(vcpu, s);
4551 }
4552 
4553 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4554 				    struct kvm_lapic_state *s)
4555 {
4556 	int r;
4557 
4558 	r = kvm_apic_set_state(vcpu, s);
4559 	if (r)
4560 		return r;
4561 	update_cr8_intercept(vcpu);
4562 
4563 	return 0;
4564 }
4565 
4566 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4567 {
4568 	/*
4569 	 * We can accept userspace's request for interrupt injection
4570 	 * as long as we have a place to store the interrupt number.
4571 	 * The actual injection will happen when the CPU is able to
4572 	 * deliver the interrupt.
4573 	 */
4574 	if (kvm_cpu_has_extint(vcpu))
4575 		return false;
4576 
4577 	/* Acknowledging ExtINT does not happen if LINT0 is masked.  */
4578 	return (!lapic_in_kernel(vcpu) ||
4579 		kvm_apic_accept_pic_intr(vcpu));
4580 }
4581 
4582 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4583 {
4584 	/*
4585 	 * Do not cause an interrupt window exit if an exception
4586 	 * is pending or an event needs reinjection; userspace
4587 	 * might want to inject the interrupt manually using KVM_SET_REGS
4588 	 * or KVM_SET_SREGS.  For that to work, we must be at an
4589 	 * instruction boundary and with no events half-injected.
4590 	 */
4591 	return (kvm_arch_interrupt_allowed(vcpu) &&
4592 		kvm_cpu_accept_dm_intr(vcpu) &&
4593 		!kvm_event_needs_reinjection(vcpu) &&
4594 		!vcpu->arch.exception.pending);
4595 }
4596 
4597 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4598 				    struct kvm_interrupt *irq)
4599 {
4600 	if (irq->irq >= KVM_NR_INTERRUPTS)
4601 		return -EINVAL;
4602 
4603 	if (!irqchip_in_kernel(vcpu->kvm)) {
4604 		kvm_queue_interrupt(vcpu, irq->irq, false);
4605 		kvm_make_request(KVM_REQ_EVENT, vcpu);
4606 		return 0;
4607 	}
4608 
4609 	/*
4610 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4611 	 * fail for in-kernel 8259.
4612 	 */
4613 	if (pic_in_kernel(vcpu->kvm))
4614 		return -ENXIO;
4615 
4616 	if (vcpu->arch.pending_external_vector != -1)
4617 		return -EEXIST;
4618 
4619 	vcpu->arch.pending_external_vector = irq->irq;
4620 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4621 	return 0;
4622 }
4623 
4624 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4625 {
4626 	kvm_inject_nmi(vcpu);
4627 
4628 	return 0;
4629 }
4630 
4631 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4632 {
4633 	kvm_make_request(KVM_REQ_SMI, vcpu);
4634 
4635 	return 0;
4636 }
4637 
4638 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4639 					   struct kvm_tpr_access_ctl *tac)
4640 {
4641 	if (tac->flags)
4642 		return -EINVAL;
4643 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
4644 	return 0;
4645 }
4646 
4647 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4648 					u64 mcg_cap)
4649 {
4650 	int r;
4651 	unsigned bank_num = mcg_cap & 0xff, bank;
4652 
4653 	r = -EINVAL;
4654 	if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4655 		goto out;
4656 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4657 		goto out;
4658 	r = 0;
4659 	vcpu->arch.mcg_cap = mcg_cap;
4660 	/* Init IA32_MCG_CTL to all 1s */
4661 	if (mcg_cap & MCG_CTL_P)
4662 		vcpu->arch.mcg_ctl = ~(u64)0;
4663 	/* Init IA32_MCi_CTL to all 1s */
4664 	for (bank = 0; bank < bank_num; bank++)
4665 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4666 
4667 	static_call(kvm_x86_setup_mce)(vcpu);
4668 out:
4669 	return r;
4670 }
4671 
4672 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4673 				      struct kvm_x86_mce *mce)
4674 {
4675 	u64 mcg_cap = vcpu->arch.mcg_cap;
4676 	unsigned bank_num = mcg_cap & 0xff;
4677 	u64 *banks = vcpu->arch.mce_banks;
4678 
4679 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4680 		return -EINVAL;
4681 	/*
4682 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4683 	 * reporting is disabled
4684 	 */
4685 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4686 	    vcpu->arch.mcg_ctl != ~(u64)0)
4687 		return 0;
4688 	banks += 4 * mce->bank;
4689 	/*
4690 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4691 	 * reporting is disabled for the bank
4692 	 */
4693 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4694 		return 0;
4695 	if (mce->status & MCI_STATUS_UC) {
4696 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4697 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4698 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4699 			return 0;
4700 		}
4701 		if (banks[1] & MCI_STATUS_VAL)
4702 			mce->status |= MCI_STATUS_OVER;
4703 		banks[2] = mce->addr;
4704 		banks[3] = mce->misc;
4705 		vcpu->arch.mcg_status = mce->mcg_status;
4706 		banks[1] = mce->status;
4707 		kvm_queue_exception(vcpu, MC_VECTOR);
4708 	} else if (!(banks[1] & MCI_STATUS_VAL)
4709 		   || !(banks[1] & MCI_STATUS_UC)) {
4710 		if (banks[1] & MCI_STATUS_VAL)
4711 			mce->status |= MCI_STATUS_OVER;
4712 		banks[2] = mce->addr;
4713 		banks[3] = mce->misc;
4714 		banks[1] = mce->status;
4715 	} else
4716 		banks[1] |= MCI_STATUS_OVER;
4717 	return 0;
4718 }
4719 
4720 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4721 					       struct kvm_vcpu_events *events)
4722 {
4723 	process_nmi(vcpu);
4724 
4725 	if (kvm_check_request(KVM_REQ_SMI, vcpu))
4726 		process_smi(vcpu);
4727 
4728 	/*
4729 	 * In guest mode, payload delivery should be deferred,
4730 	 * so that the L1 hypervisor can intercept #PF before
4731 	 * CR2 is modified (or intercept #DB before DR6 is
4732 	 * modified under nVMX). Unless the per-VM capability,
4733 	 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4734 	 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4735 	 * opportunistically defer the exception payload, deliver it if the
4736 	 * capability hasn't been requested before processing a
4737 	 * KVM_GET_VCPU_EVENTS.
4738 	 */
4739 	if (!vcpu->kvm->arch.exception_payload_enabled &&
4740 	    vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4741 		kvm_deliver_exception_payload(vcpu);
4742 
4743 	/*
4744 	 * The API doesn't provide the instruction length for software
4745 	 * exceptions, so don't report them. As long as the guest RIP
4746 	 * isn't advanced, we should expect to encounter the exception
4747 	 * again.
4748 	 */
4749 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4750 		events->exception.injected = 0;
4751 		events->exception.pending = 0;
4752 	} else {
4753 		events->exception.injected = vcpu->arch.exception.injected;
4754 		events->exception.pending = vcpu->arch.exception.pending;
4755 		/*
4756 		 * For ABI compatibility, deliberately conflate
4757 		 * pending and injected exceptions when
4758 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4759 		 */
4760 		if (!vcpu->kvm->arch.exception_payload_enabled)
4761 			events->exception.injected |=
4762 				vcpu->arch.exception.pending;
4763 	}
4764 	events->exception.nr = vcpu->arch.exception.nr;
4765 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4766 	events->exception.error_code = vcpu->arch.exception.error_code;
4767 	events->exception_has_payload = vcpu->arch.exception.has_payload;
4768 	events->exception_payload = vcpu->arch.exception.payload;
4769 
4770 	events->interrupt.injected =
4771 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4772 	events->interrupt.nr = vcpu->arch.interrupt.nr;
4773 	events->interrupt.soft = 0;
4774 	events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4775 
4776 	events->nmi.injected = vcpu->arch.nmi_injected;
4777 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
4778 	events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4779 	events->nmi.pad = 0;
4780 
4781 	events->sipi_vector = 0; /* never valid when reporting to user space */
4782 
4783 	events->smi.smm = is_smm(vcpu);
4784 	events->smi.pending = vcpu->arch.smi_pending;
4785 	events->smi.smm_inside_nmi =
4786 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4787 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4788 
4789 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4790 			 | KVM_VCPUEVENT_VALID_SHADOW
4791 			 | KVM_VCPUEVENT_VALID_SMM);
4792 	if (vcpu->kvm->arch.exception_payload_enabled)
4793 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4794 
4795 	memset(&events->reserved, 0, sizeof(events->reserved));
4796 }
4797 
4798 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm);
4799 
4800 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4801 					      struct kvm_vcpu_events *events)
4802 {
4803 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4804 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4805 			      | KVM_VCPUEVENT_VALID_SHADOW
4806 			      | KVM_VCPUEVENT_VALID_SMM
4807 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
4808 		return -EINVAL;
4809 
4810 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4811 		if (!vcpu->kvm->arch.exception_payload_enabled)
4812 			return -EINVAL;
4813 		if (events->exception.pending)
4814 			events->exception.injected = 0;
4815 		else
4816 			events->exception_has_payload = 0;
4817 	} else {
4818 		events->exception.pending = 0;
4819 		events->exception_has_payload = 0;
4820 	}
4821 
4822 	if ((events->exception.injected || events->exception.pending) &&
4823 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4824 		return -EINVAL;
4825 
4826 	/* INITs are latched while in SMM */
4827 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4828 	    (events->smi.smm || events->smi.pending) &&
4829 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4830 		return -EINVAL;
4831 
4832 	process_nmi(vcpu);
4833 	vcpu->arch.exception.injected = events->exception.injected;
4834 	vcpu->arch.exception.pending = events->exception.pending;
4835 	vcpu->arch.exception.nr = events->exception.nr;
4836 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4837 	vcpu->arch.exception.error_code = events->exception.error_code;
4838 	vcpu->arch.exception.has_payload = events->exception_has_payload;
4839 	vcpu->arch.exception.payload = events->exception_payload;
4840 
4841 	vcpu->arch.interrupt.injected = events->interrupt.injected;
4842 	vcpu->arch.interrupt.nr = events->interrupt.nr;
4843 	vcpu->arch.interrupt.soft = events->interrupt.soft;
4844 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4845 		static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4846 						events->interrupt.shadow);
4847 
4848 	vcpu->arch.nmi_injected = events->nmi.injected;
4849 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4850 		vcpu->arch.nmi_pending = events->nmi.pending;
4851 	static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4852 
4853 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4854 	    lapic_in_kernel(vcpu))
4855 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
4856 
4857 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4858 		if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm)
4859 			kvm_smm_changed(vcpu, events->smi.smm);
4860 
4861 		vcpu->arch.smi_pending = events->smi.pending;
4862 
4863 		if (events->smi.smm) {
4864 			if (events->smi.smm_inside_nmi)
4865 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4866 			else
4867 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4868 		}
4869 
4870 		if (lapic_in_kernel(vcpu)) {
4871 			if (events->smi.latched_init)
4872 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4873 			else
4874 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4875 		}
4876 	}
4877 
4878 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4879 
4880 	return 0;
4881 }
4882 
4883 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4884 					     struct kvm_debugregs *dbgregs)
4885 {
4886 	unsigned long val;
4887 
4888 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4889 	kvm_get_dr(vcpu, 6, &val);
4890 	dbgregs->dr6 = val;
4891 	dbgregs->dr7 = vcpu->arch.dr7;
4892 	dbgregs->flags = 0;
4893 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4894 }
4895 
4896 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4897 					    struct kvm_debugregs *dbgregs)
4898 {
4899 	if (dbgregs->flags)
4900 		return -EINVAL;
4901 
4902 	if (!kvm_dr6_valid(dbgregs->dr6))
4903 		return -EINVAL;
4904 	if (!kvm_dr7_valid(dbgregs->dr7))
4905 		return -EINVAL;
4906 
4907 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4908 	kvm_update_dr0123(vcpu);
4909 	vcpu->arch.dr6 = dbgregs->dr6;
4910 	vcpu->arch.dr7 = dbgregs->dr7;
4911 	kvm_update_dr7(vcpu);
4912 
4913 	return 0;
4914 }
4915 
4916 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4917 					 struct kvm_xsave *guest_xsave)
4918 {
4919 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
4920 		return;
4921 
4922 	fpu_copy_guest_fpstate_to_uabi(&vcpu->arch.guest_fpu,
4923 				       guest_xsave->region,
4924 				       sizeof(guest_xsave->region),
4925 				       vcpu->arch.pkru);
4926 }
4927 
4928 static void kvm_vcpu_ioctl_x86_get_xsave2(struct kvm_vcpu *vcpu,
4929 					  u8 *state, unsigned int size)
4930 {
4931 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
4932 		return;
4933 
4934 	fpu_copy_guest_fpstate_to_uabi(&vcpu->arch.guest_fpu,
4935 				       state, size, vcpu->arch.pkru);
4936 }
4937 
4938 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4939 					struct kvm_xsave *guest_xsave)
4940 {
4941 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
4942 		return 0;
4943 
4944 	return fpu_copy_uabi_to_guest_fpstate(&vcpu->arch.guest_fpu,
4945 					      guest_xsave->region,
4946 					      supported_xcr0, &vcpu->arch.pkru);
4947 }
4948 
4949 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4950 					struct kvm_xcrs *guest_xcrs)
4951 {
4952 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4953 		guest_xcrs->nr_xcrs = 0;
4954 		return;
4955 	}
4956 
4957 	guest_xcrs->nr_xcrs = 1;
4958 	guest_xcrs->flags = 0;
4959 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4960 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4961 }
4962 
4963 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4964 				       struct kvm_xcrs *guest_xcrs)
4965 {
4966 	int i, r = 0;
4967 
4968 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
4969 		return -EINVAL;
4970 
4971 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4972 		return -EINVAL;
4973 
4974 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4975 		/* Only support XCR0 currently */
4976 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4977 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4978 				guest_xcrs->xcrs[i].value);
4979 			break;
4980 		}
4981 	if (r)
4982 		r = -EINVAL;
4983 	return r;
4984 }
4985 
4986 /*
4987  * kvm_set_guest_paused() indicates to the guest kernel that it has been
4988  * stopped by the hypervisor.  This function will be called from the host only.
4989  * EINVAL is returned when the host attempts to set the flag for a guest that
4990  * does not support pv clocks.
4991  */
4992 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4993 {
4994 	if (!vcpu->arch.pv_time_enabled)
4995 		return -EINVAL;
4996 	vcpu->arch.pvclock_set_guest_stopped_request = true;
4997 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4998 	return 0;
4999 }
5000 
5001 static int kvm_arch_tsc_has_attr(struct kvm_vcpu *vcpu,
5002 				 struct kvm_device_attr *attr)
5003 {
5004 	int r;
5005 
5006 	switch (attr->attr) {
5007 	case KVM_VCPU_TSC_OFFSET:
5008 		r = 0;
5009 		break;
5010 	default:
5011 		r = -ENXIO;
5012 	}
5013 
5014 	return r;
5015 }
5016 
5017 static int kvm_arch_tsc_get_attr(struct kvm_vcpu *vcpu,
5018 				 struct kvm_device_attr *attr)
5019 {
5020 	u64 __user *uaddr = (u64 __user *)(unsigned long)attr->addr;
5021 	int r;
5022 
5023 	if ((u64)(unsigned long)uaddr != attr->addr)
5024 		return -EFAULT;
5025 
5026 	switch (attr->attr) {
5027 	case KVM_VCPU_TSC_OFFSET:
5028 		r = -EFAULT;
5029 		if (put_user(vcpu->arch.l1_tsc_offset, uaddr))
5030 			break;
5031 		r = 0;
5032 		break;
5033 	default:
5034 		r = -ENXIO;
5035 	}
5036 
5037 	return r;
5038 }
5039 
5040 static int kvm_arch_tsc_set_attr(struct kvm_vcpu *vcpu,
5041 				 struct kvm_device_attr *attr)
5042 {
5043 	u64 __user *uaddr = (u64 __user *)(unsigned long)attr->addr;
5044 	struct kvm *kvm = vcpu->kvm;
5045 	int r;
5046 
5047 	if ((u64)(unsigned long)uaddr != attr->addr)
5048 		return -EFAULT;
5049 
5050 	switch (attr->attr) {
5051 	case KVM_VCPU_TSC_OFFSET: {
5052 		u64 offset, tsc, ns;
5053 		unsigned long flags;
5054 		bool matched;
5055 
5056 		r = -EFAULT;
5057 		if (get_user(offset, uaddr))
5058 			break;
5059 
5060 		raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
5061 
5062 		matched = (vcpu->arch.virtual_tsc_khz &&
5063 			   kvm->arch.last_tsc_khz == vcpu->arch.virtual_tsc_khz &&
5064 			   kvm->arch.last_tsc_offset == offset);
5065 
5066 		tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio) + offset;
5067 		ns = get_kvmclock_base_ns();
5068 
5069 		__kvm_synchronize_tsc(vcpu, offset, tsc, ns, matched);
5070 		raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
5071 
5072 		r = 0;
5073 		break;
5074 	}
5075 	default:
5076 		r = -ENXIO;
5077 	}
5078 
5079 	return r;
5080 }
5081 
5082 static int kvm_vcpu_ioctl_device_attr(struct kvm_vcpu *vcpu,
5083 				      unsigned int ioctl,
5084 				      void __user *argp)
5085 {
5086 	struct kvm_device_attr attr;
5087 	int r;
5088 
5089 	if (copy_from_user(&attr, argp, sizeof(attr)))
5090 		return -EFAULT;
5091 
5092 	if (attr.group != KVM_VCPU_TSC_CTRL)
5093 		return -ENXIO;
5094 
5095 	switch (ioctl) {
5096 	case KVM_HAS_DEVICE_ATTR:
5097 		r = kvm_arch_tsc_has_attr(vcpu, &attr);
5098 		break;
5099 	case KVM_GET_DEVICE_ATTR:
5100 		r = kvm_arch_tsc_get_attr(vcpu, &attr);
5101 		break;
5102 	case KVM_SET_DEVICE_ATTR:
5103 		r = kvm_arch_tsc_set_attr(vcpu, &attr);
5104 		break;
5105 	}
5106 
5107 	return r;
5108 }
5109 
5110 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
5111 				     struct kvm_enable_cap *cap)
5112 {
5113 	int r;
5114 	uint16_t vmcs_version;
5115 	void __user *user_ptr;
5116 
5117 	if (cap->flags)
5118 		return -EINVAL;
5119 
5120 	switch (cap->cap) {
5121 	case KVM_CAP_HYPERV_SYNIC2:
5122 		if (cap->args[0])
5123 			return -EINVAL;
5124 		fallthrough;
5125 
5126 	case KVM_CAP_HYPERV_SYNIC:
5127 		if (!irqchip_in_kernel(vcpu->kvm))
5128 			return -EINVAL;
5129 		return kvm_hv_activate_synic(vcpu, cap->cap ==
5130 					     KVM_CAP_HYPERV_SYNIC2);
5131 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
5132 		if (!kvm_x86_ops.nested_ops->enable_evmcs)
5133 			return -ENOTTY;
5134 		r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
5135 		if (!r) {
5136 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
5137 			if (copy_to_user(user_ptr, &vmcs_version,
5138 					 sizeof(vmcs_version)))
5139 				r = -EFAULT;
5140 		}
5141 		return r;
5142 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
5143 		if (!kvm_x86_ops.enable_direct_tlbflush)
5144 			return -ENOTTY;
5145 
5146 		return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
5147 
5148 	case KVM_CAP_HYPERV_ENFORCE_CPUID:
5149 		return kvm_hv_set_enforce_cpuid(vcpu, cap->args[0]);
5150 
5151 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
5152 		vcpu->arch.pv_cpuid.enforce = cap->args[0];
5153 		if (vcpu->arch.pv_cpuid.enforce)
5154 			kvm_update_pv_runtime(vcpu);
5155 
5156 		return 0;
5157 	default:
5158 		return -EINVAL;
5159 	}
5160 }
5161 
5162 long kvm_arch_vcpu_ioctl(struct file *filp,
5163 			 unsigned int ioctl, unsigned long arg)
5164 {
5165 	struct kvm_vcpu *vcpu = filp->private_data;
5166 	void __user *argp = (void __user *)arg;
5167 	int r;
5168 	union {
5169 		struct kvm_sregs2 *sregs2;
5170 		struct kvm_lapic_state *lapic;
5171 		struct kvm_xsave *xsave;
5172 		struct kvm_xcrs *xcrs;
5173 		void *buffer;
5174 	} u;
5175 
5176 	vcpu_load(vcpu);
5177 
5178 	u.buffer = NULL;
5179 	switch (ioctl) {
5180 	case KVM_GET_LAPIC: {
5181 		r = -EINVAL;
5182 		if (!lapic_in_kernel(vcpu))
5183 			goto out;
5184 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
5185 				GFP_KERNEL_ACCOUNT);
5186 
5187 		r = -ENOMEM;
5188 		if (!u.lapic)
5189 			goto out;
5190 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
5191 		if (r)
5192 			goto out;
5193 		r = -EFAULT;
5194 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
5195 			goto out;
5196 		r = 0;
5197 		break;
5198 	}
5199 	case KVM_SET_LAPIC: {
5200 		r = -EINVAL;
5201 		if (!lapic_in_kernel(vcpu))
5202 			goto out;
5203 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
5204 		if (IS_ERR(u.lapic)) {
5205 			r = PTR_ERR(u.lapic);
5206 			goto out_nofree;
5207 		}
5208 
5209 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
5210 		break;
5211 	}
5212 	case KVM_INTERRUPT: {
5213 		struct kvm_interrupt irq;
5214 
5215 		r = -EFAULT;
5216 		if (copy_from_user(&irq, argp, sizeof(irq)))
5217 			goto out;
5218 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
5219 		break;
5220 	}
5221 	case KVM_NMI: {
5222 		r = kvm_vcpu_ioctl_nmi(vcpu);
5223 		break;
5224 	}
5225 	case KVM_SMI: {
5226 		r = kvm_vcpu_ioctl_smi(vcpu);
5227 		break;
5228 	}
5229 	case KVM_SET_CPUID: {
5230 		struct kvm_cpuid __user *cpuid_arg = argp;
5231 		struct kvm_cpuid cpuid;
5232 
5233 		/*
5234 		 * KVM does not correctly handle changing guest CPUID after KVM_RUN, as
5235 		 * MAXPHYADDR, GBPAGES support, AMD reserved bit behavior, etc.. aren't
5236 		 * tracked in kvm_mmu_page_role.  As a result, KVM may miss guest page
5237 		 * faults due to reusing SPs/SPTEs.  In practice no sane VMM mucks with
5238 		 * the core vCPU model on the fly, so fail.
5239 		 */
5240 		r = -EINVAL;
5241 		if (vcpu->arch.last_vmentry_cpu != -1)
5242 			goto out;
5243 
5244 		r = -EFAULT;
5245 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5246 			goto out;
5247 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
5248 		break;
5249 	}
5250 	case KVM_SET_CPUID2: {
5251 		struct kvm_cpuid2 __user *cpuid_arg = argp;
5252 		struct kvm_cpuid2 cpuid;
5253 
5254 		/*
5255 		 * KVM_SET_CPUID{,2} after KVM_RUN is forbidded, see the comment in
5256 		 * KVM_SET_CPUID case above.
5257 		 */
5258 		r = -EINVAL;
5259 		if (vcpu->arch.last_vmentry_cpu != -1)
5260 			goto out;
5261 
5262 		r = -EFAULT;
5263 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5264 			goto out;
5265 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
5266 					      cpuid_arg->entries);
5267 		break;
5268 	}
5269 	case KVM_GET_CPUID2: {
5270 		struct kvm_cpuid2 __user *cpuid_arg = argp;
5271 		struct kvm_cpuid2 cpuid;
5272 
5273 		r = -EFAULT;
5274 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5275 			goto out;
5276 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
5277 					      cpuid_arg->entries);
5278 		if (r)
5279 			goto out;
5280 		r = -EFAULT;
5281 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
5282 			goto out;
5283 		r = 0;
5284 		break;
5285 	}
5286 	case KVM_GET_MSRS: {
5287 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
5288 		r = msr_io(vcpu, argp, do_get_msr, 1);
5289 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5290 		break;
5291 	}
5292 	case KVM_SET_MSRS: {
5293 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
5294 		r = msr_io(vcpu, argp, do_set_msr, 0);
5295 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5296 		break;
5297 	}
5298 	case KVM_TPR_ACCESS_REPORTING: {
5299 		struct kvm_tpr_access_ctl tac;
5300 
5301 		r = -EFAULT;
5302 		if (copy_from_user(&tac, argp, sizeof(tac)))
5303 			goto out;
5304 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
5305 		if (r)
5306 			goto out;
5307 		r = -EFAULT;
5308 		if (copy_to_user(argp, &tac, sizeof(tac)))
5309 			goto out;
5310 		r = 0;
5311 		break;
5312 	};
5313 	case KVM_SET_VAPIC_ADDR: {
5314 		struct kvm_vapic_addr va;
5315 		int idx;
5316 
5317 		r = -EINVAL;
5318 		if (!lapic_in_kernel(vcpu))
5319 			goto out;
5320 		r = -EFAULT;
5321 		if (copy_from_user(&va, argp, sizeof(va)))
5322 			goto out;
5323 		idx = srcu_read_lock(&vcpu->kvm->srcu);
5324 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
5325 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5326 		break;
5327 	}
5328 	case KVM_X86_SETUP_MCE: {
5329 		u64 mcg_cap;
5330 
5331 		r = -EFAULT;
5332 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
5333 			goto out;
5334 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
5335 		break;
5336 	}
5337 	case KVM_X86_SET_MCE: {
5338 		struct kvm_x86_mce mce;
5339 
5340 		r = -EFAULT;
5341 		if (copy_from_user(&mce, argp, sizeof(mce)))
5342 			goto out;
5343 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
5344 		break;
5345 	}
5346 	case KVM_GET_VCPU_EVENTS: {
5347 		struct kvm_vcpu_events events;
5348 
5349 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
5350 
5351 		r = -EFAULT;
5352 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
5353 			break;
5354 		r = 0;
5355 		break;
5356 	}
5357 	case KVM_SET_VCPU_EVENTS: {
5358 		struct kvm_vcpu_events events;
5359 
5360 		r = -EFAULT;
5361 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5362 			break;
5363 
5364 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5365 		break;
5366 	}
5367 	case KVM_GET_DEBUGREGS: {
5368 		struct kvm_debugregs dbgregs;
5369 
5370 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5371 
5372 		r = -EFAULT;
5373 		if (copy_to_user(argp, &dbgregs,
5374 				 sizeof(struct kvm_debugregs)))
5375 			break;
5376 		r = 0;
5377 		break;
5378 	}
5379 	case KVM_SET_DEBUGREGS: {
5380 		struct kvm_debugregs dbgregs;
5381 
5382 		r = -EFAULT;
5383 		if (copy_from_user(&dbgregs, argp,
5384 				   sizeof(struct kvm_debugregs)))
5385 			break;
5386 
5387 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5388 		break;
5389 	}
5390 	case KVM_GET_XSAVE: {
5391 		r = -EINVAL;
5392 		if (vcpu->arch.guest_fpu.uabi_size > sizeof(struct kvm_xsave))
5393 			break;
5394 
5395 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5396 		r = -ENOMEM;
5397 		if (!u.xsave)
5398 			break;
5399 
5400 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5401 
5402 		r = -EFAULT;
5403 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5404 			break;
5405 		r = 0;
5406 		break;
5407 	}
5408 	case KVM_SET_XSAVE: {
5409 		int size = vcpu->arch.guest_fpu.uabi_size;
5410 
5411 		u.xsave = memdup_user(argp, size);
5412 		if (IS_ERR(u.xsave)) {
5413 			r = PTR_ERR(u.xsave);
5414 			goto out_nofree;
5415 		}
5416 
5417 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5418 		break;
5419 	}
5420 
5421 	case KVM_GET_XSAVE2: {
5422 		int size = vcpu->arch.guest_fpu.uabi_size;
5423 
5424 		u.xsave = kzalloc(size, GFP_KERNEL_ACCOUNT);
5425 		r = -ENOMEM;
5426 		if (!u.xsave)
5427 			break;
5428 
5429 		kvm_vcpu_ioctl_x86_get_xsave2(vcpu, u.buffer, size);
5430 
5431 		r = -EFAULT;
5432 		if (copy_to_user(argp, u.xsave, size))
5433 			break;
5434 
5435 		r = 0;
5436 		break;
5437 	}
5438 
5439 	case KVM_GET_XCRS: {
5440 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5441 		r = -ENOMEM;
5442 		if (!u.xcrs)
5443 			break;
5444 
5445 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5446 
5447 		r = -EFAULT;
5448 		if (copy_to_user(argp, u.xcrs,
5449 				 sizeof(struct kvm_xcrs)))
5450 			break;
5451 		r = 0;
5452 		break;
5453 	}
5454 	case KVM_SET_XCRS: {
5455 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5456 		if (IS_ERR(u.xcrs)) {
5457 			r = PTR_ERR(u.xcrs);
5458 			goto out_nofree;
5459 		}
5460 
5461 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5462 		break;
5463 	}
5464 	case KVM_SET_TSC_KHZ: {
5465 		u32 user_tsc_khz;
5466 
5467 		r = -EINVAL;
5468 		user_tsc_khz = (u32)arg;
5469 
5470 		if (kvm_has_tsc_control &&
5471 		    user_tsc_khz >= kvm_max_guest_tsc_khz)
5472 			goto out;
5473 
5474 		if (user_tsc_khz == 0)
5475 			user_tsc_khz = tsc_khz;
5476 
5477 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5478 			r = 0;
5479 
5480 		goto out;
5481 	}
5482 	case KVM_GET_TSC_KHZ: {
5483 		r = vcpu->arch.virtual_tsc_khz;
5484 		goto out;
5485 	}
5486 	case KVM_KVMCLOCK_CTRL: {
5487 		r = kvm_set_guest_paused(vcpu);
5488 		goto out;
5489 	}
5490 	case KVM_ENABLE_CAP: {
5491 		struct kvm_enable_cap cap;
5492 
5493 		r = -EFAULT;
5494 		if (copy_from_user(&cap, argp, sizeof(cap)))
5495 			goto out;
5496 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5497 		break;
5498 	}
5499 	case KVM_GET_NESTED_STATE: {
5500 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5501 		u32 user_data_size;
5502 
5503 		r = -EINVAL;
5504 		if (!kvm_x86_ops.nested_ops->get_state)
5505 			break;
5506 
5507 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5508 		r = -EFAULT;
5509 		if (get_user(user_data_size, &user_kvm_nested_state->size))
5510 			break;
5511 
5512 		r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5513 						     user_data_size);
5514 		if (r < 0)
5515 			break;
5516 
5517 		if (r > user_data_size) {
5518 			if (put_user(r, &user_kvm_nested_state->size))
5519 				r = -EFAULT;
5520 			else
5521 				r = -E2BIG;
5522 			break;
5523 		}
5524 
5525 		r = 0;
5526 		break;
5527 	}
5528 	case KVM_SET_NESTED_STATE: {
5529 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5530 		struct kvm_nested_state kvm_state;
5531 		int idx;
5532 
5533 		r = -EINVAL;
5534 		if (!kvm_x86_ops.nested_ops->set_state)
5535 			break;
5536 
5537 		r = -EFAULT;
5538 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5539 			break;
5540 
5541 		r = -EINVAL;
5542 		if (kvm_state.size < sizeof(kvm_state))
5543 			break;
5544 
5545 		if (kvm_state.flags &
5546 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5547 		      | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5548 		      | KVM_STATE_NESTED_GIF_SET))
5549 			break;
5550 
5551 		/* nested_run_pending implies guest_mode.  */
5552 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5553 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5554 			break;
5555 
5556 		idx = srcu_read_lock(&vcpu->kvm->srcu);
5557 		r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5558 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5559 		break;
5560 	}
5561 	case KVM_GET_SUPPORTED_HV_CPUID:
5562 		r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5563 		break;
5564 #ifdef CONFIG_KVM_XEN
5565 	case KVM_XEN_VCPU_GET_ATTR: {
5566 		struct kvm_xen_vcpu_attr xva;
5567 
5568 		r = -EFAULT;
5569 		if (copy_from_user(&xva, argp, sizeof(xva)))
5570 			goto out;
5571 		r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5572 		if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5573 			r = -EFAULT;
5574 		break;
5575 	}
5576 	case KVM_XEN_VCPU_SET_ATTR: {
5577 		struct kvm_xen_vcpu_attr xva;
5578 
5579 		r = -EFAULT;
5580 		if (copy_from_user(&xva, argp, sizeof(xva)))
5581 			goto out;
5582 		r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5583 		break;
5584 	}
5585 #endif
5586 	case KVM_GET_SREGS2: {
5587 		u.sregs2 = kzalloc(sizeof(struct kvm_sregs2), GFP_KERNEL);
5588 		r = -ENOMEM;
5589 		if (!u.sregs2)
5590 			goto out;
5591 		__get_sregs2(vcpu, u.sregs2);
5592 		r = -EFAULT;
5593 		if (copy_to_user(argp, u.sregs2, sizeof(struct kvm_sregs2)))
5594 			goto out;
5595 		r = 0;
5596 		break;
5597 	}
5598 	case KVM_SET_SREGS2: {
5599 		u.sregs2 = memdup_user(argp, sizeof(struct kvm_sregs2));
5600 		if (IS_ERR(u.sregs2)) {
5601 			r = PTR_ERR(u.sregs2);
5602 			u.sregs2 = NULL;
5603 			goto out;
5604 		}
5605 		r = __set_sregs2(vcpu, u.sregs2);
5606 		break;
5607 	}
5608 	case KVM_HAS_DEVICE_ATTR:
5609 	case KVM_GET_DEVICE_ATTR:
5610 	case KVM_SET_DEVICE_ATTR:
5611 		r = kvm_vcpu_ioctl_device_attr(vcpu, ioctl, argp);
5612 		break;
5613 	default:
5614 		r = -EINVAL;
5615 	}
5616 out:
5617 	kfree(u.buffer);
5618 out_nofree:
5619 	vcpu_put(vcpu);
5620 	return r;
5621 }
5622 
5623 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5624 {
5625 	return VM_FAULT_SIGBUS;
5626 }
5627 
5628 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5629 {
5630 	int ret;
5631 
5632 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
5633 		return -EINVAL;
5634 	ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5635 	return ret;
5636 }
5637 
5638 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5639 					      u64 ident_addr)
5640 {
5641 	return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5642 }
5643 
5644 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5645 					 unsigned long kvm_nr_mmu_pages)
5646 {
5647 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5648 		return -EINVAL;
5649 
5650 	mutex_lock(&kvm->slots_lock);
5651 
5652 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5653 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5654 
5655 	mutex_unlock(&kvm->slots_lock);
5656 	return 0;
5657 }
5658 
5659 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5660 {
5661 	return kvm->arch.n_max_mmu_pages;
5662 }
5663 
5664 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5665 {
5666 	struct kvm_pic *pic = kvm->arch.vpic;
5667 	int r;
5668 
5669 	r = 0;
5670 	switch (chip->chip_id) {
5671 	case KVM_IRQCHIP_PIC_MASTER:
5672 		memcpy(&chip->chip.pic, &pic->pics[0],
5673 			sizeof(struct kvm_pic_state));
5674 		break;
5675 	case KVM_IRQCHIP_PIC_SLAVE:
5676 		memcpy(&chip->chip.pic, &pic->pics[1],
5677 			sizeof(struct kvm_pic_state));
5678 		break;
5679 	case KVM_IRQCHIP_IOAPIC:
5680 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
5681 		break;
5682 	default:
5683 		r = -EINVAL;
5684 		break;
5685 	}
5686 	return r;
5687 }
5688 
5689 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5690 {
5691 	struct kvm_pic *pic = kvm->arch.vpic;
5692 	int r;
5693 
5694 	r = 0;
5695 	switch (chip->chip_id) {
5696 	case KVM_IRQCHIP_PIC_MASTER:
5697 		spin_lock(&pic->lock);
5698 		memcpy(&pic->pics[0], &chip->chip.pic,
5699 			sizeof(struct kvm_pic_state));
5700 		spin_unlock(&pic->lock);
5701 		break;
5702 	case KVM_IRQCHIP_PIC_SLAVE:
5703 		spin_lock(&pic->lock);
5704 		memcpy(&pic->pics[1], &chip->chip.pic,
5705 			sizeof(struct kvm_pic_state));
5706 		spin_unlock(&pic->lock);
5707 		break;
5708 	case KVM_IRQCHIP_IOAPIC:
5709 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
5710 		break;
5711 	default:
5712 		r = -EINVAL;
5713 		break;
5714 	}
5715 	kvm_pic_update_irq(pic);
5716 	return r;
5717 }
5718 
5719 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5720 {
5721 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5722 
5723 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5724 
5725 	mutex_lock(&kps->lock);
5726 	memcpy(ps, &kps->channels, sizeof(*ps));
5727 	mutex_unlock(&kps->lock);
5728 	return 0;
5729 }
5730 
5731 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5732 {
5733 	int i;
5734 	struct kvm_pit *pit = kvm->arch.vpit;
5735 
5736 	mutex_lock(&pit->pit_state.lock);
5737 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5738 	for (i = 0; i < 3; i++)
5739 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5740 	mutex_unlock(&pit->pit_state.lock);
5741 	return 0;
5742 }
5743 
5744 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5745 {
5746 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
5747 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5748 		sizeof(ps->channels));
5749 	ps->flags = kvm->arch.vpit->pit_state.flags;
5750 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5751 	memset(&ps->reserved, 0, sizeof(ps->reserved));
5752 	return 0;
5753 }
5754 
5755 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5756 {
5757 	int start = 0;
5758 	int i;
5759 	u32 prev_legacy, cur_legacy;
5760 	struct kvm_pit *pit = kvm->arch.vpit;
5761 
5762 	mutex_lock(&pit->pit_state.lock);
5763 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5764 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5765 	if (!prev_legacy && cur_legacy)
5766 		start = 1;
5767 	memcpy(&pit->pit_state.channels, &ps->channels,
5768 	       sizeof(pit->pit_state.channels));
5769 	pit->pit_state.flags = ps->flags;
5770 	for (i = 0; i < 3; i++)
5771 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5772 				   start && i == 0);
5773 	mutex_unlock(&pit->pit_state.lock);
5774 	return 0;
5775 }
5776 
5777 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5778 				 struct kvm_reinject_control *control)
5779 {
5780 	struct kvm_pit *pit = kvm->arch.vpit;
5781 
5782 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
5783 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5784 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
5785 	 */
5786 	mutex_lock(&pit->pit_state.lock);
5787 	kvm_pit_set_reinject(pit, control->pit_reinject);
5788 	mutex_unlock(&pit->pit_state.lock);
5789 
5790 	return 0;
5791 }
5792 
5793 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5794 {
5795 
5796 	/*
5797 	 * Flush all CPUs' dirty log buffers to the  dirty_bitmap.  Called
5798 	 * before reporting dirty_bitmap to userspace.  KVM flushes the buffers
5799 	 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5800 	 * VM-Exit.
5801 	 */
5802 	struct kvm_vcpu *vcpu;
5803 	unsigned long i;
5804 
5805 	kvm_for_each_vcpu(i, vcpu, kvm)
5806 		kvm_vcpu_kick(vcpu);
5807 }
5808 
5809 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5810 			bool line_status)
5811 {
5812 	if (!irqchip_in_kernel(kvm))
5813 		return -ENXIO;
5814 
5815 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5816 					irq_event->irq, irq_event->level,
5817 					line_status);
5818 	return 0;
5819 }
5820 
5821 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5822 			    struct kvm_enable_cap *cap)
5823 {
5824 	int r;
5825 
5826 	if (cap->flags)
5827 		return -EINVAL;
5828 
5829 	switch (cap->cap) {
5830 	case KVM_CAP_DISABLE_QUIRKS:
5831 		kvm->arch.disabled_quirks = cap->args[0];
5832 		r = 0;
5833 		break;
5834 	case KVM_CAP_SPLIT_IRQCHIP: {
5835 		mutex_lock(&kvm->lock);
5836 		r = -EINVAL;
5837 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5838 			goto split_irqchip_unlock;
5839 		r = -EEXIST;
5840 		if (irqchip_in_kernel(kvm))
5841 			goto split_irqchip_unlock;
5842 		if (kvm->created_vcpus)
5843 			goto split_irqchip_unlock;
5844 		r = kvm_setup_empty_irq_routing(kvm);
5845 		if (r)
5846 			goto split_irqchip_unlock;
5847 		/* Pairs with irqchip_in_kernel. */
5848 		smp_wmb();
5849 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5850 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5851 		kvm_request_apicv_update(kvm, true, APICV_INHIBIT_REASON_ABSENT);
5852 		r = 0;
5853 split_irqchip_unlock:
5854 		mutex_unlock(&kvm->lock);
5855 		break;
5856 	}
5857 	case KVM_CAP_X2APIC_API:
5858 		r = -EINVAL;
5859 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5860 			break;
5861 
5862 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5863 			kvm->arch.x2apic_format = true;
5864 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5865 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
5866 
5867 		r = 0;
5868 		break;
5869 	case KVM_CAP_X86_DISABLE_EXITS:
5870 		r = -EINVAL;
5871 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5872 			break;
5873 
5874 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5875 			kvm_can_mwait_in_guest())
5876 			kvm->arch.mwait_in_guest = true;
5877 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5878 			kvm->arch.hlt_in_guest = true;
5879 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5880 			kvm->arch.pause_in_guest = true;
5881 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5882 			kvm->arch.cstate_in_guest = true;
5883 		r = 0;
5884 		break;
5885 	case KVM_CAP_MSR_PLATFORM_INFO:
5886 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5887 		r = 0;
5888 		break;
5889 	case KVM_CAP_EXCEPTION_PAYLOAD:
5890 		kvm->arch.exception_payload_enabled = cap->args[0];
5891 		r = 0;
5892 		break;
5893 	case KVM_CAP_X86_USER_SPACE_MSR:
5894 		kvm->arch.user_space_msr_mask = cap->args[0];
5895 		r = 0;
5896 		break;
5897 	case KVM_CAP_X86_BUS_LOCK_EXIT:
5898 		r = -EINVAL;
5899 		if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5900 			break;
5901 
5902 		if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5903 		    (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5904 			break;
5905 
5906 		if (kvm_has_bus_lock_exit &&
5907 		    cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5908 			kvm->arch.bus_lock_detection_enabled = true;
5909 		r = 0;
5910 		break;
5911 #ifdef CONFIG_X86_SGX_KVM
5912 	case KVM_CAP_SGX_ATTRIBUTE: {
5913 		unsigned long allowed_attributes = 0;
5914 
5915 		r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5916 		if (r)
5917 			break;
5918 
5919 		/* KVM only supports the PROVISIONKEY privileged attribute. */
5920 		if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5921 		    !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5922 			kvm->arch.sgx_provisioning_allowed = true;
5923 		else
5924 			r = -EINVAL;
5925 		break;
5926 	}
5927 #endif
5928 	case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5929 		r = -EINVAL;
5930 		if (kvm_x86_ops.vm_copy_enc_context_from)
5931 			r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5932 		return r;
5933 	case KVM_CAP_VM_MOVE_ENC_CONTEXT_FROM:
5934 		r = -EINVAL;
5935 		if (kvm_x86_ops.vm_move_enc_context_from)
5936 			r = kvm_x86_ops.vm_move_enc_context_from(
5937 				kvm, cap->args[0]);
5938 		return r;
5939 	case KVM_CAP_EXIT_HYPERCALL:
5940 		if (cap->args[0] & ~KVM_EXIT_HYPERCALL_VALID_MASK) {
5941 			r = -EINVAL;
5942 			break;
5943 		}
5944 		kvm->arch.hypercall_exit_enabled = cap->args[0];
5945 		r = 0;
5946 		break;
5947 	case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
5948 		r = -EINVAL;
5949 		if (cap->args[0] & ~1)
5950 			break;
5951 		kvm->arch.exit_on_emulation_error = cap->args[0];
5952 		r = 0;
5953 		break;
5954 	default:
5955 		r = -EINVAL;
5956 		break;
5957 	}
5958 	return r;
5959 }
5960 
5961 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5962 {
5963 	struct kvm_x86_msr_filter *msr_filter;
5964 
5965 	msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5966 	if (!msr_filter)
5967 		return NULL;
5968 
5969 	msr_filter->default_allow = default_allow;
5970 	return msr_filter;
5971 }
5972 
5973 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5974 {
5975 	u32 i;
5976 
5977 	if (!msr_filter)
5978 		return;
5979 
5980 	for (i = 0; i < msr_filter->count; i++)
5981 		kfree(msr_filter->ranges[i].bitmap);
5982 
5983 	kfree(msr_filter);
5984 }
5985 
5986 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5987 			      struct kvm_msr_filter_range *user_range)
5988 {
5989 	unsigned long *bitmap = NULL;
5990 	size_t bitmap_size;
5991 
5992 	if (!user_range->nmsrs)
5993 		return 0;
5994 
5995 	if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
5996 		return -EINVAL;
5997 
5998 	if (!user_range->flags)
5999 		return -EINVAL;
6000 
6001 	bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
6002 	if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
6003 		return -EINVAL;
6004 
6005 	bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
6006 	if (IS_ERR(bitmap))
6007 		return PTR_ERR(bitmap);
6008 
6009 	msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
6010 		.flags = user_range->flags,
6011 		.base = user_range->base,
6012 		.nmsrs = user_range->nmsrs,
6013 		.bitmap = bitmap,
6014 	};
6015 
6016 	msr_filter->count++;
6017 	return 0;
6018 }
6019 
6020 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
6021 {
6022 	struct kvm_msr_filter __user *user_msr_filter = argp;
6023 	struct kvm_x86_msr_filter *new_filter, *old_filter;
6024 	struct kvm_msr_filter filter;
6025 	bool default_allow;
6026 	bool empty = true;
6027 	int r = 0;
6028 	u32 i;
6029 
6030 	if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
6031 		return -EFAULT;
6032 
6033 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
6034 		empty &= !filter.ranges[i].nmsrs;
6035 
6036 	default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
6037 	if (empty && !default_allow)
6038 		return -EINVAL;
6039 
6040 	new_filter = kvm_alloc_msr_filter(default_allow);
6041 	if (!new_filter)
6042 		return -ENOMEM;
6043 
6044 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
6045 		r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
6046 		if (r) {
6047 			kvm_free_msr_filter(new_filter);
6048 			return r;
6049 		}
6050 	}
6051 
6052 	mutex_lock(&kvm->lock);
6053 
6054 	/* The per-VM filter is protected by kvm->lock... */
6055 	old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
6056 
6057 	rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
6058 	synchronize_srcu(&kvm->srcu);
6059 
6060 	kvm_free_msr_filter(old_filter);
6061 
6062 	kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
6063 	mutex_unlock(&kvm->lock);
6064 
6065 	return 0;
6066 }
6067 
6068 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
6069 static int kvm_arch_suspend_notifier(struct kvm *kvm)
6070 {
6071 	struct kvm_vcpu *vcpu;
6072 	unsigned long i;
6073 	int ret = 0;
6074 
6075 	mutex_lock(&kvm->lock);
6076 	kvm_for_each_vcpu(i, vcpu, kvm) {
6077 		if (!vcpu->arch.pv_time_enabled)
6078 			continue;
6079 
6080 		ret = kvm_set_guest_paused(vcpu);
6081 		if (ret) {
6082 			kvm_err("Failed to pause guest VCPU%d: %d\n",
6083 				vcpu->vcpu_id, ret);
6084 			break;
6085 		}
6086 	}
6087 	mutex_unlock(&kvm->lock);
6088 
6089 	return ret ? NOTIFY_BAD : NOTIFY_DONE;
6090 }
6091 
6092 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state)
6093 {
6094 	switch (state) {
6095 	case PM_HIBERNATION_PREPARE:
6096 	case PM_SUSPEND_PREPARE:
6097 		return kvm_arch_suspend_notifier(kvm);
6098 	}
6099 
6100 	return NOTIFY_DONE;
6101 }
6102 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
6103 
6104 static int kvm_vm_ioctl_get_clock(struct kvm *kvm, void __user *argp)
6105 {
6106 	struct kvm_clock_data data = { 0 };
6107 
6108 	get_kvmclock(kvm, &data);
6109 	if (copy_to_user(argp, &data, sizeof(data)))
6110 		return -EFAULT;
6111 
6112 	return 0;
6113 }
6114 
6115 static int kvm_vm_ioctl_set_clock(struct kvm *kvm, void __user *argp)
6116 {
6117 	struct kvm_arch *ka = &kvm->arch;
6118 	struct kvm_clock_data data;
6119 	u64 now_raw_ns;
6120 
6121 	if (copy_from_user(&data, argp, sizeof(data)))
6122 		return -EFAULT;
6123 
6124 	/*
6125 	 * Only KVM_CLOCK_REALTIME is used, but allow passing the
6126 	 * result of KVM_GET_CLOCK back to KVM_SET_CLOCK.
6127 	 */
6128 	if (data.flags & ~KVM_CLOCK_VALID_FLAGS)
6129 		return -EINVAL;
6130 
6131 	kvm_hv_invalidate_tsc_page(kvm);
6132 	kvm_start_pvclock_update(kvm);
6133 	pvclock_update_vm_gtod_copy(kvm);
6134 
6135 	/*
6136 	 * This pairs with kvm_guest_time_update(): when masterclock is
6137 	 * in use, we use master_kernel_ns + kvmclock_offset to set
6138 	 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
6139 	 * is slightly ahead) here we risk going negative on unsigned
6140 	 * 'system_time' when 'data.clock' is very small.
6141 	 */
6142 	if (data.flags & KVM_CLOCK_REALTIME) {
6143 		u64 now_real_ns = ktime_get_real_ns();
6144 
6145 		/*
6146 		 * Avoid stepping the kvmclock backwards.
6147 		 */
6148 		if (now_real_ns > data.realtime)
6149 			data.clock += now_real_ns - data.realtime;
6150 	}
6151 
6152 	if (ka->use_master_clock)
6153 		now_raw_ns = ka->master_kernel_ns;
6154 	else
6155 		now_raw_ns = get_kvmclock_base_ns();
6156 	ka->kvmclock_offset = data.clock - now_raw_ns;
6157 	kvm_end_pvclock_update(kvm);
6158 	return 0;
6159 }
6160 
6161 long kvm_arch_vm_ioctl(struct file *filp,
6162 		       unsigned int ioctl, unsigned long arg)
6163 {
6164 	struct kvm *kvm = filp->private_data;
6165 	void __user *argp = (void __user *)arg;
6166 	int r = -ENOTTY;
6167 	/*
6168 	 * This union makes it completely explicit to gcc-3.x
6169 	 * that these two variables' stack usage should be
6170 	 * combined, not added together.
6171 	 */
6172 	union {
6173 		struct kvm_pit_state ps;
6174 		struct kvm_pit_state2 ps2;
6175 		struct kvm_pit_config pit_config;
6176 	} u;
6177 
6178 	switch (ioctl) {
6179 	case KVM_SET_TSS_ADDR:
6180 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
6181 		break;
6182 	case KVM_SET_IDENTITY_MAP_ADDR: {
6183 		u64 ident_addr;
6184 
6185 		mutex_lock(&kvm->lock);
6186 		r = -EINVAL;
6187 		if (kvm->created_vcpus)
6188 			goto set_identity_unlock;
6189 		r = -EFAULT;
6190 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
6191 			goto set_identity_unlock;
6192 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
6193 set_identity_unlock:
6194 		mutex_unlock(&kvm->lock);
6195 		break;
6196 	}
6197 	case KVM_SET_NR_MMU_PAGES:
6198 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
6199 		break;
6200 	case KVM_GET_NR_MMU_PAGES:
6201 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
6202 		break;
6203 	case KVM_CREATE_IRQCHIP: {
6204 		mutex_lock(&kvm->lock);
6205 
6206 		r = -EEXIST;
6207 		if (irqchip_in_kernel(kvm))
6208 			goto create_irqchip_unlock;
6209 
6210 		r = -EINVAL;
6211 		if (kvm->created_vcpus)
6212 			goto create_irqchip_unlock;
6213 
6214 		r = kvm_pic_init(kvm);
6215 		if (r)
6216 			goto create_irqchip_unlock;
6217 
6218 		r = kvm_ioapic_init(kvm);
6219 		if (r) {
6220 			kvm_pic_destroy(kvm);
6221 			goto create_irqchip_unlock;
6222 		}
6223 
6224 		r = kvm_setup_default_irq_routing(kvm);
6225 		if (r) {
6226 			kvm_ioapic_destroy(kvm);
6227 			kvm_pic_destroy(kvm);
6228 			goto create_irqchip_unlock;
6229 		}
6230 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
6231 		smp_wmb();
6232 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
6233 		kvm_request_apicv_update(kvm, true, APICV_INHIBIT_REASON_ABSENT);
6234 	create_irqchip_unlock:
6235 		mutex_unlock(&kvm->lock);
6236 		break;
6237 	}
6238 	case KVM_CREATE_PIT:
6239 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
6240 		goto create_pit;
6241 	case KVM_CREATE_PIT2:
6242 		r = -EFAULT;
6243 		if (copy_from_user(&u.pit_config, argp,
6244 				   sizeof(struct kvm_pit_config)))
6245 			goto out;
6246 	create_pit:
6247 		mutex_lock(&kvm->lock);
6248 		r = -EEXIST;
6249 		if (kvm->arch.vpit)
6250 			goto create_pit_unlock;
6251 		r = -ENOMEM;
6252 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
6253 		if (kvm->arch.vpit)
6254 			r = 0;
6255 	create_pit_unlock:
6256 		mutex_unlock(&kvm->lock);
6257 		break;
6258 	case KVM_GET_IRQCHIP: {
6259 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6260 		struct kvm_irqchip *chip;
6261 
6262 		chip = memdup_user(argp, sizeof(*chip));
6263 		if (IS_ERR(chip)) {
6264 			r = PTR_ERR(chip);
6265 			goto out;
6266 		}
6267 
6268 		r = -ENXIO;
6269 		if (!irqchip_kernel(kvm))
6270 			goto get_irqchip_out;
6271 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
6272 		if (r)
6273 			goto get_irqchip_out;
6274 		r = -EFAULT;
6275 		if (copy_to_user(argp, chip, sizeof(*chip)))
6276 			goto get_irqchip_out;
6277 		r = 0;
6278 	get_irqchip_out:
6279 		kfree(chip);
6280 		break;
6281 	}
6282 	case KVM_SET_IRQCHIP: {
6283 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6284 		struct kvm_irqchip *chip;
6285 
6286 		chip = memdup_user(argp, sizeof(*chip));
6287 		if (IS_ERR(chip)) {
6288 			r = PTR_ERR(chip);
6289 			goto out;
6290 		}
6291 
6292 		r = -ENXIO;
6293 		if (!irqchip_kernel(kvm))
6294 			goto set_irqchip_out;
6295 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
6296 	set_irqchip_out:
6297 		kfree(chip);
6298 		break;
6299 	}
6300 	case KVM_GET_PIT: {
6301 		r = -EFAULT;
6302 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
6303 			goto out;
6304 		r = -ENXIO;
6305 		if (!kvm->arch.vpit)
6306 			goto out;
6307 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
6308 		if (r)
6309 			goto out;
6310 		r = -EFAULT;
6311 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
6312 			goto out;
6313 		r = 0;
6314 		break;
6315 	}
6316 	case KVM_SET_PIT: {
6317 		r = -EFAULT;
6318 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
6319 			goto out;
6320 		mutex_lock(&kvm->lock);
6321 		r = -ENXIO;
6322 		if (!kvm->arch.vpit)
6323 			goto set_pit_out;
6324 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
6325 set_pit_out:
6326 		mutex_unlock(&kvm->lock);
6327 		break;
6328 	}
6329 	case KVM_GET_PIT2: {
6330 		r = -ENXIO;
6331 		if (!kvm->arch.vpit)
6332 			goto out;
6333 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
6334 		if (r)
6335 			goto out;
6336 		r = -EFAULT;
6337 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
6338 			goto out;
6339 		r = 0;
6340 		break;
6341 	}
6342 	case KVM_SET_PIT2: {
6343 		r = -EFAULT;
6344 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
6345 			goto out;
6346 		mutex_lock(&kvm->lock);
6347 		r = -ENXIO;
6348 		if (!kvm->arch.vpit)
6349 			goto set_pit2_out;
6350 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
6351 set_pit2_out:
6352 		mutex_unlock(&kvm->lock);
6353 		break;
6354 	}
6355 	case KVM_REINJECT_CONTROL: {
6356 		struct kvm_reinject_control control;
6357 		r =  -EFAULT;
6358 		if (copy_from_user(&control, argp, sizeof(control)))
6359 			goto out;
6360 		r = -ENXIO;
6361 		if (!kvm->arch.vpit)
6362 			goto out;
6363 		r = kvm_vm_ioctl_reinject(kvm, &control);
6364 		break;
6365 	}
6366 	case KVM_SET_BOOT_CPU_ID:
6367 		r = 0;
6368 		mutex_lock(&kvm->lock);
6369 		if (kvm->created_vcpus)
6370 			r = -EBUSY;
6371 		else
6372 			kvm->arch.bsp_vcpu_id = arg;
6373 		mutex_unlock(&kvm->lock);
6374 		break;
6375 #ifdef CONFIG_KVM_XEN
6376 	case KVM_XEN_HVM_CONFIG: {
6377 		struct kvm_xen_hvm_config xhc;
6378 		r = -EFAULT;
6379 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
6380 			goto out;
6381 		r = kvm_xen_hvm_config(kvm, &xhc);
6382 		break;
6383 	}
6384 	case KVM_XEN_HVM_GET_ATTR: {
6385 		struct kvm_xen_hvm_attr xha;
6386 
6387 		r = -EFAULT;
6388 		if (copy_from_user(&xha, argp, sizeof(xha)))
6389 			goto out;
6390 		r = kvm_xen_hvm_get_attr(kvm, &xha);
6391 		if (!r && copy_to_user(argp, &xha, sizeof(xha)))
6392 			r = -EFAULT;
6393 		break;
6394 	}
6395 	case KVM_XEN_HVM_SET_ATTR: {
6396 		struct kvm_xen_hvm_attr xha;
6397 
6398 		r = -EFAULT;
6399 		if (copy_from_user(&xha, argp, sizeof(xha)))
6400 			goto out;
6401 		r = kvm_xen_hvm_set_attr(kvm, &xha);
6402 		break;
6403 	}
6404 #endif
6405 	case KVM_SET_CLOCK:
6406 		r = kvm_vm_ioctl_set_clock(kvm, argp);
6407 		break;
6408 	case KVM_GET_CLOCK:
6409 		r = kvm_vm_ioctl_get_clock(kvm, argp);
6410 		break;
6411 	case KVM_MEMORY_ENCRYPT_OP: {
6412 		r = -ENOTTY;
6413 		if (kvm_x86_ops.mem_enc_op)
6414 			r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
6415 		break;
6416 	}
6417 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
6418 		struct kvm_enc_region region;
6419 
6420 		r = -EFAULT;
6421 		if (copy_from_user(&region, argp, sizeof(region)))
6422 			goto out;
6423 
6424 		r = -ENOTTY;
6425 		if (kvm_x86_ops.mem_enc_reg_region)
6426 			r = static_call(kvm_x86_mem_enc_reg_region)(kvm, &region);
6427 		break;
6428 	}
6429 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
6430 		struct kvm_enc_region region;
6431 
6432 		r = -EFAULT;
6433 		if (copy_from_user(&region, argp, sizeof(region)))
6434 			goto out;
6435 
6436 		r = -ENOTTY;
6437 		if (kvm_x86_ops.mem_enc_unreg_region)
6438 			r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, &region);
6439 		break;
6440 	}
6441 	case KVM_HYPERV_EVENTFD: {
6442 		struct kvm_hyperv_eventfd hvevfd;
6443 
6444 		r = -EFAULT;
6445 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
6446 			goto out;
6447 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
6448 		break;
6449 	}
6450 	case KVM_SET_PMU_EVENT_FILTER:
6451 		r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
6452 		break;
6453 	case KVM_X86_SET_MSR_FILTER:
6454 		r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
6455 		break;
6456 	default:
6457 		r = -ENOTTY;
6458 	}
6459 out:
6460 	return r;
6461 }
6462 
6463 static void kvm_init_msr_list(void)
6464 {
6465 	struct x86_pmu_capability x86_pmu;
6466 	u32 dummy[2];
6467 	unsigned i;
6468 
6469 	BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
6470 			 "Please update the fixed PMCs in msrs_to_saved_all[]");
6471 
6472 	perf_get_x86_pmu_capability(&x86_pmu);
6473 
6474 	num_msrs_to_save = 0;
6475 	num_emulated_msrs = 0;
6476 	num_msr_based_features = 0;
6477 
6478 	for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
6479 		if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
6480 			continue;
6481 
6482 		/*
6483 		 * Even MSRs that are valid in the host may not be exposed
6484 		 * to the guests in some cases.
6485 		 */
6486 		switch (msrs_to_save_all[i]) {
6487 		case MSR_IA32_BNDCFGS:
6488 			if (!kvm_mpx_supported())
6489 				continue;
6490 			break;
6491 		case MSR_TSC_AUX:
6492 			if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
6493 			    !kvm_cpu_cap_has(X86_FEATURE_RDPID))
6494 				continue;
6495 			break;
6496 		case MSR_IA32_UMWAIT_CONTROL:
6497 			if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
6498 				continue;
6499 			break;
6500 		case MSR_IA32_RTIT_CTL:
6501 		case MSR_IA32_RTIT_STATUS:
6502 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
6503 				continue;
6504 			break;
6505 		case MSR_IA32_RTIT_CR3_MATCH:
6506 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6507 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
6508 				continue;
6509 			break;
6510 		case MSR_IA32_RTIT_OUTPUT_BASE:
6511 		case MSR_IA32_RTIT_OUTPUT_MASK:
6512 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6513 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
6514 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
6515 				continue;
6516 			break;
6517 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
6518 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6519 				msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
6520 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
6521 				continue;
6522 			break;
6523 		case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
6524 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
6525 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6526 				continue;
6527 			break;
6528 		case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
6529 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
6530 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6531 				continue;
6532 			break;
6533 		case MSR_IA32_XFD:
6534 		case MSR_IA32_XFD_ERR:
6535 			if (!kvm_cpu_cap_has(X86_FEATURE_XFD))
6536 				continue;
6537 			break;
6538 		default:
6539 			break;
6540 		}
6541 
6542 		msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
6543 	}
6544 
6545 	for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
6546 		if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
6547 			continue;
6548 
6549 		emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
6550 	}
6551 
6552 	for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
6553 		struct kvm_msr_entry msr;
6554 
6555 		msr.index = msr_based_features_all[i];
6556 		if (kvm_get_msr_feature(&msr))
6557 			continue;
6558 
6559 		msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6560 	}
6561 }
6562 
6563 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6564 			   const void *v)
6565 {
6566 	int handled = 0;
6567 	int n;
6568 
6569 	do {
6570 		n = min(len, 8);
6571 		if (!(lapic_in_kernel(vcpu) &&
6572 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6573 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6574 			break;
6575 		handled += n;
6576 		addr += n;
6577 		len -= n;
6578 		v += n;
6579 	} while (len);
6580 
6581 	return handled;
6582 }
6583 
6584 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6585 {
6586 	int handled = 0;
6587 	int n;
6588 
6589 	do {
6590 		n = min(len, 8);
6591 		if (!(lapic_in_kernel(vcpu) &&
6592 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6593 					 addr, n, v))
6594 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6595 			break;
6596 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6597 		handled += n;
6598 		addr += n;
6599 		len -= n;
6600 		v += n;
6601 	} while (len);
6602 
6603 	return handled;
6604 }
6605 
6606 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6607 			struct kvm_segment *var, int seg)
6608 {
6609 	static_call(kvm_x86_set_segment)(vcpu, var, seg);
6610 }
6611 
6612 void kvm_get_segment(struct kvm_vcpu *vcpu,
6613 		     struct kvm_segment *var, int seg)
6614 {
6615 	static_call(kvm_x86_get_segment)(vcpu, var, seg);
6616 }
6617 
6618 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6619 			   struct x86_exception *exception)
6620 {
6621 	struct kvm_mmu *mmu = vcpu->arch.mmu;
6622 	gpa_t t_gpa;
6623 
6624 	BUG_ON(!mmu_is_nested(vcpu));
6625 
6626 	/* NPT walks are always user-walks */
6627 	access |= PFERR_USER_MASK;
6628 	t_gpa  = mmu->gva_to_gpa(vcpu, mmu, gpa, access, exception);
6629 
6630 	return t_gpa;
6631 }
6632 
6633 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6634 			      struct x86_exception *exception)
6635 {
6636 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6637 
6638 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6639 	return mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
6640 }
6641 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6642 
6643  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6644 				struct x86_exception *exception)
6645 {
6646 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6647 
6648 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6649 	access |= PFERR_FETCH_MASK;
6650 	return mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
6651 }
6652 
6653 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6654 			       struct x86_exception *exception)
6655 {
6656 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6657 
6658 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6659 	access |= PFERR_WRITE_MASK;
6660 	return mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
6661 }
6662 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6663 
6664 /* uses this to access any guest's mapped memory without checking CPL */
6665 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6666 				struct x86_exception *exception)
6667 {
6668 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6669 
6670 	return mmu->gva_to_gpa(vcpu, mmu, gva, 0, exception);
6671 }
6672 
6673 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6674 				      struct kvm_vcpu *vcpu, u32 access,
6675 				      struct x86_exception *exception)
6676 {
6677 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6678 	void *data = val;
6679 	int r = X86EMUL_CONTINUE;
6680 
6681 	while (bytes) {
6682 		gpa_t gpa = mmu->gva_to_gpa(vcpu, mmu, addr, access, exception);
6683 		unsigned offset = addr & (PAGE_SIZE-1);
6684 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6685 		int ret;
6686 
6687 		if (gpa == UNMAPPED_GVA)
6688 			return X86EMUL_PROPAGATE_FAULT;
6689 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6690 					       offset, toread);
6691 		if (ret < 0) {
6692 			r = X86EMUL_IO_NEEDED;
6693 			goto out;
6694 		}
6695 
6696 		bytes -= toread;
6697 		data += toread;
6698 		addr += toread;
6699 	}
6700 out:
6701 	return r;
6702 }
6703 
6704 /* used for instruction fetching */
6705 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6706 				gva_t addr, void *val, unsigned int bytes,
6707 				struct x86_exception *exception)
6708 {
6709 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6710 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6711 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6712 	unsigned offset;
6713 	int ret;
6714 
6715 	/* Inline kvm_read_guest_virt_helper for speed.  */
6716 	gpa_t gpa = mmu->gva_to_gpa(vcpu, mmu, addr, access|PFERR_FETCH_MASK,
6717 				    exception);
6718 	if (unlikely(gpa == UNMAPPED_GVA))
6719 		return X86EMUL_PROPAGATE_FAULT;
6720 
6721 	offset = addr & (PAGE_SIZE-1);
6722 	if (WARN_ON(offset + bytes > PAGE_SIZE))
6723 		bytes = (unsigned)PAGE_SIZE - offset;
6724 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6725 				       offset, bytes);
6726 	if (unlikely(ret < 0))
6727 		return X86EMUL_IO_NEEDED;
6728 
6729 	return X86EMUL_CONTINUE;
6730 }
6731 
6732 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6733 			       gva_t addr, void *val, unsigned int bytes,
6734 			       struct x86_exception *exception)
6735 {
6736 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6737 
6738 	/*
6739 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6740 	 * is returned, but our callers are not ready for that and they blindly
6741 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
6742 	 * uninitialized kernel stack memory into cr2 and error code.
6743 	 */
6744 	memset(exception, 0, sizeof(*exception));
6745 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6746 					  exception);
6747 }
6748 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6749 
6750 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6751 			     gva_t addr, void *val, unsigned int bytes,
6752 			     struct x86_exception *exception, bool system)
6753 {
6754 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6755 	u32 access = 0;
6756 
6757 	if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6758 		access |= PFERR_USER_MASK;
6759 
6760 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6761 }
6762 
6763 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6764 		unsigned long addr, void *val, unsigned int bytes)
6765 {
6766 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6767 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6768 
6769 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6770 }
6771 
6772 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6773 				      struct kvm_vcpu *vcpu, u32 access,
6774 				      struct x86_exception *exception)
6775 {
6776 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6777 	void *data = val;
6778 	int r = X86EMUL_CONTINUE;
6779 
6780 	while (bytes) {
6781 		gpa_t gpa = mmu->gva_to_gpa(vcpu, mmu, addr, access, exception);
6782 		unsigned offset = addr & (PAGE_SIZE-1);
6783 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6784 		int ret;
6785 
6786 		if (gpa == UNMAPPED_GVA)
6787 			return X86EMUL_PROPAGATE_FAULT;
6788 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6789 		if (ret < 0) {
6790 			r = X86EMUL_IO_NEEDED;
6791 			goto out;
6792 		}
6793 
6794 		bytes -= towrite;
6795 		data += towrite;
6796 		addr += towrite;
6797 	}
6798 out:
6799 	return r;
6800 }
6801 
6802 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6803 			      unsigned int bytes, struct x86_exception *exception,
6804 			      bool system)
6805 {
6806 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6807 	u32 access = PFERR_WRITE_MASK;
6808 
6809 	if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6810 		access |= PFERR_USER_MASK;
6811 
6812 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6813 					   access, exception);
6814 }
6815 
6816 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6817 				unsigned int bytes, struct x86_exception *exception)
6818 {
6819 	/* kvm_write_guest_virt_system can pull in tons of pages. */
6820 	vcpu->arch.l1tf_flush_l1d = true;
6821 
6822 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6823 					   PFERR_WRITE_MASK, exception);
6824 }
6825 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6826 
6827 int handle_ud(struct kvm_vcpu *vcpu)
6828 {
6829 	static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6830 	int emul_type = EMULTYPE_TRAP_UD;
6831 	char sig[5]; /* ud2; .ascii "kvm" */
6832 	struct x86_exception e;
6833 
6834 	if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6835 		return 1;
6836 
6837 	if (force_emulation_prefix &&
6838 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6839 				sig, sizeof(sig), &e) == 0 &&
6840 	    memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6841 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6842 		emul_type = EMULTYPE_TRAP_UD_FORCED;
6843 	}
6844 
6845 	return kvm_emulate_instruction(vcpu, emul_type);
6846 }
6847 EXPORT_SYMBOL_GPL(handle_ud);
6848 
6849 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6850 			    gpa_t gpa, bool write)
6851 {
6852 	/* For APIC access vmexit */
6853 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6854 		return 1;
6855 
6856 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6857 		trace_vcpu_match_mmio(gva, gpa, write, true);
6858 		return 1;
6859 	}
6860 
6861 	return 0;
6862 }
6863 
6864 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6865 				gpa_t *gpa, struct x86_exception *exception,
6866 				bool write)
6867 {
6868 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6869 	u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6870 		| (write ? PFERR_WRITE_MASK : 0);
6871 
6872 	/*
6873 	 * currently PKRU is only applied to ept enabled guest so
6874 	 * there is no pkey in EPT page table for L1 guest or EPT
6875 	 * shadow page table for L2 guest.
6876 	 */
6877 	if (vcpu_match_mmio_gva(vcpu, gva) && (!is_paging(vcpu) ||
6878 	    !permission_fault(vcpu, vcpu->arch.walk_mmu,
6879 			      vcpu->arch.mmio_access, 0, access))) {
6880 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6881 					(gva & (PAGE_SIZE - 1));
6882 		trace_vcpu_match_mmio(gva, *gpa, write, false);
6883 		return 1;
6884 	}
6885 
6886 	*gpa = mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
6887 
6888 	if (*gpa == UNMAPPED_GVA)
6889 		return -1;
6890 
6891 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6892 }
6893 
6894 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6895 			const void *val, int bytes)
6896 {
6897 	int ret;
6898 
6899 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6900 	if (ret < 0)
6901 		return 0;
6902 	kvm_page_track_write(vcpu, gpa, val, bytes);
6903 	return 1;
6904 }
6905 
6906 struct read_write_emulator_ops {
6907 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6908 				  int bytes);
6909 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6910 				  void *val, int bytes);
6911 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6912 			       int bytes, void *val);
6913 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6914 				    void *val, int bytes);
6915 	bool write;
6916 };
6917 
6918 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6919 {
6920 	if (vcpu->mmio_read_completed) {
6921 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6922 			       vcpu->mmio_fragments[0].gpa, val);
6923 		vcpu->mmio_read_completed = 0;
6924 		return 1;
6925 	}
6926 
6927 	return 0;
6928 }
6929 
6930 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6931 			void *val, int bytes)
6932 {
6933 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6934 }
6935 
6936 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6937 			 void *val, int bytes)
6938 {
6939 	return emulator_write_phys(vcpu, gpa, val, bytes);
6940 }
6941 
6942 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6943 {
6944 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6945 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
6946 }
6947 
6948 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6949 			  void *val, int bytes)
6950 {
6951 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6952 	return X86EMUL_IO_NEEDED;
6953 }
6954 
6955 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6956 			   void *val, int bytes)
6957 {
6958 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6959 
6960 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6961 	return X86EMUL_CONTINUE;
6962 }
6963 
6964 static const struct read_write_emulator_ops read_emultor = {
6965 	.read_write_prepare = read_prepare,
6966 	.read_write_emulate = read_emulate,
6967 	.read_write_mmio = vcpu_mmio_read,
6968 	.read_write_exit_mmio = read_exit_mmio,
6969 };
6970 
6971 static const struct read_write_emulator_ops write_emultor = {
6972 	.read_write_emulate = write_emulate,
6973 	.read_write_mmio = write_mmio,
6974 	.read_write_exit_mmio = write_exit_mmio,
6975 	.write = true,
6976 };
6977 
6978 static int emulator_read_write_onepage(unsigned long addr, void *val,
6979 				       unsigned int bytes,
6980 				       struct x86_exception *exception,
6981 				       struct kvm_vcpu *vcpu,
6982 				       const struct read_write_emulator_ops *ops)
6983 {
6984 	gpa_t gpa;
6985 	int handled, ret;
6986 	bool write = ops->write;
6987 	struct kvm_mmio_fragment *frag;
6988 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6989 
6990 	/*
6991 	 * If the exit was due to a NPF we may already have a GPA.
6992 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6993 	 * Note, this cannot be used on string operations since string
6994 	 * operation using rep will only have the initial GPA from the NPF
6995 	 * occurred.
6996 	 */
6997 	if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6998 	    (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6999 		gpa = ctxt->gpa_val;
7000 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
7001 	} else {
7002 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
7003 		if (ret < 0)
7004 			return X86EMUL_PROPAGATE_FAULT;
7005 	}
7006 
7007 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
7008 		return X86EMUL_CONTINUE;
7009 
7010 	/*
7011 	 * Is this MMIO handled locally?
7012 	 */
7013 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
7014 	if (handled == bytes)
7015 		return X86EMUL_CONTINUE;
7016 
7017 	gpa += handled;
7018 	bytes -= handled;
7019 	val += handled;
7020 
7021 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
7022 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
7023 	frag->gpa = gpa;
7024 	frag->data = val;
7025 	frag->len = bytes;
7026 	return X86EMUL_CONTINUE;
7027 }
7028 
7029 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
7030 			unsigned long addr,
7031 			void *val, unsigned int bytes,
7032 			struct x86_exception *exception,
7033 			const struct read_write_emulator_ops *ops)
7034 {
7035 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7036 	gpa_t gpa;
7037 	int rc;
7038 
7039 	if (ops->read_write_prepare &&
7040 		  ops->read_write_prepare(vcpu, val, bytes))
7041 		return X86EMUL_CONTINUE;
7042 
7043 	vcpu->mmio_nr_fragments = 0;
7044 
7045 	/* Crossing a page boundary? */
7046 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
7047 		int now;
7048 
7049 		now = -addr & ~PAGE_MASK;
7050 		rc = emulator_read_write_onepage(addr, val, now, exception,
7051 						 vcpu, ops);
7052 
7053 		if (rc != X86EMUL_CONTINUE)
7054 			return rc;
7055 		addr += now;
7056 		if (ctxt->mode != X86EMUL_MODE_PROT64)
7057 			addr = (u32)addr;
7058 		val += now;
7059 		bytes -= now;
7060 	}
7061 
7062 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
7063 					 vcpu, ops);
7064 	if (rc != X86EMUL_CONTINUE)
7065 		return rc;
7066 
7067 	if (!vcpu->mmio_nr_fragments)
7068 		return rc;
7069 
7070 	gpa = vcpu->mmio_fragments[0].gpa;
7071 
7072 	vcpu->mmio_needed = 1;
7073 	vcpu->mmio_cur_fragment = 0;
7074 
7075 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
7076 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
7077 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
7078 	vcpu->run->mmio.phys_addr = gpa;
7079 
7080 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
7081 }
7082 
7083 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
7084 				  unsigned long addr,
7085 				  void *val,
7086 				  unsigned int bytes,
7087 				  struct x86_exception *exception)
7088 {
7089 	return emulator_read_write(ctxt, addr, val, bytes,
7090 				   exception, &read_emultor);
7091 }
7092 
7093 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
7094 			    unsigned long addr,
7095 			    const void *val,
7096 			    unsigned int bytes,
7097 			    struct x86_exception *exception)
7098 {
7099 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
7100 				   exception, &write_emultor);
7101 }
7102 
7103 #define CMPXCHG_TYPE(t, ptr, old, new) \
7104 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
7105 
7106 #ifdef CONFIG_X86_64
7107 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
7108 #else
7109 #  define CMPXCHG64(ptr, old, new) \
7110 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
7111 #endif
7112 
7113 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
7114 				     unsigned long addr,
7115 				     const void *old,
7116 				     const void *new,
7117 				     unsigned int bytes,
7118 				     struct x86_exception *exception)
7119 {
7120 	struct kvm_host_map map;
7121 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7122 	u64 page_line_mask;
7123 	gpa_t gpa;
7124 	char *kaddr;
7125 	bool exchanged;
7126 
7127 	/* guests cmpxchg8b have to be emulated atomically */
7128 	if (bytes > 8 || (bytes & (bytes - 1)))
7129 		goto emul_write;
7130 
7131 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
7132 
7133 	if (gpa == UNMAPPED_GVA ||
7134 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
7135 		goto emul_write;
7136 
7137 	/*
7138 	 * Emulate the atomic as a straight write to avoid #AC if SLD is
7139 	 * enabled in the host and the access splits a cache line.
7140 	 */
7141 	if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
7142 		page_line_mask = ~(cache_line_size() - 1);
7143 	else
7144 		page_line_mask = PAGE_MASK;
7145 
7146 	if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
7147 		goto emul_write;
7148 
7149 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
7150 		goto emul_write;
7151 
7152 	kaddr = map.hva + offset_in_page(gpa);
7153 
7154 	switch (bytes) {
7155 	case 1:
7156 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
7157 		break;
7158 	case 2:
7159 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
7160 		break;
7161 	case 4:
7162 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
7163 		break;
7164 	case 8:
7165 		exchanged = CMPXCHG64(kaddr, old, new);
7166 		break;
7167 	default:
7168 		BUG();
7169 	}
7170 
7171 	kvm_vcpu_unmap(vcpu, &map, true);
7172 
7173 	if (!exchanged)
7174 		return X86EMUL_CMPXCHG_FAILED;
7175 
7176 	kvm_page_track_write(vcpu, gpa, new, bytes);
7177 
7178 	return X86EMUL_CONTINUE;
7179 
7180 emul_write:
7181 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
7182 
7183 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
7184 }
7185 
7186 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
7187 {
7188 	int r = 0, i;
7189 
7190 	for (i = 0; i < vcpu->arch.pio.count; i++) {
7191 		if (vcpu->arch.pio.in)
7192 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
7193 					    vcpu->arch.pio.size, pd);
7194 		else
7195 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
7196 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
7197 					     pd);
7198 		if (r)
7199 			break;
7200 		pd += vcpu->arch.pio.size;
7201 	}
7202 	return r;
7203 }
7204 
7205 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
7206 			       unsigned short port,
7207 			       unsigned int count, bool in)
7208 {
7209 	vcpu->arch.pio.port = port;
7210 	vcpu->arch.pio.in = in;
7211 	vcpu->arch.pio.count  = count;
7212 	vcpu->arch.pio.size = size;
7213 
7214 	if (!kernel_pio(vcpu, vcpu->arch.pio_data))
7215 		return 1;
7216 
7217 	vcpu->run->exit_reason = KVM_EXIT_IO;
7218 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
7219 	vcpu->run->io.size = size;
7220 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
7221 	vcpu->run->io.count = count;
7222 	vcpu->run->io.port = port;
7223 
7224 	return 0;
7225 }
7226 
7227 static int __emulator_pio_in(struct kvm_vcpu *vcpu, int size,
7228 			     unsigned short port, unsigned int count)
7229 {
7230 	WARN_ON(vcpu->arch.pio.count);
7231 	memset(vcpu->arch.pio_data, 0, size * count);
7232 	return emulator_pio_in_out(vcpu, size, port, count, true);
7233 }
7234 
7235 static void complete_emulator_pio_in(struct kvm_vcpu *vcpu, void *val)
7236 {
7237 	int size = vcpu->arch.pio.size;
7238 	unsigned count = vcpu->arch.pio.count;
7239 	memcpy(val, vcpu->arch.pio_data, size * count);
7240 	trace_kvm_pio(KVM_PIO_IN, vcpu->arch.pio.port, size, count, vcpu->arch.pio_data);
7241 	vcpu->arch.pio.count = 0;
7242 }
7243 
7244 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
7245 			   unsigned short port, void *val, unsigned int count)
7246 {
7247 	if (vcpu->arch.pio.count) {
7248 		/*
7249 		 * Complete a previous iteration that required userspace I/O.
7250 		 * Note, @count isn't guaranteed to match pio.count as userspace
7251 		 * can modify ECX before rerunning the vCPU.  Ignore any such
7252 		 * shenanigans as KVM doesn't support modifying the rep count,
7253 		 * and the emulator ensures @count doesn't overflow the buffer.
7254 		 */
7255 	} else {
7256 		int r = __emulator_pio_in(vcpu, size, port, count);
7257 		if (!r)
7258 			return r;
7259 
7260 		/* Results already available, fall through.  */
7261 	}
7262 
7263 	complete_emulator_pio_in(vcpu, val);
7264 	return 1;
7265 }
7266 
7267 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
7268 				    int size, unsigned short port, void *val,
7269 				    unsigned int count)
7270 {
7271 	return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
7272 
7273 }
7274 
7275 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
7276 			    unsigned short port, const void *val,
7277 			    unsigned int count)
7278 {
7279 	int ret;
7280 
7281 	memcpy(vcpu->arch.pio_data, val, size * count);
7282 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
7283 	ret = emulator_pio_in_out(vcpu, size, port, count, false);
7284 	if (ret)
7285                 vcpu->arch.pio.count = 0;
7286 
7287         return ret;
7288 }
7289 
7290 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
7291 				     int size, unsigned short port,
7292 				     const void *val, unsigned int count)
7293 {
7294 	return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
7295 }
7296 
7297 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
7298 {
7299 	return static_call(kvm_x86_get_segment_base)(vcpu, seg);
7300 }
7301 
7302 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
7303 {
7304 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
7305 }
7306 
7307 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
7308 {
7309 	if (!need_emulate_wbinvd(vcpu))
7310 		return X86EMUL_CONTINUE;
7311 
7312 	if (static_call(kvm_x86_has_wbinvd_exit)()) {
7313 		int cpu = get_cpu();
7314 
7315 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
7316 		on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
7317 				wbinvd_ipi, NULL, 1);
7318 		put_cpu();
7319 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
7320 	} else
7321 		wbinvd();
7322 	return X86EMUL_CONTINUE;
7323 }
7324 
7325 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
7326 {
7327 	kvm_emulate_wbinvd_noskip(vcpu);
7328 	return kvm_skip_emulated_instruction(vcpu);
7329 }
7330 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
7331 
7332 
7333 
7334 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
7335 {
7336 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
7337 }
7338 
7339 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
7340 			    unsigned long *dest)
7341 {
7342 	kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
7343 }
7344 
7345 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
7346 			   unsigned long value)
7347 {
7348 
7349 	return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
7350 }
7351 
7352 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
7353 {
7354 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
7355 }
7356 
7357 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
7358 {
7359 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7360 	unsigned long value;
7361 
7362 	switch (cr) {
7363 	case 0:
7364 		value = kvm_read_cr0(vcpu);
7365 		break;
7366 	case 2:
7367 		value = vcpu->arch.cr2;
7368 		break;
7369 	case 3:
7370 		value = kvm_read_cr3(vcpu);
7371 		break;
7372 	case 4:
7373 		value = kvm_read_cr4(vcpu);
7374 		break;
7375 	case 8:
7376 		value = kvm_get_cr8(vcpu);
7377 		break;
7378 	default:
7379 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
7380 		return 0;
7381 	}
7382 
7383 	return value;
7384 }
7385 
7386 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
7387 {
7388 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7389 	int res = 0;
7390 
7391 	switch (cr) {
7392 	case 0:
7393 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
7394 		break;
7395 	case 2:
7396 		vcpu->arch.cr2 = val;
7397 		break;
7398 	case 3:
7399 		res = kvm_set_cr3(vcpu, val);
7400 		break;
7401 	case 4:
7402 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
7403 		break;
7404 	case 8:
7405 		res = kvm_set_cr8(vcpu, val);
7406 		break;
7407 	default:
7408 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
7409 		res = -1;
7410 	}
7411 
7412 	return res;
7413 }
7414 
7415 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
7416 {
7417 	return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
7418 }
7419 
7420 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7421 {
7422 	static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
7423 }
7424 
7425 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7426 {
7427 	static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
7428 }
7429 
7430 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7431 {
7432 	static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
7433 }
7434 
7435 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7436 {
7437 	static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
7438 }
7439 
7440 static unsigned long emulator_get_cached_segment_base(
7441 	struct x86_emulate_ctxt *ctxt, int seg)
7442 {
7443 	return get_segment_base(emul_to_vcpu(ctxt), seg);
7444 }
7445 
7446 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
7447 				 struct desc_struct *desc, u32 *base3,
7448 				 int seg)
7449 {
7450 	struct kvm_segment var;
7451 
7452 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
7453 	*selector = var.selector;
7454 
7455 	if (var.unusable) {
7456 		memset(desc, 0, sizeof(*desc));
7457 		if (base3)
7458 			*base3 = 0;
7459 		return false;
7460 	}
7461 
7462 	if (var.g)
7463 		var.limit >>= 12;
7464 	set_desc_limit(desc, var.limit);
7465 	set_desc_base(desc, (unsigned long)var.base);
7466 #ifdef CONFIG_X86_64
7467 	if (base3)
7468 		*base3 = var.base >> 32;
7469 #endif
7470 	desc->type = var.type;
7471 	desc->s = var.s;
7472 	desc->dpl = var.dpl;
7473 	desc->p = var.present;
7474 	desc->avl = var.avl;
7475 	desc->l = var.l;
7476 	desc->d = var.db;
7477 	desc->g = var.g;
7478 
7479 	return true;
7480 }
7481 
7482 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
7483 				 struct desc_struct *desc, u32 base3,
7484 				 int seg)
7485 {
7486 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7487 	struct kvm_segment var;
7488 
7489 	var.selector = selector;
7490 	var.base = get_desc_base(desc);
7491 #ifdef CONFIG_X86_64
7492 	var.base |= ((u64)base3) << 32;
7493 #endif
7494 	var.limit = get_desc_limit(desc);
7495 	if (desc->g)
7496 		var.limit = (var.limit << 12) | 0xfff;
7497 	var.type = desc->type;
7498 	var.dpl = desc->dpl;
7499 	var.db = desc->d;
7500 	var.s = desc->s;
7501 	var.l = desc->l;
7502 	var.g = desc->g;
7503 	var.avl = desc->avl;
7504 	var.present = desc->p;
7505 	var.unusable = !var.present;
7506 	var.padding = 0;
7507 
7508 	kvm_set_segment(vcpu, &var, seg);
7509 	return;
7510 }
7511 
7512 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
7513 			    u32 msr_index, u64 *pdata)
7514 {
7515 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7516 	int r;
7517 
7518 	r = kvm_get_msr(vcpu, msr_index, pdata);
7519 
7520 	if (r && kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_RDMSR, 0,
7521 				    complete_emulated_rdmsr, r)) {
7522 		/* Bounce to user space */
7523 		return X86EMUL_IO_NEEDED;
7524 	}
7525 
7526 	return r;
7527 }
7528 
7529 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
7530 			    u32 msr_index, u64 data)
7531 {
7532 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7533 	int r;
7534 
7535 	r = kvm_set_msr(vcpu, msr_index, data);
7536 
7537 	if (r && kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_WRMSR, data,
7538 				    complete_emulated_msr_access, r)) {
7539 		/* Bounce to user space */
7540 		return X86EMUL_IO_NEEDED;
7541 	}
7542 
7543 	return r;
7544 }
7545 
7546 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
7547 {
7548 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7549 
7550 	return vcpu->arch.smbase;
7551 }
7552 
7553 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
7554 {
7555 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7556 
7557 	vcpu->arch.smbase = smbase;
7558 }
7559 
7560 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
7561 			      u32 pmc)
7562 {
7563 	if (kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc))
7564 		return 0;
7565 	return -EINVAL;
7566 }
7567 
7568 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
7569 			     u32 pmc, u64 *pdata)
7570 {
7571 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
7572 }
7573 
7574 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
7575 {
7576 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
7577 }
7578 
7579 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
7580 			      struct x86_instruction_info *info,
7581 			      enum x86_intercept_stage stage)
7582 {
7583 	return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
7584 					    &ctxt->exception);
7585 }
7586 
7587 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
7588 			      u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
7589 			      bool exact_only)
7590 {
7591 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7592 }
7593 
7594 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7595 {
7596 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7597 }
7598 
7599 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7600 {
7601 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7602 }
7603 
7604 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7605 {
7606 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7607 }
7608 
7609 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7610 {
7611 	return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7612 }
7613 
7614 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7615 {
7616 	kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7617 }
7618 
7619 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7620 {
7621 	static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7622 }
7623 
7624 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7625 {
7626 	return emul_to_vcpu(ctxt)->arch.hflags;
7627 }
7628 
7629 static void emulator_exiting_smm(struct x86_emulate_ctxt *ctxt)
7630 {
7631 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7632 
7633 	kvm_smm_changed(vcpu, false);
7634 }
7635 
7636 static int emulator_leave_smm(struct x86_emulate_ctxt *ctxt,
7637 				  const char *smstate)
7638 {
7639 	return static_call(kvm_x86_leave_smm)(emul_to_vcpu(ctxt), smstate);
7640 }
7641 
7642 static void emulator_triple_fault(struct x86_emulate_ctxt *ctxt)
7643 {
7644 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, emul_to_vcpu(ctxt));
7645 }
7646 
7647 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7648 {
7649 	return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7650 }
7651 
7652 static const struct x86_emulate_ops emulate_ops = {
7653 	.read_gpr            = emulator_read_gpr,
7654 	.write_gpr           = emulator_write_gpr,
7655 	.read_std            = emulator_read_std,
7656 	.write_std           = emulator_write_std,
7657 	.read_phys           = kvm_read_guest_phys_system,
7658 	.fetch               = kvm_fetch_guest_virt,
7659 	.read_emulated       = emulator_read_emulated,
7660 	.write_emulated      = emulator_write_emulated,
7661 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
7662 	.invlpg              = emulator_invlpg,
7663 	.pio_in_emulated     = emulator_pio_in_emulated,
7664 	.pio_out_emulated    = emulator_pio_out_emulated,
7665 	.get_segment         = emulator_get_segment,
7666 	.set_segment         = emulator_set_segment,
7667 	.get_cached_segment_base = emulator_get_cached_segment_base,
7668 	.get_gdt             = emulator_get_gdt,
7669 	.get_idt	     = emulator_get_idt,
7670 	.set_gdt             = emulator_set_gdt,
7671 	.set_idt	     = emulator_set_idt,
7672 	.get_cr              = emulator_get_cr,
7673 	.set_cr              = emulator_set_cr,
7674 	.cpl                 = emulator_get_cpl,
7675 	.get_dr              = emulator_get_dr,
7676 	.set_dr              = emulator_set_dr,
7677 	.get_smbase          = emulator_get_smbase,
7678 	.set_smbase          = emulator_set_smbase,
7679 	.set_msr             = emulator_set_msr,
7680 	.get_msr             = emulator_get_msr,
7681 	.check_pmc	     = emulator_check_pmc,
7682 	.read_pmc            = emulator_read_pmc,
7683 	.halt                = emulator_halt,
7684 	.wbinvd              = emulator_wbinvd,
7685 	.fix_hypercall       = emulator_fix_hypercall,
7686 	.intercept           = emulator_intercept,
7687 	.get_cpuid           = emulator_get_cpuid,
7688 	.guest_has_long_mode = emulator_guest_has_long_mode,
7689 	.guest_has_movbe     = emulator_guest_has_movbe,
7690 	.guest_has_fxsr      = emulator_guest_has_fxsr,
7691 	.set_nmi_mask        = emulator_set_nmi_mask,
7692 	.get_hflags          = emulator_get_hflags,
7693 	.exiting_smm         = emulator_exiting_smm,
7694 	.leave_smm           = emulator_leave_smm,
7695 	.triple_fault        = emulator_triple_fault,
7696 	.set_xcr             = emulator_set_xcr,
7697 };
7698 
7699 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7700 {
7701 	u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7702 	/*
7703 	 * an sti; sti; sequence only disable interrupts for the first
7704 	 * instruction. So, if the last instruction, be it emulated or
7705 	 * not, left the system with the INT_STI flag enabled, it
7706 	 * means that the last instruction is an sti. We should not
7707 	 * leave the flag on in this case. The same goes for mov ss
7708 	 */
7709 	if (int_shadow & mask)
7710 		mask = 0;
7711 	if (unlikely(int_shadow || mask)) {
7712 		static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7713 		if (!mask)
7714 			kvm_make_request(KVM_REQ_EVENT, vcpu);
7715 	}
7716 }
7717 
7718 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7719 {
7720 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7721 	if (ctxt->exception.vector == PF_VECTOR)
7722 		return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7723 
7724 	if (ctxt->exception.error_code_valid)
7725 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7726 				      ctxt->exception.error_code);
7727 	else
7728 		kvm_queue_exception(vcpu, ctxt->exception.vector);
7729 	return false;
7730 }
7731 
7732 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7733 {
7734 	struct x86_emulate_ctxt *ctxt;
7735 
7736 	ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7737 	if (!ctxt) {
7738 		pr_err("kvm: failed to allocate vcpu's emulator\n");
7739 		return NULL;
7740 	}
7741 
7742 	ctxt->vcpu = vcpu;
7743 	ctxt->ops = &emulate_ops;
7744 	vcpu->arch.emulate_ctxt = ctxt;
7745 
7746 	return ctxt;
7747 }
7748 
7749 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7750 {
7751 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7752 	int cs_db, cs_l;
7753 
7754 	static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7755 
7756 	ctxt->gpa_available = false;
7757 	ctxt->eflags = kvm_get_rflags(vcpu);
7758 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7759 
7760 	ctxt->eip = kvm_rip_read(vcpu);
7761 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
7762 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
7763 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
7764 		     cs_db				? X86EMUL_MODE_PROT32 :
7765 							  X86EMUL_MODE_PROT16;
7766 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7767 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7768 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7769 
7770 	ctxt->interruptibility = 0;
7771 	ctxt->have_exception = false;
7772 	ctxt->exception.vector = -1;
7773 	ctxt->perm_ok = false;
7774 
7775 	init_decode_cache(ctxt);
7776 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7777 }
7778 
7779 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7780 {
7781 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7782 	int ret;
7783 
7784 	init_emulate_ctxt(vcpu);
7785 
7786 	ctxt->op_bytes = 2;
7787 	ctxt->ad_bytes = 2;
7788 	ctxt->_eip = ctxt->eip + inc_eip;
7789 	ret = emulate_int_real(ctxt, irq);
7790 
7791 	if (ret != X86EMUL_CONTINUE) {
7792 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7793 	} else {
7794 		ctxt->eip = ctxt->_eip;
7795 		kvm_rip_write(vcpu, ctxt->eip);
7796 		kvm_set_rflags(vcpu, ctxt->eflags);
7797 	}
7798 }
7799 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7800 
7801 static void prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, u64 *data,
7802 					   u8 ndata, u8 *insn_bytes, u8 insn_size)
7803 {
7804 	struct kvm_run *run = vcpu->run;
7805 	u64 info[5];
7806 	u8 info_start;
7807 
7808 	/*
7809 	 * Zero the whole array used to retrieve the exit info, as casting to
7810 	 * u32 for select entries will leave some chunks uninitialized.
7811 	 */
7812 	memset(&info, 0, sizeof(info));
7813 
7814 	static_call(kvm_x86_get_exit_info)(vcpu, (u32 *)&info[0], &info[1],
7815 					   &info[2], (u32 *)&info[3],
7816 					   (u32 *)&info[4]);
7817 
7818 	run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7819 	run->emulation_failure.suberror = KVM_INTERNAL_ERROR_EMULATION;
7820 
7821 	/*
7822 	 * There's currently space for 13 entries, but 5 are used for the exit
7823 	 * reason and info.  Restrict to 4 to reduce the maintenance burden
7824 	 * when expanding kvm_run.emulation_failure in the future.
7825 	 */
7826 	if (WARN_ON_ONCE(ndata > 4))
7827 		ndata = 4;
7828 
7829 	/* Always include the flags as a 'data' entry. */
7830 	info_start = 1;
7831 	run->emulation_failure.flags = 0;
7832 
7833 	if (insn_size) {
7834 		BUILD_BUG_ON((sizeof(run->emulation_failure.insn_size) +
7835 			      sizeof(run->emulation_failure.insn_bytes) != 16));
7836 		info_start += 2;
7837 		run->emulation_failure.flags |=
7838 			KVM_INTERNAL_ERROR_EMULATION_FLAG_INSTRUCTION_BYTES;
7839 		run->emulation_failure.insn_size = insn_size;
7840 		memset(run->emulation_failure.insn_bytes, 0x90,
7841 		       sizeof(run->emulation_failure.insn_bytes));
7842 		memcpy(run->emulation_failure.insn_bytes, insn_bytes, insn_size);
7843 	}
7844 
7845 	memcpy(&run->internal.data[info_start], info, sizeof(info));
7846 	memcpy(&run->internal.data[info_start + ARRAY_SIZE(info)], data,
7847 	       ndata * sizeof(data[0]));
7848 
7849 	run->emulation_failure.ndata = info_start + ARRAY_SIZE(info) + ndata;
7850 }
7851 
7852 static void prepare_emulation_ctxt_failure_exit(struct kvm_vcpu *vcpu)
7853 {
7854 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7855 
7856 	prepare_emulation_failure_exit(vcpu, NULL, 0, ctxt->fetch.data,
7857 				       ctxt->fetch.end - ctxt->fetch.data);
7858 }
7859 
7860 void __kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, u64 *data,
7861 					  u8 ndata)
7862 {
7863 	prepare_emulation_failure_exit(vcpu, data, ndata, NULL, 0);
7864 }
7865 EXPORT_SYMBOL_GPL(__kvm_prepare_emulation_failure_exit);
7866 
7867 void kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu)
7868 {
7869 	__kvm_prepare_emulation_failure_exit(vcpu, NULL, 0);
7870 }
7871 EXPORT_SYMBOL_GPL(kvm_prepare_emulation_failure_exit);
7872 
7873 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7874 {
7875 	struct kvm *kvm = vcpu->kvm;
7876 
7877 	++vcpu->stat.insn_emulation_fail;
7878 	trace_kvm_emulate_insn_failed(vcpu);
7879 
7880 	if (emulation_type & EMULTYPE_VMWARE_GP) {
7881 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7882 		return 1;
7883 	}
7884 
7885 	if (kvm->arch.exit_on_emulation_error ||
7886 	    (emulation_type & EMULTYPE_SKIP)) {
7887 		prepare_emulation_ctxt_failure_exit(vcpu);
7888 		return 0;
7889 	}
7890 
7891 	kvm_queue_exception(vcpu, UD_VECTOR);
7892 
7893 	if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7894 		prepare_emulation_ctxt_failure_exit(vcpu);
7895 		return 0;
7896 	}
7897 
7898 	return 1;
7899 }
7900 
7901 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7902 				  bool write_fault_to_shadow_pgtable,
7903 				  int emulation_type)
7904 {
7905 	gpa_t gpa = cr2_or_gpa;
7906 	kvm_pfn_t pfn;
7907 
7908 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7909 		return false;
7910 
7911 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7912 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7913 		return false;
7914 
7915 	if (!vcpu->arch.mmu->direct_map) {
7916 		/*
7917 		 * Write permission should be allowed since only
7918 		 * write access need to be emulated.
7919 		 */
7920 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7921 
7922 		/*
7923 		 * If the mapping is invalid in guest, let cpu retry
7924 		 * it to generate fault.
7925 		 */
7926 		if (gpa == UNMAPPED_GVA)
7927 			return true;
7928 	}
7929 
7930 	/*
7931 	 * Do not retry the unhandleable instruction if it faults on the
7932 	 * readonly host memory, otherwise it will goto a infinite loop:
7933 	 * retry instruction -> write #PF -> emulation fail -> retry
7934 	 * instruction -> ...
7935 	 */
7936 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7937 
7938 	/*
7939 	 * If the instruction failed on the error pfn, it can not be fixed,
7940 	 * report the error to userspace.
7941 	 */
7942 	if (is_error_noslot_pfn(pfn))
7943 		return false;
7944 
7945 	kvm_release_pfn_clean(pfn);
7946 
7947 	/* The instructions are well-emulated on direct mmu. */
7948 	if (vcpu->arch.mmu->direct_map) {
7949 		unsigned int indirect_shadow_pages;
7950 
7951 		write_lock(&vcpu->kvm->mmu_lock);
7952 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7953 		write_unlock(&vcpu->kvm->mmu_lock);
7954 
7955 		if (indirect_shadow_pages)
7956 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7957 
7958 		return true;
7959 	}
7960 
7961 	/*
7962 	 * if emulation was due to access to shadowed page table
7963 	 * and it failed try to unshadow page and re-enter the
7964 	 * guest to let CPU execute the instruction.
7965 	 */
7966 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7967 
7968 	/*
7969 	 * If the access faults on its page table, it can not
7970 	 * be fixed by unprotecting shadow page and it should
7971 	 * be reported to userspace.
7972 	 */
7973 	return !write_fault_to_shadow_pgtable;
7974 }
7975 
7976 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7977 			      gpa_t cr2_or_gpa,  int emulation_type)
7978 {
7979 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7980 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7981 
7982 	last_retry_eip = vcpu->arch.last_retry_eip;
7983 	last_retry_addr = vcpu->arch.last_retry_addr;
7984 
7985 	/*
7986 	 * If the emulation is caused by #PF and it is non-page_table
7987 	 * writing instruction, it means the VM-EXIT is caused by shadow
7988 	 * page protected, we can zap the shadow page and retry this
7989 	 * instruction directly.
7990 	 *
7991 	 * Note: if the guest uses a non-page-table modifying instruction
7992 	 * on the PDE that points to the instruction, then we will unmap
7993 	 * the instruction and go to an infinite loop. So, we cache the
7994 	 * last retried eip and the last fault address, if we meet the eip
7995 	 * and the address again, we can break out of the potential infinite
7996 	 * loop.
7997 	 */
7998 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7999 
8000 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
8001 		return false;
8002 
8003 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
8004 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
8005 		return false;
8006 
8007 	if (x86_page_table_writing_insn(ctxt))
8008 		return false;
8009 
8010 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
8011 		return false;
8012 
8013 	vcpu->arch.last_retry_eip = ctxt->eip;
8014 	vcpu->arch.last_retry_addr = cr2_or_gpa;
8015 
8016 	if (!vcpu->arch.mmu->direct_map)
8017 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
8018 
8019 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
8020 
8021 	return true;
8022 }
8023 
8024 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
8025 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
8026 
8027 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm)
8028 {
8029 	trace_kvm_smm_transition(vcpu->vcpu_id, vcpu->arch.smbase, entering_smm);
8030 
8031 	if (entering_smm) {
8032 		vcpu->arch.hflags |= HF_SMM_MASK;
8033 	} else {
8034 		vcpu->arch.hflags &= ~(HF_SMM_MASK | HF_SMM_INSIDE_NMI_MASK);
8035 
8036 		/* Process a latched INIT or SMI, if any.  */
8037 		kvm_make_request(KVM_REQ_EVENT, vcpu);
8038 
8039 		/*
8040 		 * Even if KVM_SET_SREGS2 loaded PDPTRs out of band,
8041 		 * on SMM exit we still need to reload them from
8042 		 * guest memory
8043 		 */
8044 		vcpu->arch.pdptrs_from_userspace = false;
8045 	}
8046 
8047 	kvm_mmu_reset_context(vcpu);
8048 }
8049 
8050 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
8051 				unsigned long *db)
8052 {
8053 	u32 dr6 = 0;
8054 	int i;
8055 	u32 enable, rwlen;
8056 
8057 	enable = dr7;
8058 	rwlen = dr7 >> 16;
8059 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
8060 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
8061 			dr6 |= (1 << i);
8062 	return dr6;
8063 }
8064 
8065 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
8066 {
8067 	struct kvm_run *kvm_run = vcpu->run;
8068 
8069 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
8070 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
8071 		kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
8072 		kvm_run->debug.arch.exception = DB_VECTOR;
8073 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
8074 		return 0;
8075 	}
8076 	kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
8077 	return 1;
8078 }
8079 
8080 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
8081 {
8082 	unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
8083 	int r;
8084 
8085 	r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
8086 	if (unlikely(!r))
8087 		return 0;
8088 
8089 	kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_INSTRUCTIONS);
8090 
8091 	/*
8092 	 * rflags is the old, "raw" value of the flags.  The new value has
8093 	 * not been saved yet.
8094 	 *
8095 	 * This is correct even for TF set by the guest, because "the
8096 	 * processor will not generate this exception after the instruction
8097 	 * that sets the TF flag".
8098 	 */
8099 	if (unlikely(rflags & X86_EFLAGS_TF))
8100 		r = kvm_vcpu_do_singlestep(vcpu);
8101 	return r;
8102 }
8103 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
8104 
8105 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
8106 {
8107 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
8108 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
8109 		struct kvm_run *kvm_run = vcpu->run;
8110 		unsigned long eip = kvm_get_linear_rip(vcpu);
8111 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
8112 					   vcpu->arch.guest_debug_dr7,
8113 					   vcpu->arch.eff_db);
8114 
8115 		if (dr6 != 0) {
8116 			kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
8117 			kvm_run->debug.arch.pc = eip;
8118 			kvm_run->debug.arch.exception = DB_VECTOR;
8119 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
8120 			*r = 0;
8121 			return true;
8122 		}
8123 	}
8124 
8125 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
8126 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
8127 		unsigned long eip = kvm_get_linear_rip(vcpu);
8128 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
8129 					   vcpu->arch.dr7,
8130 					   vcpu->arch.db);
8131 
8132 		if (dr6 != 0) {
8133 			kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
8134 			*r = 1;
8135 			return true;
8136 		}
8137 	}
8138 
8139 	return false;
8140 }
8141 
8142 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
8143 {
8144 	switch (ctxt->opcode_len) {
8145 	case 1:
8146 		switch (ctxt->b) {
8147 		case 0xe4:	/* IN */
8148 		case 0xe5:
8149 		case 0xec:
8150 		case 0xed:
8151 		case 0xe6:	/* OUT */
8152 		case 0xe7:
8153 		case 0xee:
8154 		case 0xef:
8155 		case 0x6c:	/* INS */
8156 		case 0x6d:
8157 		case 0x6e:	/* OUTS */
8158 		case 0x6f:
8159 			return true;
8160 		}
8161 		break;
8162 	case 2:
8163 		switch (ctxt->b) {
8164 		case 0x33:	/* RDPMC */
8165 			return true;
8166 		}
8167 		break;
8168 	}
8169 
8170 	return false;
8171 }
8172 
8173 /*
8174  * Decode to be emulated instruction. Return EMULATION_OK if success.
8175  */
8176 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
8177 				    void *insn, int insn_len)
8178 {
8179 	int r = EMULATION_OK;
8180 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8181 
8182 	init_emulate_ctxt(vcpu);
8183 
8184 	/*
8185 	 * We will reenter on the same instruction since we do not set
8186 	 * complete_userspace_io. This does not handle watchpoints yet,
8187 	 * those would be handled in the emulate_ops.
8188 	 */
8189 	if (!(emulation_type & EMULTYPE_SKIP) &&
8190 	    kvm_vcpu_check_breakpoint(vcpu, &r))
8191 		return r;
8192 
8193 	r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
8194 
8195 	trace_kvm_emulate_insn_start(vcpu);
8196 	++vcpu->stat.insn_emulation;
8197 
8198 	return r;
8199 }
8200 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
8201 
8202 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
8203 			    int emulation_type, void *insn, int insn_len)
8204 {
8205 	int r;
8206 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8207 	bool writeback = true;
8208 	bool write_fault_to_spt;
8209 
8210 	if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
8211 		return 1;
8212 
8213 	vcpu->arch.l1tf_flush_l1d = true;
8214 
8215 	/*
8216 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
8217 	 * never reused.
8218 	 */
8219 	write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
8220 	vcpu->arch.write_fault_to_shadow_pgtable = false;
8221 
8222 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
8223 		kvm_clear_exception_queue(vcpu);
8224 
8225 		r = x86_decode_emulated_instruction(vcpu, emulation_type,
8226 						    insn, insn_len);
8227 		if (r != EMULATION_OK)  {
8228 			if ((emulation_type & EMULTYPE_TRAP_UD) ||
8229 			    (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
8230 				kvm_queue_exception(vcpu, UD_VECTOR);
8231 				return 1;
8232 			}
8233 			if (reexecute_instruction(vcpu, cr2_or_gpa,
8234 						  write_fault_to_spt,
8235 						  emulation_type))
8236 				return 1;
8237 			if (ctxt->have_exception) {
8238 				/*
8239 				 * #UD should result in just EMULATION_FAILED, and trap-like
8240 				 * exception should not be encountered during decode.
8241 				 */
8242 				WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
8243 					     exception_type(ctxt->exception.vector) == EXCPT_TRAP);
8244 				inject_emulated_exception(vcpu);
8245 				return 1;
8246 			}
8247 			return handle_emulation_failure(vcpu, emulation_type);
8248 		}
8249 	}
8250 
8251 	if ((emulation_type & EMULTYPE_VMWARE_GP) &&
8252 	    !is_vmware_backdoor_opcode(ctxt)) {
8253 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
8254 		return 1;
8255 	}
8256 
8257 	/*
8258 	 * EMULTYPE_SKIP without EMULTYPE_COMPLETE_USER_EXIT is intended for
8259 	 * use *only* by vendor callbacks for kvm_skip_emulated_instruction().
8260 	 * The caller is responsible for updating interruptibility state and
8261 	 * injecting single-step #DBs.
8262 	 */
8263 	if (emulation_type & EMULTYPE_SKIP) {
8264 		if (ctxt->mode != X86EMUL_MODE_PROT64)
8265 			ctxt->eip = (u32)ctxt->_eip;
8266 		else
8267 			ctxt->eip = ctxt->_eip;
8268 
8269 		if (emulation_type & EMULTYPE_COMPLETE_USER_EXIT) {
8270 			r = 1;
8271 			goto writeback;
8272 		}
8273 
8274 		kvm_rip_write(vcpu, ctxt->eip);
8275 		if (ctxt->eflags & X86_EFLAGS_RF)
8276 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
8277 		return 1;
8278 	}
8279 
8280 	if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
8281 		return 1;
8282 
8283 	/* this is needed for vmware backdoor interface to work since it
8284 	   changes registers values  during IO operation */
8285 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
8286 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8287 		emulator_invalidate_register_cache(ctxt);
8288 	}
8289 
8290 restart:
8291 	if (emulation_type & EMULTYPE_PF) {
8292 		/* Save the faulting GPA (cr2) in the address field */
8293 		ctxt->exception.address = cr2_or_gpa;
8294 
8295 		/* With shadow page tables, cr2 contains a GVA or nGPA. */
8296 		if (vcpu->arch.mmu->direct_map) {
8297 			ctxt->gpa_available = true;
8298 			ctxt->gpa_val = cr2_or_gpa;
8299 		}
8300 	} else {
8301 		/* Sanitize the address out of an abundance of paranoia. */
8302 		ctxt->exception.address = 0;
8303 	}
8304 
8305 	r = x86_emulate_insn(ctxt);
8306 
8307 	if (r == EMULATION_INTERCEPTED)
8308 		return 1;
8309 
8310 	if (r == EMULATION_FAILED) {
8311 		if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
8312 					emulation_type))
8313 			return 1;
8314 
8315 		return handle_emulation_failure(vcpu, emulation_type);
8316 	}
8317 
8318 	if (ctxt->have_exception) {
8319 		r = 1;
8320 		if (inject_emulated_exception(vcpu))
8321 			return r;
8322 	} else if (vcpu->arch.pio.count) {
8323 		if (!vcpu->arch.pio.in) {
8324 			/* FIXME: return into emulator if single-stepping.  */
8325 			vcpu->arch.pio.count = 0;
8326 		} else {
8327 			writeback = false;
8328 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
8329 		}
8330 		r = 0;
8331 	} else if (vcpu->mmio_needed) {
8332 		++vcpu->stat.mmio_exits;
8333 
8334 		if (!vcpu->mmio_is_write)
8335 			writeback = false;
8336 		r = 0;
8337 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8338 	} else if (vcpu->arch.complete_userspace_io) {
8339 		writeback = false;
8340 		r = 0;
8341 	} else if (r == EMULATION_RESTART)
8342 		goto restart;
8343 	else
8344 		r = 1;
8345 
8346 writeback:
8347 	if (writeback) {
8348 		unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
8349 		toggle_interruptibility(vcpu, ctxt->interruptibility);
8350 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8351 		if (!ctxt->have_exception ||
8352 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
8353 			kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_INSTRUCTIONS);
8354 			if (ctxt->is_branch)
8355 				kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_BRANCH_INSTRUCTIONS);
8356 			kvm_rip_write(vcpu, ctxt->eip);
8357 			if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
8358 				r = kvm_vcpu_do_singlestep(vcpu);
8359 			if (kvm_x86_ops.update_emulated_instruction)
8360 				static_call(kvm_x86_update_emulated_instruction)(vcpu);
8361 			__kvm_set_rflags(vcpu, ctxt->eflags);
8362 		}
8363 
8364 		/*
8365 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
8366 		 * do nothing, and it will be requested again as soon as
8367 		 * the shadow expires.  But we still need to check here,
8368 		 * because POPF has no interrupt shadow.
8369 		 */
8370 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
8371 			kvm_make_request(KVM_REQ_EVENT, vcpu);
8372 	} else
8373 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
8374 
8375 	return r;
8376 }
8377 
8378 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
8379 {
8380 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
8381 }
8382 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
8383 
8384 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
8385 					void *insn, int insn_len)
8386 {
8387 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
8388 }
8389 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
8390 
8391 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
8392 {
8393 	vcpu->arch.pio.count = 0;
8394 	return 1;
8395 }
8396 
8397 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
8398 {
8399 	vcpu->arch.pio.count = 0;
8400 
8401 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
8402 		return 1;
8403 
8404 	return kvm_skip_emulated_instruction(vcpu);
8405 }
8406 
8407 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
8408 			    unsigned short port)
8409 {
8410 	unsigned long val = kvm_rax_read(vcpu);
8411 	int ret = emulator_pio_out(vcpu, size, port, &val, 1);
8412 
8413 	if (ret)
8414 		return ret;
8415 
8416 	/*
8417 	 * Workaround userspace that relies on old KVM behavior of %rip being
8418 	 * incremented prior to exiting to userspace to handle "OUT 0x7e".
8419 	 */
8420 	if (port == 0x7e &&
8421 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
8422 		vcpu->arch.complete_userspace_io =
8423 			complete_fast_pio_out_port_0x7e;
8424 		kvm_skip_emulated_instruction(vcpu);
8425 	} else {
8426 		vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8427 		vcpu->arch.complete_userspace_io = complete_fast_pio_out;
8428 	}
8429 	return 0;
8430 }
8431 
8432 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
8433 {
8434 	unsigned long val;
8435 
8436 	/* We should only ever be called with arch.pio.count equal to 1 */
8437 	BUG_ON(vcpu->arch.pio.count != 1);
8438 
8439 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
8440 		vcpu->arch.pio.count = 0;
8441 		return 1;
8442 	}
8443 
8444 	/* For size less than 4 we merge, else we zero extend */
8445 	val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
8446 
8447 	/*
8448 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
8449 	 * the copy and tracing
8450 	 */
8451 	emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
8452 	kvm_rax_write(vcpu, val);
8453 
8454 	return kvm_skip_emulated_instruction(vcpu);
8455 }
8456 
8457 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
8458 			   unsigned short port)
8459 {
8460 	unsigned long val;
8461 	int ret;
8462 
8463 	/* For size less than 4 we merge, else we zero extend */
8464 	val = (size < 4) ? kvm_rax_read(vcpu) : 0;
8465 
8466 	ret = emulator_pio_in(vcpu, size, port, &val, 1);
8467 	if (ret) {
8468 		kvm_rax_write(vcpu, val);
8469 		return ret;
8470 	}
8471 
8472 	vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8473 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
8474 
8475 	return 0;
8476 }
8477 
8478 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
8479 {
8480 	int ret;
8481 
8482 	if (in)
8483 		ret = kvm_fast_pio_in(vcpu, size, port);
8484 	else
8485 		ret = kvm_fast_pio_out(vcpu, size, port);
8486 	return ret && kvm_skip_emulated_instruction(vcpu);
8487 }
8488 EXPORT_SYMBOL_GPL(kvm_fast_pio);
8489 
8490 static int kvmclock_cpu_down_prep(unsigned int cpu)
8491 {
8492 	__this_cpu_write(cpu_tsc_khz, 0);
8493 	return 0;
8494 }
8495 
8496 static void tsc_khz_changed(void *data)
8497 {
8498 	struct cpufreq_freqs *freq = data;
8499 	unsigned long khz = 0;
8500 
8501 	if (data)
8502 		khz = freq->new;
8503 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8504 		khz = cpufreq_quick_get(raw_smp_processor_id());
8505 	if (!khz)
8506 		khz = tsc_khz;
8507 	__this_cpu_write(cpu_tsc_khz, khz);
8508 }
8509 
8510 #ifdef CONFIG_X86_64
8511 static void kvm_hyperv_tsc_notifier(void)
8512 {
8513 	struct kvm *kvm;
8514 	int cpu;
8515 
8516 	mutex_lock(&kvm_lock);
8517 	list_for_each_entry(kvm, &vm_list, vm_list)
8518 		kvm_make_mclock_inprogress_request(kvm);
8519 
8520 	/* no guest entries from this point */
8521 	hyperv_stop_tsc_emulation();
8522 
8523 	/* TSC frequency always matches when on Hyper-V */
8524 	for_each_present_cpu(cpu)
8525 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
8526 	kvm_max_guest_tsc_khz = tsc_khz;
8527 
8528 	list_for_each_entry(kvm, &vm_list, vm_list) {
8529 		__kvm_start_pvclock_update(kvm);
8530 		pvclock_update_vm_gtod_copy(kvm);
8531 		kvm_end_pvclock_update(kvm);
8532 	}
8533 
8534 	mutex_unlock(&kvm_lock);
8535 }
8536 #endif
8537 
8538 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
8539 {
8540 	struct kvm *kvm;
8541 	struct kvm_vcpu *vcpu;
8542 	int send_ipi = 0;
8543 	unsigned long i;
8544 
8545 	/*
8546 	 * We allow guests to temporarily run on slowing clocks,
8547 	 * provided we notify them after, or to run on accelerating
8548 	 * clocks, provided we notify them before.  Thus time never
8549 	 * goes backwards.
8550 	 *
8551 	 * However, we have a problem.  We can't atomically update
8552 	 * the frequency of a given CPU from this function; it is
8553 	 * merely a notifier, which can be called from any CPU.
8554 	 * Changing the TSC frequency at arbitrary points in time
8555 	 * requires a recomputation of local variables related to
8556 	 * the TSC for each VCPU.  We must flag these local variables
8557 	 * to be updated and be sure the update takes place with the
8558 	 * new frequency before any guests proceed.
8559 	 *
8560 	 * Unfortunately, the combination of hotplug CPU and frequency
8561 	 * change creates an intractable locking scenario; the order
8562 	 * of when these callouts happen is undefined with respect to
8563 	 * CPU hotplug, and they can race with each other.  As such,
8564 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
8565 	 * undefined; you can actually have a CPU frequency change take
8566 	 * place in between the computation of X and the setting of the
8567 	 * variable.  To protect against this problem, all updates of
8568 	 * the per_cpu tsc_khz variable are done in an interrupt
8569 	 * protected IPI, and all callers wishing to update the value
8570 	 * must wait for a synchronous IPI to complete (which is trivial
8571 	 * if the caller is on the CPU already).  This establishes the
8572 	 * necessary total order on variable updates.
8573 	 *
8574 	 * Note that because a guest time update may take place
8575 	 * anytime after the setting of the VCPU's request bit, the
8576 	 * correct TSC value must be set before the request.  However,
8577 	 * to ensure the update actually makes it to any guest which
8578 	 * starts running in hardware virtualization between the set
8579 	 * and the acquisition of the spinlock, we must also ping the
8580 	 * CPU after setting the request bit.
8581 	 *
8582 	 */
8583 
8584 	smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8585 
8586 	mutex_lock(&kvm_lock);
8587 	list_for_each_entry(kvm, &vm_list, vm_list) {
8588 		kvm_for_each_vcpu(i, vcpu, kvm) {
8589 			if (vcpu->cpu != cpu)
8590 				continue;
8591 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8592 			if (vcpu->cpu != raw_smp_processor_id())
8593 				send_ipi = 1;
8594 		}
8595 	}
8596 	mutex_unlock(&kvm_lock);
8597 
8598 	if (freq->old < freq->new && send_ipi) {
8599 		/*
8600 		 * We upscale the frequency.  Must make the guest
8601 		 * doesn't see old kvmclock values while running with
8602 		 * the new frequency, otherwise we risk the guest sees
8603 		 * time go backwards.
8604 		 *
8605 		 * In case we update the frequency for another cpu
8606 		 * (which might be in guest context) send an interrupt
8607 		 * to kick the cpu out of guest context.  Next time
8608 		 * guest context is entered kvmclock will be updated,
8609 		 * so the guest will not see stale values.
8610 		 */
8611 		smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8612 	}
8613 }
8614 
8615 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
8616 				     void *data)
8617 {
8618 	struct cpufreq_freqs *freq = data;
8619 	int cpu;
8620 
8621 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
8622 		return 0;
8623 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
8624 		return 0;
8625 
8626 	for_each_cpu(cpu, freq->policy->cpus)
8627 		__kvmclock_cpufreq_notifier(freq, cpu);
8628 
8629 	return 0;
8630 }
8631 
8632 static struct notifier_block kvmclock_cpufreq_notifier_block = {
8633 	.notifier_call  = kvmclock_cpufreq_notifier
8634 };
8635 
8636 static int kvmclock_cpu_online(unsigned int cpu)
8637 {
8638 	tsc_khz_changed(NULL);
8639 	return 0;
8640 }
8641 
8642 static void kvm_timer_init(void)
8643 {
8644 	max_tsc_khz = tsc_khz;
8645 
8646 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
8647 #ifdef CONFIG_CPU_FREQ
8648 		struct cpufreq_policy *policy;
8649 		int cpu;
8650 
8651 		cpu = get_cpu();
8652 		policy = cpufreq_cpu_get(cpu);
8653 		if (policy) {
8654 			if (policy->cpuinfo.max_freq)
8655 				max_tsc_khz = policy->cpuinfo.max_freq;
8656 			cpufreq_cpu_put(policy);
8657 		}
8658 		put_cpu();
8659 #endif
8660 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
8661 					  CPUFREQ_TRANSITION_NOTIFIER);
8662 	}
8663 
8664 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
8665 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
8666 }
8667 
8668 #ifdef CONFIG_X86_64
8669 static void pvclock_gtod_update_fn(struct work_struct *work)
8670 {
8671 	struct kvm *kvm;
8672 	struct kvm_vcpu *vcpu;
8673 	unsigned long i;
8674 
8675 	mutex_lock(&kvm_lock);
8676 	list_for_each_entry(kvm, &vm_list, vm_list)
8677 		kvm_for_each_vcpu(i, vcpu, kvm)
8678 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8679 	atomic_set(&kvm_guest_has_master_clock, 0);
8680 	mutex_unlock(&kvm_lock);
8681 }
8682 
8683 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8684 
8685 /*
8686  * Indirection to move queue_work() out of the tk_core.seq write held
8687  * region to prevent possible deadlocks against time accessors which
8688  * are invoked with work related locks held.
8689  */
8690 static void pvclock_irq_work_fn(struct irq_work *w)
8691 {
8692 	queue_work(system_long_wq, &pvclock_gtod_work);
8693 }
8694 
8695 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
8696 
8697 /*
8698  * Notification about pvclock gtod data update.
8699  */
8700 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8701 			       void *priv)
8702 {
8703 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8704 	struct timekeeper *tk = priv;
8705 
8706 	update_pvclock_gtod(tk);
8707 
8708 	/*
8709 	 * Disable master clock if host does not trust, or does not use,
8710 	 * TSC based clocksource. Delegate queue_work() to irq_work as
8711 	 * this is invoked with tk_core.seq write held.
8712 	 */
8713 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8714 	    atomic_read(&kvm_guest_has_master_clock) != 0)
8715 		irq_work_queue(&pvclock_irq_work);
8716 	return 0;
8717 }
8718 
8719 static struct notifier_block pvclock_gtod_notifier = {
8720 	.notifier_call = pvclock_gtod_notify,
8721 };
8722 #endif
8723 
8724 int kvm_arch_init(void *opaque)
8725 {
8726 	struct kvm_x86_init_ops *ops = opaque;
8727 	int r;
8728 
8729 	if (kvm_x86_ops.hardware_enable) {
8730 		pr_err("kvm: already loaded vendor module '%s'\n", kvm_x86_ops.name);
8731 		r = -EEXIST;
8732 		goto out;
8733 	}
8734 
8735 	if (!ops->cpu_has_kvm_support()) {
8736 		pr_err_ratelimited("kvm: no hardware support for '%s'\n",
8737 				   ops->runtime_ops->name);
8738 		r = -EOPNOTSUPP;
8739 		goto out;
8740 	}
8741 	if (ops->disabled_by_bios()) {
8742 		pr_err_ratelimited("kvm: support for '%s' disabled by bios\n",
8743 				   ops->runtime_ops->name);
8744 		r = -EOPNOTSUPP;
8745 		goto out;
8746 	}
8747 
8748 	/*
8749 	 * KVM explicitly assumes that the guest has an FPU and
8750 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8751 	 * vCPU's FPU state as a fxregs_state struct.
8752 	 */
8753 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8754 		printk(KERN_ERR "kvm: inadequate fpu\n");
8755 		r = -EOPNOTSUPP;
8756 		goto out;
8757 	}
8758 
8759 	r = -ENOMEM;
8760 
8761 	x86_emulator_cache = kvm_alloc_emulator_cache();
8762 	if (!x86_emulator_cache) {
8763 		pr_err("kvm: failed to allocate cache for x86 emulator\n");
8764 		goto out;
8765 	}
8766 
8767 	user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8768 	if (!user_return_msrs) {
8769 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8770 		goto out_free_x86_emulator_cache;
8771 	}
8772 	kvm_nr_uret_msrs = 0;
8773 
8774 	r = kvm_mmu_module_init();
8775 	if (r)
8776 		goto out_free_percpu;
8777 
8778 	kvm_timer_init();
8779 
8780 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8781 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8782 		supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8783 	}
8784 
8785 	if (pi_inject_timer == -1)
8786 		pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8787 #ifdef CONFIG_X86_64
8788 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8789 
8790 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8791 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8792 #endif
8793 
8794 	return 0;
8795 
8796 out_free_percpu:
8797 	free_percpu(user_return_msrs);
8798 out_free_x86_emulator_cache:
8799 	kmem_cache_destroy(x86_emulator_cache);
8800 out:
8801 	return r;
8802 }
8803 
8804 void kvm_arch_exit(void)
8805 {
8806 #ifdef CONFIG_X86_64
8807 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8808 		clear_hv_tscchange_cb();
8809 #endif
8810 	kvm_lapic_exit();
8811 
8812 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8813 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8814 					    CPUFREQ_TRANSITION_NOTIFIER);
8815 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8816 #ifdef CONFIG_X86_64
8817 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8818 	irq_work_sync(&pvclock_irq_work);
8819 	cancel_work_sync(&pvclock_gtod_work);
8820 #endif
8821 	kvm_x86_ops.hardware_enable = NULL;
8822 	kvm_mmu_module_exit();
8823 	free_percpu(user_return_msrs);
8824 	kmem_cache_destroy(x86_emulator_cache);
8825 #ifdef CONFIG_KVM_XEN
8826 	static_key_deferred_flush(&kvm_xen_enabled);
8827 	WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8828 #endif
8829 }
8830 
8831 static int __kvm_emulate_halt(struct kvm_vcpu *vcpu, int state, int reason)
8832 {
8833 	/*
8834 	 * The vCPU has halted, e.g. executed HLT.  Update the run state if the
8835 	 * local APIC is in-kernel, the run loop will detect the non-runnable
8836 	 * state and halt the vCPU.  Exit to userspace if the local APIC is
8837 	 * managed by userspace, in which case userspace is responsible for
8838 	 * handling wake events.
8839 	 */
8840 	++vcpu->stat.halt_exits;
8841 	if (lapic_in_kernel(vcpu)) {
8842 		vcpu->arch.mp_state = state;
8843 		return 1;
8844 	} else {
8845 		vcpu->run->exit_reason = reason;
8846 		return 0;
8847 	}
8848 }
8849 
8850 int kvm_emulate_halt_noskip(struct kvm_vcpu *vcpu)
8851 {
8852 	return __kvm_emulate_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8853 }
8854 EXPORT_SYMBOL_GPL(kvm_emulate_halt_noskip);
8855 
8856 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8857 {
8858 	int ret = kvm_skip_emulated_instruction(vcpu);
8859 	/*
8860 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8861 	 * KVM_EXIT_DEBUG here.
8862 	 */
8863 	return kvm_emulate_halt_noskip(vcpu) && ret;
8864 }
8865 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8866 
8867 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8868 {
8869 	int ret = kvm_skip_emulated_instruction(vcpu);
8870 
8871 	return __kvm_emulate_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD,
8872 					KVM_EXIT_AP_RESET_HOLD) && ret;
8873 }
8874 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8875 
8876 #ifdef CONFIG_X86_64
8877 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8878 			        unsigned long clock_type)
8879 {
8880 	struct kvm_clock_pairing clock_pairing;
8881 	struct timespec64 ts;
8882 	u64 cycle;
8883 	int ret;
8884 
8885 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8886 		return -KVM_EOPNOTSUPP;
8887 
8888 	if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8889 		return -KVM_EOPNOTSUPP;
8890 
8891 	clock_pairing.sec = ts.tv_sec;
8892 	clock_pairing.nsec = ts.tv_nsec;
8893 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8894 	clock_pairing.flags = 0;
8895 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8896 
8897 	ret = 0;
8898 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8899 			    sizeof(struct kvm_clock_pairing)))
8900 		ret = -KVM_EFAULT;
8901 
8902 	return ret;
8903 }
8904 #endif
8905 
8906 /*
8907  * kvm_pv_kick_cpu_op:  Kick a vcpu.
8908  *
8909  * @apicid - apicid of vcpu to be kicked.
8910  */
8911 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8912 {
8913 	struct kvm_lapic_irq lapic_irq;
8914 
8915 	lapic_irq.shorthand = APIC_DEST_NOSHORT;
8916 	lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8917 	lapic_irq.level = 0;
8918 	lapic_irq.dest_id = apicid;
8919 	lapic_irq.msi_redir_hint = false;
8920 
8921 	lapic_irq.delivery_mode = APIC_DM_REMRD;
8922 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8923 }
8924 
8925 bool kvm_apicv_activated(struct kvm *kvm)
8926 {
8927 	return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8928 }
8929 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8930 
8931 static void kvm_apicv_init(struct kvm *kvm)
8932 {
8933 	init_rwsem(&kvm->arch.apicv_update_lock);
8934 
8935 	set_bit(APICV_INHIBIT_REASON_ABSENT,
8936 		&kvm->arch.apicv_inhibit_reasons);
8937 	if (!enable_apicv)
8938 		set_bit(APICV_INHIBIT_REASON_DISABLE,
8939 			&kvm->arch.apicv_inhibit_reasons);
8940 }
8941 
8942 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8943 {
8944 	struct kvm_vcpu *target = NULL;
8945 	struct kvm_apic_map *map;
8946 
8947 	vcpu->stat.directed_yield_attempted++;
8948 
8949 	if (single_task_running())
8950 		goto no_yield;
8951 
8952 	rcu_read_lock();
8953 	map = rcu_dereference(vcpu->kvm->arch.apic_map);
8954 
8955 	if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8956 		target = map->phys_map[dest_id]->vcpu;
8957 
8958 	rcu_read_unlock();
8959 
8960 	if (!target || !READ_ONCE(target->ready))
8961 		goto no_yield;
8962 
8963 	/* Ignore requests to yield to self */
8964 	if (vcpu == target)
8965 		goto no_yield;
8966 
8967 	if (kvm_vcpu_yield_to(target) <= 0)
8968 		goto no_yield;
8969 
8970 	vcpu->stat.directed_yield_successful++;
8971 
8972 no_yield:
8973 	return;
8974 }
8975 
8976 static int complete_hypercall_exit(struct kvm_vcpu *vcpu)
8977 {
8978 	u64 ret = vcpu->run->hypercall.ret;
8979 
8980 	if (!is_64_bit_mode(vcpu))
8981 		ret = (u32)ret;
8982 	kvm_rax_write(vcpu, ret);
8983 	++vcpu->stat.hypercalls;
8984 	return kvm_skip_emulated_instruction(vcpu);
8985 }
8986 
8987 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8988 {
8989 	unsigned long nr, a0, a1, a2, a3, ret;
8990 	int op_64_bit;
8991 
8992 	if (kvm_xen_hypercall_enabled(vcpu->kvm))
8993 		return kvm_xen_hypercall(vcpu);
8994 
8995 	if (kvm_hv_hypercall_enabled(vcpu))
8996 		return kvm_hv_hypercall(vcpu);
8997 
8998 	nr = kvm_rax_read(vcpu);
8999 	a0 = kvm_rbx_read(vcpu);
9000 	a1 = kvm_rcx_read(vcpu);
9001 	a2 = kvm_rdx_read(vcpu);
9002 	a3 = kvm_rsi_read(vcpu);
9003 
9004 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
9005 
9006 	op_64_bit = is_64_bit_hypercall(vcpu);
9007 	if (!op_64_bit) {
9008 		nr &= 0xFFFFFFFF;
9009 		a0 &= 0xFFFFFFFF;
9010 		a1 &= 0xFFFFFFFF;
9011 		a2 &= 0xFFFFFFFF;
9012 		a3 &= 0xFFFFFFFF;
9013 	}
9014 
9015 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
9016 		ret = -KVM_EPERM;
9017 		goto out;
9018 	}
9019 
9020 	ret = -KVM_ENOSYS;
9021 
9022 	switch (nr) {
9023 	case KVM_HC_VAPIC_POLL_IRQ:
9024 		ret = 0;
9025 		break;
9026 	case KVM_HC_KICK_CPU:
9027 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
9028 			break;
9029 
9030 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
9031 		kvm_sched_yield(vcpu, a1);
9032 		ret = 0;
9033 		break;
9034 #ifdef CONFIG_X86_64
9035 	case KVM_HC_CLOCK_PAIRING:
9036 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
9037 		break;
9038 #endif
9039 	case KVM_HC_SEND_IPI:
9040 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
9041 			break;
9042 
9043 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
9044 		break;
9045 	case KVM_HC_SCHED_YIELD:
9046 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
9047 			break;
9048 
9049 		kvm_sched_yield(vcpu, a0);
9050 		ret = 0;
9051 		break;
9052 	case KVM_HC_MAP_GPA_RANGE: {
9053 		u64 gpa = a0, npages = a1, attrs = a2;
9054 
9055 		ret = -KVM_ENOSYS;
9056 		if (!(vcpu->kvm->arch.hypercall_exit_enabled & (1 << KVM_HC_MAP_GPA_RANGE)))
9057 			break;
9058 
9059 		if (!PAGE_ALIGNED(gpa) || !npages ||
9060 		    gpa_to_gfn(gpa) + npages <= gpa_to_gfn(gpa)) {
9061 			ret = -KVM_EINVAL;
9062 			break;
9063 		}
9064 
9065 		vcpu->run->exit_reason        = KVM_EXIT_HYPERCALL;
9066 		vcpu->run->hypercall.nr       = KVM_HC_MAP_GPA_RANGE;
9067 		vcpu->run->hypercall.args[0]  = gpa;
9068 		vcpu->run->hypercall.args[1]  = npages;
9069 		vcpu->run->hypercall.args[2]  = attrs;
9070 		vcpu->run->hypercall.longmode = op_64_bit;
9071 		vcpu->arch.complete_userspace_io = complete_hypercall_exit;
9072 		return 0;
9073 	}
9074 	default:
9075 		ret = -KVM_ENOSYS;
9076 		break;
9077 	}
9078 out:
9079 	if (!op_64_bit)
9080 		ret = (u32)ret;
9081 	kvm_rax_write(vcpu, ret);
9082 
9083 	++vcpu->stat.hypercalls;
9084 	return kvm_skip_emulated_instruction(vcpu);
9085 }
9086 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
9087 
9088 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
9089 {
9090 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
9091 	char instruction[3];
9092 	unsigned long rip = kvm_rip_read(vcpu);
9093 
9094 	static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
9095 
9096 	return emulator_write_emulated(ctxt, rip, instruction, 3,
9097 		&ctxt->exception);
9098 }
9099 
9100 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
9101 {
9102 	return vcpu->run->request_interrupt_window &&
9103 		likely(!pic_in_kernel(vcpu->kvm));
9104 }
9105 
9106 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
9107 {
9108 	struct kvm_run *kvm_run = vcpu->run;
9109 
9110 	kvm_run->if_flag = static_call(kvm_x86_get_if_flag)(vcpu);
9111 	kvm_run->cr8 = kvm_get_cr8(vcpu);
9112 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
9113 
9114 	/*
9115 	 * The call to kvm_ready_for_interrupt_injection() may end up in
9116 	 * kvm_xen_has_interrupt() which may require the srcu lock to be
9117 	 * held, to protect against changes in the vcpu_info address.
9118 	 */
9119 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9120 	kvm_run->ready_for_interrupt_injection =
9121 		pic_in_kernel(vcpu->kvm) ||
9122 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
9123 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9124 
9125 	if (is_smm(vcpu))
9126 		kvm_run->flags |= KVM_RUN_X86_SMM;
9127 }
9128 
9129 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
9130 {
9131 	int max_irr, tpr;
9132 
9133 	if (!kvm_x86_ops.update_cr8_intercept)
9134 		return;
9135 
9136 	if (!lapic_in_kernel(vcpu))
9137 		return;
9138 
9139 	if (vcpu->arch.apicv_active)
9140 		return;
9141 
9142 	if (!vcpu->arch.apic->vapic_addr)
9143 		max_irr = kvm_lapic_find_highest_irr(vcpu);
9144 	else
9145 		max_irr = -1;
9146 
9147 	if (max_irr != -1)
9148 		max_irr >>= 4;
9149 
9150 	tpr = kvm_lapic_get_cr8(vcpu);
9151 
9152 	static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
9153 }
9154 
9155 
9156 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
9157 {
9158 	if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9159 		kvm_x86_ops.nested_ops->triple_fault(vcpu);
9160 		return 1;
9161 	}
9162 
9163 	return kvm_x86_ops.nested_ops->check_events(vcpu);
9164 }
9165 
9166 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
9167 {
9168 	if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
9169 		vcpu->arch.exception.error_code = false;
9170 	static_call(kvm_x86_queue_exception)(vcpu);
9171 }
9172 
9173 static int inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
9174 {
9175 	int r;
9176 	bool can_inject = true;
9177 
9178 	/* try to reinject previous events if any */
9179 
9180 	if (vcpu->arch.exception.injected) {
9181 		kvm_inject_exception(vcpu);
9182 		can_inject = false;
9183 	}
9184 	/*
9185 	 * Do not inject an NMI or interrupt if there is a pending
9186 	 * exception.  Exceptions and interrupts are recognized at
9187 	 * instruction boundaries, i.e. the start of an instruction.
9188 	 * Trap-like exceptions, e.g. #DB, have higher priority than
9189 	 * NMIs and interrupts, i.e. traps are recognized before an
9190 	 * NMI/interrupt that's pending on the same instruction.
9191 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
9192 	 * priority, but are only generated (pended) during instruction
9193 	 * execution, i.e. a pending fault-like exception means the
9194 	 * fault occurred on the *previous* instruction and must be
9195 	 * serviced prior to recognizing any new events in order to
9196 	 * fully complete the previous instruction.
9197 	 */
9198 	else if (!vcpu->arch.exception.pending) {
9199 		if (vcpu->arch.nmi_injected) {
9200 			static_call(kvm_x86_set_nmi)(vcpu);
9201 			can_inject = false;
9202 		} else if (vcpu->arch.interrupt.injected) {
9203 			static_call(kvm_x86_set_irq)(vcpu);
9204 			can_inject = false;
9205 		}
9206 	}
9207 
9208 	WARN_ON_ONCE(vcpu->arch.exception.injected &&
9209 		     vcpu->arch.exception.pending);
9210 
9211 	/*
9212 	 * Call check_nested_events() even if we reinjected a previous event
9213 	 * in order for caller to determine if it should require immediate-exit
9214 	 * from L2 to L1 due to pending L1 events which require exit
9215 	 * from L2 to L1.
9216 	 */
9217 	if (is_guest_mode(vcpu)) {
9218 		r = kvm_check_nested_events(vcpu);
9219 		if (r < 0)
9220 			goto out;
9221 	}
9222 
9223 	/* try to inject new event if pending */
9224 	if (vcpu->arch.exception.pending) {
9225 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
9226 					vcpu->arch.exception.has_error_code,
9227 					vcpu->arch.exception.error_code);
9228 
9229 		vcpu->arch.exception.pending = false;
9230 		vcpu->arch.exception.injected = true;
9231 
9232 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
9233 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
9234 					     X86_EFLAGS_RF);
9235 
9236 		if (vcpu->arch.exception.nr == DB_VECTOR) {
9237 			kvm_deliver_exception_payload(vcpu);
9238 			if (vcpu->arch.dr7 & DR7_GD) {
9239 				vcpu->arch.dr7 &= ~DR7_GD;
9240 				kvm_update_dr7(vcpu);
9241 			}
9242 		}
9243 
9244 		kvm_inject_exception(vcpu);
9245 		can_inject = false;
9246 	}
9247 
9248 	/* Don't inject interrupts if the user asked to avoid doing so */
9249 	if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ)
9250 		return 0;
9251 
9252 	/*
9253 	 * Finally, inject interrupt events.  If an event cannot be injected
9254 	 * due to architectural conditions (e.g. IF=0) a window-open exit
9255 	 * will re-request KVM_REQ_EVENT.  Sometimes however an event is pending
9256 	 * and can architecturally be injected, but we cannot do it right now:
9257 	 * an interrupt could have arrived just now and we have to inject it
9258 	 * as a vmexit, or there could already an event in the queue, which is
9259 	 * indicated by can_inject.  In that case we request an immediate exit
9260 	 * in order to make progress and get back here for another iteration.
9261 	 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
9262 	 */
9263 	if (vcpu->arch.smi_pending) {
9264 		r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
9265 		if (r < 0)
9266 			goto out;
9267 		if (r) {
9268 			vcpu->arch.smi_pending = false;
9269 			++vcpu->arch.smi_count;
9270 			enter_smm(vcpu);
9271 			can_inject = false;
9272 		} else
9273 			static_call(kvm_x86_enable_smi_window)(vcpu);
9274 	}
9275 
9276 	if (vcpu->arch.nmi_pending) {
9277 		r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
9278 		if (r < 0)
9279 			goto out;
9280 		if (r) {
9281 			--vcpu->arch.nmi_pending;
9282 			vcpu->arch.nmi_injected = true;
9283 			static_call(kvm_x86_set_nmi)(vcpu);
9284 			can_inject = false;
9285 			WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
9286 		}
9287 		if (vcpu->arch.nmi_pending)
9288 			static_call(kvm_x86_enable_nmi_window)(vcpu);
9289 	}
9290 
9291 	if (kvm_cpu_has_injectable_intr(vcpu)) {
9292 		r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
9293 		if (r < 0)
9294 			goto out;
9295 		if (r) {
9296 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
9297 			static_call(kvm_x86_set_irq)(vcpu);
9298 			WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
9299 		}
9300 		if (kvm_cpu_has_injectable_intr(vcpu))
9301 			static_call(kvm_x86_enable_irq_window)(vcpu);
9302 	}
9303 
9304 	if (is_guest_mode(vcpu) &&
9305 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
9306 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
9307 		*req_immediate_exit = true;
9308 
9309 	WARN_ON(vcpu->arch.exception.pending);
9310 	return 0;
9311 
9312 out:
9313 	if (r == -EBUSY) {
9314 		*req_immediate_exit = true;
9315 		r = 0;
9316 	}
9317 	return r;
9318 }
9319 
9320 static void process_nmi(struct kvm_vcpu *vcpu)
9321 {
9322 	unsigned limit = 2;
9323 
9324 	/*
9325 	 * x86 is limited to one NMI running, and one NMI pending after it.
9326 	 * If an NMI is already in progress, limit further NMIs to just one.
9327 	 * Otherwise, allow two (and we'll inject the first one immediately).
9328 	 */
9329 	if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
9330 		limit = 1;
9331 
9332 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
9333 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
9334 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9335 }
9336 
9337 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
9338 {
9339 	u32 flags = 0;
9340 	flags |= seg->g       << 23;
9341 	flags |= seg->db      << 22;
9342 	flags |= seg->l       << 21;
9343 	flags |= seg->avl     << 20;
9344 	flags |= seg->present << 15;
9345 	flags |= seg->dpl     << 13;
9346 	flags |= seg->s       << 12;
9347 	flags |= seg->type    << 8;
9348 	return flags;
9349 }
9350 
9351 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
9352 {
9353 	struct kvm_segment seg;
9354 	int offset;
9355 
9356 	kvm_get_segment(vcpu, &seg, n);
9357 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
9358 
9359 	if (n < 3)
9360 		offset = 0x7f84 + n * 12;
9361 	else
9362 		offset = 0x7f2c + (n - 3) * 12;
9363 
9364 	put_smstate(u32, buf, offset + 8, seg.base);
9365 	put_smstate(u32, buf, offset + 4, seg.limit);
9366 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
9367 }
9368 
9369 #ifdef CONFIG_X86_64
9370 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
9371 {
9372 	struct kvm_segment seg;
9373 	int offset;
9374 	u16 flags;
9375 
9376 	kvm_get_segment(vcpu, &seg, n);
9377 	offset = 0x7e00 + n * 16;
9378 
9379 	flags = enter_smm_get_segment_flags(&seg) >> 8;
9380 	put_smstate(u16, buf, offset, seg.selector);
9381 	put_smstate(u16, buf, offset + 2, flags);
9382 	put_smstate(u32, buf, offset + 4, seg.limit);
9383 	put_smstate(u64, buf, offset + 8, seg.base);
9384 }
9385 #endif
9386 
9387 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
9388 {
9389 	struct desc_ptr dt;
9390 	struct kvm_segment seg;
9391 	unsigned long val;
9392 	int i;
9393 
9394 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
9395 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
9396 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
9397 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
9398 
9399 	for (i = 0; i < 8; i++)
9400 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
9401 
9402 	kvm_get_dr(vcpu, 6, &val);
9403 	put_smstate(u32, buf, 0x7fcc, (u32)val);
9404 	kvm_get_dr(vcpu, 7, &val);
9405 	put_smstate(u32, buf, 0x7fc8, (u32)val);
9406 
9407 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9408 	put_smstate(u32, buf, 0x7fc4, seg.selector);
9409 	put_smstate(u32, buf, 0x7f64, seg.base);
9410 	put_smstate(u32, buf, 0x7f60, seg.limit);
9411 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
9412 
9413 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9414 	put_smstate(u32, buf, 0x7fc0, seg.selector);
9415 	put_smstate(u32, buf, 0x7f80, seg.base);
9416 	put_smstate(u32, buf, 0x7f7c, seg.limit);
9417 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
9418 
9419 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
9420 	put_smstate(u32, buf, 0x7f74, dt.address);
9421 	put_smstate(u32, buf, 0x7f70, dt.size);
9422 
9423 	static_call(kvm_x86_get_idt)(vcpu, &dt);
9424 	put_smstate(u32, buf, 0x7f58, dt.address);
9425 	put_smstate(u32, buf, 0x7f54, dt.size);
9426 
9427 	for (i = 0; i < 6; i++)
9428 		enter_smm_save_seg_32(vcpu, buf, i);
9429 
9430 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
9431 
9432 	/* revision id */
9433 	put_smstate(u32, buf, 0x7efc, 0x00020000);
9434 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
9435 }
9436 
9437 #ifdef CONFIG_X86_64
9438 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
9439 {
9440 	struct desc_ptr dt;
9441 	struct kvm_segment seg;
9442 	unsigned long val;
9443 	int i;
9444 
9445 	for (i = 0; i < 16; i++)
9446 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
9447 
9448 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
9449 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
9450 
9451 	kvm_get_dr(vcpu, 6, &val);
9452 	put_smstate(u64, buf, 0x7f68, val);
9453 	kvm_get_dr(vcpu, 7, &val);
9454 	put_smstate(u64, buf, 0x7f60, val);
9455 
9456 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
9457 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
9458 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
9459 
9460 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
9461 
9462 	/* revision id */
9463 	put_smstate(u32, buf, 0x7efc, 0x00020064);
9464 
9465 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
9466 
9467 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9468 	put_smstate(u16, buf, 0x7e90, seg.selector);
9469 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
9470 	put_smstate(u32, buf, 0x7e94, seg.limit);
9471 	put_smstate(u64, buf, 0x7e98, seg.base);
9472 
9473 	static_call(kvm_x86_get_idt)(vcpu, &dt);
9474 	put_smstate(u32, buf, 0x7e84, dt.size);
9475 	put_smstate(u64, buf, 0x7e88, dt.address);
9476 
9477 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9478 	put_smstate(u16, buf, 0x7e70, seg.selector);
9479 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
9480 	put_smstate(u32, buf, 0x7e74, seg.limit);
9481 	put_smstate(u64, buf, 0x7e78, seg.base);
9482 
9483 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
9484 	put_smstate(u32, buf, 0x7e64, dt.size);
9485 	put_smstate(u64, buf, 0x7e68, dt.address);
9486 
9487 	for (i = 0; i < 6; i++)
9488 		enter_smm_save_seg_64(vcpu, buf, i);
9489 }
9490 #endif
9491 
9492 static void enter_smm(struct kvm_vcpu *vcpu)
9493 {
9494 	struct kvm_segment cs, ds;
9495 	struct desc_ptr dt;
9496 	unsigned long cr0;
9497 	char buf[512];
9498 
9499 	memset(buf, 0, 512);
9500 #ifdef CONFIG_X86_64
9501 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9502 		enter_smm_save_state_64(vcpu, buf);
9503 	else
9504 #endif
9505 		enter_smm_save_state_32(vcpu, buf);
9506 
9507 	/*
9508 	 * Give enter_smm() a chance to make ISA-specific changes to the vCPU
9509 	 * state (e.g. leave guest mode) after we've saved the state into the
9510 	 * SMM state-save area.
9511 	 */
9512 	static_call(kvm_x86_enter_smm)(vcpu, buf);
9513 
9514 	kvm_smm_changed(vcpu, true);
9515 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
9516 
9517 	if (static_call(kvm_x86_get_nmi_mask)(vcpu))
9518 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
9519 	else
9520 		static_call(kvm_x86_set_nmi_mask)(vcpu, true);
9521 
9522 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
9523 	kvm_rip_write(vcpu, 0x8000);
9524 
9525 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
9526 	static_call(kvm_x86_set_cr0)(vcpu, cr0);
9527 	vcpu->arch.cr0 = cr0;
9528 
9529 	static_call(kvm_x86_set_cr4)(vcpu, 0);
9530 
9531 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
9532 	dt.address = dt.size = 0;
9533 	static_call(kvm_x86_set_idt)(vcpu, &dt);
9534 
9535 	kvm_set_dr(vcpu, 7, DR7_FIXED_1);
9536 
9537 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
9538 	cs.base = vcpu->arch.smbase;
9539 
9540 	ds.selector = 0;
9541 	ds.base = 0;
9542 
9543 	cs.limit    = ds.limit = 0xffffffff;
9544 	cs.type     = ds.type = 0x3;
9545 	cs.dpl      = ds.dpl = 0;
9546 	cs.db       = ds.db = 0;
9547 	cs.s        = ds.s = 1;
9548 	cs.l        = ds.l = 0;
9549 	cs.g        = ds.g = 1;
9550 	cs.avl      = ds.avl = 0;
9551 	cs.present  = ds.present = 1;
9552 	cs.unusable = ds.unusable = 0;
9553 	cs.padding  = ds.padding = 0;
9554 
9555 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9556 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
9557 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
9558 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
9559 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
9560 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
9561 
9562 #ifdef CONFIG_X86_64
9563 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9564 		static_call(kvm_x86_set_efer)(vcpu, 0);
9565 #endif
9566 
9567 	kvm_update_cpuid_runtime(vcpu);
9568 	kvm_mmu_reset_context(vcpu);
9569 }
9570 
9571 static void process_smi(struct kvm_vcpu *vcpu)
9572 {
9573 	vcpu->arch.smi_pending = true;
9574 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9575 }
9576 
9577 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
9578 				       unsigned long *vcpu_bitmap)
9579 {
9580 	kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC, vcpu_bitmap);
9581 }
9582 
9583 void kvm_make_scan_ioapic_request(struct kvm *kvm)
9584 {
9585 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
9586 }
9587 
9588 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
9589 {
9590 	bool activate;
9591 
9592 	if (!lapic_in_kernel(vcpu))
9593 		return;
9594 
9595 	down_read(&vcpu->kvm->arch.apicv_update_lock);
9596 
9597 	activate = kvm_apicv_activated(vcpu->kvm);
9598 	if (vcpu->arch.apicv_active == activate)
9599 		goto out;
9600 
9601 	vcpu->arch.apicv_active = activate;
9602 	kvm_apic_update_apicv(vcpu);
9603 	static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
9604 
9605 	/*
9606 	 * When APICv gets disabled, we may still have injected interrupts
9607 	 * pending. At the same time, KVM_REQ_EVENT may not be set as APICv was
9608 	 * still active when the interrupt got accepted. Make sure
9609 	 * inject_pending_event() is called to check for that.
9610 	 */
9611 	if (!vcpu->arch.apicv_active)
9612 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9613 
9614 out:
9615 	up_read(&vcpu->kvm->arch.apicv_update_lock);
9616 }
9617 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
9618 
9619 void __kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9620 {
9621 	unsigned long old, new;
9622 
9623 	lockdep_assert_held_write(&kvm->arch.apicv_update_lock);
9624 
9625 	if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
9626 	    !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
9627 		return;
9628 
9629 	old = new = kvm->arch.apicv_inhibit_reasons;
9630 
9631 	if (activate)
9632 		__clear_bit(bit, &new);
9633 	else
9634 		__set_bit(bit, &new);
9635 
9636 	if (!!old != !!new) {
9637 		trace_kvm_apicv_update_request(activate, bit);
9638 		/*
9639 		 * Kick all vCPUs before setting apicv_inhibit_reasons to avoid
9640 		 * false positives in the sanity check WARN in svm_vcpu_run().
9641 		 * This task will wait for all vCPUs to ack the kick IRQ before
9642 		 * updating apicv_inhibit_reasons, and all other vCPUs will
9643 		 * block on acquiring apicv_update_lock so that vCPUs can't
9644 		 * redo svm_vcpu_run() without seeing the new inhibit state.
9645 		 *
9646 		 * Note, holding apicv_update_lock and taking it in the read
9647 		 * side (handling the request) also prevents other vCPUs from
9648 		 * servicing the request with a stale apicv_inhibit_reasons.
9649 		 */
9650 		kvm_make_all_cpus_request(kvm, KVM_REQ_APICV_UPDATE);
9651 		kvm->arch.apicv_inhibit_reasons = new;
9652 		if (new) {
9653 			unsigned long gfn = gpa_to_gfn(APIC_DEFAULT_PHYS_BASE);
9654 			kvm_zap_gfn_range(kvm, gfn, gfn+1);
9655 		}
9656 	} else
9657 		kvm->arch.apicv_inhibit_reasons = new;
9658 }
9659 EXPORT_SYMBOL_GPL(__kvm_request_apicv_update);
9660 
9661 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9662 {
9663 	down_write(&kvm->arch.apicv_update_lock);
9664 	__kvm_request_apicv_update(kvm, activate, bit);
9665 	up_write(&kvm->arch.apicv_update_lock);
9666 }
9667 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
9668 
9669 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
9670 {
9671 	if (!kvm_apic_present(vcpu))
9672 		return;
9673 
9674 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
9675 
9676 	if (irqchip_split(vcpu->kvm))
9677 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
9678 	else {
9679 		static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
9680 		if (ioapic_in_kernel(vcpu->kvm))
9681 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
9682 	}
9683 
9684 	if (is_guest_mode(vcpu))
9685 		vcpu->arch.load_eoi_exitmap_pending = true;
9686 	else
9687 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
9688 }
9689 
9690 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
9691 {
9692 	u64 eoi_exit_bitmap[4];
9693 
9694 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
9695 		return;
9696 
9697 	if (to_hv_vcpu(vcpu)) {
9698 		bitmap_or((ulong *)eoi_exit_bitmap,
9699 			  vcpu->arch.ioapic_handled_vectors,
9700 			  to_hv_synic(vcpu)->vec_bitmap, 256);
9701 		static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
9702 		return;
9703 	}
9704 
9705 	static_call(kvm_x86_load_eoi_exitmap)(
9706 		vcpu, (u64 *)vcpu->arch.ioapic_handled_vectors);
9707 }
9708 
9709 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
9710 					    unsigned long start, unsigned long end)
9711 {
9712 	unsigned long apic_address;
9713 
9714 	/*
9715 	 * The physical address of apic access page is stored in the VMCS.
9716 	 * Update it when it becomes invalid.
9717 	 */
9718 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9719 	if (start <= apic_address && apic_address < end)
9720 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9721 }
9722 
9723 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9724 {
9725 	if (!lapic_in_kernel(vcpu))
9726 		return;
9727 
9728 	if (!kvm_x86_ops.set_apic_access_page_addr)
9729 		return;
9730 
9731 	static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9732 }
9733 
9734 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9735 {
9736 	smp_send_reschedule(vcpu->cpu);
9737 }
9738 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9739 
9740 /*
9741  * Returns 1 to let vcpu_run() continue the guest execution loop without
9742  * exiting to the userspace.  Otherwise, the value will be returned to the
9743  * userspace.
9744  */
9745 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9746 {
9747 	int r;
9748 	bool req_int_win =
9749 		dm_request_for_irq_injection(vcpu) &&
9750 		kvm_cpu_accept_dm_intr(vcpu);
9751 	fastpath_t exit_fastpath;
9752 
9753 	bool req_immediate_exit = false;
9754 
9755 	/* Forbid vmenter if vcpu dirty ring is soft-full */
9756 	if (unlikely(vcpu->kvm->dirty_ring_size &&
9757 		     kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9758 		vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9759 		trace_kvm_dirty_ring_exit(vcpu);
9760 		r = 0;
9761 		goto out;
9762 	}
9763 
9764 	if (kvm_request_pending(vcpu)) {
9765 		if (kvm_check_request(KVM_REQ_VM_DEAD, vcpu)) {
9766 			r = -EIO;
9767 			goto out;
9768 		}
9769 		if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9770 			if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9771 				r = 0;
9772 				goto out;
9773 			}
9774 		}
9775 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9776 			kvm_mmu_unload(vcpu);
9777 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9778 			__kvm_migrate_timers(vcpu);
9779 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9780 			kvm_update_masterclock(vcpu->kvm);
9781 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9782 			kvm_gen_kvmclock_update(vcpu);
9783 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9784 			r = kvm_guest_time_update(vcpu);
9785 			if (unlikely(r))
9786 				goto out;
9787 		}
9788 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9789 			kvm_mmu_sync_roots(vcpu);
9790 		if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9791 			kvm_mmu_load_pgd(vcpu);
9792 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9793 			kvm_vcpu_flush_tlb_all(vcpu);
9794 
9795 			/* Flushing all ASIDs flushes the current ASID... */
9796 			kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9797 		}
9798 		kvm_service_local_tlb_flush_requests(vcpu);
9799 
9800 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9801 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9802 			r = 0;
9803 			goto out;
9804 		}
9805 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9806 			if (is_guest_mode(vcpu)) {
9807 				kvm_x86_ops.nested_ops->triple_fault(vcpu);
9808 			} else {
9809 				vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9810 				vcpu->mmio_needed = 0;
9811 				r = 0;
9812 				goto out;
9813 			}
9814 		}
9815 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9816 			/* Page is swapped out. Do synthetic halt */
9817 			vcpu->arch.apf.halted = true;
9818 			r = 1;
9819 			goto out;
9820 		}
9821 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9822 			record_steal_time(vcpu);
9823 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
9824 			process_smi(vcpu);
9825 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
9826 			process_nmi(vcpu);
9827 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
9828 			kvm_pmu_handle_event(vcpu);
9829 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
9830 			kvm_pmu_deliver_pmi(vcpu);
9831 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9832 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9833 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
9834 				     vcpu->arch.ioapic_handled_vectors)) {
9835 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9836 				vcpu->run->eoi.vector =
9837 						vcpu->arch.pending_ioapic_eoi;
9838 				r = 0;
9839 				goto out;
9840 			}
9841 		}
9842 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9843 			vcpu_scan_ioapic(vcpu);
9844 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9845 			vcpu_load_eoi_exitmap(vcpu);
9846 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9847 			kvm_vcpu_reload_apic_access_page(vcpu);
9848 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9849 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9850 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9851 			r = 0;
9852 			goto out;
9853 		}
9854 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9855 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9856 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9857 			r = 0;
9858 			goto out;
9859 		}
9860 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9861 			struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9862 
9863 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9864 			vcpu->run->hyperv = hv_vcpu->exit;
9865 			r = 0;
9866 			goto out;
9867 		}
9868 
9869 		/*
9870 		 * KVM_REQ_HV_STIMER has to be processed after
9871 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9872 		 * depend on the guest clock being up-to-date
9873 		 */
9874 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9875 			kvm_hv_process_stimers(vcpu);
9876 		if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9877 			kvm_vcpu_update_apicv(vcpu);
9878 		if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9879 			kvm_check_async_pf_completion(vcpu);
9880 		if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9881 			static_call(kvm_x86_msr_filter_changed)(vcpu);
9882 
9883 		if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9884 			static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9885 	}
9886 
9887 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9888 	    kvm_xen_has_interrupt(vcpu)) {
9889 		++vcpu->stat.req_event;
9890 		r = kvm_apic_accept_events(vcpu);
9891 		if (r < 0) {
9892 			r = 0;
9893 			goto out;
9894 		}
9895 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9896 			r = 1;
9897 			goto out;
9898 		}
9899 
9900 		r = inject_pending_event(vcpu, &req_immediate_exit);
9901 		if (r < 0) {
9902 			r = 0;
9903 			goto out;
9904 		}
9905 		if (req_int_win)
9906 			static_call(kvm_x86_enable_irq_window)(vcpu);
9907 
9908 		if (kvm_lapic_enabled(vcpu)) {
9909 			update_cr8_intercept(vcpu);
9910 			kvm_lapic_sync_to_vapic(vcpu);
9911 		}
9912 	}
9913 
9914 	r = kvm_mmu_reload(vcpu);
9915 	if (unlikely(r)) {
9916 		goto cancel_injection;
9917 	}
9918 
9919 	preempt_disable();
9920 
9921 	static_call(kvm_x86_prepare_guest_switch)(vcpu);
9922 
9923 	/*
9924 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
9925 	 * IPI are then delayed after guest entry, which ensures that they
9926 	 * result in virtual interrupt delivery.
9927 	 */
9928 	local_irq_disable();
9929 	vcpu->mode = IN_GUEST_MODE;
9930 
9931 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9932 
9933 	/*
9934 	 * 1) We should set ->mode before checking ->requests.  Please see
9935 	 * the comment in kvm_vcpu_exiting_guest_mode().
9936 	 *
9937 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
9938 	 * pairs with the memory barrier implicit in pi_test_and_set_on
9939 	 * (see vmx_deliver_posted_interrupt).
9940 	 *
9941 	 * 3) This also orders the write to mode from any reads to the page
9942 	 * tables done while the VCPU is running.  Please see the comment
9943 	 * in kvm_flush_remote_tlbs.
9944 	 */
9945 	smp_mb__after_srcu_read_unlock();
9946 
9947 	/*
9948 	 * This handles the case where a posted interrupt was
9949 	 * notified with kvm_vcpu_kick.  Assigned devices can
9950 	 * use the POSTED_INTR_VECTOR even if APICv is disabled,
9951 	 * so do it even if APICv is disabled on this vCPU.
9952 	 */
9953 	if (kvm_lapic_enabled(vcpu))
9954 		static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
9955 
9956 	if (kvm_vcpu_exit_request(vcpu)) {
9957 		vcpu->mode = OUTSIDE_GUEST_MODE;
9958 		smp_wmb();
9959 		local_irq_enable();
9960 		preempt_enable();
9961 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9962 		r = 1;
9963 		goto cancel_injection;
9964 	}
9965 
9966 	if (req_immediate_exit) {
9967 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9968 		static_call(kvm_x86_request_immediate_exit)(vcpu);
9969 	}
9970 
9971 	fpregs_assert_state_consistent();
9972 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
9973 		switch_fpu_return();
9974 
9975 	if (vcpu->arch.guest_fpu.xfd_err)
9976 		wrmsrl(MSR_IA32_XFD_ERR, vcpu->arch.guest_fpu.xfd_err);
9977 
9978 	if (unlikely(vcpu->arch.switch_db_regs)) {
9979 		set_debugreg(0, 7);
9980 		set_debugreg(vcpu->arch.eff_db[0], 0);
9981 		set_debugreg(vcpu->arch.eff_db[1], 1);
9982 		set_debugreg(vcpu->arch.eff_db[2], 2);
9983 		set_debugreg(vcpu->arch.eff_db[3], 3);
9984 	} else if (unlikely(hw_breakpoint_active())) {
9985 		set_debugreg(0, 7);
9986 	}
9987 
9988 	for (;;) {
9989 		/*
9990 		 * Assert that vCPU vs. VM APICv state is consistent.  An APICv
9991 		 * update must kick and wait for all vCPUs before toggling the
9992 		 * per-VM state, and responsing vCPUs must wait for the update
9993 		 * to complete before servicing KVM_REQ_APICV_UPDATE.
9994 		 */
9995 		WARN_ON_ONCE(kvm_apicv_activated(vcpu->kvm) != kvm_vcpu_apicv_active(vcpu));
9996 
9997 		exit_fastpath = static_call(kvm_x86_run)(vcpu);
9998 		if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9999 			break;
10000 
10001 		if (kvm_lapic_enabled(vcpu))
10002 			static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
10003 
10004 		if (unlikely(kvm_vcpu_exit_request(vcpu))) {
10005 			exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
10006 			break;
10007 		}
10008 	}
10009 
10010 	/*
10011 	 * Do this here before restoring debug registers on the host.  And
10012 	 * since we do this before handling the vmexit, a DR access vmexit
10013 	 * can (a) read the correct value of the debug registers, (b) set
10014 	 * KVM_DEBUGREG_WONT_EXIT again.
10015 	 */
10016 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
10017 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
10018 		static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
10019 		kvm_update_dr0123(vcpu);
10020 		kvm_update_dr7(vcpu);
10021 	}
10022 
10023 	/*
10024 	 * If the guest has used debug registers, at least dr7
10025 	 * will be disabled while returning to the host.
10026 	 * If we don't have active breakpoints in the host, we don't
10027 	 * care about the messed up debug address registers. But if
10028 	 * we have some of them active, restore the old state.
10029 	 */
10030 	if (hw_breakpoint_active())
10031 		hw_breakpoint_restore();
10032 
10033 	vcpu->arch.last_vmentry_cpu = vcpu->cpu;
10034 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
10035 
10036 	vcpu->mode = OUTSIDE_GUEST_MODE;
10037 	smp_wmb();
10038 
10039 	/*
10040 	 * Sync xfd before calling handle_exit_irqoff() which may
10041 	 * rely on the fact that guest_fpu::xfd is up-to-date (e.g.
10042 	 * in #NM irqoff handler).
10043 	 */
10044 	if (vcpu->arch.xfd_no_write_intercept)
10045 		fpu_sync_guest_vmexit_xfd_state();
10046 
10047 	static_call(kvm_x86_handle_exit_irqoff)(vcpu);
10048 
10049 	if (vcpu->arch.guest_fpu.xfd_err)
10050 		wrmsrl(MSR_IA32_XFD_ERR, 0);
10051 
10052 	/*
10053 	 * Consume any pending interrupts, including the possible source of
10054 	 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
10055 	 * An instruction is required after local_irq_enable() to fully unblock
10056 	 * interrupts on processors that implement an interrupt shadow, the
10057 	 * stat.exits increment will do nicely.
10058 	 */
10059 	kvm_before_interrupt(vcpu, KVM_HANDLING_IRQ);
10060 	local_irq_enable();
10061 	++vcpu->stat.exits;
10062 	local_irq_disable();
10063 	kvm_after_interrupt(vcpu);
10064 
10065 	/*
10066 	 * Wait until after servicing IRQs to account guest time so that any
10067 	 * ticks that occurred while running the guest are properly accounted
10068 	 * to the guest.  Waiting until IRQs are enabled degrades the accuracy
10069 	 * of accounting via context tracking, but the loss of accuracy is
10070 	 * acceptable for all known use cases.
10071 	 */
10072 	vtime_account_guest_exit();
10073 
10074 	if (lapic_in_kernel(vcpu)) {
10075 		s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
10076 		if (delta != S64_MIN) {
10077 			trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
10078 			vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
10079 		}
10080 	}
10081 
10082 	local_irq_enable();
10083 	preempt_enable();
10084 
10085 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
10086 
10087 	/*
10088 	 * Profile KVM exit RIPs:
10089 	 */
10090 	if (unlikely(prof_on == KVM_PROFILING)) {
10091 		unsigned long rip = kvm_rip_read(vcpu);
10092 		profile_hit(KVM_PROFILING, (void *)rip);
10093 	}
10094 
10095 	if (unlikely(vcpu->arch.tsc_always_catchup))
10096 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10097 
10098 	if (vcpu->arch.apic_attention)
10099 		kvm_lapic_sync_from_vapic(vcpu);
10100 
10101 	r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
10102 	return r;
10103 
10104 cancel_injection:
10105 	if (req_immediate_exit)
10106 		kvm_make_request(KVM_REQ_EVENT, vcpu);
10107 	static_call(kvm_x86_cancel_injection)(vcpu);
10108 	if (unlikely(vcpu->arch.apic_attention))
10109 		kvm_lapic_sync_from_vapic(vcpu);
10110 out:
10111 	return r;
10112 }
10113 
10114 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
10115 {
10116 	if (!kvm_arch_vcpu_runnable(vcpu) &&
10117 	    (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
10118 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
10119 		if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
10120 			kvm_vcpu_halt(vcpu);
10121 		else
10122 			kvm_vcpu_block(vcpu);
10123 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
10124 
10125 		if (kvm_x86_ops.post_block)
10126 			static_call(kvm_x86_post_block)(vcpu);
10127 
10128 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
10129 			return 1;
10130 	}
10131 
10132 	if (kvm_apic_accept_events(vcpu) < 0)
10133 		return 0;
10134 	switch(vcpu->arch.mp_state) {
10135 	case KVM_MP_STATE_HALTED:
10136 	case KVM_MP_STATE_AP_RESET_HOLD:
10137 		vcpu->arch.pv.pv_unhalted = false;
10138 		vcpu->arch.mp_state =
10139 			KVM_MP_STATE_RUNNABLE;
10140 		fallthrough;
10141 	case KVM_MP_STATE_RUNNABLE:
10142 		vcpu->arch.apf.halted = false;
10143 		break;
10144 	case KVM_MP_STATE_INIT_RECEIVED:
10145 		break;
10146 	default:
10147 		return -EINTR;
10148 	}
10149 	return 1;
10150 }
10151 
10152 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
10153 {
10154 	if (is_guest_mode(vcpu))
10155 		kvm_check_nested_events(vcpu);
10156 
10157 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
10158 		!vcpu->arch.apf.halted);
10159 }
10160 
10161 static int vcpu_run(struct kvm_vcpu *vcpu)
10162 {
10163 	int r;
10164 	struct kvm *kvm = vcpu->kvm;
10165 
10166 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
10167 	vcpu->arch.l1tf_flush_l1d = true;
10168 
10169 	for (;;) {
10170 		if (kvm_vcpu_running(vcpu)) {
10171 			r = vcpu_enter_guest(vcpu);
10172 		} else {
10173 			r = vcpu_block(kvm, vcpu);
10174 		}
10175 
10176 		if (r <= 0)
10177 			break;
10178 
10179 		kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
10180 		if (kvm_cpu_has_pending_timer(vcpu))
10181 			kvm_inject_pending_timer_irqs(vcpu);
10182 
10183 		if (dm_request_for_irq_injection(vcpu) &&
10184 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
10185 			r = 0;
10186 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
10187 			++vcpu->stat.request_irq_exits;
10188 			break;
10189 		}
10190 
10191 		if (__xfer_to_guest_mode_work_pending()) {
10192 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
10193 			r = xfer_to_guest_mode_handle_work(vcpu);
10194 			if (r)
10195 				return r;
10196 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
10197 		}
10198 	}
10199 
10200 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
10201 
10202 	return r;
10203 }
10204 
10205 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
10206 {
10207 	int r;
10208 
10209 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
10210 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
10211 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
10212 	return r;
10213 }
10214 
10215 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
10216 {
10217 	BUG_ON(!vcpu->arch.pio.count);
10218 
10219 	return complete_emulated_io(vcpu);
10220 }
10221 
10222 /*
10223  * Implements the following, as a state machine:
10224  *
10225  * read:
10226  *   for each fragment
10227  *     for each mmio piece in the fragment
10228  *       write gpa, len
10229  *       exit
10230  *       copy data
10231  *   execute insn
10232  *
10233  * write:
10234  *   for each fragment
10235  *     for each mmio piece in the fragment
10236  *       write gpa, len
10237  *       copy data
10238  *       exit
10239  */
10240 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
10241 {
10242 	struct kvm_run *run = vcpu->run;
10243 	struct kvm_mmio_fragment *frag;
10244 	unsigned len;
10245 
10246 	BUG_ON(!vcpu->mmio_needed);
10247 
10248 	/* Complete previous fragment */
10249 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
10250 	len = min(8u, frag->len);
10251 	if (!vcpu->mmio_is_write)
10252 		memcpy(frag->data, run->mmio.data, len);
10253 
10254 	if (frag->len <= 8) {
10255 		/* Switch to the next fragment. */
10256 		frag++;
10257 		vcpu->mmio_cur_fragment++;
10258 	} else {
10259 		/* Go forward to the next mmio piece. */
10260 		frag->data += len;
10261 		frag->gpa += len;
10262 		frag->len -= len;
10263 	}
10264 
10265 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
10266 		vcpu->mmio_needed = 0;
10267 
10268 		/* FIXME: return into emulator if single-stepping.  */
10269 		if (vcpu->mmio_is_write)
10270 			return 1;
10271 		vcpu->mmio_read_completed = 1;
10272 		return complete_emulated_io(vcpu);
10273 	}
10274 
10275 	run->exit_reason = KVM_EXIT_MMIO;
10276 	run->mmio.phys_addr = frag->gpa;
10277 	if (vcpu->mmio_is_write)
10278 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
10279 	run->mmio.len = min(8u, frag->len);
10280 	run->mmio.is_write = vcpu->mmio_is_write;
10281 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
10282 	return 0;
10283 }
10284 
10285 /* Swap (qemu) user FPU context for the guest FPU context. */
10286 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
10287 {
10288 	/*
10289 	 * Exclude PKRU from restore as restored separately in
10290 	 * kvm_x86_ops.run().
10291 	 */
10292 	fpu_swap_kvm_fpstate(&vcpu->arch.guest_fpu, true);
10293 	trace_kvm_fpu(1);
10294 }
10295 
10296 /* When vcpu_run ends, restore user space FPU context. */
10297 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
10298 {
10299 	fpu_swap_kvm_fpstate(&vcpu->arch.guest_fpu, false);
10300 	++vcpu->stat.fpu_reload;
10301 	trace_kvm_fpu(0);
10302 }
10303 
10304 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
10305 {
10306 	struct kvm_run *kvm_run = vcpu->run;
10307 	int r;
10308 
10309 	vcpu_load(vcpu);
10310 	kvm_sigset_activate(vcpu);
10311 	kvm_run->flags = 0;
10312 	kvm_load_guest_fpu(vcpu);
10313 
10314 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
10315 		if (kvm_run->immediate_exit) {
10316 			r = -EINTR;
10317 			goto out;
10318 		}
10319 		kvm_vcpu_block(vcpu);
10320 		if (kvm_apic_accept_events(vcpu) < 0) {
10321 			r = 0;
10322 			goto out;
10323 		}
10324 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
10325 		r = -EAGAIN;
10326 		if (signal_pending(current)) {
10327 			r = -EINTR;
10328 			kvm_run->exit_reason = KVM_EXIT_INTR;
10329 			++vcpu->stat.signal_exits;
10330 		}
10331 		goto out;
10332 	}
10333 
10334 	if ((kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) ||
10335 	    (kvm_run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)) {
10336 		r = -EINVAL;
10337 		goto out;
10338 	}
10339 
10340 	if (kvm_run->kvm_dirty_regs) {
10341 		r = sync_regs(vcpu);
10342 		if (r != 0)
10343 			goto out;
10344 	}
10345 
10346 	/* re-sync apic's tpr */
10347 	if (!lapic_in_kernel(vcpu)) {
10348 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
10349 			r = -EINVAL;
10350 			goto out;
10351 		}
10352 	}
10353 
10354 	if (unlikely(vcpu->arch.complete_userspace_io)) {
10355 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
10356 		vcpu->arch.complete_userspace_io = NULL;
10357 		r = cui(vcpu);
10358 		if (r <= 0)
10359 			goto out;
10360 	} else
10361 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
10362 
10363 	if (kvm_run->immediate_exit)
10364 		r = -EINTR;
10365 	else
10366 		r = vcpu_run(vcpu);
10367 
10368 out:
10369 	kvm_put_guest_fpu(vcpu);
10370 	if (kvm_run->kvm_valid_regs)
10371 		store_regs(vcpu);
10372 	post_kvm_run_save(vcpu);
10373 	kvm_sigset_deactivate(vcpu);
10374 
10375 	vcpu_put(vcpu);
10376 	return r;
10377 }
10378 
10379 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10380 {
10381 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
10382 		/*
10383 		 * We are here if userspace calls get_regs() in the middle of
10384 		 * instruction emulation. Registers state needs to be copied
10385 		 * back from emulation context to vcpu. Userspace shouldn't do
10386 		 * that usually, but some bad designed PV devices (vmware
10387 		 * backdoor interface) need this to work
10388 		 */
10389 		emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
10390 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10391 	}
10392 	regs->rax = kvm_rax_read(vcpu);
10393 	regs->rbx = kvm_rbx_read(vcpu);
10394 	regs->rcx = kvm_rcx_read(vcpu);
10395 	regs->rdx = kvm_rdx_read(vcpu);
10396 	regs->rsi = kvm_rsi_read(vcpu);
10397 	regs->rdi = kvm_rdi_read(vcpu);
10398 	regs->rsp = kvm_rsp_read(vcpu);
10399 	regs->rbp = kvm_rbp_read(vcpu);
10400 #ifdef CONFIG_X86_64
10401 	regs->r8 = kvm_r8_read(vcpu);
10402 	regs->r9 = kvm_r9_read(vcpu);
10403 	regs->r10 = kvm_r10_read(vcpu);
10404 	regs->r11 = kvm_r11_read(vcpu);
10405 	regs->r12 = kvm_r12_read(vcpu);
10406 	regs->r13 = kvm_r13_read(vcpu);
10407 	regs->r14 = kvm_r14_read(vcpu);
10408 	regs->r15 = kvm_r15_read(vcpu);
10409 #endif
10410 
10411 	regs->rip = kvm_rip_read(vcpu);
10412 	regs->rflags = kvm_get_rflags(vcpu);
10413 }
10414 
10415 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10416 {
10417 	vcpu_load(vcpu);
10418 	__get_regs(vcpu, regs);
10419 	vcpu_put(vcpu);
10420 	return 0;
10421 }
10422 
10423 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10424 {
10425 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
10426 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10427 
10428 	kvm_rax_write(vcpu, regs->rax);
10429 	kvm_rbx_write(vcpu, regs->rbx);
10430 	kvm_rcx_write(vcpu, regs->rcx);
10431 	kvm_rdx_write(vcpu, regs->rdx);
10432 	kvm_rsi_write(vcpu, regs->rsi);
10433 	kvm_rdi_write(vcpu, regs->rdi);
10434 	kvm_rsp_write(vcpu, regs->rsp);
10435 	kvm_rbp_write(vcpu, regs->rbp);
10436 #ifdef CONFIG_X86_64
10437 	kvm_r8_write(vcpu, regs->r8);
10438 	kvm_r9_write(vcpu, regs->r9);
10439 	kvm_r10_write(vcpu, regs->r10);
10440 	kvm_r11_write(vcpu, regs->r11);
10441 	kvm_r12_write(vcpu, regs->r12);
10442 	kvm_r13_write(vcpu, regs->r13);
10443 	kvm_r14_write(vcpu, regs->r14);
10444 	kvm_r15_write(vcpu, regs->r15);
10445 #endif
10446 
10447 	kvm_rip_write(vcpu, regs->rip);
10448 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
10449 
10450 	vcpu->arch.exception.pending = false;
10451 
10452 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10453 }
10454 
10455 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10456 {
10457 	vcpu_load(vcpu);
10458 	__set_regs(vcpu, regs);
10459 	vcpu_put(vcpu);
10460 	return 0;
10461 }
10462 
10463 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
10464 {
10465 	struct kvm_segment cs;
10466 
10467 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10468 	*db = cs.db;
10469 	*l = cs.l;
10470 }
10471 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
10472 
10473 static void __get_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10474 {
10475 	struct desc_ptr dt;
10476 
10477 	if (vcpu->arch.guest_state_protected)
10478 		goto skip_protected_regs;
10479 
10480 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10481 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10482 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10483 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10484 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10485 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10486 
10487 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10488 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10489 
10490 	static_call(kvm_x86_get_idt)(vcpu, &dt);
10491 	sregs->idt.limit = dt.size;
10492 	sregs->idt.base = dt.address;
10493 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
10494 	sregs->gdt.limit = dt.size;
10495 	sregs->gdt.base = dt.address;
10496 
10497 	sregs->cr2 = vcpu->arch.cr2;
10498 	sregs->cr3 = kvm_read_cr3(vcpu);
10499 
10500 skip_protected_regs:
10501 	sregs->cr0 = kvm_read_cr0(vcpu);
10502 	sregs->cr4 = kvm_read_cr4(vcpu);
10503 	sregs->cr8 = kvm_get_cr8(vcpu);
10504 	sregs->efer = vcpu->arch.efer;
10505 	sregs->apic_base = kvm_get_apic_base(vcpu);
10506 }
10507 
10508 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10509 {
10510 	__get_sregs_common(vcpu, sregs);
10511 
10512 	if (vcpu->arch.guest_state_protected)
10513 		return;
10514 
10515 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
10516 		set_bit(vcpu->arch.interrupt.nr,
10517 			(unsigned long *)sregs->interrupt_bitmap);
10518 }
10519 
10520 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10521 {
10522 	int i;
10523 
10524 	__get_sregs_common(vcpu, (struct kvm_sregs *)sregs2);
10525 
10526 	if (vcpu->arch.guest_state_protected)
10527 		return;
10528 
10529 	if (is_pae_paging(vcpu)) {
10530 		for (i = 0 ; i < 4 ; i++)
10531 			sregs2->pdptrs[i] = kvm_pdptr_read(vcpu, i);
10532 		sregs2->flags |= KVM_SREGS2_FLAGS_PDPTRS_VALID;
10533 	}
10534 }
10535 
10536 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
10537 				  struct kvm_sregs *sregs)
10538 {
10539 	vcpu_load(vcpu);
10540 	__get_sregs(vcpu, sregs);
10541 	vcpu_put(vcpu);
10542 	return 0;
10543 }
10544 
10545 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
10546 				    struct kvm_mp_state *mp_state)
10547 {
10548 	int r;
10549 
10550 	vcpu_load(vcpu);
10551 	if (kvm_mpx_supported())
10552 		kvm_load_guest_fpu(vcpu);
10553 
10554 	r = kvm_apic_accept_events(vcpu);
10555 	if (r < 0)
10556 		goto out;
10557 	r = 0;
10558 
10559 	if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
10560 	     vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
10561 	    vcpu->arch.pv.pv_unhalted)
10562 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
10563 	else
10564 		mp_state->mp_state = vcpu->arch.mp_state;
10565 
10566 out:
10567 	if (kvm_mpx_supported())
10568 		kvm_put_guest_fpu(vcpu);
10569 	vcpu_put(vcpu);
10570 	return r;
10571 }
10572 
10573 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
10574 				    struct kvm_mp_state *mp_state)
10575 {
10576 	int ret = -EINVAL;
10577 
10578 	vcpu_load(vcpu);
10579 
10580 	if (!lapic_in_kernel(vcpu) &&
10581 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
10582 		goto out;
10583 
10584 	/*
10585 	 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
10586 	 * INIT state; latched init should be reported using
10587 	 * KVM_SET_VCPU_EVENTS, so reject it here.
10588 	 */
10589 	if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
10590 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
10591 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
10592 		goto out;
10593 
10594 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
10595 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
10596 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
10597 	} else
10598 		vcpu->arch.mp_state = mp_state->mp_state;
10599 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10600 
10601 	ret = 0;
10602 out:
10603 	vcpu_put(vcpu);
10604 	return ret;
10605 }
10606 
10607 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
10608 		    int reason, bool has_error_code, u32 error_code)
10609 {
10610 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
10611 	int ret;
10612 
10613 	init_emulate_ctxt(vcpu);
10614 
10615 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
10616 				   has_error_code, error_code);
10617 	if (ret) {
10618 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
10619 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
10620 		vcpu->run->internal.ndata = 0;
10621 		return 0;
10622 	}
10623 
10624 	kvm_rip_write(vcpu, ctxt->eip);
10625 	kvm_set_rflags(vcpu, ctxt->eflags);
10626 	return 1;
10627 }
10628 EXPORT_SYMBOL_GPL(kvm_task_switch);
10629 
10630 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10631 {
10632 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
10633 		/*
10634 		 * When EFER.LME and CR0.PG are set, the processor is in
10635 		 * 64-bit mode (though maybe in a 32-bit code segment).
10636 		 * CR4.PAE and EFER.LMA must be set.
10637 		 */
10638 		if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
10639 			return false;
10640 		if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
10641 			return false;
10642 	} else {
10643 		/*
10644 		 * Not in 64-bit mode: EFER.LMA is clear and the code
10645 		 * segment cannot be 64-bit.
10646 		 */
10647 		if (sregs->efer & EFER_LMA || sregs->cs.l)
10648 			return false;
10649 	}
10650 
10651 	return kvm_is_valid_cr4(vcpu, sregs->cr4);
10652 }
10653 
10654 static int __set_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs,
10655 		int *mmu_reset_needed, bool update_pdptrs)
10656 {
10657 	struct msr_data apic_base_msr;
10658 	int idx;
10659 	struct desc_ptr dt;
10660 
10661 	if (!kvm_is_valid_sregs(vcpu, sregs))
10662 		return -EINVAL;
10663 
10664 	apic_base_msr.data = sregs->apic_base;
10665 	apic_base_msr.host_initiated = true;
10666 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
10667 		return -EINVAL;
10668 
10669 	if (vcpu->arch.guest_state_protected)
10670 		return 0;
10671 
10672 	dt.size = sregs->idt.limit;
10673 	dt.address = sregs->idt.base;
10674 	static_call(kvm_x86_set_idt)(vcpu, &dt);
10675 	dt.size = sregs->gdt.limit;
10676 	dt.address = sregs->gdt.base;
10677 	static_call(kvm_x86_set_gdt)(vcpu, &dt);
10678 
10679 	vcpu->arch.cr2 = sregs->cr2;
10680 	*mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
10681 	vcpu->arch.cr3 = sregs->cr3;
10682 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
10683 	static_call_cond(kvm_x86_post_set_cr3)(vcpu, sregs->cr3);
10684 
10685 	kvm_set_cr8(vcpu, sregs->cr8);
10686 
10687 	*mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
10688 	static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
10689 
10690 	*mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
10691 	static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
10692 	vcpu->arch.cr0 = sregs->cr0;
10693 
10694 	*mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
10695 	static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
10696 
10697 	if (update_pdptrs) {
10698 		idx = srcu_read_lock(&vcpu->kvm->srcu);
10699 		if (is_pae_paging(vcpu)) {
10700 			load_pdptrs(vcpu, kvm_read_cr3(vcpu));
10701 			*mmu_reset_needed = 1;
10702 		}
10703 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
10704 	}
10705 
10706 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10707 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10708 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10709 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10710 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10711 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10712 
10713 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10714 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10715 
10716 	update_cr8_intercept(vcpu);
10717 
10718 	/* Older userspace won't unhalt the vcpu on reset. */
10719 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
10720 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
10721 	    !is_protmode(vcpu))
10722 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10723 
10724 	return 0;
10725 }
10726 
10727 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10728 {
10729 	int pending_vec, max_bits;
10730 	int mmu_reset_needed = 0;
10731 	int ret = __set_sregs_common(vcpu, sregs, &mmu_reset_needed, true);
10732 
10733 	if (ret)
10734 		return ret;
10735 
10736 	if (mmu_reset_needed)
10737 		kvm_mmu_reset_context(vcpu);
10738 
10739 	max_bits = KVM_NR_INTERRUPTS;
10740 	pending_vec = find_first_bit(
10741 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
10742 
10743 	if (pending_vec < max_bits) {
10744 		kvm_queue_interrupt(vcpu, pending_vec, false);
10745 		pr_debug("Set back pending irq %d\n", pending_vec);
10746 		kvm_make_request(KVM_REQ_EVENT, vcpu);
10747 	}
10748 	return 0;
10749 }
10750 
10751 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10752 {
10753 	int mmu_reset_needed = 0;
10754 	bool valid_pdptrs = sregs2->flags & KVM_SREGS2_FLAGS_PDPTRS_VALID;
10755 	bool pae = (sregs2->cr0 & X86_CR0_PG) && (sregs2->cr4 & X86_CR4_PAE) &&
10756 		!(sregs2->efer & EFER_LMA);
10757 	int i, ret;
10758 
10759 	if (sregs2->flags & ~KVM_SREGS2_FLAGS_PDPTRS_VALID)
10760 		return -EINVAL;
10761 
10762 	if (valid_pdptrs && (!pae || vcpu->arch.guest_state_protected))
10763 		return -EINVAL;
10764 
10765 	ret = __set_sregs_common(vcpu, (struct kvm_sregs *)sregs2,
10766 				 &mmu_reset_needed, !valid_pdptrs);
10767 	if (ret)
10768 		return ret;
10769 
10770 	if (valid_pdptrs) {
10771 		for (i = 0; i < 4 ; i++)
10772 			kvm_pdptr_write(vcpu, i, sregs2->pdptrs[i]);
10773 
10774 		kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
10775 		mmu_reset_needed = 1;
10776 		vcpu->arch.pdptrs_from_userspace = true;
10777 	}
10778 	if (mmu_reset_needed)
10779 		kvm_mmu_reset_context(vcpu);
10780 	return 0;
10781 }
10782 
10783 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
10784 				  struct kvm_sregs *sregs)
10785 {
10786 	int ret;
10787 
10788 	vcpu_load(vcpu);
10789 	ret = __set_sregs(vcpu, sregs);
10790 	vcpu_put(vcpu);
10791 	return ret;
10792 }
10793 
10794 static void kvm_arch_vcpu_guestdbg_update_apicv_inhibit(struct kvm *kvm)
10795 {
10796 	bool inhibit = false;
10797 	struct kvm_vcpu *vcpu;
10798 	unsigned long i;
10799 
10800 	down_write(&kvm->arch.apicv_update_lock);
10801 
10802 	kvm_for_each_vcpu(i, vcpu, kvm) {
10803 		if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ) {
10804 			inhibit = true;
10805 			break;
10806 		}
10807 	}
10808 	__kvm_request_apicv_update(kvm, !inhibit, APICV_INHIBIT_REASON_BLOCKIRQ);
10809 	up_write(&kvm->arch.apicv_update_lock);
10810 }
10811 
10812 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
10813 					struct kvm_guest_debug *dbg)
10814 {
10815 	unsigned long rflags;
10816 	int i, r;
10817 
10818 	if (vcpu->arch.guest_state_protected)
10819 		return -EINVAL;
10820 
10821 	vcpu_load(vcpu);
10822 
10823 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10824 		r = -EBUSY;
10825 		if (vcpu->arch.exception.pending)
10826 			goto out;
10827 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10828 			kvm_queue_exception(vcpu, DB_VECTOR);
10829 		else
10830 			kvm_queue_exception(vcpu, BP_VECTOR);
10831 	}
10832 
10833 	/*
10834 	 * Read rflags as long as potentially injected trace flags are still
10835 	 * filtered out.
10836 	 */
10837 	rflags = kvm_get_rflags(vcpu);
10838 
10839 	vcpu->guest_debug = dbg->control;
10840 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10841 		vcpu->guest_debug = 0;
10842 
10843 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10844 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
10845 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10846 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10847 	} else {
10848 		for (i = 0; i < KVM_NR_DB_REGS; i++)
10849 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10850 	}
10851 	kvm_update_dr7(vcpu);
10852 
10853 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10854 		vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
10855 
10856 	/*
10857 	 * Trigger an rflags update that will inject or remove the trace
10858 	 * flags.
10859 	 */
10860 	kvm_set_rflags(vcpu, rflags);
10861 
10862 	static_call(kvm_x86_update_exception_bitmap)(vcpu);
10863 
10864 	kvm_arch_vcpu_guestdbg_update_apicv_inhibit(vcpu->kvm);
10865 
10866 	r = 0;
10867 
10868 out:
10869 	vcpu_put(vcpu);
10870 	return r;
10871 }
10872 
10873 /*
10874  * Translate a guest virtual address to a guest physical address.
10875  */
10876 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10877 				    struct kvm_translation *tr)
10878 {
10879 	unsigned long vaddr = tr->linear_address;
10880 	gpa_t gpa;
10881 	int idx;
10882 
10883 	vcpu_load(vcpu);
10884 
10885 	idx = srcu_read_lock(&vcpu->kvm->srcu);
10886 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10887 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
10888 	tr->physical_address = gpa;
10889 	tr->valid = gpa != UNMAPPED_GVA;
10890 	tr->writeable = 1;
10891 	tr->usermode = 0;
10892 
10893 	vcpu_put(vcpu);
10894 	return 0;
10895 }
10896 
10897 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10898 {
10899 	struct fxregs_state *fxsave;
10900 
10901 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
10902 		return 0;
10903 
10904 	vcpu_load(vcpu);
10905 
10906 	fxsave = &vcpu->arch.guest_fpu.fpstate->regs.fxsave;
10907 	memcpy(fpu->fpr, fxsave->st_space, 128);
10908 	fpu->fcw = fxsave->cwd;
10909 	fpu->fsw = fxsave->swd;
10910 	fpu->ftwx = fxsave->twd;
10911 	fpu->last_opcode = fxsave->fop;
10912 	fpu->last_ip = fxsave->rip;
10913 	fpu->last_dp = fxsave->rdp;
10914 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10915 
10916 	vcpu_put(vcpu);
10917 	return 0;
10918 }
10919 
10920 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10921 {
10922 	struct fxregs_state *fxsave;
10923 
10924 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
10925 		return 0;
10926 
10927 	vcpu_load(vcpu);
10928 
10929 	fxsave = &vcpu->arch.guest_fpu.fpstate->regs.fxsave;
10930 
10931 	memcpy(fxsave->st_space, fpu->fpr, 128);
10932 	fxsave->cwd = fpu->fcw;
10933 	fxsave->swd = fpu->fsw;
10934 	fxsave->twd = fpu->ftwx;
10935 	fxsave->fop = fpu->last_opcode;
10936 	fxsave->rip = fpu->last_ip;
10937 	fxsave->rdp = fpu->last_dp;
10938 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10939 
10940 	vcpu_put(vcpu);
10941 	return 0;
10942 }
10943 
10944 static void store_regs(struct kvm_vcpu *vcpu)
10945 {
10946 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10947 
10948 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10949 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
10950 
10951 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10952 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10953 
10954 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10955 		kvm_vcpu_ioctl_x86_get_vcpu_events(
10956 				vcpu, &vcpu->run->s.regs.events);
10957 }
10958 
10959 static int sync_regs(struct kvm_vcpu *vcpu)
10960 {
10961 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10962 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
10963 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10964 	}
10965 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10966 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10967 			return -EINVAL;
10968 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10969 	}
10970 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10971 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10972 				vcpu, &vcpu->run->s.regs.events))
10973 			return -EINVAL;
10974 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10975 	}
10976 
10977 	return 0;
10978 }
10979 
10980 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10981 {
10982 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10983 		pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10984 			     "guest TSC will not be reliable\n");
10985 
10986 	return 0;
10987 }
10988 
10989 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10990 {
10991 	struct page *page;
10992 	int r;
10993 
10994 	vcpu->arch.last_vmentry_cpu = -1;
10995 	vcpu->arch.regs_avail = ~0;
10996 	vcpu->arch.regs_dirty = ~0;
10997 
10998 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10999 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11000 	else
11001 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
11002 
11003 	r = kvm_mmu_create(vcpu);
11004 	if (r < 0)
11005 		return r;
11006 
11007 	if (irqchip_in_kernel(vcpu->kvm)) {
11008 		r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
11009 		if (r < 0)
11010 			goto fail_mmu_destroy;
11011 		if (kvm_apicv_activated(vcpu->kvm))
11012 			vcpu->arch.apicv_active = true;
11013 	} else
11014 		static_branch_inc(&kvm_has_noapic_vcpu);
11015 
11016 	r = -ENOMEM;
11017 
11018 	page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
11019 	if (!page)
11020 		goto fail_free_lapic;
11021 	vcpu->arch.pio_data = page_address(page);
11022 
11023 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
11024 				       GFP_KERNEL_ACCOUNT);
11025 	if (!vcpu->arch.mce_banks)
11026 		goto fail_free_pio_data;
11027 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
11028 
11029 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
11030 				GFP_KERNEL_ACCOUNT))
11031 		goto fail_free_mce_banks;
11032 
11033 	if (!alloc_emulate_ctxt(vcpu))
11034 		goto free_wbinvd_dirty_mask;
11035 
11036 	if (!fpu_alloc_guest_fpstate(&vcpu->arch.guest_fpu)) {
11037 		pr_err("kvm: failed to allocate vcpu's fpu\n");
11038 		goto free_emulate_ctxt;
11039 	}
11040 
11041 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
11042 	vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
11043 
11044 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
11045 
11046 	kvm_async_pf_hash_reset(vcpu);
11047 	kvm_pmu_init(vcpu);
11048 
11049 	vcpu->arch.pending_external_vector = -1;
11050 	vcpu->arch.preempted_in_kernel = false;
11051 
11052 #if IS_ENABLED(CONFIG_HYPERV)
11053 	vcpu->arch.hv_root_tdp = INVALID_PAGE;
11054 #endif
11055 
11056 	r = static_call(kvm_x86_vcpu_create)(vcpu);
11057 	if (r)
11058 		goto free_guest_fpu;
11059 
11060 	vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
11061 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
11062 	kvm_vcpu_mtrr_init(vcpu);
11063 	vcpu_load(vcpu);
11064 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
11065 	kvm_vcpu_reset(vcpu, false);
11066 	kvm_init_mmu(vcpu);
11067 	vcpu_put(vcpu);
11068 	return 0;
11069 
11070 free_guest_fpu:
11071 	fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
11072 free_emulate_ctxt:
11073 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
11074 free_wbinvd_dirty_mask:
11075 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
11076 fail_free_mce_banks:
11077 	kfree(vcpu->arch.mce_banks);
11078 fail_free_pio_data:
11079 	free_page((unsigned long)vcpu->arch.pio_data);
11080 fail_free_lapic:
11081 	kvm_free_lapic(vcpu);
11082 fail_mmu_destroy:
11083 	kvm_mmu_destroy(vcpu);
11084 	return r;
11085 }
11086 
11087 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
11088 {
11089 	struct kvm *kvm = vcpu->kvm;
11090 
11091 	if (mutex_lock_killable(&vcpu->mutex))
11092 		return;
11093 	vcpu_load(vcpu);
11094 	kvm_synchronize_tsc(vcpu, 0);
11095 	vcpu_put(vcpu);
11096 
11097 	/* poll control enabled by default */
11098 	vcpu->arch.msr_kvm_poll_control = 1;
11099 
11100 	mutex_unlock(&vcpu->mutex);
11101 
11102 	if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
11103 		schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
11104 						KVMCLOCK_SYNC_PERIOD);
11105 }
11106 
11107 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
11108 {
11109 	int idx;
11110 
11111 	kvmclock_reset(vcpu);
11112 
11113 	static_call(kvm_x86_vcpu_free)(vcpu);
11114 
11115 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
11116 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
11117 	fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
11118 
11119 	kvm_hv_vcpu_uninit(vcpu);
11120 	kvm_pmu_destroy(vcpu);
11121 	kfree(vcpu->arch.mce_banks);
11122 	kvm_free_lapic(vcpu);
11123 	idx = srcu_read_lock(&vcpu->kvm->srcu);
11124 	kvm_mmu_destroy(vcpu);
11125 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
11126 	free_page((unsigned long)vcpu->arch.pio_data);
11127 	kvfree(vcpu->arch.cpuid_entries);
11128 	if (!lapic_in_kernel(vcpu))
11129 		static_branch_dec(&kvm_has_noapic_vcpu);
11130 }
11131 
11132 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
11133 {
11134 	struct kvm_cpuid_entry2 *cpuid_0x1;
11135 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
11136 	unsigned long new_cr0;
11137 
11138 	/*
11139 	 * Several of the "set" flows, e.g. ->set_cr0(), read other registers
11140 	 * to handle side effects.  RESET emulation hits those flows and relies
11141 	 * on emulated/virtualized registers, including those that are loaded
11142 	 * into hardware, to be zeroed at vCPU creation.  Use CRs as a sentinel
11143 	 * to detect improper or missing initialization.
11144 	 */
11145 	WARN_ON_ONCE(!init_event &&
11146 		     (old_cr0 || kvm_read_cr3(vcpu) || kvm_read_cr4(vcpu)));
11147 
11148 	kvm_lapic_reset(vcpu, init_event);
11149 
11150 	vcpu->arch.hflags = 0;
11151 
11152 	vcpu->arch.smi_pending = 0;
11153 	vcpu->arch.smi_count = 0;
11154 	atomic_set(&vcpu->arch.nmi_queued, 0);
11155 	vcpu->arch.nmi_pending = 0;
11156 	vcpu->arch.nmi_injected = false;
11157 	kvm_clear_interrupt_queue(vcpu);
11158 	kvm_clear_exception_queue(vcpu);
11159 
11160 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
11161 	kvm_update_dr0123(vcpu);
11162 	vcpu->arch.dr6 = DR6_ACTIVE_LOW;
11163 	vcpu->arch.dr7 = DR7_FIXED_1;
11164 	kvm_update_dr7(vcpu);
11165 
11166 	vcpu->arch.cr2 = 0;
11167 
11168 	kvm_make_request(KVM_REQ_EVENT, vcpu);
11169 	vcpu->arch.apf.msr_en_val = 0;
11170 	vcpu->arch.apf.msr_int_val = 0;
11171 	vcpu->arch.st.msr_val = 0;
11172 
11173 	kvmclock_reset(vcpu);
11174 
11175 	kvm_clear_async_pf_completion_queue(vcpu);
11176 	kvm_async_pf_hash_reset(vcpu);
11177 	vcpu->arch.apf.halted = false;
11178 
11179 	if (vcpu->arch.guest_fpu.fpstate && kvm_mpx_supported()) {
11180 		struct fpstate *fpstate = vcpu->arch.guest_fpu.fpstate;
11181 
11182 		/*
11183 		 * To avoid have the INIT path from kvm_apic_has_events() that be
11184 		 * called with loaded FPU and does not let userspace fix the state.
11185 		 */
11186 		if (init_event)
11187 			kvm_put_guest_fpu(vcpu);
11188 
11189 		fpstate_clear_xstate_component(fpstate, XFEATURE_BNDREGS);
11190 		fpstate_clear_xstate_component(fpstate, XFEATURE_BNDCSR);
11191 
11192 		if (init_event)
11193 			kvm_load_guest_fpu(vcpu);
11194 	}
11195 
11196 	if (!init_event) {
11197 		kvm_pmu_reset(vcpu);
11198 		vcpu->arch.smbase = 0x30000;
11199 
11200 		vcpu->arch.msr_misc_features_enables = 0;
11201 
11202 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
11203 	}
11204 
11205 	/* All GPRs except RDX (handled below) are zeroed on RESET/INIT. */
11206 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
11207 	kvm_register_mark_dirty(vcpu, VCPU_REGS_RSP);
11208 
11209 	/*
11210 	 * Fall back to KVM's default Family/Model/Stepping of 0x600 (P6/Athlon)
11211 	 * if no CPUID match is found.  Note, it's impossible to get a match at
11212 	 * RESET since KVM emulates RESET before exposing the vCPU to userspace,
11213 	 * i.e. it's impossible for kvm_find_cpuid_entry() to find a valid entry
11214 	 * on RESET.  But, go through the motions in case that's ever remedied.
11215 	 */
11216 	cpuid_0x1 = kvm_find_cpuid_entry(vcpu, 1, 0);
11217 	kvm_rdx_write(vcpu, cpuid_0x1 ? cpuid_0x1->eax : 0x600);
11218 
11219 	vcpu->arch.ia32_xss = 0;
11220 
11221 	static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
11222 
11223 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
11224 	kvm_rip_write(vcpu, 0xfff0);
11225 
11226 	vcpu->arch.cr3 = 0;
11227 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
11228 
11229 	/*
11230 	 * CR0.CD/NW are set on RESET, preserved on INIT.  Note, some versions
11231 	 * of Intel's SDM list CD/NW as being set on INIT, but they contradict
11232 	 * (or qualify) that with a footnote stating that CD/NW are preserved.
11233 	 */
11234 	new_cr0 = X86_CR0_ET;
11235 	if (init_event)
11236 		new_cr0 |= (old_cr0 & (X86_CR0_NW | X86_CR0_CD));
11237 	else
11238 		new_cr0 |= X86_CR0_NW | X86_CR0_CD;
11239 
11240 	static_call(kvm_x86_set_cr0)(vcpu, new_cr0);
11241 	static_call(kvm_x86_set_cr4)(vcpu, 0);
11242 	static_call(kvm_x86_set_efer)(vcpu, 0);
11243 	static_call(kvm_x86_update_exception_bitmap)(vcpu);
11244 
11245 	/*
11246 	 * Reset the MMU context if paging was enabled prior to INIT (which is
11247 	 * implied if CR0.PG=1 as CR0 will be '0' prior to RESET).  Unlike the
11248 	 * standard CR0/CR4/EFER modification paths, only CR0.PG needs to be
11249 	 * checked because it is unconditionally cleared on INIT and all other
11250 	 * paging related bits are ignored if paging is disabled, i.e. CR0.WP,
11251 	 * CR4, and EFER changes are all irrelevant if CR0.PG was '0'.
11252 	 */
11253 	if (old_cr0 & X86_CR0_PG)
11254 		kvm_mmu_reset_context(vcpu);
11255 
11256 	/*
11257 	 * Intel's SDM states that all TLB entries are flushed on INIT.  AMD's
11258 	 * APM states the TLBs are untouched by INIT, but it also states that
11259 	 * the TLBs are flushed on "External initialization of the processor."
11260 	 * Flush the guest TLB regardless of vendor, there is no meaningful
11261 	 * benefit in relying on the guest to flush the TLB immediately after
11262 	 * INIT.  A spurious TLB flush is benign and likely negligible from a
11263 	 * performance perspective.
11264 	 */
11265 	if (init_event)
11266 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
11267 }
11268 EXPORT_SYMBOL_GPL(kvm_vcpu_reset);
11269 
11270 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
11271 {
11272 	struct kvm_segment cs;
11273 
11274 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
11275 	cs.selector = vector << 8;
11276 	cs.base = vector << 12;
11277 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
11278 	kvm_rip_write(vcpu, 0);
11279 }
11280 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
11281 
11282 int kvm_arch_hardware_enable(void)
11283 {
11284 	struct kvm *kvm;
11285 	struct kvm_vcpu *vcpu;
11286 	unsigned long i;
11287 	int ret;
11288 	u64 local_tsc;
11289 	u64 max_tsc = 0;
11290 	bool stable, backwards_tsc = false;
11291 
11292 	kvm_user_return_msr_cpu_online();
11293 	ret = static_call(kvm_x86_hardware_enable)();
11294 	if (ret != 0)
11295 		return ret;
11296 
11297 	local_tsc = rdtsc();
11298 	stable = !kvm_check_tsc_unstable();
11299 	list_for_each_entry(kvm, &vm_list, vm_list) {
11300 		kvm_for_each_vcpu(i, vcpu, kvm) {
11301 			if (!stable && vcpu->cpu == smp_processor_id())
11302 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
11303 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
11304 				backwards_tsc = true;
11305 				if (vcpu->arch.last_host_tsc > max_tsc)
11306 					max_tsc = vcpu->arch.last_host_tsc;
11307 			}
11308 		}
11309 	}
11310 
11311 	/*
11312 	 * Sometimes, even reliable TSCs go backwards.  This happens on
11313 	 * platforms that reset TSC during suspend or hibernate actions, but
11314 	 * maintain synchronization.  We must compensate.  Fortunately, we can
11315 	 * detect that condition here, which happens early in CPU bringup,
11316 	 * before any KVM threads can be running.  Unfortunately, we can't
11317 	 * bring the TSCs fully up to date with real time, as we aren't yet far
11318 	 * enough into CPU bringup that we know how much real time has actually
11319 	 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
11320 	 * variables that haven't been updated yet.
11321 	 *
11322 	 * So we simply find the maximum observed TSC above, then record the
11323 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
11324 	 * the adjustment will be applied.  Note that we accumulate
11325 	 * adjustments, in case multiple suspend cycles happen before some VCPU
11326 	 * gets a chance to run again.  In the event that no KVM threads get a
11327 	 * chance to run, we will miss the entire elapsed period, as we'll have
11328 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
11329 	 * loose cycle time.  This isn't too big a deal, since the loss will be
11330 	 * uniform across all VCPUs (not to mention the scenario is extremely
11331 	 * unlikely). It is possible that a second hibernate recovery happens
11332 	 * much faster than a first, causing the observed TSC here to be
11333 	 * smaller; this would require additional padding adjustment, which is
11334 	 * why we set last_host_tsc to the local tsc observed here.
11335 	 *
11336 	 * N.B. - this code below runs only on platforms with reliable TSC,
11337 	 * as that is the only way backwards_tsc is set above.  Also note
11338 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
11339 	 * have the same delta_cyc adjustment applied if backwards_tsc
11340 	 * is detected.  Note further, this adjustment is only done once,
11341 	 * as we reset last_host_tsc on all VCPUs to stop this from being
11342 	 * called multiple times (one for each physical CPU bringup).
11343 	 *
11344 	 * Platforms with unreliable TSCs don't have to deal with this, they
11345 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
11346 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
11347 	 * guarantee that they stay in perfect synchronization.
11348 	 */
11349 	if (backwards_tsc) {
11350 		u64 delta_cyc = max_tsc - local_tsc;
11351 		list_for_each_entry(kvm, &vm_list, vm_list) {
11352 			kvm->arch.backwards_tsc_observed = true;
11353 			kvm_for_each_vcpu(i, vcpu, kvm) {
11354 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
11355 				vcpu->arch.last_host_tsc = local_tsc;
11356 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
11357 			}
11358 
11359 			/*
11360 			 * We have to disable TSC offset matching.. if you were
11361 			 * booting a VM while issuing an S4 host suspend....
11362 			 * you may have some problem.  Solving this issue is
11363 			 * left as an exercise to the reader.
11364 			 */
11365 			kvm->arch.last_tsc_nsec = 0;
11366 			kvm->arch.last_tsc_write = 0;
11367 		}
11368 
11369 	}
11370 	return 0;
11371 }
11372 
11373 void kvm_arch_hardware_disable(void)
11374 {
11375 	static_call(kvm_x86_hardware_disable)();
11376 	drop_user_return_notifiers();
11377 }
11378 
11379 int kvm_arch_hardware_setup(void *opaque)
11380 {
11381 	struct kvm_x86_init_ops *ops = opaque;
11382 	int r;
11383 
11384 	rdmsrl_safe(MSR_EFER, &host_efer);
11385 
11386 	if (boot_cpu_has(X86_FEATURE_XSAVES))
11387 		rdmsrl(MSR_IA32_XSS, host_xss);
11388 
11389 	r = ops->hardware_setup();
11390 	if (r != 0)
11391 		return r;
11392 
11393 	memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
11394 	kvm_ops_static_call_update();
11395 
11396 	kvm_register_perf_callbacks(ops->handle_intel_pt_intr);
11397 
11398 	if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
11399 		supported_xss = 0;
11400 
11401 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
11402 	cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
11403 #undef __kvm_cpu_cap_has
11404 
11405 	if (kvm_has_tsc_control) {
11406 		/*
11407 		 * Make sure the user can only configure tsc_khz values that
11408 		 * fit into a signed integer.
11409 		 * A min value is not calculated because it will always
11410 		 * be 1 on all machines.
11411 		 */
11412 		u64 max = min(0x7fffffffULL,
11413 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
11414 		kvm_max_guest_tsc_khz = max;
11415 
11416 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
11417 	}
11418 
11419 	kvm_init_msr_list();
11420 	return 0;
11421 }
11422 
11423 void kvm_arch_hardware_unsetup(void)
11424 {
11425 	kvm_unregister_perf_callbacks();
11426 
11427 	static_call(kvm_x86_hardware_unsetup)();
11428 }
11429 
11430 int kvm_arch_check_processor_compat(void *opaque)
11431 {
11432 	struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
11433 	struct kvm_x86_init_ops *ops = opaque;
11434 
11435 	WARN_ON(!irqs_disabled());
11436 
11437 	if (__cr4_reserved_bits(cpu_has, c) !=
11438 	    __cr4_reserved_bits(cpu_has, &boot_cpu_data))
11439 		return -EIO;
11440 
11441 	return ops->check_processor_compatibility();
11442 }
11443 
11444 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
11445 {
11446 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
11447 }
11448 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
11449 
11450 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
11451 {
11452 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
11453 }
11454 
11455 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
11456 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
11457 
11458 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
11459 {
11460 	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
11461 
11462 	vcpu->arch.l1tf_flush_l1d = true;
11463 	if (pmu->version && unlikely(pmu->event_count)) {
11464 		pmu->need_cleanup = true;
11465 		kvm_make_request(KVM_REQ_PMU, vcpu);
11466 	}
11467 	static_call(kvm_x86_sched_in)(vcpu, cpu);
11468 }
11469 
11470 void kvm_arch_free_vm(struct kvm *kvm)
11471 {
11472 	kfree(to_kvm_hv(kvm)->hv_pa_pg);
11473 	__kvm_arch_free_vm(kvm);
11474 }
11475 
11476 
11477 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
11478 {
11479 	int ret;
11480 	unsigned long flags;
11481 
11482 	if (type)
11483 		return -EINVAL;
11484 
11485 	ret = kvm_page_track_init(kvm);
11486 	if (ret)
11487 		return ret;
11488 
11489 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
11490 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
11491 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
11492 	INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
11493 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
11494 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
11495 
11496 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
11497 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
11498 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
11499 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
11500 		&kvm->arch.irq_sources_bitmap);
11501 
11502 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
11503 	mutex_init(&kvm->arch.apic_map_lock);
11504 	seqcount_raw_spinlock_init(&kvm->arch.pvclock_sc, &kvm->arch.tsc_write_lock);
11505 	kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
11506 
11507 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
11508 	pvclock_update_vm_gtod_copy(kvm);
11509 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
11510 
11511 	kvm->arch.guest_can_read_msr_platform_info = true;
11512 
11513 #if IS_ENABLED(CONFIG_HYPERV)
11514 	spin_lock_init(&kvm->arch.hv_root_tdp_lock);
11515 	kvm->arch.hv_root_tdp = INVALID_PAGE;
11516 #endif
11517 
11518 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
11519 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
11520 
11521 	kvm_apicv_init(kvm);
11522 	kvm_hv_init_vm(kvm);
11523 	kvm_mmu_init_vm(kvm);
11524 	kvm_xen_init_vm(kvm);
11525 
11526 	return static_call(kvm_x86_vm_init)(kvm);
11527 }
11528 
11529 int kvm_arch_post_init_vm(struct kvm *kvm)
11530 {
11531 	return kvm_mmu_post_init_vm(kvm);
11532 }
11533 
11534 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
11535 {
11536 	vcpu_load(vcpu);
11537 	kvm_mmu_unload(vcpu);
11538 	vcpu_put(vcpu);
11539 }
11540 
11541 static void kvm_free_vcpus(struct kvm *kvm)
11542 {
11543 	unsigned long i;
11544 	struct kvm_vcpu *vcpu;
11545 
11546 	/*
11547 	 * Unpin any mmu pages first.
11548 	 */
11549 	kvm_for_each_vcpu(i, vcpu, kvm) {
11550 		kvm_clear_async_pf_completion_queue(vcpu);
11551 		kvm_unload_vcpu_mmu(vcpu);
11552 	}
11553 
11554 	kvm_destroy_vcpus(kvm);
11555 }
11556 
11557 void kvm_arch_sync_events(struct kvm *kvm)
11558 {
11559 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
11560 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
11561 	kvm_free_pit(kvm);
11562 }
11563 
11564 #define  ERR_PTR_USR(e)  ((void __user *)ERR_PTR(e))
11565 
11566 /**
11567  * __x86_set_memory_region: Setup KVM internal memory slot
11568  *
11569  * @kvm: the kvm pointer to the VM.
11570  * @id: the slot ID to setup.
11571  * @gpa: the GPA to install the slot (unused when @size == 0).
11572  * @size: the size of the slot. Set to zero to uninstall a slot.
11573  *
11574  * This function helps to setup a KVM internal memory slot.  Specify
11575  * @size > 0 to install a new slot, while @size == 0 to uninstall a
11576  * slot.  The return code can be one of the following:
11577  *
11578  *   HVA:           on success (uninstall will return a bogus HVA)
11579  *   -errno:        on error
11580  *
11581  * The caller should always use IS_ERR() to check the return value
11582  * before use.  Note, the KVM internal memory slots are guaranteed to
11583  * remain valid and unchanged until the VM is destroyed, i.e., the
11584  * GPA->HVA translation will not change.  However, the HVA is a user
11585  * address, i.e. its accessibility is not guaranteed, and must be
11586  * accessed via __copy_{to,from}_user().
11587  */
11588 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
11589 				      u32 size)
11590 {
11591 	int i, r;
11592 	unsigned long hva, old_npages;
11593 	struct kvm_memslots *slots = kvm_memslots(kvm);
11594 	struct kvm_memory_slot *slot;
11595 
11596 	/* Called with kvm->slots_lock held.  */
11597 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
11598 		return ERR_PTR_USR(-EINVAL);
11599 
11600 	slot = id_to_memslot(slots, id);
11601 	if (size) {
11602 		if (slot && slot->npages)
11603 			return ERR_PTR_USR(-EEXIST);
11604 
11605 		/*
11606 		 * MAP_SHARED to prevent internal slot pages from being moved
11607 		 * by fork()/COW.
11608 		 */
11609 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
11610 			      MAP_SHARED | MAP_ANONYMOUS, 0);
11611 		if (IS_ERR((void *)hva))
11612 			return (void __user *)hva;
11613 	} else {
11614 		if (!slot || !slot->npages)
11615 			return NULL;
11616 
11617 		old_npages = slot->npages;
11618 		hva = slot->userspace_addr;
11619 	}
11620 
11621 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11622 		struct kvm_userspace_memory_region m;
11623 
11624 		m.slot = id | (i << 16);
11625 		m.flags = 0;
11626 		m.guest_phys_addr = gpa;
11627 		m.userspace_addr = hva;
11628 		m.memory_size = size;
11629 		r = __kvm_set_memory_region(kvm, &m);
11630 		if (r < 0)
11631 			return ERR_PTR_USR(r);
11632 	}
11633 
11634 	if (!size)
11635 		vm_munmap(hva, old_npages * PAGE_SIZE);
11636 
11637 	return (void __user *)hva;
11638 }
11639 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
11640 
11641 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
11642 {
11643 	kvm_mmu_pre_destroy_vm(kvm);
11644 }
11645 
11646 void kvm_arch_destroy_vm(struct kvm *kvm)
11647 {
11648 	if (current->mm == kvm->mm) {
11649 		/*
11650 		 * Free memory regions allocated on behalf of userspace,
11651 		 * unless the the memory map has changed due to process exit
11652 		 * or fd copying.
11653 		 */
11654 		mutex_lock(&kvm->slots_lock);
11655 		__x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
11656 					0, 0);
11657 		__x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
11658 					0, 0);
11659 		__x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
11660 		mutex_unlock(&kvm->slots_lock);
11661 	}
11662 	static_call_cond(kvm_x86_vm_destroy)(kvm);
11663 	kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
11664 	kvm_pic_destroy(kvm);
11665 	kvm_ioapic_destroy(kvm);
11666 	kvm_free_vcpus(kvm);
11667 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
11668 	kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
11669 	kvm_mmu_uninit_vm(kvm);
11670 	kvm_page_track_cleanup(kvm);
11671 	kvm_xen_destroy_vm(kvm);
11672 	kvm_hv_destroy_vm(kvm);
11673 }
11674 
11675 static void memslot_rmap_free(struct kvm_memory_slot *slot)
11676 {
11677 	int i;
11678 
11679 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11680 		kvfree(slot->arch.rmap[i]);
11681 		slot->arch.rmap[i] = NULL;
11682 	}
11683 }
11684 
11685 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
11686 {
11687 	int i;
11688 
11689 	memslot_rmap_free(slot);
11690 
11691 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11692 		kvfree(slot->arch.lpage_info[i - 1]);
11693 		slot->arch.lpage_info[i - 1] = NULL;
11694 	}
11695 
11696 	kvm_page_track_free_memslot(slot);
11697 }
11698 
11699 int memslot_rmap_alloc(struct kvm_memory_slot *slot, unsigned long npages)
11700 {
11701 	const int sz = sizeof(*slot->arch.rmap[0]);
11702 	int i;
11703 
11704 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11705 		int level = i + 1;
11706 		int lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11707 
11708 		if (slot->arch.rmap[i])
11709 			continue;
11710 
11711 		slot->arch.rmap[i] = kvcalloc(lpages, sz, GFP_KERNEL_ACCOUNT);
11712 		if (!slot->arch.rmap[i]) {
11713 			memslot_rmap_free(slot);
11714 			return -ENOMEM;
11715 		}
11716 	}
11717 
11718 	return 0;
11719 }
11720 
11721 static int kvm_alloc_memslot_metadata(struct kvm *kvm,
11722 				      struct kvm_memory_slot *slot)
11723 {
11724 	unsigned long npages = slot->npages;
11725 	int i, r;
11726 
11727 	/*
11728 	 * Clear out the previous array pointers for the KVM_MR_MOVE case.  The
11729 	 * old arrays will be freed by __kvm_set_memory_region() if installing
11730 	 * the new memslot is successful.
11731 	 */
11732 	memset(&slot->arch, 0, sizeof(slot->arch));
11733 
11734 	if (kvm_memslots_have_rmaps(kvm)) {
11735 		r = memslot_rmap_alloc(slot, npages);
11736 		if (r)
11737 			return r;
11738 	}
11739 
11740 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11741 		struct kvm_lpage_info *linfo;
11742 		unsigned long ugfn;
11743 		int lpages;
11744 		int level = i + 1;
11745 
11746 		lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11747 
11748 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
11749 		if (!linfo)
11750 			goto out_free;
11751 
11752 		slot->arch.lpage_info[i - 1] = linfo;
11753 
11754 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
11755 			linfo[0].disallow_lpage = 1;
11756 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
11757 			linfo[lpages - 1].disallow_lpage = 1;
11758 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
11759 		/*
11760 		 * If the gfn and userspace address are not aligned wrt each
11761 		 * other, disable large page support for this slot.
11762 		 */
11763 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
11764 			unsigned long j;
11765 
11766 			for (j = 0; j < lpages; ++j)
11767 				linfo[j].disallow_lpage = 1;
11768 		}
11769 	}
11770 
11771 	if (kvm_page_track_create_memslot(kvm, slot, npages))
11772 		goto out_free;
11773 
11774 	return 0;
11775 
11776 out_free:
11777 	memslot_rmap_free(slot);
11778 
11779 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11780 		kvfree(slot->arch.lpage_info[i - 1]);
11781 		slot->arch.lpage_info[i - 1] = NULL;
11782 	}
11783 	return -ENOMEM;
11784 }
11785 
11786 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
11787 {
11788 	struct kvm_vcpu *vcpu;
11789 	unsigned long i;
11790 
11791 	/*
11792 	 * memslots->generation has been incremented.
11793 	 * mmio generation may have reached its maximum value.
11794 	 */
11795 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
11796 
11797 	/* Force re-initialization of steal_time cache */
11798 	kvm_for_each_vcpu(i, vcpu, kvm)
11799 		kvm_vcpu_kick(vcpu);
11800 }
11801 
11802 int kvm_arch_prepare_memory_region(struct kvm *kvm,
11803 				   const struct kvm_memory_slot *old,
11804 				   struct kvm_memory_slot *new,
11805 				   enum kvm_mr_change change)
11806 {
11807 	if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
11808 		return kvm_alloc_memslot_metadata(kvm, new);
11809 
11810 	if (change == KVM_MR_FLAGS_ONLY)
11811 		memcpy(&new->arch, &old->arch, sizeof(old->arch));
11812 	else if (WARN_ON_ONCE(change != KVM_MR_DELETE))
11813 		return -EIO;
11814 
11815 	return 0;
11816 }
11817 
11818 
11819 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
11820 {
11821 	struct kvm_arch *ka = &kvm->arch;
11822 
11823 	if (!kvm_x86_ops.cpu_dirty_log_size)
11824 		return;
11825 
11826 	if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
11827 	    (!enable && --ka->cpu_dirty_logging_count == 0))
11828 		kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
11829 
11830 	WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
11831 }
11832 
11833 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
11834 				     struct kvm_memory_slot *old,
11835 				     const struct kvm_memory_slot *new,
11836 				     enum kvm_mr_change change)
11837 {
11838 	u32 old_flags = old ? old->flags : 0;
11839 	u32 new_flags = new ? new->flags : 0;
11840 	bool log_dirty_pages = new_flags & KVM_MEM_LOG_DIRTY_PAGES;
11841 
11842 	/*
11843 	 * Update CPU dirty logging if dirty logging is being toggled.  This
11844 	 * applies to all operations.
11845 	 */
11846 	if ((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES)
11847 		kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
11848 
11849 	/*
11850 	 * Nothing more to do for RO slots (which can't be dirtied and can't be
11851 	 * made writable) or CREATE/MOVE/DELETE of a slot.
11852 	 *
11853 	 * For a memslot with dirty logging disabled:
11854 	 * CREATE:      No dirty mappings will already exist.
11855 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
11856 	 *		kvm_arch_flush_shadow_memslot()
11857 	 *
11858 	 * For a memslot with dirty logging enabled:
11859 	 * CREATE:      No shadow pages exist, thus nothing to write-protect
11860 	 *		and no dirty bits to clear.
11861 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
11862 	 *		kvm_arch_flush_shadow_memslot().
11863 	 */
11864 	if ((change != KVM_MR_FLAGS_ONLY) || (new_flags & KVM_MEM_READONLY))
11865 		return;
11866 
11867 	/*
11868 	 * READONLY and non-flags changes were filtered out above, and the only
11869 	 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
11870 	 * logging isn't being toggled on or off.
11871 	 */
11872 	if (WARN_ON_ONCE(!((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES)))
11873 		return;
11874 
11875 	if (!log_dirty_pages) {
11876 		/*
11877 		 * Dirty logging tracks sptes in 4k granularity, meaning that
11878 		 * large sptes have to be split.  If live migration succeeds,
11879 		 * the guest in the source machine will be destroyed and large
11880 		 * sptes will be created in the destination.  However, if the
11881 		 * guest continues to run in the source machine (for example if
11882 		 * live migration fails), small sptes will remain around and
11883 		 * cause bad performance.
11884 		 *
11885 		 * Scan sptes if dirty logging has been stopped, dropping those
11886 		 * which can be collapsed into a single large-page spte.  Later
11887 		 * page faults will create the large-page sptes.
11888 		 */
11889 		kvm_mmu_zap_collapsible_sptes(kvm, new);
11890 	} else {
11891 		/*
11892 		 * Initially-all-set does not require write protecting any page,
11893 		 * because they're all assumed to be dirty.
11894 		 */
11895 		if (kvm_dirty_log_manual_protect_and_init_set(kvm))
11896 			return;
11897 
11898 		if (kvm_x86_ops.cpu_dirty_log_size) {
11899 			kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11900 			kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M);
11901 		} else {
11902 			kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
11903 		}
11904 	}
11905 }
11906 
11907 void kvm_arch_commit_memory_region(struct kvm *kvm,
11908 				struct kvm_memory_slot *old,
11909 				const struct kvm_memory_slot *new,
11910 				enum kvm_mr_change change)
11911 {
11912 	if (!kvm->arch.n_requested_mmu_pages &&
11913 	    (change == KVM_MR_CREATE || change == KVM_MR_DELETE)) {
11914 		unsigned long nr_mmu_pages;
11915 
11916 		nr_mmu_pages = kvm->nr_memslot_pages / KVM_MEMSLOT_PAGES_TO_MMU_PAGES_RATIO;
11917 		nr_mmu_pages = max(nr_mmu_pages, KVM_MIN_ALLOC_MMU_PAGES);
11918 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
11919 	}
11920 
11921 	kvm_mmu_slot_apply_flags(kvm, old, new, change);
11922 
11923 	/* Free the arrays associated with the old memslot. */
11924 	if (change == KVM_MR_MOVE)
11925 		kvm_arch_free_memslot(kvm, old);
11926 }
11927 
11928 void kvm_arch_flush_shadow_all(struct kvm *kvm)
11929 {
11930 	kvm_mmu_zap_all(kvm);
11931 }
11932 
11933 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
11934 				   struct kvm_memory_slot *slot)
11935 {
11936 	kvm_page_track_flush_slot(kvm, slot);
11937 }
11938 
11939 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
11940 {
11941 	return (is_guest_mode(vcpu) &&
11942 			kvm_x86_ops.guest_apic_has_interrupt &&
11943 			static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
11944 }
11945 
11946 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
11947 {
11948 	if (!list_empty_careful(&vcpu->async_pf.done))
11949 		return true;
11950 
11951 	if (kvm_apic_has_events(vcpu))
11952 		return true;
11953 
11954 	if (vcpu->arch.pv.pv_unhalted)
11955 		return true;
11956 
11957 	if (vcpu->arch.exception.pending)
11958 		return true;
11959 
11960 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11961 	    (vcpu->arch.nmi_pending &&
11962 	     static_call(kvm_x86_nmi_allowed)(vcpu, false)))
11963 		return true;
11964 
11965 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
11966 	    (vcpu->arch.smi_pending &&
11967 	     static_call(kvm_x86_smi_allowed)(vcpu, false)))
11968 		return true;
11969 
11970 	if (kvm_arch_interrupt_allowed(vcpu) &&
11971 	    (kvm_cpu_has_interrupt(vcpu) ||
11972 	    kvm_guest_apic_has_interrupt(vcpu)))
11973 		return true;
11974 
11975 	if (kvm_hv_has_stimer_pending(vcpu))
11976 		return true;
11977 
11978 	if (is_guest_mode(vcpu) &&
11979 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
11980 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
11981 		return true;
11982 
11983 	return false;
11984 }
11985 
11986 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
11987 {
11988 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
11989 }
11990 
11991 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
11992 {
11993 	if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11994 		return true;
11995 
11996 	return false;
11997 }
11998 
11999 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
12000 {
12001 	if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
12002 		return true;
12003 
12004 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
12005 		kvm_test_request(KVM_REQ_SMI, vcpu) ||
12006 		 kvm_test_request(KVM_REQ_EVENT, vcpu))
12007 		return true;
12008 
12009 	return kvm_arch_dy_has_pending_interrupt(vcpu);
12010 }
12011 
12012 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
12013 {
12014 	if (vcpu->arch.guest_state_protected)
12015 		return true;
12016 
12017 	return vcpu->arch.preempted_in_kernel;
12018 }
12019 
12020 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
12021 {
12022 	return kvm_rip_read(vcpu);
12023 }
12024 
12025 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
12026 {
12027 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
12028 }
12029 
12030 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
12031 {
12032 	return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
12033 }
12034 
12035 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
12036 {
12037 	/* Can't read the RIP when guest state is protected, just return 0 */
12038 	if (vcpu->arch.guest_state_protected)
12039 		return 0;
12040 
12041 	if (is_64_bit_mode(vcpu))
12042 		return kvm_rip_read(vcpu);
12043 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
12044 		     kvm_rip_read(vcpu));
12045 }
12046 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
12047 
12048 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
12049 {
12050 	return kvm_get_linear_rip(vcpu) == linear_rip;
12051 }
12052 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
12053 
12054 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
12055 {
12056 	unsigned long rflags;
12057 
12058 	rflags = static_call(kvm_x86_get_rflags)(vcpu);
12059 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
12060 		rflags &= ~X86_EFLAGS_TF;
12061 	return rflags;
12062 }
12063 EXPORT_SYMBOL_GPL(kvm_get_rflags);
12064 
12065 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
12066 {
12067 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
12068 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
12069 		rflags |= X86_EFLAGS_TF;
12070 	static_call(kvm_x86_set_rflags)(vcpu, rflags);
12071 }
12072 
12073 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
12074 {
12075 	__kvm_set_rflags(vcpu, rflags);
12076 	kvm_make_request(KVM_REQ_EVENT, vcpu);
12077 }
12078 EXPORT_SYMBOL_GPL(kvm_set_rflags);
12079 
12080 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
12081 {
12082 	int r;
12083 
12084 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
12085 	      work->wakeup_all)
12086 		return;
12087 
12088 	r = kvm_mmu_reload(vcpu);
12089 	if (unlikely(r))
12090 		return;
12091 
12092 	if (!vcpu->arch.mmu->direct_map &&
12093 	      work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
12094 		return;
12095 
12096 	kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
12097 }
12098 
12099 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
12100 {
12101 	BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
12102 
12103 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
12104 }
12105 
12106 static inline u32 kvm_async_pf_next_probe(u32 key)
12107 {
12108 	return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
12109 }
12110 
12111 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12112 {
12113 	u32 key = kvm_async_pf_hash_fn(gfn);
12114 
12115 	while (vcpu->arch.apf.gfns[key] != ~0)
12116 		key = kvm_async_pf_next_probe(key);
12117 
12118 	vcpu->arch.apf.gfns[key] = gfn;
12119 }
12120 
12121 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
12122 {
12123 	int i;
12124 	u32 key = kvm_async_pf_hash_fn(gfn);
12125 
12126 	for (i = 0; i < ASYNC_PF_PER_VCPU &&
12127 		     (vcpu->arch.apf.gfns[key] != gfn &&
12128 		      vcpu->arch.apf.gfns[key] != ~0); i++)
12129 		key = kvm_async_pf_next_probe(key);
12130 
12131 	return key;
12132 }
12133 
12134 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12135 {
12136 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
12137 }
12138 
12139 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12140 {
12141 	u32 i, j, k;
12142 
12143 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
12144 
12145 	if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
12146 		return;
12147 
12148 	while (true) {
12149 		vcpu->arch.apf.gfns[i] = ~0;
12150 		do {
12151 			j = kvm_async_pf_next_probe(j);
12152 			if (vcpu->arch.apf.gfns[j] == ~0)
12153 				return;
12154 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
12155 			/*
12156 			 * k lies cyclically in ]i,j]
12157 			 * |    i.k.j |
12158 			 * |....j i.k.| or  |.k..j i...|
12159 			 */
12160 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
12161 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
12162 		i = j;
12163 	}
12164 }
12165 
12166 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
12167 {
12168 	u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
12169 
12170 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
12171 				      sizeof(reason));
12172 }
12173 
12174 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
12175 {
12176 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
12177 
12178 	return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
12179 					     &token, offset, sizeof(token));
12180 }
12181 
12182 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
12183 {
12184 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
12185 	u32 val;
12186 
12187 	if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
12188 					 &val, offset, sizeof(val)))
12189 		return false;
12190 
12191 	return !val;
12192 }
12193 
12194 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
12195 {
12196 	if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
12197 		return false;
12198 
12199 	if (!kvm_pv_async_pf_enabled(vcpu) ||
12200 	    (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
12201 		return false;
12202 
12203 	return true;
12204 }
12205 
12206 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
12207 {
12208 	if (unlikely(!lapic_in_kernel(vcpu) ||
12209 		     kvm_event_needs_reinjection(vcpu) ||
12210 		     vcpu->arch.exception.pending))
12211 		return false;
12212 
12213 	if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
12214 		return false;
12215 
12216 	/*
12217 	 * If interrupts are off we cannot even use an artificial
12218 	 * halt state.
12219 	 */
12220 	return kvm_arch_interrupt_allowed(vcpu);
12221 }
12222 
12223 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
12224 				     struct kvm_async_pf *work)
12225 {
12226 	struct x86_exception fault;
12227 
12228 	trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
12229 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
12230 
12231 	if (kvm_can_deliver_async_pf(vcpu) &&
12232 	    !apf_put_user_notpresent(vcpu)) {
12233 		fault.vector = PF_VECTOR;
12234 		fault.error_code_valid = true;
12235 		fault.error_code = 0;
12236 		fault.nested_page_fault = false;
12237 		fault.address = work->arch.token;
12238 		fault.async_page_fault = true;
12239 		kvm_inject_page_fault(vcpu, &fault);
12240 		return true;
12241 	} else {
12242 		/*
12243 		 * It is not possible to deliver a paravirtualized asynchronous
12244 		 * page fault, but putting the guest in an artificial halt state
12245 		 * can be beneficial nevertheless: if an interrupt arrives, we
12246 		 * can deliver it timely and perhaps the guest will schedule
12247 		 * another process.  When the instruction that triggered a page
12248 		 * fault is retried, hopefully the page will be ready in the host.
12249 		 */
12250 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
12251 		return false;
12252 	}
12253 }
12254 
12255 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
12256 				 struct kvm_async_pf *work)
12257 {
12258 	struct kvm_lapic_irq irq = {
12259 		.delivery_mode = APIC_DM_FIXED,
12260 		.vector = vcpu->arch.apf.vec
12261 	};
12262 
12263 	if (work->wakeup_all)
12264 		work->arch.token = ~0; /* broadcast wakeup */
12265 	else
12266 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
12267 	trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
12268 
12269 	if ((work->wakeup_all || work->notpresent_injected) &&
12270 	    kvm_pv_async_pf_enabled(vcpu) &&
12271 	    !apf_put_user_ready(vcpu, work->arch.token)) {
12272 		vcpu->arch.apf.pageready_pending = true;
12273 		kvm_apic_set_irq(vcpu, &irq, NULL);
12274 	}
12275 
12276 	vcpu->arch.apf.halted = false;
12277 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
12278 }
12279 
12280 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
12281 {
12282 	kvm_make_request(KVM_REQ_APF_READY, vcpu);
12283 	if (!vcpu->arch.apf.pageready_pending)
12284 		kvm_vcpu_kick(vcpu);
12285 }
12286 
12287 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
12288 {
12289 	if (!kvm_pv_async_pf_enabled(vcpu))
12290 		return true;
12291 	else
12292 		return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
12293 }
12294 
12295 void kvm_arch_start_assignment(struct kvm *kvm)
12296 {
12297 	if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
12298 		static_call_cond(kvm_x86_start_assignment)(kvm);
12299 }
12300 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
12301 
12302 void kvm_arch_end_assignment(struct kvm *kvm)
12303 {
12304 	atomic_dec(&kvm->arch.assigned_device_count);
12305 }
12306 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
12307 
12308 bool kvm_arch_has_assigned_device(struct kvm *kvm)
12309 {
12310 	return atomic_read(&kvm->arch.assigned_device_count);
12311 }
12312 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
12313 
12314 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
12315 {
12316 	atomic_inc(&kvm->arch.noncoherent_dma_count);
12317 }
12318 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
12319 
12320 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
12321 {
12322 	atomic_dec(&kvm->arch.noncoherent_dma_count);
12323 }
12324 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
12325 
12326 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
12327 {
12328 	return atomic_read(&kvm->arch.noncoherent_dma_count);
12329 }
12330 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
12331 
12332 bool kvm_arch_has_irq_bypass(void)
12333 {
12334 	return true;
12335 }
12336 
12337 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
12338 				      struct irq_bypass_producer *prod)
12339 {
12340 	struct kvm_kernel_irqfd *irqfd =
12341 		container_of(cons, struct kvm_kernel_irqfd, consumer);
12342 	int ret;
12343 
12344 	irqfd->producer = prod;
12345 	kvm_arch_start_assignment(irqfd->kvm);
12346 	ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
12347 					 prod->irq, irqfd->gsi, 1);
12348 
12349 	if (ret)
12350 		kvm_arch_end_assignment(irqfd->kvm);
12351 
12352 	return ret;
12353 }
12354 
12355 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
12356 				      struct irq_bypass_producer *prod)
12357 {
12358 	int ret;
12359 	struct kvm_kernel_irqfd *irqfd =
12360 		container_of(cons, struct kvm_kernel_irqfd, consumer);
12361 
12362 	WARN_ON(irqfd->producer != prod);
12363 	irqfd->producer = NULL;
12364 
12365 	/*
12366 	 * When producer of consumer is unregistered, we change back to
12367 	 * remapped mode, so we can re-use the current implementation
12368 	 * when the irq is masked/disabled or the consumer side (KVM
12369 	 * int this case doesn't want to receive the interrupts.
12370 	*/
12371 	ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
12372 	if (ret)
12373 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
12374 		       " fails: %d\n", irqfd->consumer.token, ret);
12375 
12376 	kvm_arch_end_assignment(irqfd->kvm);
12377 }
12378 
12379 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
12380 				   uint32_t guest_irq, bool set)
12381 {
12382 	return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
12383 }
12384 
12385 bool kvm_arch_irqfd_route_changed(struct kvm_kernel_irq_routing_entry *old,
12386 				  struct kvm_kernel_irq_routing_entry *new)
12387 {
12388 	if (new->type != KVM_IRQ_ROUTING_MSI)
12389 		return true;
12390 
12391 	return !!memcmp(&old->msi, &new->msi, sizeof(new->msi));
12392 }
12393 
12394 bool kvm_vector_hashing_enabled(void)
12395 {
12396 	return vector_hashing;
12397 }
12398 
12399 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
12400 {
12401 	return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
12402 }
12403 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
12404 
12405 
12406 int kvm_spec_ctrl_test_value(u64 value)
12407 {
12408 	/*
12409 	 * test that setting IA32_SPEC_CTRL to given value
12410 	 * is allowed by the host processor
12411 	 */
12412 
12413 	u64 saved_value;
12414 	unsigned long flags;
12415 	int ret = 0;
12416 
12417 	local_irq_save(flags);
12418 
12419 	if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
12420 		ret = 1;
12421 	else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
12422 		ret = 1;
12423 	else
12424 		wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
12425 
12426 	local_irq_restore(flags);
12427 
12428 	return ret;
12429 }
12430 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
12431 
12432 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
12433 {
12434 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
12435 	struct x86_exception fault;
12436 	u32 access = error_code &
12437 		(PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
12438 
12439 	if (!(error_code & PFERR_PRESENT_MASK) ||
12440 	    mmu->gva_to_gpa(vcpu, mmu, gva, access, &fault) != UNMAPPED_GVA) {
12441 		/*
12442 		 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
12443 		 * tables probably do not match the TLB.  Just proceed
12444 		 * with the error code that the processor gave.
12445 		 */
12446 		fault.vector = PF_VECTOR;
12447 		fault.error_code_valid = true;
12448 		fault.error_code = error_code;
12449 		fault.nested_page_fault = false;
12450 		fault.address = gva;
12451 	}
12452 	vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
12453 }
12454 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
12455 
12456 /*
12457  * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
12458  * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
12459  * indicates whether exit to userspace is needed.
12460  */
12461 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
12462 			      struct x86_exception *e)
12463 {
12464 	if (r == X86EMUL_PROPAGATE_FAULT) {
12465 		kvm_inject_emulated_page_fault(vcpu, e);
12466 		return 1;
12467 	}
12468 
12469 	/*
12470 	 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
12471 	 * while handling a VMX instruction KVM could've handled the request
12472 	 * correctly by exiting to userspace and performing I/O but there
12473 	 * doesn't seem to be a real use-case behind such requests, just return
12474 	 * KVM_EXIT_INTERNAL_ERROR for now.
12475 	 */
12476 	kvm_prepare_emulation_failure_exit(vcpu);
12477 
12478 	return 0;
12479 }
12480 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
12481 
12482 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
12483 {
12484 	bool pcid_enabled;
12485 	struct x86_exception e;
12486 	struct {
12487 		u64 pcid;
12488 		u64 gla;
12489 	} operand;
12490 	int r;
12491 
12492 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
12493 	if (r != X86EMUL_CONTINUE)
12494 		return kvm_handle_memory_failure(vcpu, r, &e);
12495 
12496 	if (operand.pcid >> 12 != 0) {
12497 		kvm_inject_gp(vcpu, 0);
12498 		return 1;
12499 	}
12500 
12501 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
12502 
12503 	switch (type) {
12504 	case INVPCID_TYPE_INDIV_ADDR:
12505 		if ((!pcid_enabled && (operand.pcid != 0)) ||
12506 		    is_noncanonical_address(operand.gla, vcpu)) {
12507 			kvm_inject_gp(vcpu, 0);
12508 			return 1;
12509 		}
12510 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
12511 		return kvm_skip_emulated_instruction(vcpu);
12512 
12513 	case INVPCID_TYPE_SINGLE_CTXT:
12514 		if (!pcid_enabled && (operand.pcid != 0)) {
12515 			kvm_inject_gp(vcpu, 0);
12516 			return 1;
12517 		}
12518 
12519 		kvm_invalidate_pcid(vcpu, operand.pcid);
12520 		return kvm_skip_emulated_instruction(vcpu);
12521 
12522 	case INVPCID_TYPE_ALL_NON_GLOBAL:
12523 		/*
12524 		 * Currently, KVM doesn't mark global entries in the shadow
12525 		 * page tables, so a non-global flush just degenerates to a
12526 		 * global flush. If needed, we could optimize this later by
12527 		 * keeping track of global entries in shadow page tables.
12528 		 */
12529 
12530 		fallthrough;
12531 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
12532 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
12533 		return kvm_skip_emulated_instruction(vcpu);
12534 
12535 	default:
12536 		kvm_inject_gp(vcpu, 0);
12537 		return 1;
12538 	}
12539 }
12540 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
12541 
12542 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
12543 {
12544 	struct kvm_run *run = vcpu->run;
12545 	struct kvm_mmio_fragment *frag;
12546 	unsigned int len;
12547 
12548 	BUG_ON(!vcpu->mmio_needed);
12549 
12550 	/* Complete previous fragment */
12551 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
12552 	len = min(8u, frag->len);
12553 	if (!vcpu->mmio_is_write)
12554 		memcpy(frag->data, run->mmio.data, len);
12555 
12556 	if (frag->len <= 8) {
12557 		/* Switch to the next fragment. */
12558 		frag++;
12559 		vcpu->mmio_cur_fragment++;
12560 	} else {
12561 		/* Go forward to the next mmio piece. */
12562 		frag->data += len;
12563 		frag->gpa += len;
12564 		frag->len -= len;
12565 	}
12566 
12567 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
12568 		vcpu->mmio_needed = 0;
12569 
12570 		// VMG change, at this point, we're always done
12571 		// RIP has already been advanced
12572 		return 1;
12573 	}
12574 
12575 	// More MMIO is needed
12576 	run->mmio.phys_addr = frag->gpa;
12577 	run->mmio.len = min(8u, frag->len);
12578 	run->mmio.is_write = vcpu->mmio_is_write;
12579 	if (run->mmio.is_write)
12580 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
12581 	run->exit_reason = KVM_EXIT_MMIO;
12582 
12583 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12584 
12585 	return 0;
12586 }
12587 
12588 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12589 			  void *data)
12590 {
12591 	int handled;
12592 	struct kvm_mmio_fragment *frag;
12593 
12594 	if (!data)
12595 		return -EINVAL;
12596 
12597 	handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12598 	if (handled == bytes)
12599 		return 1;
12600 
12601 	bytes -= handled;
12602 	gpa += handled;
12603 	data += handled;
12604 
12605 	/*TODO: Check if need to increment number of frags */
12606 	frag = vcpu->mmio_fragments;
12607 	vcpu->mmio_nr_fragments = 1;
12608 	frag->len = bytes;
12609 	frag->gpa = gpa;
12610 	frag->data = data;
12611 
12612 	vcpu->mmio_needed = 1;
12613 	vcpu->mmio_cur_fragment = 0;
12614 
12615 	vcpu->run->mmio.phys_addr = gpa;
12616 	vcpu->run->mmio.len = min(8u, frag->len);
12617 	vcpu->run->mmio.is_write = 1;
12618 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
12619 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
12620 
12621 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12622 
12623 	return 0;
12624 }
12625 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
12626 
12627 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12628 			 void *data)
12629 {
12630 	int handled;
12631 	struct kvm_mmio_fragment *frag;
12632 
12633 	if (!data)
12634 		return -EINVAL;
12635 
12636 	handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12637 	if (handled == bytes)
12638 		return 1;
12639 
12640 	bytes -= handled;
12641 	gpa += handled;
12642 	data += handled;
12643 
12644 	/*TODO: Check if need to increment number of frags */
12645 	frag = vcpu->mmio_fragments;
12646 	vcpu->mmio_nr_fragments = 1;
12647 	frag->len = bytes;
12648 	frag->gpa = gpa;
12649 	frag->data = data;
12650 
12651 	vcpu->mmio_needed = 1;
12652 	vcpu->mmio_cur_fragment = 0;
12653 
12654 	vcpu->run->mmio.phys_addr = gpa;
12655 	vcpu->run->mmio.len = min(8u, frag->len);
12656 	vcpu->run->mmio.is_write = 0;
12657 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
12658 
12659 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12660 
12661 	return 0;
12662 }
12663 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
12664 
12665 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12666 			   unsigned int port);
12667 
12668 static int complete_sev_es_emulated_outs(struct kvm_vcpu *vcpu)
12669 {
12670 	int size = vcpu->arch.pio.size;
12671 	int port = vcpu->arch.pio.port;
12672 
12673 	vcpu->arch.pio.count = 0;
12674 	if (vcpu->arch.sev_pio_count)
12675 		return kvm_sev_es_outs(vcpu, size, port);
12676 	return 1;
12677 }
12678 
12679 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12680 			   unsigned int port)
12681 {
12682 	for (;;) {
12683 		unsigned int count =
12684 			min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
12685 		int ret = emulator_pio_out(vcpu, size, port, vcpu->arch.sev_pio_data, count);
12686 
12687 		/* memcpy done already by emulator_pio_out.  */
12688 		vcpu->arch.sev_pio_count -= count;
12689 		vcpu->arch.sev_pio_data += count * vcpu->arch.pio.size;
12690 		if (!ret)
12691 			break;
12692 
12693 		/* Emulation done by the kernel.  */
12694 		if (!vcpu->arch.sev_pio_count)
12695 			return 1;
12696 	}
12697 
12698 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_outs;
12699 	return 0;
12700 }
12701 
12702 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12703 			  unsigned int port);
12704 
12705 static void advance_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12706 {
12707 	unsigned count = vcpu->arch.pio.count;
12708 	complete_emulator_pio_in(vcpu, vcpu->arch.sev_pio_data);
12709 	vcpu->arch.sev_pio_count -= count;
12710 	vcpu->arch.sev_pio_data += count * vcpu->arch.pio.size;
12711 }
12712 
12713 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12714 {
12715 	int size = vcpu->arch.pio.size;
12716 	int port = vcpu->arch.pio.port;
12717 
12718 	advance_sev_es_emulated_ins(vcpu);
12719 	if (vcpu->arch.sev_pio_count)
12720 		return kvm_sev_es_ins(vcpu, size, port);
12721 	return 1;
12722 }
12723 
12724 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12725 			  unsigned int port)
12726 {
12727 	for (;;) {
12728 		unsigned int count =
12729 			min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
12730 		if (!__emulator_pio_in(vcpu, size, port, count))
12731 			break;
12732 
12733 		/* Emulation done by the kernel.  */
12734 		advance_sev_es_emulated_ins(vcpu);
12735 		if (!vcpu->arch.sev_pio_count)
12736 			return 1;
12737 	}
12738 
12739 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
12740 	return 0;
12741 }
12742 
12743 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
12744 			 unsigned int port, void *data,  unsigned int count,
12745 			 int in)
12746 {
12747 	vcpu->arch.sev_pio_data = data;
12748 	vcpu->arch.sev_pio_count = count;
12749 	return in ? kvm_sev_es_ins(vcpu, size, port)
12750 		  : kvm_sev_es_outs(vcpu, size, port);
12751 }
12752 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
12753 
12754 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
12755 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
12756 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
12757 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
12758 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
12759 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
12760 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
12761 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
12762 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
12763 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
12764 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
12765 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
12766 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
12767 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
12768 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
12769 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
12770 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
12771 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
12772 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
12773 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
12774 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
12775 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
12776 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
12777 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_accept_irq);
12778 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
12779 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
12780 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
12781 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);
12782